* enable func call injection on delve for ppc64le
* Function call injection on Delve/ppc64le, modified DWARF encoding and decoding for floating point registers to make floatsum test work
* Function call injection on Delve/ppc64le cleanup
* skip PIE tests for function call injection on other packages
* Address review comments
* accounted for additional skipped PIE tests for function call injection
* Code cleanup and undoing revert of previous commit
* Enable function call injection only on 1.22 and above and some cleanup
* additional cleanup, go fmt run
* Debug function call tests fail on ppc64le/PIE mode adjusted the backup_test_health.md file accordingly
Miscellaneous non-functional changes to prepare for adding support for
follow-exec mode on Windows:
- removed (*nativeProcess).wait function from Windows backend (unused).
- move close of ptraceDoneChan from release to handlePtraceFuncs, this
makes postExit callable by a function executed by execPtraceFunc.
- change addTarget to detach before creating the target object if we
don't actually want to attach to the child process, also moved the
detach call to (*processGroup).add instead of having one in addTarget
and one in the code that calls (*processGroup).add.
- changed Detach to be a method of TargetGroup/ProcessGroup, the
Windows backend will need access to the process group to call
WaitForDebugEvent.
- moved resume method to processGroup. First all threads stopped at a
breakpoint need to be stepped, then all other threads can be resumed.
This is true also for linux even though it didn't cause the current
tests to fail.
Adds a waitfor option to 'dlv attach' that waits for a process with a
name starting with a given prefix to appear before attaching to it.
Debugserver on macOS does not support follow-fork mode, but has this
feature instead which is not the same thing but still helps with
multiprocess debugging somewhat.
* logflags,proc: flag to log stacktrace execution
Add a log flag to write logs about what the stacktracer does.
* proc: read context from sigtrampgo, fixes TestCgoStacktrace2 on 1.21
Changes stacktrace code to read the signal context from the arguments
of sigtrampgo.
Also changes the automatic fatalthrow breakpoint for go 1.21.
In combination these two changes fix TestCgoStacktrace2 on Go 1.21 on
various platforms.
* pkg/proc: pad variable mem in extractVarInfoFromEntry
On 64 bit system, the byte size of the following struct is 16:
type myStruct struct {
a int
b uint32
}
But extractVarInfoFromEntry only allocates a mem of 12 bytes for it.
When calling method of this struct with the "call" command, it will
result in this error:
write out of bounds
This patch extends the mem by adding padding bytes to the end of the
mem.
Fixes#3364.
* move the padding logic into newCompositeMemory
Read the command line of the main target process as well as any other
process Delve attaches to in follow exec mode.
The command line can be viewed using the 'target list' command.
In follow exec mode this command line is used to match the follow exec
regex to decide whether or not to attach to a child process.
On macOS or when using rr the list of arguments is not available for
attached processes since there is no way to use the gdb serial protocol
to read it.
Fixes#2242
When using Step on a function that has a dynamic CALL instruction we
set a Step breakpoint on the call.
When it is hit we determine the destination of the CALL by looking at
registers, set a breakpoint there and continue.
If the Step breakpoint is hit simultaneously with a normal breakpoint
our Step logic will take precedence and the normal breakpoint hit will
be hidden from the user.
Move the Step logic to a breaklet callback so that it does not
interfere with the decision to stop.
Unrecovered-panic and fatal-throw were no longer part of the breakpoint
list because starting in 37e44bf they were created before the logical
breakpoints map was switched to the logical breakpoints map of the
target group.
Adds the ability to automatically debug child processes executed by the
target to the linux native backend.
This commit does not contain user interface or API to access this
functionality.
Updates #2551
* Test windows/arm64 pipeline
* update build script to support windows/arm64
* skip TestLaunchRequestWithRelativeExecPath is symblink can't be created
* partially fix and skip TestCgoStacktrace
* update backend health docs
* update
* log test output
* skip starbind test on windows arm64
* skip starbind test on windows arm64
* skip rtype test on windows arm64
* skip pie backend tests on windows/arm64
* fix tests
* skip function calls test on windows/arm64
* fix tests
* revert hardware breakpoint test relax
* add pie test clarification
* skip symlink test only on windows
* skip TestStepConcurrentDirect
* readd exp.winarm64
* fix param
* add exp.winarm64 tags
* skip TestGeneratedDoc on winarm64
Go 1.20 switched to uint64 to represent goroutine IDs, we can't
actually follow suit because we have allowed clients to use -1 to refer
to the currently selected goroutine, however we should at least switch
to int64 and also update the rtype check to accept the 1.20 type.
