kernel/relay.c

   1 /*
   2  * Public API and common code for kernel->userspace relay file support.
   3  *
   4  * See Documentation/filesystems/relay.rst 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  * fault() vm_op implementation for relay file mapping.
  32  */
  33 static vm_fault_t relay_buf_fault(struct vm_fault *vmf)
  34 {
  35         struct page *page;
  36         struct rchan_buf *buf = vmf->vma->vm_private_data;
  37         pgoff_t pgoff = vmf->pgoff;
  38
  39         if (!buf)
  40                 return VM_FAULT_OOM;
  41
  42         page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
  43         if (!page)
  44                 return VM_FAULT_SIGBUS;
  45         get_page(page);
  46         vmf->page = page;
  47
  48         return 0;
  49 }
  50
  51 /*
  52  * vm_ops for relay file mappings.
  53  */
  54 static const struct vm_operations_struct relay_file_mmap_ops = {
  55         .fault = relay_buf_fault,
  56 };
  57
  58 /*
  59  * allocate an array of pointers of struct page
  60  */
  61 static struct page **relay_alloc_page_array(unsigned int n_pages)
  62 {
  63         const size_t pa_size = n_pages * sizeof(struct page *);
  64         if (pa_size > PAGE_SIZE)
  65                 return vzalloc(pa_size);
  66         return kzalloc(pa_size, GFP_KERNEL);
  67 }
  68
  69 /*
  70  * free an array of pointers of struct page
  71  */
  72 static void relay_free_page_array(struct page **array)
  73 {
  74         kvfree(array);
  75 }
  76
  77 /**
  78  *      relay_mmap_buf: - mmap channel buffer to process address space
  79  *      @buf: relay channel buffer
  80  *      @vma: vm_area_struct describing memory to be mapped
  81  *
  82  *      Returns 0 if ok, negative on error
  83  *
  84  *      Caller should already have grabbed mmap_lock.
  85  */
  86 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
  87 {
  88         unsigned long length = vma->vm_end - vma->vm_start;
  89
  90         if (!buf)
  91                 return -EBADF;
  92
  93         if (length != (unsigned long)buf->chan->alloc_size)
  94                 return -EINVAL;
  95
  96         vma->vm_ops = &relay_file_mmap_ops;
  97         vma->vm_flags |= VM_DONTEXPAND;
  98         vma->vm_private_data = buf;
  99
 100         return 0;
 101 }
 102
 103 /**
 104  *      relay_alloc_buf - allocate a channel buffer
 105  *      @buf: the buffer struct
 106  *      @size: total size of the buffer
 107  *
 108  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
 109  *      passed in size will get page aligned, if it isn't already.
 110  */
 111 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
 112 {
 113         void *mem;
 114         unsigned int i, j, n_pages;
 115
 116         *size = PAGE_ALIGN(*size);
 117         n_pages = *size >> PAGE_SHIFT;
 118
 119         buf->page_array = relay_alloc_page_array(n_pages);
 120         if (!buf->page_array)
 121                 return NULL;
 122
 123         for (i = 0; i < n_pages; i++) {
 124                 buf->page_array[i] = alloc_page(GFP_KERNEL);
 125                 if (unlikely(!buf->page_array[i]))
 126                         goto depopulate;
 127                 set_page_private(buf->page_array[i], (unsigned long)buf);
 128         }
 129         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
 130         if (!mem)
 131                 goto depopulate;
 132
 133         memset(mem, 0, *size);
 134         buf->page_count = n_pages;
 135         return mem;
 136
 137 depopulate:
 138         for (j = 0; j < i; j++)
 139                 __free_page(buf->page_array[j]);
 140         relay_free_page_array(buf->page_array);
 141         return NULL;
 142 }
 143
 144 /**
 145  *      relay_create_buf - allocate and initialize a channel buffer
 146  *      @chan: the relay channel
 147  *
 148  *      Returns channel buffer if successful, %NULL otherwise.
 149  */
 150 static struct rchan_buf *relay_create_buf(struct rchan *chan)
 151 {
 152         struct rchan_buf *buf;
 153
 154         if (chan->n_subbufs > KMALLOC_MAX_SIZE / sizeof(size_t *))
 155                 return NULL;
 156
 157         buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
 158         if (!buf)
 159                 return NULL;
 160         buf->padding = kmalloc_array(chan->n_subbufs, sizeof(size_t *),
 161                                      GFP_KERNEL);
 162         if (!buf->padding)
 163                 goto free_buf;
 164
 165         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
 166         if (!buf->start)
 167                 goto free_buf;
 168
 169         buf->chan = chan;
 170         kref_get(&buf->chan->kref);
 171         return buf;
 172
 173 free_buf:
 174         kfree(buf->padding);
 175         kfree(buf);
 176         return NULL;
 177 }
 178
 179 /**
 180  *      relay_destroy_channel - free the channel struct
 181  *      @kref: target kernel reference that contains the relay channel
 182  *
 183  *      Should only be called from kref_put().
