hugetlbfs: fix mmap failure in unaligned size request
[linux-2.6.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         if (is_vmalloc_addr(array))
85                 vfree(array);
86         else
87                 kfree(array);
88 }
89
90 /**
91  *      relay_mmap_buf: - mmap channel buffer to process address space
92  *      @buf: relay channel buffer
93  *      @vma: vm_area_struct describing memory to be mapped
94  *
95  *      Returns 0 if ok, negative on error
96  *
97  *      Caller should already have grabbed mmap_sem.
98  */
99 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
100 {
101         unsigned long length = vma->vm_end - vma->vm_start;
102         struct file *filp = vma->vm_file;
103
104         if (!buf)
105                 return -EBADF;
106
107         if (length != (unsigned long)buf->chan->alloc_size)
108                 return -EINVAL;
109
110         vma->vm_ops = &relay_file_mmap_ops;
111         vma->vm_flags |= VM_DONTEXPAND;
112         vma->vm_private_data = buf;
113         buf->chan->cb->buf_mapped(buf, filp);
114
115         return 0;
116 }
117
118 /**
119  *      relay_alloc_buf - allocate a channel buffer
120  *      @buf: the buffer struct
121  *      @size: total size of the buffer
122  *
123  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
124  *      passed in size will get page aligned, if it isn't already.
125  */
126 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
127 {
128         void *mem;
129         unsigned int i, j, n_pages;
130
131         *size = PAGE_ALIGN(*size);
132         n_pages = *size >> PAGE_SHIFT;
133
134         buf->page_array = relay_alloc_page_array(n_pages);
135         if (!buf->page_array)
136                 return NULL;
137
138         for (i = 0; i < n_pages; i++) {
139                 buf->page_array[i] = alloc_page(GFP_KERNEL);
140                 if (unlikely(!buf->page_array[i]))
141                         goto depopulate;
142                 set_page_private(buf->page_array[i], (unsigned long)buf);
143         }
144         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
145         if (!mem)
146                 goto depopulate;
147
148         memset(mem, 0, *size);
149         buf->page_count = n_pages;
150         return mem;
151
152 depopulate:
153         for (j = 0; j < i; j++)
154                 __free_page(buf->page_array[j]);
155         relay_free_page_array(buf->page_array);
156         return NULL;
157 }
158
159 /**
160  *      relay_create_buf - allocate and initialize a channel buffer
161  *      @chan: the relay channel
162  *
163  *      Returns channel buffer if successful, %NULL otherwise.
164  */
165 static struct rchan_buf *relay_create_buf(struct rchan *chan)
166 {
167         struct rchan_buf *buf;
168
169         if (chan->n_subbufs > UINT_MAX / sizeof(size_t *))
170                 return NULL;
171
172         buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
173         if (!buf)
174                 return NULL;
175         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
176         if (!buf->padding)
177                 goto free_buf;
178
179         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
180         if (!buf->start)
181                 goto free_buf;
182
183         buf->chan = chan;
184         kref_get(&buf->chan->kref);
185         return buf;
186
187 free_buf:
188         kfree(buf->padding);
189         kfree(buf);
190         return NULL;
191 }
192
193 /**
194  *      relay_destroy_channel - free the channel struct
195  *      @kref: target kernel reference that contains the relay channel
196  *
197  *      Should only be called from kref_put().
198  */
199 static void relay_destroy_channel(struct kref *kref)
200 {
201         struct rchan *chan = container_of(kref, struct rchan, kref);
202         kfree(chan);
203 }
204
205 /**
206  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
207  *      @buf: the buffer struct
208  */
209 static void relay_destroy_buf(struct rchan_buf *buf)
210 {
211         struct rchan *chan = buf->chan;
212         unsigned int i;
213
214         if (likely(buf->start)) {
215                 vunmap(buf->start);
216                 for (i = 0; i < buf->page_count; i++)
217                         __free_page(buf->page_array[i]);
218                 relay_free_page_array(buf->page_array);
219         }
220         chan->buf[buf->cpu] = NULL;
221         kfree(buf->padding);
222         kfree(buf);
223         kref_put(&chan->kref, relay_destroy_channel);
224 }
225
226 /**
227  *      relay_remove_buf - remove a channel buffer
228  *      @kref: target kernel reference that contains the relay buffer
229  *
230  *      Removes the file from the fileystem, which also frees the
231  *      rchan_buf_struct and the channel buffer.  Should only be called from
232  *      kref_put().
233  */
234 static void relay_remove_buf(struct kref *kref)
235 {
236         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
237         buf->chan->cb->remove_buf_file(buf->dentry);
238         relay_destroy_buf(buf);
239 }
240
241 /**
242  *      relay_buf_empty - boolean, is the channel buffer empty?
243  *      @buf: channel buffer
244  *
245  *      Returns 1 if the buffer is empty, 0 otherwise.
