Re: [tbox-l] rsync: writefd_unbuffered failed to write 32768 bytes: phase "unknown"

[Michael W. <mw...@un...>]

Philip and other IBM folks,

Following up the rsync issue, I have found more useful information.

Problem:
rsync as packaged by IBM can not write beyond 2,147,483,647 bytes.

Proof:

(1) while rsync from a remote box (Solaris 8) to IBM AIX 5.2:

$ /opt/csw/bin/rsync -av --rsh=ssh users01.dbf oracle@zimu:/max-ds1/ora...
...
rsync: writefd_unbuffered failed to write 4 bytes: phase "unknown": Broken pipe
rsync error: error in rsync protocol data stream (code 12) at io.c(515)

I monitor the temp file:

$ while true; do ls -ls .u*; df -k .; done

Filesystem    1024-blocks      Free %Used    Iused %Iused Mounted on
/dev/lv01         7503872    292200   97%    23225     2% /max-ds1
2097152 -rw-------   1 oracle   dba      2147483647 Jun 15 09:06 .users01.dbf.OS2Oaa

Filesystem    1024-blocks      Free %Used    Iused %Iused Mounted on
/dev/lv01         7503872    292200   97%    23225     2% /max-ds1
ls: 0653-341 The file .u* does not exist.

The write failed exactly at 2,147,483,647 which is 2*1024*1024*1024 - 1.

And the truss -f -p <rsync_pid> shows:

28448:  kread(0, "\0\0\0\0\0\0\0\0\0\0\0\0".., 19688)   = 16384
28448:  _select(1, 0x2FF205D8, 0x00000000, 0x00000000, 0x2FF205D0) = 1
28448:  kread(0, "\0\0\0\0\0\0\0\0\0\0\0\0".., 3304)    = 3304
28448:  kwrite(1, "\002\0\00203 ÿ ü\0\0\0\0".., 32768)  = 32767
28448:  kwrite(6, " &\0\0\b w r i t e   f a".., 42)     = 42
28448:  _sigaction(30, 0x2FF21AC0, 0x2FF21AD0)          = 0
28448:  _sigaction(31, 0x2FF21AC0, 0x2FF21AD0)          = 0
28448:  unlink(".users01.dbf.OS2Oaa")                   = 0
28448:  kwrite(6, " ;\0\0\b r s y n c   e r".., 63)     = 63
28448:  kfcntl(1, F_GETFL, 0x00000000)                  = 2
28448:  kfcntl(2, F_GETFL, 0x00000000)                  = 2
28448:  _exit(11)

I would think it should use *write64(). Is kwrite() capable handling
2+GB file?

(2) Could this be the remote rsync problem?

No. Because:

(2.1) rsync 6+GB file to the remote box succeeded as I reported
earlier, which means the remote rsync is capable write 6+GB file.
And this appears that rsync on IBM AIX 5.2 is capable reading
6+GB file.
(2.2) rsync on IBM AIX 5.2 locally also fails. 
(2.3) rsync on remote box to itself succeeded for the same 2+GB file.

Environment:
# oslevel -r
5200-02
# rpm -ql rsync-2.5.4-1
/opt/freeware/bin/rsync
/opt/freeware/doc/rsync-2.5.4
/opt/freeware/doc/rsync-2.5.4/COPYING
/opt/freeware/doc/rsync-2.5.4/README
/opt/freeware/doc/rsync-2.5.4/tech_report.tex
/opt/freeware/man/man1/rsync.1
/opt/freeware/man/man5/rsyncd.conf.5
/usr/bin/rsync

Other info:

- Straightforward compilation for the latest version of rsync 2.6.2
  have the same problem. Modify config.h not to mkstemp() may fix it.

- This problem should be very easy to reproduce.

-- 
Michael Wang * http://www.unixlabplus.com/ * mw...@un...