by W.Richard Ste vens. EINTR handling suggested by bde@freebsd.org).
Code cleanup:
1. Add missing return type.
2. Replace 'union wait' by int.
3. Use Posix-style signal handling instead of signal().
4. Use fork() instead of deprecated vfork().
5. Block signals before fork()'ing, instead of after.
6. Return -1 if fork() fails, instead of 0.
7. Add EINTR handling for waitpid() call.
Also add claim of Posix conformance to man page.
Now that we preserve RPC handles instead of rebuilding them each time
a ypcln function is called, we have to be careful about keeping our sockets
in a sane state. It's possible that the caller may call a ypclnt
function, and then decide to close all its file descriptors. This would
also close the socket descriptor held by the yplib code. Worse, it
could re-open the same descriptor number for its own use. If it then calls
another ypclnt function, the subsequent RPC will fail because the socket
will either be gone or replaced with Something Completely Different. The
yplib code will recover by rebinding, but it doing so it may wreck the
descriptor which now belongs to the caller.
To fix this, _yp_dobind() needs to label the descriptor somehow so
that it can test it later to make sure it hasn't been altered between
ypclnt calls. It does this by binding the socket, thus associating a port
number with it. It then saves this port number in the dom_local_port member
of the dom_binding structure for the given domain. When _yp_dobind() is
called again (which it is at the start of each ypclnt function), it checks
to see if the domain is already bound, and if it is, it does a getsockname()
on the socket and compares the port number to the one it saved. If the
getsockname() fails, or the port number doesn't match, it abandons the
socket and sets up a new client handle.
This still incurs some syscall overhead, which is what I was trying to
avoid, but it's still not as bad as before.
functions are implimented as wrappers around getservent(), which means it's
up to getservent() to do all the work. The NIS support in getservent()
only allows it to scan through the services.byname map one entry at a
time until it finds the requested service name/port. This can be painfully
slow due to the overhead involved (lots and lots of successive RPCs).
To fix this, we allow getservbyname() and getservbyport() to signal
getservent() that if NIS is turned on (there's a '+' in /etc/services),
the usual yp_first()/yp_next() linear search should be abandoned and
yp_match() used instead. This causes getservent() to immediately
locate the requested entry instead of wasting time groping through the
whole map.
The downside is that this trick is accomplished by exporting a couple of
pointers from getservent.c which getservbyname.c and getservbyport.c can
preset in order to tell getservent() what to do. If all three functions
were in the same source module, then the extra cruft could be delcared
static to avoid poluting the global symbol space. Maybe they should be
combined anyway. For now I've settled on prepending lots of underscores.
privileged port within a single bind(), rather than looping through
attempts to bind over and over again over progressively lower ports.
This should speed up rlogin/rsh etc, and will probably cure some of the
strange rlogin hangs that have been reported in the past where rresvport()
managed to bind() to a port address that it shouldn't have.
not based on gpl'ed code, just prototype and usage. I'm not 100% certain
they behave the same while the system is in trouble (eg: malloc() failing)
but in those circumstances all bets would be off anyway.
These routines work like sprintf() and vsprintf(), except that instead of
using a fixed buffer, they allocate memory and return it to the user
and it's the user's responsibility to free() it. They have allocate as
much memory as they need (and can get), so the size of strings it can deal
with is limited only by the amount of memory it can malloc() on your
behalf.
There are a few gpl'ed programs starting to use this interface, and it's
becoming more common with the scares about security risks with sprintf().
I dont like the look of the code that the various programs (including
cvs, gdb, libg++, etc) provide if configure can't find it on the system.
It should be possible to modify the stdio core code to provide this
interface more efficiently, I was more worried about having something
that worked and was secure. :-) (I noticed that there was once intended
to be a smprintf() routine when our stdio was written for 4.4BSD, but it
looks pretty stillborn, and it's intended interface is not clear). Since
Linux and gnu libc have this interface, it seemed silly to bring yet
another one onto the scene.
to call clnt_destroy() on a potentially NULL RPC handle. Somebody should
bang on this a bit to make sure the problem is really gone; I seem to
have difficulty reproducing it. Patch provided by Peter Wemm and
slightly tweaked by me.
