Lorenzo Stoakes [Thu, 28 Apr 2011 04:13:49 +0000 (00:13 -0400)]
gc: correctly handle fields of pointer type to recursive forward references
Previously, whether declaring a type which copied the structure of a type it was referenced in via a pointer field would work depended on whether you declared it before or after the type it copied, e.g. type T2 T1; type T1 struct { F *T2 } would work, however type T1 struct { F *T2 }; type T2 T1 wouldn't.
The g->sched.sp saved stack pointer and the
g->stackbase and g->stackguard stack bounds
can change even while "the world is stopped",
because a goroutine has to call functions (and
therefore might split its stack) when exiting a
system call to check whether the world is stopped
(and if so, wait until the world continues).
That means the garbage collector cannot access
those values safely (without a race) for goroutines
executing system calls. Instead, save a consistent
triple in g->gcsp, g->gcstack, g->gcguard during
entersyscall and have the garbage collector refer
to those.
The old code was occasionally seeing (because of
the race) an sp and stk that did not correspond to
each other, so that stk - sp was not the number of
stack bytes following sp. In that case, if sp < stk
then the call scanblock(sp, stk - sp) scanned too
many bytes (anything between the two pointers,
which pointed into different allocation blocks).
If sp > stk then stk - sp wrapped around.
On 32-bit, stk - sp is a uintptr (uint32) converted
to int64 in the call to scanblock, so a large (~4G)
but positive number. Scanblock would try to scan
that many bytes and eventually fault accessing
unmapped memory. On 64-bit, stk - sp is a uintptr (uint64)
promoted to int64 in the call to scanblock, so a negative
number. Scanblock would not scan anything, possibly
causing in-use blocks to be freed.
In short, 32-bit platforms would have seen either
ineffective garbage collection or crashes during garbage
collection, while 64-bit platforms would have seen
either ineffective or incorrect garbage collection.
You can see the invalid arguments to scanblock in the
stack traces in issue 1620.
Fixes #1620.
Fixes #1746.
R=iant, r
CC=golang-dev
https://golang.org/cl/4437075
runtime: memory allocated by OS not in usable range
runtime: out of memory: cannot allocate 1114112-byte block (2138832896 in use)
throw: out of memory
Peter Mundy [Wed, 27 Apr 2011 19:47:12 +0000 (15:47 -0400)]
runtime: fix mkversion to output valid path separators
In a GOROOT path a backslash is a path separator
not an escape character. For example, `C:\go`.
Fixes gotest error:
version.go:3: unknown escape sequence: g
Andrew Gerrand [Wed, 27 Apr 2011 01:00:34 +0000 (11:00 +1000)]
goinstall: support GOPATH; building and installing outside the Go tree
For example, with GOPATH set like so
GOPATH=/home/adg/gocode
And after creating some subdirectories
mkdir /home/adg/gocode/{bin,pkg,src}
I can use goinstall to install the github.com/nf/goto web server,
which depends on the github.com/nf/stat package, with
goinstall github.com/nf/goto
This downloads and installs all dependencies (that aren't already
installed) like so
/home/adg/gocode/bin/goto
/home/adg/gocode/pkg/darwin_amd64/github.com/nf/stat.a
/home/adg/gocode/src/github.com/nf/goto/...
/home/adg/gocode/src/github.com/nf/stat/...
Adam Langley [Tue, 26 Apr 2011 14:26:22 +0000 (10:26 -0400)]
crypto/x509: memorize chain building.
I ran the new verification code against a large number of certificates
with a huge (>1000) number of intermediates.
I had previously convinced myself that a cycle in the certificate
graph implied a cycle in the hash graph (and thus, a contradiction).
This is bogus because the signatures don't cover each other.
Secondly, I managed to drive the verification into a time explosion
with a fully connected graph of certificates. The code would try to
walk the factorial number of paths.
This change switches the CertPool to dealing with indexes of
certificates rather than pointers: this makes equality easy. (I didn't
want to compare pointers because a reasonable gc could move objects
around over time.)
Secondly, verification now memorizes the chains from a given
certificate. This is dynamic programming for the lazy, but there's a
solid reason behind it: dynamic programming would ignore the Issuer
hints that we can exploit by walking up the chain rather than down.