 184  */
 185 static void relay_destroy_channel(struct kref *kref)
 186 {
 187         struct rchan *chan = container_of(kref, struct rchan, kref);
 188         free_percpu(chan->buf);
 189         kfree(chan);
 190 }
 191
 192 /**
 193  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
 194  *      @buf: the buffer struct
 195  */
 196 static void relay_destroy_buf(struct rchan_buf *buf)
 197 {
 198         struct rchan *chan = buf->chan;
 199         unsigned int i;
 200
 201         if (likely(buf->start)) {
 202                 vunmap(buf->start);
 203                 for (i = 0; i < buf->page_count; i++)
 204                         __free_page(buf->page_array[i]);
 205                 relay_free_page_array(buf->page_array);
 206         }
 207         *per_cpu_ptr(chan->buf, buf->cpu) = NULL;
 208         kfree(buf->padding);
 209         kfree(buf);
 210         kref_put(&chan->kref, relay_destroy_channel);
 211 }
 212
 213 /**
 214  *      relay_remove_buf - remove a channel buffer
 215  *      @kref: target kernel reference that contains the relay buffer
 216  *
 217  *      Removes the file from the filesystem, which also frees the
 218  *      rchan_buf_struct and the channel buffer.  Should only be called from
 219  *      kref_put().
 220  */
 221 static void relay_remove_buf(struct kref *kref)
 222 {
 223         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
 224         relay_destroy_buf(buf);
 225 }
 226
 227 /**
 228  *      relay_buf_empty - boolean, is the channel buffer empty?
 229  *      @buf: channel buffer
 230  *
 231  *      Returns 1 if the buffer is empty, 0 otherwise.
 232  */
 233 static int relay_buf_empty(struct rchan_buf *buf)
 234 {
 235         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
 236 }
 237
 238 /**
 239  *      relay_buf_full - boolean, is the channel buffer full?
 240  *      @buf: channel buffer
 241  *
 242  *      Returns 1 if the buffer is full, 0 otherwise.
 243  */
 244 int relay_buf_full(struct rchan_buf *buf)
 245 {
 246         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
 247         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
 248 }
 249 EXPORT_SYMBOL_GPL(relay_buf_full);
 250
 251 /*
 252  * High-level relay kernel API and associated functions.
 253  */
 254
 255 /*
 256  * rchan_callback implementations defining default channel behavior.  Used
 257  * in place of corresponding NULL values in client callback struct.
 258  */
 259
 260 /*
 261  * subbuf_start() default callback.  Does nothing.
 262  */
 263 static int subbuf_start_default_callback (struct rchan_buf *buf,
 264                                           void *subbuf,
 265                                           void *prev_subbuf,
 266                                           size_t prev_padding)
 267 {
 268         if (relay_buf_full(buf))
 269                 return 0;
 270
 271         return 1;
 272 }
 273
 274 /*
 275  * create_buf_file_create() default callback.  Does nothing.
 276  */
 277 static struct dentry *create_buf_file_default_callback(const char *filename,
 278                                                        struct dentry *parent,
 279                                                        umode_t mode,
 280                                                        struct rchan_buf *buf,
 281                                                        int *is_global)
 282 {
 283         return NULL;
 284 }
 285
 286 /*
 287  * remove_buf_file() default callback.  Does nothing.
 288  */
 289 static int remove_buf_file_default_callback(struct dentry *dentry)
 290 {
 291         return -EINVAL;
 292 }
 293
 294 /**
 295  *      wakeup_readers - wake up readers waiting on a channel
 296  *      @work: contains the channel buffer
 297  *
 298  *      This is the function used to defer reader waking
 299  */
 300 static void wakeup_readers(struct irq_work *work)
 301 {
 302         struct rchan_buf *buf;
 303
 304         buf = container_of(work, struct rchan_buf, wakeup_work);
 305         wake_up_interruptible(&buf->read_wait);
 306 }
 307
 308 /**
 309  *      __relay_reset - reset a channel buffer
 310  *      @buf: the channel buffer
 311  *      @init: 1 if this is a first-time initialization
 312  *
 313  *      See relay_reset() for description of effect.
 314  */
 315 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
 316 {
 317         size_t i;
 318
 319         if (init) {
 320                 init_waitqueue_head(&buf->read_wait);
 321                 kref_init(&buf->kref);
 322                 init_irq_work(&buf->wakeup_work, wakeup_readers);
 323         } else {
 324                 irq_work_sync(&buf->wakeup_work);
 325         }
 326
 327         buf->subbufs_produced = 0;
 328         buf->subbufs_consumed = 0;
 329         buf->bytes_consumed = 0;
 330         buf->finalized = 0;
 331         buf->data = buf->start;
 332         buf->offset = 0;
 333
 334         for (i = 0; i < buf->chan->n_subbufs; i++)
 335                 buf->padding[i] = 0;
 336
 337         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
 338 }
 339
 340 /**
 341  *      relay_reset - reset the channel
 342  *      @chan: the channel
 343  *
 344  *      This has the effect of erasing all data from all channel buffers
 345  *      and restarting the channel in its initial state.  The buffers
 346  *      are not freed, so any mappings are still in effect.
 347  *
 348  *      NOTE. Care should be taken that the channel isn't actually
 349  *      being used by anything when this call is made.
