kernel/kernel/relay.c

   1 /*
   2  * Public API and common code for kernel->userspace relay file support.
   3  *
   4  * See Documentation/filesystems/relay.txt for an overview.
   5  *
   6  * Copyright (C) 2002-2005 - Tom Zanussi (zanussi@us.ibm.com), IBM Corp
   7  * Copyright (C) 1999-2005 - Karim Yaghmour (karim@opersys.com)
   8  *
   9  * Moved to kernel/relay.c by Paul Mundt, 2006.
  10  * November 2006 - CPU hotplug support by Mathieu Desnoyers
  11  *      (mathieu.desnoyers@polymtl.ca)
  12  *
  13  * This file is released under the GPL.
  14  */
  15 #include <linux/errno.h>
  16 #include <linux/stddef.h>
  17 #include <linux/slab.h>
  18 #include <linux/export.h>
  19 #include <linux/string.h>
  20 #include <linux/relay.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/mm.h>
  23 #include <linux/cpu.h>
  24 #include <linux/splice.h>
  25
  26 /* list of open channels, for cpu hotplug */
  27 static DEFINE_MUTEX(relay_channels_mutex);
  28 static LIST_HEAD(relay_channels);
  29
  30 /*
  31  * close() vm_op implementation for relay file mapping.
  32  */
  33 static void relay_file_mmap_close(struct vm_area_struct *vma)
  34 {
  35         struct rchan_buf *buf = vma->vm_private_data;
  36         buf->chan->cb->buf_unmapped(buf, vma->vm_file);
  37 }
  38
  39 /*
  40  * fault() vm_op implementation for relay file mapping.
  41  */
  42 static int relay_buf_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
  43 {
  44         struct page *page;
  45         struct rchan_buf *buf = vma->vm_private_data;
  46         pgoff_t pgoff = vmf->pgoff;
  47
  48         if (!buf)
  49                 return VM_FAULT_OOM;
  50
  51         page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
  52         if (!page)
  53                 return VM_FAULT_SIGBUS;
  54         get_page(page);
  55         vmf->page = page;
  56
  57         return 0;
  58 }
  59
  60 /*
  61  * vm_ops for relay file mappings.
  62  */
  63 static const struct vm_operations_struct relay_file_mmap_ops = {
  64         .fault = relay_buf_fault,
  65         .close = relay_file_mmap_close,
  66 };
  67
  68 /*
  69  * allocate an array of pointers of struct page
  70  */
  71 static struct page **relay_alloc_page_array(unsigned int n_pages)
  72 {
  73         const size_t pa_size = n_pages * sizeof(struct page *);
  74         if (pa_size > PAGE_SIZE)
  75                 return vzalloc(pa_size);
  76         return kzalloc(pa_size, GFP_KERNEL);
  77 }
  78
  79 /*
  80  * free an array of pointers of struct page
  81  */
  82 static void relay_free_page_array(struct page **array)
  83 {
  84         kvfree(array);
  85 }
  86
  87 /**
  88  *      relay_mmap_buf: - mmap channel buffer to process address space
  89  *      @buf: relay channel buffer
  90  *      @vma: vm_area_struct describing memory to be mapped
  91  *
  92  *      Returns 0 if ok, negative on error
  93  *
  94  *      Caller should already have grabbed mmap_sem.
  95  */
  96 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
  97 {
  98         unsigned long length = vma->vm_end - vma->vm_start;
  99         struct file *filp = vma->vm_file;
 100
 101         if (!buf)
 102                 return -EBADF;
 103
 104         if (length != (unsigned long)buf->chan->alloc_size)
 105                 return -EINVAL;
 106
 107         vma->vm_ops = &relay_file_mmap_ops;
 108         vma->vm_flags |= VM_DONTEXPAND;
 109         vma->vm_private_data = buf;
 110         buf->chan->cb->buf_mapped(buf, filp);
 111
 112         return 0;
 113 }
 114
 115 /**
 116  *      relay_alloc_buf - allocate a channel buffer
 117  *      @buf: the buffer struct
 118  *      @size: total size of the buffer
 119  *
 120  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
 121  *      passed in size will get page aligned, if it isn't already.
 122  */
 123 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
 124 {
 125         void *mem;
 126         unsigned int i, j, n_pages;
 127
 128         *size = PAGE_ALIGN(*size);
 129         n_pages = *size >> PAGE_SHIFT;
 130
 131         buf->page_array = relay_alloc_page_array(n_pages);
 132         if (!buf->page_array)
 133                 return NULL;
 134
 135         for (i = 0; i < n_pages; i++) {
 136                 buf->page_array[i] = alloc_page(GFP_KERNEL);
 137                 if (unlikely(!buf->page_array[i]))
 138                         goto depopulate;
 139                 set_page_private(buf->page_array[i], (unsigned long)buf);
 140         }
 141         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
 142         if (!mem)
 143                 goto depopulate;
 144
 145         memset(mem, 0, *size);
 146         buf->page_count = n_pages;
 147         return mem;
 148
 149 depopulate:
 150         for (j = 0; j < i; j++)
 151                 __free_page(buf->page_array[j]);
 152         relay_free_page_array(buf->page_array);
 153         return NULL;
 154 }
 155
 156 /**
 157  *      relay_create_buf - allocate and initialize a channel buffer
 158  *      @chan: the relay channel
 159  *
 160  *      Returns channel buffer if successful, %NULL otherwise.
