Replace global swhash in swap pager with per-object trie to track swap
blocks assigned to the object pages.
- The global swhash_mtx is removed, trie is synchronized by the
corresponding object lock.
- The swp_pager_meta_free_all() function used during object
termination is optimized by only looking at the trie instead of
having to search whole hash for the swap blocks owned by the object.
- On swap_pager_swapoff(), instead of iterating over the swhash,
global object list have to be inspected. There, we have to ensure
that we do see valid trie content if we see that the object type is
swap.
Sizing of the swblk zone is same as for swblock zone, each swblk maps
SWAP_META_PAGES pages.
John Baldwin [Fri, 25 Aug 2017 23:10:27 +0000 (23:10 +0000)]
Don't grab SOCK_LOCK for soref() when queuing an AIO request.
The AIO job holds a reference on the associated file descriptor, so the
socket's count should already be > 0. This fixes a LOR with the socket
buffer lock after recent socket locking changes in HEAD.
Sean Bruno [Fri, 25 Aug 2017 22:38:55 +0000 (22:38 +0000)]
Add a different #define for the maximum number of transmit and
recieve descriptors for the igb(4) class of devices. This will
allow a better definition for maximum going forward. Some igb(4)
devices support more than the default 4K.
Reported by: Jason (j@nitrology.com)
Sponsored by: Limelight Networks
Warner Losh [Fri, 25 Aug 2017 21:38:43 +0000 (21:38 +0000)]
Fill in reserved areas from NVMe spec in the IDENTIFY structure
(struct nvme_controller_data) as defined in the NVM Express
specification, revsion 1.3.
Warner Losh [Fri, 25 Aug 2017 21:38:29 +0000 (21:38 +0000)]
Add feature codes from NVMe 1.3 specification:
o Automomous Power State Transition
o Host Memory Buffer
o Timestamp
o Keep Alive Timer
o Host Controlled Thermal Management
o Non-Operational Power State Config
Also note that feature codes 0x78-0x7f are reserved for the NVMe
Management Interface.
Alan Cox [Fri, 25 Aug 2017 18:47:23 +0000 (18:47 +0000)]
Correct a regression in the previous change, r322459. Specifically, the
removal of the "blk" parameter from blst_meta_alloc() had the unintended
effect of generating an out-of-range allocation when the cursor reaches
the end of the tree if the number of managed blocks in the tree equals
the so-called "radix" (which in the blist code is not the standard notion
of what a radix is but rather the maximum number of leaves in a tree of
the current height.) In other words, only certain swap configurations
were affected, which is why earlier testing did not reveal the problem.
Conrad Meyer [Fri, 25 Aug 2017 16:38:21 +0000 (16:38 +0000)]
getmntinfo(3): Scale faster, and return sooner
getmntinfo(3) is designed around a relatively static or slow growing set of
current mounts. It tried to detect a race with somewhat concurrent mount
and re-call getfsstat(2) in that case, looping indefinitely. It also
allocated space for a single extra mount as slop.
In the case where the user has a large number of mounts and is adding them
at a rapid pace, it fell over.
This patch makes two functional changes:
1. Allocate even more slop. Double whatever the last getfsstat(2) returned.
2. Abort and return some known results after looping a few times
(arbitrarily, 3). If the list is constantly changing, we can't guarantee
we return a full result to the user at any point anyways.
While here, add very basic functional tests for getmntinfo(3) to the libc
suite.
PR: 221743
Submitted by: Peter Eriksson <peter AT ifm.liu.se> (earlier version)
Sponsored by: Dell EMC Isilon
Benno Rice [Fri, 25 Aug 2017 16:10:16 +0000 (16:10 +0000)]
Replace makefs' hand-rolled unescaping with strunvis
mtree path names and link attributes are encoded, generally using strvis. Newer
versions of mtree will use C-style escapes but previously the accepted form was
octal escapes. makefs' mtree code spots the C-style escapes but fails to deal
with octal escapes correctly.
