lib/ringbuffer/ring_buffer_iterator.c

   1 /*
   2  * ring_buffer_iterator.c
   3  *
   4  * Ring buffer and channel iterators. Get each event of a channel in order. Uses
   5  * a prio heap for per-cpu buffers, giving a O(log(NR_CPUS)) algorithmic
   6  * complexity for the "get next event" operation.
   7  *
   8  * Copyright (C) 2010-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   9  *
  10  * This library is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; only
  13  * version 2.1 of the License.
  14  *
  15  * This library is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with this library; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  *
  24  * Author:
  25  *      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  26  */
  27
  28 #include "../../wrapper/ringbuffer/iterator.h"
  29 #include "../../wrapper/file.h"
  30 #include <linux/jiffies.h>
  31 #include <linux/delay.h>
  32 #include <linux/module.h>
  33
  34 /*
  35  * Safety factor taking into account internal kernel interrupt latency.
  36  * Assuming 250ms worse-case latency.
  37  */
  38 #define MAX_SYSTEM_LATENCY      250
  39
  40 /*
  41  * Maximum delta expected between trace clocks. At most 1 jiffy delta.
  42  */
  43 #define MAX_CLOCK_DELTA         (jiffies_to_usecs(1) * 1000)
  44
  45 /**
  46  * lib_ring_buffer_get_next_record - Get the next record in a buffer.
  47  * @chan: channel
  48  * @buf: buffer
  49  *
  50  * Returns the size of the event read, -EAGAIN if buffer is empty, -ENODATA if
  51  * buffer is empty and finalized. The buffer must already be opened for reading.
  52  */
  53 ssize_t lib_ring_buffer_get_next_record(struct channel *chan,
  54                                         struct lib_ring_buffer *buf)
  55 {
  56         const struct lib_ring_buffer_config *config = &chan->backend.config;
  57         struct lib_ring_buffer_iter *iter = &buf->iter;
  58         int ret;
  59
  60 restart:
  61         switch (iter->state) {
  62         case ITER_GET_SUBBUF:
  63                 ret = lib_ring_buffer_get_next_subbuf(buf);
  64                 if (ret && !ACCESS_ONCE(buf->finalized)
  65                     && config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
  66                         /*
  67                          * Use "pull" scheme for global buffers. The reader
  68                          * itself flushes the buffer to "pull" data not visible
  69                          * to readers yet. Flush current subbuffer and re-try.
  70                          *
  71                          * Per-CPU buffers rather use a "push" scheme because
  72                          * the IPI needed to flush all CPU's buffers is too
  73                          * costly. In the "push" scheme, the reader waits for
  74                          * the writer periodic deferrable timer to flush the
  75                          * buffers (keeping track of a quiescent state
  76                          * timestamp). Therefore, the writer "pushes" data out
  77                          * of the buffers rather than letting the reader "pull"
  78                          * data from the buffer.
