Commit | Line | Data |
---|---|---|
f3bc08c5 MD |
1 | /* |
2 | * ring_buffer_iterator.c | |
3 | * | |
f3bc08c5 MD |
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 | * | |
886d51a3 MD |
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 | * | |
f3bc08c5 MD |
24 | * Author: |
25 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
f3bc08c5 MD |
26 | */ |
27 | ||
c075712b MD |
28 | #include <wrapper/ringbuffer/iterator.h> |
29 | #include <wrapper/file.h> | |
f3bc08c5 MD |
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 | { | |
5a8fd222 | 56 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
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); | |
acb98454 | 64 | if (ret && !READ_ONCE(buf->finalized) |
f3bc08c5 MD |
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 | |
da9f3fb7 | 74 | * the writer periodic timer to flush the |
f3bc08c5 MD |
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 | { | |
a88db018 | 136 | struct lttng_ptr_heap *heap = &chan->iter.heap; |
f3bc08c5 MD |
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 | |
ab2277d6 | 167 | * list. |
f3bc08c5 MD |
168 | */ |
169 | CHAN_WARN_ON(chan, len < 0); | |
170 | list_del(&buf->iter.empty_node); | |
a88db018 | 171 | CHAN_WARN_ON(chan, lttng_heap_insert(heap, buf)); |
f3bc08c5 MD |
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 | { | |
5a8fd222 | 241 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 | 242 | struct lib_ring_buffer *buf; |
a88db018 | 243 | struct lttng_ptr_heap *heap; |
f3bc08c5 MD |
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 | */ | |
a88db018 | 256 | buf = lttng_heap_maximum(heap); |
f3bc08c5 MD |
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 */ | |
a88db018 | 268 | CHAN_WARN_ON(chan, lttng_heap_remove(heap) != buf); |
f3bc08c5 MD |
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 | */ | |
a88db018 | 276 | CHAN_WARN_ON(chan, lttng_heap_remove(heap) != buf); |
f3bc08c5 MD |
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 | |
a88db018 | 285 | * lttng_heap_replace_max(). |
f3bc08c5 MD |
286 | */ |
287 | CHAN_WARN_ON(chan, len < 0); | |
a88db018 | 288 | CHAN_WARN_ON(chan, lttng_heap_replace_max(heap, buf) != buf); |
f3bc08c5 MD |
289 | break; |
290 | } | |
291 | } | |
292 | ||
a88db018 | 293 | buf = lttng_heap_maximum(heap); |
f3bc08c5 MD |
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 | ||
a88db018 | 303 | *ret_buf = buf = lttng_heap_maximum(heap); |
f3bc08c5 MD |
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 */ | |
5a8fd222 | 349 | if (chan->backend.config.alloc == RING_BUFFER_ALLOC_PER_CPU) |
f3bc08c5 MD |
350 | list_add(&buf->iter.empty_node, &chan->iter.empty_head); |
351 | } | |
352 | ||
1e367326 MD |
353 | #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)) |
354 | ||
355 | int lttng_cpuhp_rb_iter_online(unsigned int cpu, | |
356 | struct lttng_cpuhp_node *node) | |
357 | { | |
358 | struct channel *chan = container_of(node, struct channel, | |
359 | cpuhp_iter_online); | |
360 | struct lib_ring_buffer *buf = per_cpu_ptr(chan->backend.buf, cpu); | |
361 | const struct lib_ring_buffer_config *config = &chan->backend.config; | |
362 | ||
363 | CHAN_WARN_ON(chan, config->alloc == RING_BUFFER_ALLOC_GLOBAL); | |
364 | ||
365 | lib_ring_buffer_iterator_init(chan, buf); | |
366 | return 0; | |
367 | } | |
368 | EXPORT_SYMBOL_GPL(lttng_cpuhp_rb_iter_online); | |
369 | ||
370 | #else /* #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)) */ | |