246  */
247 static int relay_buf_empty(struct rchan_buf *buf)
248 {
249         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
250 }
251
252 /**
253  *      relay_buf_full - boolean, is the channel buffer full?
254  *      @buf: channel buffer
255  *
256  *      Returns 1 if the buffer is full, 0 otherwise.
257  */
258 int relay_buf_full(struct rchan_buf *buf)
259 {
260         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
261         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
262 }
263 EXPORT_SYMBOL_GPL(relay_buf_full);
264
265 /*
266  * High-level relay kernel API and associated functions.
267  */
268
269 /*
270  * rchan_callback implementations defining default channel behavior.  Used
271  * in place of corresponding NULL values in client callback struct.
272  */
273
274 /*
275  * subbuf_start() default callback.  Does nothing.
276  */
277 static int subbuf_start_default_callback (struct rchan_buf *buf,
278                                           void *subbuf,
279                                           void *prev_subbuf,
280                                           size_t prev_padding)
281 {
282         if (relay_buf_full(buf))
283                 return 0;
284
285         return 1;
286 }
287
288 /*
289  * buf_mapped() default callback.  Does nothing.
290  */
291 static void buf_mapped_default_callback(struct rchan_buf *buf,
292                                         struct file *filp)
293 {
294 }
295
296 /*
297  * buf_unmapped() default callback.  Does nothing.
298  */
299 static void buf_unmapped_default_callback(struct rchan_buf *buf,
300                                           struct file *filp)
301 {
302 }
303
304 /*
305  * create_buf_file_create() default callback.  Does nothing.
306  */
307 static struct dentry *create_buf_file_default_callback(const char *filename,
308                                                        struct dentry *parent,
309                                                        umode_t mode,
310                                                        struct rchan_buf *buf,
311                                                        int *is_global)
312 {
313         return NULL;
314 }
315
316 /*
317  * remove_buf_file() default callback.  Does nothing.
318  */
319 static int remove_buf_file_default_callback(struct dentry *dentry)
320 {
321         return -EINVAL;
322 }
323
324 /* relay channel default callbacks */
325 static struct rchan_callbacks default_channel_callbacks = {
326         .subbuf_start = subbuf_start_default_callback,
327         .buf_mapped = buf_mapped_default_callback,
328         .buf_unmapped = buf_unmapped_default_callback,
329         .create_buf_file = create_buf_file_default_callback,
330         .remove_buf_file = remove_buf_file_default_callback,
331 };
332
333 /**
334  *      wakeup_readers - wake up readers waiting on a channel
335  *      @data: contains the channel buffer
336  *
337  *      This is the timer function used to defer reader waking.
338  */
339 static void wakeup_readers(unsigned long data)
340 {
341         struct rchan_buf *buf = (struct rchan_buf *)data;
342         wake_up_interruptible(&buf->read_wait);
343 }
344
345 /**
346  *      __relay_reset - reset a channel buffer
347  *      @buf: the channel buffer
348  *      @init: 1 if this is a first-time initialization
349  *
350  *      See relay_reset() for description of effect.
351  */
352 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
353 {
354         size_t i;
355
356         if (init) {
357                 init_waitqueue_head(&buf->read_wait);
358                 kref_init(&buf->kref);
359                 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
360         } else
361                 del_timer_sync(&buf->timer);
362
363         buf->subbufs_produced = 0;
364         buf->subbufs_consumed = 0;
365         buf->bytes_consumed = 0;
366         buf->finalized = 0;
367         buf->data = buf->start;
368         buf->offset = 0;
369
370         for (i = 0; i < buf->chan->n_subbufs; i++)
371                 buf->padding[i] = 0;
372
373         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
374 }
375
376 /**
377  *      relay_reset - reset the channel
378  *      @chan: the channel
379  *
380  *      This has the effect of erasing all data from all channel buffers
381  *      and restarting the channel in its initial state.  The buffers
382  *      are not freed, so any mappings are still in effect.
383  *
384  *      NOTE. Care should be taken that the channel isn't actually
385  *      being used by anything when this call is made.
386  */
387 void relay_reset(struct rchan *chan)
388 {
389         unsigned int i;
390
391         if (!chan)
392                 return;
393
394         if (chan->is_global && chan->buf[0]) {
395                 __relay_reset(chan->buf[0], 0);
396                 return;
397         }
398
399         mutex_lock(&relay_channels_mutex);
400         for_each_possible_cpu(i)
401                 if (chan->buf[i])
402                         __relay_reset(chan->buf[i], 0);
403         mutex_unlock(&relay_channels_mutex);
404 }
405 EXPORT_SYMBOL_GPL(relay_reset);
406
407 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
408                                         struct dentry *dentry)
409 {
410         buf->dentry = dentry;
411         buf->dentry->d_inode->i_size = buf->early_bytes;
412 }
413
414 static struct dentry *relay_create_buf_file(struct rchan *chan,
415                                             struct rchan_buf *buf,
416                                             unsigned int cpu)
417 {
418         struct dentry *dentry;
419         char *tmpname;
420
421         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
422         if (!tmpname)
423                 return NULL;
424         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
425
426         /* Create file in fs */
427         dentry = chan->cb->create_buf_file(tmpname, chan->parent,
428                                            S_IRUSR, buf,
429                                            &chan->is_global);
430
431         kfree(tmpname);
432
433         return dentry;
434 }
435
436 /*
437  *      relay_open_buf - create a new relay channel buffer
438  *
439  *      used by relay_open() and CPU hotplug.