 161  */
 162 static struct rchan_buf *relay_create_buf(struct rchan *chan)
 163 {
 164         struct rchan_buf *buf;
 165
 166         if (chan->n_subbufs > UINT_MAX / sizeof(size_t *))
 167                 return NULL;
 168
 169         buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
 170         if (!buf)
 171                 return NULL;
 172         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
 173         if (!buf->padding)
 174                 goto free_buf;
 175
 176         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
 177         if (!buf->start)
 178                 goto free_buf;
 179
 180         buf->chan = chan;
 181         kref_get(&buf->chan->kref);
 182         return buf;
 183
 184 free_buf:
 185         kfree(buf->padding);
 186         kfree(buf);
 187         return NULL;
 188 }
 189
 190 /**
 191  *      relay_destroy_channel - free the channel struct
 192  *      @kref: target kernel reference that contains the relay channel
 193  *
 194  *      Should only be called from kref_put().
 195  */
 196 static void relay_destroy_channel(struct kref *kref)
 197 {
 198         struct rchan *chan = container_of(kref, struct rchan, kref);
 199         kfree(chan);
 200 }
 201
 202 /**
 203  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
 204  *      @buf: the buffer struct
 205  */
 206 static void relay_destroy_buf(struct rchan_buf *buf)
 207 {
 208         struct rchan *chan = buf->chan;
 209         unsigned int i;
 210
 211         if (likely(buf->start)) {
 212                 vunmap(buf->start);
 213                 for (i = 0; i < buf->page_count; i++)
 214                         __free_page(buf->page_array[i]);
 215                 relay_free_page_array(buf->page_array);
 216         }
 217         chan->buf[buf->cpu] = NULL;
 218         kfree(buf->padding);
 219         kfree(buf);
 220         kref_put(&chan->kref, relay_destroy_channel);
 221 }
 222
 223 /**
 224  *      relay_remove_buf - remove a channel buffer
 225  *      @kref: target kernel reference that contains the relay buffer
 226  *
 227  *      Removes the file from the filesystem, which also frees the
 228  *      rchan_buf_struct and the channel buffer.  Should only be called from
 229  *      kref_put().
 230  */
 231 static void relay_remove_buf(struct kref *kref)
 232 {
 233         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
 234         relay_destroy_buf(buf);
 235 }
 236
 237 /**
 238  *      relay_buf_empty - boolean, is the channel buffer empty?
 239  *      @buf: channel buffer
 240  *
 241  *      Returns 1 if the buffer is empty, 0 otherwise.
 242  */
 243 static int relay_buf_empty(struct rchan_buf *buf)
 244 {
 245         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
 246 }
 247
 248 /**
 249  *      relay_buf_full - boolean, is the channel buffer full?
 250  *      @buf: channel buffer
 251  *
 252  *      Returns 1 if the buffer is full, 0 otherwise.
 253  */
 254 int relay_buf_full(struct rchan_buf *buf)
 255 {
 256         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
 257         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
 258 }
 259 EXPORT_SYMBOL_GPL(relay_buf_full);
 260
 261 /*
 262  * High-level relay kernel API and associated functions.
 263  */
 264
 265 /*
 266  * rchan_callback implementations defining default channel behavior.  Used
 267  * in place of corresponding NULL values in client callback struct.
 268  */
 269
 270 /*
 271  * subbuf_start() default callback.  Does nothing.
 272  */
 273 static int subbuf_start_default_callback (struct rchan_buf *buf,
 274                                           void *subbuf,
 275                                           void *prev_subbuf,
 276                                           size_t prev_padding)
 277 {
 278         if (relay_buf_full(buf))
 279                 return 0;
 280
 281         return 1;
 282 }
 283
 284 /*
 285  * buf_mapped() default callback.  Does nothing.
 286  */
 287 static void buf_mapped_default_callback(struct rchan_buf *buf,
 288                                         struct file *filp)
 289 {
 290 }
 291
 292 /*
 293  * buf_unmapped() default callback.  Does nothing.
 294  */
 295 static void buf_unmapped_default_callback(struct rchan_buf *buf,
 296                                           struct file *filp)
 297 {
 298 }
 299
 300 /*
 301  * create_buf_file_create() default callback.  Does nothing.
 302  */
 303 static struct dentry *create_buf_file_default_callback(const char *filename,
 304                                                        struct dentry *parent,
 305                                                        umode_t mode,
 306                                                        struct rchan_buf *buf,
 307                                                        int *is_global)
 308 {
 309         return NULL;
 310 }
 311
 312 /*
 313  * remove_buf_file() default callback.  Does nothing.
 314  */
 315 static int remove_buf_file_default_callback(struct dentry *dentry)
 316 {
 317         return -EINVAL;
 318 }
 319
 320 /* relay channel default callbacks */
 321 static struct rchan_callbacks default_channel_callbacks = {
 322         .subbuf_start = subbuf_start_default_callback,
 323         .buf_mapped = buf_mapped_default_callback,
 324         .buf_unmapped = buf_unmapped_default_callback,
 325         .create_buf_file = create_buf_file_default_callback,
 326         .remove_buf_file = remove_buf_file_default_callback,
 327 };
 328
 329 /**
 330  *      wakeup_readers - wake up readers waiting on a channel
 331  *      @data: contains the channel buffer
 332  *
 333  *      This is the timer function used to defer reader waking.