 350  */
 351 void relay_reset(struct rchan *chan)
 352 {
 353         struct rchan_buf *buf;
 354         unsigned int i;
 355
 356         if (!chan)
 357                 return;
 358
 359         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
 360                 __relay_reset(buf, 0);
 361                 return;
 362         }
 363
 364         mutex_lock(&relay_channels_mutex);
 365         for_each_possible_cpu(i)
 366                 if ((buf = *per_cpu_ptr(chan->buf, i)))
 367                         __relay_reset(buf, 0);
 368         mutex_unlock(&relay_channels_mutex);
 369 }
 370 EXPORT_SYMBOL_GPL(relay_reset);
 371
 372 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
 373                                         struct dentry *dentry)
 374 {
 375         buf->dentry = dentry;
 376         d_inode(buf->dentry)->i_size = buf->early_bytes;
 377 }
 378
 379 static struct dentry *relay_create_buf_file(struct rchan *chan,
 380                                             struct rchan_buf *buf,
 381                                             unsigned int cpu)
 382 {
 383         struct dentry *dentry;
 384         char *tmpname;
 385
 386         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
 387         if (!tmpname)
 388                 return NULL;
 389         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
 390
 391         /* Create file in fs */
 392         dentry = chan->cb->create_buf_file(tmpname, chan->parent,
 393                                            S_IRUSR, buf,
 394                                            &chan->is_global);
 395         if (IS_ERR(dentry))
 396                 dentry = NULL;
 397
 398         kfree(tmpname);
 399
 400         return dentry;
 401 }
 402
 403 /*
 404  *      relay_open_buf - create a new relay channel buffer
 405  *
 406  *      used by relay_open() and CPU hotplug.
 407  */
 408 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
 409 {
 410         struct rchan_buf *buf = NULL;
 411         struct dentry *dentry;
 412
 413         if (chan->is_global)
 414                 return *per_cpu_ptr(chan->buf, 0);
 415
 416         buf = relay_create_buf(chan);
 417         if (!buf)
 418                 return NULL;
 419
 420         if (chan->has_base_filename) {
 421                 dentry = relay_create_buf_file(chan, buf, cpu);
 422                 if (!dentry)
 423                         goto free_buf;
 424                 relay_set_buf_dentry(buf, dentry);
 425         } else {
 426                 /* Only retrieve global info, nothing more, nothing less */
 427                 dentry = chan->cb->create_buf_file(NULL, NULL,
 428                                                    S_IRUSR, buf,
 429                                                    &chan->is_global);
 430                 if (IS_ERR_OR_NULL(dentry))
 431                         goto free_buf;
 432         }
 433
 434         buf->cpu = cpu;
 435         __relay_reset(buf, 1);
 436
 437         if(chan->is_global) {
 438                 *per_cpu_ptr(chan->buf, 0) = buf;
 439                 buf->cpu = 0;
 440         }
 441
 442         return buf;
 443
 444 free_buf:
 445         relay_destroy_buf(buf);
 446         return NULL;
 447 }
 448
 449 /**
 450  *      relay_close_buf - close a channel buffer
 451  *      @buf: channel buffer
 452  *
 453  *      Marks the buffer finalized and restores the default callbacks.
 454  *      The channel buffer and channel buffer data structure are then freed
 455  *      automatically when the last reference is given up.
 456  */
 457 static void relay_close_buf(struct rchan_buf *buf)
 458 {
 459         buf->finalized = 1;
 460         irq_work_sync(&buf->wakeup_work);
 461         buf->chan->cb->remove_buf_file(buf->dentry);
 462         kref_put(&buf->kref, relay_remove_buf);
 463 }
 464
 465 static void setup_callbacks(struct rchan *chan,
 466                                    struct rchan_callbacks *cb)
 467 {
 468         if (!cb->subbuf_start)
 469                 cb->subbuf_start = subbuf_start_default_callback;
 470         if (!cb->create_buf_file)
 471                 cb->create_buf_file = create_buf_file_default_callback;
 472         if (!cb->remove_buf_file)
 473                 cb->remove_buf_file = remove_buf_file_default_callback;
 474         chan->cb = cb;
 475 }
 476
 477 int relay_prepare_cpu(unsigned int cpu)
 478 {
 479         struct rchan *chan;
 480         struct rchan_buf *buf;
 481
 482         mutex_lock(&relay_channels_mutex);
 483         list_for_each_entry(chan, &relay_channels, list) {
 484                 if ((buf = *per_cpu_ptr(chan->buf, cpu)))
 485                         continue;
 486                 buf = relay_open_buf(chan, cpu);
 487                 if (!buf) {
 488                         pr_err("relay: cpu %d buffer creation failed\n", cpu);
 489                         mutex_unlock(&relay_channels_mutex);
 490                         return -ENOMEM;
 491                 }
 492                 *per_cpu_ptr(chan->buf, cpu) = buf;
 493         }
 494         mutex_unlock(&relay_channels_mutex);
 495         return 0;
 496 }
 497
 498 /**
 499  *      relay_open - create a new relay channel
 500  *      @base_filename: base name of files to create, %NULL for buffering only
 501  *      @parent: dentry of parent directory, %NULL for root directory or buffer
 502  *      @subbuf_size: size of sub-buffers
 503  *      @n_subbufs: number of sub-buffers
 504  *      @cb: client callback functions
 505  *      @private_data: user-defined data
 506  *
 507  *      Returns channel pointer if successful, %NULL otherwise.