Remove mtree's escape-decoding code (except for a few instances where it's
needed) and instead pass pathnames and link targets through strunvis prior to
use.
Alan Somers [Fri, 25 Aug 2017 15:31:55 +0000 (15:31 +0000)]
dd(1): Incorrect casting of arguments
dd(1) casts many of its numeric arguments from uintmax_t to intmax_t and
back again to detect whether or not the original arguments were negative.
This is not correct, and causes problems with boundary cases, for example
when count is SSIZE_MAX-1.
Ed Schouten [Fri, 25 Aug 2017 11:01:39 +0000 (11:01 +0000)]
Sync CloudABI compatibility against the latest upstream version (v0.13).
With Flower (CloudABI's network connection daemon) becoming more
complete, there is no longer any need for creating any unconnected
sockets. Socket pairs in combination with file descriptor passing is all
that is necessary, as that is what is used by Flower to pass network
connections from the public internet to listening processes.
Remove all of the kernel bits that were used to implement socket(),
listen(), bindat() and connectat(). In principle, accept() and
SO_ACCEPTCONN may also be removed, but there are still some consumers
left.
Bruce Evans [Fri, 25 Aug 2017 10:57:17 +0000 (10:57 +0000)]
Fix bugs in (mostly) not-yet-activated parts of early/emergency output:
- map the hard-coded frame buffer address above KERNBASE. Using the
physical address only worked because of larger mapping bugs.
The hard-coded frame buffer address only works on x86. Use messy ifdefs
to try to avoid warnings about unused code for other arches.
- remove the sysctl for reading and writing the table kernel console
attributes. Writing only worked for emergency output since normal
output uses unalterd copies.
- fix the test for the emergency console being usable
- explain why a hard-coded attribute is used very early. Emergency output
works on x86 even before the pcpu pointer is initialized.
Bruce Evans [Fri, 25 Aug 2017 07:04:41 +0000 (07:04 +0000)]
Support setting the colors of cursors for the VGA renderer.
Advertise this by changing the defaults to mostly red. If you don't like
this, change them (almost) back using:
vidcontrol -c charcolors,base=7,height=0
vidcontrol -c mousecolors,base=0[,height=15]
The (graphics mode only) mouse cursor colors were hard-coded to a black
border and lightwhite interior. Black for the border is the worst
possible default, since it is the same as the default black background
and not good for any dark background. Reversing this gives the better
default of X Windows. Coloring everything works better still. Now
the coloring defaults to a lightwhite border and red interior.
Coloring for the character cursor is more complicated and mode
dependent. The new coloring doesn't apply for hardware cursors. For
non-block cursors, it only applies in graphics mode. In text mode,
the cursor color was usually a hard-coded (dull)white for the background
only, unless the foreground was white when it was a hard-coded black
for the background only, unless the foreground was white and the
background was black it was reverse video. In graphics mode, it was
always reverse video for the block cursor. Reverse video is worse,
especially over cutmarking regions, since cutmarking still uses simple
reverse video (nothing better is possible in text mode) and double
reverse video for the cursor gives normal video. Now, graphics mode
uses the same algorithm as the best case for text mode in all cases
for graphics mode. The hard-coded sequence { white, black, } for the
background is now { red, white, blue, } where the first 2 colors can
be configured. The blue color at the end is a sentinel which prevents
reverse video being used in most cases but breaks the compatibility
setting for white on black and black on white characters. This will
be fixed later. The compatibility setting is most needed for mono modes.
The previous commit to syscons.c changed sc_cnterm() to be more careful.
It followed null pointers in some cases. But sc_cnterm() has been
unreachable for 15+ years since changes for multiple consoles turned
off calls to the the cnterm destructor for all console drivers. Before
them, it was only called at boot time. So no driver with an attached
console has ever been unloadable and not even the non-console destructors
have been tested much.
