Gwendal Grignou [Thu, 13 Oct 2016 21:25:33 +0000 (14:25 -0700)]
Fix compilation errors:
While compiling nvme-cli for ChromeOS, the compiler founds minor issues
in the code:
- unsigned and sizet_t can never be negative.
- use fabs instead of llabs when argument is a double.
Christoph Hellwig [Wed, 12 Oct 2016 08:45:51 +0000 (10:45 +0200)]
fine tune the nvme-discover manpage
- remove the device argument, which as far as I can tell does not
exist.
- mention the /etc/nvme/hostnqn file, the default host nqn and explain
host nqns a bit more
- mention the loopback transport
- be a bit more specific about IP addressing as a concept separate from
the RDMA transport
- improve the introduction a bit.
Jay Freyensee [Fri, 7 Oct 2016 19:09:20 +0000 (12:09 -0700)]
nvme-cli: tweak discover man pages w/argconfig fix
With Sagi's fix to argconfig.c, the nvme-discover man
pages need to be tweaked to show the true intent of
using discovery.conf, a default with multiple discover
command-lines.
Signed-off-by: Jay Freyensee <james_p_freyensee@linux.intel.com>
Jay Freyensee [Fri, 7 Oct 2016 18:46:16 +0000 (11:46 -0700)]
nvme-cli: have cfg var get set more than 1 execution
Without this fix, the struct config cfg variable only
gets set one time when argconfig_parse() is called. Any other
time it is attempted to be set, it acts as a read-only value
and continues to use the initial value it was set with.
This poses a problem for the discovery feature, discovery.conf file
as it is desired to make the file multiple line instead of a single
line, such as:
-t rdma -a 192.168.69.33 -s 4420 -q host2-proper-nqn
-t rdma -a 192.168.1.4 -s 4420 -q host2-proper-nqn
-t rdma -a 192.168.69.33 -s 99 -q host2-proper-nqn
I've tried the fix with a few existing commands and nothing
appears to be broken, but I haven't done an exhaustive test.
Reported-by: Jay Freyensee <james_p_freyensee@linux.intel.com> Originally-by: Sagi Grimberg <sagi@lightbits.io> Tested-by: Jay Freyensee <james_p_freyensee@linux.intel.com>
Modified: print the subnqn (NVM Subsystem NVMe Qualified Name) field for both normal and jason formats upon the id_ctrl() command. Also, sn, mn, and fr fields are printed as the jason format.
Jay Freyensee [Fri, 23 Sep 2016 16:49:50 +0000 (09:49 -0700)]
nvme-cli: change 'r' to 'c' for connect flag
To make things a tad easier for human usage understanding,
refine the single letter short option of 'r' to 'c' for the
'connect' command as 'r' is already used by 'discover'.
Signed-off-by: Jay Freyensee <james_p_freyensee@linux.intel.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Keith Busch [Thu, 22 Sep 2016 16:11:25 +0000 (12:11 -0400)]
nvme-format: Add optional reset parameter
Reset after format shouldn't be necessary, but some controllers require
it. This patch adds a parameter a user can set to trigger an automatic
controller reset after a successful format.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Scott Bauer [Thu, 22 Sep 2016 15:50:54 +0000 (09:50 -0600)]
Fix stack-based overflow in id-ctrl json output.
Fixes:
==15224==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7fffffffcf55 at pc 0x7ffff6ecb9f5 bp 0x7fffffffcd80 sp 0x7fffffffc510
WRITE of size 22 at 0x7fffffffcf55 thread T0
#0 0x7ffff6ecb9f4 in __interceptor_vsprintf (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x619f4)
#1 0x7ffff6ecbcc9 in __interceptor_sprintf (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x61cc9)
#2 0x420357 in json_nvme_id_ctrl /home/sbauer/nvme_code/nvme-cli/nvme-print.c:1266
#3 0x4072f1 in __id_ctrl /home/sbauer/nvme_code/nvme-cli/nvme.c:935
#4 0x40742f in id_ctrl /home/sbauer/nvme_code/nvme-cli/nvme.c:950
#5 0x42fa63 in handle_plugin /home/sbauer/nvme_code/nvme-cli/plugin.c:140
#6 0x410c46 in main /home/sbauer/nvme_code/nvme-cli/nvme.c:2729
#7 0x7ffff6ac182f in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x2082f)
#8 0x402268 in _start (/home/sbauer/nvme_code/nvme-cli/nvme+0x402268)
Address 0x7fffffffcf55 is located in stack of thread T0 at offset 117 in frame
#0 0x42011f in json_nvme_id_ctrl /home/sbauer/nvme_code/nvme-cli/nvme-print.c:1254
This frame has 3 object(s):
[32, 41) 'fr'
[96, 117) 'sn' <== Memory access at offset 117 overflows this variable
[160, 201) 'mn'
Signed-off-by: Scott Bauer <scott.bauer@intel.com>
Scott Bauer [Mon, 19 Sep 2016 15:47:15 +0000 (09:47 -0600)]
json: Print to stderr and abort() instead of returning NULL and faulting.
