The idea here is to avoid a memory access and conditional branch per
probe site. Instead, the probe is represented by an "unreachable"
unconditional function call. asm goto is used to store the address of
the probe site (represented by a no-op sled) and the address of the
function call into a tracepoint record. Each SDT probe carries a list
of tracepoints.
When the probe is enabled, the no-op sled corresponding to each
tracepoint is overwritten with a jmp to the corresponding label. The
implementation uses smp_rendezvous() to park all other CPUs while the
instruction is being overwritten, as this can't be done atomically in
general. The compiler moves argument marshalling code and the
sdt_probe() function call out-of-line, i.e., to the end of the function.
Per gallatin@ in D43504, this approach has less overhead when probes are
disabled. To make the implementation a bit simpler, I removed support
for probes with 7 arguments; nothing makes use of this except a
regression test case. It could be re-added later if need be.
The approach taken in this patch enables some more improvements:
1. We can now automatically fill out the "function" field of SDT probe
names. The SDT macros let the programmer specify the function and
module names, but this is really a bug and shouldn't have been
allowed. The intent was to be able to have the same probe in
multiple functions and to let the user restrict which probes actually
get enabled by specifying a function name or glob.
2. We can avoid branching on SDT_PROBES_ENABLED() by adding the ability
to include blocks of code in the out-of-line path. For example:
if (SDT_PROBES_ENABLED()) {
int reason = CLD_EXITED;
if (WCOREDUMP(signo))
reason = CLD_DUMPED;
else if (WIFSIGNALED(signo))
reason = CLD_KILLED;
SDT_PROBE1(proc, , , exit, reason);
}
could be written
SDT_PROBE1_EXT(proc, , , exit, reason,
int reason;
reason = CLD_EXITED;
if (WCOREDUMP(signo))
reason = CLD_DUMPED;
else if (WIFSIGNALED(signo))
reason = CLD_KILLED;
);
In the future I would like to use this mechanism more generally, e.g.,
to remove branches and marshalling code used by hwpmc, and generally to
make it easier to add new tracepoint consumers without having to add
more conditional branches to hot code paths.
Reviewed by: Domagoj Stolfa, avg
MFC after: 2 months
Differential Revision: https://reviews.freebsd.org/D44483
When compiling dt_lex.l, flex produces warnings of the form:
dt_lex.l:413: warning, trailing context made variable due to preceding '|' action
dt_lex.l:412: warning, dangerous trailing context
dt_lex.l:412: warning, dangerous trailing context
Here, trailing context refers to the use of "$", which expands to "/\n".
The meaning behind these warnings is described in the first two
paragraphs of the flex manual's DEFICIENCIES/BUGS section:
Some trailing context patterns cannot be properly matched and generate
warning messages ("dangerous trailing context"). These are patterns
where the ending of the first part of the rule matches the beginning of
the second part, such as "zx*/xy*", where the 'x*' matches the 'x' at
the beginning of the trailing context. (Note that the POSIX draft
states that the text matched by such patterns is undefined.)
For some trailing context rules, parts which are actually fixed-length
are not recognized as such, leading to the above mentioned performance
loss. In particular, parts using '|' or {n} (such as "foo{3}") are
always considered variable-length.
Here, the warnings appear to be bogus in this case. The lexer has no
problem matching either of the referenced patterns, e.g.,
printf("foobar
or
# 1 "asdfasdf
Introduce a small amount of code duplication to silence the warning.
MFC after: 2 weeks
If there are no CTF sections then ctfmerge just has nothing to do; it
should not be an error.
Note that ctfmerge has an option to require CTF:
-t Make sure that all object files have a CTF section.
Before this change, this option explicitly exited without error if none
of the object files have CTF sections, with the comment:
If we're verifying that C files have CTF, it's safe to
assume that in this case, we're building only from assembly
inputs.
PR: 276930
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D43878
In particular, avoid loading the user's .profile file, since that can
have undesirable side effects. Most tests were already careful to do
this.