440  */
441 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
442 {
443         struct rchan_buf *buf = NULL;
444         struct dentry *dentry;
445
446         if (chan->is_global)
447                 return chan->buf[0];
448
449         buf = relay_create_buf(chan);
450         if (!buf)
451                 return NULL;
452
453         if (chan->has_base_filename) {
454                 dentry = relay_create_buf_file(chan, buf, cpu);
455                 if (!dentry)
456                         goto free_buf;
457                 relay_set_buf_dentry(buf, dentry);
458         }
459
460         buf->cpu = cpu;
461         __relay_reset(buf, 1);
462
463         if(chan->is_global) {
464                 chan->buf[0] = buf;
465                 buf->cpu = 0;
466         }
467
468         return buf;
469
470 free_buf:
471         relay_destroy_buf(buf);
472         return NULL;
473 }
474
475 /**
476  *      relay_close_buf - close a channel buffer
477  *      @buf: channel buffer
478  *
479  *      Marks the buffer finalized and restores the default callbacks.
480  *      The channel buffer and channel buffer data structure are then freed
481  *      automatically when the last reference is given up.
482  */
483 static void relay_close_buf(struct rchan_buf *buf)
484 {
485         buf->finalized = 1;
486         del_timer_sync(&buf->timer);
487         kref_put(&buf->kref, relay_remove_buf);
488 }
489
490 static void setup_callbacks(struct rchan *chan,
491                                    struct rchan_callbacks *cb)
492 {
493         if (!cb) {
494                 chan->cb = &default_channel_callbacks;
495                 return;
496         }
497
498         if (!cb->subbuf_start)
499                 cb->subbuf_start = subbuf_start_default_callback;
500         if (!cb->buf_mapped)
501                 cb->buf_mapped = buf_mapped_default_callback;
502         if (!cb->buf_unmapped)
503                 cb->buf_unmapped = buf_unmapped_default_callback;
504         if (!cb->create_buf_file)
505                 cb->create_buf_file = create_buf_file_default_callback;
506         if (!cb->remove_buf_file)
507                 cb->remove_buf_file = remove_buf_file_default_callback;
508         chan->cb = cb;
509 }
510
511 /**
512  *      relay_hotcpu_callback - CPU hotplug callback
513  *      @nb: notifier block
514  *      @action: hotplug action to take
515  *      @hcpu: CPU number
516  *
517  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
518  */
519 static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
520                                 unsigned long action,
521                                 void *hcpu)
522 {
523         unsigned int hotcpu = (unsigned long)hcpu;
524         struct rchan *chan;
525
526         switch(action) {
527         case CPU_UP_PREPARE:
528         case CPU_UP_PREPARE_FROZEN:
529                 mutex_lock(&relay_channels_mutex);
530                 list_for_each_entry(chan, &relay_channels, list) {
531                         if (chan->buf[hotcpu])
532                                 continue;
533                         chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
534                         if(!chan->buf[hotcpu]) {
535                                 printk(KERN_ERR
536                                         "relay_hotcpu_callback: cpu %d buffer "
537                                         "creation failed\n", hotcpu);
538                                 mutex_unlock(&relay_channels_mutex);
539                                 return notifier_from_errno(-ENOMEM);
540                         }
541                 }
542                 mutex_unlock(&relay_channels_mutex);
543                 break;
544         case CPU_DEAD:
545         case CPU_DEAD_FROZEN:
546                 /* No need to flush the cpu : will be flushed upon
547                  * final relay_flush() call. */
548                 break;
549         }
550         return NOTIFY_OK;
551 }
552
553 /**
554  *      relay_open - create a new relay channel
555  *      @base_filename: base name of files to create, %NULL for buffering only
556  *      @parent: dentry of parent directory, %NULL for root directory or buffer
557  *      @subbuf_size: size of sub-buffers
558  *      @n_subbufs: number of sub-buffers
559  *      @cb: client callback functions
560  *      @private_data: user-defined data
561  *
562  *      Returns channel pointer if successful, %NULL otherwise.
563  *
564  *      Creates a channel buffer for each cpu using the sizes and
565  *      attributes specified.  The created channel buffer files
566  *      will be named base_filename0...base_filenameN-1.  File
567  *      permissions will be %S_IRUSR.
