uprobe: collapse THP pmd after removing all uprobes
[platform/kernel/linux-rpi.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 vm_fault_t relay_buf_fault(struct vm_fault *vmf)
43 {
44         struct page *page;
45         struct rchan_buf *buf = vmf->vma->vm_private_data;
46         pgoff_t pgoff = vmf->pgoff;
47
48         if (!buf)
49                 return VM_FAULT_OOM;
50
51         page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
52         if (!page)
53                 return VM_FAULT_SIGBUS;
54         get_page(page);
55         vmf->page = page;
56
57         return 0;
58 }
59
60 /*
61  * vm_ops for relay file mappings.
62  */
63 static const struct vm_operations_struct relay_file_mmap_ops = {
64         .fault = relay_buf_fault,
65         .close = relay_file_mmap_close,
66 };
67
68 /*
69  * allocate an array of pointers of struct page
70  */
71 static struct page **relay_alloc_page_array(unsigned int n_pages)
72 {
73         const size_t pa_size = n_pages * sizeof(struct page *);
74         if (pa_size > PAGE_SIZE)
75                 return vzalloc(pa_size);
76         return kzalloc(pa_size, GFP_KERNEL);
77 }
78
79 /*
80  * free an array of pointers of struct page
81  */
82 static void relay_free_page_array(struct page **array)
83 {
84         kvfree(array);
85 }
86
87 /**
88  *      relay_mmap_buf: - mmap channel buffer to process address space
89  *      @buf: relay channel buffer
90  *      @vma: vm_area_struct describing memory to be mapped
91  *
92  *      Returns 0 if ok, negative on error
93  *
94  *      Caller should already have grabbed mmap_sem.
95  */
96 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
97 {
98         unsigned long length = vma->vm_end - vma->vm_start;
99         struct file *filp = vma->vm_file;
100
101         if (!buf)
102                 return -EBADF;
103
104         if (length != (unsigned long)buf->chan->alloc_size)
105                 return -EINVAL;
106
107         vma->vm_ops = &relay_file_mmap_ops;
108         vma->vm_flags |= VM_DONTEXPAND;
109         vma->vm_private_data = buf;
110         buf->chan->cb->buf_mapped(buf, filp);
111
112         return 0;
113 }
114
115 /**
116  *      relay_alloc_buf - allocate a channel buffer
117  *      @buf: the buffer struct
118  *      @size: total size of the buffer
119  *
120  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
121  *      passed in size will get page aligned, if it isn't already.
122  */
123 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
124 {
125         void *mem;
126         unsigned int i, j, n_pages;
127
128         *size = PAGE_ALIGN(*size);
129         n_pages = *size >> PAGE_SHIFT;
130
131         buf->page_array = relay_alloc_page_array(n_pages);
132         if (!buf->page_array)
133                 return NULL;
134
135         for (i = 0; i < n_pages; i++) {
136                 buf->page_array[i] = alloc_page(GFP_KERNEL);
137                 if (unlikely(!buf->page_array[i]))
138                         goto depopulate;
139                 set_page_private(buf->page_array[i], (unsigned long)buf);
140         }
141         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
142         if (!mem)
143                 goto depopulate;
144
145         memset(mem, 0, *size);
146         buf->page_count = n_pages;
147         return mem;
148
149 depopulate:
150         for (j = 0; j < i; j++)
151                 __free_page(buf->page_array[j]);
152         relay_free_page_array(buf->page_array);
153         return NULL;
154 }
155
156 /**
157  *      relay_create_buf - allocate and initialize a channel buffer
158  *      @chan: the relay channel
159  *
160  *      Returns channel buffer if successful, %NULL otherwise.
161  */
162 static struct rchan_buf *relay_create_buf(struct rchan *chan)
163 {
164         struct rchan_buf *buf;
165
166         if (chan->n_subbufs > KMALLOC_MAX_SIZE / sizeof(size_t *))
167                 return NULL;
168
169         buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
170         if (!buf)
171                 return NULL;
172         buf->padding = kmalloc_array(chan->n_subbufs, sizeof(size_t *),
173                                      GFP_KERNEL);
174         if (!buf->padding)
175                 goto free_buf;
176
177         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
178         if (!buf->start)
179                 goto free_buf;
180
181         buf->chan = chan;
182         kref_get(&buf->chan->kref);
183         return buf;
184
185 free_buf:
186         kfree(buf->padding);
187         kfree(buf);
188         return NULL;
189 }
190
191 /**
192  *      relay_destroy_channel - free the channel struct
193  *      @kref: target kernel reference that contains the relay channel
194  *
195  *      Should only be called from kref_put().
