That was both redundant as zfs_znode_sa_init() already does the job and
insufficient as the root vnode can be reached via other means.
MFC after: 1 weeks
gfs code is (almsot) completely agnostic of FreeBSD VFS locking, so it
does not handle doomed but not yet dead vnodes and may return them.
Check for those vnodes here and retry a lookup.
Note that ZFS and gfs have additional protections that ensure that a
parent vnode of the current vnode is never doomed.
The fixed problem is an occasional failure to lookup a 'snapshot' or
'shares' directories under .zfs.
Note that for the above reason all uses of zfsctl_root_lookup() are
better be replaced with VOP_LOOKUP.
MFC after: 5 weeks
Speedup is hard to measure because the only time vdev_geom_open_by_guids
gets called on many drives at the same time is during boot. But with
vdev_geom_open hacked to always call vdev_geom_open_by_guids, operations
like "zpool create" speed up by 65%.
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c
* Read all of a vdev's labels in parallel instead of sequentially.
* In vdev_geom_read_config, don't read the entire label, including
the uberblock. That's a waste of RAM. Just read the vdev config
nvlist. Reduces the IO and RAM involved with tasting from 1MB to
448KB.
Reviewed by: avg
MFC after: 4 weeks
Sponsored by: Spectra Logic Corp
Differential Revision: https://reviews.freebsd.org/D6153
FreeBSD zfs_ioc_rename() has an option, not present upstream, that
allows to rename snapshots without unmounting them first. I am not sure
what is a rationale for that option, but its actual behavior was the
opposite of the intended behavior. That is, by default the snapshots
were not unmounted.
The option was introduced as part of a large update from upstream in
r248498.
One of the consequences was a havoc under .zfs/snapshot after the rename.
The snapshots got new names but were mounted on top of directories with
old names, so readdir would list the new names, but lookup would still
find the old mounts.
PR: 209093
Reported by: Frédéric VANNIÈRE <f.vanniere@planet-work.com>
MFC after: 5 days
That was just wrong. In fact, we can safely keep this static entry when
it's inactive.
Now the destructive action is moved to the reclaim method and the
function is renamed from zfsctl_snapdir_inactive(0 to
zfsctl_snapdir_reclaim().
Also, we can use gfs_vop_reclaim() instead of gfs_dir_inactive() +
kmem_free().
Lastly, we can just assert that the node does not any children when it
is reclaimed, even on the force unmount. That's because zfs_umount()
does an extra vflush() pass which should destroy all snapshot-mountpoint
vnodes that are the snapdir's children.
MFC after: 5 weeks
Those vnodes should not linger. "Stale" nodes may get out of
synchronization with actual snapshots. For example if we destroy a
snapshot and create a new one with the same name. Or when we rename a
snapshot.
While there fix the argument type for zfsctl_snapshot_reclaim().
Also, its original argument can be passed to gfs_vop_reclaim() directly.
Bug 209093 could be related although I have not specifically verified
that. Referencing just in case.
PR: 209093
MFC after: 5 weeks
Dropping the root vnode's lock after VFS_ROOT() didn't really help the
fact that we acquired the lock while holding its child's, .zfs, lock
while performing the operaiton.
So, directly use zfs_zget() to get the root vnode.
While there simplify the code in zfsctl_freebsd_root_lookup.
We know that .zfs is always exclusively locked.
We know that there is already a reference on *vpp, so no need for an
extra one.
Account for the fact that .. lookup may ask for a different lock type,
not necessarily LK_EXCLUSIVE. And handle a possible failure to acquire
the lock given the lock flags.
MFC after: 5 weeks
In fact, that was dangerous. For example, zfsctl_snapshot_reclaim()
calls gfs_dir_lookup() on ".." path and that ends up calling
gfs_lookup_dot() which violated locking order by acquiring the parent's
directory vnode lock after the child's vnode lock.
Also, the previous behavior was inconsistent as gfs_dir_lookup()
returned a locked vnode for . and .. lookups, but not for any other.
Now gfs_lookup_dot() just references a resulting vnode and the locking
is done in its consumers, where necessary.
Note that we do not enable shared locking support for any gfs / zfsctl
vnodes.
This commit partially reverts r273641.
MFC after: 5 weeks
The former acquired a snap vnode lock while holding sd_lock while the
latter does the opposite.
The solution is drop sd_lock before acquiring the vnode lock. That
should be okay as we are still holding a lock on the 'snapshot'
directory in the exclusive mode. That lock ensures that there are no
concurrent lookups in the directory and thus no concurrent mount attempts.