568  */
569 struct rchan *relay_open(const char *base_filename,
570                          struct dentry *parent,
571                          size_t subbuf_size,
572                          size_t n_subbufs,
573                          struct rchan_callbacks *cb,
574                          void *private_data)
575 {
576         unsigned int i;
577         struct rchan *chan;
578
579         if (!(subbuf_size && n_subbufs))
580                 return NULL;
581         if (subbuf_size > UINT_MAX / n_subbufs)
582                 return NULL;
583
584         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
585         if (!chan)
586                 return NULL;
587
588         chan->version = RELAYFS_CHANNEL_VERSION;
589         chan->n_subbufs = n_subbufs;
590         chan->subbuf_size = subbuf_size;
591         chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
592         chan->parent = parent;
593         chan->private_data = private_data;
594         if (base_filename) {
595                 chan->has_base_filename = 1;
596                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
597         }
598         setup_callbacks(chan, cb);
599         kref_init(&chan->kref);
600
601         mutex_lock(&relay_channels_mutex);
602         for_each_online_cpu(i) {
603                 chan->buf[i] = relay_open_buf(chan, i);
604                 if (!chan->buf[i])
605                         goto free_bufs;
606         }
607         list_add(&chan->list, &relay_channels);
608         mutex_unlock(&relay_channels_mutex);
609
610         return chan;
611
612 free_bufs:
613         for_each_possible_cpu(i) {
614                 if (chan->buf[i])
615                         relay_close_buf(chan->buf[i]);
616         }
617
618         kref_put(&chan->kref, relay_destroy_channel);
619         mutex_unlock(&relay_channels_mutex);
620         return NULL;
621 }
622 EXPORT_SYMBOL_GPL(relay_open);
623
624 struct rchan_percpu_buf_dispatcher {
625         struct rchan_buf *buf;
626         struct dentry *dentry;
627 };
628
629 /* Called in atomic context. */
630 static void __relay_set_buf_dentry(void *info)
631 {
632         struct rchan_percpu_buf_dispatcher *p = info;
633
634         relay_set_buf_dentry(p->buf, p->dentry);
635 }
636
637 /**
638  *      relay_late_setup_files - triggers file creation
639  *      @chan: channel to operate on
640  *      @base_filename: base name of files to create
641  *      @parent: dentry of parent directory, %NULL for root directory
642  *
643  *      Returns 0 if successful, non-zero otherwise.
644  *
645  *      Use to setup files for a previously buffer-only channel.
646  *      Useful to do early tracing in kernel, before VFS is up, for example.
647  */
648 int relay_late_setup_files(struct rchan *chan,
649                            const char *base_filename,
650                            struct dentry *parent)
651 {
652         int err = 0;
653         unsigned int i, curr_cpu;
654         unsigned long flags;
655         struct dentry *dentry;
656         struct rchan_percpu_buf_dispatcher disp;
657
658         if (!chan || !base_filename)
659                 return -EINVAL;
660
661         strlcpy(chan->base_filename, base_filename, NAME_MAX);
662
663         mutex_lock(&relay_channels_mutex);
664         /* Is chan already set up? */
665         if (unlikely(chan->has_base_filename)) {
666                 mutex_unlock(&relay_channels_mutex);
667                 return -EEXIST;
668         }
669         chan->has_base_filename = 1;
670         chan->parent = parent;
671         curr_cpu = get_cpu();
672         /*
673          * The CPU hotplug notifier ran before us and created buffers with
674          * no files associated. So it's safe to call relay_setup_buf_file()
675          * on all currently online CPUs.
676          */
677         for_each_online_cpu(i) {
678                 if (unlikely(!chan->buf[i])) {
679                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
680                         err = -EINVAL;
681                         break;
682                 }
683
684                 dentry = relay_create_buf_file(chan, chan->buf[i], i);
685                 if (unlikely(!dentry)) {
686                         err = -EINVAL;
687                         break;
688                 }
689
690                 if (curr_cpu == i) {
691                         local_irq_save(flags);
692                         relay_set_buf_dentry(chan->buf[i], dentry);
693                         local_irq_restore(flags);
694                 } else {
695                         disp.buf = chan->buf[i];
696                         disp.dentry = dentry;
697                         smp_mb();
698                         /* relay_channels_mutex must be held, so wait. */
699                         err = smp_call_function_single(i,
700                                                        __relay_set_buf_dentry,
701                                                        &disp, 1);
702                 }
703                 if (unlikely(err))
704                         break;
705         }
706         put_cpu();
707         mutex_unlock(&relay_channels_mutex);
708
709         return err;
710 }
711
712 /**
713  *      relay_switch_subbuf - switch to a new sub-buffer
714  *      @buf: channel buffer
715  *      @length: size of current event
716  *
717  *      Returns either the length passed in or 0 if full.
718  *
719  *      Performs sub-buffer-switch tasks such as invoking callbacks,
720  *      updating padding counts, waking up readers, etc.
