lib/ringbuffer/ring_buffer_iterator.c

   1 /* SPDX-License-Identifier: (GPL-2.0-only OR LGPL-2.1-only)
   2  *
   3  * ring_buffer_iterator.c
   4  *
   5  * Ring buffer and channel iterators. Get each event of a channel in order. Uses
   6  * a prio heap for per-cpu buffers, giving a O(log(NR_CPUS)) algorithmic
   7  * complexity for the "get next event" operation.
   8  *
   9  * Copyright (C) 2010-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  10  */
  11
  12 #include <ringbuffer/iterator.h>
  13 #include <linux/file.h>
  14 #include <linux/uaccess.h>
  15 #include <linux/jiffies.h>
  16 #include <linux/delay.h>
  17 #include <linux/module.h>
  18
  19 /*
  20  * Safety factor taking into account internal kernel interrupt latency.
  21  * Assuming 250ms worse-case latency.
  22  */
  23 #define MAX_SYSTEM_LATENCY      250
  24
  25 /*
  26  * Maximum delta expected between trace clocks. At most 1 jiffy delta.
  27  */
  28 #define MAX_CLOCK_DELTA         (jiffies_to_usecs(1) * 1000)
  29
  30 /**
  31  * lib_ring_buffer_get_next_record - Get the next record in a buffer.
  32  * @chan: channel
  33  * @buf: buffer
  34  *
  35  * Returns the size of the event read, -EAGAIN if buffer is empty, -ENODATA if
  36  * buffer is empty and finalized. The buffer must already be opened for reading.
  37  */
  38 ssize_t lib_ring_buffer_get_next_record(struct channel *chan,
  39                                         struct lib_ring_buffer *buf)
  40 {
  41         const struct lib_ring_buffer_config *config = &chan->backend.config;
  42         struct lib_ring_buffer_iter *iter = &buf->iter;
  43         int ret;
  44
  45 restart:
  46         switch (iter->state) {
  47         case ITER_GET_SUBBUF:
  48                 ret = lib_ring_buffer_get_next_subbuf(buf);
  49                 if (ret && !READ_ONCE(buf->finalized)
  50                     && config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
  51                         /*
  52                          * Use "pull" scheme for global buffers. The reader
  53                          * itself flushes the buffer to "pull" data not visible
  54                          * to readers yet. Flush current subbuffer and re-try.
  55                          *
  56                          * Per-CPU buffers rather use a "push" scheme because
  57                          * the IPI needed to flush all CPU's buffers is too
  58                          * costly. In the "push" scheme, the reader waits for
  59                          * the writer periodic timer to flush the
  60                          * buffers (keeping track of a quiescent state
  61                          * timestamp). Therefore, the writer "pushes" data out
  62                          * of the buffers rather than letting the reader "pull"
  63                          * data from the buffer.
  64                          */
  65                         lib_ring_buffer_switch_slow(buf, SWITCH_ACTIVE);
  66                         ret = lib_ring_buffer_get_next_subbuf(buf);
  67                 }
  68                 if (ret)
  69                         return ret;
  70                 iter->consumed = buf->cons_snapshot;
  71                 iter->data_size = lib_ring_buffer_get_read_data_size(config, buf);
  72                 iter->read_offset = iter->consumed;
  73                 /* skip header */
  74                 iter->read_offset += config->cb.subbuffer_header_size();
  75                 iter->state = ITER_TEST_RECORD;
  76                 goto restart;
  77         case ITER_TEST_RECORD:
  78                 if (iter->read_offset - iter->consumed >= iter->data_size) {
  79                         iter->state = ITER_PUT_SUBBUF;
  80                 } else {
  81                         CHAN_WARN_ON(chan, !config->cb.record_get);
  82                         config->cb.record_get(config, chan, buf,
  83                                               iter->read_offset,
  84                                               &iter->header_len,
  85                                               &iter->payload_len,
  86                                               &iter->timestamp);
  87                         iter->read_offset += iter->header_len;
  88                         subbuffer_consume_record(config, &buf->backend);
  89                         iter->state = ITER_NEXT_RECORD;
  90                         return iter->payload_len;
  91                 }
  92                 goto restart;
  93         case ITER_NEXT_RECORD:
  94                 iter->read_offset += iter->payload_len;