MFC after: 1 week
This option can be used to specify a format to use in DTrace output.
The following formats are supported:
- json
- xml
- html
- none (default DTrace output)
This is implemented using libxo and integrated into libdtrace. Client
code only works with the following API:
- dtrace_oformat_setup(dtrace_hdl_t *) -- to be called when output is starting.
- dtrace_oformat_teardown(dtrace_hdl_t *) -- to be called when output is finished
- dtrace_oformat(dtrace_hdl_t *) -- check if oformat is enabled.
- dtrace_set_outfp(FILE *) -- sets the output file for oformat.
- Ensure that oformat is correctly checked in the drop handler and record
processing callbacks.
This commit also adds tests which check if the generated output is
valid (JSON, XML) and extends the dtrace(1) describing the structured output.
Reviewed by: markj
Discussed with: phil
MFC after: 2 months
Sponsored by: Innovate UK
Differential Revision: https://reviews.freebsd.org/D41745
libproc's PR_REQUESTED is not implemented on FreeBSD. Remove dead code
in dtrace that would handle it.
Reviewed by: markj
MFC after: 1 week
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D41225
dtrace stored its PR_RLC and PR_KLC flags in the proc_handle's flags,
where they collided with PATTACH_FORCE and PATTACH_RDONLY, respectively.
Thus, Psetflags(PR_KLC) effectively also set the PATTACH_RDONLY flag.
Since the flags are private to dtrace (at least on FreeBSD), store them in
dtrace's own dt_proc structure instead.
On FreeBSD, either PR_RLC or PR_KLC was always set, so remove code that
would handle the case where neither was set.
Reviewed by: markj
MFC after: 1 week
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D41121
The illumos #ifdef's in this file make it harder to read and maintain.
There is little change upstream, and increasing changes for FreeBSD.
Remove the illumos code with `unifdef`. The only manual changes here
are the #includes and #defines at the top, and removing blank lines.
Reviewed by: markj
MFC after: 1 week
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D41173
The pid/killonerror test uses an invalid probe specifier to verify that
the child process is killed. It occasionally fails because the "date"
command is allowed to run long enough to print the date. This is harmless
in this case, but is clearly not ideal.
When the dt_proc_control thread is about to exit, and the dtrace command
forked the child, do not make the child runnable.
Reviewed by: markj
MFC after: 1 week
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D40976
Tracing memcpy() would crash the kernel, because we'd also trace the
memcpy() calls from kinst_invop(). To fix this, introduce kinst_memcpy()
whose arguments are 'volatile', so that we avoid having the compiler
replace it with a regular memcpy().
Reviewed by: markj
Approved by: markj (mentor)
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D40284
By specifying the -d flag, libdtrace will dump the D script after it has
applied syntactical sugar transformations (e.g if/else). This is useful
for both understanding what dt_sugar does, as well as debugging it.
Reviewed by: markj
Approved by: markj (mentor)
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D38732
Because dt_module_update() is highly OS-specific, the ifdefs make it
hard to read and follow what is going on. Also handle dm_modid, and
remove handling of the ".filename" section, since we can easily fetch
the filename from the module's pathname (k_stat->pathname).
Reviewed by: markj
Approved by: markj (mentor)
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D39177
The section will contain static relocations which do not need to be
preserved after linking, and moreover these relocations may reference
symbols that end up getting removed.
Do not set SHF_ALLOC and instead let the linker decide what needs to be
done.
PR: 258872
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Largely untested, as we can't really do anything with user probes
without an implementation of fasttrap. However, this is enough to
generate an embedded dtrace program with `dtrace -G` and link the
generated ELF file.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D38301
When dtrace starts, it tries to detect if the dtrace klds are loaded,
and if not, it loads them by loading the dtraceall kld. This module
depends on most dtrace modules, including systrace for the native
freebsd and freebsd32 ABIs. However, it does not depend on the
systrace_linux klds, as they in turn depend on the linux ABI klds, and
we don't want to load an ABI module that the user has not explicitly
requested. This can leave a naive user in a state where they think all
syscall providers have been loaded, yet linux ABI syscalls are
"invisible" to dtrace.
