Bug 8679 - recvfile code path using splice() on Linux leaves data in the pipe on short write.
Summary: recvfile code path using splice() on Linux leaves data in the pipe on short w...
Status: RESOLVED FIXED
Alias: None
Product: Samba 3.6
Classification: Unclassified
Component: File services (show other bugs)
Version: unspecified
Hardware: All All
: P5 normal
Target Milestone: ---
Assignee: Karolin Seeger
QA Contact: Samba QA Contact
URL:
Keywords:
Depends on:
Blocks: 8595
  Show dependency tree / graph
 
Reported: 2011-12-25 05:01 UTC by Jeremy Allison
Modified: 2012-06-06 12:27 UTC (History)
0 users

See Also:

Attachments
git am fix for 3.6.x and 3.5.x. (988 bytes, patch)
2011-12-31 02:01 UTC, Jeremy Allison 		no flags Details
git-am fix for 3.5.x and 3.6.x. (6.39 KB, patch)
2012-01-03 19:36 UTC, Jeremy Allison 		abartlet: review+
Details
Show Obsolete (1) View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Jeremy Allison 2011-12-25 05:01:42 UTC 
Reported by Andrew Bartlett:

> In particular, I'm thinking
>
> if (total_written < count) {
> 
> is not the correct test for draining the socket.  Why does draining the
> socket have anything to do with total_written?  Shouldn't it be:
> 
> if (count) {
> 
> That is, during a multi-part recvfile of 48k of data (due to the 16k
> chunks) I think that we could get a pattern like:
> 
> count = 48000
> total_written = 0
> 
> nread = 16000
> 
> end loop 1:
> total_written = 16000
> count = 32000
> 
> error:
> total_written is < count
> drain 16000 from fromfd, leaving 16000 in fromfd
> 
> or 
> 
> start loop 2:
> nread = 16000
> total_written = 32000
> count = 16000
> 
> error:
> 
> total_written is > count
> no drain, but 16000 still in fromfd

Patch to follow. This is a problem in 3.6.x and also 3.5.x.
Comment 1 Jeremy Allison 2011-12-31 02:01:00 UTC 
Created attachment 7217 [details]
git am fix for 3.6.x and 3.5.x.
Comment 2 Jeremy Allison 2011-12-31 05:13:22 UTC 
Comment on attachment 7217 [details]
git am fix for 3.6.x and 3.5.x.

Canceling this review as the patchset isn't yet complete.
Comment 3 Jeremy Allison 2012-01-03 19:36:32 UTC 
Created attachment 7223 [details]
git-am fix for 3.5.x and 3.6.x.

Full patch fixing all issues. Has gone into master.
Comment 4 Andrew Bartlett 2012-01-03 23:56:12 UTC 
Comment on attachment 7223 [details]
git-am fix for 3.5.x and 3.6.x.

Tested with the sparc recvfile patch and a full filesystem on 3.5.11
Comment 5 Jeremy Allison 2012-01-04 00:03:50 UTC 
Re-assigning to Karolin for inclusion in 3.5.next and 3.6.next.
Jeremy.
Comment 6 Karolin Seeger 2012-01-04 20:32:58 UTC 
Pushed to v3-5-test and v3-6-test.
Closing out bug report.

Thanks!