Annotation of wikisrc/zfs.mdwn, revision 1.42

1.1       gdt         1: # ZFS on NetBSD
                      2: 
1.11      gdt         3: This page attempts to do two things: provide enough orientation and
1.13      wiz         4: pointers to standard ZFS documentation for NetBSD users who are new to
1.11      gdt         5: ZFS, and to describe NetBSD-specific ZFS information.  It is
1.1       gdt         6: emphatically not a tutorial or an introduction to ZFS.
                      7: 
                      8: Many things are marked with \todo because they need a better
1.28      gdt         9: explanation, and some have question marks
1.1       gdt        10: 
1.42    ! gdt        11: This HOWTO describes the most recent state of branches, and does not
        !            12: attempt to describe formal releases.  This is a clue; if you are using
        !            13: NetBSD 9 and ZFS, you should update along the branch.
        !            14: 
1.1       gdt        15: # Status of ZFS in NetBSD
                     16: 
1.27      gdt        17: ## NetBSD 8
1.23      gdt        18: 
1.27      gdt        19: NetBSD 8 has an old version of ZFS, and it is not recommended for use
                     20: at all.  There is no evidence that anyone is interested in helping
                     21: with ZFS on 8.  Those wishing to use ZFS on NetBSD 8 should therefore
                     22: update to NetBSD 9.
1.1       gdt        23: 
                     24: ## NetBSD 9
                     25: 
1.27      gdt        26: NetBSD-9 has ZFS that is considered to work well.  There have been
                     27: fixes since 9.0_RELEASE.  As always, people running NetBSD 9 are
                     28: likely best served by the most recent version of the netbsd-9 stable
1.42    ! gdt        29: branch.  As of 2021-03, ZFS in the NetBSD 9.1 release is very close to
        !            30: netbsd-9, except that the mkdir fix is newly in netbsd-9.
1.1       gdt        31: 
1.42    ! gdt        32: There was a crash with mkdir over NFS with maproot, resolved in March
        !            33: 2021 in 9 and current.  See http://gnats.netbsd.org/55042
1.41      gdt        34: 
1.42    ! gdt        35: There is a workaround where removing a file will commit the ZIL
        !            36: (normally this would not be done), to avoid crashes due to vnode
        !            37: reclaims.  \todo Link to PR.
1.41      gdt        38: 
                     39: There has been a report of an occasional panic somewhere in
                     40: zfs_putpages.
                     41: 
1.27      gdt        42: ## NetBSD-current
1.1       gdt        43: 
1.42    ! gdt        44: NetBSD-current (as of 2021-03) has similar ZFS code to 9.
1.5       gdt        45: 
1.42    ! gdt        46: There is initial support for [[ZFS root|wiki/RootOnZFS]], via booting
        !            47: from ffs and pivoting.
1.1       gdt        48: 
1.27      gdt        49: ## NetBSD/xen special issues
1.1       gdt        50: 
1.36      gdt        51: Summary: if you are using NetBSD, xen and zfs, use NetBSD-current.
                     52: 
1.27      gdt        53: In NetBSD-9, MAXPHYS is 64KB in most places, but because of xbd(4) it
                     54: is set to 32KB for XEN kernels.  Thus the standard zfs kernel modules
                     55: do not work under xen.  In NetBSD-current, xbd(4) supports 64 KB
1.36      gdt        56: MAXPHYS and this is no longer an issue.  Xen and zfs on current are
                     57: reported to work well together, as of 2021-02.
1.1       gdt        58: 
                     59: ## Architectures
                     60: 
                     61: Most people seem to be using amd64.
                     62: 
                     63: To build zfs, one puts MKZFS=yes in mk.conf.  This is default on amd64
                     64: and aarch64 on netbsd-9.  In current, it is also default on sparc64.
                     65: 
                     66: More or less, zfs can be enabled on an architecture when it is known
                     67: to build and run reliably.  (Of course, users are welcome to build it
                     68: and report.)
