relay: Use per CPU constructs for the relay channel buffer pointers
[linux-block.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 /**
517  *      relay_hotcpu_callback - CPU hotplug callback
518  *      @nb: notifier block
519  *      @action: hotplug action to take
520  *      @hcpu: CPU number
521  *
522  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
523  */
524 static int relay_hotcpu_callback(struct notifier_block *nb,
525                                 unsigned long action,
526                                 void *hcpu)
527 {
528         unsigned int hotcpu = (unsigned long)hcpu;
529         struct rchan *chan;
530         struct rchan_buf *buf;
531
532         switch(action) {
533         case CPU_UP_PREPARE:
534         case CPU_UP_PREPARE_FROZEN:
535                 mutex_lock(&relay_channels_mutex);
536                 list_for_each_entry(chan, &relay_channels, list) {
537                         if ((buf = *per_cpu_ptr(chan->buf, hotcpu)))
538                                 continue;
539                         buf = relay_open_buf(chan, hotcpu);
540                         if (!buf) {
541                                 printk(KERN_ERR
542                                         "relay_hotcpu_callback: cpu %d buffer "
543                                         "creation failed\n", hotcpu);
544                                 mutex_unlock(&relay_channels_mutex);
545                                 return notifier_from_errno(-ENOMEM);
546                         }
547                         *per_cpu_ptr(chan->buf, hotcpu) = buf;
548                 }
549                 mutex_unlock(&relay_channels_mutex);
550                 break;
551         case CPU_DEAD:
552         case CPU_DEAD_FROZEN:
553                 /* No need to flush the cpu : will be flushed upon
554                  * final relay_flush() call. */
555                 break;
556         }
557         return NOTIFY_OK;
558 }
559
560 /**
561  *      relay_open - create a new relay channel
562  *      @base_filename: base name of files to create, %NULL for buffering only
563  *      @parent: dentry of parent directory, %NULL for root directory or buffer
564  *      @subbuf_size: size of sub-buffers
565  *      @n_subbufs: number of sub-buffers
566  *      @cb: client callback functions
567  *      @private_data: user-defined data
568  *
569  *      Returns channel pointer if successful, %NULL otherwise.
570  *
571  *      Creates a channel buffer for each cpu using the sizes and
572  *      attributes specified.  The created channel buffer files
573  *      will be named base_filename0...base_filenameN-1.  File
574  *      permissions will be %S_IRUSR.
575  *
576  *      If opening a buffer (@parent = NULL) that you later wish to register
577  *      in a filesystem, call relay_late_setup_files() once the @parent dentry
578  *      is available.
579  */
580 struct rchan *relay_open(const char *base_filename,
581                          struct dentry *parent,
582                          size_t subbuf_size,
583                          size_t n_subbufs,
584                          struct rchan_callbacks *cb,
585                          void *private_data)
586 {
587         unsigned int i;
588         struct rchan *chan;
589         struct rchan_buf *buf;
590
591         if (!(subbuf_size && n_subbufs))
592                 return NULL;
593         if (subbuf_size > UINT_MAX / n_subbufs)
594                 return NULL;
595
596         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
597         if (!chan)
598                 return NULL;
599
600         chan->buf = alloc_percpu(struct rchan_buf *);
601         chan->version = RELAYFS_CHANNEL_VERSION;
602         chan->n_subbufs = n_subbufs;
603         chan->subbuf_size = subbuf_size;
604         chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
605         chan->parent = parent;
606         chan->private_data = private_data;
607         if (base_filename) {
608                 chan->has_base_filename = 1;
609                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
610         }
611         setup_callbacks(chan, cb);
612         kref_init(&chan->kref);
613
614         mutex_lock(&relay_channels_mutex);
615         for_each_online_cpu(i) {
616                 buf = relay_open_buf(chan, i);
617                 if (!buf)
618                         goto free_bufs;
619                 *per_cpu_ptr(chan->buf, i) = buf;
620         }
621         list_add(&chan->list, &relay_channels);
622         mutex_unlock(&relay_channels_mutex);
623
624         return chan;
625
626 free_bufs:
627         for_each_possible_cpu(i) {
628                 if ((buf = *per_cpu_ptr(chan->buf, i)))
629                         relay_close_buf(buf);
630         }
631
632         kref_put(&chan->kref, relay_destroy_channel);
633         mutex_unlock(&relay_channels_mutex);
634         kfree(chan);
635         return NULL;
636 }
637 EXPORT_SYMBOL_GPL(relay_open);
638
639 struct rchan_percpu_buf_dispatcher {
640         struct rchan_buf *buf;
641         struct dentry *dentry;
642 };
643
644 /* Called in atomic context. */
645 static void __relay_set_buf_dentry(void *info)
646 {
647         struct rchan_percpu_buf_dispatcher *p = info;
648
649         relay_set_buf_dentry(p->buf, p->dentry);
650 }
651
652 /**
653  *      relay_late_setup_files - triggers file creation
654  *      @chan: channel to operate on
655  *      @base_filename: base name of files to create
656  *      @parent: dentry of parent directory, %NULL for root directory
657  *
658  *      Returns 0 if successful, non-zero otherwise.
