>date: 2004/11/30 01:51:13; author: jwise; state: Exp; lines: +21 -3
>As discussed on tech-toolchain@netbsd.org, make cpp refuse to attempt to
>parse a #include'd file which does not pass S_ISREG() if the environment variable
>CPP_RESTRICTED is set.
>
>This is primarily intended for use by programs such as calendar(1) which
>use cpp to parse untrusted user files -- without this change (and the corresponding
>change to calendar(1)), any user can cause a denial-of-service for the daily
>calendar -a run by #include'ing a named pipe.
>
>Many thanks to christos@netbsd for his help in polishing this.
>revision 1.2
>date: 2003/10/29 21:25:03; author: mycroft; state: Exp; lines: +40 -11
>As seen on tech-toolchain:
>
>So the problem is thus:
>
>* The {u,}mulsidi3 generate two parallel sets which modify the upper and lower
>halves of the target register.
>
>* life_analysis() does not track subregister modifications -- if you don't
>modify the whole register with a single set, it considers the register
>unused.
>
>The simple, if klugy, solution to this is to stick an explicit clobber in. It
>seems to work.
>
>While doing this, I noticed that constant folding was not happening for
>32x32->64 multiplies. This is because the parallel set generated by
>{u,}mulsidi3 cannot be folded at all. To solve this, I first expand to a
>normal multiply, and then use a define_insn_and_split to convert it to the
>parallel set after CSE and constant folding.
>
>This patch has also been submitted to GCC bugzilla, but who knows if I'll get
>a reply to that.
- add GPLUSPLUS_BACKWARD_INCLUDE_DIR to the INCLUDE_DEFAULTS includes
- define GPLUSPLUS_BACKWARD_INCLUDE_DIR, TOOLDIR_BASE_PREFIX,
STANDARD_BINDIR_PREFIX and STANDARD_LIBEXEC_PREFIX
- delete NETBSD_LIBGCC_SPEC
- define LIBGCC_PICSUFFIX as "_pic".
>PR/29832: J.T. Conklin: libstdc++ missing std::wstring support
>This wonderful patch together with yesterdays wcsftime addition, makes
>us able to use wide strings in c++
old gdb, teach it new flags, so that it can recognize binaries compiled
with new binutils. For new flags set arch to an approximate superset
that this version of bfd knows about. This is good enough for gdb.
PR port-sh3/32972
The rest of the gnu/dist/diffutils tree remains at version 2.8.1.
Fixes PR 23162 from George Michaelson (etcupdate/sdiff report
error on applying merge edits)
Approved by wiz, christos
- our kerberos support
- nolock
- setuid support / setxid in child
- avoid dangerous commands and only allow admin group to execute them
- selectable CVS directory name
- symlinked repository fixes
- log admin commands in history
- default to ssh instead of rsh
- localid keyword
- null revision on re-added files
- umask fixes
- t flag in log
- don't recursively re-enter signal error handler
- xasprintf in selected places
- ipv6 support
delay slot (e.g. PLT entries use such code). We can not set the
breakpoint there, as trapa is illegal in the delay slot. Set the
breakpoint to the next instruction instead. We are guaranteed to
arrive there, as control transfers are illegal in the delay slot.
Kaz Kojima <kkojima@rr.iij4u.or.jp>
* elf32-sh.c (sh_elf_relocate_section): Don't reset relocation
for R_SH_REL32 in shared objects if the symbol is locally called.
Makes hidden and protected symbols in shared objects work.
This will be committed to the binutils mainline.
Changelog for 2.2.5 -> 2.2.6:
20050806
Workaround: accept(2) fails with EPROTO when the client
already disconnected (SunOS 5.5.1). File: sane_accept.c.
20050815
Workaround: old Solaris compilers can't link an archive
without globally visible symbols. File: tls/tls_misc.c.
20050922
Bugfix: the *SQL clients did not uniformly choose the
database host from the available pool of servers due to an
off-by-one error, so that the "last" available server was
not selected. Leandro Santi. Files: dict_mysql.c, dict_pgsql.c.