721  */
722 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
723 {
724         void *old, *new;
725         size_t old_subbuf, new_subbuf;
726
727         if (unlikely(length > buf->chan->subbuf_size))
728                 goto toobig;
729
730         if (buf->offset != buf->chan->subbuf_size + 1) {
731                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
732                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
733                 buf->padding[old_subbuf] = buf->prev_padding;
734                 buf->subbufs_produced++;
735                 if (buf->dentry)
736                         buf->dentry->d_inode->i_size +=
737                                 buf->chan->subbuf_size -
738                                 buf->padding[old_subbuf];
739                 else
740                         buf->early_bytes += buf->chan->subbuf_size -
741                                             buf->padding[old_subbuf];
742                 smp_mb();
743                 if (waitqueue_active(&buf->read_wait))
744                         /*
745                          * Calling wake_up_interruptible() from here
746                          * will deadlock if we happen to be logging
747                          * from the scheduler (trying to re-grab
748                          * rq->lock), so defer it.
749                          */
750                         mod_timer(&buf->timer, jiffies + 1);
751         }
752
753         old = buf->data;
754         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
755         new = buf->start + new_subbuf * buf->chan->subbuf_size;
756         buf->offset = 0;
757         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
758                 buf->offset = buf->chan->subbuf_size + 1;
759                 return 0;
760         }
761         buf->data = new;
762         buf->padding[new_subbuf] = 0;
763
764         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
765                 goto toobig;
766
767         return length;
768
769 toobig:
770         buf->chan->last_toobig = length;
771         return 0;
772 }
773 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
774
775 /**
776  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
777  *      @chan: the channel
778  *      @cpu: the cpu associated with the channel buffer to update
779  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
780  *
781  *      Adds to the channel buffer's consumed sub-buffer count.
782  *      subbufs_consumed should be the number of sub-buffers newly consumed,
783  *      not the total consumed.
784  *
785  *      NOTE. Kernel clients don't need to call this function if the channel
786  *      mode is 'overwrite'.
787  */
788 void relay_subbufs_consumed(struct rchan *chan,
789                             unsigned int cpu,
790                             size_t subbufs_consumed)
791 {
792         struct rchan_buf *buf;
793
794         if (!chan)
795                 return;
796
797         if (cpu >= NR_CPUS || !chan->buf[cpu] ||
798                                         subbufs_consumed > chan->n_subbufs)
799                 return;
800
801         buf = chan->buf[cpu];
802         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
803                 buf->subbufs_consumed = buf->subbufs_produced;
804         else
805                 buf->subbufs_consumed += subbufs_consumed;
806 }
807 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
808
809 /**
810  *      relay_close - close the channel
811  *      @chan: the channel
812  *
813  *      Closes all channel buffers and frees the channel.
814  */
815 void relay_close(struct rchan *chan)
816 {
817         unsigned int i;
818
819         if (!chan)
820                 return;
821
822         mutex_lock(&relay_channels_mutex);
823         if (chan->is_global && chan->buf[0])
824                 relay_close_buf(chan->buf[0]);
825         else
826                 for_each_possible_cpu(i)
827                         if (chan->buf[i])
828                                 relay_close_buf(chan->buf[i]);
829
830         if (chan->last_toobig)
831                 printk(KERN_WARNING "relay: one or more items not logged "
832                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
833                        chan->last_toobig, chan->subbuf_size);
834
835         list_del(&chan->list);
836         kref_put(&chan->kref, relay_destroy_channel);
837         mutex_unlock(&relay_channels_mutex);
838 }
839 EXPORT_SYMBOL_GPL(relay_close);
840
841 /**
842  *      relay_flush - close the channel
843  *      @chan: the channel
844  *
845  *      Flushes all channel buffers, i.e. forces buffer switch.
846  */
847 void relay_flush(struct rchan *chan)
848 {
849         unsigned int i;
850
851         if (!chan)
852                 return;
853
854         if (chan->is_global && chan->buf[0]) {
855                 relay_switch_subbuf(chan->buf[0], 0);
856                 return;
857         }
858
859         mutex_lock(&relay_channels_mutex);
860         for_each_possible_cpu(i)
861                 if (chan->buf[i])
862                         relay_switch_subbuf(chan->buf[i], 0);
863         mutex_unlock(&relay_channels_mutex);
864 }
865 EXPORT_SYMBOL_GPL(relay_flush);
866
867 /**
868  *      relay_file_open - open file op for relay files
869  *      @inode: the inode
870  *      @filp: the file
871  *
872  *      Increments the channel buffer refcount.
873  */
874 static int relay_file_open(struct inode *inode, struct file *filp)
875 {
876         struct rchan_buf *buf = inode->i_private;
877         kref_get(&buf->kref);
878         filp->private_data = buf;
879
880         return nonseekable_open(inode, filp);
881 }
882
883 /**
884  *      relay_file_mmap - mmap file op for relay files
885  *      @filp: the file
886  *      @vma: the vma describing what to map
887  *
888  *      Calls upon relay_mmap_buf() to map the file into user space.