  79                          */
  80                         lib_ring_buffer_switch_slow(buf, SWITCH_ACTIVE);
  81                         ret = lib_ring_buffer_get_next_subbuf(buf);
  82                 }
  83                 if (ret)
  84                         return ret;
  85                 iter->consumed = buf->cons_snapshot;
  86                 iter->data_size = lib_ring_buffer_get_read_data_size(config, buf);
  87                 iter->read_offset = iter->consumed;
  88                 /* skip header */
  89                 iter->read_offset += config->cb.subbuffer_header_size();
  90                 iter->state = ITER_TEST_RECORD;
  91                 goto restart;
  92         case ITER_TEST_RECORD:
  93                 if (iter->read_offset - iter->consumed >= iter->data_size) {
  94                         iter->state = ITER_PUT_SUBBUF;
  95                 } else {
  96                         CHAN_WARN_ON(chan, !config->cb.record_get);
  97                         config->cb.record_get(config, chan, buf,
  98                                               iter->read_offset,
  99                                               &iter->header_len,
 100                                               &iter->payload_len,
 101                                               &iter->timestamp);
 102                         iter->read_offset += iter->header_len;
 103                         subbuffer_consume_record(config, &buf->backend);
 104                         iter->state = ITER_NEXT_RECORD;
 105                         return iter->payload_len;
 106                 }
 107                 goto restart;
 108         case ITER_NEXT_RECORD:
 109                 iter->read_offset += iter->payload_len;
 110                 iter->state = ITER_TEST_RECORD;
 111                 goto restart;
 112         case ITER_PUT_SUBBUF:
 113                 lib_ring_buffer_put_next_subbuf(buf);
 114                 iter->state = ITER_GET_SUBBUF;
 115                 goto restart;
 116         default:
 117                 CHAN_WARN_ON(chan, 1);  /* Should not happen */
 118                 return -EPERM;
 119         }
 120 }
 121 EXPORT_SYMBOL_GPL(lib_ring_buffer_get_next_record);
 122
 123 static int buf_is_higher(void *a, void *b)
 124 {
 125         struct lib_ring_buffer *bufa = a;
 126         struct lib_ring_buffer *bufb = b;
 127
 128         /* Consider lowest timestamps to be at the top of the heap */
 129         return (bufa->iter.timestamp < bufb->iter.timestamp);
 130 }
 131
 132 static
 133 void lib_ring_buffer_get_empty_buf_records(const struct lib_ring_buffer_config *config,
 134                                            struct channel *chan)
 135 {
 136         struct lttng_ptr_heap *heap = &chan->iter.heap;
 137         struct lib_ring_buffer *buf, *tmp;
 138         ssize_t len;
 139
 140         list_for_each_entry_safe(buf, tmp, &chan->iter.empty_head,
 141                                  iter.empty_node) {
 142                 len = lib_ring_buffer_get_next_record(chan, buf);
 143
 144                 /*
 145                  * Deal with -EAGAIN and -ENODATA.
 146                  * len >= 0 means record contains data.
 147                  * -EBUSY should never happen, because we support only one
 148                  * reader.
 149                  */
 150                 switch (len) {
 151                 case -EAGAIN:
 152                         /* Keep node in empty list */
 153                         break;
 154                 case -ENODATA:
 155                         /*
 156                          * Buffer is finalized. Don't add to list of empty
 157                          * buffer, because it has no more data to provide, ever.
 158                          */
 159                         list_del(&buf->iter.empty_node);
 160                         break;
 161                 case -EBUSY:
 162                         CHAN_WARN_ON(chan, 1);
 163                         break;
 164                 default:
 165                         /*
 166                          * Insert buffer into the heap, remove from empty buffer
 167                          * list.
 168                          */
 169                         CHAN_WARN_ON(chan, len < 0);
 170                         list_del(&buf->iter.empty_node);
 171                         CHAN_WARN_ON(chan, lttng_heap_insert(heap, buf));
 172                 }
 173         }
 174 }
 175
 176 static
 177 void lib_ring_buffer_wait_for_qs(const struct lib_ring_buffer_config *config,
 178                                  struct channel *chan)
 179 {
 180         u64 timestamp_qs;
 181         unsigned long wait_msecs;
 182
 183         /*
 184          * No need to wait if no empty buffers are present.
 185          */
 186         if (list_empty(&chan->iter.empty_head))
 187                 return;
 188
 189         timestamp_qs = config->cb.ring_buffer_clock_read(chan);
 190         /*
 191          * We need to consider previously empty buffers.
 192          * Do a get next buf record on each of them. Add them to
 193          * the heap if they have data. If at least one of them
 194          * don't have data, we need to wait for
 195          * switch_timer_interval + MAX_SYSTEM_LATENCY (so we are sure the
 196          * buffers have been switched either by the timer or idle entry) and
 197          * check them again, adding them if they have data.
