illumos/illumos-gate@dbfd9f9300dbfd9f9300https://www.illumos.org/issues/8156
dbuf_evict_notify() holds the dbuf_evict_lock while checking if it should do
the eviction itself (because the evict thread is not able to keep up).
This can result in massive lock contention.
It isn't necessary to hold the lock, because if we make the wrong choice
occasionally, nothing bad will happen.
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: Paul Dagnelie <pcd@delphix.com>
Approved by: Robert Mustacchi <rm@joyent.com>
Author: Matthew Ahrens <mahrens@delphix.com>
MFC after: 1 week
illumos/illumos-gate@5b062782535b06278253https://www.illumos.org/issues/8005
RAID-Z requires that space be allocated in multiples of P+1 sectors,
because this is the minimum size block that can have the required amount
of parity. Thus blocks on RAIDZ1 must be allocated in a multiple of 2
sectors; on RAIDZ2 multiple of 3; and on RAIDZ3 multiple of 4. A sector
is a unit of 2^ashift bytes, typically 512B or 4KB.
To satisfy this constraint, the allocation size is rounded up to the
proper multiple, resulting in up to 3 "pad sectors" at the end of some
blocks. The contents of these pad sectors are not used, so we do not
need to read or write these sectors. However, some storage hardware
performs much worse (around 1/2 as fast) on mostly-contiguous writes
when there are small gaps of non-overwritten data between the writes.
Therefore, ZFS creates "optional" zio's when writing RAID-Z blocks that
include pad sectors. If writing a pad sector will fill the gap between
two (required) writes, we will issue the optional zio, thus doubling
performance. The gap-filling performance improvement was introduced in
July 2009.
Writing the optional zio is done by the io aggregation code in
vdev_queue.c. The problem is that it is also subject to the limit on
the size of aggregate writes, zfs_vdev_aggregation_limit, which is by
default 128KB. For a given block, if the amount of data plus padding
written to a leaf device exceeds zfs_vdev_aggregation_limit, the
optional zio will not be written, resulting in a ~2x performance
degradation.
The problem occurs only for certain values of ashift, compressed block
size, and RAID-Z configuration (number of parity and data disks). It
cannot occur with the default recordsize=128KB. If compression is
enabled, all configurations with recordsize=1MB or larger will be
impacted to some degree.
The problem notably occurs with recordsize=1MB, compression=off, with 10
disks in a RAIDZ2 or RAIDZ3 group (with 512B or 4KB sectors). Therefore
Reviewed by: Saso Kiselkov <saso.kiselkov@nexenta.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com>
Approved by: Robert Mustacchi <rm@joyent.com>
Author: Matthew Ahrens <mahrens@delphix.com>
MFC after: 10 days
illumos/illumos-gate@adaec86ad2adaec86ad2https://www.illumos.org/issues/8155
When writing pre-compressed buffers, arc_write() requires that the compression
algorithm used to compress the buffer matches the compression algorithm
requested by the zio_prop_t, which is set by dmu_write_policy().
This makes dmu_write_policy() and its callers a bit more complicated.
We can simplify this by making arc_write() trust the caller to supply the type
of pre-compressed buffer that it wants to write, and override the compression
setting in the zio_prop_t.
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Robert Mustacchi <rm@joyent.com>
Author: Matthew Ahrens <mahrens@delphix.com>
MFC after: 10 days
illumos/illumos-gate@79809f9cf479809f9cf4https://www.illumos.org/issues/8269
It seems that currently normalization of stddev aggregation is done
incorrectly.
We divide both the sum of values and the sum of their squares by the
normalization factor. But we should divide the sum of squares by the
normalization factor squared to scale the original values properly.
FreeBSD note: the actual change was committed in r316853, this commit
adds the test files and record merge information.
Reviewed by: Bryan Cantrill <bryan@joyent.com>
Approved by: Robert Mustacchi <rm@joyent.com>
Author: Andriy Gapon <avg@FreeBSD.org>
MFC after: 1 week
Sponsored by: Panzura
Its purpose was to translate the values for msdosfs inode numbers,
which is calculated from the msdosfs structures describing the file,
into the range representable by 32bit ino_t. The translation acted
for filesystems larger than 128Gb, it reserved the range 0xf0000000
(FILENO_FIRST_DYN) to UINT32_MAX and remembered some arbitrary
translation of ino >= FILENO_FIRST_DYN into this range. It consumed
memory that could be only freed by unmount, and the translation was
not stable across remounts.