889  */
890 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
891 {
892         struct rchan_buf *buf = filp->private_data;
893         return relay_mmap_buf(buf, vma);
894 }
895
896 /**
897  *      relay_file_poll - poll file op for relay files
898  *      @filp: the file
899  *      @wait: poll table
900  *
901  *      Poll implemention.
902  */
903 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
904 {
905         unsigned int mask = 0;
906         struct rchan_buf *buf = filp->private_data;
907
908         if (buf->finalized)
909                 return POLLERR;
910
911         if (filp->f_mode & FMODE_READ) {
912                 poll_wait(filp, &buf->read_wait, wait);
913                 if (!relay_buf_empty(buf))
914                         mask |= POLLIN | POLLRDNORM;
915         }
916
917         return mask;
918 }
919
920 /**
921  *      relay_file_release - release file op for relay files
922  *      @inode: the inode
923  *      @filp: the file
924  *
925  *      Decrements the channel refcount, as the filesystem is
926  *      no longer using it.
927  */
928 static int relay_file_release(struct inode *inode, struct file *filp)
929 {
930         struct rchan_buf *buf = filp->private_data;
931         kref_put(&buf->kref, relay_remove_buf);
932
933         return 0;
934 }
935
936 /*
937  *      relay_file_read_consume - update the consumed count for the buffer
938  */
939 static void relay_file_read_consume(struct rchan_buf *buf,
940                                     size_t read_pos,
941                                     size_t bytes_consumed)
942 {
943         size_t subbuf_size = buf->chan->subbuf_size;
944         size_t n_subbufs = buf->chan->n_subbufs;
945         size_t read_subbuf;
946
947         if (buf->subbufs_produced == buf->subbufs_consumed &&
948             buf->offset == buf->bytes_consumed)
949                 return;
950
951         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
952                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
953                 buf->bytes_consumed = 0;
954         }
955
956         buf->bytes_consumed += bytes_consumed;
957         if (!read_pos)
958                 read_subbuf = buf->subbufs_consumed % n_subbufs;
959         else
960                 read_subbuf = read_pos / buf->chan->subbuf_size;
961         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
962                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
963                     (buf->offset == subbuf_size))
964                         return;
965                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
966                 buf->bytes_consumed = 0;
967         }
968 }
969
970 /*
971  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
972  */
973 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
974 {
975         size_t subbuf_size = buf->chan->subbuf_size;
976         size_t n_subbufs = buf->chan->n_subbufs;
977         size_t produced = buf->subbufs_produced;
978         size_t consumed = buf->subbufs_consumed;
979
980         relay_file_read_consume(buf, read_pos, 0);
981
982         consumed = buf->subbufs_consumed;
983
984         if (unlikely(buf->offset > subbuf_size)) {
985                 if (produced == consumed)
986                         return 0;
987                 return 1;
988         }
989
990         if (unlikely(produced - consumed >= n_subbufs)) {
991                 consumed = produced - n_subbufs + 1;
992                 buf->subbufs_consumed = consumed;
993                 buf->bytes_consumed = 0;
994         }
995
996         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
997         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
998
999         if (consumed > produced)
1000                 produced += n_subbufs * subbuf_size;
1001
1002         if (consumed == produced) {
1003                 if (buf->offset == subbuf_size &&
1004                     buf->subbufs_produced > buf->subbufs_consumed)
1005                         return 1;
1006                 return 0;
1007         }
1008
1009         return 1;
1010 }
1011
1012 /**
1013  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1014  *      @read_pos: file read position
1015  *      @buf: relay channel buffer
1016  */
1017 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1018                                            struct rchan_buf *buf)
1019 {
1020         size_t padding, avail = 0;
1021         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1022         size_t subbuf_size = buf->chan->subbuf_size;
1023
1024         write_subbuf = (buf->data - buf->start) / subbuf_size;
1025         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1026         read_subbuf = read_pos / subbuf_size;
1027         read_offset = read_pos % subbuf_size;
1028         padding = buf->padding[read_subbuf];
1029
1030         if (read_subbuf == write_subbuf) {
1031                 if (read_offset + padding < write_offset)
1032                         avail = write_offset - (read_offset + padding);
1033         } else
1034                 avail = (subbuf_size - padding) - read_offset;
1035
1036         return avail;
1037 }
1038
1039 /**
1040  *      relay_file_read_start_pos - find the first available byte to read
1041  *      @read_pos: file read position
1042  *      @buf: relay channel buffer
1043  *
1044  *      If the @read_pos is in the middle of padding, return the
1045  *      position of the first actually available byte, otherwise
1046  *      return the original value.