 508  *
 509  *      Creates a channel buffer for each cpu using the sizes and
 510  *      attributes specified.  The created channel buffer files
 511  *      will be named base_filename0...base_filenameN-1.  File
 512  *      permissions will be %S_IRUSR.
 513  *
 514  *      If opening a buffer (@parent = NULL) that you later wish to register
 515  *      in a filesystem, call relay_late_setup_files() once the @parent dentry
 516  *      is available.
 517  */
 518 struct rchan *relay_open(const char *base_filename,
 519                          struct dentry *parent,
 520                          size_t subbuf_size,
 521                          size_t n_subbufs,
 522                          struct rchan_callbacks *cb,
 523                          void *private_data)
 524 {
 525         unsigned int i;
 526         struct rchan *chan;
 527         struct rchan_buf *buf;
 528
 529         if (!(subbuf_size && n_subbufs))
 530                 return NULL;
 531         if (subbuf_size > UINT_MAX / n_subbufs)
 532                 return NULL;
 533         if (!cb)
 534                 return NULL;
 535
 536         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
 537         if (!chan)
 538                 return NULL;
 539
 540         chan->buf = alloc_percpu(struct rchan_buf *);
 541         if (!chan->buf) {
 542                 kfree(chan);
 543                 return NULL;
 544         }
 545
 546         chan->version = RELAYFS_CHANNEL_VERSION;
 547         chan->n_subbufs = n_subbufs;
 548         chan->subbuf_size = subbuf_size;
 549         chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
 550         chan->parent = parent;
 551         chan->private_data = private_data;
 552         if (base_filename) {
 553                 chan->has_base_filename = 1;
 554                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
 555         }
 556         setup_callbacks(chan, cb);
 557         kref_init(&chan->kref);
 558
 559         mutex_lock(&relay_channels_mutex);
 560         for_each_online_cpu(i) {
 561                 buf = relay_open_buf(chan, i);
 562                 if (!buf)
 563                         goto free_bufs;
 564                 *per_cpu_ptr(chan->buf, i) = buf;
 565         }
 566         list_add(&chan->list, &relay_channels);
 567         mutex_unlock(&relay_channels_mutex);
 568
 569         return chan;
 570
 571 free_bufs:
 572         for_each_possible_cpu(i) {
 573                 if ((buf = *per_cpu_ptr(chan->buf, i)))
 574                         relay_close_buf(buf);
 575         }
 576
 577         kref_put(&chan->kref, relay_destroy_channel);
 578         mutex_unlock(&relay_channels_mutex);
 579         return NULL;
 580 }
 581 EXPORT_SYMBOL_GPL(relay_open);
 582
 583 struct rchan_percpu_buf_dispatcher {
 584         struct rchan_buf *buf;
 585         struct dentry *dentry;
 586 };
 587
 588 /* Called in atomic context. */
 589 static void __relay_set_buf_dentry(void *info)
 590 {
 591         struct rchan_percpu_buf_dispatcher *p = info;
 592
 593         relay_set_buf_dentry(p->buf, p->dentry);
 594 }
 595
 596 /**
 597  *      relay_late_setup_files - triggers file creation
 598  *      @chan: channel to operate on
 599  *      @base_filename: base name of files to create
 600  *      @parent: dentry of parent directory, %NULL for root directory
 601  *
 602  *      Returns 0 if successful, non-zero otherwise.
 603  *
 604  *      Use to setup files for a previously buffer-only channel created
 605  *      by relay_open() with a NULL parent dentry.
 606  *
 607  *      For example, this is useful for perfomring early tracing in kernel,
 608  *      before VFS is up and then exposing the early results once the dentry
 609  *      is available.
 610  */
 611 int relay_late_setup_files(struct rchan *chan,
 612                            const char *base_filename,
 613                            struct dentry *parent)
 614 {
 615         int err = 0;
 616         unsigned int i, curr_cpu;
 617         unsigned long flags;
 618         struct dentry *dentry;
 619         struct rchan_buf *buf;
 620         struct rchan_percpu_buf_dispatcher disp;
 621
 622         if (!chan || !base_filename)
 623                 return -EINVAL;
 624
 625         strlcpy(chan->base_filename, base_filename, NAME_MAX);
 626
 627         mutex_lock(&relay_channels_mutex);
 628         /* Is chan already set up? */
 629         if (unlikely(chan->has_base_filename)) {
 630                 mutex_unlock(&relay_channels_mutex);
 631                 return -EEXIST;
 632         }
 633         chan->has_base_filename = 1;
 634         chan->parent = parent;
 635
 636         if (chan->is_global) {
 637                 err = -EINVAL;
 638                 buf = *per_cpu_ptr(chan->buf, 0);
 639                 if (!WARN_ON_ONCE(!buf)) {
 640                         dentry = relay_create_buf_file(chan, buf, 0);
 641                         if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
 642                                 relay_set_buf_dentry(buf, dentry);
 643                                 err = 0;
 644                         }
 645                 }
 646                 mutex_unlock(&relay_channels_mutex);
 647                 return err;
 648         }
 649
 650         curr_cpu = get_cpu();
 651         /*
 652          * The CPU hotplug notifier ran before us and created buffers with
 653          * no files associated. So it's safe to call relay_setup_buf_file()
 654          * on all currently online CPUs.
