glusterfs.git/xlators/performance/io-cache/src/io-cache.c, branch v3.10.1

io-cache: Fix a read hang

2016-11-23T13:11:07+00:00

Issue:
=====
In certain cases, there was no unwind of read
from read-ahead xlator, thus resulting in hang.

RCA:
====
In certain cases, ioc_readv() issues STACK_WIND_TAIL() instead
of STACK_WIND(). One such case is when inode_ctx for that file
is not present (can happen if readdirp was called, and populates
md-cache and serves all the lookups from cache).

Consider the following graph:
...
io-cache (parent)
   |
readdir-ahead
   |
read-ahead
...

Below is the code snippet of ioc_readv calling STACK_WIND_TAIL:
ioc_readv()
{
...
 if (!inode_ctx)
   STACK_WIND_TAIL (frame, FIRST_CHILD (frame->this),
                    FIRST_CHILD (frame->this)->fops->readv, fd,
                    size, offset, flags, xdata);
   /* Ideally, this stack_wind should wind to readdir-ahead:readv()
      but it winds to read-ahead:readv(). See below for
      explaination.
    */
...
}

STACK_WIND_TAIL (frame, obj, fn, ...)
{
  frame->this = obj;
  /* for the above mentioned graph, frame->this will be readdir-ahead
   * frame->this = FIRST_CHILD (frame->this) i.e. readdir-ahead, which
   * is as expected
   */
  ...
  THIS = obj;
  /* THIS will be read-ahead instead of readdir-ahead!, as obj expands
   * to "FIRST_CHILD (frame->this)" and frame->this was pointing
   * to readdir-ahead in the previous statement.
   */
  ...
  fn (frame, obj, params);
  /* fn will call read-ahead:readv() instead of readdir-ahead:readv()!
   * as fn expands to "FIRST_CHILD (frame->this)->fops->readv" and
   * frame->this was pointing ro readdir-ahead in the first statement
   */
  ...
}

Thus, the readdir-ahead's readv() implementation will be skipped, and
ra_readv() will be called with frame->this = "readdir-ahead" and
this = "read-ahead". This can lead to corruption / hang / other problems.
But in this perticular case, when 'frame->this' and 'this' passed
to ra_readv() doesn't match, it causes ra_readv() to call ra_readv()
again!. Thus the logic of read-ahead readv() falls apart and leads to
hang.

Solution:
=========
Ideally, STACK_WIND_TAIL() should be modified as:
STACK_WIND_TAIL (frame, obj, fn, ...)
{
  next_xl = obj /* resolve obj as the variables passed in obj macro
                   can be overwritten in the further instrucions */
  next_xl_fn = fn /* resolve fn and store in a tmp variable, before
                     modifying any variables */
  frame->this = next_xl;
  ...
  THIS = next_xl;
  ...
  next_xl_fn (frame, next_xl, params);
  ...
}
But for this solution, knowing the type of variable 'next_xl_fn' is
a challenge and is not easy. Hence just modifying all the existing
callers to pass "FIRST_CHILD (this)" as obj, instead of
"FIRST_CHILD (frame->this)".

Change-Id: I179ffe3d1f154bc5a1935fd2ee44e912eb0fbb61
BUG: 1388292
Signed-off-by: Poornima G 
Reviewed-on: http://review.gluster.org/15901
Smoke: Gluster Build System 
Reviewed-by: Raghavendra G 
NetBSD-regression: NetBSD Build System 
CentOS-regression: Gluster Build System

performance/open-behind: Pass O_DIRECT flags for anon fd reads when required

2016-09-23T05:32:14+00:00

Writes are already passing the correct flags at the time of open().

Also, make io-cache honor direct-io for anon-fds with
O_DIRECT flag during reads.