- Don't call _yp_unbind() in individual ypclnt functions unless we encounter
an RPC error while making a clnt_call().
bugs in your code is to put it in the -stable branch. (Corollary: the
day you discover the bug is the day the Internet decides to route your
telnet session to the repository box via Zimbabwe.)
Remove one bogus free(result) (from _havemaster()) that slipped by me.
Flagged by: phkmalloc
Pointed out to me by: Stefan Esser
In a nutshell, this macroizes the local/global symbol scoping rules
that are different in a.out and ELF. It also makes the i386 assembler
stubs conform to i386 PIC calling conventions - the a.out ld.so didn't
object, but the ELF one needs it as it implements PIC jumps via PLT's as
well as calls. The a.out rtld only worked because it was accidently
snooping the grandparent calling function's return address off the stack..
This also affects the libc_r code a little, because of cpp macro nesting.
Each of the ypclnt functions does a _yp_dobind() when it starts and then
a _yp_unbind() when it finishes. This is not strictly necessary and it
wastes cycles: it means we do a new clnt_create() and clnt_destroy()
for each yp_whatever() call. In fact, you can do multiple clnt_call()s
using a single RPC client handle returned by clnt_create(). Ideally we only
have to create a handle to ypserv once (the first time we call a ypclnt
function) and then destroy it and rebind only if a call to ypserv fails.
- Modify _yp_dobind() so that it only creates a new RPC client handle
when establishing a new binding or when one of the ypclnt calls
invalidates an existing binding and calls _yp_dobind() to establish
a new one.
- Modify the various ypclnt functions to only call _yp_unbind() if a
call to ypserv fails.
If _ANSI_SOURCE or _POSIX_SOURCE is defined, then <ctype.h> had to
be included before <stddef.h> or <stdlib.h> to get rune_t declared.
Now rune_t is declared perfectly bogusly in all cases when <ctype.h>
is included.
This change breaks similar (but more convoluted) convolutions in the
stddef.h in gcc distributions. Ports of gcc should avoid using the
gcc headers.
In __initdb(), a failure to open the local password database is supposed
to result in a warning message being syslog()ed. This warning is only
supposed to be generated as long as the 'warned' flag hasn't been yet;
once the warning is generated, the flag should be set so that the message
is only syslog()ed once. However, while the state of the flag is checked
properly, the flag's state is never changed, so you always get multiple
warnings instead of just one.
Pointed out by: Peter Wemm
This commit covers the man pages for most of the ANSI library functions.
A few others such as strtol.3 have to mention <sys/types.h> because they
mix ANSI interfaces with less well designed extensions.
getnetgrent.c:
- Catch one bogon that snuck by: in _listmatch(), check for '\0'
rather than '\n'; strings returned from yp_match() are terminated
with a nul, not a newline.
getpwent.c:
- Rip out all of the +inclusion/-exclusion stuff from before and
replace it with something a little less grotty. The main problem
with the old mechanism was that it wasted many cycles processing
NIS entries even after it already knew they were to be exlcuded
(or not included, depending on your pointof view). The highlights
of these changes include:
o Uses an in-memory hash database table to keep track of all the
-@netgroup, -user, and -@group exclusions.
o Tries harder to duplicate the behavior normally obtained when using
NIS inclusions/exclusions on a flat /etc/passwd file (meaning things
come out in much the same order).
o Uses seperate methods for handling getpwent() and getpwnam()/getpwuid()
operations instead of trying to do everything with one general
function, which didn't work as well as I thought it would.
o Uses both getnetgrent() and innetgr() to try to save time where
possible.
o Use only one special token in the local password database
(_PW_KEYYPBYNUM) instead of seperate tokens to mark + and -
entries (and stop using the counter tokens too). If this new
token doesn't exist, the code will make due with the standard
_PW_KEYBYNUM token in order to support older databases that
won't have the new token in them.
All this is an attempt to make this stuff work better in environments
with large NIS passwd databases.
- Clear the _yp_innetgr flag immediately after calling setnetgrent() from
innetgr(). We only need the flag set to temporarily alter setnetgrent()'s
behavior. Previously, it was being cleared too late.
- When in NIS-only mode, innetgr() was wasting time doing unecessary
extra processing after it had already found a match.
- Remember to free memory allocated by the NIS functions during innetgr()
searches.