  95                 iter->state = ITER_TEST_RECORD;
  96                 goto restart;
  97         case ITER_PUT_SUBBUF:
  98                 lib_ring_buffer_put_next_subbuf(buf);
  99                 iter->state = ITER_GET_SUBBUF;
 100                 goto restart;
 101         default:
 102                 CHAN_WARN_ON(chan, 1);  /* Should not happen */
 103                 return -EPERM;
 104         }
 105 }
 106 EXPORT_SYMBOL_GPL(lib_ring_buffer_get_next_record);
 107
 108 static int buf_is_higher(void *a, void *b)
 109 {
 110         struct lib_ring_buffer *bufa = a;
 111         struct lib_ring_buffer *bufb = b;
 112
 113         /* Consider lowest timestamps to be at the top of the heap */
 114         return (bufa->iter.timestamp < bufb->iter.timestamp);
 115 }
 116
 117 static
 118 void lib_ring_buffer_get_empty_buf_records(const struct lib_ring_buffer_config *config,
 119                                            struct channel *chan)
 120 {
 121         struct lttng_ptr_heap *heap = &chan->iter.heap;
 122         struct lib_ring_buffer *buf, *tmp;
 123         ssize_t len;
 124
 125         list_for_each_entry_safe(buf, tmp, &chan->iter.empty_head,
 126                                  iter.empty_node) {
 127                 len = lib_ring_buffer_get_next_record(chan, buf);
 128
 129                 /*
 130                  * Deal with -EAGAIN and -ENODATA.
 131                  * len >= 0 means record contains data.
 132                  * -EBUSY should never happen, because we support only one
 133                  * reader.
 134                  */
 135                 switch (len) {
 136                 case -EAGAIN:
 137                         /* Keep node in empty list */
 138                         break;
 139                 case -ENODATA:
 140                         /*
 141                          * Buffer is finalized. Don't add to list of empty
 142                          * buffer, because it has no more data to provide, ever.
 143                          */
 144                         list_del(&buf->iter.empty_node);
 145                         break;
 146                 case -EBUSY:
 147                         CHAN_WARN_ON(chan, 1);
 148                         break;
 149                 default:
 150                         /*
 151                          * Insert buffer into the heap, remove from empty buffer
 152                          * list.
 153                          */
 154                         CHAN_WARN_ON(chan, len < 0);
 155                         list_del(&buf->iter.empty_node);
 156                         CHAN_WARN_ON(chan, lttng_heap_insert(heap, buf));
 157                 }
 158         }
 159 }
 160
 161 static
 162 void lib_ring_buffer_wait_for_qs(const struct lib_ring_buffer_config *config,
 163                                  struct channel *chan)
 164 {
 165         u64 timestamp_qs;
 166         unsigned long wait_msecs;
 167
 168         /*
 169          * No need to wait if no empty buffers are present.
 170          */
 171         if (list_empty(&chan->iter.empty_head))
 172                 return;
 173
 174         timestamp_qs = config->cb.ring_buffer_clock_read(chan);
 175         /*
 176          * We need to consider previously empty buffers.
 177          * Do a get next buf record on each of them. Add them to
 178          * the heap if they have data. If at least one of them
 179          * don't have data, we need to wait for
 180          * switch_timer_interval + MAX_SYSTEM_LATENCY (so we are sure the
 181          * buffers have been switched either by the timer or idle entry) and
 182          * check them again, adding them if they have data.
 183          */
 184         lib_ring_buffer_get_empty_buf_records(config, chan);
 185
 186         /*
 187          * No need to wait if no empty buffers are present.
 188          */
 189         if (list_empty(&chan->iter.empty_head))
 190                 return;
 191
 192         /*
 193          * We need to wait for the buffer switch timer to run. If the
 194          * CPU is idle, idle entry performed the switch.
 195          * TODO: we could optimize further by skipping the sleep if all
 196          * empty buffers belong to idle or offline cpus.
 197          */
 198         wait_msecs = jiffies_to_msecs(chan->switch_timer_interval);
 199         wait_msecs += MAX_SYSTEM_LATENCY;
 200         msleep(wait_msecs);
 201         lib_ring_buffer_get_empty_buf_records(config, chan);
 202         /*
 203          * Any buffer still in the empty list here cannot possibly
 204          * contain an event with a timestamp prior to "timestamp_qs".
