matiasbjorling / zdm-device-mapper Goto Github PK

Device Mapper for Zoned based devices: ZDM for short. This project aims to present a traditional block device for Host Aware and Host Managed zoned block device drives. ZDM presents to the drive a workload that conforms to the Host Aware intended use or Host Managed constraints.

ZDM implements a translation layer between a the tradition block interface and a zoned block device (ZBD) supporting the ZBC or ZAC command set. ZDM utilizes the available conventional space on a ZBD drive for storing the translation layer mapping data. The device configuration should include a minimum of 0.1% of the drive capacity as conventional space, however 0.2% is currently preferred and most tested. The current design assumes the conventional space is located at the low LBA range of the partition or the drive. The preferred layout is a typical GPT partitioned drive with a large single partition [ex: parted -s /dev/sdX mkpart 1MiB -- -1].

It is recommended that ZDM be configured to allocate 1%, the default, of the drive capacity as over provisioning used to manage the garbage collection of stale blocks.

ZDM supports the BLKDISCARD IOCTL issued by blkdiscard(). It is recommended that filesystems are mounted with the -o discard option.

The initial implementation focuses on drives with same sized zones of 256 MiB. In future the support for other zones sizes will be added. Internally all addressing is on 4-KiB (4k) boundaries. Currently a 4k PAGE_SIZE is assumed. Architectures with 8k (or other) PAGE_SIZE values have not been tested and are likely broken at the moment.

Drives of the more restrictive Host Managed device type are expected to work provided the required amount of conventional space is available.

• Current Linux Kernel v4.4 to v.4.10 with ZDM patches
• Recommended: Util Linux with ZDM patches
• Recommended: sg3utils (1.41 or later)

• ZDM software is a work in progress, subject to change without notice and is provided without warranty for any particular intended purpose beyond testing and reference. The ZDM may become unstable, resulting in system crash or hang, including the possibility of data loss. No responsibility for lost data or system damage is assumed by the creators or distributors of ZDM software. Users explicitly assume all risk associated with running the ZDM software.

• All zones are 256 MiB.
• 4k page or block size.
• Host Aware zoned block device, possibly with conventional zones.
• Host Managed zoned block device with conventional zones.
• Currently 256 MiB of RAM per drive is recommended.
• SAS controller SAT supports ZBC <-> ZAC translation.

• zdm-tools: zdmadm, zdm-status, zdm-zones, zdmon and others ...
• util-linux: blkid, blkreport and blkzonecmd

• Reset all WPs on drive:

      sg_reset_wp --all /dev/sdX

or

      blkzonecmd --reset --all /dev/sdX

or

      zdm-zonecmd --reset --all --sat /dev/sdX

• Partition the drive to start the partition at a WP boundary.

      parted -s /dev/sdX mklabel gpt
      parted -s mkpart primary 256MiB -- -1

• Place ZDM drive mapper on /dev/sdX

      zdmadm -F -c /dev/sdX1

• Format:

      mkfs -t ext4 -E discard /dev/mapper/zdm_sdX1

or

      mkfs -t ext4 -b 4096 -g 32768 -G 32 \
        -E offset=0,num_backup_sb=0,packed_meta_blocks=1,discard \
        -O flex_bg,extent,sparse_super2 /dev/mapper/zdm_sdX1

• Mounting the filesystem.

      mount -o discard /dev/mapper/zdm_sdX1 /mnt/zdm_sdX1

Building:

• Normal kernel build with CONFIG_DM_ZDM

ZAC/ZBC standards are still being developed. The Zoned Device ATA command Set v1 is finalized, and v2 is under development. Changes to the command set and command interface can be expected, but are not expected to be included in subsequent ZDM work.

ZDM is distributed under the terms of GPL v2 or any later version.

ZDM and all and all utilities here in are distributed "as is," without technical support, and WITHOUT ANY WARRANTY, without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. Along with ZDM, you should have received a copy of the GNU General Public License. If not, please see http://www.gnu.org/licenses/.

• Shaun Tancheff (author)
• Adrian Palmer [email protected]

• CEPH: Enable ceph to created and used an OSD that used ZDM. NOTE: depends on blkid support from util-linux or util-linux-ng.
  • 0.94.5 ceph patch
• util-linux -- Added: blkreport, blkzonecmd
  • 2.27.1 For Ubuntu 16.04
• util-linux-ng -- Missing: blkreport, blkzonecmd
  • 2.17.2 For CentOS 6.7

• Patches
  • v4.8 ZDM r127 patches for linux v4.8

• Notice for MD-RAID level 4/5/6. The bio queue feature is off by default.
  • Workaround: Add -q 256 to zdmadm command-line when creating and restoring ZDM instances. Can be changed at run-time.
  • Data corruption has been noted with GC. Use --gc off to disable automatic GC activity. Can be changed at run-time.
  • Data corruption has been noted with metadata journaling. Use --journal-age 0 to disable metadata journaling. Can be changed at run-time.

