mirror of
https://github.com/SCST-project/scst.git
synced 2026-05-19 11:41:26 +00:00
svn+ssh://vlnb@svn.code.sf.net/p/scst/svn/trunk ........ r5955 | bvassche | 2014-12-22 05:10:41 -0800 (Mon, 22 Dec 2014) | 1 line Update for kernel 3.18 ........ r5959 | bvassche | 2015-01-06 05:25:28 -0800 (Tue, 06 Jan 2015) | 1 line scst_calc_block_shift: Log block shift and sector size upon mismatch ........ r5960 | bvassche | 2015-01-07 01:20:06 -0800 (Wed, 07 Jan 2015) | 4 lines scst_local: Fix unique per session sas address Signed-off-by: Sebastian Herbszt <herbszt@gmx.de> ........ r5961 | bvassche | 2015-01-09 04:23:25 -0800 (Fri, 09 Jan 2015) | 4 lines scst_sysfs: return EINVAL on too big LUN Signed-off-by: Sebastian Herbszt <herbszt@gmx.de> ........ r5965 | bvassche | 2015-01-13 00:55:46 -0800 (Tue, 13 Jan 2015) | 68 lines qla2x00t: Copy entire SCST sense buffer to q2x ctio There seems to be a bug in passing sense information to QLA HBAs, where the last 2 bytes of the sense data (ASC, ASCQ) are not copied to the low level sense buffer. We encountered this in ESX, which relies on these 2 bytes to parse the MISCOMPARE sense code (0xE1, 0x1D, 0x00). Bellow is a simple test to recreate this issue, but during vMotion operations (where VMs are moved from one host to another), this may cause the operation to fail leaving the VM in an inconsistent state. The test I ran to verify that we are indeed missing the bytes is the following: 1. Create a SCST based device 2. Expose the device to 2 ESX hosts 3. Format the device as VMFS5, create a test directory 4. From both hosts, I start writing to this directory (no VMs involved, just write normal files) At this stage, both ESX hosts try to take access to the directory. The VMFS filesystem contains a per-directory lock which is managed by COMPARE AND WRITE command. Each ESX will attempt to change the VMFS lock location from unlocked to locked to create the new file. Obviously there are bound to be failures (which are equivalent to programming locking conflicts), these are reported by the MISCOMPARE sense code. Upon these MISCOMPARE errors, the host will re-try taking the lock until it succeeds, and will then proceed to perform the write operation on the directory. Due to the bug in copying the sense buffer from the SCST core to the QLA ctio, instead of the full sense code, only the key (0xE) is sent, and ESX does not know how to handle it resulting in IO error. Here are the errors as they appear on the command line: /vmfs/volumes/54a297c4-ca5af1cc-7f94-002219d20f28/ats_test # ./open_close_test-esx2.sh ./open_close_test-esx2.sh: line 8: can't create ats_fileoptest-esx2_1.txt: Input/output error ./open_close_test-esx2.sh: line 8: can't create ats_fileoptest-esx2_21.txt: Input/output error ./open_close_test-esx2.sh: line 8: can't create ats_fileoptest-esx2_110.txt: Input/output error ./open_close_test-esx2.sh: line 8: can't create ats_fileoptest-esx2_111.txt: Input/output error In the /var/log/vmkernel.log, we can see that the sense information is missing (0xE, 0x0, 0x0) instead of (0xE, 0x1D, 0x0). 2014-12-30T12:13:20.714Z cpu6:33519)ScsiDeviceIO: 2338: Cmd(0x412e84f957c0) 0x89, CmdSN 0x234d from world 519051 to dev "eui.0024f400d5020007" failed H:0x0 D:0x2 P:0x0 Valid sense data: 0xe 0x0 0x0. 2014-12-30T12:13:20.766Z cpu6:33519)ScsiDeviceIO: 2338: Cmd(0x412e84f91d00) 0x89, CmdSN 0x2350 from world 519051 to dev "eui.0024f400d5020007" failed H:0x0 D:0x2 P:0x0 Valid sense data: 0xe 0x0 0x0. 2014-12-30T12:13:20.766Z cpu6:33519)ScsiDeviceIO: 2338: Cmd(0x412e80449fc0) 0x89, CmdSN 0x234f from world 519051 to dev "eui.0024f400d5020007" failed H:0x0 D:0x2 P:0x0 Valid sense data: 0xe 0x0 0x0. This patch fixes this issue, the test will run without a problem with the fix (no IO errors, all the files are properly written to the directory). Signed-off-by: Shahar Salzman <shahar.salzman@kaminario.com> Reviewed-by: Eran Mann <eran.mann@kaminario.com> [bvanassche: simplified implementation] Signed-off-by: Bart Van Assche <bvanassche@acm.org> ........ git-svn-id: http://svn.code.sf.net/p/scst/svn/branches/3.0.x@6110 d57e44dd-8a1f-0410-8b47-8ef2f437770f
SCST Local ...