 205          * The new quiescent state timestamp is the one we grabbed
 206          * before waiting for buffer data.  It is therefore safe to
 207          * ignore empty buffers up to last_qs timestamp for fusion
 208          * merge.
 209          */
 210         chan->iter.last_qs = timestamp_qs;
 211 }
 212
 213 /**
 214  * channel_get_next_record - Get the next record in a channel.
 215  * @chan: channel
 216  * @ret_buf: the buffer in which the event is located (output)
 217  *
 218  * Returns the size of new current event, -EAGAIN if all buffers are empty,
 219  * -ENODATA if all buffers are empty and finalized. The channel must already be
 220  * opened for reading.
 221  */
 222
 223 ssize_t channel_get_next_record(struct channel *chan,
 224                                 struct lib_ring_buffer **ret_buf)
 225 {
 226         const struct lib_ring_buffer_config *config = &chan->backend.config;
 227         struct lib_ring_buffer *buf;
 228         struct lttng_ptr_heap *heap;
 229         ssize_t len;
 230
 231         if (config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
 232                 *ret_buf = channel_get_ring_buffer(config, chan, 0);
 233                 return lib_ring_buffer_get_next_record(chan, *ret_buf);
 234         }
 235
 236         heap = &chan->iter.heap;
 237
 238         /*
 239          * get next record for topmost buffer.
 240          */
 241         buf = lttng_heap_maximum(heap);
 242         if (buf) {
 243                 len = lib_ring_buffer_get_next_record(chan, buf);
 244                 /*
 245                  * Deal with -EAGAIN and -ENODATA.
 246                  * len >= 0 means record contains data.
 247                  */
 248                 switch (len) {
 249                 case -EAGAIN:
 250                         buf->iter.timestamp = 0;
 251                         list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 252                         /* Remove topmost buffer from the heap */
 253                         CHAN_WARN_ON(chan, lttng_heap_remove(heap) != buf);
 254                         break;
 255                 case -ENODATA:
 256                         /*
 257                          * Buffer is finalized. Remove buffer from heap and
 258                          * don't add to list of empty buffer, because it has no
 259                          * more data to provide, ever.
 260                          */
 261                         CHAN_WARN_ON(chan, lttng_heap_remove(heap) != buf);
 262                         break;
 263                 case -EBUSY:
 264                         CHAN_WARN_ON(chan, 1);
 265                         break;
 266                 default:
 267                         /*
 268                          * Reinsert buffer into the heap. Note that heap can be
 269                          * partially empty, so we need to use
 270                          * lttng_heap_replace_max().
 271                          */
 272                         CHAN_WARN_ON(chan, len < 0);
 273                         CHAN_WARN_ON(chan, lttng_heap_replace_max(heap, buf) != buf);
 274                         break;
 275                 }
 276         }
 277
 278         buf = lttng_heap_maximum(heap);
 279         if (!buf || buf->iter.timestamp > chan->iter.last_qs) {
 280                 /*
 281                  * Deal with buffers previously showing no data.
 282                  * Add buffers containing data to the heap, update
 283                  * last_qs.
 284                  */
 285                 lib_ring_buffer_wait_for_qs(config, chan);
 286         }
 287
 288         *ret_buf = buf = lttng_heap_maximum(heap);
 289         if (buf) {
 290                 /*
 291                  * If this warning triggers, you probably need to check your
 292                  * system interrupt latency. Typical causes: too many printk()
 293                  * output going to a serial console with interrupts off.
 294                  * Allow for MAX_CLOCK_DELTA ns timestamp delta going backward.
 295                  * Observed on SMP KVM setups with trace_clock().
