Write to pipe log using vmsplice() (Trac #3)

[edsiper]

Linux kernel provides a syscall named vmsplice() which allows to write an
iov struct with zero copy to a pipe, we must use vmsplice() instead of
writev() for that procedure.

Migrated from http://bugs.monkey-project.com/ticket/3

{
    "status": "closed", 
    "changetime": "2011-09-06T19:05:21", 
    "description": "Linux kernel provides a syscall named vmsplice() which allows to write an\niov struct with zero copy to a pipe, we must use vmsplice() instead of\nwritev() for that procedure.\n", 
    "reporter": "edsiper", 
    "cc": "", 
    "resolution": "wontfix", 
    "_ts": "1315335921386186", 
    "component": "Logger", 
    "summary": "Write to pipe log using vmsplice()", 
    "priority": "minor", 
    "keywords": "", 
    "version": "", 
    "time": "2010-05-15T12:48:30", 
    "milestone": "Future Release", 
    "owner": "edsiper", 
    "type": "enhancement"
}

[edsiper]

Trac update at 20100726T17:59:53: charlie_brown commented:

I was skimming through the code and it came to my attention the following commented snippet:

        /* for some reason, vmsplice is not working as expected for us, 
         * maybe we need to fix something here, at the moment
         * we will keep using writev to push the iovec struct to the pipe
         *
         *
         * n = vmsplice(fd, 
         *            (const struct iovec *) mk_io->io, 
         *            mk_io->iov_idx, 
         *            SPLICE_F_GIFT);
         * return n;
         */

1. vmsplice's man page states that when the SPLICE_F_GIFT flag is present, the memory pointed by each iovec's iov_base MUST be aligned to page boundary:
   ''"Data must also be properly page aligned, both in memory and length."''
   which, if one dumps the iov_base addresses at run-time, they're not.

2. BUT if we remove SPLICE_F_GIFT, the logger writes to the files with interleaved garbage from the response headers; I haven't checked why.

[edsiper]

Trac update at 20110227T00:07:34:

• edsiper changed milestone from "" to "Future Release"
• edsiper changed priority from "major" to "minor"

[edsiper]

Trac update at 20110906T19:05:21:

• davidlohr commented:

This code has been determined to be unapplicable and has already been removed.

• davidlohr changed resolution from "" to "wontfix"
• davidlohr changed status from "new" to "closed"