659  *
660  *      Use to setup files for a previously buffer-only channel created
661  *      by relay_open() with a NULL parent dentry.
662  *
663  *      For example, this is useful for perfomring early tracing in kernel,
664  *      before VFS is up and then exposing the early results once the dentry
665  *      is available.
666  */
667 int relay_late_setup_files(struct rchan *chan,
668                            const char *base_filename,
669                            struct dentry *parent)
670 {
671         int err = 0;
672         unsigned int i, curr_cpu;
673         unsigned long flags;
674         struct dentry *dentry;
675         struct rchan_buf *buf;
676         struct rchan_percpu_buf_dispatcher disp;
677
678         if (!chan || !base_filename)
679                 return -EINVAL;
680
681         strlcpy(chan->base_filename, base_filename, NAME_MAX);
682
683         mutex_lock(&relay_channels_mutex);
684         /* Is chan already set up? */
685         if (unlikely(chan->has_base_filename)) {
686                 mutex_unlock(&relay_channels_mutex);
687                 return -EEXIST;
688         }
689         chan->has_base_filename = 1;
690         chan->parent = parent;
691
692         if (chan->is_global) {
693                 err = -EINVAL;
694                 buf = *per_cpu_ptr(chan->buf, 0);
695                 if (!WARN_ON_ONCE(!buf)) {
696                         dentry = relay_create_buf_file(chan, buf, 0);
697                         if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
698                                 relay_set_buf_dentry(buf, dentry);
699                                 err = 0;
700                         }
701                 }
702                 mutex_unlock(&relay_channels_mutex);
703                 return err;
704         }
705
706         curr_cpu = get_cpu();
707         /*
708          * The CPU hotplug notifier ran before us and created buffers with
709          * no files associated. So it's safe to call relay_setup_buf_file()
710          * on all currently online CPUs.
711          */
712         for_each_online_cpu(i) {
713                 buf = *per_cpu_ptr(chan->buf, i);
714                 if (unlikely(!buf)) {
715                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
716                         err = -EINVAL;
717                         break;
718                 }
719
720                 dentry = relay_create_buf_file(chan, buf, i);
721                 if (unlikely(!dentry)) {
722                         err = -EINVAL;
723                         break;
724                 }
725
726                 if (curr_cpu == i) {
727                         local_irq_save(flags);
728                         relay_set_buf_dentry(buf, dentry);
729                         local_irq_restore(flags);
730                 } else {
731                         disp.buf = buf;
732                         disp.dentry = dentry;
733                         smp_mb();
734                         /* relay_channels_mutex must be held, so wait. */
735                         err = smp_call_function_single(i,
736                                                        __relay_set_buf_dentry,
737                                                        &disp, 1);
738                 }
739                 if (unlikely(err))
740                         break;
741         }
742         put_cpu();
743         mutex_unlock(&relay_channels_mutex);
744
745         return err;
746 }
747 EXPORT_SYMBOL_GPL(relay_late_setup_files);
748
749 /**
750  *      relay_switch_subbuf - switch to a new sub-buffer
751  *      @buf: channel buffer
752  *      @length: size of current event
753  *
754  *      Returns either the length passed in or 0 if full.