Change-Id: I215cb09ef1b607b9f95cabf0ef3065c00edd9e78
BUG: 1377556
Signed-off-by: Krutika Dhananjay 
Reviewed-on: http://review.gluster.org/15537
Smoke: Gluster Build System 
NetBSD-regression: NetBSD Build System 
Reviewed-by: Raghavendra G 
CentOS-regression: Gluster Build System

all: reduce "inline" usage

2015-09-01T11:55:15+00:00

There are three kinds of inline functions: plain inline, extern inline,
and static inline.  All three have been removed from .c files, except
those in "contrib" which aren't our problem.  Inlines in .h files, which
are overwhelmingly "static inline" already, have generally been left
alone.  Over time we should be able to "lower" these into .c files, but
that has to be done in a case-by-case fashion requiring more manual
effort.  This part was easy to do automatically without (as far as I can
tell) any ill effect.

In the process, several pieces of dead code were flagged by the
compiler, and were removed.

Change-Id: I56a5e614735c9e0a6ee420dab949eac22e25c155
BUG: 1245331
Signed-off-by: Jeff Darcy 
Reviewed-on: http://review.gluster.org/11769
Tested-by: NetBSD Build System 
Reviewed-by: Dan Lambright 
Reviewed-by: Kaleb KEITHLEY 
Reviewed-by: Niels de Vos 
Reviewed-by: Raghavendra Bhat 
Reviewed-by: Venky Shankar

Logging: Porting the performance translator

2015-06-27T09:29:27+00:00

         logs to new logging framework.

Change-Id: Ie6aaf8d30bd4457bb73c48e23e6b1dea27598644
BUG: 1194640
Signed-off-by: arao 
Reviewed-on: http://review.gluster.org/9822
Tested-by: NetBSD Build System 
Tested-by: Gluster Build System 
Reviewed-by: Raghavendra G 
Tested-by: Raghavendra G

build: do not #include "config.h" in each file

2015-05-29T12:43:17+00:00

Instead of including config.h in each file, and have the additional
config.h included from the compiler commandline (-include option).

When a .c file tests for a certain #define, and config.h was not
included, incorrect assumtions were made. With this change, it can not
happen again.

BUG: 1222319
Change-Id: I4f9097b8740b81ecfe8b218d52ca50361f74cb64
Signed-off-by: Niels de Vos 
Reviewed-on: http://review.gluster.org/10808
Tested-by: Gluster Build System 
Tested-by: NetBSD Build System
Reviewed-by: Kaleb KEITHLEY 
Reviewed-by: Pranith Kumar Karampuri

Avoid conflict between contrib/uuid and system uuid

2015-04-04T17:48:35+00:00

glusterfs relies on Linux uuid implementation, which
API is incompatible with most other systems's uuid. As
a result, libglusterfs has to embed contrib/uuid,
which is the Linux implementation, on non Linux systems.
This implementation is incompatible with systtem's
built in, but the symbols have the same names.

Usually this is not a problem because when we link
with -lglusterfs, libc's symbols are trumped. However
there is a problem when a program not linked with
-lglusterfs will dlopen() glusterfs component. In
such a case, libc's uuid implementation is already
loaded in the calling program, and it will be used
instead of libglusterfs's implementation, causing
crashes.

A possible workaround is to use pre-load libglusterfs
in the calling program (using LD_PRELOAD on NetBSD for
instance), but such a mechanism is not portable, nor
is it flexible. A much better approach is to rename
libglusterfs's uuid_* functions to gf_uuid_* to avoid
any possible conflict. This is what this change attempts.

BUG: 1206587
Change-Id: I9ccd3e13afed1c7fc18508e92c7beb0f5d49f31a
Signed-off-by: Emmanuel Dreyfus 
Reviewed-on: http://review.gluster.org/10017
Tested-by: Gluster Build System 
Reviewed-by: Niels de Vos

glfs_fini: Clean up all the resources allocated in glfs_new.

2015-03-04T15:15:12+00:00

Initially even after calling glfs_fini(), all the threads created
during init and many other resources like memory pool, iobuf pool,
event pool and other memory allocs were not being freed.
With this patch these resources are freed in glfs_fini().

The two thumb rules followed in this patch are:
- The threads are not killed, they are made to exit voluntarily,
once the queued tasks are completed. The main thread waits for
the other threads to exit.
- Free the memory pools and destroy the graphs only after all the
other threads are stopped, so that there are less chances of
hitting access after free.