196  */
197 static void relay_destroy_channel(struct kref *kref)
198 {
199         struct rchan *chan = container_of(kref, struct rchan, kref);
200         kfree(chan);
201 }
202
203 /**
204  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
205  *      @buf: the buffer struct
206  */
207 static void relay_destroy_buf(struct rchan_buf *buf)
208 {
209         struct rchan *chan = buf->chan;
210         unsigned int i;
211
212         if (likely(buf->start)) {
213                 vunmap(buf->start);
214                 for (i = 0; i < buf->page_count; i++)
215                         __free_page(buf->page_array[i]);
216                 relay_free_page_array(buf->page_array);
217         }
218         *per_cpu_ptr(chan->buf, buf->cpu) = NULL;
219         kfree(buf->padding);
220         kfree(buf);
221         kref_put(&chan->kref, relay_destroy_channel);
222 }
223
224 /**
225  *      relay_remove_buf - remove a channel buffer
226  *      @kref: target kernel reference that contains the relay buffer
227  *
228  *      Removes the file from the filesystem, which also frees the
229  *      rchan_buf_struct and the channel buffer.  Should only be called from
230  *      kref_put().
231  */
232 static void relay_remove_buf(struct kref *kref)
233 {
234         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
235         relay_destroy_buf(buf);
236 }
237
238 /**
239  *      relay_buf_empty - boolean, is the channel buffer empty?
240  *      @buf: channel buffer
241  *
242  *      Returns 1 if the buffer is empty, 0 otherwise.
243  */
244 static int relay_buf_empty(struct rchan_buf *buf)
245 {
246         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
247 }
248
249 /**
250  *      relay_buf_full - boolean, is the channel buffer full?
251  *      @buf: channel buffer
252  *
253  *      Returns 1 if the buffer is full, 0 otherwise.
254  */
255 int relay_buf_full(struct rchan_buf *buf)
256 {
257         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
258         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
259 }
260 EXPORT_SYMBOL_GPL(relay_buf_full);
261
262 /*
263  * High-level relay kernel API and associated functions.
264  */
265
266 /*
267  * rchan_callback implementations defining default channel behavior.  Used
268  * in place of corresponding NULL values in client callback struct.
269  */
270
271 /*
272  * subbuf_start() default callback.  Does nothing.
273  */
274 static int subbuf_start_default_callback (struct rchan_buf *buf,
275                                           void *subbuf,
276                                           void *prev_subbuf,
277                                           size_t prev_padding)
278 {
279         if (relay_buf_full(buf))
280                 return 0;
281
282         return 1;
283 }
284
285 /*
286  * buf_mapped() default callback.  Does nothing.
287  */
288 static void buf_mapped_default_callback(struct rchan_buf *buf,
289                                         struct file *filp)
290 {
291 }
292
293 /*
294  * buf_unmapped() default callback.  Does nothing.
295  */
296 static void buf_unmapped_default_callback(struct rchan_buf *buf,
297                                           struct file *filp)
298 {
299 }
300
301 /*
302  * create_buf_file_create() default callback.  Does nothing.
303  */
304 static struct dentry *create_buf_file_default_callback(const char *filename,
305                                                        struct dentry *parent,
306                                                        umode_t mode,
307                                                        struct rchan_buf *buf,
308                                                        int *is_global)
309 {
310         return NULL;
311 }
312
313 /*
314  * remove_buf_file() default callback.  Does nothing.
315  */
316 static int remove_buf_file_default_callback(struct dentry *dentry)
317 {
318         return -EINVAL;
319 }
320
321 /* relay channel default callbacks */
322 static struct rchan_callbacks default_channel_callbacks = {
323         .subbuf_start = subbuf_start_default_callback,
324         .buf_mapped = buf_mapped_default_callback,
325         .buf_unmapped = buf_unmapped_default_callback,
326         .create_buf_file = create_buf_file_default_callback,
327         .remove_buf_file = remove_buf_file_default_callback,
328 };
329
330 /**
331  *      wakeup_readers - wake up readers waiting on a channel
332  *      @work: contains the channel buffer
333  *
334  *      This is the function used to defer reader waking
335  */
336 static void wakeup_readers(struct irq_work *work)
337 {
338         struct rchan_buf *buf;
339
340         buf = container_of(work, struct rchan_buf, wakeup_work);
341         wake_up_interruptible(&buf->read_wait);
342 }
343
344 /**
345  *      __relay_reset - reset a channel buffer
346  *      @buf: the channel buffer
347  *      @init: 1 if this is a first-time initialization
348  *
349  *      See relay_reset() for description of effect.
350  */
351 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
352 {
353         size_t i;
354
355         if (init) {
356                 init_waitqueue_head(&buf->read_wait);
357                 kref_init(&buf->kref);
358                 init_irq_work(&buf->wakeup_work, wakeup_readers);
359         } else {
360                 irq_work_sync(&buf->wakeup_work);
361         }
362
363         buf->subbufs_produced = 0;
364         buf->subbufs_consumed = 0;
365         buf->bytes_consumed = 0;
366         buf->finalized = 0;
367         buf->data = buf->start;
368         buf->offset = 0;
369
370         for (i = 0; i < buf->chan->n_subbufs; i++)
371                 buf->padding[i] = 0;
372
373         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
374 }
375
376 /**
377  *      relay_reset - reset the channel
378  *      @chan: the channel
379  *
380  *      This has the effect of erasing all data from all channel buffers
381  *      and restarting the channel in its initial state.  The buffers
382  *      are not freed, so any mappings are still in effect.