371 | ||
f3bc08c5 MD |
372 | #ifdef CONFIG_HOTPLUG_CPU |
373 | static | |
e8f071d5 | 374 | int channel_iterator_cpu_hotplug(struct notifier_block *nb, |
f3bc08c5 MD |
375 | unsigned long action, |
376 | void *hcpu) | |
377 | { | |
378 | unsigned int cpu = (unsigned long)hcpu; | |
379 | struct channel *chan = container_of(nb, struct channel, | |
380 | hp_iter_notifier); | |
381 | struct lib_ring_buffer *buf = per_cpu_ptr(chan->backend.buf, cpu); | |
5a8fd222 | 382 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
383 | |
384 | if (!chan->hp_iter_enable) | |
385 | return NOTIFY_DONE; | |
386 | ||
387 | CHAN_WARN_ON(chan, config->alloc == RING_BUFFER_ALLOC_GLOBAL); | |
388 | ||
389 | switch (action) { | |
390 | case CPU_DOWN_FAILED: | |
391 | case CPU_DOWN_FAILED_FROZEN: | |
392 | case CPU_ONLINE: | |
393 | case CPU_ONLINE_FROZEN: | |
394 | lib_ring_buffer_iterator_init(chan, buf); | |
395 | return NOTIFY_OK; | |
396 | default: | |
397 | return NOTIFY_DONE; | |
398 | } | |
399 | } | |
400 | #endif | |
401 | ||
1e367326 MD |
402 | #endif /* #else #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)) */ |
403 | ||
f3bc08c5 MD |
404 | int channel_iterator_init(struct channel *chan) |
405 | { | |
5a8fd222 | 406 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
407 | struct lib_ring_buffer *buf; |
408 | ||
409 | if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) { | |
1e367326 | 410 | int ret; |
f3bc08c5 MD |
411 | |
412 | INIT_LIST_HEAD(&chan->iter.empty_head); | |
a88db018 | 413 | ret = lttng_heap_init(&chan->iter.heap, |
41affe31 | 414 | num_possible_cpus(), |
f3bc08c5 MD |
415 | GFP_KERNEL, buf_is_higher); |
416 | if (ret) | |
417 | return ret; | |
1e367326 MD |
418 | |
419 | #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)) | |
420 | chan->cpuhp_iter_online.component = LTTNG_RING_BUFFER_ITER; | |
421 | ret = cpuhp_state_add_instance(lttng_rb_hp_online, | |
422 | &chan->cpuhp_iter_online.node); | |
423 | if (ret) | |
424 | return ret; | |
425 | #else /* #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)) */ | |
426 | { | |
427 | int cpu; | |
428 | ||
429 | /* | |
430 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
431 | * in early initcall, it will not be notified of secondary cpus. | |
432 | * In that off case, we need to allocate for all possible cpus. | |
433 | */ | |
f3bc08c5 | 434 | #ifdef CONFIG_HOTPLUG_CPU |
1e367326 MD |
435 | chan->hp_iter_notifier.notifier_call = |
436 | channel_iterator_cpu_hotplug; | |
437 | chan->hp_iter_notifier.priority = 10; | |
438 | register_cpu_notifier(&chan->hp_iter_notifier); | |
439 | ||
440 | get_online_cpus(); | |
441 | for_each_online_cpu(cpu) { | |
442 | buf = per_cpu_ptr(chan->backend.buf, cpu); | |
443 | lib_ring_buffer_iterator_init(chan, buf); | |
444 | } | |
445 | chan->hp_iter_enable = 1; | |
446 | put_online_cpus(); | |
f3bc08c5 | 447 | #else |
1e367326 MD |
448 | for_each_possible_cpu(cpu) { |
449 | buf = per_cpu_ptr(chan->backend.buf, cpu); | |
450 | lib_ring_buffer_iterator_init(chan, buf); | |
451 | } | |
f3bc08c5 | 452 | #endif |
1e367326 MD |
453 | } |
454 | #endif /* #else #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)) */ | |
f3bc08c5 MD |
455 | } else { |
456 | buf = channel_get_ring_buffer(config, chan, 0); | |
457 | lib_ring_buffer_iterator_init(chan, buf); | |
458 | } | |
459 | return 0; | |
460 | } | |
461 | ||
462 | void channel_iterator_unregister_notifiers(struct channel *chan) | |