Richard Sharpe, 30-Nov-2008
This is the SCST Local driver. Its function is to allow you to access devices
that are exported via SCST directly on the same Linux system that they are
exported from.
No assumptions are made in the code about the device types on the target, so
any device handlers that you load in SCST should be visible, including tapes
and so forth.
You can freely use any sg, sd, st, etc. devices imported from target,
except the following: you can't mount file systems or put swap on them
for all dev handlers, except BLOCKIO and pass-through, because it can
lead to recursive memory allocation deadlock. This is a limitation of
Linux memory/cache manager. See SCST README file for details. For
BLOCKIO and pass-through dev handlers there's no such limitation, so you
can freely mount file systems over them.
To build, simply issue 'make' in the scst_local directory.
Try 'modinfo scst_local' for a listing of module parameters so far.
Here is how I have used it so far:
1. Load up scst:
modprobe scst
modprobe scst_vdisk
2. Create a virtual disk (or your own device handler):
dd if=/dev/zero of=/some/path/vdisk1.img bs=16384 count=1000000
echo "add_device vm_disk1 filename=/some/path/vdisk1.img" >/sys/kernel/scst_tgt/handlers/vdisk_fileio/mgmt
3. Load the scst_local driver:
insmod scst_local
echo "add vm_disk1 0" >/sys/kernel/scst_tgt/targets/scst_local/scst_local_tgt/luns/mgmt
4. Check what you have
cat /proc/scsi/scsi
Attached devices:
Host: scsi0 Channel: 00 Id: 00 Lun: 00
Vendor: ATA Model: ST9320320AS Rev: 0303
Type: Direct-Access ANSI SCSI revision: 05
Host: scsi4 Channel: 00 Id: 00 Lun: 00
Vendor: TSSTcorp Model: CD/DVDW TS-L632D Rev: TO04
Type: CD-ROM ANSI SCSI revision: 05
Host: scsi7 Channel: 00 Id: 00 Lun: 00
Vendor: SCST_FIO Model: vm_disk1 Rev: 200
Type: Direct-Access ANSI SCSI revision: 04
Or instead of manually "add_device" in (2) and step (3) write a
scstadmin config:
HANDLER vdisk_fileio {
DEVICE vm_disk1 {
filename /some/path/vdisk1.img
}
}
TARGET_DRIVER scst_local {
TARGET scst_local_tgt {
LUN 0 vm_disk1
}
}
then:
insmod scst_local
scstadmin -config conf_file.cfg
More advanced examples:
For (3) you can:
insmod scst_local add_default_tgt=0
echo "add_target scst_local_tgt session_name=scst_local_host" >/sys/kernel/scst_tgt/targets/scst_local//mgmt
echo "add vm_disk1 0" >/sys/kernel/scst_tgt/targets/scst_local/scst_local_tgt/luns/mgmt
Scst_local module's parameter add_default_tgt disables creation of
default target "scst_local_tgt" and session "scst_local_host", so you
needed to create it manually.
There can be any number of targets and sessions created. Each SCST
session corresponds to SCSI host. You can change which LUNs assigned to
each session by using SCST access control. This mode is intended for
user space target drivers (see below).
Alternatively, you can write an scstadmin's config file conf_file.cfg:
HANDLER vdisk_fileio {
DEVICE vm_disk1 {
filename /some/path/vdisk1.img
}
}
TARGET_DRIVER scst_local {
TARGET scst_local_tgt {
session_name scst_local_host
LUN 0 vm_disk1
}
}
then:
insmod scst_local add_default_tgt=0
scstadmin -config conf_file.cfg
NOTE! Although scstadmin allows to create scst_local's sessions using
"session_name" expression, it doesn't save existing sessions during
writing config file by "write_config" command. If you need this
functionality, feel free to send a request for it in SCST development
mailing list.
5. Have fun.
Some of this was coded while in Santa Clara, some in Bangalore, and some in
Hyderabad. Noe doubt some will be coded on the way back to Santa Clara.
The code still has bugs, so if you encounter any, email me the fixes at:
realrichardsharpe@gmail.com
I am thinking of renaming this to something more interesting.
Sysfs interface
===============
See SCST's README for a common SCST sysfs description.
Root of this driver is /sys/kernel/scst_tgt/targets/scst_local. It has
the following additional entry:
- stats - read-only attribute with some statistical information.
Each target subdirectory contains the following additional entries:
- phys_transport_version - contains and allows to change physical
transport version descriptor. It determines by which physical
interface this target will look like. See SPC for more details. By
default, it is not defined (0).
- scsi_transport_version - contains and allows to change SCSI
transport version descriptor. It determines by which SCSI
transport this target will look like. See SPC for more details. By
default, it is SAS.