383  *
384  *      NOTE. Care should be taken that the channel isn't actually
385  *      being used by anything when this call is made.
386  */
387 void relay_reset(struct rchan *chan)
388 {
389         struct rchan_buf *buf;
390         unsigned int i;
391
392         if (!chan)
393                 return;
394
395         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
396                 __relay_reset(buf, 0);
397                 return;
398         }
399
400         mutex_lock(&relay_channels_mutex);
401         for_each_possible_cpu(i)
402                 if ((buf = *per_cpu_ptr(chan->buf, i)))
403                         __relay_reset(buf, 0);
404         mutex_unlock(&relay_channels_mutex);
405 }
406 EXPORT_SYMBOL_GPL(relay_reset);
407
408 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
409                                         struct dentry *dentry)
410 {
411         buf->dentry = dentry;
412         d_inode(buf->dentry)->i_size = buf->early_bytes;
413 }
414
415 static struct dentry *relay_create_buf_file(struct rchan *chan,
416                                             struct rchan_buf *buf,
417                                             unsigned int cpu)
418 {
419         struct dentry *dentry;
420         char *tmpname;
421
422         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
423         if (!tmpname)
424                 return NULL;
425         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
426
427         /* Create file in fs */
428         dentry = chan->cb->create_buf_file(tmpname, chan->parent,
429                                            S_IRUSR, buf,
430                                            &chan->is_global);
431         if (IS_ERR(dentry))
432                 dentry = NULL;
433
434         kfree(tmpname);
435
436         return dentry;
437 }
438
439 /*
440  *      relay_open_buf - create a new relay channel buffer
441  *
442  *      used by relay_open() and CPU hotplug.
443  */
444 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
445 {
446         struct rchan_buf *buf = NULL;
447         struct dentry *dentry;
448
449         if (chan->is_global)
450                 return *per_cpu_ptr(chan->buf, 0);
451
452         buf = relay_create_buf(chan);
453         if (!buf)
454                 return NULL;
455
456         if (chan->has_base_filename) {
457                 dentry = relay_create_buf_file(chan, buf, cpu);
458                 if (!dentry)
459                         goto free_buf;
460                 relay_set_buf_dentry(buf, dentry);
461         } else {
462                 /* Only retrieve global info, nothing more, nothing less */
463                 dentry = chan->cb->create_buf_file(NULL, NULL,
464                                                    S_IRUSR, buf,
465                                                    &chan->is_global);
466                 if (IS_ERR_OR_NULL(dentry))
467                         goto free_buf;
468         }
469
470         buf->cpu = cpu;
471         __relay_reset(buf, 1);
472
473         if(chan->is_global) {
474                 *per_cpu_ptr(chan->buf, 0) = buf;
475                 buf->cpu = 0;
476         }
477
478         return buf;
479
480 free_buf:
481         relay_destroy_buf(buf);
482         return NULL;
483 }
484
485 /**
486  *      relay_close_buf - close a channel buffer
487  *      @buf: channel buffer
488  *
489  *      Marks the buffer finalized and restores the default callbacks.
490  *      The channel buffer and channel buffer data structure are then freed
491  *      automatically when the last reference is given up.
492  */
493 static void relay_close_buf(struct rchan_buf *buf)
494 {
495         buf->finalized = 1;
496         irq_work_sync(&buf->wakeup_work);
497         buf->chan->cb->remove_buf_file(buf->dentry);
498         kref_put(&buf->kref, relay_remove_buf);
499 }
500
501 static void setup_callbacks(struct rchan *chan,
502                                    struct rchan_callbacks *cb)
503 {
504         if (!cb) {
505                 chan->cb = &default_channel_callbacks;
506                 return;
507         }
508
509         if (!cb->subbuf_start)
510                 cb->subbuf_start = subbuf_start_default_callback;
511         if (!cb->buf_mapped)
512                 cb->buf_mapped = buf_mapped_default_callback;
513         if (!cb->buf_unmapped)
514                 cb->buf_unmapped = buf_unmapped_default_callback;
515         if (!cb->create_buf_file)
516                 cb->create_buf_file = create_buf_file_default_callback;
517         if (!cb->remove_buf_file)
518                 cb->remove_buf_file = remove_buf_file_default_callback;
519         chan->cb = cb;
520 }
521
522 int relay_prepare_cpu(unsigned int cpu)
523 {
524         struct rchan *chan;
525         struct rchan_buf *buf;
526
527         mutex_lock(&relay_channels_mutex);
528         list_for_each_entry(chan, &relay_channels, list) {
529                 if ((buf = *per_cpu_ptr(chan->buf, cpu)))
530                         continue;
531                 buf = relay_open_buf(chan, cpu);
532                 if (!buf) {
533                         pr_err("relay: cpu %d buffer creation failed\n", cpu);
534                         mutex_unlock(&relay_channels_mutex);
535                         return -ENOMEM;
536                 }
537                 *per_cpu_ptr(chan->buf, cpu) = buf;
538         }
539         mutex_unlock(&relay_channels_mutex);
540         return 0;
541 }
542
543 /**
544  *      relay_open - create a new relay channel
545  *      @base_filename: base name of files to create, %NULL for buffering only
546  *      @parent: dentry of parent directory, %NULL for root directory or buffer
547  *      @subbuf_size: size of sub-buffers
548  *      @n_subbufs: number of sub-buffers
549  *      @cb: client callback functions
550  *      @private_data: user-defined data
551  *
552  *      Returns channel pointer if successful, %NULL otherwise.