463 | { | |
5a8fd222 | 464 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
465 | |
466 | if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) { | |
1e367326 MD |
467 | #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)) |
468 | { | |
469 | int ret; | |
470 | ||
471 | ret = cpuhp_state_remove_instance(lttng_rb_hp_online, | |
472 | &chan->cpuhp_iter_online.node); | |
473 | WARN_ON(ret); | |
474 | } | |
475 | #else /* #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)) */ | |
f3bc08c5 MD |
476 | chan->hp_iter_enable = 0; |
477 | unregister_cpu_notifier(&chan->hp_iter_notifier); | |
1e367326 | 478 | #endif /* #else #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,10,0)) */ |
f3bc08c5 MD |
479 | } |
480 | } | |
481 | ||
482 | void channel_iterator_free(struct channel *chan) | |
483 | { | |
5a8fd222 | 484 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
485 | |
486 | if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) | |
a88db018 | 487 | lttng_heap_free(&chan->iter.heap); |
f3bc08c5 MD |
488 | } |
489 | ||
490 | int lib_ring_buffer_iterator_open(struct lib_ring_buffer *buf) | |
491 | { | |
492 | struct channel *chan = buf->backend.chan; | |
5a8fd222 | 493 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
494 | CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR); |
495 | return lib_ring_buffer_open_read(buf); | |
496 | } | |
497 | EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_open); | |
498 | ||
499 | /* | |
500 | * Note: Iterators must not be mixed with other types of outputs, because an | |
501 | * iterator can leave the buffer in "GET" state, which is not consistent with | |
502 | * other types of output (mmap, splice, raw data read). | |
503 | */ | |
504 | void lib_ring_buffer_iterator_release(struct lib_ring_buffer *buf) | |
505 | { | |
506 | lib_ring_buffer_release_read(buf); | |
507 | } | |
508 | EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_release); | |
509 | ||
510 | int channel_iterator_open(struct channel *chan) | |
511 | { | |
5a8fd222 | 512 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
513 | struct lib_ring_buffer *buf; |
514 | int ret = 0, cpu; | |
515 | ||
516 | CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR); | |
517 | ||
518 | if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) { | |
519 | get_online_cpus(); | |
520 | /* Allow CPU hotplug to keep track of opened reader */ | |
521 | chan->iter.read_open = 1; | |
522 | for_each_channel_cpu(cpu, chan) { | |
523 | buf = channel_get_ring_buffer(config, chan, cpu); | |
524 | ret = lib_ring_buffer_iterator_open(buf); | |
525 | if (ret) | |
526 | goto error; | |
527 | buf->iter.read_open = 1; | |
528 | } | |
529 | put_online_cpus(); | |
530 | } else { | |
531 | buf = channel_get_ring_buffer(config, chan, 0); | |
532 | ret = lib_ring_buffer_iterator_open(buf); | |
533 | } | |
534 | return ret; | |
535 | error: | |
536 | /* Error should always happen on CPU 0, hence no close is required. */ | |
537 | CHAN_WARN_ON(chan, cpu != 0); | |
538 | put_online_cpus(); | |
539 | return ret; | |
540 | } | |
541 | EXPORT_SYMBOL_GPL(channel_iterator_open); | |
542 | ||
543 | void channel_iterator_release(struct channel *chan) | |
544 | { | |
5a8fd222 | 545 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
546 | struct lib_ring_buffer *buf; |
547 | int cpu; | |
548 | ||
549 | if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) { | |
