[lttng-dev] [PATCH lttng-ust v2] Fix: allocate ts_end in ringbuffer shared memory

Fri May 10 15:24:44 EDT 2019

On 2019-05-10 2:08 p.m., Jonathan Rajotte-Julien wrote:
> Hi,
> 
> On Fri, May 10, 2019 at 12:33:42PM -0400, Sebastien Boisvert wrote:
>> On 2019-05-10 11:26 a.m., Gabriel-Andrew Pollo-Guilbert wrote:
>>
>> Hello Gabriel-Andrew,
>>
>>> This patch allocates the memory used by the ts_end field added by commit
>>> 6c737d05. When allocating lots of subbuffer for a channel (512 or more),
>>> zalloc_shm() will fail to allocate all the objects because the allocated memory
>>> map didn't take account the newly added field.
>>>
>>> lttng-tools version: b14f53d4 (2.12.0-pre)
>>>
>>> Steps to reproduce the bug:
>>>
>>> 	1. lttng-sessiond -vvv --verbose-consumer
>>
>> I need to killall existing lttng-sessiond daemon, right ?
> 
> No, since we are in userspace this can be done with a root lttng-sessiond
> present. But in real life -> yes.
> 
>>
>>> 	2. start a traced application
>>
>> I used lttng-ust/doc/examples/demo/.
> 
> Make sure that the app is long lived (easy-ust with a while loop for example).
> This is only to force the allocation on lttng start.
> 
>>
>>> 	3. lttng create "test-sesssion"
>>                                 ^^^
>> There are too many s in session.
> 
> Does not matter since it is the session name here and subsequent call are done
> in the context on that session. Still a good catch.

Well, I thought that this typo was going to make it into the "git log"..

> 
>>
>>
>>> 	4. lttng enable-channel --userspace --num-subbuf 512 --subbuf-size 8k --overwrite channel
>>> 	5. lttng enable-event -u -a -c channel
>>> 	6. lttng start
>>>
>>> After these steps, the following error message show should be thrown:
>>>
>>> 	Error: ask_channel_creation consumer command failed
>>> 	Error: Error creating UST channel "channel" on the consumer daemon
>>
>> When I build lttng-tools b14f53d4
>>
>> I get this error:
>>
>> make[3]: Entering directory '/home/sboisvert/open-source/lttng.org/Source/lttng-tools/src/common/ust-consumer'
>>   CC       ust-consumer.lo
>> ust-consumer.c: In function 'lttng_ustconsumer_recv_cmd':
>> ust-consumer.c:1459:7: error: 'struct ustctl_consumer_channel_attr' has no member named 'blocking_timeout'
>>    attr.blocking_timeout= msg.u.ask_channel.blocking_timeout;
>>        ^
>> ust-consumer.c: In function 'lttng_ustconsumer_sample_snapshot_positions':
>> ust-consumer.c:2225:9: warning: implicit declaration of function 'ustctl_snapshot_sample_positions'; did you mean 'ustctl_snapshot_get_produced'? [-Wimplicit-function-declaration]
>>   return ustctl_snapshot_sample_positions(stream->ustream);
>>          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>          ustctl_snapshot_get_produced
>> Makefile:548: recipe for target 'ust-consumer.lo' failed
> 
> Are you sure you are building lttng-tools against lttng-ust
> 6c737d0594cac0d969e1948ea1ed55c15be9cec8 or master?

My email said that I was using lttng-ust, git tag v2.9.4.

> 
> In any case, this does not seems related to this particular issue. Please start
> a new email thread.
> 

I wanted to test the change "Fix: allocate ts_end in ringbuffer shared memory" on my computer.

Building userspace-rcu, lttng-ust, and lttng-tools was a prerequisite.

But, fair enough, I should have started a new thread.

