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