20050929
Paranoia: don't ignore garbage in SMTP or LMTP server replies
when ESMTP command pipelining is turned on. For example,
after sending ".<CR><LF>QUIT<CR><LF>", Postfix could recognize
the server's 2XX QUIT reply as a 2XX END-OF-DATA reply after
garbage, causing mail to be lost. The SMTP and LMTP clients
now report a remote protocol error and defer delivery.
Files: smtp/smtp_chat.c, smtp/smtp_trouble.c, lmtp/lmtp_chat.c,
lmtp/lmtp_trouble.c.
20051011
Bugfix: raise the "policy violation" flag when a client
request exceeds a concurrency or rate limit. File:
smtpd/smtpd.c.
Bugfix (cut-and-paste error): don't reply with 421 (too
many MAIL FROM or RCPT TO commands) when we aren't closing
the connection. File: smtpd/smtpd.c.
20051013
Bugfix: don't do smtpd_end_of_data_restrictions after the
transaction failed due to, e.g., a write error. File:
smtpd/smtpd.c.
Cleanup: the SMTP server now enforces the message_size_limit
even when the client did not send SIZE information with the
MAIL FROM command. This protects before-queue content
filters against over-size messages. File: smtpd/smtpd.c.
20051105
Workaround: the next-hop logical destination information
for connection caching was reset only after a good non-TLS
connection, so that cached connections to non-TLS backup
servers could suck away traffic from TLS primary servers
(the Postfix SMTP client cannot cache an open TLS connection).
Found during code review. Fixing this requires more change
than is allowed in a stable release. File: smtp/smtp_connect.c.
20051108
Bugfix: two messages could get the same message ID due to
a race condition. This time window was increased when queue
file creation was postponed from MAIL FROM until the first
accepted RCPT TO. The window is closed again. Found by
Victor. Files: global/mail_stream.c, global/mail_queue.c,
cleanup/cleanup_message.c. This code is back-ported from
the Postfix 2.3 snapshot release.
20051119
Bugfix: the queue manager did not write a per-recipient
defer logfile record when the delivery agent crashed after
the initial handshake with the queue manager, and before
reporting the delivery status to the queue manager. Files:
*qmgr/qmgr_deliver.c.
20051126
Log warning when REDIRECT, FILTER, HOLD and DISCARD are
used in smtpd_etrn_restrictions. File: smtpd/smtpd_check.c.
20051128
Bugfix: moved code around from one place to another to make
REDIRECT, FILTER, HOLD and DISCARD access(5) table actions
work in smtpd_end_of_data_restrictions. PREPEND will not
be fixed; it must be specified before the message content
is received. Files: smtpd/smtpd.c, smtpd/smtpd_check.c,
cleanup/cleanup_extracted.c, pickup/pickup.c.
Changelog for 2.2.5 -> 2.2.6:
20050806
Workaround: accept(2) fails with EPROTO when the client
already disconnected (SunOS 5.5.1). File: sane_accept.c.
20050815
Workaround: old Solaris compilers can't link an archive
without globally visible symbols. File: tls/tls_misc.c.
20050922
Bugfix: the *SQL clients did not uniformly choose the
database host from the available pool of servers due to an
off-by-one error, so that the "last" available server was
not selected. Leandro Santi. Files: dict_mysql.c, dict_pgsql.c.
20050929
Paranoia: don't ignore garbage in SMTP or LMTP server replies
when ESMTP command pipelining is turned on. For example,
after sending ".<CR><LF>QUIT<CR><LF>", Postfix could recognize
the server's 2XX QUIT reply as a 2XX END-OF-DATA reply after
garbage, causing mail to be lost. The SMTP and LMTP clients
now report a remote protocol error and defer delivery.
Files: smtp/smtp_chat.c, smtp/smtp_trouble.c, lmtp/lmtp_chat.c,
lmtp/lmtp_trouble.c.
20051011
Bugfix: raise the "policy violation" flag when a client
request exceeds a concurrency or rate limit. File:
smtpd/smtpd.c.
Bugfix (cut-and-paste error): don't reply with 421 (too
many MAIL FROM or RCPT TO commands) when we aren't closing
the connection. File: smtpd/smtpd.c.
20051013
Bugfix: don't do smtpd_end_of_data_restrictions after the
transaction failed due to, e.g., a write error. File:
smtpd/smtpd.c.