                     69: 
1.27      gdt        70: # Quick Start
                     71: 
                     72: See the [FreeBSD Quickstart
                     73: Guide](https://www.freebsd.org/doc/handbook/zfs-quickstart.html); only
                     74: the first item is NetBSD specific.
                     75: 
                     76:   - Put zfs=YES in rc.conf.
                     77: 
                     78:   - Create a pool as "zpool create pool1 /dev/dk0".
                     79: 
                     80:   - df and see /pool1
                     81: 
                     82:   - Create a filesystem mounted on /n0 as "zfs create -o
                     83:     mountpoint=/n0 pool1/n0".
                     84: 
1.36      gdt        85:   - Read the documentation referenced in the next section.
                     86: 
                     87: ## Documentation Pointers
                     88: 
                     89: See the man pages for zfs(8), zpool(8).  Also see zdb(8), if only for
                     90: seeing pool config info when run with no arguments.
                     91: 
                     92:   - [OpenZFS Documentation](https://openzfs.github.io/openzfs-docs/)
                     93:   - [OpenZFS admin docs index page](https://github.com/openzfs/zfs/wiki/Admin-Documentation)
                     94:   - [FreeBSD Handbook ZFS Chapter](https://www.freebsd.org/doc/handbook/zfs.html)
                     95:   - [Oracle ZFS Administration Manual](https://docs.oracle.com/cd/E26505_01/html/E37384/index.html)
                     96:   - [Wikipedia](https://en.wikipedia.org/wiki/ZFS)
1.27      gdt        97: 
1.1       gdt        98: # NetBSD-specific information
                     99: 
                    100: ## rc.conf
                    101: 
                    102: The main configuration is to put zfs=YES in rc.conf, so that the rc.d
1.13      wiz       103: scripts bring up ZFS and mount ZFS file systems.
1.1       gdt       104: 
1.14      gdt       105: ## pool locations
                    106: 
                    107: One can add disks or parts of disks into pools.  Methods of specifying
                    108: areas to be included include:
                    109: 
1.29      gdt       110:   - entire disks (e.g., /dev/wd0d on amd64, or /dev/wd0 which has the same major/minor)
1.14      gdt       111:   - disklabel partitions (e.g., /dev/sd0e)
                    112:   - wedges (e.g., /dev/dk0)
                    113: 
1.33      gdt       114: Information about created or imported pools is stored in
                    115: /etc/zfs/zpool.cache.
                    116: 
1.40      gdt       117: Conventional wisdom is that a pool that is more than 80% used gets
                    118: unhappy; so far there is not NetBSD-specific wisdom to confirm or
                    119: refute that.
                    120: 
1.37      gdt       121: ## pool native blocksize mismatch
1.36      gdt       122: 
                    123: ZFS attempts to find out the native blocksize for a disk when using it
                    124: in a pool; this is almost always 512 or 4096.  Somewhere between 9.0
                    125: and 9.1, at least some disks on some controllers that used to report
                    126: 512 now report 4096.  This provokes a blocksize mismatch warning.
                    127: 
                    128: Given that the native blocksize of the disk didn't change, and things
                    129: seemed OK using the 512 emulated blocks, the warning is likely not
                    130: critical.  However, it is also likely that rebuilding the pool with
                    131: the 4096 blocksize is likely to result in better behavior because ZFS
                    132: will only try to do 4096-byte writes.  \todo Verify this and find the
                    133: actual change and explain better.
                    134: 
1.33      gdt       135: ## pool importing problems
                    136: 
                    137: While one can "zpool pool0 /dev/wd0f" and have a working pool, this
                    138: pool cannot be exported and imported straigthforwardly.  "zpool
1.35      gdt       139: export" works fine, and deletes zpool.cache.  "zpool import", however,
                    140: only looks at entire disks (e.g. /dev/wd0), and might look at slices
                    141: (e.g. /dev/dk0).  It does not look at partitions like /dev/wd0f, and
                    142: there is no way on the command line to ask that specific devices be
                    143: examined.  Thus, export/import fails for pools with disklabel
                    144: partitions.