550 | get_online_cpus(); | |
551 | for_each_channel_cpu(cpu, chan) { | |
552 | buf = channel_get_ring_buffer(config, chan, cpu); | |
553 | if (buf->iter.read_open) { | |
554 | lib_ring_buffer_iterator_release(buf); | |
555 | buf->iter.read_open = 0; | |
556 | } | |
557 | } | |
558 | chan->iter.read_open = 0; | |
559 | put_online_cpus(); | |
560 | } else { | |
561 | buf = channel_get_ring_buffer(config, chan, 0); | |
562 | lib_ring_buffer_iterator_release(buf); | |
563 | } | |
564 | } | |
565 | EXPORT_SYMBOL_GPL(channel_iterator_release); | |
566 | ||
567 | void lib_ring_buffer_iterator_reset(struct lib_ring_buffer *buf) | |
568 | { | |
569 | struct channel *chan = buf->backend.chan; | |
570 | ||
571 | if (buf->iter.state != ITER_GET_SUBBUF) | |
572 | lib_ring_buffer_put_next_subbuf(buf); | |
573 | buf->iter.state = ITER_GET_SUBBUF; | |
574 | /* Remove from heap (if present). */ | |
a88db018 | 575 | if (lttng_heap_cherrypick(&chan->iter.heap, buf)) |
f3bc08c5 MD |
576 | list_add(&buf->iter.empty_node, &chan->iter.empty_head); |
577 | buf->iter.timestamp = 0; | |
578 | buf->iter.header_len = 0; | |
579 | buf->iter.payload_len = 0; | |
580 | buf->iter.consumed = 0; | |
581 | buf->iter.read_offset = 0; | |
582 | buf->iter.data_size = 0; | |
583 | /* Don't reset allocated and read_open */ | |
584 | } | |
585 | ||
586 | void channel_iterator_reset(struct channel *chan) | |
587 | { | |
5a8fd222 | 588 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
589 | struct lib_ring_buffer *buf; |
590 | int cpu; | |
591 | ||
592 | /* Empty heap, put into empty_head */ | |
a88db018 | 593 | while ((buf = lttng_heap_remove(&chan->iter.heap)) != NULL) |
f3bc08c5 MD |
594 | list_add(&buf->iter.empty_node, &chan->iter.empty_head); |
595 | ||
596 | for_each_channel_cpu(cpu, chan) { | |
597 | buf = channel_get_ring_buffer(config, chan, cpu); | |
598 | lib_ring_buffer_iterator_reset(buf); | |
599 | } | |
600 | /* Don't reset read_open */ | |
601 | chan->iter.last_qs = 0; | |
602 | chan->iter.last_timestamp = 0; | |
603 | chan->iter.last_cpu = 0; | |
604 | chan->iter.len_left = 0; | |
605 | } | |
606 | ||
607 | /* | |
608 | * Ring buffer payload extraction read() implementation. | |
609 | */ | |
610 | static | |
611 | ssize_t channel_ring_buffer_file_read(struct file *filp, | |
612 | char __user *user_buf, | |
613 | size_t count, | |
614 | loff_t *ppos, | |
615 | struct channel *chan, | |
616 | struct lib_ring_buffer *buf, | |
617 | int fusionmerge) | |
618 | { | |
5a8fd222 | 619 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
620 | size_t read_count = 0, read_offset; |
621 | ssize_t len; | |
622 | ||
623 | might_sleep(); | |
624 | if (!access_ok(VERIFY_WRITE, user_buf, count)) | |
625 | return -EFAULT; | |
626 | ||
627 | /* Finish copy of previous record */ | |
628 | if (*ppos != 0) { | |
629 | if (read_count < count) { | |
630 | len = chan->iter.len_left; | |
631 | read_offset = *ppos; | |
632 | if (config->alloc == RING_BUFFER_ALLOC_PER_CPU | |
633 | && fusionmerge) | |
a88db018 | 634 | buf = lttng_heap_maximum(&chan->iter.heap); |
f3bc08c5 MD |
635 | CHAN_WARN_ON(chan, !buf); |
636 | goto skip_get_next; | |
637 | } | |
638 | } | |
639 | ||
640 | while (read_count < count) { | |
641 | size_t copy_len, space_left; | |
642 | ||
643 | if (fusionmerge) | |
644 | len = channel_get_next_record(chan, &buf); | |
645 | else | |
646 | len = lib_ring_buffer_get_next_record(chan, buf); | |
647 | len_test: | |