• ZDM #91

  • Kernel from v4.2 tag is pre-patched.
  • Added version number to track alpha/beta releases.
  • Numerous bug fixes and better stability.
  • Much improved GC / zone compaction handling.
  • Patch set is cleaner and better defined.
  • Patches provided to apply cleanly against 4.1 (stable), 4.2, and 4.3.

• ZDM #96

  • Support for ZDM metadata on conventional space.
  • Modified queue limits to not confuse sgdisk.
  • Initial ceph support for ZDM OSD (ceph-disk prepare --zdm).
  • Block layer fixes for ZAC commands.
  • Move ZDM metadata initialization to zdmadm userspace.
  • Kernel from v4.2 tag is pre-patched.
  • More code cleanups: Endian (__le/__be), style and formatting, started adding inline documentation.
  • Numerous bug fixes and better stability.
  • More GC / zone compaction improvements.
  • Patches provided to apply cleanly against 4.1 (stable), 4.2, and 4.3.

• ZDM #97

  • Fix block ata zoned_command (Reset WP). Using wrong psuedo constant.
  • Fix a deadlock when gc_immediate is called from multiple threads.

• ZDM #101

  • Added Stream Id to manage co-location of data with similar expected lifetimes.
  • Remove the fixed address segmentation (Megazone) scheme.
  • Added support for large conventional space in low LBAs and utilize the space for translation tables.

• ZDM #103

  • Added read-ahead for LT entries.
  • Removed big io_mutex
  • Added discard extent cache
  • Enable discard support for md-raid to be enabled by ZDM block devices.
  • Added an experimental prototype hack to use PID as stream id.
  • Changed userspace ioctl to use procfs as ioctl are disabled in v4.4 and later kernels.

• ZDM #104

  • Transition to using streamid patches from Jens Axobe.
  • Add ZAC->ZBC report zones little to big endian translation.
  • Add flags to struct bio to support ZBC commands: Report zones, reset, open and close zone(s).
  • Reworked flush handling to ensure metadata is durable before releasing.

• ZDM #105

  • Bug Fixes
    • Memory leak during undrop.
    • Bio set not allocated.
    • GC queuing bugs .. wrap under on free zone count.
    • Reworked immediate/wait logic.
  • Update to v4.5 kernel.

• ZDM #106

  • Bug Fix: Change lazy-drop timeout back to 15s

• ZDM #107

  • Bug Fix: sync() NULL dereference bug fix

• ZDM #108

  • Bug Fix: Rework BIO flags, fix zone off-by-1 error when partition is not zone aligned.

• ZDM #109

  • Updated to v4.6-rc2
  • Re based on separate operations from flags in the bio/request structs https://lkml.kernel.org/r/[email protected]
  • Re based on libata: SATL update: https://lkml.kernel.org/r/[email protected]
  • Re based on libata: ZAC support: https://lkml.kernel.org/r/[email protected]
  • Re based on Update version of write stream ID patchset https://lkml.kernel.org/r/[email protected]
  • Forced GC now starts 2 zones earlier and will always attempt a GC even when only 1 block is known to be stale (due to the deferred discard the stale counting can be far out of date.
  • Reduced memory used by arrays of map page pointers to a sparse array using a hash map. 8TB should now use ~12 MiB during first fill and ~25MiB during GC.
  • Initial implementation of metadata journal to CoW metadata changes between sync / cache flush events.
  • Changed ZDM superblock crc64 to crc32 as crc64 was removed from util-linux.
  • util-linux updated and required (blkreport and blkzonecmd).
  • zdm-tools updated to used blkreport and blkzonecmd IOCTLs as well as crc32 instead of crc64.

• ZDM #110

  • Metadata journal write back is is expanded to include 2 journal passes before overwriting origin lba (every 3rd flush). This also includes a bug fix to force migration out of journal to free up journal space.
  • Added Hannes caching device mapper (dm-zoned) to being quantitative comparison with ZDM (dm-zdm).
  • Rename circus to accommodate both device mappers (dm-zoned -> dm-zdm then added dm-zoned).
  • Re-Sync zdmadm tools to match name changes and refreshed libzdm code base.

• ZDM #112

  • Fix memory corruption due to over writting allocated space for WB Journal tracking
  • Updated to rc6 and removed pending bio split patch series.
  • Updated SCSI_ZBC/BLK_DEV_ZONED features to support current HA drives.
  • Added support for Write Same via SCT for SATA drives.

• ZDM #114

  • Disabled WB journal path by default.
  • Expanded de-dupe to include any 4k aligned bio.
  • Updated to 4.6 kerenl.

• ZDM #119

  • Updated to 4.8 kerenl.
  • Numerous bug fixes and stability enhancements.

• ZDM #120

  • Bug fix: Dynamically expand extent cache.

• ZDM #127

  • Incorported ZBC feature patches planned for v4.10
  • Added dynamic configuration features such as:
    • cache age and size
    • GC on/off as well as GC profiles
    • metadata journaling and aging