Like we do with unrecovered panics, create a default breakpoint to
catch runtime errors that will cause the program to terminate.
Primarily intended to give users the opportunity to examine the state
of a deadlocked process.
runtime.clone (on some operating systems?) work similarly to fork:
when a thread calls runtime.clone a new thread is created. For a
short period of time both the parent thread and the child thread
appear to be running the same goroutine, until the child thread
adjusts its TLS to point to the correct goroutine.
This means that proc.GetG for a thread that's currently running
'runtime.clone' could be wrong and, consequently, the field
proc.(G).thread of a G struct returned by GoroutinesInfo could be
also wrong. And, finally, that FindGoroutine could sometimes return
a *G with a bad associated thread if the goroutine of interest
recently called 'runtime.clone'.
To work around this problem this commit makes two changes:
1. proc.GetG will return nil for all threads executing runtime.clone.
2. FindGoroutine will return the selected goroutine as long as the
ID matches the one requested.
Change (1) takes care of the 'runtime.clone' problem. If we stop
the target process shortly after a thread executed the SYSCALL
instruction in 'runtime.clone' there are three possibilities:
a. Both the parent thread and the child thread are stopped inside
'runtime.clone'. In this case the state we report is slightly
incorrect, because both threads will be reported as not running any
goroutine when we do know which goorutine one of them (the parent)
is running. This doesn't actually matter since runtime.clone is
always called on the system stack and therefore the goroutine in
runtime.allgs will have the correct location.
b. The child thread managed to exit 'runtime.clone' but the parent
thread didn't. This is similar to (a) but in this case GetG on the
child thread will return the correct goroutine. GetG on the parent
thread will still return (incorrectly) nil but this doesn't matter
for the samer reason as described in (a).
c. The parent thread managed to exit 'runtime.clone' but the child
thread didn't. In this case GetG will return the correct goroutine
both for the parent thread (because it's not executing runtime.clone)
and the child thread.
Change (2) means that even if a thread has a completely nonsensical
TLS (for example because it's set through cgo) evaluating a variable
with a valid GoroutineID will still work as long as it's the current
goroutine (which is the most common case). This change also doubles
as an optimization for FindGoroutine.
Fixes#1469
When compression is applied by default running the DWZ tool on the
resulting binary will crash.
The actual default compression code will look and see if compression
makes any difference and if so replace the normal `.debug_*` section
with `.zdebug_*`. This is why it may not have been hit before. On one of
my workstations I build with 1.12rc1 and no compression happens, but on
a Fedora VM I build and the binary results in compressed DWARF sections.
Adding this flag will make this test more consistent overall.
Add new version to CHANGELOG and update internal version.
Thank you @sbromberger @chainhelen, @dishmaev, @kevin-cantwell,
@Russtopia, @slp, @zavla, @the4thamigo-uk, @altimac and
@GregorioMartinez.
* config/config.go
This change checks for the environmental variable XDG_CONFIG_HOME
before storing the configuration file in the home directory.
Closes#1454
* config/config.go
This change attempts to move the previous config file location to the
new XDG_CONFIG_HOME compliant location on linux systems or when
XDG_CONFIG_HOME is set.
This change removes the old config file location if the file is
successfully moved.
* config/config.go
Consolidate messages about config moving and clean up into one.
Remove extraneous newlines.
* pkg/config: config path error messages print to stderr
This change updates the output location of error messages related to
the config path. This will prevent clients from getting unexpected output
when an error occurs.
* pkg/config: Update path used when returning unable to move config error
This change fixes the error message when Delve was unable to move the config. Previously returned a FileInfo object not the file path.
* pkg/config: Remove check for stderr when printing error messages
* pkg/config: Move success message for moving config to stderr
This prevents the success message from being printed to stdout and
breaking frontends.
* pkg/config: Rename variable to be more descriptive
FindGoroutine can be slow when there are many goroutines running. This
can not be fixed in the general case, however:
1. Instead of getting the entire list of goroutines at once just get a
few at a time and return as soon as we find the one we want.
2. Since FindGoroutine is mostly called by ConvertEvalScope and users
are more likely to request informations about a goroutine running on a
thread, look within the threads first.
As specified in line dwarf/godwarf/type.go:507 the typeCache entry
should always be set before recursive calls to readType to avoid infite
recursion.
Most code in readType already does this but some of the code added
later to handle Go types was wrong.
Fix this bug and also fix the String and Size methods of Type so that
they handle recursive types "correctly" (i.e. they don't recur
forever).
No test is added for this since all legitimate uses of cyclical types
were already handled correctly and the malformed types emitted by the
go compiler will probably be removed in 1.12.
See: https://github.com/golang/go/issues/29264Fixes#1444
The repository is being switched from the personal account
github.com/derekparker/delve to the organization account
github.com/go-delve/delve. This patch updates imports and docs, while
preserving things which should not be changed such as my name in the
CHANGELOG and in TODO comments.
Some build environments (such as when building RPMs) enjoy symlinking
things. This unfortunately causes our tests to fail as we record the
path of fixtures and use that when looking up file:line information.
However, the debug info in the binary records the original file
location, not the location of the symlink.
When casting an integer into a struct pointer we make a fake pointer
variable that doesn't have an address, maybeDereference and
structMember should still work on this kind of Variable.
Fixes#1432
The size of the TLS memory arena needs to be aligned to pointer sized
boundaries on 86x64 architectures, otherwise some programs using cgo
will not have the correct offset for the g struct.
No tests because reproducing this problem depends on behavior of the
GNU ld linker caused by unclear influences.
Fixes#1428.
Goroutine id == 0 is special (there can be many goroutines with id 0).
If the caller of FindGoroutine asks for gid==0 and current thread is
running a goroutine 0 (i.e. either no goroutine or a special
goroutine) return whatever goroutine is running on the current thread.
Updates #1428
* pkg/config: Using current directory to store config if getting user details fails.
This patch is to use current directory to store config file.
user.Current() call fails when dlv is run inside a container. (Alpine)
* Addressing comments
* Removed whitespace
If DebugInfoDirectories is empty, set it to same value as we have in the
default configuration.
This is useful for users which already have an older config file.
Otherwise the later call to ioutil.ReadAll will return an empty array.
This issue is visible from the user's perspective due to the new
debug-info-directories parameter, which does have a value in the default
config.