Warner Losh [Fri, 25 Aug 2017 03:15:18 +0000 (03:15 +0000)]
Enable bus mastering on the device before resetting the device. The
card has to do PCIe transactions to complete the reset process, but
can't do them, per the PCIe spec, unless bus mastering is enabled.
Bruce Evans [Fri, 25 Aug 2017 02:37:32 +0000 (02:37 +0000)]
Fix missing switching of the terminal emulator when switching the
terminal state for kernel console output.
r56043 in 2000 added many complications to support dynamic selection
of the terminal emulator using modules and the ioctl CONS_SETTERM.
This was never completed. There are still no modules, but it is easy
to restore the scterm and dumb emulators at compile time. Then
boot-time configuration for the preferred one doesn't work right, but
CONS_SETTERM almost works after fixing this bug. CONS_SETTERM only
switches the emulator for the user state, leaving the kernel state(s)
still using the boot-time emulator. The fix is especially important
when switching from sc to scteken, since the scteken state has pointers
in it.
Eric Joyner [Thu, 24 Aug 2017 22:56:22 +0000 (22:56 +0000)]
ixv(4): Add more robust mailbox API negotiation
The previous update to the driver to 3.2.12-k changed the VF's API version
to 1.2, but did not let the VF fall back to 1.1 or 1.0 versions. So, this
patch tries 1.2 first, then the older versions in succession if that fails.
This should allow the VF driver to negotiate 1.1 and work with older PF
drivers, such as the one used in Amazon's EC2 service.
Warner Losh [Thu, 24 Aug 2017 22:11:10 +0000 (22:11 +0000)]
Expand the latency tracking array from 1.024s to 8.192s to help track
extreme outliers from dodgy drives. Adjust comments to reflect this,
and make sure that the number of latency buckets match in the two
places where it matters.
Warner Losh [Thu, 24 Aug 2017 22:10:58 +0000 (22:10 +0000)]
Fix 32-bit overflow on latency measurements
o Allow I/O scheduler to gather times on 32-bit systems. We do this by shifting
the sbintime_t over by 8 bits and truncating to 32-bits. This gives us 8.24
time. This is sufficient both in range (256 seconds is about 128x what current
users need) and precision (60ns easily meets the 1ms smallest bucket size
measurements). 64-bit systems are unchanged. Centralize all the time math so
it's easy to tweak tha range / precision tradeoffs in the future.
o While I'm here, the I/O scheduler should be using periph_data rather than
sim_data since it is operating on behalf of the periph.
Gleb Smirnoff [Thu, 24 Aug 2017 20:52:02 +0000 (20:52 +0000)]
Add a test case for a connection on accept queue that is reset before
it is accepted. In that case accept(2) shall return ECONNABORTED.
Accept filters provide help with easily replicating that case.
Gleb Smirnoff [Thu, 24 Aug 2017 20:49:19 +0000 (20:49 +0000)]
Third take on the r319685 and r320480. Actually allow for call soisconnected()
via soisdisconnected(), and in the earlier unlock earlier to avoid lock
recursion.
This fixes a situation when a socket on accept queue is reset before being
accepted.
Reported by: Jason Eggleston <jeggleston llnw.com>
Alan Somers [Thu, 24 Aug 2017 19:48:41 +0000 (19:48 +0000)]
zfsd(8): Close a race condition when onlining a disk paritition
When inserting a partitioned disk, devfs and geom will announce the whole
disk before they announce the partition. If the partition containing ZFS
extends to one of the disk's extents, then zfsd will see a ZFS label on the
whole disk and attempt to online it. ZFS is smart enough to activate the
partition instead of the whole disk, but only if GEOM has already created
the partition's provider.
cddl/contrib/opensolaris/lib/libzfs/common/libzfs.h
cddl/contrib/opensolaris/lib/libzfs/common/libzfs_import.c
Add a zpool_read_all_labels method. It's similar to
zpool_read_label, but it will return the number of labels found.
cddl/usr.sbin/zfsd/zfsd_event.cc
When processing a DevFS CREATE event, only online a VDEV if we can
read all four ZFS labels.