 655          */
 656         for_each_online_cpu(i) {
 657                 buf = *per_cpu_ptr(chan->buf, i);
 658                 if (unlikely(!buf)) {
 659                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
 660                         err = -EINVAL;
 661                         break;
 662                 }
 663
 664                 dentry = relay_create_buf_file(chan, buf, i);
 665                 if (unlikely(!dentry)) {
 666                         err = -EINVAL;
 667                         break;
 668                 }
 669
 670                 if (curr_cpu == i) {
 671                         local_irq_save(flags);
 672                         relay_set_buf_dentry(buf, dentry);
 673                         local_irq_restore(flags);
 674                 } else {
 675                         disp.buf = buf;
 676                         disp.dentry = dentry;
 677                         smp_mb();
 678                         /* relay_channels_mutex must be held, so wait. */
 679                         err = smp_call_function_single(i,
 680                                                        __relay_set_buf_dentry,
 681                                                        &disp, 1);
 682                 }
 683                 if (unlikely(err))
 684                         break;
 685         }
 686         put_cpu();
 687         mutex_unlock(&relay_channels_mutex);
 688
 689         return err;
 690 }
 691 EXPORT_SYMBOL_GPL(relay_late_setup_files);
 692
 693 /**
 694  *      relay_switch_subbuf - switch to a new sub-buffer
 695  *      @buf: channel buffer
 696  *      @length: size of current event
 697  *
 698  *      Returns either the length passed in or 0 if full.
 699  *
 700  *      Performs sub-buffer-switch tasks such as invoking callbacks,
 701  *      updating padding counts, waking up readers, etc.
 702  */
 703 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
 704 {
 705         void *old, *new;
 706         size_t old_subbuf, new_subbuf;
 707
 708         if (unlikely(length > buf->chan->subbuf_size))
 709                 goto toobig;
 710
 711         if (buf->offset != buf->chan->subbuf_size + 1) {
 712                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
 713                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 714                 buf->padding[old_subbuf] = buf->prev_padding;
 715                 buf->subbufs_produced++;
 716                 if (buf->dentry)
 717                         d_inode(buf->dentry)->i_size +=
 718                                 buf->chan->subbuf_size -
 719                                 buf->padding[old_subbuf];
 720                 else
 721                         buf->early_bytes += buf->chan->subbuf_size -
 722                                             buf->padding[old_subbuf];
 723                 smp_mb();
 724                 if (waitqueue_active(&buf->read_wait)) {
 725                         /*
 726                          * Calling wake_up_interruptible() from here
 727                          * will deadlock if we happen to be logging
 728                          * from the scheduler (trying to re-grab
 729                          * rq->lock), so defer it.
 730                          */
 731                         irq_work_queue(&buf->wakeup_work);
 732                 }
 733         }
 734
 735         old = buf->data;
 736         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
 737         new = buf->start + new_subbuf * buf->chan->subbuf_size;
 738         buf->offset = 0;
 739         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
 740                 buf->offset = buf->chan->subbuf_size + 1;
 741                 return 0;
 742         }
 743         buf->data = new;
 744         buf->padding[new_subbuf] = 0;
 745
 746         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
 747                 goto toobig;
 748
 749         return length;
 750
 751 toobig:
 752         buf->chan->last_toobig = length;
 753         return 0;
 754 }
 755 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
 756
 757 /**
 758  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
 759  *      @chan: the channel
 760  *      @cpu: the cpu associated with the channel buffer to update
 761  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
 762  *
 763  *      Adds to the channel buffer's consumed sub-buffer count.
 764  *      subbufs_consumed should be the number of sub-buffers newly consumed,
 765  *      not the total consumed.
 766  *
 767  *      NOTE. Kernel clients don't need to call this function if the channel
 768  *      mode is 'overwrite'.
 769  */
 770 void relay_subbufs_consumed(struct rchan *chan,
 771                             unsigned int cpu,
 772                             size_t subbufs_consumed)
 773 {
 774         struct rchan_buf *buf;
 775
 776         if (!chan || cpu >= NR_CPUS)
 777                 return;
 778
 779         buf = *per_cpu_ptr(chan->buf, cpu);
 780         if (!buf || subbufs_consumed > chan->n_subbufs)
 781                 return;
 782
 783         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
 784                 buf->subbufs_consumed = buf->subbufs_produced;
 785         else
 786                 buf->subbufs_consumed += subbufs_consumed;
 787 }
 788 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
 789
 790 /**
 791  *      relay_close - close the channel
 792  *      @chan: the channel
 793  *
 794  *      Closes all channel buffers and frees the channel.