 296                  */
 297                 if (chan->iter.last_timestamp
 298                     > (buf->iter.timestamp + MAX_CLOCK_DELTA)) {
 299                         printk(KERN_WARNING "ring_buffer: timestamps going "
 300                                "backward. Last time %llu ns, cpu %d, "
 301                                "current time %llu ns, cpu %d, "
 302                                "delta %llu ns.\n",
 303                                chan->iter.last_timestamp, chan->iter.last_cpu,
 304                                buf->iter.timestamp, buf->backend.cpu,
 305                                chan->iter.last_timestamp - buf->iter.timestamp);
 306                         CHAN_WARN_ON(chan, 1);
 307                 }
 308                 chan->iter.last_timestamp = buf->iter.timestamp;
 309                 chan->iter.last_cpu = buf->backend.cpu;
 310                 return buf->iter.payload_len;
 311         } else {
 312                 /* Heap is empty */
 313                 if (list_empty(&chan->iter.empty_head))
 314                         return -ENODATA;        /* All buffers finalized */
 315                 else
 316                         return -EAGAIN;         /* Temporarily empty */
 317         }
 318 }
 319 EXPORT_SYMBOL_GPL(channel_get_next_record);
 320
 321 static
 322 void lib_ring_buffer_iterator_init(struct channel *chan, struct lib_ring_buffer *buf)
 323 {
 324         if (buf->iter.allocated)
 325                 return;
 326
 327         buf->iter.allocated = 1;
 328         if (chan->iter.read_open && !buf->iter.read_open) {
 329                 CHAN_WARN_ON(chan, lib_ring_buffer_open_read(buf) != 0);
 330                 buf->iter.read_open = 1;
 331         }
 332
 333         /* Add to list of buffers without any current record */
 334         if (chan->backend.config.alloc == RING_BUFFER_ALLOC_PER_CPU)
 335                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 336 }
 337
 338 int lttng_cpuhp_rb_iter_online(unsigned int cpu,
 339                 struct lttng_cpuhp_node *node)
 340 {
 341         struct channel *chan = container_of(node, struct channel,
 342                                             cpuhp_iter_online);
 343         struct lib_ring_buffer *buf = per_cpu_ptr(chan->backend.buf, cpu);
 344         const struct lib_ring_buffer_config *config = &chan->backend.config;
 345
 346         CHAN_WARN_ON(chan, config->alloc == RING_BUFFER_ALLOC_GLOBAL);
 347
 348         lib_ring_buffer_iterator_init(chan, buf);
 349         return 0;
 350 }
 351 EXPORT_SYMBOL_GPL(lttng_cpuhp_rb_iter_online);
 352
 353 int channel_iterator_init(struct channel *chan)
 354 {
 355         const struct lib_ring_buffer_config *config = &chan->backend.config;
 356         struct lib_ring_buffer *buf;
 357
 358         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 359                 int ret;
 360
 361                 INIT_LIST_HEAD(&chan->iter.empty_head);
 362                 ret = lttng_heap_init(&chan->iter.heap,
 363                                 num_possible_cpus(),
 364                                 GFP_KERNEL, buf_is_higher);
 365                 if (ret)
 366                         return ret;
 367
 368                 chan->cpuhp_iter_online.component = LTTNG_RING_BUFFER_ITER;
 369                 ret = cpuhp_state_add_instance(lttng_rb_hp_online,
 370                         &chan->cpuhp_iter_online.node);
 371                 if (ret)
 372                         return ret;
 373         } else {
 374                 buf = channel_get_ring_buffer(config, chan, 0);
 375                 lib_ring_buffer_iterator_init(chan, buf);
 376         }
 377         return 0;
 378 }
 379
 380 void channel_iterator_unregister_notifiers(struct channel *chan)
 381 {
 382         const struct lib_ring_buffer_config *config = &chan->backend.config;
 383
 384         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 385                 {
 386                         int ret;
 387
 388                         ret = cpuhp_state_remove_instance(lttng_rb_hp_online,
 389                                 &chan->cpuhp_iter_online.node);
 390                         WARN_ON(ret);
 391                 }
 392         }
 393 }
 394
 395 void channel_iterator_free(struct channel *chan)
 396 {
 397         const struct lib_ring_buffer_config *config = &chan->backend.config;
 398
 399         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
 400                 lttng_heap_free(&chan->iter.heap);
 401 }
 402
 403 int lib_ring_buffer_iterator_open(struct lib_ring_buffer *buf)
 404 {
 405         struct channel *chan = buf->backend.chan;
 406         const struct lib_ring_buffer_config *config = &chan->backend.config;
 407         CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR);
 408         return lib_ring_buffer_open_read(buf);
 409 }
 410 EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_open);
 411
 412 /*
 413  * Note: Iterators must not be mixed with other types of outputs, because an
 414  * iterator can leave the buffer in "GET" state, which is not consistent with
 415  * other types of output (mmap, splice, raw data read).