Also, 6g was passing uninitialized
Node &n2 to regalloc, causing non-deterministic
register collisions (but only when both left and
right hand side of comparison had function calls).
Anthony Martin [Sat, 23 Apr 2011 14:53:49 +0000 (10:53 -0400)]
ld: fix Plan 9 symbol table
Static symbols were not being marked as such.
I also made the 'z' symbols use the first byte of
the name instead of an explicit NUL so that if
the symbol table format is ever changed, the only
place that would need updating is addhist().
Ian Lance Taylor [Fri, 22 Apr 2011 23:59:21 +0000 (16:59 -0700)]
fmt: decrease recursion depth
This permits the test to run when using gccgo on system
without split-stack support. See
http://gcc.gnu.org/ml/gcc-patches/2011-04/msg01420.html
http://gcc.gnu.org/PR48553
Adam Langley [Fri, 22 Apr 2011 19:33:41 +0000 (15:33 -0400)]
crypto/rsa: support > 3 primes.
With full multi-prime support we can support version 1 PKCS#1 private
keys. This means exporting all the members of rsa.PrivateKey, thus
making the API a little messy. However there has already been another
request to export this so it seems to be something that's needed.
Over time, rsa.GenerateMultiPrimeKey will replace rsa.GenerateKey, but
I need to work on the prime balance first because we're no longer
generating primes which are a multiples of 8 bits.
syscall: add BPF support for darwin/386, darwin/amd64
note: due to issue 1466 the Msghdr and BpfProgram
struct for src/pkg/syscall/ztypes_darwin_386.go,
src/pkg/syscall/ztypes_darwin_amd64.go had to be
edited after the godefs generation.
Ian Lance Taylor [Thu, 21 Apr 2011 15:32:58 +0000 (08:32 -0700)]
runtime: skip functions with no lines when building src line table
Avoid getting out of synch when a function, such as main.init,
has no associated line number information. Without this the
function before main.init can skip the PC all the way to the
next function, which will cause the next function's line table
to be associated with main.init, and leave subsequent
functions with the wrong line numbers.
On Mac X 10.6 /etc/resolv.conf is changed dynamically,
and may not exist at all when all network connections
are turned off, thus any lookup, even for "localhost"
would fail with "error reading DNS config: open
/etc/resolv.conf: no such file or directory". This
change avoids the error by trying to lookup addresses
in /etc/hosts before loading DNS config.
gc: correct handling of unexported method names in embedded interfaces
go/types: update for export data format change
reflect: require package qualifiers to match during interface check
runtime: require package qualifiers to match during interface check
test: fixed bug324, adapt to be silent
Fixes #1550.
Issue 1536 remains open.
R=gri, ken2, r
CC=golang-dev
https://golang.org/cl/4442071
reflect: add Type.Implements, Type.AssignableTo, Value.CallSlice; make Set match Go
This CL makes reflect require that values be assignable to the target type
in exactly the same places where that is the rule in Go. It also adds
the Implements and AssignableTo methods so that callers can check
the types themselves so as to avoid a panic.
Before this CL, reflect required strict type identity.
This CL expands Call to accept and correctly marshal arbitrary
argument lists for variadic functions; it introduces CallSlice for use
in the case where the slice for the variadic argument is already known.
net: use C library resolver on FreeBSD, Linux, OS X / amd64, 386
This CL makes it possible to resolve DNS names on OS X
without offending the Application-Level Firewall.
It also means that cross-compiling from one operating
system to another is no longer possible when using
package net, because cgo needs to be able to sniff around
the local C libraries. We could special-case this one use
and check in generated files, but it seems more trouble
than it's worth. Cross compiling is dead anyway.
It is still possible to use either GOARCH=amd64 or GOARCH=386
on typical Linux and OS X x86 systems.
It is also still possible to build GOOS=linux GOARCH=arm on
any system, because arm is for now excluded from this change
(there is no cgo for arm yet).
Andrew Gerrand [Wed, 20 Apr 2011 02:02:29 +0000 (12:02 +1000)]
goinstall: support building executable commands
This CL gives goinstall the ability to build commands,
not just packages.
"goinstall foo.googlecode.com/hg/bar" will build the command named
"bar" and install it to GOBIN. "goinstall ." will use the name of the
local directory as the command name.