 795  */
 796 void relay_close(struct rchan *chan)
 797 {
 798         struct rchan_buf *buf;
 799         unsigned int i;
 800
 801         if (!chan)
 802                 return;
 803
 804         mutex_lock(&relay_channels_mutex);
 805         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0)))
 806                 relay_close_buf(buf);
 807         else
 808                 for_each_possible_cpu(i)
 809                         if ((buf = *per_cpu_ptr(chan->buf, i)))
 810                                 relay_close_buf(buf);
 811
 812         if (chan->last_toobig)
 813                 printk(KERN_WARNING "relay: one or more items not logged "
 814                        "[item size (%zd) > sub-buffer size (%zd)]\n",
 815                        chan->last_toobig, chan->subbuf_size);
 816
 817         list_del(&chan->list);
 818         kref_put(&chan->kref, relay_destroy_channel);
 819         mutex_unlock(&relay_channels_mutex);
 820 }
 821 EXPORT_SYMBOL_GPL(relay_close);
 822
 823 /**
 824  *      relay_flush - close the channel
 825  *      @chan: the channel
 826  *
 827  *      Flushes all channel buffers, i.e. forces buffer switch.
 828  */
 829 void relay_flush(struct rchan *chan)
 830 {
 831         struct rchan_buf *buf;
 832         unsigned int i;
 833
 834         if (!chan)
 835                 return;
 836
 837         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
 838                 relay_switch_subbuf(buf, 0);
 839                 return;
 840         }
 841
 842         mutex_lock(&relay_channels_mutex);
 843         for_each_possible_cpu(i)
 844                 if ((buf = *per_cpu_ptr(chan->buf, i)))
 845                         relay_switch_subbuf(buf, 0);
 846         mutex_unlock(&relay_channels_mutex);
 847 }
 848 EXPORT_SYMBOL_GPL(relay_flush);
 849
 850 /**
 851  *      relay_file_open - open file op for relay files
 852  *      @inode: the inode
 853  *      @filp: the file
 854  *
 855  *      Increments the channel buffer refcount.
 856  */
 857 static int relay_file_open(struct inode *inode, struct file *filp)
 858 {
 859         struct rchan_buf *buf = inode->i_private;
 860         kref_get(&buf->kref);
 861         filp->private_data = buf;
 862
 863         return nonseekable_open(inode, filp);
 864 }
 865
 866 /**
 867  *      relay_file_mmap - mmap file op for relay files
 868  *      @filp: the file
 869  *      @vma: the vma describing what to map
 870  *
 871  *      Calls upon relay_mmap_buf() to map the file into user space.
 872  */
 873 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
 874 {
 875         struct rchan_buf *buf = filp->private_data;
 876         return relay_mmap_buf(buf, vma);
 877 }
 878
 879 /**
 880  *      relay_file_poll - poll file op for relay files
 881  *      @filp: the file
 882  *      @wait: poll table
 883  *
 884  *      Poll implemention.
 885  */
 886 static __poll_t relay_file_poll(struct file *filp, poll_table *wait)
 887 {
 888         __poll_t mask = 0;
 889         struct rchan_buf *buf = filp->private_data;
 890
 891         if (buf->finalized)
 892                 return EPOLLERR;
 893
 894         if (filp->f_mode & FMODE_READ) {
 895                 poll_wait(filp, &buf->read_wait, wait);
 896                 if (!relay_buf_empty(buf))
 897                         mask |= EPOLLIN | EPOLLRDNORM;
 898         }
 899
 900         return mask;
 901 }
 902
 903 /**
 904  *      relay_file_release - release file op for relay files
 905  *      @inode: the inode
 906  *      @filp: the file
 907  *
 908  *      Decrements the channel refcount, as the filesystem is
 909  *      no longer using it.
