getsockname() and getpeername() fail when using high memory

[rudi]

The functions getsockname() and getpeername() will fail, if the variable containing the size of "struct sockaddr" is located in high memory.

It is suggested not to pass the address of that variable directly to the IP stack but use an intermediate local instead.

See ​http://svn.netlabs.org/qt4/ticket/179#comment:4

[bird]

(In [3674]) libsocket: Made it possible to pass sockaddr parameters in high memory to accept, getsockname, bind, connect and getpeername. (untested) References #233.

[bird]

(In [3675]) 0.6: Backporting 3674 - part 1. References #233.

[bird]

(In [3676]) 0.6: Backporting r3674 - part 2. Fixes #233.

[Silvan Scherrer]

in my tests with samba and libc064 connect() still does not work when built with -Zhigh-mem

[Silvan Scherrer]

as a addition socketpair() doesn't work also when the fd is in high-mem

[Silvan Scherrer]

see ​http://svn.netlabs.org/qt4/ticket/179 for a little testcase. sockname_fail.zip

[Silvan Scherrer]

getpeername() always returns -1 now.
http://svn.netlabs.org/libc/changeset?reponame=&new=3674%40trunk%2Flibc%2Fsrc%2Flibsocket%2Fgetpeername.c&old=2446%40trunk%2Flibc%2Fsrc%2Flibsocket%2Fgetpeername.c
the LIBCLOG_RETURN_INT(rc); on line 48 was eliminated by accident i guess

[dmik]

No, I guess it was eliminated on purpose, but the last statement should be LIBCLOG_RETURN_INT(rc);. Anyway, for the time being, the hot fix of libc_dll.lib is here ​ftp://ftp.netlabs.org/pub/odin/test/libc_dll_233_regression_fixed.zip.

[bird]

Changesets r3745 and r3746 addresses the regressions. r3747 addresses the socketpair() high memory issue. (0.6 branch)

[bird]

r3748 forward ports the changes to the trunk. Case closed.