But now we have to account for the possibility that the snap vnode
might get reclaim after we drop sd_lock and before we can get
the node lock. So, check for that case and retry.
MFC after: 5 weeks
This commit partially reverts r273641 which introduced the leak.
It did so to accomodate for some consumers of traverse() that expected
the starting vnode to stay as-is. But that introduced the leak in the
case when a mounted filesystem was found and its root vnode was
returned.
r299914 removed the troublesome consumers and now there is no reason to
keep the starting vnode. So, now the new rules are:
- if there is no mounted filesystem, then nothing is changed
- otherwise the starting vnode is always released
- the root vnode of the mounted filesystem is returned locked and
referenced in the case of success
MFC after: 5 weeks
X-MFC after: r299914
We pretend that snapshots mounted under .zfs are part of the original
filesystem and we try very hard to hide vnodes on top of which the snapshots
are mounted. Given that I believe that the removed operations should
never be called. They might have been called previously because
of issues fixed in r299906, r299908 and r299913.
MFC after: 5 weeks
Previosuly we did that only if the snapshot was mounted earlier, its
root vnode got recycled and then we accessed it again.
We never cleared the flag for a freshly mounted snapshot.
That was very inconsistent and probably a source of some bugs.
Or maybe that painted over some bugs which might get revealed now.
We should consistently clear the flag because we try very hard to
pretend that snapshots auto-mounted under .zfs are part of their
original filesystem. In other words, we try to hide the fact that they
are different filesystems / mountpoints.
MFC after: 5 weeks
The logic is similar to that already present in zfs_dirlook() to handle
a dot-dot lookup on a root vnode of a snapshot mounted under
.zfs/snapshot/.
illumos does not have an equivalent of vop_vptocnp, so there only the
lookup had to be patched up.
MFC after: 4 weeks
* Remove excessive references on a snapshot mountpoint vnode.
zfsctl_snapdir_lookup() called VN_HOLD() on a vnode returned from
zfsctl_snapshot_mknode() and the latter also had a call to VN_HOLD()
on the same vnode.
On top of that gfs_dir_create() already returns the vnode with the
use count of 1 (set in getnewvnode).
So there was 3 references on the vnode.
* mount_snapshot() should keep a reference to a covered vnode.
That reference is owned by the mountpoint (mounted snapshot filesystem).
* Remove cryptic manipulations of a covered vnode in zfs_umount().
FreeBSD dounmount() already does the right thing and releases the covered
vnode.
PR: 207464
Reported by: dustinwenz@ebureau.com
Tested by: Howard Powell <hpowell@lighthouseinstruments.com>
MFC after: 3 weeks
Currently, Application Processors (non-boot CPUs) are started by
MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until
SI_SUB_SMP at which point they are released to run kernel threads.
SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter
the scheduler and start running threads until fairly late in the
boot.
This change moves SI_SUB_SMP up to just before software interrupt
threads are created allowing the APs to start executing kernel
threads much sooner (before any devices are probed). This allows
several initialization routines that need to perform initialization
on all CPUs to now perform that initialization in one step rather
than having to defer the AP initialization to a second SYSINIT run
at SI_SUB_SMP. It also permits all CPUs to be available for
handling interrupts before any devices are probed.
This last feature fixes a problem on with interrupt vector exhaustion.
Specifically, in the old model all device interrupts were routed
onto the boot CPU during boot. Later after the APs were released at
SI_SUB_SMP, interrupts were redistributed across all CPUs.
However, several drivers for multiqueue hardware allocate N interrupts
per CPU in the system. In a system with many CPUs, just a few drivers
doing this could exhaust the available pool of interrupt vectors on
the boot CPU as each driver was allocating N * mp_ncpu vectors on the
boot CPU. Now, drivers will allocate interrupts on their desired CPUs
during boot meaning that only N interrupts are allocated from the boot
CPU instead of N * mp_ncpu.
Some other bits of code can also be simplified as smp_started is
now true much earlier and will now always be true for these bits of
code. This removes the need to treat the single-CPU boot environment
as a special case.
As a transition aid, the new behavior is available under a new kernel
option (EARLY_AP_STARTUP). This will allow the option to be turned off
if need be during initial testing. I plan to enable this on x86 by
default in a followup commit in the next few days and to have all
platforms moved over before 11.0. Once the transition is complete,
the option will be removed along with the !EARLY_AP_STARTUP code.
