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