 334  */
 335 static void wakeup_readers(unsigned long data)
 336 {
 337         struct rchan_buf *buf = (struct rchan_buf *)data;
 338         wake_up_interruptible(&buf->read_wait);
 339         /*
 340          * Stupid polling for now:
 341          */
 342         mod_timer(&buf->timer, jiffies + 1);
 343 }
 344
 345 /**
 346  *      __relay_reset - reset a channel buffer
 347  *      @buf: the channel buffer
 348  *      @init: 1 if this is a first-time initialization
 349  *
 350  *      See relay_reset() for description of effect.
 351  */
 352 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
 353 {
 354         size_t i;
 355
 356         if (init) {
 357                 init_waitqueue_head(&buf->read_wait);
 358                 kref_init(&buf->kref);
 359                 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
 360                 mod_timer(&buf->timer, jiffies + 1);
 361         } else
 362                 del_timer_sync(&buf->timer);
 363
 364         buf->subbufs_produced = 0;
 365         buf->subbufs_consumed = 0;
 366         buf->bytes_consumed = 0;
 367         buf->finalized = 0;
 368         buf->data = buf->start;
 369         buf->offset = 0;
 370
 371         for (i = 0; i < buf->chan->n_subbufs; i++)
 372                 buf->padding[i] = 0;
 373
 374         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
 375 }
 376
 377 /**
 378  *      relay_reset - reset the channel
 379  *      @chan: the channel
 380  *
 381  *      This has the effect of erasing all data from all channel buffers
 382  *      and restarting the channel in its initial state.  The buffers
 383  *      are not freed, so any mappings are still in effect.
 384  *
 385  *      NOTE. Care should be taken that the channel isn't actually
 386  *      being used by anything when this call is made.
 387  */
 388 void relay_reset(struct rchan *chan)
 389 {
 390         unsigned int i;
 391
 392         if (!chan)
 393                 return;
 394
 395         if (chan->is_global && chan->buf[0]) {
 396                 __relay_reset(chan->buf[0], 0);
 397                 return;
 398         }
 399
 400         mutex_lock(&relay_channels_mutex);
 401         for_each_possible_cpu(i)
 402                 if (chan->buf[i])
 403                         __relay_reset(chan->buf[i], 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
 432         kfree(tmpname);
 433
 434         return dentry;
 435 }
 436
 437 /*
 438  *      relay_open_buf - create a new relay channel buffer
 439  *
 440  *      used by relay_open() and CPU hotplug.
 441  */
 442 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
 443 {
 444         struct rchan_buf *buf = NULL;
 445         struct dentry *dentry;
 446
 447         if (chan->is_global)
 448                 return chan->buf[0];
 449
 450         buf = relay_create_buf(chan);
 451         if (!buf)
 452                 return NULL;
 453
 454         if (chan->has_base_filename) {
 455                 dentry = relay_create_buf_file(chan, buf, cpu);
 456                 if (!dentry)
 457                         goto free_buf;
 458                 relay_set_buf_dentry(buf, dentry);
 459         }
 460
 461         buf->cpu = cpu;
 462         __relay_reset(buf, 1);
 463
 464         if(chan->is_global) {
 465                 chan->buf[0] = buf;
 466                 buf->cpu = 0;
 467         }
 468
 469         return buf;
 470
 471 free_buf:
 472         relay_destroy_buf(buf);
 473         return NULL;
 474 }
 475
 476 /**
 477  *      relay_close_buf - close a channel buffer
 478  *      @buf: channel buffer
 479  *
 480  *      Marks the buffer finalized and restores the default callbacks.
 481  *      The channel buffer and channel buffer data structure are then freed
 482  *      automatically when the last reference is given up.
