libtracecmd: Support changing /proc/kallsyms
v3 cover libtracecmd: Support changing /proc/kallsyms
v2: https://lore.kernel.org/linux-trace-devel/[email protected]/
v2 -> v3: Add a comment and some newlines (Steven).
v1: https://lore.kernel.org/linux-trace-devel/[email protected]/
v1 -> v2: Rewrite the size instead of reading the file twice (Steven).
Turns out exposing do_lseek() is not necessary, since everything is in the same C file.
Hi,
this series fixes tracing in presence of frequent BPF or kernel module loads/unloads. Patch 1 is a small required improvement; patch 2 is the actual implementation.
The problem can be reproduced like this:
# while true; do rmmod loop; modprobe loop; done &
# while trace-cmd record -p function_graph /bin/true; do :; done
plugin 'function_graph'
libtracecmd: Invalid argument
error in size of file '/proc/kallsyms'
trace-cmd: Invalid argument
Error creating output fileBest regards,
Ilya
Ilya Leoshkevich (2):
libtracecmd: Support querying position within a new compressed block
libtracecmd: Support changing /proc/kallsyms
lib/trace-cmd/trace-compress.c | 5 ++++- lib/trace-cmd/trace-output.c | 38 +++++++++++++++++++++++++++++++++- 2 files changed, 41 insertions(+), 2 deletions(-)
Signature · 1 line
2.50.1
PATCH v3 1/2 libtracecmd: Support querying position within a new compressed block
tracecmd_compress_lseek() refuses to return position within a new block, because handle->buffer is NULL. Add support for this; do not implement actually seeking within a new block, because this is not required at the moment.
--- lib/trace-cmd/trace-compress.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
lib/trace-cmd/trace-compress.c +4 -1
@@ -132,9 +132,12 @@ off_t tracecmd_compress_lseek(struct tracecmd_compression *handle, off_t offset, {unsigned long p; -if (!handle || !handle->buffer) +if (!handle)return (off_t)-1 ; +if (!handle->buffer) +return (whence ==SEEK_CUR && offset ==0 ) ?0 : (off_t)-1 ; +switch (whence) {case SEEK_CUR : p = handle->pointer + offset;
Signature · 1 line
2.50.1
PATCH v3 2/2 libtracecmd: Support changing /proc/kallsyms
In reply to Ilya LeoshkevichRunning BPF selftests under trace-cmd intermittently fails with:
error in size of file '/proc/kallsyms'
This is because these selftests load and unload BPF programs. bpf_prog_put() uses workqueues and RCU, so these programs disappear from /proc/kallsyms after a delay.
trace-cmd reads /proc/kallsyms twice: the first time to compute its size, and the second time to copy it into the trace file. If the resulting sizes don't match, which is what happens in this case, recording fails.
Fix by updating the size. This will work even for sending trace data over the network thanks to trace_msg_cache.
--- lib/trace-cmd/trace-output.c | 38 +++++++++++++++++++++++++++++++++++- 1 file changed, 37 insertions(+), 1 deletion(-)
lib/trace-cmd/trace-output.c +37 -1
@@ -577,6 +577,39 @@ __hidden int tcmd_out_update_section_header(struct tracecmd_output *handle, tsizreturn 0 ; } +/* +* For files that can change while reading them (like /proc/kallsyms) the +* normal "size != check_size" can fail. That's because from the time the file +* size is determined and written into the output to the time the actual data +* is copied into the output, the size can change. In this case, the size that +* was stored in the output needs to be updated. +*/ +static int update_endian_4 (struct tracecmd_output *handle, +off_t offset,int val) +{ +tsize_t current; +int endian4; + +if (offset == (off_t)-1 ) +return -1 ; + + current =do_lseek (handle,0 ,SEEK_CUR ); +if (current == (off_t)-1 ) +return -1 ; + +if (do_lseek (handle, offset,SEEK_SET ) == (off_t)-1 ) +return -1 ; + + endian4 =convert_endian_4 (handle, val); +if (tcmd_do_write_check (handle, &endian4,4 )) +return -1 ; + +if (do_lseek (handle, current,SEEK_SET ) == (off_t)-1 ) +return -1 ; + +return 0 ; +} +static int save_string_section (struct tracecmd_output *handle,bool compress) {enum tracecmd_section_flags flags =0 ;
@@ -1135,6 +1168,7 @@ static int read_proc_kallsyms(struct tracecmd_output *handle, bool compress)enum tracecmd_section_flags flags =0 ;unsigned int size, check_size, endian4;const char *path ="/proc/kallsyms" ; +off_t size_offset;tsize_t offset;struct stat st;int ret;
@@ -1166,13 +1200,15 @@ static int read_proc_kallsyms(struct tracecmd_output *handle, bool compress) } size =get_size (path); endian4 =convert_endian_4 (handle, size); + size_offset =do_lseek (handle,0 ,SEEK_CUR ); ret =tcmd_do_write_check (handle, &endian4,4 );if (ret) goto out;set_proc_kptr_restrict (0 ); check_size =copy_file (handle, path); -if (size != check_size) { +if (size != check_size && +update_endian_4 (handle, size_offset, check_size)) { errno =EINVAL ;tracecmd_warning ("error in size of file '%s'" , path);set_proc_kptr_restrict (1 );
Signature · 1 line
2.50.1
Quoted 38 lines · Ilya Leoshkevich
v2:
https://lore.kernel.org/linux-trace-devel/[email protected]/
v2 -> v3: Add a comment and some newlines (Steven).v1:
https://lore.kernel.org/linux-trace-devel/[email protected]/
v1 -> v2: Rewrite the size instead of reading the file twice (Steven).
Turns out exposingdo_lseek()is not necessary, since everything is in the same C file.Hi,
this series fixes tracing in presence of frequent BPF or kernel module
loads/unloads. Patch 1 is a small required improvement; patch 2 is the
actual implementation.The problem can be reproduced like this:
# while true; do rmmod loop; modprobe loop; done &# while trace-cmd record -p function_graph /bin/true; do :; done plugin 'function_graph' libtracecmd: Invalid argument error in size of file '/proc/kallsyms' trace-cmd: Invalid argument Error creating output fileBest regards,
IlyaIlya Leoshkevich (2):
libtracecmd: Support querying position within a new compressed block
libtracecmd: Support changing/proc/kallsyms
Quoted 4 lines
lib/trace-cmd/trace-compress.c | 5 ++++-
lib/trace-cmd/trace-output.c | 38
+++++++++++++++++++++++++++++++++-2 files changed, 41 insertions(+), 2 deletions(-)
Gentle ping.