755  *
756  *      Performs sub-buffer-switch tasks such as invoking callbacks,
757  *      updating padding counts, waking up readers, etc.
758  */
759 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
760 {
761         void *old, *new;
762         size_t old_subbuf, new_subbuf;
763
764         if (unlikely(length > buf->chan->subbuf_size))
765                 goto toobig;
766
767         if (buf->offset != buf->chan->subbuf_size + 1) {
768                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
769                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
770                 buf->padding[old_subbuf] = buf->prev_padding;
771                 buf->subbufs_produced++;
772                 if (buf->dentry)
773                         d_inode(buf->dentry)->i_size +=
774                                 buf->chan->subbuf_size -
775                                 buf->padding[old_subbuf];
776                 else
777                         buf->early_bytes += buf->chan->subbuf_size -
778                                             buf->padding[old_subbuf];
779                 smp_mb();
780                 if (waitqueue_active(&buf->read_wait))
781                         /*
782                          * Calling wake_up_interruptible() from here
783                          * will deadlock if we happen to be logging
784                          * from the scheduler (trying to re-grab
785                          * rq->lock), so defer it.
786                          */
787                         mod_timer(&buf->timer, jiffies + 1);
788         }
789
790         old = buf->data;
791         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
792         new = buf->start + new_subbuf * buf->chan->subbuf_size;
793         buf->offset = 0;
794         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
795                 buf->offset = buf->chan->subbuf_size + 1;
796                 return 0;
797         }
798         buf->data = new;
799         buf->padding[new_subbuf] = 0;
800
801         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
802                 goto toobig;
803
804         return length;
805
806 toobig:
807         buf->chan->last_toobig = length;
808         return 0;
809 }
810 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
811
812 /**
813  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
814  *      @chan: the channel
815  *      @cpu: the cpu associated with the channel buffer to update
816  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
817  *
818  *      Adds to the channel buffer's consumed sub-buffer count.
819  *      subbufs_consumed should be the number of sub-buffers newly consumed,
820  *      not the total consumed.
821  *
822  *      NOTE. Kernel clients don't need to call this function if the channel
823  *      mode is 'overwrite'.
824  */
825 void relay_subbufs_consumed(struct rchan *chan,
826                             unsigned int cpu,
827                             size_t subbufs_consumed)
828 {
829         struct rchan_buf *buf;
830
831         if (!chan)
832                 return;
833
834         buf = *per_cpu_ptr(chan->buf, cpu);
835         if (cpu >= NR_CPUS || !buf || subbufs_consumed > chan->n_subbufs)
836                 return;
837
838         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
839                 buf->subbufs_consumed = buf->subbufs_produced;
840         else
841                 buf->subbufs_consumed += subbufs_consumed;
842 }
843 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
844
845 /**
846  *      relay_close - close the channel
847  *      @chan: the channel
848  *
849  *      Closes all channel buffers and frees the channel.
850  */
851 void relay_close(struct rchan *chan)
852 {
853         struct rchan_buf *buf;
854         unsigned int i;
855
856         if (!chan)
857                 return;
858
859         mutex_lock(&relay_channels_mutex);
860         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0)))
861                 relay_close_buf(buf);
862         else
863                 for_each_possible_cpu(i)
864                         if ((buf = *per_cpu_ptr(chan->buf, i)))
865                                 relay_close_buf(buf);
866
867         if (chan->last_toobig)
868                 printk(KERN_WARNING "relay: one or more items not logged "
869                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
870                        chan->last_toobig, chan->subbuf_size);
871
872         list_del(&chan->list);
873         kref_put(&chan->kref, relay_destroy_channel);
874         mutex_unlock(&relay_channels_mutex);
875 }
876 EXPORT_SYMBOL_GPL(relay_close);
877
878 /**
879  *      relay_flush - close the channel
880  *      @chan: the channel
881  *
882  *      Flushes all channel buffers, i.e. forces buffer switch.
