Trust argument order to determine argument frame layout when calling
functions, this allows calling optimized functions and removes the
special cases for runtime.mallocgc.
Fixes#1589
Add options to start a stacktrace from the values saved in the
runtime.g struct as well as a way to disable the stackSwitch logic and
just get a normal stacktrace.
If a closure captures a variable but also defines a variable of the
same name in its root scope the shadowed flag would, sometimes, not be
appropriately applied to the captured variable.
This change:
1. sorts the variable list by depth *and* declaration line, so that
closure captured variables always appear before other root-scope
variables, regardless of the order used by the compiler
2. marks variable with the same name as shadowed even if there is only
one scope at play.
This fixes the problem but as a side effect:
1. programs compiled with Go prior to version 1.9 will have the
shadowed flag applied arbitrarily (previously the shadowed flag was not
applied at all)
2. programs compiled with Go prior to versoin 1.11 will still exhibit
the bug, as they do not have DeclLine information.
Fixes#1672
The fix for #1428 was buggy, partly because I communicated poorly. Sorry
about that.
The size of the TLS segment should be padded such that TLS addresses
are congruent in the file to where they will end up memory, i.e.
(tlsoffset%align) == (vaddr%align). In most cases, vaddr will be aligned
and it won't matter, but if not then simply aligning the end of the
segment is incorrect. This should be right.
(For the record, the current rounding logic is working in bits, but
PtrSize is in bytes, so it wasn't working as originally intended
either.)
Intent here is to bring optargorder up to date with delve
and keep it in sync (and to use optargorder to help monitor
compiler output for debugging quality regressions).
proc.Next and proc.Step will call, after setting their temp
breakpoints, curthread.SetCurrentBreakpoint. This is intended to find
if one of the newly created breakpoints happens to be at the same
instruction that curthread is stopped at.
However SetCurrentBreakpoint is intended to be called after a Continue
and StepInstruction operation so it will also detect if curthread is
stopped one byte after a breakpoint.
If the instruction immediately preceeding the current instruction of
curthread happens to:
1. have one of the newly created temp breakpoints
2. be one byte long
SetCurrentBreakpoint will believe that we just hit that breakpoint and
therefore the instruction should be repeated, and thus rewind the PC of
curthread by 1.
We should distinguish between the two uses of SetCurrentBreakpoint and
disable the check for "just hit" breakpoints when inappropriate.
Fixes#1656
Moves EvalScope methods to the proper file and organizes everything
together. Also makes some EvalScope methods no longer methods and just
pure functions.
* proc: fix stacktraces when a SIGSEGV happens during a cgo call
When a SIGSEGV happens in a cgo call (for example as a result of
dereferencing a NULL pointer) the stack layout will look like this:
(system stack) runtime.fatalthrow
(system stack) runtime.throw
(system stack) runtime.sigpanic
(system stack) offending C function
... other C functions...
(system stack) runtime.asmcgocall
(goroutine stack) call inside cgo
The code in switchStack would switch directly from the
runtime.fatalthrow frame to the first frame in the goroutine stack,
hiding important information.
Disable this switch for runtime.fatalthrow and reintroduce the check
for runtime.mstart that existed before this version of the code was
implemented in commit 7bec20.
This problem was reported in comment:
https://github.com/go-delve/delve/issues/935#issuecomment-512182533
* cmd/dlv: actually disable C compiler optimizations when building
* *: Add .cirrus.yml for FreeBSD testing
* *: run go mod tidy
* service/test: prefer 127.0.0.1 over localhost
* dwarf/line: fix TestDebugLinePrologueParser
* vendor: rerun go mod vendor
Propagate signals when stepping because debugserver will report them,
from the issue:
2019-07-11T16:31:25+02:00 debug layer=gdbconn <- $z0,105525d,1#c9
2019-07-11T16:31:25+02:00 debug layer=gdbconn -> $OK#00
2019-07-11T16:31:25+02:00 debug layer=gdbconn <- $vCont;s:c41c3#50
2019-07-11T16:31:25+02:00 debug layer=gdbconn -> $T1cthread:c41c3;threads:c41c3,c41d7,c41d8,c41d9,c41da;thread-pcs:105525d,7fffc464bf46,7fffc464bbf2,7fffc464bbf2,7fffc46...
2019-07-11T16:31:25+02:00 debug layer=gdbconn <- $Z0,105525d,1#a9
2019-07-11T16:31:25+02:00 debug layer=gdbconn -> $OK#00
in this case we request a single step on thread c41c3 but debugserver
reports instead a signal (in this case SIGWINCH).
Fixes#1610
Add variables flag to mark variables that are allocated on a register
(and have no address) and variables that we read as result of a
function call (and are allocated on a stack that no longer exists when
we show them to the user).