Resources freed and its order:
1. Destroy the inode table of all the graphs - Call forget on all the inodes.
This will not be required when the cleanup during graph switch is
implemented to perform inode table destroy.
2. Deactivate the current graph, call fini of all the xlators.
3. Syncenv destroy - Join the synctask threads and cleanup syncenv resources
Sets the destroy mode, complete the existing synctasks, then join the
synctask threads.
After entering the destroy mode,
-if a new synctask is submitted, it fails.
-if syncenv_new() is called, it will end up creating new threads,
but this is called only during init.
4. Poller thread destroy
Register an event handler which sets the destroy mode for the poller.
Once the poller is done processing all the events, it exits.
5. Tear down the logging framework
The log file is closed and the log level is set to none, after this
point no log messages appear either in log file or in stderr.
6. Destroy the timer thread
Set the destroy bit, once the pending timer events are processed
the timer thread exits.
Note: Log infrastructure should be shutdown before destroying the timer
thread as gf_log uses timers.
7. Destroy the glusterfs_ctx_t
For all the graphs(active and passive), free graph, xlator structs and few other lists.
Free the memory pools - iobuf pool, event pool, dict, logbuf pool,
stub mem pool, stack mem pool, frame mem pool.

Few things not addressed in this patch:
1. rpc_transport object not destroyed, the PARENT_DOWN should have
destroyed this object but has not, needs to be addressed as a part
of different patch
2. Each xlator fini should clean up the local pool allocated by its xlator.
Needs to be addresses as a part of different patch.
3. Each xlator should implement forget to free its inode_ctx.
Needs to be addresses as a part of different patch.
3. Few other leaks reported by valgrind.
4. fd and fd contexts

The numbers:
The resource usage by the test case in this patch:
Without the fix, Memory: ~3GB; Threads: ~81
With this fix, Memory: 300MB; Threads: 1(main thread)

Change-Id: I96b9277541737aa8372b4e6c9eed380cb871e7c2
BUG: 1093594
Signed-off-by: Poornima G
Reviewed-on: http://review.gluster.org/7642
Tested-by: Gluster Build System
Reviewed-by: Rajesh Joseph
Reviewed-by: Raghavendra Talur
Reviewed-by: Krishnan Parthasarathi
Reviewed-by: Pranith Kumar Karampuri
Reviewed-by: Shyamsundar Ranganathan

build: MacOSX Porting fixes

2014-04-24T21:41:48+00:00

git@forge.gluster.org:~schafdog/glusterfs-core/osx-glusterfs

Working functionality on MacOSX

 - GlusterD (management daemon)
 - GlusterCLI (management cli)
 - GlusterFS FUSE (using OSXFUSE)
 - GlusterNFS (without NLM - issues with rpc.statd)

Change-Id: I20193d3f8904388e47344e523b3787dbeab044ac
BUG: 1089172
Signed-off-by: Harshavardhana 
Signed-off-by: Dennis Schafroth 
Tested-by: Harshavardhana 
Tested-by: Dennis Schafroth 
Reviewed-on: http://review.gluster.org/7503
Tested-by: Gluster Build System 
Reviewed-by: Anand Avati

io-cache: Unlock and then goto out in failure case.

2014-02-17T16:09:55+00:00

Fix for coverity bug CID:1124625

Change-Id: I76df453a17f2af7c48a80b6fc0ccd411ab96e371
BUG: 789278
Signed-off-by: Raghavendra Talur 
Reviewed-on: http://review.gluster.org/6949
Reviewed-by: Santosh Pradhan 
Tested-by: Gluster Build System

zerofill: Change the type of len argument of glfs_zerofill() to off_t

2013-11-15T07:29:48+00:00

glfs_zerofill() can be potentially called to zero-out entire file and
hence allow for bigger value of length parameter.

Change-Id: I75f1d11af298915049a3f3a7cb3890a2d72fca63
BUG: 1028673
Signed-off-by: Bharata B Rao 
Reviewed-on: http://review.gluster.org/6266
Tested-by: Gluster Build System 
Reviewed-by: M. Mohan Kumar 
Tested-by: M. Mohan Kumar 
Reviewed-by: Anand Avati