553  *
554  *      Creates a channel buffer for each cpu using the sizes and
555  *      attributes specified.  The created channel buffer files
556  *      will be named base_filename0...base_filenameN-1.  File
557  *      permissions will be %S_IRUSR.
558  *
559  *      If opening a buffer (@parent = NULL) that you later wish to register
560  *      in a filesystem, call relay_late_setup_files() once the @parent dentry
561  *      is available.
562  */
563 struct rchan *relay_open(const char *base_filename,
564                          struct dentry *parent,
565                          size_t subbuf_size,
566                          size_t n_subbufs,
567                          struct rchan_callbacks *cb,
568                          void *private_data)
569 {
570         unsigned int i;
571         struct rchan *chan;
572         struct rchan_buf *buf;
573
574         if (!(subbuf_size && n_subbufs))
575                 return NULL;
576         if (subbuf_size > UINT_MAX / n_subbufs)
577                 return NULL;
578
579         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
580         if (!chan)
581                 return NULL;
582
583         chan->buf = alloc_percpu(struct rchan_buf *);
584         chan->version = RELAYFS_CHANNEL_VERSION;
585         chan->n_subbufs = n_subbufs;
586         chan->subbuf_size = subbuf_size;
587         chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
588         chan->parent = parent;
589         chan->private_data = private_data;
590         if (base_filename) {
591                 chan->has_base_filename = 1;
592                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
593         }
594         setup_callbacks(chan, cb);
595         kref_init(&chan->kref);
596
597         mutex_lock(&relay_channels_mutex);
598         for_each_online_cpu(i) {
599                 buf = relay_open_buf(chan, i);
600                 if (!buf)
601                         goto free_bufs;
602                 *per_cpu_ptr(chan->buf, i) = buf;
603         }
604         list_add(&chan->list, &relay_channels);
605         mutex_unlock(&relay_channels_mutex);
606
607         return chan;
608
609 free_bufs:
610         for_each_possible_cpu(i) {
611                 if ((buf = *per_cpu_ptr(chan->buf, i)))
612                         relay_close_buf(buf);
613         }
614
615         kref_put(&chan->kref, relay_destroy_channel);
616         mutex_unlock(&relay_channels_mutex);
617         return NULL;
618 }
619 EXPORT_SYMBOL_GPL(relay_open);
620
621 struct rchan_percpu_buf_dispatcher {
622         struct rchan_buf *buf;
623         struct dentry *dentry;
624 };
625
626 /* Called in atomic context. */
627 static void __relay_set_buf_dentry(void *info)
628 {
629         struct rchan_percpu_buf_dispatcher *p = info;
630
631         relay_set_buf_dentry(p->buf, p->dentry);
632 }
633
634 /**
635  *      relay_late_setup_files - triggers file creation
636  *      @chan: channel to operate on
637  *      @base_filename: base name of files to create
638  *      @parent: dentry of parent directory, %NULL for root directory
639  *
640  *      Returns 0 if successful, non-zero otherwise.
641  *
642  *      Use to setup files for a previously buffer-only channel created
643  *      by relay_open() with a NULL parent dentry.
644  *
645  *      For example, this is useful for perfomring early tracing in kernel,
646  *      before VFS is up and then exposing the early results once the dentry
647  *      is available.
648  */
649 int relay_late_setup_files(struct rchan *chan,
650                            const char *base_filename,
651                            struct dentry *parent)
652 {
653         int err = 0;
654         unsigned int i, curr_cpu;
655         unsigned long flags;
656         struct dentry *dentry;
657         struct rchan_buf *buf;
658         struct rchan_percpu_buf_dispatcher disp;
659
660         if (!chan || !base_filename)
661                 return -EINVAL;
662
663         strlcpy(chan->base_filename, base_filename, NAME_MAX);
664
665         mutex_lock(&relay_channels_mutex);
666         /* Is chan already set up? */
667         if (unlikely(chan->has_base_filename)) {
668                 mutex_unlock(&relay_channels_mutex);
669                 return -EEXIST;
670         }
671         chan->has_base_filename = 1;
672         chan->parent = parent;
673
674         if (chan->is_global) {
675                 err = -EINVAL;
676                 buf = *per_cpu_ptr(chan->buf, 0);
677                 if (!WARN_ON_ONCE(!buf)) {
678                         dentry = relay_create_buf_file(chan, buf, 0);
679                         if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
680                                 relay_set_buf_dentry(buf, dentry);
681                                 err = 0;
682                         }
683                 }
684                 mutex_unlock(&relay_channels_mutex);
685                 return err;
686         }
687
688         curr_cpu = get_cpu();
689         /*
690          * The CPU hotplug notifier ran before us and created buffers with
691          * no files associated. So it's safe to call relay_setup_buf_file()
692          * on all currently online CPUs.