To fix this, check to see if the linux ABI modules are loaded. If they
are, then load their systrace klds.
Reviewed by: markj, (emaste & jhb, earlier versions)
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D37986
Otherwise, if multiple object files contain references to the same
probe, newish lld will refuse to link them by default, raising a
duplicate global symbol definition error. Previously, duplicate global
symbols with identical absolute st_values were permitted by both lld and
GNU ld.
Since dtrace has no use for probe function symbols after the relocation
performed by dtrace -G, make the symbols weak as well, following a
suggestion from MaskRay.
Reported by: dim
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
kinst does not instantiate its probes automatically, it only does so on
demand via an ioctl interface implemented by /dev/kinst. This change
modifies libdtrace to perform that work when the script references the
kinst provider, similar to the way pid provider probes are implemented.
Reviewed by: markj
MFC after: 3 months
Sponsored by: Google, Inc. (GSoC 2022)
Differential Revision: https://reviews.freebsd.org/D36852
This allows invop-based providers (i.e., fbt and kinst) to expose the
register file of the CPU at the point where the probe fired. It does
not work for SDT providers because their probes are implemented as plain
function calls and so don't save registers. It's not clear what
semantics "regs" should have for them anyway.
This is akin to "uregs", which nominally provides access to the
userspace registers. In fact, DIF already had a DIF_VAR_REGS variable
defined, it was simply unimplemented.
Usage example: print the contents of %rdi upon each call to
amd64_syscall():
fbt::amd64_syscall:entry {printf("%x", regs[R_RDI]);}
Note that the R_* constants are defined in /usr/lib/dtrace/regs_x86.d.
Currently there are no similar definitions for non-x86 platforms.
Reviewed by: christos
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D36799
CTF integers have an explicit width and so can be used to represent
bitfields. Bitfield types emitted by ctfconvert(1) share the name of
the base integer type, so a struct field with type "unsigned int : 15"
will have a type named "unsigned int".
To avoid ambiguity when looking up types by name, add a suffix to names
of bitfield types to distinguish them from the base type. Then, if
ctfmerge happens to order bitfield types before the corresponding base
type in a CTF file, a name lookup will return the base type, which is
always going to be the desired behaviour.
PR: 265403
Reported by: cy
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Passing "0x%p" to sprintf results in double "0x" being printed.
This causes a dtrace script compilation failure when "-d" flag
is specified.
Fix that by removing the extraneous "0x".
Reviewed by: markj
Approved by: mw(mentor)
Obtained from: Semihalf
Sponsored by: Alstom
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D35690
The presence of the latter causes a link error when building a
statically linked ctfdump(1) because libc defines the same symbol.
libc's warn() is defined as a weak symbol and so does not cause the same
problem, but let's just use libc's version.
Reported by: stephane rochoy <stephane.rochoy@stormshield.eu>
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Notable upstream pull request merges:
#9078: log xattr=sa create/remove/update to ZIL
#11919: Cross-platform xattr user namespace compatibility
#13014: Report dnodes with faulty bonuslen
#13016: FreeBSD: Fix zvol_cdev_open locking
#13019: spl: Don't check FreeBSD rwlocks for double initialization
#13027: Fix clearing set-uid and set-gid bits on a file when
replying a write
#13031: Add enumerated vdev names to 'zpool iostat -v' and
'zpool list -v'
#13074: Enable encrypted raw sending to pools with greater ashift
#13076: Receive checks should allow unencrypted child datasets
#13098: Avoid dirtying the final TXGs when exporting a pool
#13172: Fix ENOSPC when unlinking multiple files from full pool
Obtained from: OpenZFS
OpenZFS commit: a86e089415
For some reason I can't reproduce this locally, but Jenkins complains.
Reported by: Jenkins
Fixes: bdf290cd3e ("ctf: Add v3 support to CTF tools, ctf{convert,dump,merge}")
