When a process is exiting, there is a narrow window where p_cred may be
NULL while its threads are still executing. Specifically, the last thread
to exit a process sets the process state to PRS_ZOMBIE with the proc
spinlock held and then calls thread_exit(). thread_exit() drops the spin
lock, permitting the process to be reaped and thus causing its cred struct
to be released. However, the exiting thread may still cause DTrace probes
to fire by calling sched_throw(), resulting in a double fault if such a
probe enabling attempts to access the GID or UID DIF variables.
The thread's cred reference is not susceptible to this race since it is not
released until after the thread has exited.
MFC after: 1 week
As ZFS requires a more kernel stack pages than is the default on some
architectures e.g. i386, warn if KSTACK_PAGES is less than
ZFS_MIN_KSTACK_PAGES (which is 4 at the time of writing).
MFC after: 3 days
Sponsored by: Multiplay
Transitions 0->1 and 1->0 (which decide e.g. on putting the vnode on the free
list) of either counter are still guarded with vnode interlock.
Reviewed by: kib (earlier version)
Tested by: pho
We now take z_teardown_lock as a writer to ensure that there is no I/O
while the filesystem state is in a flux. Also, zfs_suspend_fs() ->
zfsvfs_teardown() call zfs_unregister_callbacks() and zfs_resume_fs() ->
zfsvfs_setup() call zfs_unregister_callbacks(). Previously there was no
synchronization between those calls and the calls in the re-mounting
case. That could lead to concurrent execution and a crash.
PR: 180060
Differential Revision: https://reviews.freebsd.org/D2865
Suggested by: mahrens
Reviewed by: delphij, pho, mahrens, will
MFC after: 13 days
Sponsored by: ClusterHQ
Reviewed by: Bayard Bell <buffer.g.overflow@gmail.com>
Reviewed by: Alek Pinchuk <alek@nexenta.com>
Reviewed by: Simon Klinkert <simon.klinkert@gmail.com>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Approved by: Richard Lowe <richlowe@richlowe.net>
Author: Matthew Ahrens <mahrens@delphix.com>
illumos/illumos-gate@46e1baa6cfhttps://www.illumos.org/issues/5911
Sometimes ZFS appears to hang while deleting a file. It is actually
making slow progress at the file deletion, but other operations
(administrative and writes via the data path) "hang" until the file
removal completes, which can take a long time if the file has many
blocks. The deletion (or most of it) happens in a single txg, and the
sync thread spends most of its time reading indirect blocks via this
stack trace:
swtch+0x141()
cv_wait+0x70()
zio_wait+0x5b()
dbuf_read+0x2c0()
free_children+0x50()
free_children+0x12a()
free_children+0x12a()
free_children+0x12a()
dnode_sync_free_range_impl+0xdf()
dnode_sync_free_range+0x52()
range_tree_vacate+0x65()
dnode_sync+0x1d8()
dmu_objset_sync_dnodes+0x77()
dmu_objset_sync+0x19f()
dsl_dataset_sync+0x51()
dsl_pool_sync+0x9a()
spa_sync+0x2ff()
txg_sync_thread+0x21f()
thread_start+8()
One way to reproduce the problem is if we are over the arc_meta_limit,
e.g. because lots of indirect blocks are pinned because we have L0
dbufs under them. It could be that most of the L1 indirects are cached,
in which case when dmu_free_long_range_impl() calls dmu_tx_hold_free(),
it will complete very quickly. This allows dmu_free_long_range_impl() to
put many (perhaps all of its) transactions in the same TXG. However,
dmu_free_long_range_impl() calls dnode_evict_dbufs (and
dnode_free_range()), which removes the L0 dbufs, thus reducing the hold
count on the L1 indirect blocks above it, allowing them to be evicted.
Because we are over the arc_meta_limit(), these L1 blocks will be
evicted ASAP. Thus when we get to syncing context, the L1 indirects are
no longer cached and must be read in.
Obtained from: illumos
MFC after: 15 days
Do that only when when fast inline versions are available.
At the moment that can be the case only in the kernel and not for all
platforms.
The original code uses the binary search and that's kept as a fallback.
This is a micro optimization.
Differential Revision: https://reviews.freebsd.org/D2839
Reviewed by: delphij, mahrens, mav
MFC after: 17 days
o Provide an extensive set of assertions for input array of pages.
o Remove now duplicate assertions from different pagers.
Sponsored by: Nginx, Inc.
Sponsored by: Netflix
The problem is that when inline versions of flsl and flsll are not
available, then libkern.h must be included for their declarations
in kernel sources.