693          */
694         for_each_online_cpu(i) {
695                 buf = *per_cpu_ptr(chan->buf, i);
696                 if (unlikely(!buf)) {
697                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
698                         err = -EINVAL;
699                         break;
700                 }
701
702                 dentry = relay_create_buf_file(chan, buf, i);
703                 if (unlikely(!dentry)) {
704                         err = -EINVAL;
705                         break;
706                 }
707
708                 if (curr_cpu == i) {
709                         local_irq_save(flags);
710                         relay_set_buf_dentry(buf, dentry);
711                         local_irq_restore(flags);
712                 } else {
713                         disp.buf = buf;
714                         disp.dentry = dentry;
715                         smp_mb();
716                         /* relay_channels_mutex must be held, so wait. */
717                         err = smp_call_function_single(i,
718                                                        __relay_set_buf_dentry,
719                                                        &disp, 1);
720                 }
721                 if (unlikely(err))
722                         break;
723         }
724         put_cpu();
725         mutex_unlock(&relay_channels_mutex);
726
727         return err;
728 }
729 EXPORT_SYMBOL_GPL(relay_late_setup_files);
730
731 /**
732  *      relay_switch_subbuf - switch to a new sub-buffer
733  *      @buf: channel buffer
734  *      @length: size of current event
735  *
736  *      Returns either the length passed in or 0 if full.
737  *
738  *      Performs sub-buffer-switch tasks such as invoking callbacks,
739  *      updating padding counts, waking up readers, etc.
740  */
741 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
742 {
743         void *old, *new;
744         size_t old_subbuf, new_subbuf;
745
746         if (unlikely(length > buf->chan->subbuf_size))
747                 goto toobig;
748
749         if (buf->offset != buf->chan->subbuf_size + 1) {
750                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
751                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
752                 buf->padding[old_subbuf] = buf->prev_padding;
753                 buf->subbufs_produced++;
754                 if (buf->dentry)
755                         d_inode(buf->dentry)->i_size +=
756                                 buf->chan->subbuf_size -
757                                 buf->padding[old_subbuf];
758                 else
759                         buf->early_bytes += buf->chan->subbuf_size -
760                                             buf->padding[old_subbuf];
761                 smp_mb();
762                 if (waitqueue_active(&buf->read_wait)) {
763                         /*
764                          * Calling wake_up_interruptible() from here
765                          * will deadlock if we happen to be logging
766                          * from the scheduler (trying to re-grab
767                          * rq->lock), so defer it.
768                          */
769                         irq_work_queue(&buf->wakeup_work);
770                 }
771         }
772
773         old = buf->data;
774         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
775         new = buf->start + new_subbuf * buf->chan->subbuf_size;
776         buf->offset = 0;
777         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
778                 buf->offset = buf->chan->subbuf_size + 1;
779                 return 0;
780         }
781         buf->data = new;
782         buf->padding[new_subbuf] = 0;
783
784         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
785                 goto toobig;
786
787         return length;
788
789 toobig:
790         buf->chan->last_toobig = length;
791         return 0;
792 }
793 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
794
795 /**
796  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
797  *      @chan: the channel
798  *      @cpu: the cpu associated with the channel buffer to update
799  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
800  *
801  *      Adds to the channel buffer's consumed sub-buffer count.
802  *      subbufs_consumed should be the number of sub-buffers newly consumed,
803  *      not the total consumed.
804  *
805  *      NOTE. Kernel clients don't need to call this function if the channel
806  *      mode is 'overwrite'.
807  */
808 void relay_subbufs_consumed(struct rchan *chan,
809                             unsigned int cpu,
810                             size_t subbufs_consumed)
811 {
812         struct rchan_buf *buf;
813
814         if (!chan || cpu >= NR_CPUS)
815                 return;
816
817         buf = *per_cpu_ptr(chan->buf, cpu);
818         if (!buf || subbufs_consumed > chan->n_subbufs)
819                 return;
820
821         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
822                 buf->subbufs_consumed = buf->subbufs_produced;
823         else
824                 buf->subbufs_consumed += subbufs_consumed;
825 }
826 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
827
828 /**
829  *      relay_close - close the channel
830  *      @chan: the channel
831  *
832  *      Closes all channel buffers and frees the channel.