 483  */
 484 static void relay_close_buf(struct rchan_buf *buf)
 485 {
 486         buf->finalized = 1;
 487         del_timer_sync(&buf->timer);
 488         buf->chan->cb->remove_buf_file(buf->dentry);
 489         kref_put(&buf->kref, relay_remove_buf);
 490 }
 491
 492 static void setup_callbacks(struct rchan *chan,
 493                                    struct rchan_callbacks *cb)
 494 {
 495         if (!cb) {
 496                 chan->cb = &default_channel_callbacks;
 497                 return;
 498         }
 499
 500         if (!cb->subbuf_start)
 501                 cb->subbuf_start = subbuf_start_default_callback;
 502         if (!cb->buf_mapped)
 503                 cb->buf_mapped = buf_mapped_default_callback;
 504         if (!cb->buf_unmapped)
 505                 cb->buf_unmapped = buf_unmapped_default_callback;
 506         if (!cb->create_buf_file)
 507                 cb->create_buf_file = create_buf_file_default_callback;
 508         if (!cb->remove_buf_file)
 509                 cb->remove_buf_file = remove_buf_file_default_callback;
 510         chan->cb = cb;
 511 }
 512
 513 /**
 514  *      relay_hotcpu_callback - CPU hotplug callback
 515  *      @nb: notifier block
 516  *      @action: hotplug action to take
 517  *      @hcpu: CPU number
 518  *
 519  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
 520  */
 521 static int relay_hotcpu_callback(struct notifier_block *nb,
 522                                 unsigned long action,
 523                                 void *hcpu)
 524 {
 525         unsigned int hotcpu = (unsigned long)hcpu;
 526         struct rchan *chan;
 527
 528         switch(action) {
 529         case CPU_UP_PREPARE:
 530         case CPU_UP_PREPARE_FROZEN:
 531                 mutex_lock(&relay_channels_mutex);
 532                 list_for_each_entry(chan, &relay_channels, list) {
 533                         if (chan->buf[hotcpu])
 534                                 continue;
 535                         chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
 536                         if(!chan->buf[hotcpu]) {
 537                                 printk(KERN_ERR
 538                                         "relay_hotcpu_callback: cpu %d buffer "
 539                                         "creation failed\n", hotcpu);
 540                                 mutex_unlock(&relay_channels_mutex);
 541                                 return notifier_from_errno(-ENOMEM);
 542                         }
 543                 }
 544                 mutex_unlock(&relay_channels_mutex);
 545                 break;
 546         case CPU_DEAD:
 547         case CPU_DEAD_FROZEN:
 548                 /* No need to flush the cpu : will be flushed upon
 549                  * final relay_flush() call. */
 550                 break;
 551         }
 552         return NOTIFY_OK;
 553 }
 554
 555 /**
 556  *      relay_open - create a new relay channel
 557  *      @base_filename: base name of files to create, %NULL for buffering only
 558  *      @parent: dentry of parent directory, %NULL for root directory or buffer
 559  *      @subbuf_size: size of sub-buffers
 560  *      @n_subbufs: number of sub-buffers
 561  *      @cb: client callback functions
 562  *      @private_data: user-defined data
 563  *
 564  *      Returns channel pointer if successful, %NULL otherwise.
 565  *
 566  *      Creates a channel buffer for each cpu using the sizes and
 567  *      attributes specified.  The created channel buffer files
 568  *      will be named base_filename0...base_filenameN-1.  File
 569  *      permissions will be %S_IRUSR.
 570  */
 571 struct rchan *relay_open(const char *base_filename,
 572                          struct dentry *parent,
 573                          size_t subbuf_size,
 574                          size_t n_subbufs,
 575                          struct rchan_callbacks *cb,
 576                          void *private_data)
 577 {
 578         unsigned int i;
 579         struct rchan *chan;
 580
 581         if (!(subbuf_size && n_subbufs))
 582                 return NULL;
 583         if (subbuf_size > UINT_MAX / n_subbufs)
 584                 return NULL;
 585
 586         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
 587         if (!chan)
 588                 return NULL;
 589
 590         chan->version = RELAYFS_CHANNEL_VERSION;
 591         chan->n_subbufs = n_subbufs;
 592         chan->subbuf_size = subbuf_size;
 593         chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
 594         chan->parent = parent;
 595         chan->private_data = private_data;
 596         if (base_filename) {
 597                 chan->has_base_filename = 1;
 598                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
 599         }
 600         setup_callbacks(chan, cb);
 601         kref_init(&chan->kref);
 602
 603         mutex_lock(&relay_channels_mutex);
 604         for_each_online_cpu(i) {
 605                 chan->buf[i] = relay_open_buf(chan, i);
 606                 if (!chan->buf[i])
 607                         goto free_bufs;
 608         }
 609         list_add(&chan->list, &relay_channels);
 610         mutex_unlock(&relay_channels_mutex);
 611
 612         return chan;
 613
 614 free_bufs:
 615         for_each_possible_cpu(i) {
 616                 if (chan->buf[i])
 617                         relay_close_buf(chan->buf[i]);
 618         }
 619
 620         kref_put(&chan->kref, relay_destroy_channel);
 621         mutex_unlock(&relay_channels_mutex);
 622         return NULL;
 623 }
 624 EXPORT_SYMBOL_GPL(relay_open);
 625
 626 struct rchan_percpu_buf_dispatcher {
 627         struct rchan_buf *buf;
 628         struct dentry *dentry;
 629 };
 630
 631 /* Called in atomic context. */
 632 static void __relay_set_buf_dentry(void *info)
 633 {
 634         struct rchan_percpu_buf_dispatcher *p = info;
 635
 636         relay_set_buf_dentry(p->buf, p->dentry);
 637 }
 638
 639 /**
 640  *      relay_late_setup_files - triggers file creation
 641  *      @chan: channel to operate on
 642  *      @base_filename: base name of files to create
 643  *      @parent: dentry of parent directory, %NULL for root directory
 644  *
 645  *      Returns 0 if successful, non-zero otherwise.
 646  *
 647  *      Use to setup files for a previously buffer-only channel.
 648  *      Useful to do early tracing in kernel, before VFS is up, for example.
 649  */
 650 int relay_late_setup_files(struct rchan *chan,
 651                            const char *base_filename,
 652                            struct dentry *parent)
 653 {
 654         int err = 0;
 655         unsigned int i, curr_cpu;
 656         unsigned long flags;
 657         struct dentry *dentry;
 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         curr_cpu = get_cpu();
 674         /*
 675          * The CPU hotplug notifier ran before us and created buffers with
 676          * no files associated. So it's safe to call relay_setup_buf_file()
 677          * on all currently online CPUs.