1.33      gdt       145: 
                    146: One can make wd0 be a link to wd0f temporarily, and the pool will then
                    147: be importable.  However, "wd0" is stored in zpool.cache and on the
                    148: next boot that will attempt to be used.  This is obviously not a good
                    149: approach.
                    150: 
                    151: One an mkdir e.g. /etc/zfs/pool0 and in it have a symlink to
                    152: /dev/wd0f.  Then, zpool import -d /etc/zfs/pool0 will scan
                    153: /etc/zfs/pool0/wd0f and succeed.  The resulting zpool.cache will have
                    154: that path, but having symlinks in /etc/zfs/POOLNAME seems acceptable.
                    155: 
                    156: \todo Determine a good fix, perhaps man page changes only, fix it
                    157: upstream, in curent, and in 9, before removing this discussion.
                    158: 
1.37      gdt       159: ## mountpoint conventions
                    160: 
                    161: By default, datasets are mounted as /poolname/datasetname.  One can
                    162: also set a mountpoint; see zfs(8).
                    163: 
                    164: There does not appear to be any reason to choose explicit mountpoints
                    165: vs the default (and either using data in place or symlinking to it).
                    166: 
1.1       gdt       167: ## mount order
                    168: 
1.13      wiz       169: NetBSD 9 mounts other file systems and then ZFS file systems.  This can
1.3       gdt       170: be a problem if /usr/pkgsrc is on ZFS and /usr/pkgsrc/distfiles is on
                    171: NFS.  A workaround is to use noauto and do the mounts in
                    172: /etc/rc.local.
                    173: 
1.27      gdt       174: NetBSD current after 20200301 mounts ZFS first.  The same issues and
                    175: workarounds apply in different circumstances.
1.1       gdt       176: 
1.19      gdt       177: ## NFS
                    178: 
1.31      gdt       179: zfs filesystems can be exported via NFS, simply by placing them in
                    180: /etc/exports like any other filesystem.
1.27      gdt       181: 
1.31      gdt       182: The "zfs share" command adds a line for each filesystem with the
                    183: sharenfs property set to /etc/zfs/exports, and "zfs unshare" removes
1.33      gdt       184: it.  This file is ignored on NetBSD-9 and current before 20210216; on
                    185: current after 20210216 those filesystems should be exported (assuming
                    186: NFS is enabled).  It does not appear to be possible to set options
                    187: like maproot and network restrictions via this method.
1.31      gdt       188: 
1.33      gdt       189: On current before 20210216, a remote mkdir of a filesystem mounted via
1.31      gdt       190: -maproot=0:10 causes a kernel NULL pointer dereference.  This is now
1.33      gdt       191: fixed.
                    192: 
1.14      gdt       193: ## zvol
                    194: 
                    195: Within a ZFS pool, the standard approach is to have file systems, but
                    196: one can also create a zvol, which is a block device of a certain size.
                    197: 
1.38      gdt       198: As an example, "zfs create -V 16G tank0/xen-netbsd-9-amd64" creates a
                    199: zvol (intended to be a virtual disk for a domU).
                    200: 
                    201: The zvol in the example will appear as
                    202: /dev/zvol/rdsk/tank0/xen-netbsd-9-amd64 and
                    203: /dev/zvol/dsk/tank0/xen-netbsd-9-amd64 and can be used like a
                    204: disklabel partition or wedge.  However, the system will not read
                    205: disklabels and gpt labels from a zvol.
                    206: 
                    207: Doing "swapctl -a" on a zvol device node fails.  \todo Is it really
1.39      gdt       208: true that NetBSD can't swap on a zvol?  (When using a zvol for swap,
                    209: standard advice is to avoid the "-s" option which avoids reserving the
                    210: allocated space.  Standard advice is also to consider using a
                    211: dedicated pool.)
1.14      gdt       212: 
                    213: \todo Explain that one can export a zvol via iscsi.