As a consequence, the first 'dlv' invocation fails to find the separate
debug info files, while the second one works as expected:
$ ./dlv core /usr/bin/dockerd-current /case/dockerd.1234
could not find external debug info file
$ ./dlv core /usr/bin/dockerd-current /case/dockerd.1234
Type 'help' for list of commands.
(dlv) exit
Minidumps are the windows equivalent of unix core files.
This commit updates pkg/proc/core so that it can open and read windows
minidumps.
Updates #794
If a function can be inlined it will appear as two entries in
debug_info. A DW_TAG_subprogram entry with DW_AT_inlined = true (that
will be used as the abstract origin) and a second DW_TAG_subprogram
entry with an abstract origin.
To retrieve the name of this second entry we must load its abstract
origin.
If proc.Step encounters a CALL instruction that points to an address
that isn't associated with any function it should still follow the
CALL.
The circumstances creating this problem do not normally occur, it was
encountered in the process of fixing a bug created by Go1.12.
It was never true that return variables were in the inverse order.
Instead in Go1.11 return variables are saved in debug_info in an
arbitrary order and inverting them just happened to work for this
specific example.
This bug was fixed in Go 1.12, regardless we should attempt to
rearrange return variables anyway.
When a location expression requests a register check that we have as
many bytes in the register as requested and if we don't report the
error.
Updates #1416
Instead of unconditionally returning all present goroutines,
GoroutinesInfo now allows specifying a range (start and count). In
addition to the array of goroutines and the error, it now also returns
the next goroutine to be processed, to be used as 'start' argument on
the next call, or 0 if all present goroutines have already been
processed.
This way clients can avoid eating large amounts of RAM while debugging
core dumps and processes with a exceptionally high amount of goroutines.
Fixes#1403
This patch is a slight refactor to share more code used for genericprocess initialization. There will always be OS/backend specificinitialization, but as much as can be shared should be to preventduplicating of any logic (setting internal breakpoints, loading bininfo,etc).
Users can create sparse maps in two ways, either by:
a) adding lots of entries to a map and then deleting most of them, or
b) using the make(mapType, N) expression with a very large N
When this happens reading the resulting map will be very slow
because loadMap needs to scan many buckets for each entry it finds.
Technically this is not a bug, the user just created a map that's
very sparse and therefore very slow to read. However it's very
annoying to have the debugger hang for several seconds when trying
to read the local variables just because one of them (which you
might not even be interested into) happens to be a very sparse map.
There is an easy mitigation to this problem: not reading any
additional buckets once we know that we have already read all
entries of the map, or as many entries as we need to fulfill the
MaxArrayValues parameter.
Unfortunately this is mostly useless, a VLSM (Very Large Sparse Map)
with a single entry will still be slow to access, because the single
entry in the map could easily end up in the last bucket.
The obvious solution to this problem is to set a limit to the
number of buckets we read when loading a map. However there is no
good way to set this limit.
If we hardcode it there will be no way to print maps that are beyond
whatever limit we pick.
We could let users (or clients) specify it but the meaning of such
knob would be arcane and they would have no way of picking a good
value (because there is no objectively good value for it).
The solution used in this commit is to set an arbirtray limit on
the number of buckets we read but only when loadMap is invoked
through API calls ListLocalVars and ListFunctionArgs. In this way
`ListLocalVars` and `ListFunctionArgs` (which are often invoked
automatically by GUI clients) remain fast even in presence of a
VLSM, but the contents of the VLSM can still be inspected using
`EvalVariable`.
The linux version of proc/native and proc/core contained largely
overlapping implementations of the register handling code, deduplicate
it by moving it into proc/linutil.
This patch allows the `trace` CLI subcommand to display return values of
a function. Additionally, it will also display information on where the
function exited, which could also be helpful in determining the path
taken during function execution.
Fixes#388
Some libraries (for example steam_api64.dll) will send this exception
code to set the thread name on Microsoft VisualC.
In theory it should be fine to send the exception back to the target,
which is responsible for setting a handler for it, in practice in some
cases (steam_api64.dll) this will crash the program. So we'll mask it
instead.
Fixes#1383
Continue did not resume execution after a call to CallFunction if the
point where the process was stopped, before the call CallFunction, was
a breakpoint.
Fixes#1374
The name "where" may confuse users into thinking that this parameter
actually does something where in fact it's just arbitrary text used to
identify the checkpoint.
Fixes#1373
Support for position independent executables (PIE) on the native linux
backend, the gdbserver backend on linux and the core backend.
Also implemented in the windows native backend, but it can't be tested
because go doesn't support PIE on windows yet.
On macOS 10.14 Apple changed the command line tools so that system
headers now need to be manually installed.
Instead of adding one extra install step to the install procedure add a
build tag to allow compilation of delve without the native backend on
macOS. By default (i.e. when using `go get`) this is how delve will be
compiled on macOS, the make script is changed to enable compiling the
native backend if the required dependencies have been installed.
Insure that both configuration still build correctly on Travis CI and
change the documentation to describe how to compile the native backend
and that it isn't normally needed.
Fixes#1359
With this syntax users do not need to type the concrete type of an
interface variable to access its contents. This also sidesteps the
problem where the serialization of a type by go/printer is different
from the one used for debug_info type names.
Updates #1328
We should print something when we exit from continue/next/step/stepout
even if we don't have a source file for the instruction that we are
stopped on.
This is mostly important on macOS where a SIGSEGV will cause 'continue'
to fail with a 'bad access' error (see #852) and the output can be
confusing.
Fixes#1244
Go allows converting a single integer value to string, resulting in a
string containing a single unicode rune with the same code as the value
of the integer.
Allow the same conversion to happen.
Fixes#1322
Add a flag to Stackframe that indicates where the stack frame is the
bottom-most frame of the stack. This allows clients to know whether the
stack trace terminated normally or if it was truncated because the
maximum depth was reached.
Add a truncation message to the 'stack' command.
Fncall.go was written with the assumption that the object returned by
proc.Thread.Registers does not change after we call
proc.Thread.SetPC/etc.
This is true for the native backend but not for gdbserial. I had
anticipated this problem and introduced the Save/SavedRegisters
mechanism during the first implementation of fncall.go but that's
insufficient.
Instead:
1. clarify that the object returned by proc.Thread.Registers could
change when the CPU registers are modified.