Each session subdirectory contains the following additional entries:
- transport_id - contains this host's TransportID. This TransportID
used to identify initiator in Persisten Reservation commands. If you
change scsi_transport_version for a target, make sure you set for all
its sessions correct TransportID. See SPC for more details.
- host - links to the corresponding SCSI host. Using it you can find
local sg/bsg/sd/etc. devices of this session. For instance, this
links points out to host12, so you can find your sg devices by:
$ lsscsi -g|grep "\[12:"
[12:0:0:0] disk SCST_FIO rd1 200 /dev/sdc /dev/sg2
[12:0:0:1] disk SCST_FIO nullio 200 /dev/sdd /dev/sg3
They are /dev/sg2 and /dev/sg3.
The following management commands available via /sys/kernel/scst_tgt/targets/scst_local/mgmt:
- add_target target_name [session_name=sess_name; [session_name=sess_name1;] [...]] -
creates a target with optionally one or more sessions.
- del_target target_name - deletes a target.
- add_session target_name session_name - adds to target target_name
session (SCSI host) with name session_name.
- del_session target_name session_name - deletes session session_name
from target target_name.
Note on performance
===================
Although this driver implemented in the most performance effective way,
including zero-copy passing data between SCSI/block subsystems and SCST,
in many cases it is NOT suited to measure performance as a NULL link.
For example, it is not suited for max IOPS measurements. This is because
for such cases not performance of the link between the target and
initiator is the bottleneck, but CPU or memory speed on the target or
initiator. For scst_local you have both initiator and target on the same
system, which means each your initiator and target are much less
CPU/memory powerful.
User space target drivers
=========================
Scst_local can be used to write full featured SCST target drivers in
user space:
1. For each SCSI target a user space target driver should create an
scst_local's target using "add_target" command.
2. Then the user space target driver should, if needed, set its SCSI and
physical transport version descriptors using attributes
scsi_transport_version and phys_transport_version correspondingly in
/sys/kernel/scst_tgt/targets/scst_local/target_name directory.
3. For incoming session (I_T nexus) from an initiator the user space
target driver should create scst_local's session using "add_session"
command.
4. Then, if needed, the user space target driver should set TransportID
for this session (I_T nexus) using attribute
/sys/kernel/scst_tgt/targets/scst_local/target_name/sessions/session_name/transport_id
5. Then the user space target driver should find out sg/bsg devices for
the LUNs the created session has using link
/sys/kernel/scst_tgt/targets/scst_local/target_name/sessions/session_name/host
as described above.
6. Then the user space target driver can start serving the initiator using
found sg/bsg devices.
For other connected initiators steps 3-6 should be repeated.
Compilation options
===================
There are the following compilation options, that could be commented
in/out in Makefile:
- CONFIG_SCST_LOCAL_DIRECT_PROCESSING - by default, when this option
is not defined, scst_local reschedules all commands for processing in
one of the SCST threads. If this option is defined, scst_local tries
to not do it, if possible (sometimes queuecommand() called under
various locks held), but instead process them in the submitter's
context. This is to increase performance, but as on 2.6.37 and below
Linux block layer doesn't work with such kind of reentrance, hence
this option disabled by default. Note! At the moment in
scst_estimate_context*() returning DIRECT contexts disabled, so this
option doesn't have any real effect.
Change log
==========
V0.1 24-Sep-2008 (Hyderabad) Initial coding, pretty chatty and messy,
but worked.
V0.2 25-Sep-2008 (Hong Kong) Cleaned up the code a lot, reduced the log
chatter, fixed a bug where multiple LUNs did not
work. Also, added logging control. Tested with
five virtual disks. They all came up as /dev/sdb
through /dev/sdf and I could dd to them. Also
fixed a bug preventing multiple adapters.
V0.3 26-Sep-2008 (Santa Clara) Added back a copyright plus cleaned up some
unused functions and structures.
V0.4 5-Oct-2008 (Santa Clara) Changed name to scst_local as suggested, cleaned
up some unused variables (made them used) and
change allocation to a kmem_cache pool.
V0.5 5-Oct-2008 (Santa Clara) Added mgmt commands to handle dev reset and
aborts. Not sure if aborts works. Also corrected
the version info and renamed readme to README.
V0.6 7-Oct-2008 (Santa Clara) Removed some redundant code and made some
changes suggested by Vladislav.
V0.7 11-Oct-2008 (Santa Clara) Moved into the scst tree. Cleaned up some
unused functions, used TRACE macros etc.
V0.9 30-Nov-2008 (Mtn View) Cleaned up an additional problem with symbols not
being defined in older version of the kernel. Also
fixed some English and cleaned up this doc.
V1.0 10-Sep-2010 (Moscow) Sysfs management added. Reviewed and cleaned up.
V2.1 Update for kernels up to 3.0. Cleanups.