| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
|
|
|
|
|
|
| |
Extended CHAP authentication modes using newer hash digest functions
than MD5.
Makes OpenSSL libcrypto use a requirement, replacing the in tree MD5
implementation as well (and removing the unused SHA1 code).
|
|
|
|
| |
Updated Changelog and two include files.
|
|\
| |
| | |
Add make target to install systemd units
|
|/ |
|
| |
|
|\
| |
| | |
The iscsi login/logout service requires iscsid.
|
|/
|
|
|
|
| |
The iscsi.service systemd unit file, which controls
iscsi logins and logouts, should be stopped if the
iscsid service file is stopped.
|
|\
| |
| | |
rec update: disable the idbm_lock in read/write when updating the rec
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The iscsiadm is getting a file lock while writing out records, and
updates are a read-modify-write operation and currently only the
write is locked.
So the parallel requests are reading in the pre-update record and
then overwriting each other with the writes.
Fixing RHBZ#1624670
Signed-off-by: Xiubo Li <xiubli@redhat.com>
|
|\ \
| |/
|/| |
iscsiuio bug fix
|
| |
| |
| |
| | |
Signed-off-by: Nilesh Javali <njavali@marvell.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
In the event of DHCP failure, killing of enable_nic_thread did not
process any iscsid requests leading to error,
iscsistart: Could not broadcast to uIP after 5 tries
and login failure to next active path.
Do not kill enable_nic_thread and allow further processing of iscsid
requests and performing login to next active path.
Signed-off-by: Nilesh Javali <njavali@marvell.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Commit c1870ae6dddb7bc2dbdc750df68edbbe1a80763f added systemd
support to iscsiuio, but it was not integrated with the
autoconf script there, so this commit rectifies that. Now one
can build without systemd support, from the top, using:
$ make OPTFLAGS="... -DNO_SYSTEMD ...
which now works for both iscsid and iscsiuio.
|
|\ \
| | |
| | | |
Make iscsid systemd usage optional
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
You can compile without system now by using something
like:
make OPTFLAGS="-DNO_SYSTEMD ..." NO_SYSTEMD=1
This will skip systemd code for iscsid and iscsiuio.
|
|\ \ \
| | | |
| | | | |
Fix systemd startup etc
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Add in example socket and service files for iscsiuio, such
that it can socket-activated, as is iscsid. Also, update
main in iscsiuio to use sd_notify() when it is ready for
business. All of the systemd code can be skipped if
NO_SYSTEMD is defined when building.
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Only send a message down the pipe if it is open, i.e.
only in daemon (non-foreground) mode.
Fixes: 9483ee87ceea3ec8816e6a39d93aee242e7e19d2
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Add a pipe from child to parent when running in the
background (daemon mode), so that the parent does not exit
until the child is ready for business, so that systemd does
not send us work too early. Also, fix error messages,
adding newlines where needed, and printing them to
stderr.
|
| |/ /
| | |
| | |
| | |
| | | |
Convert /var/run to /run, and convert /var/lock to /run/lock,
as per modern requirements on use of the /var directory.
|
|\ \ \
| |/ /
|/| | |
Fix possible discovery hang when timing out
|
|/ /
| |
| |
| |
| |
| |
| | |
If session reopening failed during discovery, the CPU
could peg at 100% because the code that gives up when
tretires are exhausted was not setting an error
return value.
|
|\ \
| | |
| | | |
Fix output of node printing for multiple paths.
|
|/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The output of "iscsiadmm -m node -P*" changed when
libopeniscsiusr was merged, and an earlier attempt was
made to clean it up (see commit 59806c99fb3f). This
commit fixes output so that. Before this fix, if one
had multiple paths to the same target, the output
of "iscsiadm -m node -P1" would be:
> Target: iqn.2008-04.net.gonzoleeman:01:disk1
> Portal: 10.0.0.1:3260,1
> Iface Name: default
> Target: iqn.2008-04.net.gonzoleeman:01:disk1
> Portal: 10.0.1.1:3260,2
> Iface Name: default
But this commit fixes the output to be that of iscsiadm
before the libopeniscsiusr merge, i.e.:
> Target: iqn.2008-04.net.gonzoleeman:01:disk1
> Portal: 10.0.0.1:3260,1
> Iface Name: default
> Portal: 10.0.1.1:3260,2
> Iface Name: default
|
|\ \
| | |
| | | |
Fix printing of node database again.
|
|/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This fixes commit ec2afb8a1c997bac706de4f70a3a59fdd82b5741,
which aimed to make the return value from "iscsiadm -m node -P[01]"
return the correct value of "21", for no targets found. But
instead it make the output of all "iscsiadm -m node -P*" act like
the output of "iscsiadm -m node --op show", which was just wrong.
Instead, revert the above-mentioned commit, then add a simple
check for "no nodes" in print_nodes().
|
|\ \
| | |
| | | |
Add Restart=on-failure option to iscsid.service
|
| | |
| | |
| | |
| | | |
This change adds Restart=on-failure systemd service option to iscsid.service config file so that iscsid daemon is restarted on failure. This option is particularly useful when using iscsi boot device and iscsid daemon crashes or is inadvertently killed.
|
|\ \ \
| |/ /
|/| | |
Fix node print return value when no nodes.
|
|/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
When we switched to using libopeniscsiusr, the behavior of iscsiadm
when running "iscsiadm -m node" diverged with that seen when
running "iscsiadm -m node --op show", whereas before that time
(version 2.0.873) was the same for these two cases. Note that
iscsiadm return value used to be the same in both cases, i.e.
returning 21, meaning "no targets found", but now returns
21 for the "--op show" case and zero for the other.