2. add a Copy method to Registers that returns a copy of the registers
that are guaranteed not to change when the CPU registers change.
3. remove the Save/SavedRegisters mechanism.
This solution leaves us the option, in the future, to cache the output
of proc.(Thread).Registers, avoiding a system call every time it's
called.
Normally variables that have a named struct as a type will get a
typedef entry as their type, sometimes however the Go linker will
decide to use the DW_TAG_structure_type entry instead.
For consistency always wrap a struct type into a typedef when we are
creating a new variables (see comment in newVariable for exceptions).
This fixes a bug where it would be impossible to call methods on a
global variable.
Add new version to CHANGELOG and update internal version.
Thank you @jaym, @slp, @yasushi-saito, @acshekhara1, @benc153,
@yuval-k, @functionary, @psanford @giuscri, @jsoref, @Carpetsmoker,
@PatrickSchuster, @aarzilli, @derekparker, @ramya-rao-a and @dlsniper.
Evaluates var.method expressions into a variable holding the
corresponding method with the receiver variable as a child, in
preparation for extending CallFunction so that it can call methods.
Changes (*Variable).setValue so that it can be used in CallFunction to
set up the argument frame for the function call, adding the ability to:
- nil nillable types
- set strings to the empty string
- copy from one structure to another (including strings and slices)
- convert any interface type to interface{}
- convert pointer shaped types (map, chan, pointers, and structs
consisting of a single pointer field) to interface{}
This covers all cases where an assignment statement can be evaluated
without allocating memory or calling functions in the target process.
This patch makes it so inlined functions are returned in the
function
list, and also allows users to set breakpoints on the call site of
inlined functions.
Fixes#1261
If we send a process to foreground while the headless instance may get
a SIGTTOU/SIGTTIN, if not ignored this signal will stop the headless.
It's not clear why this only happens the second time we do this but
that's how it is.
Also removes the direct syscall to TIOCSPGRP and lets the go runtime do
it instead.
Fixes#1279
Adds a test that compares the output of our state machine with the
output of the debug_line reader in the standard library and checks that
they produce the same output for the debug_line section of grafana as
compiled on macOS (which is the most interesting case since it uses cgo
and therefore goes through dsymutil).
A few bugs were uncovered and fixed:
1. is_stmt was reset improperly after a DW_LNS_end_sequence instruction
2. basic_block, prologue_end and epilogue_begin were not reset after a
DW_LNS_copy instruction
3. some opcodes were not decoded properly if the debug_line section
declares fewer standard opcodes than we know about.
Fixes#1282
Adds -defer flag to the stack command that decorates the stack traces
by associating each stack frame with its deferred calls.
Reworks proc.next to use this feature instead of using proc.DeferPC,
laying the groundwork to implement #1240.
next/step/stepout should work even if the current frame isn't the
topmost stack frame, but their behavior should be different in that
case (they should continue inside the function of the selected frame).
Most of the logic of next/step/stepout would work correctly if we
simply replaced the call to proc.topframe with something that took a
frame index. However the breakpoint they set on the first deferred
function is wrong, and fixing it requires scanning the defer stack and
matching it to the call stack, something we can't do yet.
Given that enhancing next/step/stepout will take time and the current
behavior confuses users (see issue #1240) return an error if
next/step/stepout are called while the currently selected frame isn't
frame 0.
Updates #1240
There is no guarantee that files will end up stored contiguously in the
debug_line section which makes this optimization wrong in the general
case.
In particular with recent versions of go1.11 and a go.mod file present
the go compiler seems to sometimes produce executables that actually
violate this assumption.
Implements the function call injection protocol introduced in go 1.11
by https://go-review.googlesource.com/c/go/+/109699.
This is only the basic support, see TODO comments in pkg/proc/fncall.go
for a list of missing features.
Updates #119
The JSON-RPC layer doesn't like non-nil error that return an empty string
when the Error method is called and when this happens it shuts down the
connection to the server.
Since we can return a ThreadBlockedError to the client it can't have an
empty string as return value.
Fixes#1251
Core files created by gdb can have sections missing that would be
present in OS created core files.
We work around this by first reading PT_LOAD entries from the exe and
then reading them from the core.
Fixes#1121
Setting the Level field of a logrus logger doesn't actually do anything
since the Level field simply reports the log level of the last log
message emitted on the logger.
The right way to do that is to set logger.Logger.Level.
Also cleans up newline characters from log messages emitted through
logrus and fixes the direction of the arrows in the messages emitted by
rpccommon, which was inconsistent with the arrows of gdbserial.
This pull request makes several changes to delve to allow headless
instancess that are started with the --accept-multiclient flag to
keep running even if there is no connected client. Specifically:
1. Makes a headless instance started with --accept-multiclient quit
after one of the clients sends a Detach request (previously they
would never ever quit, which was a bug).
2. Changes proc/gdbserial and proc/native so that they mark the
Process as exited after they detach, even if they did not kill the
process during detach. This prevents bugs such as #1231 where we
attempt to manipulate a target process after we detached from it.
3. On non --accept-multiclient instances do not kill the target
process unless we started it or the client specifically requests
it (previously if the client did not Detach before closing the
connection we would kill the target process unconditionally)
4. Add a -c option to the quit command that detaches from the
headless server after restarting the target.
5. Change terminal so that, when attached to --accept-multiclient,
pressing ^C will prompt the user to either disconnect from the
server or pause the target process. Also extend the exit prompt to
ask if the user wants to keep the headless server running.
Implements #245, #952, #1159, #1231
Go1.11 switched to the zlib-gnu compression format for debug sections.
Change proc and and a test in dwarf/line to support this change.
Also deletes some dead code from pkg/proc/bininfo.go that hadn't been
used in a long time.
Implements structured logging via Logrus. This gives us a logger per
boundry that we care about, allowing for easier parsing of logs if users
have more than one log option enabled. Also, cleans up a lot of
conditionals in the code by simply silencing the logger at creation as
opposed to conditionally logging everywhere.
To save disk space, some distributions strip the debugging information
from the binaries, putting it in separate files, usually distributed in
separate packages.