1047  */
1048 static size_t relay_file_read_start_pos(size_t read_pos,
1049                                         struct rchan_buf *buf)
1050 {
1051         size_t read_subbuf, padding, padding_start, padding_end;
1052         size_t subbuf_size = buf->chan->subbuf_size;
1053         size_t n_subbufs = buf->chan->n_subbufs;
1054         size_t consumed = buf->subbufs_consumed % n_subbufs;
1055
1056         if (!read_pos)
1057                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1058         read_subbuf = read_pos / subbuf_size;
1059         padding = buf->padding[read_subbuf];
1060         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1061         padding_end = (read_subbuf + 1) * subbuf_size;
1062         if (read_pos >= padding_start && read_pos < padding_end) {
1063                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1064                 read_pos = read_subbuf * subbuf_size;
1065         }
1066
1067         return read_pos;
1068 }
1069
1070 /**
1071  *      relay_file_read_end_pos - return the new read position
1072  *      @read_pos: file read position
1073  *      @buf: relay channel buffer
1074  *      @count: number of bytes to be read
1075  */
1076 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1077                                       size_t read_pos,
1078                                       size_t count)
1079 {
1080         size_t read_subbuf, padding, end_pos;
1081         size_t subbuf_size = buf->chan->subbuf_size;
1082         size_t n_subbufs = buf->chan->n_subbufs;
1083
1084         read_subbuf = read_pos / subbuf_size;
1085         padding = buf->padding[read_subbuf];
1086         if (read_pos % subbuf_size + count + padding == subbuf_size)
1087                 end_pos = (read_subbuf + 1) * subbuf_size;
1088         else
1089                 end_pos = read_pos + count;
1090         if (end_pos >= subbuf_size * n_subbufs)
1091                 end_pos = 0;
1092
1093         return end_pos;
1094 }
1095
1096 /*
1097  *      subbuf_read_actor - read up to one subbuf's worth of data
1098  */
1099 static int subbuf_read_actor(size_t read_start,
1100                              struct rchan_buf *buf,
1101                              size_t avail,
1102                              read_descriptor_t *desc,
1103                              read_actor_t actor)
1104 {
1105         void *from;
1106         int ret = 0;
1107
1108         from = buf->start + read_start;
1109         ret = avail;
1110         if (copy_to_user(desc->arg.buf, from, avail)) {
1111                 desc->error = -EFAULT;
1112                 ret = 0;
1113         }
1114         desc->arg.data += ret;
1115         desc->written += ret;
1116         desc->count -= ret;
1117
1118         return ret;
1119 }
1120
1121 typedef int (*subbuf_actor_t) (size_t read_start,
1122                                struct rchan_buf *buf,
1123                                size_t avail,
1124                                read_descriptor_t *desc,
1125                                read_actor_t actor);
1126
1127 /*
1128  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1129  */
1130 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1131                                         subbuf_actor_t subbuf_actor,
1132                                         read_actor_t actor,
1133                                         read_descriptor_t *desc)
1134 {
1135         struct rchan_buf *buf = filp->private_data;
1136         size_t read_start, avail;
1137         int ret;
1138
1139         if (!desc->count)
1140                 return 0;
1141
1142         mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
1143         do {
1144                 if (!relay_file_read_avail(buf, *ppos))
1145                         break;
1146
1147                 read_start = relay_file_read_start_pos(*ppos, buf);
1148                 avail = relay_file_read_subbuf_avail(read_start, buf);
1149                 if (!avail)
1150                         break;
1151
1152                 avail = min(desc->count, avail);
1153                 ret = subbuf_actor(read_start, buf, avail, desc, actor);
1154                 if (desc->error < 0)
1155                         break;
1156
1157                 if (ret) {
1158                         relay_file_read_consume(buf, read_start, ret);
1159                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1160                 }
1161         } while (desc->count && ret);
1162         mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1163
1164         return desc->written;
1165 }
1166
1167 static ssize_t relay_file_read(struct file *filp,
1168                                char __user *buffer,
1169                                size_t count,
1170                                loff_t *ppos)
1171 {
1172         read_descriptor_t desc;
1173         desc.written = 0;
1174         desc.count = count;
1175         desc.arg.buf = buffer;
1176         desc.error = 0;
1177         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1178                                        NULL, &desc);
1179 }
1180
1181 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1182 {
1183         rbuf->bytes_consumed += bytes_consumed;
1184
1185         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1186                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1187                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1188         }
1189 }
1190
1191 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1192                                    struct pipe_buffer *buf)
1193 {
1194         struct rchan_buf *rbuf;
1195
1196         rbuf = (struct rchan_buf *)page_private(buf->page);
1197         relay_consume_bytes(rbuf, buf->private);
1198 }
1199
1200 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1201         .can_merge = 0,
1202         .map = generic_pipe_buf_map,
1203         .unmap = generic_pipe_buf_unmap,