 910  */
 911 static int relay_file_release(struct inode *inode, struct file *filp)
 912 {
 913         struct rchan_buf *buf = filp->private_data;
 914         kref_put(&buf->kref, relay_remove_buf);
 915
 916         return 0;
 917 }
 918
 919 /*
 920  *      relay_file_read_consume - update the consumed count for the buffer
 921  */
 922 static void relay_file_read_consume(struct rchan_buf *buf,
 923                                     size_t read_pos,
 924                                     size_t bytes_consumed)
 925 {
 926         size_t subbuf_size = buf->chan->subbuf_size;
 927         size_t n_subbufs = buf->chan->n_subbufs;
 928         size_t read_subbuf;
 929
 930         if (buf->subbufs_produced == buf->subbufs_consumed &&
 931             buf->offset == buf->bytes_consumed)
 932                 return;
 933
 934         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
 935                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 936                 buf->bytes_consumed = 0;
 937         }
 938
 939         buf->bytes_consumed += bytes_consumed;
 940         if (!read_pos)
 941                 read_subbuf = buf->subbufs_consumed % n_subbufs;
 942         else
 943                 read_subbuf = read_pos / buf->chan->subbuf_size;
 944         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
 945                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
 946                     (buf->offset == subbuf_size))
 947                         return;
 948                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
 949                 buf->bytes_consumed = 0;
 950         }
 951 }
 952
 953 /*
 954  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
 955  */
 956 static int relay_file_read_avail(struct rchan_buf *buf)
 957 {
 958         size_t subbuf_size = buf->chan->subbuf_size;
 959         size_t n_subbufs = buf->chan->n_subbufs;
 960         size_t produced = buf->subbufs_produced;
 961         size_t consumed;
 962
 963         relay_file_read_consume(buf, 0, 0);
 964
 965         consumed = buf->subbufs_consumed;
 966
 967         if (unlikely(buf->offset > subbuf_size)) {
 968                 if (produced == consumed)
 969                         return 0;
 970                 return 1;
 971         }
 972
 973         if (unlikely(produced - consumed >= n_subbufs)) {
 974                 consumed = produced - n_subbufs + 1;
 975                 buf->subbufs_consumed = consumed;
 976                 buf->bytes_consumed = 0;
 977         }
 978
 979         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
 980         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
 981
 982         if (consumed > produced)
 983                 produced += n_subbufs * subbuf_size;
 984
 985         if (consumed == produced) {
 986                 if (buf->offset == subbuf_size &&
 987                     buf->subbufs_produced > buf->subbufs_consumed)
 988                         return 1;
 989                 return 0;
 990         }
 991
 992         return 1;
 993 }
 994
 995 /**
 996  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
 997  *      @read_pos: file read position
 998  *      @buf: relay channel buffer
 999  */
1000 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1001                                            struct rchan_buf *buf)
1002 {
1003         size_t padding, avail = 0;
1004         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1005         size_t subbuf_size = buf->chan->subbuf_size;
1006
1007         write_subbuf = (buf->data - buf->start) / subbuf_size;
1008         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1009         read_subbuf = read_pos / subbuf_size;
1010         read_offset = read_pos % subbuf_size;
1011         padding = buf->padding[read_subbuf];
1012
1013         if (read_subbuf == write_subbuf) {
1014                 if (read_offset + padding < write_offset)
1015                         avail = write_offset - (read_offset + padding);
1016         } else
1017                 avail = (subbuf_size - padding) - read_offset;
1018
1019         return avail;
1020 }
1021
1022 /**
1023  *      relay_file_read_start_pos - find the first available byte to read
1024  *      @buf: relay channel buffer
1025  *
1026  *      If the read_pos is in the middle of padding, return the
1027  *      position of the first actually available byte, otherwise
1028  *      return the original value.
1029  */
1030 static size_t relay_file_read_start_pos(struct rchan_buf *buf)
1031 {
1032         size_t read_subbuf, padding, padding_start, padding_end;
1033         size_t subbuf_size = buf->chan->subbuf_size;
1034         size_t n_subbufs = buf->chan->n_subbufs;
1035         size_t consumed = buf->subbufs_consumed % n_subbufs;
1036         size_t read_pos = consumed * subbuf_size + buf->bytes_consumed;
1037
1038         read_subbuf = read_pos / subbuf_size;
1039         padding = buf->padding[read_subbuf];
1040         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1041         padding_end = (read_subbuf + 1) * subbuf_size;
1042         if (read_pos >= padding_start && read_pos < padding_end) {
1043                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1044                 read_pos = read_subbuf * subbuf_size;
1045         }
1046
1047         return read_pos;
1048 }
1049
1050 /**
1051  *      relay_file_read_end_pos - return the new read position
1052  *      @read_pos: file read position
1053  *      @buf: relay channel buffer
1054  *      @count: number of bytes to be read
1055  */
1056 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1057                                       size_t read_pos,
1058                                       size_t count)
1059 {
1060         size_t read_subbuf, padding, end_pos;
1061         size_t subbuf_size = buf->chan->subbuf_size;
1062         size_t n_subbufs = buf->chan->n_subbufs;
1063
1064         read_subbuf = read_pos / subbuf_size;
1065         padding = buf->padding[read_subbuf];
1066         if (read_pos % subbuf_size + count + padding == subbuf_size)
1067                 end_pos = (read_subbuf + 1) * subbuf_size;
1068         else
1069                 end_pos = read_pos + count;
1070         if (end_pos >= subbuf_size * n_subbufs)
1071                 end_pos = 0;
1072
1073         return end_pos;
1074 }
1075
1076 static ssize_t relay_file_read(struct file *filp,
1077                                char __user *buffer,
1078                                size_t count,
1079                                loff_t *ppos)
1080 {
1081         struct rchan_buf *buf = filp->private_data;
1082         size_t read_start, avail;
1083         size_t written = 0;
1084         int ret;
1085
1086         if (!count)
1087                 return 0;
1088
1089         inode_lock(file_inode(filp));
1090         do {
1091                 void *from;
1092
1093                 if (!relay_file_read_avail(buf))
1094                         break;
1095
1096                 read_start = relay_file_read_start_pos(buf);
1097                 avail = relay_file_read_subbuf_avail(read_start, buf);