 198          */
 199         lib_ring_buffer_get_empty_buf_records(config, chan);
 200
 201         /*
 202          * No need to wait if no empty buffers are present.
 203          */
 204         if (list_empty(&chan->iter.empty_head))
 205                 return;
 206
 207         /*
 208          * We need to wait for the buffer switch timer to run. If the
 209          * CPU is idle, idle entry performed the switch.
 210          * TODO: we could optimize further by skipping the sleep if all
 211          * empty buffers belong to idle or offline cpus.
 212          */
 213         wait_msecs = jiffies_to_msecs(chan->switch_timer_interval);
 214         wait_msecs += MAX_SYSTEM_LATENCY;
 215         msleep(wait_msecs);
 216         lib_ring_buffer_get_empty_buf_records(config, chan);
 217         /*
 218          * Any buffer still in the empty list here cannot possibly
 219          * contain an event with a timestamp prior to "timestamp_qs".
 220          * The new quiescent state timestamp is the one we grabbed
 221          * before waiting for buffer data.  It is therefore safe to
 222          * ignore empty buffers up to last_qs timestamp for fusion
 223          * merge.
 224          */
 225         chan->iter.last_qs = timestamp_qs;
 226 }
 227
 228 /**
 229  * channel_get_next_record - Get the next record in a channel.
 230  * @chan: channel
 231  * @ret_buf: the buffer in which the event is located (output)
 232  *
 233  * Returns the size of new current event, -EAGAIN if all buffers are empty,
 234  * -ENODATA if all buffers are empty and finalized. The channel must already be
 235  * opened for reading.
 236  */
 237
 238 ssize_t channel_get_next_record(struct channel *chan,
 239                                 struct lib_ring_buffer **ret_buf)
 240 {
 241         const struct lib_ring_buffer_config *config = &chan->backend.config;
 242         struct lib_ring_buffer *buf;
 243         struct lttng_ptr_heap *heap;
 244         ssize_t len;
 245
 246         if (config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
 247                 *ret_buf = channel_get_ring_buffer(config, chan, 0);
 248                 return lib_ring_buffer_get_next_record(chan, *ret_buf);
 249         }
 250
 251         heap = &chan->iter.heap;
 252
 253         /*
 254          * get next record for topmost buffer.
 255          */
 256         buf = lttng_heap_maximum(heap);
 257         if (buf) {
 258                 len = lib_ring_buffer_get_next_record(chan, buf);
 259                 /*
 260                  * Deal with -EAGAIN and -ENODATA.
 261                  * len >= 0 means record contains data.
 262                  */
 263                 switch (len) {
 264                 case -EAGAIN:
 265                         buf->iter.timestamp = 0;
 266                         list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 267                         /* Remove topmost buffer from the heap */
 268                         CHAN_WARN_ON(chan, lttng_heap_remove(heap) != buf);
 269                         break;
 270                 case -ENODATA:
 271                         /*
 272                          * Buffer is finalized. Remove buffer from heap and
 273                          * don't add to list of empty buffer, because it has no
 274                          * more data to provide, ever.
 275                          */
 276                         CHAN_WARN_ON(chan, lttng_heap_remove(heap) != buf);
 277                         break;
 278                 case -EBUSY:
 279                         CHAN_WARN_ON(chan, 1);
 280                         break;
 281                 default:
 282                         /*
 283                          * Reinsert buffer into the heap. Note that heap can be
 284                          * partially empty, so we need to use
 285                          * lttng_heap_replace_max().
 286                          */
 287                         CHAN_WARN_ON(chan, len < 0);
 288                         CHAN_WARN_ON(chan, lttng_heap_replace_max(heap, buf) != buf);
 289                         break;
 290                 }
 291         }
 292
 293         buf = lttng_heap_maximum(heap);
 294         if (!buf || buf->iter.timestamp > chan->iter.last_qs) {
 295                 /*
 296                  * Deal with buffers previously showing no data.