To locate the file containing the debug information for a certain
binary, an ELF note named ".note.gnu.build-id" is added to the latter,
which contains a header and a build identification. This identification
can be used to compose a path with this form:
/usr/lib/debug/.build-id/BUILDID[:2]/BUILDID[2:].debug
With this patch, if Delve can't find the debug information in the main
binary, it'll try to locate and parse ".note.gnu.build-id", to compose
and attempt to open a path with the format described above.
Fixes#1206
A user complained on the mailing list about having continuous
"optimized function warnings" on non-optimized functions when using 1.9.
This commit fixes the problem by disabling optimized function detection
on 1.9 and earlier (where it's impossible) and adds a test so we don't
break it again in the future.
If the application being debugged imports two packages with the same
name (but different paths) there was no way to disambiguate the two,
since the character '/' can not appear inside a go identifier.
By allowing users to use a string literal as the package name a package
path can be specified.
We occasionally receive bug reports from users of VSCode-go and GoLand.
GoLand has its own way of capturing the packet exchange between itself
and delve but VSCode-go (supposedly) doesn't.
So far this hasn't been a problem since all bug reports were obvious
bugs on the plugin or easy to reproduce without VSCode-go, but it might
be helpful in the future to have a way to log the packet exchange
between dlv and a frontend.
This commit adds a --log-output option to enable logging of all rpc
messages and changes service/rpccommon accordingly.
Displays the return values of the current function when we step out of
it after executing a step, next or stepout command.
Implementation of this feature is tricky: when the function has
returned the return variables are not in scope anymore. Implementing
this feature requires evaluating variables that are out of scope, using
a stack frame that doesn't exist anymore.
We can't calculate the address of these variables when the
next/step/stepout command is initiated either, because between that
point and the time where the stepout breakpoint is actually hit the
goroutine stack could grow and be moved to a different memory address.
Add a new method "Common" to proc.Process that returns a pointer to a
struct that pkg/proc can use to store its things, independently of the
backend.
This is used here to replace the AllGCache typecasts, it will also be
used to store the return values of the stepout breakpoint and the state
for injected function calls.
go1.11 adds a new extended attribute to all type DIEs containing the
address of the corresponding runtime._type struct, use this attribute
to find the DIE of the concrete type of interface variables when
available.
Go1.11 uses the is_stmt flag of .debug_line to communicate which
assembly instructions are good places for breakpoints, we should
respect this flag.
These changes were introduced by:
* https://go-review.googlesource.com/c/go/+/102435/
Additionally when setting next breakpoints ignore all PC addresses that
belong to the same line as the one currently under at the cursor. This
matches the behavior of gdb and avoids stopping multiple times at the
heading line of a for statement with go1.11.
Change: https://go-review.googlesource.com/c/go/+/110416 adds the
prologue_end flag to the .debug_line section to communicate the end of
the stack-split prologue. We should use it instead of pattern matching
the disassembly when available.
Fixes#550
type of interfaces
'c7cde8b'.
Maps were always loaded with using the default configuration during a
reslice. This is probably a remnant from when we didn't let clients
configure the load parameters.
If dwz binary is available in the system, test delve's ability to find
deduplicated symbols in the DWARF information.
dwzcompression.go contains a small C function (void fortytwo()) which
calls glibc's fprintf with stdin as first argument. Normally, stdin
will be present as a DW_TAG_variable as part of a DW_TAG_compile_unit
named dwzcompression.cgo2.c.
After running dwz on the binary, stdin is moved to a
DW_TAG_partial_unit, which is imported from dwzcompression.cgo2.c with
a DW_TAG_imported_unit.
This test verifies that delve is able to find stdin symbol's type, as a
way to confirm it understands dwz's compressed/deduplicated DWARF
information.
The EnableDWZCompression flag allows tests to request BuildFixture to
run "dwz" on the Fixture's resulting binary to compress/deduplicate its
DWARF sections.
'dwz' is a tool that reduces the size of DWARF sections by
deduplicating symbols. The deduplicated symbols are moved from their
original 'compile unit' to a 'partial unit', which is then referenced
from its original location with an 'imported unit' tag.
In the case of Go binaries, all symbols are located in a single
'compile unit', and the name of each symbol contains a reference to its
package, so 'dwz' is not able to deduplicate them. But still, some C
symbols included in the binary are deduplicated, which also alters the
structure of the DWARF sections, making delve unable to parse them
(crashing in the attempt).
While it would've been possible to simply ignore the C symbols, or
blindly loading all then into BinaryInfo members (packageVars,
Functions...), for correctness sake this change tries to do the right
thing, staging symbols into temporary partialUnit objects, moving them
to BinaryInfo when they are actually requested by a 'imported unit'
tag.
Change the linux verison of proc/native and proc/gdbserial (with
debugserver) so that they let the target process use the terminal when
delve is launched in headless mode.
Windows already worked, proc/gdbserial (with rr) already worked.
I couldn't find a way to make proc/gdbserial (with lldb-server) work.
No tests are added because I can't think of a way to test for
foregroundness of a process.
Fixes#65
Go seems to be generating multiple compilation units that have
the same file. I think this happens for functions that get inlined.
Without this patch, those inlined functions break the ability to set
a breakpoint at other lines in the file. I was able to load the same
binary in gdb and set a breakpoints throughout the file without issue.
```
➜ objdump --dwarf=decodedline automate-gateway | grep handler/users.go
.../handler/users.go:[++]
s/.../handler/users.go 20 0xb6dd88
.../handler/users.go:[++]
s/.../handler/users.go 20 0xb6e50f
.../handler/users.go:[++]
s/automate-gateway/handler/users.go 32 0xb66640
```
Inlined functions are still a little weird. setting a breakpoint on
a function that gets inlined picks the first occurence. That being
said, I think delve should still do something reasonable for the rest
of the lines in the file.
Previously to 4e177bb99acc511897f9cdbfc6cbc50d92ae4725 it was possible
to use --log without arguments to enable logging, this commit reenables
that use case.
The new functionality of the --log flag moved to a new flag name
--logx.
Fixes#1188
Caching the frame in variablesByTag is problematic:
1. accounting for variables that are (partially) stored in registers is
complicated (see issue #1106)
2. for some types (strings, interfaces...) simply creating the Variable
object reads memory, which therefore happens before we can do any
caching.
Instead cache the entire frame when the EvalScope object is created.
The cached range is between the SP value of the current frame and the
CFA of the preceeding frame, if available, or the CFA of the current
frame otherwise.