Increases the maximum string length from 64 to 1MB when loading strings
for a binary operator, also delays the loading until it's necessary.
This ensures that comparison between strings will always succeed in
reasonable situations.
Fixes#1615
Backports debug/dwarf commit: 535741a69a1300d1fe2800778b99c8a1b75d7fdd
CL: https://go-review.googlesource.com/18459
The x/debug/dwarf that we used for dwarf/godwarf/type.go was forked
from debug/dwarf long before this commit.
Original description:
Currently readType simultaneously constructs a type graph and resolves
the sizes of the types. However, these two operations are
fundamentally at odds: the order we parse a cyclic structure in may be
different than the order we need to resolve type sizes in. As a
result, it's possible that when readType attempts to resolve the size
of a typedef, it may dereference a nil Type field of another typedef
retrieved from the type cache that's only partially constructed.
To fix this, we delay resolving typedef sizes until the end of the
readType recursion, when the full type graph is constructed.
Fixes#1601
* proc: allow simultaneous call injection to multiple goroutines
Changes the call injection code so that we can have multiple call
injections going on at the same time as long as they happen on distinct
goroutines.
* proc: fix EvalExpressionWithCalls for constant expressions
The lack of address of constant expressions would confuse EvalExpressionWithCalls
Fixes#1577
* tests: fix tests for Go 1.13
- Go 1.13 doesn't autogenerate init functions anymore, tests that
expected that now fail and should be skipped.
- Plugin tests now need -gcflags'all=-N -l' now, we were probably
getting lucky with -gcflags='-N -l' before.
* proc: allow signed integers as shift counts
Go1.13 allows signed integers to be used as the right hand side of a
shift operator, change eval to match.
* goversion: update maximum supported version
* travis: force Go to use vendor directory
Travis scripts get confused by "go: downloading" lines, the exact
reason is not clear. Testing that the vendor directory is up to date is
a good idea anyway.
Support for bulk queries makes the DWARF quality checker
(github.com/dr2chase/dwarf-goodness/cmd/dwarf-goodness)
run much more efficiently (replace quadratic cost with
linear).
The location specified '<fnname>:0' could be used to set a breakpoint
on the entry point of the function (as opposed to locspec '<fnname>'
which sets it after the prologue).
Setting a breakpoint on an entry point is almost never useful, the way
this feature was implemented could cause it to be used accidentally and
there are other ways to accomplish the same task (by setting a
breakpoint on the PC address directly).
Also fixes findCompileUnitForOffset which was broken in some edge cases
(when looking up an offset inside the last child of the compilation
unit) which don't happen in normal executables (we only look up types, and those
are always direct childs of compile units).
Allow changing the value of a string variable to a new literal string,
which requires calling runtime.mallocgc to allocate the string into the
target process.
This means that a command like:
call f("some string")
is now supported.
Additionally the command:
call s = "some string"
is also supported.
Fixes#826
The bitmasks for transforming a 64-bit register into it's lower-bit
counterparts (e.g. RAX -> EAX -> AX -> AH/AL) were incorrect. This patch
fixes the bitmasks and adds an additional test.
Childless compile units would confuse loadDebugInfoMaps.
No test because I don't know what causes go to invoke GNU As in such a
way that it produces a childless compile unit.
Fixes#1572
Instead of reading partial units as we see them skip them entirely and
then re-read them when they are imported, directly into the destination
compile unit.
This avoids a lot of duplicate code in the loadDebugInfoMaps function
and will simplify implementing logical breakpoints and support for the
new DW_AT_go_package_name attribute added in Go 1.13.
The current wording is confusing - the file exists and the line exists, so what is the problem? I suspect this ambiguity is behind #1496 and likely others.
Also I updated the style to return values like the rest of the code in the file, which is also more readable (IMO and per https://github.com/golang/go/wiki/CodeReviewComments#named-result-parameters)
* proc: support nested function calls
Changes the code in fncall.go to support nested function calls.
This changes delays argument evaluation until after we have used
the call injection protocol to allocate an argument frame. When
evaluating the parse tree of an expression we'll initiate each
function call we find on the way down and then complete the function
call on the way up.
For example. in:
f(g(x))
we will:
1. initiate the call injection protocol for f(...)
2. progress it until the point where we have space for the arguments
of 'f' (i.e. when we receive the debugCallAXCompleteCall message
from the target runtime)
3. inititate the call injection protocol for g(...)
4. progress it until the point where we have space for the arguments
of 'g'
5. copy the value of x into the argument frame of 'g'
6. finish the call to g(...)
7. copy the return value of g(x) into the argument frame of 'f'
8. finish the call to f(...)
Updates #119
* proc: bugfix: closure addr was wrong for non-closure functions
The initial implementation of the 'call' command required the
function call to be the root expression, i.e. something like:
double(3) + 1
was not allowed, because the root expression was the binary operator
'+', not the function call.