 297                  * Add buffers containing data to the heap, update
 298                  * last_qs.
 299                  */
 300                 lib_ring_buffer_wait_for_qs(config, chan);
 301         }
 302
 303         *ret_buf = buf = lttng_heap_maximum(heap);
 304         if (buf) {
 305                 /*
 306                  * If this warning triggers, you probably need to check your
 307                  * system interrupt latency. Typical causes: too many printk()
 308                  * output going to a serial console with interrupts off.
 309                  * Allow for MAX_CLOCK_DELTA ns timestamp delta going backward.
 310                  * Observed on SMP KVM setups with trace_clock().
 311                  */
 312                 if (chan->iter.last_timestamp
 313                     > (buf->iter.timestamp + MAX_CLOCK_DELTA)) {
 314                         printk(KERN_WARNING "ring_buffer: timestamps going "
 315                                "backward. Last time %llu ns, cpu %d, "
 316                                "current time %llu ns, cpu %d, "
 317                                "delta %llu ns.\n",
 318                                chan->iter.last_timestamp, chan->iter.last_cpu,
 319                                buf->iter.timestamp, buf->backend.cpu,
 320                                chan->iter.last_timestamp - buf->iter.timestamp);
 321                         CHAN_WARN_ON(chan, 1);
 322                 }
 323                 chan->iter.last_timestamp = buf->iter.timestamp;
 324                 chan->iter.last_cpu = buf->backend.cpu;
 325                 return buf->iter.payload_len;
 326         } else {
 327                 /* Heap is empty */
 328                 if (list_empty(&chan->iter.empty_head))
 329                         return -ENODATA;        /* All buffers finalized */
 330                 else
 331                         return -EAGAIN;         /* Temporarily empty */
 332         }
 333 }
 334 EXPORT_SYMBOL_GPL(channel_get_next_record);
 335
 336 static
 337 void lib_ring_buffer_iterator_init(struct channel *chan, struct lib_ring_buffer *buf)
 338 {
 339         if (buf->iter.allocated)
 340                 return;
 341
 342         buf->iter.allocated = 1;
 343         if (chan->iter.read_open && !buf->iter.read_open) {
 344                 CHAN_WARN_ON(chan, lib_ring_buffer_open_read(buf) != 0);
 345                 buf->iter.read_open = 1;
 346         }
 347
 348         /* Add to list of buffers without any current record */
 349         if (chan->backend.config.alloc == RING_BUFFER_ALLOC_PER_CPU)
 350                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 351 }
 352
 353 #ifdef CONFIG_HOTPLUG_CPU
 354 static
 355 int channel_iterator_cpu_hotplug(struct notifier_block *nb,
 356                                            unsigned long action,
 357                                            void *hcpu)
 358 {
 359         unsigned int cpu = (unsigned long)hcpu;
 360         struct channel *chan = container_of(nb, struct channel,
 361                                             hp_iter_notifier);
 362         struct lib_ring_buffer *buf = per_cpu_ptr(chan->backend.buf, cpu);
 363         const struct lib_ring_buffer_config *config = &chan->backend.config;
 364
 365         if (!chan->hp_iter_enable)
 366                 return NOTIFY_DONE;
 367
 368         CHAN_WARN_ON(chan, config->alloc == RING_BUFFER_ALLOC_GLOBAL);
 369
 370         switch (action) {
 371         case CPU_DOWN_FAILED:
 372         case CPU_DOWN_FAILED_FROZEN:
 373         case CPU_ONLINE:
 374         case CPU_ONLINE_FROZEN:
 375                 lib_ring_buffer_iterator_init(chan, buf);
 376                 return NOTIFY_OK;
 377         default:
 378                 return NOTIFY_DONE;
 379         }
 380 }
 381 #endif
 382
 383 int channel_iterator_init(struct channel *chan)
 384 {
 385         const struct lib_ring_buffer_config *config = &chan->backend.config;
 386         struct lib_ring_buffer *buf;
 387
 388         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 389                 int cpu, ret;
 390
 391                 INIT_LIST_HEAD(&chan->iter.empty_head);
 392                 ret = lttng_heap_init(&chan->iter.heap,
 393                                 num_possible_cpus(),
 394                                 GFP_KERNEL, buf_is_higher);
 395                 if (ret)
 396                         return ret;
 397                 /*
 398                  * In case of non-hotplug cpu, if the ring-buffer is allocated
 399                  * in early initcall, it will not be notified of secondary cpus.