 416  */
 417 void lib_ring_buffer_iterator_release(struct lib_ring_buffer *buf)
 418 {
 419         lib_ring_buffer_release_read(buf);
 420 }
 421 EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_release);
 422
 423 int channel_iterator_open(struct channel *chan)
 424 {
 425         const struct lib_ring_buffer_config *config = &chan->backend.config;
 426         struct lib_ring_buffer *buf;
 427         int ret = 0, cpu;
 428
 429         CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR);
 430
 431         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 432                 get_online_cpus();
 433                 /* Allow CPU hotplug to keep track of opened reader */
 434                 chan->iter.read_open = 1;
 435                 for_each_channel_cpu(cpu, chan) {
 436                         buf = channel_get_ring_buffer(config, chan, cpu);
 437                         ret = lib_ring_buffer_iterator_open(buf);
 438                         if (ret)
 439                                 goto error;
 440                         buf->iter.read_open = 1;
 441                 }
 442                 put_online_cpus();
 443         } else {
 444                 buf = channel_get_ring_buffer(config, chan, 0);
 445                 ret = lib_ring_buffer_iterator_open(buf);
 446         }
 447         return ret;
 448 error:
 449         /* Error should always happen on CPU 0, hence no close is required. */
 450         CHAN_WARN_ON(chan, cpu != 0);
 451         put_online_cpus();
 452         return ret;
 453 }
 454 EXPORT_SYMBOL_GPL(channel_iterator_open);
 455
 456 void channel_iterator_release(struct channel *chan)
 457 {
 458         const struct lib_ring_buffer_config *config = &chan->backend.config;
 459         struct lib_ring_buffer *buf;
 460         int cpu;
 461
 462         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 463                 get_online_cpus();
 464                 for_each_channel_cpu(cpu, chan) {
 465                         buf = channel_get_ring_buffer(config, chan, cpu);
 466                         if (buf->iter.read_open) {
 467                                 lib_ring_buffer_iterator_release(buf);
 468                                 buf->iter.read_open = 0;
 469                         }
 470                 }
 471                 chan->iter.read_open = 0;
 472                 put_online_cpus();
 473         } else {
 474                 buf = channel_get_ring_buffer(config, chan, 0);
 475                 lib_ring_buffer_iterator_release(buf);
 476         }
 477 }
 478 EXPORT_SYMBOL_GPL(channel_iterator_release);
 479
 480 void lib_ring_buffer_iterator_reset(struct lib_ring_buffer *buf)
 481 {
 482         struct channel *chan = buf->backend.chan;
 483
 484         if (buf->iter.state != ITER_GET_SUBBUF)
 485                 lib_ring_buffer_put_next_subbuf(buf);
 486         buf->iter.state = ITER_GET_SUBBUF;
 487         /* Remove from heap (if present). */
 488         if (lttng_heap_cherrypick(&chan->iter.heap, buf))
 489                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 490         buf->iter.timestamp = 0;
 491         buf->iter.header_len = 0;
 492         buf->iter.payload_len = 0;
 493         buf->iter.consumed = 0;
 494         buf->iter.read_offset = 0;
 495         buf->iter.data_size = 0;
 496         /* Don't reset allocated and read_open */
 497 }
 498
 499 void channel_iterator_reset(struct channel *chan)
 500 {
 501         const struct lib_ring_buffer_config *config = &chan->backend.config;
 502         struct lib_ring_buffer *buf;
 503         int cpu;
 504
 505         /* Empty heap, put into empty_head */
 506         while ((buf = lttng_heap_remove(&chan->iter.heap)) != NULL)
 507                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 508
 509         for_each_channel_cpu(cpu, chan) {
 510                 buf = channel_get_ring_buffer(config, chan, cpu);
 511                 lib_ring_buffer_iterator_reset(buf);
 512         }
 513         /* Don't reset read_open */
 514         chan->iter.last_qs = 0;
 515         chan->iter.last_timestamp = 0;
 516         chan->iter.last_cpu = 0;
 517         chan->iter.len_left = 0;
 518 }
 519
 520 /*
 521  * Ring buffer payload extraction read() implementation.
