Add option for user space core dump on Linux #6688
Conversation
mikezhang1234567890
requested review from
charliegracie,
youngar,
rwy7 and
babsingh
as code owners
Add an environment variable toggle to switch to creating a system core dump manually in the current directory rather than rely on the kernel to create a dump automatically from a signal and storing it where the OS dictates. Signed-off-by: Mike Zhang <[email protected]>
mikezhang1234567890
force-pushed
the
linuxddr
branch
from
f3636ab
to
b864523
Compare
|
After discussion with @babsingh, we determined the following changes to improve this PR:
|
|
re #6688 (comment): The following points were missed:
All the above feedback was verbally given on 7 Sept 22. |
|
@mikezhang1234567890 Remove the draft state once the PR is ready for review. @keithc-ca Can you help with reviewing the PR once it is GTG? |
| pthread_kill(pthread_self(), J9_DUMP_SIGNAL); | ||
| #endif /* defined(LINUX) || defined(OSX) */ | ||
| #if defined(LINUX) | ||
| if (getenv("J9_USER_DUMP")) { |
There was a problem hiding this comment.
I think there should be a programmatic way of requesting a user-space dump rather than using the environment, but this should compare with NULL:
if (NULL != getenv("J9_USER_DUMP")) {
There was a problem hiding this comment.
J9_USER_DUMP might be a name that suits openj9, but other downstream projects would probably disagree.
| #if defined(LINUX) | ||
| if (getenv("J9_USER_DUMP")) { | ||
| if (0 != userspace_dump_create(portLibrary, filename)) { | ||
| fprintf(stderr, "core dump creation failed\n"); |
There was a problem hiding this comment.
The failure message should be written to filename as is the documented behavior of this function.
portLibrary->str_printf(portLibrary, filename, EsMaxPath, "core dump creation failed");
| @@ -40,6 +40,5 @@ typedef struct MarkAllPagesWritableHeader { | |||
| uintptr_t renameDump(struct OMRPortLibrary *portLibrary, char *filename, pid_t pid, int signalNumber); | |||
| char *markAllPagesWritable(struct OMRPortLibrary *portLibrary); | |||
|
|
|||
|
|
|||
|
|
|||
| uintptr_t userspace_dump_create(struct OMRPortLibrary *portLibrary, char *filename); | |||
|
|
|||
There was a problem hiding this comment.
Can you please remove this extra blank line.
| @@ -146,6 +149,10 @@ renameDump(struct OMRPortLibrary *portLibrary, char *filename, pid_t pid, int si | |||
|
|
|||
| memset(corePatternFormat, 0, sizeof(corePatternFormat)); | |||
|
|
|||
| if (getenv("J9_USER_DUMP")) { | |||
There was a problem hiding this comment.
Comparison with NULL should be explicit.
| @@ -146,6 +149,10 @@ renameDump(struct OMRPortLibrary *portLibrary, char *filename, pid_t pid, int si | |||
|
|
|||
| memset(corePatternFormat, 0, sizeof(corePatternFormat)); | |||
|
|
|||
| if (getenv("J9_USER_DUMP")) { | |||
| return stat(filename, &attrBuf); | |||
There was a problem hiding this comment.
If this fails, filename is supposed to say why.
| size_t bytes_read = 0; | ||
| char read_buffer[EsMaxPath]; | ||
|
|
||
| snprintf(proc_filename + 6, 32 - 6, "%d/stat", ppid); |
There was a problem hiding this comment.
Same /proc/ comment as before.
| /* fill out prpsinfo and general process prstatus info */ | ||
| prpsinfo->pr_uid = getuid(); | ||
| prpsinfo->pr_gid = getgid(); | ||
| if (read_word_from_stat(proc_fd, read_buffer)) { |
There was a problem hiding this comment.
Explicit comparisons with zero.
|
|
||
| static int | ||
| write_thread_notes(int addrsize, int outfile_fd, off_t *outfile_off, prstatus_t *global_prstatus) { | ||
| char corename[8] = "CORE"; /* size 8 for alignment */ |
| } | ||
|
|
||
| /* skip first 13 entries in stat file */ | ||
| for (i=0; i < 13; i++) { |
| } | ||
| file_off += sizeof corename; | ||
|
|
||
| if(!write_note_auxv_data(outfile_fd, &file_off, ¬e_hdr.n_descsz)) { |
Add an environment variable toggle to switch to creating a system core dump
manually in the current directory rather than rely on the kernel to create a
dump automatically from a signal and storing it where the OS dictates.
Fixes #6300 by providing an alternative to the kernel made system dump.