With ino_t type extended to 64 bit, there is no such issue and values
can be returned without compaction to 32bit. That is, for the native
environments, the translation layer is not necessary and adds
significant undeserved code complexity. For compat ABIs which use
32bit ino_t, the vfs.ino64_trunc_error sysctl provides some measures
to soften the failure mode when inode numbers truncation is not safe.
Discussed with: bde
Sponsored by: The FreeBSD Foundation
Provide a new mode "2" which returns a special overflow indicator in
the non-representable field instead of the silent truncation (mode
"0") or EOVERFLOW (mode "1").
In particular, the typical use of st_ino to detect hard links with
mode "2" reports false positives, which might be more suitable for
some uses.
Discussed with: bde
Sponsored by: The FreeBSD Foundation
internal zones only. This allows to create new zones at early stages
of boot, without need to mark them as internal to UMA, which isn't
always true.
Reviewed by: alc
o Separate fields of struct socket that belong to listening from
fields that belong to normal dataflow, and unionize them. This
shrinks the structure a bit.
- Take out selinfo's from the socket buffers into the socket. The
first reason is to support braindamaged scenario when a socket is
added to kevent(2) and then listen(2) is cast on it. The second
reason is that there is future plan to make socket buffers pluggable,
so that for a dataflow socket a socket buffer can be changed, and
in this case we also want to keep same selinfos through the lifetime
of a socket.
- Remove struct struct so_accf. Since now listening stuff no longer
affects struct socket size, just move its fields into listening part
of the union.
- Provide sol_upcall field and enforce that so_upcall_set() may be called
only on a dataflow socket, which has buffers, and for listening sockets
provide solisten_upcall_set().
o Remove ACCEPT_LOCK() global.
- Add a mutex to socket, to be used instead of socket buffer lock to lock
fields of struct socket that don't belong to a socket buffer.
- Allow to acquire two socket locks, but the first one must belong to a
listening socket.
- Make soref()/sorele() to use atomic(9). This allows in some situations
to do soref() without owning socket lock. There is place for improvement
here, it is possible to make sorele() also to lock optionally.
- Most protocols aren't touched by this change, except UNIX local sockets.
See below for more information.
o Reduce copy-and-paste in kernel modules that accept connections from
listening sockets: provide function solisten_dequeue(), and use it in
the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4),
infiniband, rpc.
o UNIX local sockets.
- Removal of ACCEPT_LOCK() global uncovered several races in the UNIX
local sockets. Most races exist around spawning a new socket, when we
are connecting to a local listening socket. To cover them, we need to
hold locks on both PCBs when spawning a third one. This means holding
them across sonewconn(). This creates a LOR between pcb locks and
unp_list_lock.
- To fix the new LOR, abandon the global unp_list_lock in favor of global
unp_link_lock. Indeed, separating these two locks didn't provide us any
extra parralelism in the UNIX sockets.
- Now call into uipc_attach() may happen with unp_link_lock hold if, we
are accepting, or without unp_link_lock in case if we are just creating
a socket.
- Another problem in UNIX sockets is that uipc_close() basicly did nothing
for a listening socket. The vnode remained opened for connections. This
is fixed by removing vnode in uipc_close(). Maybe the right way would be
to do it for all sockets (not only listening), simply move the vnode
teardown from uipc_detach() to uipc_close()?
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9770
- For HMAC requests, construct a special input buffer to request an empty
hash result.
- For plain cipher requests and requests that chain an AES cipher with an
HMAC, fail with EINVAL if there is no cipher payload. If needed in
the future, chained requests that only contain AAD could be serviced as
HMAC-only requests.
- For GCM requests, the hardware does not support generating the tag for
an AAD-only request. Instead, complete these requests synchronously
in software on the assumption that such requests are rare.
Sponsored by: Chelsio Communications
during bootup. Debugging information shows that softclock_call_cc() is
trying to execute the vt_consdev.vd_timer callout, and the callout
structure contains a NULL c_func.
This appears to be due to a race between vt_upgrade() running
callout_reset() and vt_resume_flush_timer() calling callout_schedule().
Fix the race by ensuring that vd_timer_armed is always set before
attempting to (re)schedule the callout.
Discussed with: emaste
MFC after: 2 weeks
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D9828
stack modules.
It adds support for mangling symbols exported by a module by prepending
a string to them. (This avoids overlapping symbols in the kernel linker.)
It allows the use of a macro as the module name in the DECLARE_MACRO()
and MACRO_VERSION() macros.