These changes have only been tested on x86. Other platform maintainers
are encouraged to port their architectures over as well. The main
things to check for are any uses of smp_started in MD code that can be
simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in
the EARLY_AP_STARTUP case (e.g. the interrupt shuffling).
PR: kern/199321
Reviewed by: markj, gnn, kib
Sponsored by: Netflix
delete permissions for ACLs
Reviewed by: Gordon Ross <gwr@nexenta.com>
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Author: Kevin Crowe <kevin.crowe@nexenta.com>
openzfs/openzfs@a40149b935
aclmode=passthrough
Reviewed by: Gordon Ross <gwr@nexenta.com>
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Author: Albert Lee <trisk@nexenta.com>
openzfs/openzfs@1bcf0d240b
perms (groupmask)
Reviewed by: Gordon Ross <gwr@nexenta.com>
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Author: Albert Lee <trisk@nexenta.com>
openzfs/openzfs@eebb483d0c
some additional considerations
Reviewed by: Gordon Ross <gwr@nexenta.com>
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Author: Kevin Crowe <kevin.crowe@nexenta.com>
openzfs/openzfs@d316fffc9c
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: John Kennedy <john.kennedy@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Don Brady <don.brady@intel.com>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Approved by: Richard Lowe <richlowe@richlowe.net>
Author: Joe Stein <joe.stein@delphix.com>
openzfs/openzfs@215198a6ad
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Robert Mustacchi <rm@joyent.com>
Author: Tim Chase <tim@chase2k.com>
openzfs/openzfs@445e67805d
clear vd->vdev_tsd in vdev_geom_close_locked instead of vdev_geom_detach.
In the latter function, it would fail to happen in certain circumstances
where cp->private was unset. Ideally, the latter should never happen, but
it can happen when vdev open fails, or where spares are involved.
MFC after: 4 weeks
X-MFC-With: 298786
Sponsored by: Spectra Logic Corp
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c
Move checks for provider's sectorsize and mediasize into a single
location in vdev_geom_attach. Remove the zfs::vdev::taste class;
it's ok to use the regular vdev class for tasting. Consolidate guid
checks into a single location in vdev_attach_ok. Consolidate some
error handling code from vdev_geom_attach into vdev_geom_detach,
closing a resource leak of geom consumers in the process.
Reviewed by: avg
MFC after: 4 weeks
Sponsored by: Spectra Logic Corp
Differential Revision: https://reviews.freebsd.org/D5974
This allows for the long function components encountered in www/firefox.
This constant is part of DTrace's userland ABI, so this change may not be
MFC'ed.
PR: 207735
Without this change, DTrace will refuse to load a DOF section if the
function component of any of its probes exceeds DTRACE_FUNCNAMELEN (128).
Probes in C++ programs can have very long function components. Rather than
rejecting all probes if a single probe exceeds the limit, simply skip the
invalid probe and emit a warning. This ensures that valid probes are
instantiated.
PR: 207735
MFC after: 2 weeks
When this flag is turned on, DOF and DIF validation errors are printed to
the kernel message buffer. This is useful for debugging.
Also remove the debug.dtrace.debug sysctl, which has no effect.
While the instructions were not included into the original instruction
set, their support can be indicated by a special feature bit.
For example:
CPU: AMD Phenom(tm) II X4 955 Processor (3214.71-MHz K8-class CPU)
...
AMD Features2=0x37ff<LAHF, ...>
Clang 3.8 uses lahf/sahf as a faster alternative to pushf/popf where
possible.
MFC after: 2 weeks
illumos/illumos-gate@26455f9efc26455f9efchttps://www.illumos.org/issues/6052
At the moment type parameter of lzc_create() is of dmu_objset_type_t type.
That exposes an implementation detail and requires sys/fs/zfs.h to be included
in libzfs_core.h creating unnecessary coupling between libzfs_core interface
and ZFS internals.
I think that dmu_objset_type_t should be replaced with a libzfs_core
enumeration of supported dataset types.
For ABI reasons the new enumeration could be bit-compatible with
dmu_objset_type_t.
For example:
typedef enum {
LZC_DST_ZFS = 2,
LZC_DST_ZVOL
} lzc_dataset_type_t;
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Richard Lowe <richlowe@richlowe.net>
Author: Andriy Gapon <andriy.gapon@clusterhq.com>
MFC after: 2 weeks
Sponsored by: ClusterHQ
Currently this argument is a pointer into the stack which is used by FBT
to fetch the first five probe arguments. On all non-x86 architectures it's
simply the trapframe address, so this change has no functional impact. On
amd64 it's a pointer into the trapframe such that stack[1 .. 5] gives the
first five argument registers, which are deliberately grouped together in
the amd64 trapframe definition.