 400                  * In that off case, we need to allocate for all possible cpus.
 401                  */
 402 #ifdef CONFIG_HOTPLUG_CPU
 403                 chan->hp_iter_notifier.notifier_call =
 404                         channel_iterator_cpu_hotplug;
 405                 chan->hp_iter_notifier.priority = 10;
 406                 register_cpu_notifier(&chan->hp_iter_notifier);
 407                 get_online_cpus();
 408                 for_each_online_cpu(cpu) {
 409                         buf = per_cpu_ptr(chan->backend.buf, cpu);
 410                         lib_ring_buffer_iterator_init(chan, buf);
 411                 }
 412                 chan->hp_iter_enable = 1;
 413                 put_online_cpus();
 414 #else
 415                 for_each_possible_cpu(cpu) {
 416                         buf = per_cpu_ptr(chan->backend.buf, cpu);
 417                         lib_ring_buffer_iterator_init(chan, buf);
 418                 }
 419 #endif
 420         } else {
 421                 buf = channel_get_ring_buffer(config, chan, 0);
 422                 lib_ring_buffer_iterator_init(chan, buf);
 423         }
 424         return 0;
 425 }
 426
 427 void channel_iterator_unregister_notifiers(struct channel *chan)
 428 {
 429         const struct lib_ring_buffer_config *config = &chan->backend.config;
 430
 431         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 432                 chan->hp_iter_enable = 0;
 433                 unregister_cpu_notifier(&chan->hp_iter_notifier);
 434         }
 435 }
 436
 437 void channel_iterator_free(struct channel *chan)
 438 {
 439         const struct lib_ring_buffer_config *config = &chan->backend.config;
 440
 441         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
 442                 lttng_heap_free(&chan->iter.heap);
 443 }
 444
 445 int lib_ring_buffer_iterator_open(struct lib_ring_buffer *buf)
 446 {
 447         struct channel *chan = buf->backend.chan;
 448         const struct lib_ring_buffer_config *config = &chan->backend.config;
 449         CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR);
 450         return lib_ring_buffer_open_read(buf);
 451 }
 452 EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_open);
 453
 454 /*
 455  * Note: Iterators must not be mixed with other types of outputs, because an
 456  * iterator can leave the buffer in "GET" state, which is not consistent with
 457  * other types of output (mmap, splice, raw data read).