It allows the code to register stack aliases (e.g. both a generic name
["default"] and version-specific name ["default_10_3p1"]).
With these changes, it is trivial to compile TCP stack modules with
the name defined in the Makefile and to load multiple versions of the
same stack simultaneously. This functionality can be used to enable
side-by-side testing of an old and new version of the same TCP stack.
It also could support upgrading the TCP stack without a reboot.
Reviewed by: gnn, sjg (makefiles only)
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D11086
With Clang 5.0 the .arch directive is required, otherwise Clang
complains "error: instruction requires: crc".
This was reported in D10499 but not added initially, because clang 3.8
available on a ref machine reported unknown directive. Clang 4.0 allows
but does not require the directive.
Submitted by: andrew
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Add pcie-controller node as a bus-parent of pcie nodes for Armada38x
boards. This reduces diff between Linux and FreeBSD PCIe device tree
representation to the minimum. This commit also allows for using multiple
PCIe ports, thanks to the recent driver updates, which support such
hierarchy. Restore original PCIe nodes in armada-385.dtsi and
apply necessary changes in hitherto unused armada-380.dtsi.
Submitted by: Michal Mazur <mkm@semihalf.com>
Marcin Wojtas <mw@semihalf.com>
Obtained from: Semihalf
Sponsored by: Stormshield, Netgate
Differential revision: https://reviews.freebsd.org/D10907
This commit is another part of preparation for PCIe multi-port
support for Marvell SoCs. Some device trees include pcie-controller
node as a bus-parent of pcie nodes. This patch adds support for
new bus, collects and configures device informations and finally
adds PCIB devices as a childs of pcie-controller in Newbus hierarchy.
Submitted by: Marcin Mazurek <mma@semihalf.com>
Obtained form: Semihalf
Sponsored by: Stormshield
Reviewed by: https://reviews.freebsd.org/D10906
Original PCIe nodes for Marvell SoCs consists of ports' nodes
under main controller node. In order to properly parse
this kind of representation in DT a mechanism for traversing
through the tree required an update. Moreover, processing FDT
data consisting of more than 2 cells had to be fixed,
because the 'reg' property of mrvl,pcie node have additional
parameter in front of 64-bit address. It should be skipped
by default. This commit works properly with old mrvl,pcie
representation for Kirkwood and ArmadaXP SoCs.
Submitted by: Wojciech Macek <wma@semihalf.com>
Michal Mazur <mkm@semihalf.com>
Obtained from: Semihalf
Sponsored by: Stormshield, Netgate
Differential revision: https://reviews.freebsd.org/D10905
This patch fixes sporadic problems with updating time
with mv_rtc driver by configuring access to it via MBUS.
For this purpose already existing second set of resources
in rtc@3800 node of Armada 38x DT is used.
Submitted by: Dominik Ermel <der@semihalf.com>
Obtained from: Semihalf
Sponsored by: Stormshield
Differential revision: https://reviews.freebsd.org/D10901
This commit enables optional reset of the RTC, in case
its registers' contents did not sustain the reboot or power-off/on
sequence. Without it, further usage of RTC is impossible
(e.g. writing values to RTC_TIME register will not succeed).
The reset is performed only if Clock Correction register
does not comprise RTC_NOMINAL_TIMING, what helps to distinguish,
whether the software configured RTC before or it comprises
the default value.
Submitted by: Bartosz Szczepanek <bsz@semihalf.com>
Obtained from: Semihalf
Sponsored by: Stormshield
Differential revision: https://reviews.freebsd.org/D10900
r31386 changed how the size of the VM page array was calculated to be
less wasteful. For most systems, the amount of memory is divided by
the overhead required by each page (a page of data plus a struct vm_page)
to determine the maximum number of available pages. However, if the
remainder for the first non-available page was at least a page of data
(so that the only memory missing was a struct vm_page), this last page
was left in phys_avail[] but was not allocated an entry in the VM page
array. Handle this case by explicitly excluding the page from
phys_avail[].
Reviewed by: alc
Sponsored by: DARPA / AFRL
Differential Revision: https://reviews.freebsd.org/D11000
subtree is already zero, then setting the "largest contiguous free block"
hint for that subtree to anything other than zero makes no sense. To be
clear, assigning a value to the hint that is too large is not a correctness
problem, only a pessimization.
Dragonfly BSD has applied the same change to blst_meta_alloc() but not
blst_meta_fill().