883  */
884 void relay_flush(struct rchan *chan)
885 {
886         struct rchan_buf *buf;
887         unsigned int i;
888
889         if (!chan)
890                 return;
891
892         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
893                 relay_switch_subbuf(buf, 0);
894                 return;
895         }
896
897         mutex_lock(&relay_channels_mutex);
898         for_each_possible_cpu(i)
899                 if ((buf = *per_cpu_ptr(chan->buf, i)))
900                         relay_switch_subbuf(buf, 0);
901         mutex_unlock(&relay_channels_mutex);
902 }
903 EXPORT_SYMBOL_GPL(relay_flush);
904
905 /**
906  *      relay_file_open - open file op for relay files
907  *      @inode: the inode
908  *      @filp: the file
909  *
910  *      Increments the channel buffer refcount.
911  */
912 static int relay_file_open(struct inode *inode, struct file *filp)
913 {
914         struct rchan_buf *buf = inode->i_private;
915         kref_get(&buf->kref);
916         filp->private_data = buf;
917
918         return nonseekable_open(inode, filp);
919 }
920
921 /**
922  *      relay_file_mmap - mmap file op for relay files
923  *      @filp: the file
924  *      @vma: the vma describing what to map
925  *
926  *      Calls upon relay_mmap_buf() to map the file into user space.
927  */
928 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
929 {
930         struct rchan_buf *buf = filp->private_data;
931         return relay_mmap_buf(buf, vma);
932 }
933
934 /**
935  *      relay_file_poll - poll file op for relay files
936  *      @filp: the file
937  *      @wait: poll table
938  *
939  *      Poll implemention.
940  */
941 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
942 {
943         unsigned int mask = 0;
944         struct rchan_buf *buf = filp->private_data;
945
946         if (buf->finalized)
947                 return POLLERR;
948
949         if (filp->f_mode & FMODE_READ) {
950                 poll_wait(filp, &buf->read_wait, wait);
951                 if (!relay_buf_empty(buf))
952                         mask |= POLLIN | POLLRDNORM;
953         }
954
955         return mask;
956 }
957
958 /**
959  *      relay_file_release - release file op for relay files
960  *      @inode: the inode
961  *      @filp: the file
962  *
963  *      Decrements the channel refcount, as the filesystem is
964  *      no longer using it.
965  */
966 static int relay_file_release(struct inode *inode, struct file *filp)
967 {
968         struct rchan_buf *buf = filp->private_data;
969         kref_put(&buf->kref, relay_remove_buf);
970
971         return 0;
972 }
973
974 /*
975  *      relay_file_read_consume - update the consumed count for the buffer
976  */
977 static void relay_file_read_consume(struct rchan_buf *buf,
978                                     size_t read_pos,
979                                     size_t bytes_consumed)
980 {
981         size_t subbuf_size = buf->chan->subbuf_size;
982         size_t n_subbufs = buf->chan->n_subbufs;
983         size_t read_subbuf;
984
985         if (buf->subbufs_produced == buf->subbufs_consumed &&
986             buf->offset == buf->bytes_consumed)
987                 return;
988
989         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
990                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
991                 buf->bytes_consumed = 0;
992         }
993
994         buf->bytes_consumed += bytes_consumed;
995         if (!read_pos)
996                 read_subbuf = buf->subbufs_consumed % n_subbufs;
997         else
998                 read_subbuf = read_pos / buf->chan->subbuf_size;
999         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
1000                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
1001                     (buf->offset == subbuf_size))
1002                         return;
1003                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
1004                 buf->bytes_consumed = 0;
1005         }
1006 }
1007
1008 /*
1009  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
1010  */
1011 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
1012 {
1013         size_t subbuf_size = buf->chan->subbuf_size;
1014         size_t n_subbufs = buf->chan->n_subbufs;
1015         size_t produced = buf->subbufs_produced;
1016         size_t consumed = buf->subbufs_consumed;
1017
1018         relay_file_read_consume(buf, read_pos, 0);
1019
1020         consumed = buf->subbufs_consumed;
1021
1022         if (unlikely(buf->offset > subbuf_size)) {
1023                 if (produced == consumed)