833  */
834 void relay_close(struct rchan *chan)
835 {
836         struct rchan_buf *buf;
837         unsigned int i;
838
839         if (!chan)
840                 return;
841
842         mutex_lock(&relay_channels_mutex);
843         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0)))
844                 relay_close_buf(buf);
845         else
846                 for_each_possible_cpu(i)
847                         if ((buf = *per_cpu_ptr(chan->buf, i)))
848                                 relay_close_buf(buf);
849
850         if (chan->last_toobig)
851                 printk(KERN_WARNING "relay: one or more items not logged "
852                        "[item size (%zd) > sub-buffer size (%zd)]\n",
853                        chan->last_toobig, chan->subbuf_size);
854
855         list_del(&chan->list);
856         kref_put(&chan->kref, relay_destroy_channel);
857         mutex_unlock(&relay_channels_mutex);
858 }
859 EXPORT_SYMBOL_GPL(relay_close);
860
861 /**
862  *      relay_flush - close the channel
863  *      @chan: the channel
864  *
865  *      Flushes all channel buffers, i.e. forces buffer switch.
866  */
867 void relay_flush(struct rchan *chan)
868 {
869         struct rchan_buf *buf;
870         unsigned int i;
871
872         if (!chan)
873                 return;
874
875         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
876                 relay_switch_subbuf(buf, 0);
877                 return;
878         }
879
880         mutex_lock(&relay_channels_mutex);
881         for_each_possible_cpu(i)
882                 if ((buf = *per_cpu_ptr(chan->buf, i)))
883                         relay_switch_subbuf(buf, 0);
884         mutex_unlock(&relay_channels_mutex);
885 }
886 EXPORT_SYMBOL_GPL(relay_flush);
887
888 /**
889  *      relay_file_open - open file op for relay files
890  *      @inode: the inode
891  *      @filp: the file
892  *
893  *      Increments the channel buffer refcount.
894  */
895 static int relay_file_open(struct inode *inode, struct file *filp)
896 {
897         struct rchan_buf *buf = inode->i_private;
898         kref_get(&buf->kref);
899         filp->private_data = buf;
900
901         return nonseekable_open(inode, filp);
902 }
903
904 /**
905  *      relay_file_mmap - mmap file op for relay files
906  *      @filp: the file
907  *      @vma: the vma describing what to map
908  *
909  *      Calls upon relay_mmap_buf() to map the file into user space.
910  */
911 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
912 {
913         struct rchan_buf *buf = filp->private_data;
914         return relay_mmap_buf(buf, vma);
915 }
916
917 /**
918  *      relay_file_poll - poll file op for relay files
919  *      @filp: the file
920  *      @wait: poll table
921  *
922  *      Poll implemention.
923  */
924 static __poll_t relay_file_poll(struct file *filp, poll_table *wait)
925 {
926         __poll_t mask = 0;
927         struct rchan_buf *buf = filp->private_data;
928
929         if (buf->finalized)
930                 return EPOLLERR;
931
932         if (filp->f_mode & FMODE_READ) {
933                 poll_wait(filp, &buf->read_wait, wait);
934                 if (!relay_buf_empty(buf))
935                         mask |= EPOLLIN | EPOLLRDNORM;
936         }
937
938         return mask;
939 }
940
941 /**
942  *      relay_file_release - release file op for relay files
943  *      @inode: the inode
944  *      @filp: the file
945  *
946  *      Decrements the channel refcount, as the filesystem is
947  *      no longer using it.
948  */
949 static int relay_file_release(struct inode *inode, struct file *filp)
950 {
951         struct rchan_buf *buf = filp->private_data;
952         kref_put(&buf->kref, relay_remove_buf);
953
954         return 0;
955 }
956
957 /*
958  *      relay_file_read_consume - update the consumed count for the buffer
959  */
960 static void relay_file_read_consume(struct rchan_buf *buf,
961                                     size_t read_pos,
962                                     size_t bytes_consumed)
963 {
964         size_t subbuf_size = buf->chan->subbuf_size;
965         size_t n_subbufs = buf->chan->n_subbufs;
966         size_t read_subbuf;
967
968         if (buf->subbufs_produced == buf->subbufs_consumed &&
969             buf->offset == buf->bytes_consumed)
970                 return;
971
972         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
973                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
974                 buf->bytes_consumed = 0;
975         }
976
977         buf->bytes_consumed += bytes_consumed;
978         if (!read_pos)
979                 read_subbuf = buf->subbufs_consumed % n_subbufs;
980         else
981                 read_subbuf = read_pos / buf->chan->subbuf_size;
982         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
983                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
984                     (buf->offset == subbuf_size))
985                         return;
986                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
987                 buf->bytes_consumed = 0;
988         }
989 }
990
991 /*
992  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
993  */
994 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
995 {
996         size_t subbuf_size = buf->chan->subbuf_size;
997         size_t n_subbufs = buf->chan->n_subbufs;
998         size_t produced = buf->subbufs_produced;
999         size_t consumed = buf->subbufs_consumed;
1000
1001         relay_file_read_consume(buf, read_pos, 0);
1002
1003         consumed = buf->subbufs_consumed;
1004
1005         if (unlikely(buf->offset > subbuf_size)) {
1006                 if (produced == consumed)
1007                         return 0;
1008                 return 1;
1009         }
1010
1011         if (unlikely(produced - consumed >= n_subbufs)) {
1012                 consumed = produced - n_subbufs + 1;
1013                 buf->subbufs_consumed = consumed;