No one who uses the JSON API is checking for return values.
Instead of littering the json code with if-else chains,
we will error on any memory allocation failure. Even though
incredibly unlikely malloc will return NULL it's easier to
check for it and print an error instead of faulting.
Signed-off-by: Scott Bauer <scott.bauer@intel.com>
fix inappropriate size of resv10 [] array field in the nvme_reservation_status based NVMe 1.21 spec.
Add cdw10 field to convey Number of Dwords (NUMD) for the Reservation Report command
Keith Busch [Wed, 31 Aug 2016 17:15:54 +0000 (11:15 -0600)]
Release version v0.9
New extensions, fabrics support, a bug fixes. The 'list' command
is liberated from its udev dependency, and no longer requires
CONFIG_IO_STRICT_DEVMEM be disabled.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Keith Busch [Tue, 30 Aug 2016 20:33:54 +0000 (14:33 -0600)]
Simplify list command
Use scandir with an appropriate filter to find all our nvme namespaces
and handle allocating all the dirent's. Makes the implementation a
little simpler.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Keith Busch [Tue, 30 Aug 2016 14:49:19 +0000 (08:49 -0600)]
Remove version from list
We can't reliably map controller registers due to kernel memory
protections added in kernel commit 90a545e98, so we are removing its
dependency for the 'list' command to run.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Roy Shterman [Wed, 24 Aug 2016 06:53:06 +0000 (09:53 +0300)]
nvme-cli/fabrics: Added hostnqn parameter to connect command
Added user-defined 'hostnqn' to 'connect' command
so users could use the allowed_hosts attribute on the target and still
connect to target with nvme-cli tool.
Scott Bauer [Mon, 22 Aug 2016 23:19:24 +0000 (17:19 -0600)]
nvme-cli: Implement list command that doesn't rely on libudev
The previous list command required libudev. We can accomplish the same
thing with standard functions. Libudev is still required for some code
in fabrics.c, but the depencency has been removed from nvme.c.
Signed-off-by: Scott Bauer <scott.bauer@intel.com>
Sagi Grimberg [Tue, 16 Aug 2016 09:46:26 +0000 (12:46 +0300)]
fabrics: Allow discover params to come from a conf file
Allow the user to just run "nvme discover" or "nvme connect-all"
in case it finds a default /etc/nvme/nvmf_disc conf file.
We allow multiple discovery addresses by iterating over the
lines of the file and executing a discover (with or without
connect) for each line. We allow newlines and '#' prefixed comments.
The return value is or'ed on all discover attempts.
In order to minimize some parsing code, I just convert the
file line into an (argc, argv) pair and feed it to argconfig_parse()
which dictates that the file lines are identical to what one would
pass nvme discover <params>. I'm open to better ideas.
Jay Freyensee [Thu, 28 Jul 2016 19:23:59 +0000 (12:23 -0700)]
nvme-cli: user-defined hostnqn option for discover
The nvme-cli will always use the default hostnqn
in /dev/nvme-fabrics for the discovery query, even though
both the NVMe Target and NVMe Host rdma implementations allow
user-defined hostnqn naming.
For example, this is the current, somewhat broken behavior if you
used your own hostnqn provision naming on the NVMe kernel target:
Keith Busch [Mon, 1 Aug 2016 21:25:38 +0000 (15:25 -0600)]
Use different error code when command not found
A failed system call returns -1 with errno set, so lets differentiate
a failed system call from not finding a command at all. This suppresses
large help messages on any generic system call failure.
Signed-off-by: Keith Busch <keith.busch@intel.com>
projects / users / hch / nvme-cli.git / log