648 | if (len < 0) { | |
649 | /* | |
650 | * Check if buffer is finalized (end of file). | |
651 | */ | |
652 | if (len == -ENODATA) { | |
653 | /* A 0 read_count will tell about end of file */ | |
654 | goto nodata; | |
655 | } | |
656 | if (filp->f_flags & O_NONBLOCK) { | |
657 | if (!read_count) | |
658 | read_count = -EAGAIN; | |
659 | goto nodata; | |
660 | } else { | |
661 | int error; | |
662 | ||
663 | /* | |
664 | * No data available at the moment, return what | |
665 | * we got. | |
666 | */ | |
667 | if (read_count) | |
668 | goto nodata; | |
669 | ||
670 | /* | |
671 | * Wait for returned len to be >= 0 or -ENODATA. | |
672 | */ | |
673 | if (fusionmerge) | |
674 | error = wait_event_interruptible( | |
675 | chan->read_wait, | |
676 | ((len = channel_get_next_record(chan, | |
677 | &buf)), len != -EAGAIN)); | |
678 | else | |
679 | error = wait_event_interruptible( | |
680 | buf->read_wait, | |
681 | ((len = lib_ring_buffer_get_next_record( | |
682 | chan, buf)), len != -EAGAIN)); | |
683 | CHAN_WARN_ON(chan, len == -EBUSY); | |
684 | if (error) { | |
685 | read_count = error; | |
686 | goto nodata; | |
687 | } | |
688 | CHAN_WARN_ON(chan, len < 0 && len != -ENODATA); | |
689 | goto len_test; | |
690 | } | |
691 | } | |
692 | read_offset = buf->iter.read_offset; | |
693 | skip_get_next: | |
694 | space_left = count - read_count; | |
695 | if (len <= space_left) { | |
696 | copy_len = len; | |
697 | chan->iter.len_left = 0; | |
698 | *ppos = 0; | |
699 | } else { | |
700 | copy_len = space_left; | |
701 | chan->iter.len_left = len - copy_len; | |
702 | *ppos = read_offset + copy_len; | |
703 | } | |
704 | if (__lib_ring_buffer_copy_to_user(&buf->backend, read_offset, | |
705 | &user_buf[read_count], | |
706 | copy_len)) { | |
707 | /* | |
708 | * Leave the len_left and ppos values at their current | |
709 | * state, as we currently have a valid event to read. | |
710 | */ | |
711 | return -EFAULT; | |
712 | } | |
713 | read_count += copy_len; | |
714 | }; | |
715 | return read_count; | |
716 | ||
717 | nodata: | |
718 | *ppos = 0; | |
719 | chan->iter.len_left = 0; | |
720 | return read_count; | |
721 | } | |
722 | ||
723 | /** | |
724 | * lib_ring_buffer_file_read - Read buffer record payload. | |
725 | * @filp: file structure pointer. | |
726 | * @buffer: user buffer to read data into. | |
727 | * @count: number of bytes to read. | |
728 | * @ppos: file read position. | |
729 | * | |
730 | * Returns a negative value on error, or the number of bytes read on success. | |
731 | * ppos is used to save the position _within the current record_ between calls | |
732 | * to read(). | |
733 | */ | |
734 | static | |
735 | ssize_t lib_ring_buffer_file_read(struct file *filp, | |
736 | char __user *user_buf, | |
737 | size_t count, | |
738 | loff_t *ppos) | |
739 | { | |
b06ed645 | 740 | struct inode *inode = filp->lttng_f_dentry->d_inode; |
f3bc08c5 MD |
741 | struct lib_ring_buffer *buf = inode->i_private; |
742 | struct channel *chan = buf->backend.chan; | |
743 | ||
744 | return channel_ring_buffer_file_read(filp, user_buf, count, ppos, | |
745 | chan, buf, 0); | |
746 | } | |
747 | ||
748 | /** | |
749 | * channel_file_read - Read channel record payload. | |
750 | * @filp: file structure pointer. | |
751 | * @buffer: user buffer to read data into. | |
752 | * @count: number of bytes to read. | |
753 | * @ppos: file read position. | |
754 | * | |