 522  */
 523 static
 524 ssize_t channel_ring_buffer_file_read(struct file *filp,
 525                                       char __user *user_buf,
 526                                       size_t count,
 527                                       loff_t *ppos,
 528                                       struct channel *chan,
 529                                       struct lib_ring_buffer *buf,
 530                                       int fusionmerge)
 531 {
 532         const struct lib_ring_buffer_config *config = &chan->backend.config;
 533         size_t read_count = 0, read_offset;
 534         ssize_t len;
 535
 536         might_sleep();
 537         if (!access_ok(user_buf, count))
 538                 return -EFAULT;
 539
 540         /* Finish copy of previous record */
 541         if (*ppos != 0) {
 542                 if (read_count < count) {
 543                         len = chan->iter.len_left;
 544                         read_offset = *ppos;
 545                         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU
 546                             && fusionmerge)
 547                                 buf = lttng_heap_maximum(&chan->iter.heap);
 548                         CHAN_WARN_ON(chan, !buf);
 549                         goto skip_get_next;
 550                 }
 551         }
 552
 553         while (read_count < count) {
 554                 size_t copy_len, space_left;
 555
 556                 if (fusionmerge)
 557                         len = channel_get_next_record(chan, &buf);
 558                 else
 559                         len = lib_ring_buffer_get_next_record(chan, buf);
 560 len_test:
 561                 if (len < 0) {
 562                         /*
 563                          * Check if buffer is finalized (end of file).
 564                          */
 565                         if (len == -ENODATA) {
 566                                 /* A 0 read_count will tell about end of file */
 567                                 goto nodata;
 568                         }
 569                         if (filp->f_flags & O_NONBLOCK) {
 570                                 if (!read_count)
 571                                         read_count = -EAGAIN;
 572                                 goto nodata;
 573                         } else {
 574                                 int error;
 575
 576                                 /*
 577                                  * No data available at the moment, return what
 578                                  * we got.
 579                                  */
 580                                 if (read_count)
 581                                         goto nodata;
 582
 583                                 /*
 584                                  * Wait for returned len to be >= 0 or -ENODATA.
 585                                  */
 586                                 if (fusionmerge)
 587                                         error = wait_event_interruptible(
 588                                           chan->read_wait,
 589                                           ((len = channel_get_next_record(chan,
 590                                                 &buf)), len != -EAGAIN));
 591                                 else
 592                                         error = wait_event_interruptible(
 593                                           buf->read_wait,
 594                                           ((len = lib_ring_buffer_get_next_record(
 595                                                   chan, buf)), len != -EAGAIN));
 596                                 CHAN_WARN_ON(chan, len == -EBUSY);
 597                                 if (error) {
 598                                         read_count = error;
 599                                         goto nodata;
 600                                 }
 601                                 CHAN_WARN_ON(chan, len < 0 && len != -ENODATA);
 602                                 goto len_test;
 603                         }
 604                 }
 605                 read_offset = buf->iter.read_offset;
 606 skip_get_next:
 607                 space_left = count - read_count;
 608                 if (len <= space_left) {
 609                         copy_len = len;
 610                         chan->iter.len_left = 0;
 611                         *ppos = 0;
 612                 } else {
 613                         copy_len = space_left;
 614                         chan->iter.len_left = len - copy_len;
 615                         *ppos = read_offset + copy_len;
 616                 }
 617                 if (__lib_ring_buffer_copy_to_user(&buf->backend, read_offset,
 618                                                &user_buf[read_count],
 619                                                copy_len)) {
 620                         /*
 621                          * Leave the len_left and ppos values at their current
 622                          * state, as we currently have a valid event to read.
 623                          */
 624                         return -EFAULT;
 625                 }
 626                 read_count += copy_len;
 627         };
 628         return read_count;
 629
 630 nodata:
 631         *ppos = 0;
 632         chan->iter.len_left = 0;
 633         return read_count;
 634 }
 635
 636 /**
 637  * lib_ring_buffer_file_read - Read buffer record payload.
 638  * @filp: file structure pointer.
 639  * @buffer: user buffer to read data into.
 640  * @count: number of bytes to read.
 641  * @ppos: file read position.
 642  *
 643  * Returns a negative value on error, or the number of bytes read on success.
 644  * ppos is used to save the position _within the current record_ between calls
 645  * to read().