This patch removes the old error-prone way of tracking
whether the tracepoint is for a function entry or
return. Instead of trying to guess, let the data structure
simply tell us directly.
Introduces a new TargetGroup abstraction that can be used to manage
multiple related targets.
No actual management of child processes is implemented here, this is
just a refactoring to make it possible to do that in the future.
Updates #2551
* pkg/proc: convert freebsd ptrace code to cgo
There is little point in having cgo call a custom C function, when the same
can be done directly from cgo (with less code and effort). Split the amd64
specific code into ptrace_freebsd_amd64.go. Also avoid mixing C.ptrace()
with syscall.SYS_PTRACE.
This will make further changes easier - no functional change intended.
* pkg/proc: check return values of ptrace calls on freebsd
The return values of the PT_GETNUMLWPS and PT_GETLWPLIST ptrace calls were
previously unchecked. While these should not fail, panic instead of using
-1 with slice allocation/handling.
* pkg/proc: return *amd64util.AMD64Xstate from freebsd ptraceGetRegset
Return a pointer to a struct, rather than a struct - this simplifies the
code in both the caller and the ptraceGetRegset function, while also avoiding
struct copying.
* pkg/proc: fix floating point register setting on freebsd
The original code could never work - PT_SETREGS on freebsd does not
take an iovec, nor does it set FP registers. Furthermore, the xsave
bytes were not stored in the amd64util.AMD64Xstate struct.
Updates #3001
* pkg/proc: re-enable function call injection on freebsd
Floating point registers can now be set and restored correctly.
This is a partial revert of 51090f003bace1f8cc37b8480ffdb6f6cc91fa5a.
Fixes#3001
* pkg/proc: deduplicate register setting code on freebsd
Adds a LogicalBreakpoint type to represent logical breakpoints
explicitly. Until now logical breakpoints were constructed implicitly
by grouping physical breakpoints together by their LogicalID.
Having logical breakpoints represented explicitly allows for a simpler
implementation of disabled breakpoints, as well as allowing a simple
implementation of delayed breakpoints (#1653, #2551) and in general of
breakpoints spanning multiple processes if we implement debugging
process trees (#2551).
Updates #1653
Updates #2551
* Upgrade FreeBSD version
* proc: fixes concerning call injection on freebsd and rr
On FreeBSD it seems we have problems restoring and setting floating
point registers, since at least restoring is necessary for call
injection to function properly fully disable call injection on FreeBSD.
On rr the same problem exists, however due to the fact that we are
acting on a recording and ending a diversion will restore register
values anyway simply disable the floatsum test.
See also: https://github.com/rr-debugger/rr/issues/3208
Updates #3001
* _scripts/test_linux.sh,_scripts/test_windows.ps1: always return exit code 0 when testing on tip
Same as what we do for test_mac.sh
* proc: support function call injection on arm64
Support function call injection on arm64 with go1.19
Moves breakpoindIDCounter out of BreakpointsMap and into
service/debugger.Debuggger to simplify proc.Target's API and aid with
implementing fork follow mode, where we'll have to debug multiple
processes simultaneously.
* made Pid a method of Target instead of a method of Process
* changed argument of NewTarget to ProcessInternal, since that's the
interface that backends have to implement
* removed warnings about ProcessInternal since there is no way for
users of pkg/proc to access those methods anyway
* made RecordingManipulation an optional interface for backends, Target
supplies its own dummy implementation when the backend doesn't
* inlined small interfaces that only existed to be inlined in
proc.Process anyway
* removed unused function findExecutable in the Windows and no-native
darwin backends
* removed (*EvalScope).EvalVariable, an old synonym for EvalExpression
Normally calls can't be performed on recorded processes, becuase the
future instructions executed by the target are predetermined. The rr
debugger however has a mechanism that allows this by taking the current
state of the recording and allowing it to diverge from the recording,
temporarily.