With this change expressions like the one above and others are
allowed.
This is the first step necessary to implement nested function calls
(where the result of a function call is used as argument to another
function call).
This is implemented by replacing proc.CallFunction with
proc.EvalExpressionWithCalls. EvalExpressionWithCalls will run
proc.(*EvalScope).EvalExpression in a different goroutine. This
goroutine, the 'eval' goroutine, will communicate with the main
goroutine of the debugger by means of two channels: continueRequest
and continueCompleted.
The eval goroutine evaluates the expression recursively, when
a function call is encountered it takes care of setting up the
function call on the target program and writes a request to the
continueRequest channel, this causes the 'main' goroutine to restart
the target program by calling proc.Continue.
Whenever Continue encounters a breakpoint that belongs to the
function call injection protocol (runtime.debugCallV1 and associated
functions) it writes to continueCompleted which resumes the 'eval'
goroutine.
The 'eval' goroutine takes care of implementing the function call
injection protocol.
When the expression is fully evaluated the 'eval' goroutine will
write a special message to 'continueRequest' signaling that the
expression evaluation is terminated which will cause Continue to
return to the user.
Updates #119
This change splits the BinaryInfo object into a slice of Image objects
containing information about the base executable and each loaded shared
library (note: go plugins are shared libraries).
Delve backens are supposed to call BinaryInfo.AddImage whenever they
detect that a new shared library has been loaded.
Member fields of BinaryInfo that are used to speed up access to dwarf
(Functions, packageVars, consts, etc...) remain part of BinaryInfo and
are updated to reference the correct image object. This simplifies this
change.
This approach has a few shortcomings:
1. Multiple shared libraries can define functions or globals with the
same name and we have no way to disambiguate between them.
2. We don't have a way to handle library unloading.
Both of those affect C shared libraries much more than they affect go
plugins. Go plugins can't be unloaded at all and a lot of name
collisions are prevented by import paths.
There's only one problem that is concerning: if two plugins both import
the same package they will end up with multiple definition for the same
function.
For example if two plugins use fmt.Printf the final in-memory image
(and therefore our BinaryInfo object) will end up with two copies of
fmt.Printf at different memory addresses. If a user types
break fmt.Printf
a breakpoint should be created at *both* locations.
Allowing this is a relatively complex change that should be done in a
different PR than this.
For this reason I consider this approach an acceptable and sustainable
stopgap.
Updates #865
Remove the breakpoint set in TestCallConcurrent so that it doesn't
interfere with the call injection protocol.
The fact that it can is a bug but that bug is better addressed after
PRs #1503 and #1504 are merged, this keeps tests happy in the meantime.
Fixes#1542
Go 1.12 introduced a change to the internal map representation where
empty map cells can be marked with a tophash value of 1 instead of just
0.
Fixes#1531
RestoreRegisters on linux would also restore FS_BASE and GS_BASE, if
the target goroutine migrated to a different thread during the call
injection this would result in two threads of the target process
pointing to the same TLS area which would greatly confuse the target
runtime, leading to fatal panics with nonsensical stack traces.
Other backends are unaffected:
- native/windows doesn't store the TLS in the same CONTEXT struct as
the other register values.
- native/darwin doesn't support function calls (and wouldn't store the
TLS value in the same struct)
- gdbserial/rr doesn't support function calls (because it's a
recording)
- gsdbserial/lldb extracts the value of TLS by executing code in the
target process.
* *: use loglevel to control what gets logged instead of output redirection
This stops logrus from doing all the formatting just to discard it
immediately afterwards.
* logflags: replace default formatter of logrus
The default formatter of logrus emits logs in two different formats
depending on whether or not the output is going to a terminal. The
output format for non-terminals is indented to be machine readable, but
we mostly read logs ourselves and the excessive quoting makes that
format unreadable.
When outputting to terminals it uses ANSI escape codes unconditionally,
without checking whether the terminal it is connected to actually
supports colors.
This commit replaces the default formatter with a much simpler
formatter that always uses a more readable format, doesn't use colors
and places the key-value pairs at the beginning of the line (which is a
better match for how we use them).
* cmd/dlv: add command line options to redirect logs
Adds two options, --log-to-file and --log-to-fd, to redirect logs to a
file or to a file descriptor.
When one of those two options is specified the "API server listening
at:" message will also be redirected to the specified file/file
descriptor.
This allows clients that want to use the "API server listening at:"
message to do so even if they want to redirect the target's stdout to
another file or device.
Implements #1179, #1523
Adds initial support for plugins, this is only the code needed to keep
track of loaded plugins on linux (both native and gdbserial backend).
It does not actually implement support for debugging plugins on linux.
Updates #865
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.
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.
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
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
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
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