1014                 buf->bytes_consumed = 0;
1015         }
1016
1017         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
1018         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
1019
1020         if (consumed > produced)
1021                 produced += n_subbufs * subbuf_size;
1022
1023         if (consumed == produced) {
1024                 if (buf->offset == subbuf_size &&
1025                     buf->subbufs_produced > buf->subbufs_consumed)
1026                         return 1;
1027                 return 0;
1028         }
1029
1030         return 1;
1031 }
1032
1033 /**
1034  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1035  *      @read_pos: file read position
1036  *      @buf: relay channel buffer
1037  */
1038 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1039                                            struct rchan_buf *buf)
1040 {
1041         size_t padding, avail = 0;
1042         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1043         size_t subbuf_size = buf->chan->subbuf_size;
1044
1045         write_subbuf = (buf->data - buf->start) / subbuf_size;
1046         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1047         read_subbuf = read_pos / subbuf_size;
1048         read_offset = read_pos % subbuf_size;
1049         padding = buf->padding[read_subbuf];
1050
1051         if (read_subbuf == write_subbuf) {
1052                 if (read_offset + padding < write_offset)
1053                         avail = write_offset - (read_offset + padding);
1054         } else
1055                 avail = (subbuf_size - padding) - read_offset;
1056
1057         return avail;
1058 }
1059
1060 /**
1061  *      relay_file_read_start_pos - find the first available byte to read
1062  *      @read_pos: file read position
1063  *      @buf: relay channel buffer
1064  *
1065  *      If the @read_pos is in the middle of padding, return the
1066  *      position of the first actually available byte, otherwise
1067  *      return the original value.
1068  */
1069 static size_t relay_file_read_start_pos(size_t read_pos,
1070                                         struct rchan_buf *buf)
1071 {
1072         size_t read_subbuf, padding, padding_start, padding_end;
1073         size_t subbuf_size = buf->chan->subbuf_size;
1074         size_t n_subbufs = buf->chan->n_subbufs;
1075         size_t consumed = buf->subbufs_consumed % n_subbufs;
1076
1077         if (!read_pos)
1078                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1079         read_subbuf = read_pos / subbuf_size;
1080         padding = buf->padding[read_subbuf];
1081         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1082         padding_end = (read_subbuf + 1) * subbuf_size;
1083         if (read_pos >= padding_start && read_pos < padding_end) {
1084                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1085                 read_pos = read_subbuf * subbuf_size;
1086         }
1087
1088         return read_pos;
1089 }
1090
1091 /**
1092  *      relay_file_read_end_pos - return the new read position
1093  *      @read_pos: file read position
1094  *      @buf: relay channel buffer
1095  *      @count: number of bytes to be read
1096  */
1097 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1098                                       size_t read_pos,
1099                                       size_t count)
1100 {
1101         size_t read_subbuf, padding, end_pos;
1102         size_t subbuf_size = buf->chan->subbuf_size;
1103         size_t n_subbufs = buf->chan->n_subbufs;
1104
1105         read_subbuf = read_pos / subbuf_size;
1106         padding = buf->padding[read_subbuf];
1107         if (read_pos % subbuf_size + count + padding == subbuf_size)
1108                 end_pos = (read_subbuf + 1) * subbuf_size;
1109         else
1110                 end_pos = read_pos + count;
1111         if (end_pos >= subbuf_size * n_subbufs)
1112                 end_pos = 0;
1113
1114         return end_pos;
1115 }
1116
1117 static ssize_t relay_file_read(struct file *filp,
1118                                char __user *buffer,
1119                                size_t count,
1120                                loff_t *ppos)
1121 {
1122         struct rchan_buf *buf = filp->private_data;
1123         size_t read_start, avail;
1124         size_t written = 0;
1125         int ret;
1126
1127         if (!count)
1128                 return 0;
1129
1130         inode_lock(file_inode(filp));
1131         do {
1132                 void *from;
1133
1134                 if (!relay_file_read_avail(buf, *ppos))
1135                         break;
1136
1137                 read_start = relay_file_read_start_pos(*ppos, buf);
1138                 avail = relay_file_read_subbuf_avail(read_start, buf);
1139                 if (!avail)
1140                         break;
1141
1142                 avail = min(count, avail);
1143                 from = buf->start + read_start;
1144                 ret = avail;
1145                 if (copy_to_user(buffer, from, avail))
1146                         break;
1147
1148                 buffer += ret;
1149                 written += ret;
1150                 count -= ret;
1151
1152                 relay_file_read_consume(buf, read_start, ret);
1153                 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1154         } while (count);
1155         inode_unlock(file_inode(filp));
1156
1157         return written;
1158 }
1159
1160 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1161 {
1162         rbuf->bytes_consumed += bytes_consumed;
1163
1164         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1165                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1166                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1167         }
1168 }
1169
1170 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1171                                    struct pipe_buffer *buf)