The fix would be trivial, but I would like to figure out first if
it even makes sense to use the libkern provided implementations.
Reported by: bz
Pointyhat to: avg
b_size is a logical size of a buffer in memory, b_asize is its physical
size that accounts for possible compression.
Currently the latter is the best approximation for the allocated, on-disk
size.
L2ARC TRIM support was committed a few weeks before L2ARC compression
was imported, so originally the code was correct, because b_size was
the size.
Further thoughts. Given that the cache device is being overwritten
in a circular fashion it is not clear if a TRIM per each evicted L2ARC
buffer has any benefits.
Maybe it would be sufficient to issue a single trim request for the whole
device when it is loaded, e.g. after a bootup, or when it is unloaded, e.g.
before a shutdown. At least as long as L2ARC is not persistent across
reboots.
Discussed with: smh
MFC after: 19 says
This is a micro optimization.
The upstream code uses the binary search.
Differential Revision: https://reviews.freebsd.org/D2839
Reviewed by: delphij, mav
MFC after: 15 days
5946 zfs_ioc_space_snaps must check that firstsnap and lastsnap refer to snapshots
5945 zfs_ioc_send_space must ensure that fromsnap refers to a snapshot
Reviewed by: Steven Hartland <killing@multiplay.co.uk>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Gordon Ross <gordon.ross@nexenta.com>
illumos/illumos-gate@24218bebb4
Note that the upstream commit is modified during MFV: in the upstream
the check is done by inspecting ds_is_snapshot field while in FreeBSD
we call dsl_dataset_is_snapshot().
This is because illumos/illumos-gate@bc9014e6a8
(r277428 in vendor-sys/illumos) is not MFV-ed yet.
MFC after: 10 days
Nothing stops a parallel unmount to suceed before the given call to
dounmount() checks and locks the covered vnode. Prevent dounmount()
from acting on the freed (although type-stable) memory by changing the
interface to require the mount point to be referenced. dounmount()
consumes the reference on return, regardless of the sucessfull or
erronous result.
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
- this should fail early unless the force flag is set
- if the force flag is set then any local modifications including
snapshots should be undone
See: https://www.illumos.org/issues/5912
See: https://reviews.csiden.org/r/220/
Reviewed by: mahrens, Paul Dagnelie <pcd@delphix.com>
MFC after: 15 days
Sponsored by: ClusterHQ
years for head. However, it is continuously misused as the mpsafe argument
for callout_init(9). Deprecate the flag and clean up callout_init() calls
to make them more consistent.
Differential Revision: https://reviews.freebsd.org/D2613
Reviewed by: jhb
MFC after: 2 weeks
This allows to ensure that we do not write to a file that was opened
for reading only or vice versa.
Also, use the correct capability in in zfs_ioc_send_new().
Differential Revision: https://reviews.freebsd.org/D2382
Reviewed by: delphij
MFC after: 17 days
Sponsored by: ClusterHQ
At present zfs_domount() acquires a reference on the filesystem's root vnode
and that reference is kept until zfs_umount.
The latter calls vflush(rootrefs = 1) to dispose of the extra reference.
There is no explanation of why that reference is kept - what problem it
solves or what behavior it improves.
Also, that logic is FreeBSD specific.
There is one real problem with that reference, though.
zfs recv -F may receive a full, non-incremental stream to a mounted filesystem.
In that case the received root object is likely to have a different z_gen
attribute value. Because of that, zfs_rezget will leave the previous root znode
and vnode disassociated from the actual object (z_sa_hdl == NULL).
Thus, future calls to VFS_ROOT() -> zfs_root() will produce a new vnode-znode
pair, while the old one will be kept alive by the outstanding reference.
So, the outstanding reference will not actually be for the new root vnode
(or, more precisely, vnodes - because a root vnode may be recycled and a newer
one can be created).
As a result, when vflush(rootrefs = 1) s called there will be two problems:
- a leaked reference on the old root vnode preventing a graceful unmount
- insufficient references on the actual root vnode leading to a crash upon
access to the vnode after it is destroyed by vgone() + vdrop()
The second issue will actually override the first one.
Differential Revision: https://reviews.freebsd.org/D2353
Reviewed by: delphij, kib, smh
MFC after: 17 days
In traverse_visitbp(), the input argument dnp is modified in the middle
to point to a temporary buffer. Originally this doesn't matter, because
no user of TRAVERSE_POST dereferences it. However, in fbeddd6 a piece of
code is added dereferencing dnp after the modification, creating a possible
bug.