1024                         return 0;
1025                 return 1;
1026         }
1027
1028         if (unlikely(produced - consumed >= n_subbufs)) {
1029                 consumed = produced - n_subbufs + 1;
1030                 buf->subbufs_consumed = consumed;
1031                 buf->bytes_consumed = 0;
1032         }
1033
1034         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
1035         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
1036
1037         if (consumed > produced)
1038                 produced += n_subbufs * subbuf_size;
1039
1040         if (consumed == produced) {
1041                 if (buf->offset == subbuf_size &&
1042                     buf->subbufs_produced > buf->subbufs_consumed)
1043                         return 1;
1044                 return 0;
1045         }
1046
1047         return 1;
1048 }
1049
1050 /**
1051  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1052  *      @read_pos: file read position
1053  *      @buf: relay channel buffer
1054  */
1055 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1056                                            struct rchan_buf *buf)
1057 {
1058         size_t padding, avail = 0;
1059         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1060         size_t subbuf_size = buf->chan->subbuf_size;
1061
1062         write_subbuf = (buf->data - buf->start) / subbuf_size;
1063         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1064         read_subbuf = read_pos / subbuf_size;
1065         read_offset = read_pos % subbuf_size;
1066         padding = buf->padding[read_subbuf];
1067
1068         if (read_subbuf == write_subbuf) {
1069                 if (read_offset + padding < write_offset)
1070                         avail = write_offset - (read_offset + padding);
1071         } else
1072                 avail = (subbuf_size - padding) - read_offset;
1073
1074         return avail;
1075 }
1076
1077 /**
1078  *      relay_file_read_start_pos - find the first available byte to read
1079  *      @read_pos: file read position
1080  *      @buf: relay channel buffer
1081  *
1082  *      If the @read_pos is in the middle of padding, return the
1083  *      position of the first actually available byte, otherwise
1084  *      return the original value.
1085  */
1086 static size_t relay_file_read_start_pos(size_t read_pos,
1087                                         struct rchan_buf *buf)
1088 {
1089         size_t read_subbuf, padding, padding_start, padding_end;
1090         size_t subbuf_size = buf->chan->subbuf_size;
1091         size_t n_subbufs = buf->chan->n_subbufs;
1092         size_t consumed = buf->subbufs_consumed % n_subbufs;
1093
1094         if (!read_pos)
1095                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1096         read_subbuf = read_pos / subbuf_size;
1097         padding = buf->padding[read_subbuf];
1098         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1099         padding_end = (read_subbuf + 1) * subbuf_size;
1100         if (read_pos >= padding_start && read_pos < padding_end) {
1101                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1102                 read_pos = read_subbuf * subbuf_size;
1103         }
1104
1105         return read_pos;
1106 }
1107
1108 /**
1109  *      relay_file_read_end_pos - return the new read position
1110  *      @read_pos: file read position
1111  *      @buf: relay channel buffer
1112  *      @count: number of bytes to be read
1113  */
1114 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1115                                       size_t read_pos,
1116                                       size_t count)
1117 {
1118         size_t read_subbuf, padding, end_pos;
1119         size_t subbuf_size = buf->chan->subbuf_size;
1120         size_t n_subbufs = buf->chan->n_subbufs;
1121
1122         read_subbuf = read_pos / subbuf_size;
1123         padding = buf->padding[read_subbuf];
1124         if (read_pos % subbuf_size + count + padding == subbuf_size)
1125                 end_pos = (read_subbuf + 1) * subbuf_size;
1126         else
1127                 end_pos = read_pos + count;
1128         if (end_pos >= subbuf_size * n_subbufs)
1129                 end_pos = 0;
1130
1131         return end_pos;
1132 }
1133
1134 /*
1135  *      subbuf_read_actor - read up to one subbuf's worth of data
1136  */
1137 static int subbuf_read_actor(size_t read_start,
1138                              struct rchan_buf *buf,
1139                              size_t avail,
1140                              read_descriptor_t *desc)
1141 {
1142         void *from;