This commit adds support for starting and ending such diversions around
function calls.
Note: this requires rr version 5.5 of later to work, see:
https://github.com/rr-debugger/rr/pull/2748
Internal breakpoints do not need IDs and assigning them from a counter
separate from the user ID counter can be a cause of confusion.
If a user breakpoint is overlayed on top of a pre-existing internal
breakpoint the temporary ID will be surfaced as if it was a user ID,
possibly conflicting with another user ID.
If a temporary breakpoint is overlayed on top of a pre-existing user
breakpoint and the user breakpoint is first deleted and then
re-created, the user ID will be resurrected along with the breakpoint,
instead of allocating a fresh one.
This change removes internal breakpoint IDs entirely, only user
breakpoints receive an ID.
This patch enables the eBPF tracer backend to parse the ID of the
Goroutine which hit the uprobe. This implementation is specific to AMD64
and will have to be generalized further in order to be used on other
architectures.
Adds a script that check automatically that the the assumptions that
pkg/proc makes about runtime are met by the actual runtime, using a
combination of parsing and magic comments.
Also emits a file describing all the struct fields, constants and
variables of the runtime that we use in pkg/proc.
In order for DAP to support halting the program (either manually or on a breakpoint) performing some action and then resuming execution, there needs to be a way to stop the program without clearing the internal breakpoints. This is necessary for log points and stopping the program to set breakpoints.
The debugging UI makes it seem like a user should be able to set or clear a breakpoint at any time. Adding this ability to complete synchronous requests while the program is running is thus important to create a seamless user experience.
This change just adds a configuration to determine whether the target should clear the stepping breakpoints, and changes the server to use this new mode. Using the new mode means that the DAP server must determine when it expect the next to be canceled and do this manually.
* proc: move breakpoint condition evaluation out of backends
Moves breakpoint condition evaluation from the point where breakpoints
are set, inside ContinueOnce, to (*Target).Continue.
This accomplishes three things:
1. the breakpoint evaluation method needs not be exported anymore
2. breakpoint condition evaluation can be done with a full scope,
containing a Target object, something that wasn't possible before
because ContinueOnce doesn't have access to the Target object.
3. moves breakpoint condition evaluation out of the critical section
where some of the threads of the target process might be still
running.
* proc/native: handle process death during stop() on Windows
It is possible that the thread dies while we are inside the stop()
function. This results in an Access is denied error being returned by
SuspendThread being called on threads that no longer exist.
Delay the reporting the error from SuspendThread until the end of
stop() and only report it if the thread still exists at that point.
Fixes flakyness with TestIssue1101 that was exacerbated by moving
breakpoint condition evaluation outside of the backends.
* pkg/proc: Prefer throw instead of fatalthrow
Currently there is a breakpoint set at runtime.fatalthrow to catch any
situation where the runtime crashes (e.g. deadlock).
When we do this, we go up a frame in order to parse the crash reason.
The problem is that there is no guarentee the "s" variable we attempt to
parse will still be considered "live".
Since runtime.fatalthrow is never called directly, set a breakpoint on
runtime.throw instead and prevent having
to search up a stack frame in order to
get the throw reason.
Fixes#2602
* service/dap: Fix TestFatalThrowBreakpoint
* Reenable TestFatalThrow DAP test
* service/dap: Don't skip test on < 1.17
* service/dap: Update test constraint for 1.16
* pkg/proc: Reinstate runtime.fatalthrow as switchstack exception
* proc: support new Go 1.17 panic/defer mechanism
Go 1.17 will create wrappers for deferred calls that take arguments.
Change defer reading code so that wrappers are automatically unwrapped.
Also the deferred function is called directly by runtime.gopanic, without going through runtime.callN which means that sometimes when a panic happens the stack is either:
0. deferred function call
1. deferred call wrapper
2. runtime.gopanic
or:
0. deferred function call
1. runtime.gopanic
instead of always being:
0. deferred function call
1. runtime.callN
2. runtime.gopanic
the isPanicCall check is changed accordingly.
* test: miscellaneous minor test fixes for Go 1.17
* proc: resolve inlined calls when stepping out of runtime.breakpoint
Calls to runtime.Breakpoint are inlined in Go 1.17 when inlining is
enabled, resolve inlined calls in stepInstructionOut.