We fix this by creating a new local variable cdnp for the DMU_OT_DNODE case,
so we don't modify the input argument. Also we introduce different local
variables in the DMU_OT_OBJSET case to prevent confusion between the input
argument.
Obtained from: zfsonlinux (a585f2f844ed3d4270221fed88f5e494eb55d932)
MFC after: 2 weeks
Sponsored by: Multiplay
The old comment in zfs_rezget explains what situation the code handles,
the new comment also describes how the situation can arise.
Also, re-join a line that became sufficiently shorti some time ago.
Differential Revision: https://reviews.freebsd.org/D2352
Reviewed by: delphij, smh
MFC after: 12 days
/dev/zfs always has per-open data, so when it is missing the file
descriptor is for some other file. Returning ENOENT in this case
is confusing as a variety of other conditions (like a missing dataset)
may result in the same error. It's better to consistently return
EBADF for any problems with the file descriptor.
Note that zfs_onexit_fd_hold() is used with 'automatic cleanup fd'
- when that fd is closed, typically because a process is terminated,
some cleanup action is taken by ZFS driver. E.g. a temporary
snapshot hold is released.
Perhaps, it would even be worthwhile changing devfs_get_cdevpriv()
to return EBADF if there is no associated data.
Differential Revision: https://reviews.freebsd.org/D2370
Reviewed by: delphij, smh
MFC after: 12 days
It would previously call into some unfinished Solaris compatibility code and
return without actually calling panic(9). The compatibility code is
unneeded, however, so just remove it and have dtrace_panic() call vpanic(9)
directly.
Differential Revision: https://reviews.freebsd.org/D2349
Reviewed by: avg
MFC after: 2 weeks
Sponsored by: EMC / Isilon Storage Division
Passing "-x lazyload" to dtrace -G during compilation causes dtrace(1) to
not link drti.o into the output object file, so the USDT probes are not created
during process startup. Instead, dtrace(1) will automatically discover and
create probes on the process' behalf when attaching.
Differential Revision: https://reviews.freebsd.org/D2203
Reviewed by: rpaulo
MFC after: 1 month
Even on Illumos, with its much larger KVA, ZFS ARC steps back if KVA usage
reaches certain threshold (3/4 on i386 or 16/17 otherwise). FreeBSD has
even less KVA, but had no such limit on archs with direct map as amd64.
As result, on machines with a lot of RAM, during load with very small user-
space memory pressure, such as `zfs send`, it was possible to reach state,
when there is enough both physical RAM and KVA (I've seen up to 25-30%),
but no continuous KVA range to allocate even single 128KB I/O request.
Address this situation from two sides:
- restore KVA usage limitations in a way the most close to Illumos;
- introduce new requirement for KVA fragmentation, specifying that we
should have at least one sequential KVA range of zfs_max_recordsize bytes.
Experiments show that first limitation done alone is not sufficient. On
machine with 64GB of RAM it is sometimes needed to drop up to half of ARC
size to get at leats one 1MB KVA chunk. Statically limiting ARC to half
of KVA/RAM is too strict, so second limitation makes it to work in cycles:
accumulate trash up to certain critical mass, do massive spring-cleaning,
and then start littering again. :)
MFC after: 1 month
If zvol_geom_start is called with a BIO_DELETE from a thread which can
sleep it queues it for later processing by the zvol_geom_worker. The
zvol_geom_worker didn't have a delete case so would simply loose the bio
hence preventing the original caller from every completing. In addition
an other unknown types would suffer the same fate.
Allow zvol_geom_worker to process BIO_DELETE's via zvol_strategy and
return unsupported for all unknown bio types.
MFC after: 2 weeks
Sponsored by: Multiplay
emulate the instructions used in function entry and exit.
For function entry ARM will use a push instruction to push up to 16
registers to the stack. While we don't expect all 16 to be used we need to
handle any combination the compiler may generate, even if it doesn't make
sense (e.g. pushing the program counter).
On function return we will either have a pop or branch instruction. The
former is similar to the push instruction, but with care to make sure we
update the stack pointer and program counter correctly in the cases they
are either in the list of registers or not. For branch we need to take the
24-bit offset, sign-extend it, and add that number of 4-byte words to the
program counter. Care needs to be taken as, due to historical reasons, the
address the branch is relative to is not the current instruction, but 8
bytes later.
This allows us to use the following probes on ARM boards:
dtrace -n 'fbt::malloc:entry { stack() }'
and
dtrace -n 'fbt:🆓return { stack() }'
Differential Revision: https://reviews.freebsd.org/D2007
Reviewed by: gnn, rpaulo
Sponsored by: ABT Systems Ltd