1143         int ret = 0;
1144
1145         from = buf->start + read_start;
1146         ret = avail;
1147         if (copy_to_user(desc->arg.buf, from, avail)) {
1148                 desc->error = -EFAULT;
1149                 ret = 0;
1150         }
1151         desc->arg.data += ret;
1152         desc->written += ret;
1153         desc->count -= ret;
1154
1155         return ret;
1156 }
1157
1158 typedef int (*subbuf_actor_t) (size_t read_start,
1159                                struct rchan_buf *buf,
1160                                size_t avail,
1161                                read_descriptor_t *desc);
1162
1163 /*
1164  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1165  */
1166 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1167                                         subbuf_actor_t subbuf_actor,
1168                                         read_descriptor_t *desc)
1169 {
1170         struct rchan_buf *buf = filp->private_data;
1171         size_t read_start, avail;
1172         int ret;
1173
1174         if (!desc->count)
1175                 return 0;
1176
1177         inode_lock(file_inode(filp));
1178         do {
1179                 if (!relay_file_read_avail(buf, *ppos))
1180                         break;
1181
1182                 read_start = relay_file_read_start_pos(*ppos, buf);
1183                 avail = relay_file_read_subbuf_avail(read_start, buf);
1184                 if (!avail)
1185                         break;
1186
1187                 avail = min(desc->count, avail);
1188                 ret = subbuf_actor(read_start, buf, avail, desc);
1189                 if (desc->error < 0)
1190                         break;
1191
1192                 if (ret) {
1193                         relay_file_read_consume(buf, read_start, ret);
1194                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1195                 }
1196         } while (desc->count && ret);
1197         inode_unlock(file_inode(filp));
1198
1199         return desc->written;
1200 }
1201
1202 static ssize_t relay_file_read(struct file *filp,
1203                                char __user *buffer,
1204                                size_t count,
1205                                loff_t *ppos)
1206 {
1207         read_descriptor_t desc;
1208         desc.written = 0;
1209         desc.count = count;
1210         desc.arg.buf = buffer;
1211         desc.error = 0;
1212         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc);
1213 }
1214
1215 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1216 {
1217         rbuf->bytes_consumed += bytes_consumed;
1218
1219         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1220                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1221                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1222         }
1223 }
1224
1225 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1226                                    struct pipe_buffer *buf)
1227 {
1228         struct rchan_buf *rbuf;
1229
1230         rbuf = (struct rchan_buf *)page_private(buf->page);
1231         relay_consume_bytes(rbuf, buf->private);
1232 }
1233
1234 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1235         .can_merge = 0,
1236         .confirm = generic_pipe_buf_confirm,
1237         .release = relay_pipe_buf_release,
1238         .steal = generic_pipe_buf_steal,
1239         .get = generic_pipe_buf_get,
1240 };
1241
1242 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1243 {
1244 }
1245
1246 /*
1247  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1248  */
1249 static ssize_t subbuf_splice_actor(struct file *in,
1250                                loff_t *ppos,
1251                                struct pipe_inode_info *pipe,
1252                                size_t len,
1253                                unsigned int flags,
1254                                int *nonpad_ret)
1255 {
1256         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1257         struct rchan_buf *rbuf = in->private_data;
1258         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1259         uint64_t pos = (uint64_t) *ppos;
1260         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1261         size_t read_start = (size_t) do_div(pos, alloc_size);
1262         size_t read_subbuf = read_start / subbuf_size;
1263         size_t padding = rbuf->padding[read_subbuf];
1264         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1265         struct page *pages[PIPE_DEF_BUFFERS];
1266         struct partial_page partial[PIPE_DEF_BUFFERS];