The fix is simple: rip out the code in iscsiadm that treats
"iscsiadm -m node" as a special case, allowing the code to fall
through to its previous behavior.
|
|\ \
| | |
| | | |
Fixed iscsi.service considering every signal and exit code as successful
|
|/ /
| |
| |
| | |
Now only code 21 (no objects found to execute on) and normal exit conditions are valid.
|
| |
| |
| |
| |
| |
| |
| |
| | |
This is a regression, in that "iface.example" was
ignored in pre-libopeniscsiusr days. But now, if
you run "iscsiadm -m iface -P1" you see a warning
about iface.example not being a valid interface file.
This commit skips that interface file in this case.
|
|\ \
| | |
| | | |
Make iscsid.service a requirement.
|
|/ /
| |
| | |
This will ensure daemon is loaded before iscsi.service starts. Also in case daemon fails to load iscsi.service will not be started.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The output from 'iscsiadm -m node -P1' and 'iscsiadm -m iface -P1'
has regressed with the introduction of libopeniscsiusr.
The pre-lib output for 'iscsiadm -m node -P1' before was:
>Target: <TARGET-IQN>
> Portal: <IP-ADDRESS>:<PORT>,1
> Iface Name: default
repeated for each Node in the node datbase, and the output
of 'iscsiadm -m iface -P1' before the change was:
>Iface: default
> Target: <TARGET-IQN>
> Portal: <IP-ADDRESS>:<PORT>,1
repeated once for each Node in the database that uses the
'default' interface (i.e. usually all nodes).
In both of these cases, the output contained details for each
and every node in the database. But this evidently wasn't the
intention if one looks at the code. The code appears to try
to only print the details for each 'Target' if and only if
those details are different than the previous line. In other
words, the code looks like it wants to suppress the 'Portal'
and 'Iface Name' lines if they are the same as the previously-
listed target. This explains why libopeniscsiusr implemented
the output of iscsiadm to suppress this info on all but the
first target, assuming the Portal and IfaceName were the
same.
The fix is to simplify print_nodes_tree() to print details
for every Node, since that is how the output worked
before the libopeniscsiusr update.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Commit 1a8d7036d0f9 added the "onboot" parameter for use
with iscsiadm's "-L" and "-U" options, so update the man
page to mention them.
Simple cleanups of the man page were done as well, with
no functional change. A diff of the old man page and the
new show only a few non-functional changes in addition
to adding the "onboot" information.
|
|\ \
| |/
| | |
iscsiuio bug fixes
|
| |
| |
| |
| | |
Signed-off-by: Nilesh Javali <nilesh.javali@cavium.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
1. Use HW BD prod index to read packet buffer descriptor.
2. Use BD producer and consumer index to check for new data instead
of UIO ring buffer index.
NOTE - This patch has dependency on upstream qedi driver v8.33.0.21,
commit "qedi: Move LL2 producer index processing in BH.",
(https://marc.info/?l=linux-scsi&m=154271028608808)
Signed-off-by: Manish Rangankar <manish.rangankar@cavium.com>
|
| |
| |
| |
| |
| |
| | |
Invalid ether len was resulting into showing CRC error on wire.
Signed-off-by: Manish Rangankar <manish.rangankar@cavium.com>
|
|\ \
| | |
| | | |
Use pkg-config in Makefiles for newer libraries.
|
|/ /
| |
| |
| |
| |
| | |
These two recently-added libraries can be in different
locations on different distros, so use pkg-config to
added the appropriate actions in the make files.
|
|\ \
| | |
| | | |
Fix i586 build issues with string length overflow.
|
|/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Gcc7 warns of possible string print overflow, on i586,
when printing password length (via a macro), generating
errors like:
[ 59s] ^~~~~~~~~~~~~~~~~~~~
[ 59s] In file included from /usr/include/stdio.h:862:0,
[ 59s] from idbm.h:27,
[ 59s] from context.h:22,
[ 59s] from idbm.c:59:
[ 59s] /usr/include/bits/stdio2.h:64:10: note:
'__builtin___snprintf_chk' output between 2 and 11 bytes into a
destination of size 8
[ 59s] return __builtin___snprintf_chk (__s, __n,
__USE_FORTIFY_LEVEL - 1,
[ 59s] ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~
[ 59s] __bos (__s), __fmt, __va_arg_pack ());
[ 59s] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[ 59s] cc1: all warnings being treated as errors
[ 59s] make[1]: *** [<builtin>: idbm.o] Error 1
[ 59s] make[1]: Leaving directory
The fix is to limit the size of the string printed, so that no
overflow is possible.
The print macros in usr/idbm.c were updated, as well, to match
the newer version in libopeniscsiusr/idbm.c, also to help the
i586 build.
|
| |
| |
| |
| |
| | |
Found by dualdm on github, who suggested a fix
but then went away, so this is submitted for him.
|
|\ \
| |/
| | |
iscsiuio: Do not flush tx queue on each uio interrupt.
|
| |
| |
| |
| |
| |
| |
| |
| | |
Unlike bnx2x, qedi start_xmit netlink provide us guarantee
of transmitting LL2 packet, so there is no need to call clear_tx_intr for
each LL2 packet. This help us in reducing iscsiuio lock contention.
Signed-off-by: Manish Rangankar <manish.rangankar@cavium.com>
|
| |
| |
| |
| |
| | |
Updated Changelog with "git shortlog 2.0.876..HEAD", and updated
version number in two include files.
|