                    214: 
1.38      gdt       215: One can use ccd to create a normal-looking disk from a zvol.  This
                    216: allows reading a GPT label from the zvol, which is useful in case the
                    217: zvol had been exported via iscsi and some other system created a
                    218: label.
1.14      gdt       219: 
1.1       gdt       220: # Memory usage
                    221: 
                    222: Basically, ZFS uses lots of memory and most people run it on systems
                    223: with large amounts of memory.  NetBSD works well on systems with
                    224: comparatively small amounts of memory.  So a natural question is how
1.27      gdt       225: well ZFS works on one's VAX with 2M of RAM :-) More seriously, one
                    226: might ask if it is reasonable to run ZFS on a RPI3 with 1G of RAM, or
                    227: if it is reasonable on a system with 4G.
                    228: 
                    229: The prevailing wisdom is more or less that ZFS consumes 1G plus 1G per
                    230: 1T of disk.  32-bit architectures are viewed as too small to run ZFS.
                    231: 
                    232: Besides RAM, zfs requires that architecture kernel stack size is at
                    233: least 12KB or more -- some operations cause stack overflow with 8KB
                    234: kernel stack. On NetBSD, the architectures with 16KB kernel stack are
                    235: amd64, sparc64, powerpc, and experimental ia64, hppa. mac68k and sh3
                    236: have 12KB kernel stack. All others use only 8KB stack, which is not
                    237: enough to run zfs.
                    238: 
                    239: NetBSD has many statistics provided via sysctl; see "sysctl
                    240: kstat.zfs".
                    241: 
                    242: FreeBSD has tunables that NetBSD does not seem to have, described in
                    243: [FreeBSD Handbook ZFS Advanced
                    244: section](https://docs.freebsd.org/en/books/handbook/zfs/#zfs-advanced).
1.1       gdt       245: 
1.27      gdt       246: # Interoperability with other systems
1.1       gdt       247: 
1.27      gdt       248: Modern ZFS uses pool version 5000 and feature flags.
1.1       gdt       249: 
1.27      gdt       250: It is in general possible to export a pool and them import the pool on
                    251: some other system, as long as the other system supports all the used
                    252: features.
1.10      gdt       253: 
1.27      gdt       254: \todo Explain how to do this and what is known to work.
1.26      wiki      255: 
1.27      gdt       256: \todo Explain feature flags relationship to FreeBSD, Linux, iIllumos,
                    257: macOS.
1.1       gdt       258: 
1.27      gdt       259: # Sources of ZFS code
1.2       gdt       260: 
1.27      gdt       261: Currently, there are multiple ZFS projects and codebases:
1.1       gdt       262: 
1.27      gdt       263:   - [OpenZFS](http://www.open-zfs.org/wiki/Main_Page)
1.30      wiz       264:   - [openzfs repository](https://github.com/openzfs/zfs)
1.27      gdt       265:   - [zfsonlinux](https://zfsonlinux.org/)
                    266:   - [OpenZFS on OS X ](https://openzfsonosx.org/) [repo](https://github.com/openzfsonosx)
                    267:   - proprietary ZFS in Solaris (not relevant in open source)
                    268:   - ZFS as released under the CDDL (common ancestor, now of historical interest)
1.1       gdt       269: 
1.27      gdt       270: OpenZFS is a coordinating project to align open ZFS codebases.  There
                    271: is a notion of a shared core codebase and OS-specific adaptation code.
1.1       gdt       272: 
1.27      gdt       273:   - [zfsonlinux relationship to OpenZFS](https://github.com/openzfs/zfs/wiki/OpenZFS-Patches)
1.28      gdt       274:   - FreeBSD more or less imports code from openzfs and pushes back fixes. \todo Verify this.
                    275:   - NetBSD has imported code from FreeBSD.
                    276:   - The status of ZFS on macOS is unclear (2021-02).

CVSweb for NetBSD wikisrc <wikimaster@NetBSD.org> software: FreeBSD-CVSweb