1267         struct splice_pipe_desc spd = {
1268                 .pages = pages,
1269                 .nr_pages = 0,
1270                 .nr_pages_max = PIPE_DEF_BUFFERS,
1271                 .partial = partial,
1272                 .flags = flags,
1273                 .ops = &relay_pipe_buf_ops,
1274                 .spd_release = relay_page_release,
1275         };
1276         ssize_t ret;
1277
1278         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1279                 return 0;
1280         if (splice_grow_spd(pipe, &spd))
1281                 return -ENOMEM;
1282
1283         /*
1284          * Adjust read len, if longer than what is available
1285          */
1286         if (len > (subbuf_size - read_start % subbuf_size))
1287                 len = subbuf_size - read_start % subbuf_size;
1288
1289         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1290         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1291         poff = read_start & ~PAGE_MASK;
1292         nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1293
1294         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1295                 unsigned int this_len, this_end, private;
1296                 unsigned int cur_pos = read_start + total_len;
1297
1298                 if (!len)
1299                         break;
1300
1301                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1302                 private = this_len;
1303
1304                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1305                 spd.partial[spd.nr_pages].offset = poff;
1306
1307                 this_end = cur_pos + this_len;
1308                 if (this_end >= nonpad_end) {
1309                         this_len = nonpad_end - cur_pos;
1310                         private = this_len + padding;
1311                 }
1312                 spd.partial[spd.nr_pages].len = this_len;
1313                 spd.partial[spd.nr_pages].private = private;
1314
1315                 len -= this_len;
1316                 total_len += this_len;
1317                 poff = 0;
1318                 pidx = (pidx + 1) % subbuf_pages;
1319
1320                 if (this_end >= nonpad_end) {
1321                         spd.nr_pages++;
1322                         break;
1323                 }
1324         }
1325
1326         ret = 0;
1327         if (!spd.nr_pages)
1328                 goto out;
1329
1330         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1331         if (ret < 0 || ret < total_len)
1332                 goto out;
1333
1334         if (read_start + ret == nonpad_end)
1335                 ret += padding;
1336
1337 out:
1338         splice_shrink_spd(&spd);
1339         return ret;
1340 }
1341
1342 static ssize_t relay_file_splice_read(struct file *in,
1343                                       loff_t *ppos,
1344                                       struct pipe_inode_info *pipe,
1345                                       size_t len,
1346                                       unsigned int flags)
1347 {
1348         ssize_t spliced;
1349         int ret;
1350         int nonpad_ret = 0;
1351
1352         ret = 0;
1353         spliced = 0;
1354
1355         while (len && !spliced) {
1356                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1357                 if (ret < 0)
1358                         break;
1359                 else if (!ret) {
1360                         if (flags & SPLICE_F_NONBLOCK)
1361                                 ret = -EAGAIN;
1362                         break;
1363                 }
1364
1365                 *ppos += ret;
1366                 if (ret > len)
1367                         len = 0;
1368                 else
1369                         len -= ret;
1370                 spliced += nonpad_ret;
1371                 nonpad_ret = 0;
1372         }
1373
1374         if (spliced)
1375                 return spliced;
1376
1377         return ret;
1378 }
1379
1380 const struct file_operations relay_file_operations = {
1381         .open           = relay_file_open,
1382         .poll           = relay_file_poll,
1383         .mmap           = relay_file_mmap,
1384         .read           = relay_file_read,
1385         .llseek         = no_llseek,
1386         .release        = relay_file_release,
1387         .splice_read    = relay_file_splice_read,
1388 };
1389 EXPORT_SYMBOL_GPL(relay_file_operations);
1390
1391 static __init int relay_init(void)
1392 {
1393
1394         hotcpu_notifier(relay_hotcpu_callback, 0);
1395         return 0;
1396 }
1397
1398 early_initcall(relay_init);