Fixes#1106
Change memCache so that the preloaded memory is not read immediately
but only after the actual read to the preloaded range.
This allows us to request caching the entire stack frame every time we
create an eval scope and no unnecessary reads will be made even if the
user is just trying to evaluate a global variable.
Vet found the following errors:
pkg/proc/moduledata.go:152: namedata[1] (8 bits) too small for shift of 8
pkg/proc/moduledata.go:170: taglendata[0] (8 bits) too small for shift of 8
The fix is to convert before shifting.
I've seen TestFrameEvaluation fail in CI in the past. It's been a while
since the last time and I couldn't reproduce it locally at all. I'd
like to have some instrumentation in case it happens again.
printcontext should use SelectedGoroutine instead of trusting that the
goroutine running on current thread matches the SelectedGoroutine.
When the user switches to a parked goroutine CurrentThread and
SelectedGoroutine will diverge.
Almost all calls to printcontext are safe, they happen after a continue
command returns when SelectedGoroutine and CurrentThread always agree,
but the calls in frameCommand and listCommand are wrong.
Additionally we should stop reporting an error when the debugger is
stopped on an unknown PC address.
When gdbserial can not find debugserver or lldb-server the error
message is always the same and it complains about lldb-server not being
found.
This is fine on linux (where the backend is unnecessary) but incomplete
on macOS (where the backend is actually used).
Make the error message clearer so that users who do not bother reading
install instructions are not confused.
Go 1.10 added inlined calls to debug_info, this commit adds support
for DW_TAG_inlined_call to delve, both for stack traces (where
inlined calls will appear as normal stack frames) and to correct
the behavior of next, step and stepout.
The calls to Next and Frame of stackIterator continue to work
unchanged and only return real stack frames, after reading each line
appendInlinedCalls is called to unpacked all the inlined calls that
involve the current PC.
The fake stack frames produced by appendInlinedCalls are
distinguished from real stack frames by having the Inlined attribute
set to true. Also their Current and Call locations are treated
differently. The Call location will be changed to represent the
position inside the inlined call, while the Current location will
always reference the real stack frame. This is done because:
* next, step and stepout need to access the debug_info entry of
the real function they are stepping through
* we are already manipulating Call in different ways while Current
is just what we read from the call stack
The strategy remains mostly the same, we disassemble the function
and we set a breakpoint on each instruction corresponding to a
different file:line. The function in question will be the one
corresponding to the first real (i.e. non-inlined) stack frame.
* If the current function contains inlined calls, 'next' will not
set any breakpoints on instructions that belong to inlined calls. We
do not do this for 'step'.
* If we are inside an inlined call that makes other inlined
functions, 'next' will not set any breakpoints that belong to
inlined calls that are children of the current inlined call.
* If the current function is inlined the breakpoint on the return
address won't be set, because inlined frames don't have a return
address.
* The code we use for stepout doesn't work at all if we are inside
an inlined call, instead we call 'next' but instruct it to remove
all PCs belonging to the current inlined call.
* Extend the "frame" command to set the current frame.
Command
frame 3
sets up so that subsequent "print", "set", "whatis" command
will operate on frame 3.
frame 3 print foo
continues to work.
Added "up", "down". They move the current frame up or down.
Implementation note:
This changes removes "scopePrefix" mode from the terminal/command.go and instead
have the command examine the goroutine/frame value to see if it is invoked in a
scoped context.
* Rename Command.Frame -> Command.frame.
updates vendored version of x86asm, adds a symbol lookup function to
pass to the disassembler.
This will show global symbol names in the disassembly like go tool
objdump does.
Registers XMM1 and XMM2 get sometimes clobbered between the time we set
them and the panic. There is no guarantee that they won't in the go
spec so we shouldn't expect any register to keep its value. However
since this seems to only affect 1 and 2 let's try to use 9 and 10
instead.
Every time we read an empty string we accidentally issue a read for 0
bytes at address 0, this is fine for real memory but the core file
reader doesn't like it.
Fixes an issue reported on the mailing list.
If a breakpoint is hit close to process death on a thread that isn't
the group leader the process could die while we are trying to stop it.
This can be easily reproduced by having the goroutine that's executing
main.main (which will almost always run on the thread group leader)
wait for a second goroutine before exiting, then setting a breakpoint
on the second goroutine and stepping through it (see TestIssue1101 in
proc_test.go).
When stepping over the return instruction of main.f the deferred
wg.Done() call will be executed which will cause the main goroutine to
resume and proceed to exit. Both the temporary breakpoint on wg.Done
and the temporary breakpoint on the return address of main.f will be in
close proximity to main.main calling os.Exit() and causing the death of
the thread group leader.
Under these circumstances the call to native.(*Thread).waitFast in
native.(*Thread).halt can hang forever due to a bug similar to
https://sourceware.org/bugzilla/show_bug.cgi?id=12702 (see comment in
native.(*Thread).wait for an explanation).
Replacing waitFast with a normal wait work in most circumstances,
however, besides the performance hit, it looks like in this
circumstances trapWait sometimes receives a spurious SIGTRAP on the
dying group leader which would cause the subsequent call to wait in
halt to accidentally reap the process without noting that it did exit.
Instead this patch removes the call to wait from halt and instead calls
trapWait in a loop in setCurrentBreakpoints until all threads are set
to running=false. This is also a better fix than the workaround to
ESRCH error while setting current breakpoints implemented in 94b50d.
Fixes#1101
If the last entry of the package path contains a '.' the corresponding
DIEs for its types will replace the '.' character with '%2e'. We must
do the same when resolving the package path of the concrete type of an
interface variable.
Fixes#1137
Previously the file handle for the newly created default config was
being closed and thrown away as opposed to returned to the caller to
finish setting up config for the rest of the process. This patch changes
to return a handle to the newly created config so setup can happen as
normal. This fixes a bug where Delve can crash on first run when a
config is not present on the system.
Fixes#1129
debug_info entries can use DW_AT_abstract_origin to inherit the
attributes of another entry, supporting this attribute is necessary to
support DW_TAG_inlined_subroutine.
Go, starting with 1.10, emits DW_TAG_inlined_subroutine entries when
inlining is enabled.
* Handle race between fork and task_for_pid
On macOS a call to fork and a subsequent call to task_for_pid will race each other. This is because the macOS kernel assigns a new proc_t structure early but the new task, thread and uthread come much later. The function exec_mach_imgact in the XNU sources contains this logic.