1098                 if (!avail)
1099                         break;
1100
1101                 avail = min(count, avail);
1102                 from = buf->start + read_start;
1103                 ret = avail;
1104                 if (copy_to_user(buffer, from, avail))
1105                         break;
1106
1107                 buffer += ret;
1108                 written += ret;
1109                 count -= ret;
1110
1111                 relay_file_read_consume(buf, read_start, ret);
1112                 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1113         } while (count);
1114         inode_unlock(file_inode(filp));
1115
1116         return written;
1117 }
1118
1119 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1120 {
1121         rbuf->bytes_consumed += bytes_consumed;
1122
1123         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1124                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1125                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1126         }
1127 }
1128
1129 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1130                                    struct pipe_buffer *buf)
1131 {
1132         struct rchan_buf *rbuf;
1133
1134         rbuf = (struct rchan_buf *)page_private(buf->page);
1135         relay_consume_bytes(rbuf, buf->private);
1136 }
1137
1138 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1139         .release        = relay_pipe_buf_release,
1140         .try_steal      = generic_pipe_buf_try_steal,
1141         .get            = generic_pipe_buf_get,
1142 };
1143
1144 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1145 {
1146 }
1147
1148 /*
1149  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1150  */
1151 static ssize_t subbuf_splice_actor(struct file *in,
1152                                loff_t *ppos,
1153                                struct pipe_inode_info *pipe,
1154                                size_t len,
1155                                unsigned int flags,
1156                                int *nonpad_ret)
1157 {
1158         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1159         struct rchan_buf *rbuf = in->private_data;
1160         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1161         uint64_t pos = (uint64_t) *ppos;
1162         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1163         size_t read_start = (size_t) do_div(pos, alloc_size);
1164         size_t read_subbuf = read_start / subbuf_size;
1165         size_t padding = rbuf->padding[read_subbuf];
1166         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1167         struct page *pages[PIPE_DEF_BUFFERS];
1168         struct partial_page partial[PIPE_DEF_BUFFERS];
1169         struct splice_pipe_desc spd = {
1170                 .pages = pages,
1171                 .nr_pages = 0,
1172                 .nr_pages_max = PIPE_DEF_BUFFERS,
1173                 .partial = partial,
1174                 .ops = &relay_pipe_buf_ops,
1175                 .spd_release = relay_page_release,
1176         };
1177         ssize_t ret;
1178
1179         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1180                 return 0;
1181         if (splice_grow_spd(pipe, &spd))
1182                 return -ENOMEM;
1183
1184         /*
1185          * Adjust read len, if longer than what is available
1186          */
1187         if (len > (subbuf_size - read_start % subbuf_size))
1188                 len = subbuf_size - read_start % subbuf_size;
1189
1190         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1191         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1192         poff = read_start & ~PAGE_MASK;
1193         nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1194
1195         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1196                 unsigned int this_len, this_end, private;
1197                 unsigned int cur_pos = read_start + total_len;
1198
1199                 if (!len)
1200                         break;
1201
1202                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1203                 private = this_len;
1204
1205                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1206                 spd.partial[spd.nr_pages].offset = poff;
1207
1208                 this_end = cur_pos + this_len;
1209                 if (this_end >= nonpad_end) {
1210                         this_len = nonpad_end - cur_pos;
1211                         private = this_len + padding;
1212                 }
1213                 spd.partial[spd.nr_pages].len = this_len;
1214                 spd.partial[spd.nr_pages].private = private;
1215
1216                 len -= this_len;
1217                 total_len += this_len;
1218                 poff = 0;
1219                 pidx = (pidx + 1) % subbuf_pages;
1220
1221                 if (this_end >= nonpad_end) {
1222                         spd.nr_pages++;
1223                         break;
1224                 }
1225         }
1226
1227         ret = 0;
1228         if (!spd.nr_pages)
1229                 goto out;
1230
1231         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1232         if (ret < 0 || ret < total_len)
1233                 goto out;
1234
1235         if (read_start + ret == nonpad_end)
1236                 ret += padding;
1237
1238 out:
1239         splice_shrink_spd(&spd);
1240         return ret;
1241 }
1242
1243 static ssize_t relay_file_splice_read(struct file *in,
1244                                       loff_t *ppos,
1245                                       struct pipe_inode_info *pipe,
1246                                       size_t len,
1247                                       unsigned int flags)
1248 {
1249         ssize_t spliced;
1250         int ret;
1251         int nonpad_ret = 0;
1252
1253         ret = 0;
1254         spliced = 0;
1255
1256         while (len && !spliced) {
1257                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1258                 if (ret < 0)
1259                         break;
1260                 else if (!ret) {
1261                         if (flags & SPLICE_F_NONBLOCK)
1262                                 ret = -EAGAIN;
1263                         break;
1264                 }
1265
1266                 *ppos += ret;
1267                 if (ret > len)
1268                         len = 0;
1269                 else
1270                         len -= ret;
1271                 spliced += nonpad_ret;
1272                 nonpad_ret = 0;
1273         }
1274
1275         if (spliced)
1276                 return spliced;
1277
1278         return ret;
1279 }
1280
1281 const struct file_operations relay_file_operations = {
1282         .open           = relay_file_open,
1283         .poll           = relay_file_poll,
1284         .mmap           = relay_file_mmap,
1285         .read           = relay_file_read,
1286         .llseek         = no_llseek,
1287         .release        = relay_file_release,
1288         .splice_read    = relay_file_splice_read,
1289 };
1290 EXPORT_SYMBOL_GPL(relay_file_operations);