 458  */
 459 void lib_ring_buffer_iterator_release(struct lib_ring_buffer *buf)
 460 {
 461         lib_ring_buffer_release_read(buf);
 462 }
 463 EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_release);
 464
 465 int channel_iterator_open(struct channel *chan)
 466 {
 467         const struct lib_ring_buffer_config *config = &chan->backend.config;
 468         struct lib_ring_buffer *buf;
 469         int ret = 0, cpu;
 470
 471         CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR);
 472
 473         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 474                 get_online_cpus();
 475                 /* Allow CPU hotplug to keep track of opened reader */
 476                 chan->iter.read_open = 1;
 477                 for_each_channel_cpu(cpu, chan) {
 478                         buf = channel_get_ring_buffer(config, chan, cpu);
 479                         ret = lib_ring_buffer_iterator_open(buf);
 480                         if (ret)
 481                                 goto error;
 482                         buf->iter.read_open = 1;
 483                 }
 484                 put_online_cpus();
 485         } else {
 486                 buf = channel_get_ring_buffer(config, chan, 0);
 487                 ret = lib_ring_buffer_iterator_open(buf);
 488         }
 489         return ret;
 490 error:
 491         /* Error should always happen on CPU 0, hence no close is required. */
 492         CHAN_WARN_ON(chan, cpu != 0);
 493         put_online_cpus();
 494         return ret;
 495 }
 496 EXPORT_SYMBOL_GPL(channel_iterator_open);
 497
 498 void channel_iterator_release(struct channel *chan)
 499 {
 500         const struct lib_ring_buffer_config *config = &chan->backend.config;
 501         struct lib_ring_buffer *buf;
 502         int cpu;
 503
 504         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 505                 get_online_cpus();
 506                 for_each_channel_cpu(cpu, chan) {
 507                         buf = channel_get_ring_buffer(config, chan, cpu);
 508                         if (buf->iter.read_open) {
 509                                 lib_ring_buffer_iterator_release(buf);
 510                                 buf->iter.read_open = 0;
 511                         }
 512                 }
 513                 chan->iter.read_open = 0;
 514                 put_online_cpus();
 515         } else {
 516                 buf = channel_get_ring_buffer(config, chan, 0);
 517                 lib_ring_buffer_iterator_release(buf);
 518         }
 519 }
 520 EXPORT_SYMBOL_GPL(channel_iterator_release);
 521
 522 void lib_ring_buffer_iterator_reset(struct lib_ring_buffer *buf)
 523 {
 524         struct channel *chan = buf->backend.chan;
 525
 526         if (buf->iter.state != ITER_GET_SUBBUF)
 527                 lib_ring_buffer_put_next_subbuf(buf);
 528         buf->iter.state = ITER_GET_SUBBUF;
 529         /* Remove from heap (if present). */
 530         if (lttng_heap_cherrypick(&chan->iter.heap, buf))
 531                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 532         buf->iter.timestamp = 0;
 533         buf->iter.header_len = 0;
 534         buf->iter.payload_len = 0;
 535         buf->iter.consumed = 0;
 536         buf->iter.read_offset = 0;
 537         buf->iter.data_size = 0;
 538         /* Don't reset allocated and read_open */
 539 }
 540
 541 void channel_iterator_reset(struct channel *chan)
 542 {
 543         const struct lib_ring_buffer_config *config = &chan->backend.config;
 544         struct lib_ring_buffer *buf;
 545         int cpu;
 546
 547         /* Empty heap, put into empty_head */
 548         while ((buf = lttng_heap_remove(&chan->iter.heap)) != NULL)
 549                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 550
 551         for_each_channel_cpu(cpu, chan) {
 552                 buf = channel_get_ring_buffer(config, chan, cpu);
 553                 lib_ring_buffer_iterator_reset(buf);
 554         }
 555         /* Don't reset read_open */
 556         chan->iter.last_qs = 0;
 557         chan->iter.last_timestamp = 0;
 558         chan->iter.last_cpu = 0;
 559         chan->iter.len_left = 0;
 560 }
 561
 562 /*
 563  * Ring buffer payload extraction read() implementation.
 564  */
 565 static
 566 ssize_t channel_ring_buffer_file_read(struct file *filp,
 567                                       char __user *user_buf,
 568                                       size_t count,
 569                                       loff_t *ppos,
 570                                       struct channel *chan,
 571                                       struct lib_ring_buffer *buf,
 572                                       int fusionmerge)
 573 {
 574         const struct lib_ring_buffer_config *config = &chan->backend.config;
 575         size_t read_count = 0, read_offset;
 576         ssize_t len;
 577
 578         might_sleep();
 579         if (!access_ok(VERIFY_WRITE, user_buf, count))
 580                 return -EFAULT;
 581
 582         /* Finish copy of previous record */
 583         if (*ppos != 0) {
 584                 if (read_count < count) {
 585                         len = chan->iter.len_left;
 586                         read_offset = *ppos;
 587                         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU
 588                             && fusionmerge)
 589                                 buf = lttng_heap_maximum(&chan->iter.heap);
 590                         CHAN_WARN_ON(chan, !buf);
 591                         goto skip_get_next;
 592                 }
 593         }
 594
 595         while (read_count < count) {
 596                 size_t copy_len, space_left;
 597
 598                 if (fusionmerge)
 599                         len = channel_get_next_record(chan, &buf);
 600                 else
 601                         len = lib_ring_buffer_get_next_record(chan, buf);
 602 len_test:
 603                 if (len < 0) {
 604                         /*
 605                          * Check if buffer is finalized (end of file).