 678          */
 679         for_each_online_cpu(i) {
 680                 if (unlikely(!chan->buf[i])) {
 681                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
 682                         err = -EINVAL;
 683                         break;
 684                 }
 685
 686                 dentry = relay_create_buf_file(chan, chan->buf[i], i);
 687                 if (unlikely(!dentry)) {
 688                         err = -EINVAL;
 689                         break;
 690                 }
 691
 692                 if (curr_cpu == i) {
 693                         local_irq_save(flags);
 694                         relay_set_buf_dentry(chan->buf[i], dentry);
 695                         local_irq_restore(flags);
 696                 } else {
 697                         disp.buf = chan->buf[i];
 698                         disp.dentry = dentry;
 699                         smp_mb();
 700                         /* relay_channels_mutex must be held, so wait. */
 701                         err = smp_call_function_single(i,
 702                                                        __relay_set_buf_dentry,
 703                                                        &disp, 1);
 704                 }
 705                 if (unlikely(err))
 706                         break;
 707         }
 708         put_cpu();
 709         mutex_unlock(&relay_channels_mutex);
 710
 711         return err;
 712 }
 713
 714 /**
 715  *      relay_switch_subbuf - switch to a new sub-buffer
 716  *      @buf: channel buffer
 717  *      @length: size of current event
 718  *
 719  *      Returns either the length passed in or 0 if full.
 720  *
 721  *      Performs sub-buffer-switch tasks such as invoking callbacks,
 722  *      updating padding counts, waking up readers, etc.
 723  */
 724 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
 725 {
 726         void *old, *new;
 727         size_t old_subbuf, new_subbuf;
 728
 729         if (unlikely(length > buf->chan->subbuf_size))
 730                 goto toobig;
 731
 732         if (buf->offset != buf->chan->subbuf_size + 1) {
 733                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
 734                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 735                 buf->padding[old_subbuf] = buf->prev_padding;
 736                 buf->subbufs_produced++;
 737                 if (buf->dentry)
 738                         d_inode(buf->dentry)->i_size +=
 739                                 buf->chan->subbuf_size -
 740                                 buf->padding[old_subbuf];
 741                 else
 742                         buf->early_bytes += buf->chan->subbuf_size -
 743                                             buf->padding[old_subbuf];
 744         }
 745
 746         old = buf->data;
 747         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 748         new = buf->start + new_subbuf * buf->chan->subbuf_size;
 749         buf->offset = 0;
 750         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
 751                 buf->offset = buf->chan->subbuf_size + 1;
 752                 return 0;
 753         }
 754         buf->data = new;
 755         buf->padding[new_subbuf] = 0;
 756
 757         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
 758                 goto toobig;
 759
 760         return length;
 761
 762 toobig:
 763         buf->chan->last_toobig = length;
 764         return 0;
 765 }
 766 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
 767
 768 /**
 769  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
 770  *      @chan: the channel
 771  *      @cpu: the cpu associated with the channel buffer to update
 772  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
 773  *
 774  *      Adds to the channel buffer's consumed sub-buffer count.
 775  *      subbufs_consumed should be the number of sub-buffers newly consumed,
 776  *      not the total consumed.
 777  *
 778  *      NOTE. Kernel clients don't need to call this function if the channel
 779  *      mode is 'overwrite'.
 780  */
 781 void relay_subbufs_consumed(struct rchan *chan,
 782                             unsigned int cpu,
 783                             size_t subbufs_consumed)
 784 {
 785         struct rchan_buf *buf;
 786
 787         if (!chan)
 788                 return;
 789
 790         if (cpu >= NR_CPUS || !chan->buf[cpu] ||
 791                                         subbufs_consumed > chan->n_subbufs)
 792                 return;
 793
 794         buf = chan->buf[cpu];
 795         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
 796                 buf->subbufs_consumed = buf->subbufs_produced;
 797         else
 798                 buf->subbufs_consumed += subbufs_consumed;
 799 }
 800 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
 801
 802 /**
 803  *      relay_close - close the channel
 804  *      @chan: the channel
 805  *
 806  *      Closes all channel buffers and frees the channel.
 807  */
 808 void relay_close(struct rchan *chan)
 809 {
 810         unsigned int i;
 811
 812         if (!chan)
 813                 return;
 814
 815         mutex_lock(&relay_channels_mutex);
 816         if (chan->is_global && chan->buf[0])
 817                 relay_close_buf(chan->buf[0]);
 818         else
 819                 for_each_possible_cpu(i)
 820                         if (chan->buf[i])
 821                                 relay_close_buf(chan->buf[i]);
 822
 823         if (chan->last_toobig)
 824                 printk(KERN_WARNING "relay: one or more items not logged "
 825                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
 826                        chan->last_toobig, chan->subbuf_size);
 827
 828         list_del(&chan->list);
 829         kref_put(&chan->kref, relay_destroy_channel);
 830         mutex_unlock(&relay_channels_mutex);
 831 }
 832 EXPORT_SYMBOL_GPL(relay_close);
 833
 834 /**
 835  *      relay_flush - close the channel
 836  *      @chan: the channel
 837  *
 838  *      Flushes all channel buffers, i.e. forces buffer switch.