In a system under load or one with delays in fork processing (i.e. various security software), task_for_pid as currently called by Delve often returns the parent task. This can be seen by printing out the task number around line 86. In a normal system we would see three calls:
-> ~/go/bin/dlv --listen=localhost:59115 --headless=true --api-version=2 --backend=native exec ./___main_go --
Task: 9731
Task: 9731
Task: 9731
API server listening at: 127.0.0.1:59115
This is the result on a system where the race is lost:
-> ~/go/bin/dlv --listen=localhost:59115 --headless=true --api-version=2 --backend=native exec ./___main_go --
Task: 8707
Task: 10499
Task: 10499
could not launch process: could not get thread count
In this latter case, task 8707 is the parent task. The child task of 10499 was desired and hence the error.
This code change checks to make sure the returned task is not that of the parent. If it is, it retries. It's possible other macOS reported Delve issues are the result of this failed race.
* proc: correct formatting
Much like the bug in issue #1031 and commit
f6f6f0bf13e4c708cb501202b83a6327a0f00e31 pointers can also escape to
the heap and then have a zero address (and no children) when we
autodereference.
1. Mark autodereferenced escaped variables with a 0 address as
unreadable.
2. Add guards to the pretty printers for unsafe.Pointer and pointers.
Fixes#1075
Depending on how the runtime schedules our goroutines we can get
unlucky and have the first call to runtime.newstack we intercept be for
a different goroutine (usually the garbage collector).
Only check stacktraces that happen on the same goroutine that executed
main.main.
* command/terminal: allow restart to change process args
Add -args flag to "restart" command. For example, "restart -args a b c" will
pass args a b c to the new process.
Add "-c" flag to pass the checkpoint name. This is needed to disambiguate the
checkpoint name and arglist.
Reverted unnecessary changes.
* Applied reviewer comments.
Vendored argv.
Change the syntax of restart. When the target is is in recording mode, it always
interprets the args as a checkpoint. Otherwise, it interprets the args as
commandline args. The flag "-args" is still there, to handle the case in which
the user wants to pass an empty args on restart.
* Add restartargs.go.
Change "restart -args" to "restart -noargs" to clarify that this flag is used to
start a process with an empty arg.
The runtime calls into g0 in many places, not necessarily using
runtime.systemstack or runtime.asmcgocall.
One example of this is the call to runtime.newstack inside
runtime.morestack.
If we stop the process while one goroutine is executing
runtime.newstack we would be unable to fully scan its stack because we
don't know that we have to switch back to the goroutine stack after
runtime.newstack.
Instead of tracking down every possible way that the runtime switches
to g0 we switch to the goroutine stack immediately after the top of the
stack, unless cgo is being executed on the systemstack.
Fixes#1066
Adds a configuration option (show-location-expr) that when activated
will cause the whatis command to also print the DWARF location
expression for a variable.
So far we have evaluated the locspec "+0" the same way we evaluate all
"+n" locspecs, this means that we turn the current PC into a file:line
pair, then we turn back the file:line into a PC address.
Normally this is harmless, however all autogenerated code returns the
source position "<autogenerated>:1" which resolves back to the very
first autogenerated instruction in the code.
This messes up the behaviour of the "disassemble" command which uses
the locspec "+0" to figure out what code to disassemble if no arguments
are passed.
We should make +0 always resolve to the current PC (of the given scope)
so that clients can use +0 as a default locspec.
On macOS, externally linked programs will have an abbrev for
DW_TAG_subprogram without the haschildren flag set. We should handle
this case instead of expecting all DW_TAG_subprogram entries to have
list of children.
Fixes#1034
When creating a stack trace we should switch between the goroutine
stack and the system stack (where cgo code is executed) as appropriate
to reconstruct the logical stacktrace.
Goroutines that are currently executing on the system stack will have
the SystemStack flag set, frames of the goroutine stack will have a
negative FrameOffset (like always) and frames of the system stack will
have a positive FrameOffset (which is actually just the CFA value for
the frame).
Updates #935
Either the CPU or the kernel may not support the calls we do when
retrieving floating point registers, this isn't an error we should
propagate.
Also improve the error reporint of pkg/proc/native.fpRegisters.
Fixes#1022
StepBreakpoints are set on CALL instructions, when they are hit we
disassemble the current instruction, figure out the destination address
and set a breakpoint after the prologue of the called function.
In order to disassemble the current instruction we disassemble the area
of memory starting from PC and going to PC+15 (because 15 bytes is the
maximum length of one instruction on AMD64). This means that we won't
just disassemble one instruction but also a few instructions following
it ending with one truncated instruction.
This usually works fine but sometimes the disassembler will panic with
an array out of bounds error when trying to disassemble a truncated
instruction. To avoid this problem this commit changes the funciton
disassemble to take one extra parameter, singleInstr, when singleInstr
is set disassemble will quit after disassembling a single instruction.
Replace the unsafe.Pointer type of the buf field of channels with the
appropriate array type, allow expressions accessing member field of the
channel struct.
Fixes#962
Conditional breakpoints with unmet conditions would cause next and step
to skip the line.
This breakpoint changes the Kind field of proc.Breakpoint from a single
value to a bit field, each breakpoint object can represent
simultaneously a user breakpoint and one internal breakpoint (of which
we have several different kinds).
The breakpoint condition for internal breakpoints is stored in the new
internalCond field of proc.Breakpoint so that it will not conflict with
user specified conditions.
The breakpoint setting code is changed to allow overlapping one
internal breakpoint on a user breakpoint, or a user breakpoint on an
existing internal breakpoint. All other combinations are rejected. The
breakpoint clearing code is changed to clear the UserBreakpoint bit and
only remove the phisical breakpoint if no other bits are set in the
Kind field. ClearInternalBreakpoints does the same thing but clearing
all bits that aren't the UserBreakpoint bit.
Fixes#844
Move some duplicate code, related to breakpoints, that was in both
backends into a single place.
This is in preparation to solve issue #844 (conditional breakpoints
make step and next fail) which will make this common breakpoint code
more complicated.
Instead of only tracking a few cherrypicked registers in stack.go track
all DWARF registers.