 606                          */
 607                         if (len == -ENODATA) {
 608                                 /* A 0 read_count will tell about end of file */
 609                                 goto nodata;
 610                         }
 611                         if (filp->f_flags & O_NONBLOCK) {
 612                                 if (!read_count)
 613                                         read_count = -EAGAIN;
 614                                 goto nodata;
 615                         } else {
 616                                 int error;
 617
 618                                 /*
 619                                  * No data available at the moment, return what
 620                                  * we got.
 621                                  */
 622                                 if (read_count)
 623                                         goto nodata;
 624
 625                                 /*
 626                                  * Wait for returned len to be >= 0 or -ENODATA.
 627                                  */
 628                                 if (fusionmerge)
 629                                         error = wait_event_interruptible(
 630                                           chan->read_wait,
 631                                           ((len = channel_get_next_record(chan,
 632                                                 &buf)), len != -EAGAIN));
 633                                 else
 634                                         error = wait_event_interruptible(
 635                                           buf->read_wait,
 636                                           ((len = lib_ring_buffer_get_next_record(
 637                                                   chan, buf)), len != -EAGAIN));
 638                                 CHAN_WARN_ON(chan, len == -EBUSY);
 639                                 if (error) {
 640                                         read_count = error;
 641                                         goto nodata;
 642                                 }
 643                                 CHAN_WARN_ON(chan, len < 0 && len != -ENODATA);
 644                                 goto len_test;
 645                         }
 646                 }
 647                 read_offset = buf->iter.read_offset;
 648 skip_get_next:
 649                 space_left = count - read_count;
 650                 if (len <= space_left) {
 651                         copy_len = len;
 652                         chan->iter.len_left = 0;
 653                         *ppos = 0;
 654                 } else {
 655                         copy_len = space_left;
 656                         chan->iter.len_left = len - copy_len;
 657                         *ppos = read_offset + copy_len;
 658                 }
 659                 if (__lib_ring_buffer_copy_to_user(&buf->backend, read_offset,
 660                                                &user_buf[read_count],
 661                                                copy_len)) {
 662                         /*
 663                          * Leave the len_left and ppos values at their current
 664                          * state, as we currently have a valid event to read.
 665                          */
 666                         return -EFAULT;
 667                 }
 668                 read_count += copy_len;
 669         };
 670         return read_count;
 671
 672 nodata:
 673         *ppos = 0;
 674         chan->iter.len_left = 0;
 675         return read_count;
 676 }
 677
 678 /**
 679  * lib_ring_buffer_file_read - Read buffer record payload.
 680  * @filp: file structure pointer.
 681  * @buffer: user buffer to read data into.
 682  * @count: number of bytes to read.
 683  * @ppos: file read position.
 684  *
 685  * Returns a negative value on error, or the number of bytes read on success.
 686  * ppos is used to save the position _within the current record_ between calls
 687  * to read().
 688  */
 689 static
 690 ssize_t lib_ring_buffer_file_read(struct file *filp,
 691                                   char __user *user_buf,
 692                                   size_t count,
 693                                   loff_t *ppos)
 694 {
 695         struct inode *inode = filp->lttng_f_dentry->d_inode;
 696         struct lib_ring_buffer *buf = inode->i_private;
 697         struct channel *chan = buf->backend.chan;
 698
 699         return channel_ring_buffer_file_read(filp, user_buf, count, ppos,
 700                                              chan, buf, 0);
 701 }
 702
 703 /**
 704  * channel_file_read - Read channel record payload.