>>
>>
>> There seems to be 2 declarations of struct ustctl_consumer_channel_attr.
>>
>> In file lttng-tools/src/bin/lttng-sessiond/lttng-ust-ctl.h,
>> struct ustctl_consumer_channel_attr has the attribute blocking_timeout.
>>
>> In file lttng-ust/include/lttng/ust-ctl.h, struct ustctl_consumer_channel_attr
>> does not have the attribute.
> 
> Currently on 6c737d0594cac0d969e1948ea1ed55c15be9cec8 (almost master):
> 
> struct ustctl_consumer_channel_attr {
> 	enum lttng_ust_chan_type type;
> 	uint64_t subbuf_size;			/* bytes */
> 	uint64_t num_subbuf;			/* power of 2 */
> 	int overwrite;				/* 1: overwrite, 0: discard */
> 	unsigned int switch_timer_interval;	/* usec */
> 	unsigned int read_timer_interval;	/* usec */
> 	enum lttng_ust_output output;		/* splice, mmap */
> 	uint32_t chan_id;			/* channel ID */
> 	unsigned char uuid[LTTNG_UST_UUID_LEN]; /* Trace session unique ID */
> 	int64_t blocking_timeout;			/* Blocking timeout (usec) */
> } LTTNG_PACKED;
> 
> This field was introduced in b2c5f61a9. Suggesting that you lttng-ust tree is
> out of date.
> 
>>
>>
>> They are exactly the same, except the missing blocking_timeout attribute.
>>
>>
>> [sboisvert at GT480:Source]$ grep -A 10 "struct ustctl_consumer_channel_attr {" lttng-ust/include/lttng/ust-ctl.h
>> struct ustctl_consumer_channel_attr {
>> 	enum lttng_ust_chan_type type;
>> 	uint64_t subbuf_size;			/* bytes */
>> 	uint64_t num_subbuf;			/* power of 2 */
>> 	int overwrite;				/* 1: overwrite, 0: discard */
>> 	unsigned int switch_timer_interval;	/* usec */
>> 	unsigned int read_timer_interval;	/* usec */
>> 	enum lttng_ust_output output;		/* splice, mmap */
>> 	uint32_t chan_id;			/* channel ID */
>> 	unsigned char uuid[LTTNG_UST_UUID_LEN]; /* Trace session unique ID */
>> } LTTNG_PACKED;
>>
>> [sboisvert at GT480:Source]$ grep -A 11 "struct ustctl_consumer_channel_attr {" lttng-tools/src/bin/lttng-sessiond/lttng-ust-ctl.h
>> struct ustctl_consumer_channel_attr {
>> 	enum lttng_ust_chan_type type;
>> 	uint64_t subbuf_size;			/* bytes */
>> 	uint64_t num_subbuf;			/* power of 2 */
>> 	int overwrite;				/* 1: overwrite, 0: discard */
>> 	unsigned int switch_timer_interval;	/* usec */
>> 	unsigned int read_timer_interval;	/* usec */
>> 	enum lttng_ust_output output;		/* splice, mmap */
>> 	uint32_t chan_id;			/* channel ID */
>> 	unsigned char uuid[LTTNG_UST_UUID_LEN]; /* Trace session unique ID */
>> 	int64_t blocking_timeout;			/* Blocking timeout (usec) */
>> } LTTNG_PACKED;
>>
>> [sboisvert at GT480:Source]$ diff -u <(grep -A 11 "struct ustctl_consumer_channel_attr {" lttng-tools/src/bin/lttng-sessiond/lttng-ust-ctl.h) <(grep -A 10 "struct ustctl_consumer_channel_attr {" lttng-ust/include/lttng/ust-ctl.h)
>> --- /dev/fd/63	2019-05-10 12:32:23.562246820 -0400
>> +++ /dev/fd/62	2019-05-10 12:32:23.562246820 -0400
>> @@ -8,5 +8,4 @@
>>  	enum lttng_ust_output output;		/* splice, mmap */
>>  	uint32_t chan_id;			/* channel ID */
>>  	unsigned char uuid[LTTNG_UST_UUID_LEN]; /* Trace session unique ID */
>> -	int64_t blocking_timeout;			/* Blocking timeout (usec) */
>>  } LTTNG_PACKED;
>>
>>  
>> My configure commands (followed by "make install"):
>>
>> [sboisvert at GT480:userspace-rcu]$ ./configure --prefix=/home/sboisvert/open-source/lttng.org/Build/userspace-rcu/v0.11.0
>>
>> [sboisvert at GT480:lttng-ust]$ ./configure --prefix=/home/sboisvert/open-source/lttng.org/Build/lttng-ust/v2.9.4 LDFLAGS=-L/home/sboisvert/open-source/lttng.org/Build/userspace-rcu/v0.11.0/lib
>>
>> [sboisvert at GT480:lttng-tools]$ ./configure --prefix=/home/sboisvert/open-source/lttng.org/Build/lttng-tools/b14f53d4 LDFLAGS=-L/home/sboisvert/open-source/lttng.org/Build/lttng-ust/v2.9.4/lib CPPFLAGS=-I/home/sboisvert/open-source/lttng.org/Build/lttng-ust/v2.9.4/include
>>
> 
> This is probably an artefact but all your path have 2.9 in them. This work is
> done against master of lttng-tools and lttng-ust.
> 
> Our CI does not indicate any failure for the master branches. [1]
> 
> [1] 
> 
> Cheers
> 