This is needed for cgo code and for the locationlists emitted by go in
1.10:
* The debug_frame sections emitted by C compilers can not be used
without tracking all registers
* the loclists emitted by go1.10 need all registers of a frame to be
interpreted.
A thread could terminate between the point when we stop for a
breakpoint and the point where we send a stop signal to all threads, if
this happens setCurrentBreakpoints will fail with an error.
We should tolerate this.
For some reason this happens very frequently when running delve on
processes with the race detector enabed.
Adds test command line flag to compile target fixtures using the -race flag.
Multiple tests will fail because of https://github.com/golang/go/issues/22600
but eventually this should work.
1. Use a slice instead of a map to access standard and extended opcodes
(reduces BenchmarkStateMachine from ~12ms/op to ~7ms/op)
2. Cache StateMachine values for the entry point of functions.
gosymtab and gopclntab only contain informations about go code, linked
C code isn't there, we should use debug_line instead to also cover C.
Updates #935
* string to []rune
* string to []byte
* []rune to string
* []byte to string
* any pointer to uintptr
The string, []rune, []byte conversion pairs aligns this to the go
language.
The pointer -> uintptr conversion pair is symmetric to the uintptr ->
pointer that we already have.
Also lets the user specify any size for byte array types instead of
just the ones already used by the program, this can be used to read
arbitrary memory.
Fixes#548, #867
Make 'next' skip deferred functions unless they are called via a panic.
Call to a deferred function through 'return' are predictable, if the
user wants to step into them 'step' can be used but without this change
there is no way to avoid stepping into them.
Implements #956
On macOS we can also stop when we receive a signal,
propagate this reason upwards to the client.
Also clear internal breakpoints after an unrecovered-panic since they
can not be reached anymore.
Fixes#872
debugserver doesn't support qXfer:exec-file:read, and it doesn't return
the executable path in the response to qProcessInfoPID, however we can
find out the executable path by using jGetLoadedDynamicLibrariesInfos.
While we are waiting for the process to exit in native.(*Process).Kill
we could receive queued exception events, those must be continued or
the wait will never finish.
Before go1.9 embedded struct fields had name == "" in runtime and ==
type name in DWARF. After go1.9 both runtime and DWARF use a simplified
version of the type as name.
Embedded structs are distinguished from normal fields by setting a flag
in the runtime.structfield, for runtime, and by adding a custom
attribute in DWARF.
Splits out type parsing and go-specific Type hierarchy from
x/debug/dwarf, replace x/debug/dwarf with debug/dwarf everywhere,
remove x/debug/dwarf from vendoring.
Instead of panicing for sending on a closed channel, detect that the
process has exited and return a proper error message.
This patch also cleans up some spots where the Pid is omitted from the
error.
Fixes#920
* proc/native: make sure debugged executable can be deleted on windows
Delve opens debugged executable to read binary info it
contains, but it never closes the file. Windows will not
let you delete file that is opened. So close Process.bi
in Process.postExit, and actually call Process.postExit
from windows Process.Kill.
Also Windows sends some debugging events
(EXIT_PROCESS_DEBUG_EVENT event in particular) after Delve
calls TerminateProcess. The events need to be consumed by
debugger before debugged process will be released by
Windows. So call Process.waitForDebugEvent after
TerminateProcess in Process.Kill.
Fixes#398
* cmd/dlv: make TestIssue398 pass on darwin
* cmd/dlv: add comment for TestIssue398
* proc/native: wait for debuggee to exit before returning from windows Process.Kill
* proc/native: close process handle before returning from windows killProcess
* proc/native: remove not used Process.Process
When stepping through runtime sometimes the current goroutine will
change. It is impossible to handle this in Next, Step and StepOut but
StepInstruction can reset the current goroutine correctly.
Next will add internal breakpoints with nil condition if it can't find
the current goroutine (possibly because there isn't a current goroutine
because the runtime hasn't been initialized yet).
onNextGoroutine should skip breakpoints with nil condition, otherwise
we'll end up with an internal debugger error trying to walk a nil
expression.
Updates #893
When there's a error reading the stack trace the call stack itself
could be corrupted and we should return the partial stacktrace that we
have.
Fixes#868
Variable lookup is slow because it requires a full scan of debug_info
to check for package variables, this doesn't matter much in interactive
use but can slow down evaluation of breakpoint conditions
significantly.
Providing benchmark proof for this is hard since this effect doesn't
show for small programs with small debug_info sections.
* proc: fix interaction of RequestManualStop and conditional breakpoints
A conditional breakpoint that is hit but has the condition evaluate to
false can block a RequestManualStop from working. If the conditional
breakpoint is set on an instruction that is executed very frequently by
multiple goroutines (or many conditional breakpoints are set) it could
prevent all calls to RequestManualStop from working.
This commit fixes the problem by changing proc.Continue to exit
unconditionally after a RequestManualStop is called.
* proc/gdbserial: fix ContinueOnce getting stuck on macOS
Fixes#902
The concrete value of an interface is always stored as a pointer inside
an interface variable. So far we have followed the memory layout and
reported the type of the 'data' attribute of interfaces as a pointer,
however this makes it impossible to distinguish interfaces with
concrete value of type 'A' from interfaces of concrete value of type
'*A'.
With this changeset when we autodereference pointers when the concrete
type of an interface is not a pointer.
* Fix various issues detected by megacheck
I've ran honnef.co/go/tools/cmd/megacheck and fixed a few of the
things that came up there.
* Cleanup using Gogland
When a Go program is externally linked, the external linker is
responsible for picking the TLS offset. It records its decision in the
runtime.tlsg symbol. Read the offset from that rather than guessing -16.
This implementation causes a regression: 1.4 and earlier will no longer
work.
Other debuggers can be instructed to decorate the stacktrace with the
value of SP. Our SP equivalent is the frame offset, since we can add it
to the Stackframe structure without incurring into added costs we
should, so that frontends can use it if they want.
Implementing proc.Process.Running in a thread safe way is complicated
and nothing actually uses it besides tests, so we are better off
rewriting the tests without Running and removing it.
In particular:
* The call to d.target.Running() in service/debugger/debugger.go
(Restart) can never return true because that line executes while
holding processMutex and all continue operations are also executed
while holding processMutex.
* The call to dbp.Running() pkg/proc/native/proc.go (Detach) can never
return true, because it's only called from
debugger.(*Debugger).detach() which is also always called while
holding processMutex.