755 | * Returns a negative value on error, or the number of bytes read on success. | |
756 | * ppos is used to save the position _within the current record_ between calls | |
757 | * to read(). | |
758 | */ | |
759 | static | |
760 | ssize_t channel_file_read(struct file *filp, | |
761 | char __user *user_buf, | |
762 | size_t count, | |
763 | loff_t *ppos) | |
764 | { | |
b06ed645 | 765 | struct inode *inode = filp->lttng_f_dentry->d_inode; |
f3bc08c5 | 766 | struct channel *chan = inode->i_private; |
5a8fd222 | 767 | const struct lib_ring_buffer_config *config = &chan->backend.config; |
f3bc08c5 MD |
768 | |
769 | if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) | |
770 | return channel_ring_buffer_file_read(filp, user_buf, count, | |
771 | ppos, chan, NULL, 1); | |
772 | else { | |
773 | struct lib_ring_buffer *buf = | |
774 | channel_get_ring_buffer(config, chan, 0); | |
775 | return channel_ring_buffer_file_read(filp, user_buf, count, | |
776 | ppos, chan, buf, 0); | |
777 | } | |
778 | } | |
779 | ||
780 | static | |
781 | int lib_ring_buffer_file_open(struct inode *inode, struct file *file) | |
782 | { | |
783 | struct lib_ring_buffer *buf = inode->i_private; | |
784 | int ret; | |
785 | ||
786 | ret = lib_ring_buffer_iterator_open(buf); | |
787 | if (ret) | |
788 | return ret; | |
789 | ||
790 | file->private_data = buf; | |
791 | ret = nonseekable_open(inode, file); | |
792 | if (ret) | |
793 | goto release_iter; | |
794 | return 0; | |
795 | ||
796 | release_iter: | |
797 | lib_ring_buffer_iterator_release(buf); | |
798 | return ret; | |
799 | } | |
800 | ||
801 | static | |
802 | int lib_ring_buffer_file_release(struct inode *inode, struct file *file) | |
803 | { | |
804 | struct lib_ring_buffer *buf = inode->i_private; | |
805 | ||
806 | lib_ring_buffer_iterator_release(buf); | |
807 | return 0; | |
808 | } | |
809 | ||
810 | static | |
811 | int channel_file_open(struct inode *inode, struct file *file) | |
812 | { | |
813 | struct channel *chan = inode->i_private; | |
814 | int ret; | |
815 | ||
816 | ret = channel_iterator_open(chan); | |
817 | if (ret) | |
818 | return ret; | |
819 | ||
820 | file->private_data = chan; | |
821 | ret = nonseekable_open(inode, file); | |
822 | if (ret) | |
823 | goto release_iter; | |
824 | return 0; | |
825 | ||
826 | release_iter: | |
827 | channel_iterator_release(chan); | |
828 | return ret; | |
829 | } | |
830 | ||
831 | static | |
832 | int channel_file_release(struct inode *inode, struct file *file) | |
833 | { | |
834 | struct channel *chan = inode->i_private; | |
835 | ||
836 | channel_iterator_release(chan); | |
837 | return 0; | |
838 | } | |
839 | ||
840 | const struct file_operations channel_payload_file_operations = { | |
a33c9927 | 841 | .owner = THIS_MODULE, |
f3bc08c5 MD |
842 | .open = channel_file_open, |
843 | .release = channel_file_release, | |
844 | .read = channel_file_read, | |
d83004aa | 845 | .llseek = vfs_lib_ring_buffer_no_llseek, |
f3bc08c5 MD |
846 | }; |
847 | EXPORT_SYMBOL_GPL(channel_payload_file_operations); | |
848 | ||
849 | const struct file_operations lib_ring_buffer_payload_file_operations = { | |
a33c9927 | 850 | .owner = THIS_MODULE, |
f3bc08c5 MD |
851 | .open = lib_ring_buffer_file_open, |
852 | .release = lib_ring_buffer_file_release, | |
853 | .read = lib_ring_buffer_file_read, | |
d83004aa | 854 | .llseek = vfs_lib_ring_buffer_no_llseek, |
f3bc08c5 MD |
855 | }; |
856 | EXPORT_SYMBOL_GPL(lib_ring_buffer_payload_file_operations); |