Bug 829172 - reopen_fd_count is becoming -ve because of stale fdctx
Summary: reopen_fd_count is becoming -ve because of stale fdctx
Keywords:
Status: CLOSED EOL
Alias: None
Product: GlusterFS
Classification: Community
Component: protocol
Version: pre-release
Hardware: Unspecified
OS: Unspecified
high
high
Target Milestone: ---
Assignee: bugs@gluster.org
QA Contact:
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2012-06-06 06:59 UTC by Pranith Kumar K
Modified: 2015-10-22 15:40 UTC (History)
3 users (show)

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Clone Of:
Environment:
Last Closed: 2015-10-22 15:40:20 UTC
Regression: ---
Mount Type: ---
Documentation: ---
CRM:
Verified Versions:
Embargoed:


Attachments (Terms of Use)

Description Pranith Kumar K 2012-06-06 06:59:00 UTC
Description of problem:
Reopen of a fd is performed after a brick comes online. If the reopen is for an fd which is already marked as released and the reopen fails  decrementing of reopen_fd_count is done but the fdctx corresponding to the released fd is still added to the saved_fds list. That stale fdctx remains there for ever leading to -ve count of reopen_fd_count on subsequent 'CHILD_DOWN' then 'CHILD_UP'.

Please see the following gdb logs:

Here the reopen_fd_count becomes zero.
(gdb) s
decrement_reopen_fd_count (this=0x22768f0, conf=0x22beb70) at client-lk.c:591
591	        uint64_t fd_count = 0;
(gdb) n
593	        LOCK (&conf->rec_lock);
(gdb) 
595	                fd_count = --(conf->reopen_fd_count);
(gdb) 
597	        UNLOCK (&conf->rec_lock);
(gdb) 
599	        if (fd_count == 0) {
(gdb) 
600	                gf_log (this->name, GF_LOG_INFO,
(gdb) 
602	                client_set_lk_version (this);
(gdb) 
603	                client_notify_parents_child_up (this);
(gdb) 
606	        return fd_count;
(gdb) 
607	}

Breakpoint 10, clnt_release_reopen_fd_cbk (req=0x7f5f5f16733c, iov=0x7f5f5f16737c, count=1, myframe=0x7f5f677a4894)
    at client-handshake.c:595
595	        xlator_t       *this   = NULL;
(gdb) n
596	        call_frame_t   *frame  = NULL;
(gdb) 
597	        clnt_conf_t    *conf   = NULL;
(gdb) 
598	        clnt_fd_ctx_t  *fdctx  = NULL;
(gdb) 
600	        frame  = myframe;
(gdb) 
601	        this   = frame->this;
(gdb) 
602	        fdctx  = (clnt_fd_ctx_t *) frame->local;
(gdb) 
603	        conf   = (clnt_conf_t *) this->private;
(gdb) 
605	        clnt_fd_lk_reacquire_failed (this, fdctx, conf);
(gdb) 
607	        decrement_reopen_fd_count (this, conf);
(gdb) s
decrement_reopen_fd_count (this=0x22768f0, conf=0x22beb70) at client-lk.c:591
591	        uint64_t fd_count = 0;
(gdb) n
593	        LOCK (&conf->rec_lock);
(gdb) 
595	                fd_count = --(conf->reopen_fd_count);
(gdb) 
597	        UNLOCK (&conf->rec_lock);
(gdb) p fd_count
$18 = 18446744073709551615
(gdb) p conf->reopen_fd_count 
$19 = 18446744073709551615


Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.
  
Actual results:


Expected results:


Additional info:

Comment 1 Junaid 2012-06-06 07:19:17 UTC
The check for this is there when we enable lock-self healing. The above condition was not checked because lock self healing was expected to be always on. Since now its optional we must handle this case.

Comment 3 Kaleb KEITHLEY 2015-10-22 15:40:20 UTC
pre-release version is ambiguous and about to be removed as a choice.

If you believe this is still a bug, please change the status back to NEW and choose the appropriate, applicable version for it.


Note You need to log in before you can comment on or make changes to this bug.