Since some tests are hard to write correctly without Process.Running a
simpler interface, Process.NotifyResumed, is introduced.
Fixes#830
RequestManualStop will run concurrently with trapWait, since one writes
dbp.halt and the other reads it dbp.halt should be protected by a
mutex.
Updates #830
type M struct was never used (as far as I know).
type VariableEval interface was used for a brief period of time during
the refactoring, now both its methods are functions.
The authorization prompt on macOS can take a long time to be
acknowledged by the user, we should keep waiting for a connection as
long as the debugserver instance we launched remains alive.
Before this commit our temp breakpoints only checked that we would stay
on the same goroutine.
However this isn't enough for recursive functions we must check that we
stay on the same goroutine AND on the same stack frame (or, in the case
of the StepOut breakpoint, the previous stack frame).
This commit:
1. adds a new synthetic variable runtime.frameoff that returns the
offset of the current frame from the base of the call stack.
This is similar to runtime.curg
2. Changes the condition used for breakpoints on the lines of the
current function to check that runtime.frameoff hasn't changed.
3. Changes the condition used for breakpoints on the return address to
check that runtime.frameoff corresponds to the previous frame in the
stack.
4. All other temporary breakpoints (the step-into breakpoints and defer
breakpoints) remain unchanged.
Fixes#828
Debugserver does not work as documented, "--" needs to be specified to
pass arguments to the target process (but only if it's an argument that
starts with a dash).
Fixes#839
While implementing the gdbserial backend everything was changed to call
Detach to "close" a process so that gdbserial could do its clean up in
a single place. However the native implementation of Detach does not
actually kill processes we launched.
Fixes#821
Lldb-server does not currently support windows/amd64 (it's in
development). And delve's native backend seems to work well on windows.
Disable the lldb backend on windows for now.
Fixes#817
- moved target.Interface into proc as proc.Process
- rename proc.IThread to proc.Thread
- replaced interfaces DisassembleInfo, Continuable and
EvalScopeConvertible with Process.
- removed superfluous Gdbserver prefix from types in the gdbserial
backend.
- removed superfluous Core prefix from types in the core backend.
- always print a header with the path to the file being displayed
- always evaluate the linespec argument, even if a scope prefix is present
Fixes#711, #713
Core files contain a variety of memory mappings either to files or
anonymous regions stored in the core file. These regions can overlap, so
figuring out what exactly to read can be tricky. This commit contains
a data structure, SplicedMemory, which accumulates mappings and reads
from the correct sources.
* pkg/proc: use golang.org/x/arch/x86/x86asm
instead of rsc.io/x86/x86asm
* pkg/dwarf: migrate to github.com/pkg/profile
from github.com/davecheney/profile
* scripts: keep script go files from being considered for the build
scripts/gen-*.go files are scripts for generating documentation
files and don't follow the typical Go package layout. Expected
usage is like
go run scripts/gen-cli-docs.go
* vendor: update vendored packages
There were many changes in delve, and go tool chains since last
vendored package update. I just rerun godpes from scratch.
$ rm vendor/*
$ rm Godeps/Godeps.json
$ go list ./... | grep -v /vendor/ | grep -v /scripts/ | go get -u -t
$ go get -u github.com/mattn/go-colorable
$ go get -u github.com/mattn/go-isatty
$ go list ./... | grep -v /vendor/ | grep -v /scripts/ | godeps save
* proc: Refactor stackIterator to use memoryReadWriter and BinaryInfo
* proc: refactor EvalScope to use memoryReadWriter and BinaryInfo
* proc: refactor Disassemble to use memoryReadWriter and BinaryInfo
* proc: refactor BinaryInfo part of proc.Process to own type
The data structures and associated code used by proc.Process
to implement target.BinaryInfo will also be useful to support a
backend for examining core dumps, split this part of proc.Process
to a different type.
* proc: compile support for all executable formats unconditionally
So far we only compiled in support for loading the executable format
supported by the host operating system.
Once support for core files is introduced it is however useful to
support loading in all executable formats, there is no reason why it
shouldn't be possible to examine a linux coredump on windows, or
viceversa.
* proc: bugfix: do not resume threads on detach if killing
* Replace BinaryInfo interface with BinInfo() method returning proc.BinaryInfo
Detach did not work for processes we attach to via PID.
Linux: we were only detaching from the main thread, all threads are
detached independently
Windows: we must resume all threads before detaching.
macOS: still broken.
Updates #772
Stack barriers were removed in Go 1.9, and thus code that
expected various stack-barrier-related symbols to exist
does not find them. Check for their absence and do not
crash when they are missing. Disable stack-barrier-handling
test for 1.9 and beyond.
Fixes#754.
The rest of the code uses Detach to "close" a proc.Process, the tests
should do the same.
Any cleanup that proc.Process needs to do can then be put inside Detach
and the tests will run it.
The test in question tries to 'next' over a call to wg.Done, this is
not guaranteed to succeed, if the goroutine gets suspended after
wg.Done has notified the waiting group but before returning to
main.dostuff the program could quit before the goroutine is resumed.
It was possible for TestStepParked to pick a runtime goroutine and
attempt to step it. Nothing guarantees that a goroutine other than the
ones we are using to run the code would actually ever resume before the
end of the program.
This makes the test more discerning in its choice of goroutines.
On Windows we can sometimes encounter threads stopped in locations for
which we do not have entries in debug_frame.
These cases seem to be due to calls to Windows API in the go runtime,
we can still produce a (partial) stack trace in this circumstance by
following frame pointers (starting with BP).
We still prefer debug_frame entries when available since go functions
do not have frame pointers before go1.8.
Sometimes windows will send us events about breakpoints we have
already removed from the code despite the fact that we go to great
lengths to avoid this already.
Change waitForDebugEvent to check that when we receive a breakpoint
event the corresponding memory actually contains an INT 3
instruction, if it doesn't ignore the event and restart the thread.
Changes to the GC in 1.8 make this test fail because the GC isn't
inserting stack barriers into our test program anymore. This isn't the
same thing as being unable to print stacktraces in presence of stack
barriers so the test shouldn't fail.
The PC we have is relative to the first instruction after the CALL
instruction currently being executed.
Anyone watching a disassembly will understand what's happening if we
report the return PC, but reporting the first PC of the current line is
useless and confusing.
