View Issue Details
| ID | Project | Category | View Status | Date Submitted | Last Update |
|---|---|---|---|---|---|
| 0003911 | unreal | ircd | public | 2010-06-06 17:34 | 2010-06-21 18:47 |
| Reporter | mykro | Assigned To | |||
| Priority | normal | Severity | major | Reproducibility | always |
| Status | closed | Resolution | duplicate | ||
| Product Version | 3.2.8 | ||||
| Summary | 0003911: ircd does not start with MAXCONNECTIONS > 1024 | ||||
| Description | Hello, the ircd does not start when MAXCONNECTIONS is set to a value higher than 1024 in the ./Config script. The ulimit hard and soft value are correctly set but FD_SETSIZE is limited to 1024. Actually, ./configure tries to set FD_SETSIZE via the gcc -D switch which failes because the macros defined via -D are set before the code is preprocessed so that any later #define's in the source code or in any header overwrites the value given with -D. The solution (rather a dirty hack) proposed in the FAQs (see http://www.unrealircd.com/faq.php#46) is not acceptable because it requires to change a central header file that is managed normally with the system's package system (e.g. apt-get). SOLUTION: ========= I solved the problem by first undefining the FD_SETSIZE macro (which is being set by the system's header files) and then redefine it within a properly selected unrealircd header file. I've chosen setup.h because it is included quite late so that the preprocessor commands are executed after the system's defaults: #undef FD_SETSIZE #define FD_SETSIZE 8192 I recommend to change the ./configure script: - remove the -D FD_SETSIZE switch which is completely useless - add the undef and the define to setup.h (or any other suitable header file) if MAXCONNECTIONS is bigger than the system's FD_SETSIZE default value Regards mykro (irc.paranode.net) | ||||
| 3rd party modules | |||||
|
|
The solution you propose is not valid. Basically, you redefine FD_SETSIZE to some value so the warning goes away, but the fd_set struct itself stays 1024 bits (if 1024 is what the system header uses) no matter what you define later on. The result is stack and/or heap corruption. Unfortunately the way proposed in the FAQ is the only solution on Linux. It can only go away by 0001245 / 0003361 (which are not for 3.2*, by the way). |
|
|
Hello, I started a discussion on the glibc mailing list (http://www.cygwin.com/ml/libc-help/2010-06/) about the idea to add a possibility to raise __FD_SETSIZE. I also opened a glibc enhancement request (http://sources.redhat.com/bugzilla/show_bug.cgi?id=11687). My proposal would finally resolve the unrealircd maxconnections problem without the need of modifying system headers. There is a living discussion about this issue with supporters and opponents. Please follow up the discussion on the mailing list and in bugzilla. What we need to explain is that a #ifndef __FD_SETSIZE would not create any API or ABI issues. Thank you. mykro (irc.paranode.net) |
|
|
thanks for the info & effort :). Though I don't think it will help much, but we'll see. other than that, this is a dup of (or very related to) 0001245 |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2010-06-06 17:34 | mykro | New Issue | |
| 2010-06-06 19:32 | syzop | QA | => Not touched yet by developer |
| 2010-06-06 19:32 | syzop | U4: Need for upstream patch | => No need for upstream InspIRCd patch |
| 2010-06-06 19:32 | syzop | Note Added: 0016113 | |
| 2010-06-06 19:32 | syzop | Status | new => closed |
| 2010-06-06 19:32 | syzop | Resolution | open => not fixable |
| 2010-06-21 14:05 | mykro | Note Added: 0016123 | |
| 2010-06-21 14:05 | mykro | Status | closed => feedback |
| 2010-06-21 14:05 | mykro | Resolution | not fixable => reopened |
| 2010-06-21 18:47 | syzop | Note Added: 0016127 | |
| 2010-06-21 18:47 | syzop | Status | feedback => closed |
| 2010-06-21 18:47 | syzop | Resolution | reopened => duplicate |
