For a Feedback: this fix worked for me. There were no crashes since I
installed the new package.
Marcus
Aníbal Monsalve Salazar escreveu:
On Wed, Sep 02, 2009 at 03:48:46PM +0200, Sandro Serafini wrote:
To fix this bug, you have to modify the 05_handle-short-read.dpatch
patch, in this way:
The THREE istance of (row number 58, 94, 133 of the
05_handle-short-read.dpatch file):
+ num_bytes += bytes_read;
Should be modified in:
+ if (bytes_read>0) num_bytes += bytes_read;
Because when read returns error, this should not influence the number
of bytes already read.
AND
The THREE istance of (row number 60, 96, 135 of the
05_handle-short-read.dpatch file):
+ while ((bytes_read > 0 || errno == EINTR) [...]
Should be modified in:
+ while ((bytes_read > 0 || (bytes_read==-1 && errno == EINTR)) [...]
Because errno is only valid when read returns -1, errno is not set if
there are no errors from the "read"; so if "read" gets interrupted by an
EINTR, then when "read" reaches end-of-file, "read" will return 0
with errno still set to EINTR, and the loop become an infinite loop.
Please someone upstream the fix and inform the author of the patch,
thanks!
Sandro
Thank you Sandro for the fix.
Marcus and Artur, could you please test Sandro's fix for this bug?
Cheers!