 705  * @filp: file structure pointer.
 706  * @buffer: user buffer to read data into.
 707  * @count: number of bytes to read.
 708  * @ppos: file read position.
 709  *
 710  * Returns a negative value on error, or the number of bytes read on success.
 711  * ppos is used to save the position _within the current record_ between calls
 712  * to read().
 713  */
 714 static
 715 ssize_t channel_file_read(struct file *filp,
 716                           char __user *user_buf,
 717                           size_t count,
 718                           loff_t *ppos)
 719 {
 720         struct inode *inode = filp->lttng_f_dentry->d_inode;
 721         struct channel *chan = inode->i_private;
 722         const struct lib_ring_buffer_config *config = &chan->backend.config;
 723
 724         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
 725                 return channel_ring_buffer_file_read(filp, user_buf, count,
 726                                                      ppos, chan, NULL, 1);
 727         else {
 728                 struct lib_ring_buffer *buf =
 729                         channel_get_ring_buffer(config, chan, 0);
 730                 return channel_ring_buffer_file_read(filp, user_buf, count,
 731                                                      ppos, chan, buf, 0);
 732         }
 733 }
 734
 735 static
 736 int lib_ring_buffer_file_open(struct inode *inode, struct file *file)
 737 {
 738         struct lib_ring_buffer *buf = inode->i_private;
 739         int ret;
 740
 741         ret = lib_ring_buffer_iterator_open(buf);
 742         if (ret)
 743                 return ret;
 744
 745         file->private_data = buf;
 746         ret = nonseekable_open(inode, file);
 747         if (ret)
 748                 goto release_iter;
 749         return 0;
 750
 751 release_iter:
 752         lib_ring_buffer_iterator_release(buf);
 753         return ret;
 754 }
 755
 756 static
 757 int lib_ring_buffer_file_release(struct inode *inode, struct file *file)
 758 {
 759         struct lib_ring_buffer *buf = inode->i_private;
 760
 761         lib_ring_buffer_iterator_release(buf);
 762         return 0;
 763 }
 764
 765 static
 766 int channel_file_open(struct inode *inode, struct file *file)
 767 {
 768         struct channel *chan = inode->i_private;
 769         int ret;
 770
 771         ret = channel_iterator_open(chan);
 772         if (ret)
 773                 return ret;
 774
 775         file->private_data = chan;
 776         ret = nonseekable_open(inode, file);
 777         if (ret)
 778                 goto release_iter;
 779         return 0;
 780
 781 release_iter:
 782         channel_iterator_release(chan);
 783         return ret;
 784 }
 785
 786 static
 787 int channel_file_release(struct inode *inode, struct file *file)
 788 {
 789         struct channel *chan = inode->i_private;
 790
 791         channel_iterator_release(chan);
 792         return 0;
 793 }
 794
 795 const struct file_operations channel_payload_file_operations = {
 796         .owner = THIS_MODULE,
 797         .open = channel_file_open,
 798         .release = channel_file_release,
 799         .read = channel_file_read,
 800         .llseek = vfs_lib_ring_buffer_no_llseek,
 801 };
 802 EXPORT_SYMBOL_GPL(channel_payload_file_operations);
 803
 804 const struct file_operations lib_ring_buffer_payload_file_operations = {
 805         .owner = THIS_MODULE,
 806         .open = lib_ring_buffer_file_open,
 807         .release = lib_ring_buffer_file_release,
 808         .read = lib_ring_buffer_file_read,
 809         .llseek = vfs_lib_ring_buffer_no_llseek,
 810 };
 811 EXPORT_SYMBOL_GPL(lib_ring_buffer_payload_file_operations);