Compute max_transfer args as long, not int, to avoid overflow when
max_sg_cnt=32768 and PAGE_SHIFT=16 (64K PAGE_SIZE). Note it is not
only the min() call, but also the << PAGE_SHIFT needs to be long.
Signed-off-by: David Butterfield <dab21774@gmail.com>
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@7189 d57e44dd-8a1f-0410-8b47-8ef2f437770f
From: Quinn Tran <quinn.tran@cavium.com>
Out of order(OOO) processing requires initiator, switch
and target to support OOO. In today¹s environment, none
of the switches support OOO. OOO requires extra buffer
space which affect performance. By turning ON this feature
in QLogic's FW, it delays error recovery because droped
frame is treated as out of order frame. We¹re turning OFF
this option of speed up error recovery.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Merge of mainline commit 300af14bdb28157090f0c6f89d244fda940082da
EXPERIMENTAL!
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@7188 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Waiting for other threads to release an object using code like
"while (object->refcnt > 0) msleep(100)" without holding a reference
on 'object' is wrong because the memory object points at may be freed
before or while this loop is in progress. Hence introduce a global
portal object count and wait on that count instead of waiting for
the per-portal reference count to reach zero.
The use-after-free was introduced in r6952 ("isert: faster release
of isert_scst module").
See also https://sourceforge.net/p/scst/tickets/2/.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@7184 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Change scst_finish_cmd() so that it does not attempt to left-shift lba when it
is -1 (the error value). Left-shifting of a negative value is undefined in C.
Signed-off-by: David Butterfield <dab21774@gmail.com>
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@7172 d57e44dd-8a1f-0410-8b47-8ef2f437770f
This patch does not change any functionality but makes it easier
to port SCST to Linux kernel v4.11.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@7152 d57e44dd-8a1f-0410-8b47-8ef2f437770f
This change helped a secondary problem I had under valgrind (due to some other
bug) with initiators timing out and reconnecting their sessions faster than the
threads were getting cleaned up (for one thing, valgrind only ever runs one
thread at a time).
It seems logical that the code in conn_lookup() would want to skip any
connection that's already known closing in this case, for the same reason it
searches the list in reverse. And the dropping connections don't have to drop
in order, so searching in reverse doesn't seem sufficient to avoid finding a
wrong (stale, closing) connection structure.
Signed-off-by: David Butterfield <dab21774@gmail.com>
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@7144 d57e44dd-8a1f-0410-8b47-8ef2f437770f
In an error path in iscsi_threads_pool_get(), when a new pool cannot be
allocated, if there is a pool on iscsi_thread_pools_list, it passes that back as
an alternative, so return zero in that case.
Signed-off-by: David Butterfield <dab21774@gmail.com>
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@7143 d57e44dd-8a1f-0410-8b47-8ef2f437770f
In scsi_cmnd_start() conn->rx_task is intended to be set only during the
call to _stage1_done, but there was no code to reset it. It doesn't
actually matter in execution, but it was a little confusing to wonder
and try to understand why it wasn't reset.
Signed-off-by: David Butterfield <dab21774@gmail.com>
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@7142 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Vladislav Bolkhovitin
Bart Van Assche
Israel Rukshin