 839  */
 840 void relay_flush(struct rchan *chan)
 841 {
 842         unsigned int i;
 843
 844         if (!chan)
 845                 return;
 846
 847         if (chan->is_global && chan->buf[0]) {
 848                 relay_switch_subbuf(chan->buf[0], 0);
 849                 return;
 850         }
 851
 852         mutex_lock(&relay_channels_mutex);
 853         for_each_possible_cpu(i)
 854                 if (chan->buf[i])
 855                         relay_switch_subbuf(chan->buf[i], 0);
 856         mutex_unlock(&relay_channels_mutex);
 857 }
 858 EXPORT_SYMBOL_GPL(relay_flush);
 859
 860 /**
 861  *      relay_file_open - open file op for relay files
 862  *      @inode: the inode
 863  *      @filp: the file
 864  *
 865  *      Increments the channel buffer refcount.
 866  */
 867 static int relay_file_open(struct inode *inode, struct file *filp)
 868 {
 869         struct rchan_buf *buf = inode->i_private;
 870         kref_get(&buf->kref);
 871         filp->private_data = buf;
 872
 873         return nonseekable_open(inode, filp);
 874 }
 875
 876 /**
 877  *      relay_file_mmap - mmap file op for relay files
 878  *      @filp: the file
 879  *      @vma: the vma describing what to map
 880  *
 881  *      Calls upon relay_mmap_buf() to map the file into user space.
 882  */
 883 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
 884 {
 885         struct rchan_buf *buf = filp->private_data;
 886         return relay_mmap_buf(buf, vma);
 887 }
 888
 889 /**
 890  *      relay_file_poll - poll file op for relay files
 891  *      @filp: the file
 892  *      @wait: poll table
 893  *
 894  *      Poll implemention.
 895  */
 896 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
 897 {
 898         unsigned int mask = 0;
 899         struct rchan_buf *buf = filp->private_data;
 900
 901         if (buf->finalized)
 902                 return POLLERR;
 903
 904         if (filp->f_mode & FMODE_READ) {
 905                 poll_wait(filp, &buf->read_wait, wait);
 906                 if (!relay_buf_empty(buf))
 907                         mask |= POLLIN | POLLRDNORM;
 908         }
 909
 910         return mask;
 911 }
 912
 913 /**
 914  *      relay_file_release - release file op for relay files
 915  *      @inode: the inode
 916  *      @filp: the file
 917  *
 918  *      Decrements the channel refcount, as the filesystem is
 919  *      no longer using it.
 920  */
 921 static int relay_file_release(struct inode *inode, struct file *filp)
 922 {
 923         struct rchan_buf *buf = filp->private_data;
 924         kref_put(&buf->kref, relay_remove_buf);
 925
 926         return 0;
 927 }
 928
 929 /*
 930  *      relay_file_read_consume - update the consumed count for the buffer
 931  */
 932 static void relay_file_read_consume(struct rchan_buf *buf,
 933                                     size_t read_pos,
 934                                     size_t bytes_consumed)
 935 {
 936         size_t subbuf_size = buf->chan->subbuf_size;
 937         size_t n_subbufs = buf->chan->n_subbufs;
 938         size_t read_subbuf;
 939
 940         if (buf->subbufs_produced == buf->subbufs_consumed &&
 941             buf->offset == buf->bytes_consumed)
 942                 return;
 943
 944         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
 945                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 946                 buf->bytes_consumed = 0;
 947         }
 948
 949         buf->bytes_consumed += bytes_consumed;
 950         if (!read_pos)
 951                 read_subbuf = buf->subbufs_consumed % n_subbufs;
 952         else
 953                 read_subbuf = read_pos / buf->chan->subbuf_size;
 954         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
 955                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
 956                     (buf->offset == subbuf_size))
 957                         return;
 958                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 959                 buf->bytes_consumed = 0;
 960         }
 961 }
 962
 963 /*
 964  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
 965  */
 966 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
 967 {
 968         size_t subbuf_size = buf->chan->subbuf_size;
 969         size_t n_subbufs = buf->chan->n_subbufs;
 970         size_t produced = buf->subbufs_produced;
 971         size_t consumed = buf->subbufs_consumed;
 972
 973         relay_file_read_consume(buf, read_pos, 0);
 974
 975         consumed = buf->subbufs_consumed;
 976
 977         if (unlikely(buf->offset > subbuf_size)) {
 978                 if (produced == consumed)
 979                         return 0;
 980                 return 1;
 981         }
 982
 983         if (unlikely(produced - consumed >= n_subbufs)) {
 984                 consumed = produced - n_subbufs + 1;
 985                 buf->subbufs_consumed = consumed;
 986                 buf->bytes_consumed = 0;
 987         }
 988
 989         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
 990         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
 991
 992         if (consumed > produced)
 993                 produced += n_subbufs * subbuf_size;
 994
 995         if (consumed == produced) {
 996                 if (buf->offset == subbuf_size &&
 997                     buf->subbufs_produced > buf->subbufs_consumed)
 998                         return 1;
 999                 return 0;
1000         }
1001
1002         return 1;
1003 }
1004
1005 /**
1006  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1007  *      @read_pos: file read position
1008  *      @buf: relay channel buffer
1009  */
1010 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1011                                            struct rchan_buf *buf)
1012 {
1013         size_t padding, avail = 0;
1014         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1015         size_t subbuf_size = buf->chan->subbuf_size;
1016
1017         write_subbuf = (buf->data - buf->start) / subbuf_size;
1018         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1019         read_subbuf = read_pos / subbuf_size;
1020         read_offset = read_pos % subbuf_size;
1021         padding = buf->padding[read_subbuf];
1022
1023         if (read_subbuf == write_subbuf) {
1024                 if (read_offset + padding < write_offset)
1025                         avail = write_offset - (read_offset + padding);
1026         } else
1027                 avail = (subbuf_size - padding) - read_offset;
1028
1029         return avail;
1030 }
1031
1032 /**
1033  *      relay_file_read_start_pos - find the first available byte to read
1034  *      @read_pos: file read position
1035  *      @buf: relay channel buffer
1036  *
1037  *      If the @read_pos is in the middle of padding, return the
1038  *      position of the first actually available byte, otherwise
1039  *      return the original value.