1172 {
1173         struct rchan_buf *rbuf;
1174
1175         rbuf = (struct rchan_buf *)page_private(buf->page);
1176         relay_consume_bytes(rbuf, buf->private);
1177 }
1178
1179 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1180         .confirm = generic_pipe_buf_confirm,
1181         .release = relay_pipe_buf_release,
1182         .steal = generic_pipe_buf_steal,
1183         .get = generic_pipe_buf_get,
1184 };
1185
1186 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1187 {
1188 }
1189
1190 /*
1191  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1192  */
1193 static ssize_t subbuf_splice_actor(struct file *in,
1194                                loff_t *ppos,
1195                                struct pipe_inode_info *pipe,
1196                                size_t len,
1197                                unsigned int flags,
1198                                int *nonpad_ret)
1199 {
1200         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1201         struct rchan_buf *rbuf = in->private_data;
1202         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1203         uint64_t pos = (uint64_t) *ppos;
1204         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1205         size_t read_start = (size_t) do_div(pos, alloc_size);
1206         size_t read_subbuf = read_start / subbuf_size;
1207         size_t padding = rbuf->padding[read_subbuf];
1208         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1209         struct page *pages[PIPE_DEF_BUFFERS];
1210         struct partial_page partial[PIPE_DEF_BUFFERS];
1211         struct splice_pipe_desc spd = {
1212                 .pages = pages,
1213                 .nr_pages = 0,
1214                 .nr_pages_max = PIPE_DEF_BUFFERS,
1215                 .partial = partial,
1216                 .ops = &relay_pipe_buf_ops,
1217                 .spd_release = relay_page_release,
1218         };
1219         ssize_t ret;
1220
1221         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1222                 return 0;
1223         if (splice_grow_spd(pipe, &spd))
1224                 return -ENOMEM;
1225
1226         /*
1227          * Adjust read len, if longer than what is available
1228          */
1229         if (len > (subbuf_size - read_start % subbuf_size))
1230                 len = subbuf_size - read_start % subbuf_size;
1231
1232         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1233         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1234         poff = read_start & ~PAGE_MASK;
1235         nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1236
1237         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1238                 unsigned int this_len, this_end, private;
1239                 unsigned int cur_pos = read_start + total_len;
1240
1241                 if (!len)
1242                         break;
1243
1244                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1245                 private = this_len;
1246
1247                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1248                 spd.partial[spd.nr_pages].offset = poff;
1249
1250                 this_end = cur_pos + this_len;
1251                 if (this_end >= nonpad_end) {
1252                         this_len = nonpad_end - cur_pos;
1253                         private = this_len + padding;
1254                 }
1255                 spd.partial[spd.nr_pages].len = this_len;
1256                 spd.partial[spd.nr_pages].private = private;
1257
1258                 len -= this_len;
1259                 total_len += this_len;
1260                 poff = 0;
1261                 pidx = (pidx + 1) % subbuf_pages;
1262
1263                 if (this_end >= nonpad_end) {
1264                         spd.nr_pages++;
1265                         break;
1266                 }
1267         }
1268
1269         ret = 0;
1270         if (!spd.nr_pages)
1271                 goto out;
1272
1273         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1274         if (ret < 0 || ret < total_len)
1275                 goto out;
1276
1277         if (read_start + ret == nonpad_end)
1278                 ret += padding;
1279
1280 out:
1281         splice_shrink_spd(&spd);
1282         return ret;
1283 }
1284
1285 static ssize_t relay_file_splice_read(struct file *in,
1286                                       loff_t *ppos,
1287                                       struct pipe_inode_info *pipe,
1288                                       size_t len,
1289                                       unsigned int flags)
1290 {
1291         ssize_t spliced;
1292         int ret;
1293         int nonpad_ret = 0;
1294
1295         ret = 0;
1296         spliced = 0;
1297
1298         while (len && !spliced) {
1299                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1300                 if (ret < 0)
1301                         break;
1302                 else if (!ret) {
1303                         if (flags & SPLICE_F_NONBLOCK)
1304                                 ret = -EAGAIN;
1305                         break;
1306                 }
1307
1308                 *ppos += ret;
1309                 if (ret > len)
1310                         len = 0;
1311                 else
1312                         len -= ret;
1313                 spliced += nonpad_ret;
1314                 nonpad_ret = 0;
1315         }
1316
1317         if (spliced)
1318                 return spliced;
1319
1320         return ret;
1321 }
1322
1323 const struct file_operations relay_file_operations = {
1324         .open           = relay_file_open,
1325         .poll           = relay_file_poll,
1326         .mmap           = relay_file_mmap,
1327         .read           = relay_file_read,
1328         .llseek         = no_llseek,
1329         .release        = relay_file_release,
1330         .splice_read    = relay_file_splice_read,
1331 };
1332 EXPORT_SYMBOL_GPL(relay_file_operations);