* proc: add support for debugCallV2 with regabi
This change adds support for the new debug call protocol which had to
change for the new register ABI introduced in Go 1.17.
Summary of changes:
- Abstracts over the debug call version depending on the Go version
found in the binary.
- Uses R12 instead of RAX as the debug protocol register when the binary
is from Go 1.17 or later.
- Creates a variable directly from the DWARF entry for function
arguments to support passing arguments however the ABI expects.
- Computes a very conservative stack frame size for the call when
injecting a call into a Go process whose version is >=1.17.
Co-authored-by: Michael Anthony Knyszek <mknyszek@google.com>
Co-authored-by: Alessandro Arzilli <alessandro.arzilli@gmail.com>
* TeamCity: enable tests on go-tip
* goversion: version compatibility bump
* TeamCity: fix go-tip builds on macOS/arm64
Co-authored-by: Michael Anthony Knyszek <mknyszek@google.com>
We told clients that further loading of variables can be done by
specifying a type cast using the address of a variable that we
returned.
This does not work for registerized variables (or, in general,
variables that have a complex location expression) because we don't
give them unique addresses and we throw away the compositeMemory object
we made to read them.
This commit changes proc so that:
1. variables with location expression divided in pieces do get a unique
memory address
2. the compositeMemory object is saved somewhere
3. when an integer is cast back into a pointer type we look through our
saved compositeMemory objects to see if there is one that covers the
specified address and use it.
The unique memory addresses we generate have the MSB set to 1, as
specified by the Intel 86x64 manual addresses in this form are reserved
for kernel memory (which we can not read anyway) so we are guaranteed
to never generate a fake memory address that overlaps a real memory
address of the application.
The unfortunate side effect of this is that it will break clients that
do not deserialize the address to a 64bit integer. This practice is
contrary to how we defined our types and contrary to the specification
of the JSON format, as of json.org, however it is also fairly common,
due to javascript itself having only 53bit integers.
We could come up with a new mechanism but then even more old clients
would have to be changed.
Ensure that any command executed after the process we are trying to
debug prints a correct and consistent exit status.
Previously the exit code was being lost after the first time we printed
that a process has exited. Additionally, certain commands would print
the PID of the process and other would not. This change makes everything
more correct and consistent.
Adds the low-level support for watchpoints (aka data breakpoints) to
the native linux/amd64 backend.
Does not add user interface or functioning support for watchpoints
on stack variables.
Updates #279
When restarting we must take care of setting breakpoint IDs correctly
so that enabled breakpoints do not end up having the same ID as a
disabled breakpoint, also we must make sure that breakpoints created
after restart will not get an ID already used by a disabled breakpoint.
* Adds toggle command
Also adds two rpc2 tests for testing the new functionality
* Removes Debuggers' ToggleBreakpoint method
rpc2's ToggleBreakpoint now calls AmendBreakpoint
Refactors the ClearBreakpoint to avoid a lock.
This adds a workaround for the bug described at:
https://github.com/golang/go/issues/25841
Because dsymutil running on PIE does not adjust the address of
debug_frame entries (but adjusts debug_info entries) we try to do the
adjustment ourselves.
Updates #2346
When cgo is used the address of the g struct is saved on the special
register TPIDR_EL0. Because executing C code could overwrite the
contents of R28 that normally contains the address of g we should read
it from TPIDR_EL0 instead when runtime.iscgo is set.
* proc/core: off-by-one error reading ELF core files
core.(*splicedMemory).ReadMemory checked the entry interval
erroneously when dealing with contiguous entries.
* terminal,service,proc/*: adds dump command (gcore equivalent)
Adds the `dump` command that creates a core file from the target process.
Backends will need to implement a new, optional, method `MemoryMap` that
returns a list of mapped memory regions.
Additionally the method `DumpProcessNotes` can be implemented to write out
to the core file notes describing the target process and its threads. If
DumpProcessNotes is not implemented `proc.Dump` will write a description of
the process and its threads in a OS/arch-independent format (that only Delve
understands).
Currently only linux/amd64 implements `DumpProcessNotes`.
Core files are only written in ELF, there is no minidump or macho-o writers.
# Conflicts:
# pkg/proc/proc_test.go
