Brad Mouring [Tue, 3 May 2022 03:38:24 +0000 (22:38 -0500)]
completions: Collapse declaration and attribute assignment
Previously, the associative arrays for the vendor/subcommands and
vendor/functions listings were split into a declaration of type
(using "typeset") and attribute definition (using "readonly"). On
bash 5.1.16 (at least), this lead to the following error (reported
after enabling -xv to expand and print shell inputs)
...
+ . /usr/share/bash-completion/completions/nvme
# bash tab completion for the nvme command line utility
# (unfortunately, bash won't let me add descriptions to cmds)
# Kelly Kaoudis kelly.n.kaoudis at intel.com, Aug. 2015
# Constant to indicate command has no options
readonly NO_OPTS=""
++ readonly NO_OPTS=
++ NO_OPTS=
# Associative array of plugins and associated subcommands
# Order here is same as PLUGIN_OBJS in Makefile
typeset -A _plugin_subcmds
++ typeset -A _plugin_subcmds
readonly _plugin_subcmds=(
[intel]="id-ctrl internal-log lat-stats \
...
++ _plugin_subcmds=(['intel']='id-ctrl internal-log lat-stats...
bash: 'intel': syntax error: operand expected (error token is "'intel'")
...
Using the available flags for "typeset" to declare the variables as
readonly arrays resolved the issue (and allows for bash completion
to work as-expected)
Hannes Reinecke [Fri, 22 Apr 2022 11:10:01 +0000 (13:10 +0200)]
plugins/seagate: fixup byte calculation
When calculating the number of bytes from the number of blocks one
has to ensure that the values can't overflow, so use a large data
type for the number of bytes than for the number of blocks.
Hannes Reinecke [Fri, 22 Apr 2022 11:04:37 +0000 (13:04 +0200)]
plugins/wdc: coverity fixes
Coverity found an off-by-one error when printing raw values; looks
like someone forgot convert the 1-based NVMe spec convention into
the 0-based C convention.
And, of course, various resource leaks.
Hannes Reinecke [Fri, 22 Apr 2022 10:31:40 +0000 (12:31 +0200)]
plugins: close filedescriptors
For some reasons the plugin authors don't seem to believe in having
to close filedecriptors. Coverity, OTOH, does.
So fix them up to close filedescriptors after use.
Steven Seungcheol Lee [Tue, 12 Apr 2022 05:45:46 +0000 (14:45 +0900)]
nvme: Fix error when passthru does not match opcode based flags
when write option is set(but opcode is not), file should open with read flag
same for read
it occur 'Bad file descriptor' error when access the file
write & read flags are set by opcode as well
Spec mentioned below about opcode(01:00 Data Transfer)
All options to the command shall transfer data as specified or transfer no data.
All commands, including vendor specific commands
01b = host to controller; 10b = controller to host
Reported-by: Kyungsik Shin <ks1012.shin@samsung.com> Signed-off-by: Steven Seungcheol Lee <sc108.lee@samsung.com>
Daniel Wagner [Wed, 13 Apr 2022 19:31:02 +0000 (21:31 +0200)]
nvme: output also libnvme version
Print the used libnvme version which the nvme-cli fronent is using.
In order to be backward compatible don't expect that the newly
introduce function in libnvme to be 1available.
This is done by providing a weak symbol which is overwritten when
statically build or when libnvme is is used as shared libray the
dlsym() call checks if the function is available.
Daniel Wagner [Tue, 12 Apr 2022 11:52:59 +0000 (13:52 +0200)]
nvme: Add git ref to the binary
In order to be able to figure out which binary is in use (for example
in debugging situation) it's really helpful to have the 'git describe'
ref added to the binary.
$ .build/nvme --version
nvme version 2.0 (git 2.0-2-gcbd3f8d+)
$ .build/nvme wdc --version
nvme wdc version 1.16.4 (git 2.0-2-gcbd3f8d+)
Martin George [Wed, 13 Apr 2022 15:16:35 +0000 (20:46 +0530)]
fabrics: skip connect if the transport types don't match
Discovery log page data may include records belonging to different
transport types. If during a nvme connect-all, a connect is attempted
on a record that doesn't match the transport type passed here, it
would end up in a connect failure for that record. For e.g. one would
see the following error if a connect is attempted on a tcp record,
but the transport type passed here is 'fc' and its associated params:
Martin George [Mon, 11 Apr 2022 04:06:20 +0000 (09:36 +0530)]
fabrics: avoid segfault when nvme discover fails with the -p option
If the initial discovery controller creation fails while passing
the -p option to the nvme discover (or connect-all) command, it
could lead to the below segfault:
nvme discover -t rdma -w 192.168.1.114 -a 192.168.1.113 -p
Failed to write to /dev/nvme-fabrics: Connection reset by peer
Segmentation fault (core dumped)
Avoid this segfault by checking if the controller created in
__create_discover_ctrl() is valid before proceeding.
Martin George [Mon, 11 Apr 2022 12:03:17 +0000 (17:33 +0530)]
fabrics: error message for nvme discover/connect-all with no params
Currently when nvme discover or connect-all is called without
passing any params, it simply blanks out without any meaningful
error. So add an appropriate error message for such scenarios,
similar to what's done in the erstwhile nvme-cli-monolithic branch.
Daniel Wagner [Fri, 8 Apr 2022 13:23:41 +0000 (15:23 +0200)]
fabrics: Refactor argument passing for transport configuration
We passing all the attributes of the transport configuration
individually. Group them together to reduce the number of arguments
one has to pass into the functions.
Daniel Wagner [Fri, 8 Apr 2022 12:08:47 +0000 (14:08 +0200)]
fabrics: Set KATO for discovery controller when connecting
In case we are connecting to the discovered controllers we should also
set the KATO value for persistent discocovery controllers correctly.
Unfortunatly, the compiler seems to get confused by only conditionally
setting tmo and warns about uninitialized variable. Hence we always
store and restore keep_alive_tmo.
Daniel Wagner [Fri, 8 Apr 2022 11:58:01 +0000 (13:58 +0200)]
fabrics: Do no modify default config for discovery controller
When a discovery controller is setup we are currently modifying the
default config. As this configuration might also be used for normal
I/O controllers don't modify it permanently.
Hannes Reinecke [Wed, 6 Apr 2022 13:24:20 +0000 (15:24 +0200)]
nvme: update to nvme_scan_filter_t modifications
The scan filter in libnvme has been updated, so update the code
to make use of it.
Signed-off-by: Hannes Reinecke <hare@suse.de>
[dwagner: list also the paths for non multipath namespaces] Signed-off-by: Daniel Wagner <dwagner@suse.de>
Arthur Shau [Thu, 3 Mar 2022 22:08:15 +0000 (14:08 -0800)]
Add new OCP plugin
New OCP plugin that consolidates some of the functionality from the other vendor plugins. For now, implements getting of C0 and C3 log pages.
Daniel Wagner [Mon, 4 Apr 2022 15:46:11 +0000 (17:46 +0200)]
fabrics: Set default trsvcid ports for TCP and RDMA
libnvme doesn't know at controller creation if the new controller is
going to be discovery controller. But nvme-cli knows it, so just set
defaults on the caller level.
Daniel Wagner [Mon, 4 Apr 2022 10:48:55 +0000 (12:48 +0200)]
fabrics: Do not free static string
38f5a54a7306 ("fabrics: Support connect even when no /etc/nvme/hostnqn
file exists") introduce a regression. arg_parser() might assign a
static string to hostnqn and/or hostid. Can't free this.
projects / users / sagi / nvme-cli.git / log