 646  */
 647 static
 648 ssize_t lib_ring_buffer_file_read(struct file *filp,
 649                                   char __user *user_buf,
 650                                   size_t count,
 651                                   loff_t *ppos)
 652 {
 653         struct inode *inode = filp->f_path.dentry->d_inode;
 654         struct lib_ring_buffer *buf = inode->i_private;
 655         struct channel *chan = buf->backend.chan;
 656
 657         return channel_ring_buffer_file_read(filp, user_buf, count, ppos,
 658                                              chan, buf, 0);
 659 }
 660
 661 /**
 662  * channel_file_read - Read channel record payload.
 663  * @filp: file structure pointer.
 664  * @buffer: user buffer to read data into.
 665  * @count: number of bytes to read.
 666  * @ppos: file read position.
 667  *
 668  * Returns a negative value on error, or the number of bytes read on success.
 669  * ppos is used to save the position _within the current record_ between calls
 670  * to read().
 671  */
 672 static
 673 ssize_t channel_file_read(struct file *filp,
 674                           char __user *user_buf,
 675                           size_t count,
 676                           loff_t *ppos)
 677 {
 678         struct inode *inode = filp->f_path.dentry->d_inode;
 679         struct channel *chan = inode->i_private;
 680         const struct lib_ring_buffer_config *config = &chan->backend.config;
 681
 682         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
 683                 return channel_ring_buffer_file_read(filp, user_buf, count,
 684                                                      ppos, chan, NULL, 1);
 685         else {
 686                 struct lib_ring_buffer *buf =
 687                         channel_get_ring_buffer(config, chan, 0);
 688                 return channel_ring_buffer_file_read(filp, user_buf, count,
 689                                                      ppos, chan, buf, 0);
 690         }
 691 }
 692
 693 static
 694 int lib_ring_buffer_file_open(struct inode *inode, struct file *file)
 695 {
 696         struct lib_ring_buffer *buf = inode->i_private;
 697         int ret;
 698
 699         ret = lib_ring_buffer_iterator_open(buf);
 700         if (ret)
 701                 return ret;
 702
 703         file->private_data = buf;
 704         ret = nonseekable_open(inode, file);
 705         if (ret)
 706                 goto release_iter;
 707         return 0;
 708
 709 release_iter:
 710         lib_ring_buffer_iterator_release(buf);
 711         return ret;
 712 }
 713
 714 static
 715 int lib_ring_buffer_file_release(struct inode *inode, struct file *file)
 716 {
 717         struct lib_ring_buffer *buf = inode->i_private;
 718
 719         lib_ring_buffer_iterator_release(buf);
 720         return 0;
 721 }
 722
 723 static
 724 int channel_file_open(struct inode *inode, struct file *file)
 725 {
 726         struct channel *chan = inode->i_private;
 727         int ret;
 728
 729         ret = channel_iterator_open(chan);
 730         if (ret)
 731                 return ret;
 732
 733         file->private_data = chan;
 734         ret = nonseekable_open(inode, file);
 735         if (ret)
 736                 goto release_iter;
 737         return 0;
 738
 739 release_iter:
 740         channel_iterator_release(chan);
 741         return ret;
 742 }
 743
 744 static
 745 int channel_file_release(struct inode *inode, struct file *file)
 746 {
 747         struct channel *chan = inode->i_private;
 748
 749         channel_iterator_release(chan);
 750         return 0;
 751 }
 752
 753 const struct file_operations channel_payload_file_operations = {
 754         .owner = THIS_MODULE,
 755         .open = channel_file_open,
 756         .release = channel_file_release,
 757         .read = channel_file_read,
 758         .llseek = vfs_lib_ring_buffer_no_llseek,
 759 };
 760 EXPORT_SYMBOL_GPL(channel_payload_file_operations);
 761
 762 const struct file_operations lib_ring_buffer_payload_file_operations = {
 763         .owner = THIS_MODULE,
 764         .open = lib_ring_buffer_file_open,
 765         .release = lib_ring_buffer_file_release,
 766         .read = lib_ring_buffer_file_read,
 767         .llseek = vfs_lib_ring_buffer_no_llseek,
 768 };
 769 EXPORT_SYMBOL_GPL(lib_ring_buffer_payload_file_operations);