A trapframe argument simplifies the invop handlers on !x86 and makes the
x86 FBT invop handler easier to understand. Moreover, it allows for invop
handlers that may want to modify the register set of the interrupted thread.
Note that now we have to account for possible partial writes
in dmu_write_uio_dbuf(). It seems that on illumos either all or none
of the data are expected to be written. But the partial writes are
quite expected when vn_io_fault support is enabled.
Reviewed by: kib
MFC after: 7 weeks
Differential Revision: https://reviews.freebsd.org/D2790
Prior to this change, vdev_geom_open_by_path would call vdev_geom_attach
prior to verifying the device's GUIDs. vdev_geom_attach calls
vdev_geom_attrchange to set the physpath in the vdev object. The result is
that if the disk could not be found, then the labels for other disks in the
same TLD would overwrite the missing disk's physpath with the physpath of
whichever disk currently has the same devname as the missing one used to
have.
MFC after: 4 weeks
Sponsored by: Spectra Logic Corp
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c
Don't drop the g_topology_lock before freeing old_physpath. That
opens up a race where one thread can call vdev_geom_attrchanged,
set old_physpath, drop the g_topology_lock, then block trying to
acquire the SCL_STATE lock. Then another thread can come into
vdev_geom_attrchanged, set old_physpath to the same value, and
proceed to free it. When the first thread resumes, it will free
the same location.
It turns out that the SCL_STATE lock isn't needed. It was
originally added by gibbs to protect vd->vdev_physpath while
updating the same. However, the update process subsequently was
switched to an atomic operation (a pointer swap). Now, there is
no need for the SCL_STATE lock, and hence no need to drop the
g_topology_lock.
Reviewed by: delphij
MFC after: 4 weeks
Sponsored by: Spectra Logic Corp
Differential Revision: https://reviews.freebsd.org/D5413
Previously uncompressed buffers did not obey that rule.
Type of b_asize is changed to uint64_t for consistency,
given that this is a zeta-byte filesystem.
l2arc_compress_buf is renamed to l2arc_transform_buf to better reflect
its new utility. Now not only we ensure that a compressed buffer has
a size aligned to ashift, but we also allocate a properly sized
temporary buffer if the original buffer is not compressed and it has
an odd size. This ensures that all I/O to the cache device is always
ashift-aligned, in terms of both a request offset and a request size.
If the aligned data is larger than the original data, then we have to use
a temporary buffer when reading it as well.
Also, enhance physical zio alignment checks using vdev_logical_ashift.
On FreeBSD we have this information, so we can make stricter assertions.
Reviewed by: smh, mav
MFC after: 1 month
Sponsored by: ClusterHQ
Differential Revision: https://reviews.freebsd.org/D2789
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Paul Dagnelie <pcd@delphix.com>
Author: Alexander Motin <mav@FreeBSD.org>
Improve speculative prefetch of indirect blocks.
Scalability of many operations on wide ZFS pool can be limited by
requirement to prefetch indirect blocks first. Recently added
asynchronous indirect block read partially helped, but did not
solve the problem completely. This patch extends existing prefetcher
functionality to explicitly work with indirect blocks.
Before this change prefetcher issued reads for up to 8MB of data in
advance. With this change it also issues indirect block reads
for up to 64MB of data in advance, so that when it will be time to
actually read those data, it can be done immediately. Alike effect
can be achieved by just increasing maximal data prefetch distance,
but at higher memory cost.
Also this change introduces indirect block prefetch for rewrite
operations, that was never done before. Previously ARC miss for
Indirect blocks regularly blocked rewrites, converting perfectly
aligned asynchronous operations into synchronous read-write pairs,
significantly reducing maximal rewrite speed.
While being there this issue was also fixed:
- prefetch was done always, even if caching for the dataset was
completely disabled.
Testing on FreeBSD with zvol on top of 6x striped 2x mirrored pool
of 12 assorted HDDs shown me such performance numbers:
------- BEFORE --------
Write 491363677 bytes/sec
Read 312430631 bytes/sec
Rewrite 97680464 bytes/sec
-------- AFTER --------
Write 493524146 bytes/sec
Read 438598079 bytes/sec
Rewrite 277506044 bytes/sec
Closes#65Closes#80openzfs/openzfs@792fd28ac0