1204         .confirm = generic_pipe_buf_confirm,
1205         .release = relay_pipe_buf_release,
1206         .steal = generic_pipe_buf_steal,
1207         .get = generic_pipe_buf_get,
1208 };
1209
1210 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1211 {
1212 }
1213
1214 /*
1215  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1216  */
1217 static ssize_t subbuf_splice_actor(struct file *in,
1218                                loff_t *ppos,
1219                                struct pipe_inode_info *pipe,
1220                                size_t len,
1221                                unsigned int flags,
1222                                int *nonpad_ret)
1223 {
1224         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1225         struct rchan_buf *rbuf = in->private_data;
1226         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1227         uint64_t pos = (uint64_t) *ppos;
1228         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1229         size_t read_start = (size_t) do_div(pos, alloc_size);
1230         size_t read_subbuf = read_start / subbuf_size;
1231         size_t padding = rbuf->padding[read_subbuf];
1232         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1233         struct page *pages[PIPE_DEF_BUFFERS];
1234         struct partial_page partial[PIPE_DEF_BUFFERS];
1235         struct splice_pipe_desc spd = {
1236                 .pages = pages,
1237                 .nr_pages = 0,
1238                 .nr_pages_max = PIPE_DEF_BUFFERS,
1239                 .partial = partial,
1240                 .flags = flags,
1241                 .ops = &relay_pipe_buf_ops,
1242                 .spd_release = relay_page_release,
1243         };
1244         ssize_t ret;
1245
1246         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1247                 return 0;
1248         if (splice_grow_spd(pipe, &spd))
1249                 return -ENOMEM;
1250
1251         /*
1252          * Adjust read len, if longer than what is available
1253          */
1254         if (len > (subbuf_size - read_start % subbuf_size))
1255                 len = subbuf_size - read_start % subbuf_size;
1256
1257         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1258         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1259         poff = read_start & ~PAGE_MASK;
1260         nr_pages = min_t(unsigned int, subbuf_pages, pipe->buffers);
1261
1262         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1263                 unsigned int this_len, this_end, private;
1264                 unsigned int cur_pos = read_start + total_len;
1265
1266                 if (!len)
1267                         break;
1268
1269                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1270                 private = this_len;
1271
1272                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1273                 spd.partial[spd.nr_pages].offset = poff;
1274
1275                 this_end = cur_pos + this_len;
1276                 if (this_end >= nonpad_end) {
1277                         this_len = nonpad_end - cur_pos;
1278                         private = this_len + padding;
1279                 }
1280                 spd.partial[spd.nr_pages].len = this_len;
1281                 spd.partial[spd.nr_pages].private = private;
1282
1283                 len -= this_len;
1284                 total_len += this_len;
1285                 poff = 0;
1286                 pidx = (pidx + 1) % subbuf_pages;
1287
1288                 if (this_end >= nonpad_end) {
1289                         spd.nr_pages++;
1290                         break;
1291                 }
1292         }
1293
1294         ret = 0;
1295         if (!spd.nr_pages)
1296                 goto out;
1297
1298         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1299         if (ret < 0 || ret < total_len)
1300                 goto out;
1301
1302         if (read_start + ret == nonpad_end)
1303                 ret += padding;
1304
1305 out:
1306         splice_shrink_spd(&spd);
1307         return ret;
1308 }
1309
1310 static ssize_t relay_file_splice_read(struct file *in,
1311                                       loff_t *ppos,
1312                                       struct pipe_inode_info *pipe,
1313                                       size_t len,
1314                                       unsigned int flags)
1315 {
1316         ssize_t spliced;
1317         int ret;
1318         int nonpad_ret = 0;
1319
1320         ret = 0;
1321         spliced = 0;
1322
1323         while (len && !spliced) {
1324                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1325                 if (ret < 0)
1326                         break;
1327                 else if (!ret) {
1328                         if (flags & SPLICE_F_NONBLOCK)
1329                                 ret = -EAGAIN;
1330                         break;
1331                 }
1332
1333                 *ppos += ret;
1334                 if (ret > len)
1335                         len = 0;
1336                 else
1337                         len -= ret;
1338                 spliced += nonpad_ret;
1339                 nonpad_ret = 0;
1340         }
1341
1342         if (spliced)
1343                 return spliced;
1344
1345         return ret;
1346 }
1347
1348 const struct file_operations relay_file_operations = {
1349         .open           = relay_file_open,
1350         .poll           = relay_file_poll,
1351         .mmap           = relay_file_mmap,
1352         .read           = relay_file_read,
1353         .llseek         = no_llseek,
1354         .release        = relay_file_release,
1355         .splice_read    = relay_file_splice_read,
1356 };
1357 EXPORT_SYMBOL_GPL(relay_file_operations);
1358
1359 static __init int relay_init(void)
1360 {
1361
1362         hotcpu_notifier(relay_hotcpu_callback, 0);
1363         return 0;
1364 }
1365
1366 early_initcall(relay_init);