Cleanup: the SMTP server now enforces the message_size_limit
even when the client did not send SIZE information with the
MAIL FROM command. This protects before-queue content
filters against over-size messages. File: smtpd/smtpd.c.
20051105
Workaround: the next-hop logical destination information
for connection caching was reset only after a good non-TLS
connection, so that cached connections to non-TLS backup
servers could suck away traffic from TLS primary servers
(the Postfix SMTP client cannot cache an open TLS connection).
Found during code review. Fixing this requires more change
than is allowed in a stable release. File: smtp/smtp_connect.c.
20051108
Bugfix: two messages could get the same message ID due to
a race condition. This time window was increased when queue
file creation was postponed from MAIL FROM until the first
accepted RCPT TO. The window is closed again. Found by
Victor. Files: global/mail_stream.c, global/mail_queue.c,
cleanup/cleanup_message.c. This code is back-ported from
the Postfix 2.3 snapshot release.
20051119
Bugfix: the queue manager did not write a per-recipient
defer logfile record when the delivery agent crashed after
the initial handshake with the queue manager, and before
reporting the delivery status to the queue manager. Files:
*qmgr/qmgr_deliver.c.
20051126
Log warning when REDIRECT, FILTER, HOLD and DISCARD are
used in smtpd_etrn_restrictions. File: smtpd/smtpd_check.c.
20051128
Bugfix: moved code around from one place to another to make
REDIRECT, FILTER, HOLD and DISCARD access(5) table actions
work in smtpd_end_of_data_restrictions. PREPEND will not
be fixed; it must be specified before the message content
is received. Files: smtpd/smtpd.c, smtpd/smtpd_check.c,
cleanup/cleanup_extracted.c, pickup/pickup.c.
Reset it to the default (command line dependend) before each turn
of "config.gcc", so each invocation gets a clean state and only the
changes from the last round persist.
This fixes the build of a sparc toolchain on a sparc64 host.
rather than putting those longer than 30 chars into .rodata.str1.32.
Apparantly gcc 3.4.4 has this change...
In spite of everything being compressed, this saves over 13k in the boot
floppies.
ptid before invoking the suspend-all-but-this-thread logic. Otherwise,
when passed an LWP (as a result of 1.15 making LWPs the primary
identifier of live execution contexts), we try to suspend everything,
which goes poorly.
warn about the comparison always being true. Fixes build for macppc
target.
Problem noted by me, analysis and fix from Thomas Klausner.
Approved by wiz.
wrong registers with some opcodes, the error is:
Error: Illegal operands: There are only 32 single precision f registers; [0-31]
And the code to cause this is:
return (cond) ? x : -x;
where x is double.
AdminGroup
AdminOptions
that control who has the right to run cvs admin and cvs init.
For fine control over cvs commands, AdminOptions can be used.
I suggest AdminOptions=kmIqc
NetBSD options. Allows powerpc-*-netbsd* to accept "-R <arg>"
when <arg> doesn't exist in the host environment (as all other
NetBSD/ELF platforms do). As with MIPS and as before, accept
-G NUM, too.
The more portable way to pass arguments to the linker is, of course,
still -Wl,-R,/some/path (note that there are two commas).
anymore. Look in the ucontext_t for the next frame
instead. Backtrace through a signal handler invocation
works now.
(Since this change is NetBSD specific, this code shouldn't
be in a file which is shared among BSDs.)
* elf64-sparc.c (sparc64_elf_adjust_dynamic_symbol): When linking a
non-shared object, do not reserve space in .plt and .rela.plt for
regular symbols neither defined nor referenced in shared objects.
This fixes firefox for Havard
"tradlittlemips" and "tradbigmips" vectors.
The old vectors are meant for IRIX with it prelink shared library
support, and changes to better support this in binutils 2.15 broke
shared library support under NetBSD. The "trad" vectors are what 32bit
Linux and OpenBSD also use.
These vectors are also compatible with older NetBSD-built shared
libraries and programs so no flag day is necessary (Yay!).
These changes from Nick Hudson. Help also from Charles Hannum.
not if the 64 bit type is defined (BFD_HOST_64_BIT) because the second
is always defined, even if it falls back to a 32 bit type. In our case,
this was not the problem; the problem was that we bfd_vma is a 32 bit
type when BFD64 is not defined, so the shift was out of range.
