Backport #6127: IO#ioctl range error in 1.9.3

• Status: Closed
• Assignee: naruse (Yui NARUSE)

[dlampa (Donovan Lampa)]

In 32-bit implementations of 1.9.3 passing an (apparently) unsigned value (BIOCIMMEDIATE = 0x80044270) to ioctl throws range error:

ruby-1.9.3-p125 :001 > a,b = IO.pipe
=> [#<IO:fd 5>, #<IO:fd 6>]
ruby-1.9.3-p125 :002 > a.ioctl(0x80044270, "\x01")
RangeError: bignum too big to convert into long' from (irb):2:in ioctl'
from (irb):2
from /Users/dlampa/.rvm/rubies/ruby-1.9.3-p125/bin/irb:16:in `'
ruby-1.9.3-p125 :001 > 0x80044270.class
=> Bignum

When in 32-bit 1.9.1 and 1.9.2 it does not:

ruby-1.9.1-p243 :001 > a,b = IO.pipe
=> [#IO:0x202fefc, #IO:0x202fec4]
ruby-1.9.1-p243 :002 > a.ioctl(0x80044270, "\x01")
Errno::ENOTTY: Inappropriate ioctl for device
from (irb):2:in ioctl' from (irb):2 from /Users/dlampa/.rvm/rubies/ruby-1.9.1-p243/bin/irb:16:in '
ruby-1.9.1-p243 :001 > 0x80044270.class
=> Bignum

OS is [x86_64-darwin10.8.0] however ruby was compiled as 32-bit through rvm using:
rvm_archflags="-arch i386" CFLAGS="-arch i386" LDFLAGS="-arch i386" rvm install [ruby-version] --patch osx-arch-fix

The patch is installed in ~/.rvm/patches/ruby/1.9.3/ and is as follows:

--- a/Makefile.in 2010-11-11 15:54:26.000000000 -0600
+++ b/Makefile.in 2010-11-11 15:55:24.000000000 -0600
@@ -55,7 +55,7 @@ XLDFLAGS = @XLDFLAGS@ $(EXTLDFLAGS)
EXTLIBS =
LIBS = @LIBS@ $(EXTLIBS)
MISSING = @LIBOBJS@ @ALLOCA@
-LDSHARED = @LIBRUBY_LDSHARED@
+LDSHARED = @LIBRUBY_LDSHARED@ $(LDFLAGS)
DLDFLAGS = @LIBRUBY_DLDFLAGS@ $(EXTLDFLAGS) @ARCH_FLAG@
SOLIBS = @SOLIBS@
MAINLIBS = @MAINLIBS@

this issue appears to be related (http://bugs.ruby-lang.org/issues/5429), but I don't speak Japanese.

[ko1 (Koichi Sasada)]

• Assignee set to naruse (Yui NARUSE)

[nobu (Nobuyoshi Nakada)]

• Tracker changed from Bug to Backport
• Project changed from Ruby master to Backport193

[naruse (Yui NARUSE)]

This is fixed on trunk with commits 33711,33713,33714,33715,33716,33717,33718,33719,33720,33721,33724,33727,33728,33752,33753.
But can't apply as is.
Anyone can apply it?

[dlampa (Donovan Lampa)]

Any updates on this? Has it been applied or is there a scheduled date for application?

[naruse (Yui NARUSE)]

• Status changed from Open to Feedback

[dlampa (Donovan Lampa)]

Apologies for the hiatus also apologies if I've misunderstood the meaning of the Feedback status.

1.9.3-p194 doesn't seem to have addressed this:

ruby-1.9.3-p194 :002 > a,b = IO.pipe
=> [#<IO:fd 5>, #<IO:fd 6>]
ruby-1.9.3-p194 :003 > a.ioctl(0x80044270, "\x01")
RangeError: bignum too big to convert into long' from (irb):3:in ioctl'
from (irb):3
from /Users/dlampa/.rvm/rubies/ruby-1.9.3-p194/bin/irb:16:in `'
ruby-1.9.3-p194 :003 > 0x80044270.class
=> Bignum

[naruse (Yui NARUSE)]

dlampa (Donovan Lampa) wrote:

Apologies for the hiatus also apologies if I've misunderstood the meaning of the Feedback status.

In general, "Feedback" means that an action of other than assignee is required.
In this ticket, I need a patch for backport which covers r33711, r33713-r33721, r33724, r33727,
r33728, r33752, and r33753.

[dlampa (Donovan Lampa)]

• File ruby-io-c.patch ruby-io-c.patch added

For those that are interested, I've taken the liberty of putting a patch together for this issue (attached).

The patch is confirmed to work and address the issue as it exists in 1.9.3-p194.

Thanks to Naruse for providing all the relevant commits for this issue.

[naruse (Yui NARUSE)]

• Status changed from Feedback to Closed
• % Done changed from 0 to 100

This issue was solved with changeset r37138.
Donovan, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

merge revision(s) 33711,33713,33714,33715,33716,33717,33718,33719,33720,33721,33724,33727,33728,33752,33753: [Backport #6127]

    * io.c (ioctl_req_t): Type of req argument of ioctl() depend on platform.
      Moreover almost all linux ioctl can't be represented by 32bit integer
      (i.e. MSB is 1). We need wrap ioctl argument type.
    [Bug #5429] [ruby-dev:44589]
    * io.c (struct ioctl_arg): ditto.
    * io.c (rb_ioctl): ditto.
    * test/ruby/test_io.rb (test_ioctl_linux): add a testcase for ioctl
    * backporting patch is created by Donovan Lampa.

[h.shirosaki (Hiroshi Shirosaki)]

I have a build error of ruby_1_9_3 branch at r37146 with mingw.
It seems due to r37138.

compiling ../io.c
../io.c: In function 'rb_cloexec_fcntl_dupfd':
../io.c:213:21: error: 'F_DUPFD' undeclared (first use in this function)
../io.c:213:21: note: each undeclared identifier is reported only once for each function it appears in
../io.c: In function 'rb_io_each_byte':
../io.c:3081:15: warning: variable 'e' set but not used [-Wunused-but-set-variable]
../io.c: In function 'rb_io_advise':
../io.c:7892:16: warning: variable 'l' set but not used [-Wunused-but-set-variable]
../io.c:7892:11: warning: variable 'off' set but not used [-Wunused-but-set-variable]
../io.c: At top level:
../io.c:1575:14: warning: 'nogvl_fsync' defined but not used [-Wunused-function]
make: *** [io.o] Error 1

Here is a patch.

Index: io.c

--- io.c (revision 37146)
+++ io.c (working copy)
@@ -209,8 +209,10 @@
else {
ret = fcntl(fd, F_DUPFD, minfd);
}
+#elif defined(F_DUPFD)

• ret = fcntl(fd, F_DUPFD, minfd);
  #else

• ret = fcntl(fd, F_DUPFD, minfd);

• ret = -1;
  #endif
  if (ret == -1) return -1;
  rb_maygvl_fd_fix_cloexec(ret);