1040  */
1041 static size_t relay_file_read_start_pos(size_t read_pos,
1042                                         struct rchan_buf *buf)
1043 {
1044         size_t read_subbuf, padding, padding_start, padding_end;
1045         size_t subbuf_size = buf->chan->subbuf_size;
1046         size_t n_subbufs = buf->chan->n_subbufs;
1047         size_t consumed = buf->subbufs_consumed % n_subbufs;
1048
1049         if (!read_pos)
1050                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1051         read_subbuf = read_pos / subbuf_size;
1052         padding = buf->padding[read_subbuf];
1053         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1054         padding_end = (read_subbuf + 1) * subbuf_size;
1055         if (read_pos >= padding_start && read_pos < padding_end) {
1056                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1057                 read_pos = read_subbuf * subbuf_size;
1058         }
1059
1060         return read_pos;
1061 }
1062
1063 /**
1064  *      relay_file_read_end_pos - return the new read position
1065  *      @read_pos: file read position
1066  *      @buf: relay channel buffer
1067  *      @count: number of bytes to be read
1068  */
1069 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1070                                       size_t read_pos,
1071                                       size_t count)
1072 {
1073         size_t read_subbuf, padding, end_pos;
1074         size_t subbuf_size = buf->chan->subbuf_size;
1075         size_t n_subbufs = buf->chan->n_subbufs;
1076
1077         read_subbuf = read_pos / subbuf_size;
1078         padding = buf->padding[read_subbuf];
1079         if (read_pos % subbuf_size + count + padding == subbuf_size)
1080                 end_pos = (read_subbuf + 1) * subbuf_size;
1081         else
1082                 end_pos = read_pos + count;
1083         if (end_pos >= subbuf_size * n_subbufs)
1084                 end_pos = 0;
1085
1086         return end_pos;
1087 }
1088
1089 /*
1090  *      subbuf_read_actor - read up to one subbuf's worth of data
1091  */
1092 static int subbuf_read_actor(size_t read_start,
1093                              struct rchan_buf *buf,
1094                              size_t avail,
1095                              read_descriptor_t *desc)
1096 {
1097         void *from;
1098         int ret = 0;
1099
1100         from = buf->start + read_start;
1101         ret = avail;
1102         if (copy_to_user(desc->arg.buf, from, avail)) {
1103                 desc->error = -EFAULT;
1104                 ret = 0;
1105         }
1106         desc->arg.data += ret;
1107         desc->written += ret;
1108         desc->count -= ret;
1109
1110         return ret;
1111 }
1112
1113 typedef int (*subbuf_actor_t) (size_t read_start,
1114                                struct rchan_buf *buf,
1115                                size_t avail,
1116                                read_descriptor_t *desc);
1117
1118 /*
1119  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1120  */
1121 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1122                                         subbuf_actor_t subbuf_actor,
1123                                         read_descriptor_t *desc)
1124 {
1125         struct rchan_buf *buf = filp->private_data;
1126         size_t read_start, avail;
1127         int ret;
1128
1129         if (!desc->count)
1130                 return 0;
1131
1132         mutex_lock(&file_inode(filp)->i_mutex);
1133         do {
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(desc->count, avail);
1143                 ret = subbuf_actor(read_start, buf, avail, desc);
1144                 if (desc->error < 0)
1145                         break;
1146
1147                 if (ret) {
1148                         relay_file_read_consume(buf, read_start, ret);
1149                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1150                 }
1151         } while (desc->count && ret);
1152         mutex_unlock(&file_inode(filp)->i_mutex);
1153
1154         return desc->written;
1155 }
1156
1157 static ssize_t relay_file_read(struct file *filp,
1158                                char __user *buffer,
1159                                size_t count,
1160                                loff_t *ppos)
1161 {
1162         read_descriptor_t desc;
1163         desc.written = 0;
1164         desc.count = count;
1165         desc.arg.buf = buffer;
1166         desc.error = 0;
1167         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc);
1168 }
1169
1170 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1171 {
1172         rbuf->bytes_consumed += bytes_consumed;
1173
1174         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1175                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1176                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1177         }
1178 }
1179
1180 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1181                                    struct pipe_buffer *buf)
1182 {
1183         struct rchan_buf *rbuf;
1184
1185         rbuf = (struct rchan_buf *)page_private(buf->page);
1186         relay_consume_bytes(rbuf, buf->private);
1187 }
1188
1189 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1190         .can_merge = 0,
1191         .confirm = generic_pipe_buf_confirm,
1192         .release = relay_pipe_buf_release,
1193         .steal = generic_pipe_buf_steal,
1194         .get = generic_pipe_buf_get,
1195 };
1196
1197 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1198 {
1199 }
1200
1201 /*
1202  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1203  */