MFC after: 6 weeks
Currently the PCI domain is initialized with the instance GUID in
vmbus_pcib_attach(). It turns out the GUID can change across VM reboot,
while some users want a persistent value for PCI domain. The solution is
that we can change to use the device serial number, which starts with 1
and is unique within a VM.
Obtained from: Haiyang Zhang
MFC after: 1 day
Sponsored by: Microsoft
This happens when closing a socket with upcall, and trace is: soclose()->
... protocol ... -> soisdisconnected() -> socantrcvmore_locked() ->
sowakeup() -> soisconnected().
Right now this case is innocent for two reasons. First, soisconnected()
doesn't clear SS_ISDISCONNECTED flag. Second, the mutex to lock the
socket is the socket receive buffer mutex, and sodisconnected() first
disables the receive buffer. But in future code, the mutex to lock
socket is different to buffer mutex, and we would get undesired mutex
recursion.
The fix is to check SS_ISDISCONNECTED flag before calling upcall.
ARM and MIPS fail universe builds.
ARM and MIPS are missing the following:
* VM_MEMATTR_WRITE_THROUGH
* VM_MEMATTR_WRITE_COMBINING
Pointy-hat to: jhibbits
arm, mips, and powerpc all implement pmap_mapdev_attr() and pmap_unmapdev(),
so add those archs to the checks. powerpc also includes the atomic_swap_*()
functions, so add that to the supported list as well. Not tested except by
compiling powerpc.
Reviewed by: markj
testing purposes. However, over the years, various changes to the kernel
have broken this feature. This revision applies some fixes to get user-
space compilation working again. There are no changes in this revision
to code that is used by the kernel.
MFC after: 3 days
beginning of a swap area for a disk label. However, neither r118390 nor
r118544, which increased the reservation from one to two blocks, correctly
accounted for these blocks when updating the variable "swap_pager_avail".
This change corrects that error.
Reviewed by: kib
MFC after: 5 days
CPU_FOREACH() is not available until SI_SUB_CPU at SI_ORDER_ANY
when the LinuxKPI is loaded as part of the kernel.
MFC after: 1 week
Sponsored by: Mellanox Technologies
pager used a different scheme for striping the allocation of swap space
across multiple devices. And, although blist_fill() was intended to support
fill operations with large counts, the old striping scheme never performed a
fill larger than the stripe size. Consequently, the misplacement of a
sanity check in blst_meta_fill() went undetected. Now, moving forward in
time to r118390, a new scheme for striping was introduced that maintained a
blist allocator per device, but as noted in r318995, swapoff_one() was not
fully and correctly converted to the new scheme. This change completes what
was started in r318995 by fixing the underlying bug in blst_meta_fill() that
stops swapoff_one() from simply performing a single blist_fill() operation.
Reviewed by: kib
MFC after: 5 days
Differential Revision: https://reviews.freebsd.org/D11043
You may now optionally specify allow.noreserved_ports to prevent root
inside a jail from using privileged ports (less than 1024)
PR: 217728
Submitted by: Matt Miller <mattm916@pulsar.neomailbox.ch>
Reviewed by: jamie, cem, smh
Relnotes: yes
Differential Revision: https://reviews.freebsd.org/D10202
short, half of the memory that is allocated to implement the radix tree is
wasted because we did not change "u_daddr_t" to be a 64-bit unsigned int
when we changed "daddr_t" to be a 64-bit (signed) int. (See r96849 and
r96851.)
Reviewed by: kib, markj
Tested by: pho
MFC after: 5 days
Differential Revision: https://reviews.freebsd.org/D11028
inode number or link count for the ABI compat binaries.
Right now, and by default after the change, too large 64bit values are
silently truncated to 32 bits. Enabling the knob causes the system to
return EOVERFLOW for stat(2) family of compat syscalls when some
values cannot be completely represented by the old structures. For
getdirentries(2), knob skips the dirents which would cause non-trivial
truncation of d_ino.
EOVERFLOW error is specified by the X/Open 1996 LFS document
('Adding Support for Arbitrary File Sizes to the Single UNIX
Specification').
Based on the discussion with: bde
Sponsored by: The FreeBSD Foundation
It turns out that this is useful on hornet and wasp SoCs but it isn't
enabled in ye olde HAL /unless/ you were using a version from one of the
business units building USB targetted devices. It eventually got fixed
for all of them as people started wanting to use the USB ports on their
SoCs (eg for flash storage, bluetooth, 4G/LTE widgets, etc.)
This is actually a fix from ath9k but I'm merging it with the available-but-
disabled code in the QCA reference HAL.
Tested:
* AR9331 SoC