1204 static ssize_t subbuf_splice_actor(struct file *in,
1205                                loff_t *ppos,
1206                                struct pipe_inode_info *pipe,
1207                                size_t len,
1208                                unsigned int flags,
1209                                int *nonpad_ret)
1210 {
1211         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1212         struct rchan_buf *rbuf = in->private_data;
1213         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1214         uint64_t pos = (uint64_t) *ppos;
1215         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1216         size_t read_start = (size_t) do_div(pos, alloc_size);
1217         size_t read_subbuf = read_start / subbuf_size;
1218         size_t padding = rbuf->padding[read_subbuf];
1219         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1220         struct page *pages[PIPE_DEF_BUFFERS];
1221         struct partial_page partial[PIPE_DEF_BUFFERS];
1222         struct splice_pipe_desc spd = {
1223                 .pages = pages,
1224                 .nr_pages = 0,
1225                 .nr_pages_max = PIPE_DEF_BUFFERS,
1226                 .partial = partial,
1227                 .flags = flags,
1228                 .ops = &relay_pipe_buf_ops,
1229                 .spd_release = relay_page_release,
1230         };
1231         ssize_t ret;
1232
1233         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1234                 return 0;
1235         if (splice_grow_spd(pipe, &spd))
1236                 return -ENOMEM;
1237
1238         /*
1239          * Adjust read len, if longer than what is available
1240          */
1241         if (len > (subbuf_size - read_start % subbuf_size))
1242                 len = subbuf_size - read_start % subbuf_size;
1243
1244         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1245         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1246         poff = read_start & ~PAGE_MASK;
1247         nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1248
1249         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1250                 unsigned int this_len, this_end, private;
1251                 unsigned int cur_pos = read_start + total_len;
1252
1253                 if (!len)
1254                         break;
1255
1256                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1257                 private = this_len;
1258
1259                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1260                 spd.partial[spd.nr_pages].offset = poff;
1261
1262                 this_end = cur_pos + this_len;
1263                 if (this_end >= nonpad_end) {
1264                         this_len = nonpad_end - cur_pos;
1265                         private = this_len + padding;
1266                 }
1267                 spd.partial[spd.nr_pages].len = this_len;
1268                 spd.partial[spd.nr_pages].private = private;
1269
1270                 len -= this_len;
1271                 total_len += this_len;
1272                 poff = 0;
1273                 pidx = (pidx + 1) % subbuf_pages;
1274
1275                 if (this_end >= nonpad_end) {
1276                         spd.nr_pages++;
1277                         break;
1278                 }
1279         }
1280
1281         ret = 0;
1282         if (!spd.nr_pages)
1283                 goto out;
1284
1285         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1286         if (ret < 0 || ret < total_len)
1287                 goto out;
1288
1289         if (read_start + ret == nonpad_end)
1290                 ret += padding;
1291
1292 out:
1293         splice_shrink_spd(&spd);
1294         return ret;
1295 }
1296
1297 static ssize_t relay_file_splice_read(struct file *in,
1298                                       loff_t *ppos,
1299                                       struct pipe_inode_info *pipe,
1300                                       size_t len,
1301                                       unsigned int flags)
1302 {
1303         ssize_t spliced;
1304         int ret;
1305         int nonpad_ret = 0;
1306
1307         ret = 0;
1308         spliced = 0;
1309
1310         while (len && !spliced) {
1311                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1312                 if (ret < 0)
1313                         break;
1314                 else if (!ret) {
1315                         if (flags & SPLICE_F_NONBLOCK)
1316                                 ret = -EAGAIN;
1317                         break;
1318                 }
1319
1320                 *ppos += ret;
1321                 if (ret > len)
1322                         len = 0;
1323                 else
1324                         len -= ret;
1325                 spliced += nonpad_ret;
1326                 nonpad_ret = 0;
1327         }
1328
1329         if (spliced)
1330                 return spliced;
1331
1332         return ret;
1333 }
1334
1335 const struct file_operations relay_file_operations = {
1336         .open           = relay_file_open,
1337         .poll           = relay_file_poll,
1338         .mmap           = relay_file_mmap,
1339         .read           = relay_file_read,
1340         .llseek         = no_llseek,
1341         .release        = relay_file_release,
1342         .splice_read    = relay_file_splice_read,
1343 };
1344 EXPORT_SYMBOL_GPL(relay_file_operations);
1345
1346 static __init int relay_init(void)
1347 {
1348
1349         hotcpu_notifier(relay_hotcpu_callback, 0);
1350         return 0;
1351 }
1352
1353 early_initcall(relay_init);