Stop masking FSGSBASE and SMEP features under monitors.
Not enabling FSGSBASE in %cr4 does not prevent reporting of the
feature by the CPUID instruction (blame Int*l). As result, kernels
which were run under monitors pretended that usermode cannot modify
TLS base without the syscall, while libc noted right combination of
capable CPU and the new kernel version, trying to use the WRFSBASE
instruction.
Really old hypervisors that cannot handle enablement of these features
in %cr4 would require the manual configuration, by setting the loader
tunable hw.cpu_stdext_disable=0x81
Reported by: lwhsu, mjoras
Sponsored by: The FreeBSD Foundation
MFC after: 18 days
Kyle Evans [Thu, 24 Aug 2017 01:23:33 +0000 (01:23 +0000)]
bsdgrep: add a primitive literal matcher
fgrep/grep -F will error out at runtime if compiled with a regex(3)
that does not define REG_NOSPEC or REG_LITERAL. glibc is one such regex(3)
implementation, and as it turns out they don't support literal matching at
all.
Provide a primitive literal matcher for use with glibc and other
implementations that don't support literal matching so that we don't
completely lose fgrep/grep -F if compiled against libgnuregex on stable/10,
stable/11, or other systems that we don't necessarily support.
This is a wholly unoptimized implementation with no plans to optimize it as
of now. This is due to both its use-case being primarily on unsupported
systems in the near-distant future and that it's reinventing the wheel that
we already have available as a feature of regex(3).
Kyle Evans [Thu, 24 Aug 2017 01:20:52 +0000 (01:20 +0000)]
bsdgrep: add some additional tests for fgrep
Previously added tests only check that fgrep is somewhat sane and works. Add
some more tests that check that the implementation is basically functional
and not producing incorrect results with various flags.
John Baldwin [Wed, 23 Aug 2017 23:30:25 +0000 (23:30 +0000)]
Improve the coverage of debug symbols for MK_DEBUG_FILES.
- Include debug symbols in static libraries. This permits binaries
to include debug symbols for functions obtained from static libraries.
- Permit the C/C++ compiler flags added for MK_DEBUG_FILES to be
overridden by setting DEBUG_FILES_CFLAGS. Use this to limit the debug
information for llvm libraries and binaries.
Ed Maste [Wed, 23 Aug 2017 17:56:55 +0000 (17:56 +0000)]
top: use __mips__ and __NetBSD__ for consistency
r322767 fixed the mips64 build failure with Clang with a minimal change
to use __FreeBSD__ instead of FreeBSD in a #if test. For consistency
and to facilitate possible upstreaming change the other macros in the
test to their canonical form.
Ed Maste [Wed, 23 Aug 2017 12:47:10 +0000 (12:47 +0000)]
Set MK_LLD_IS_LD to MK_LLD_BOOTSTRAP for cross-tools
LLD_BOOTSTRAP is intended to control the linker used to link world and
kernel, while LLD_IS_LD is intended to control the linker installed in
that world.
Force LLD_IS_LD equal to LLD_BOOTSTRAP for the cross-tools build and
install phase, so that lld will be installed as the ld to run on the
host, when LLD_BOOTSTRAP is set.
PR: 221543
Reviewed by: dim
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D12072
Add new mlx5ib(4) driver to the kernel source tree which supports
Remote DMA over Converged Ethernet, RoCE, for the ConnectX-4 series of
PCI express network cards.
There is currently no user-space support and this driver only supports
kernel side non-routable RoCE V1. The krping kernel module can be used
to test this driver. Full user-space support including RoCE V2 will be
added as part of the ongoing upgrade to ibcore from Linux 4.9. Otherwise
this driver is feature equivalent to mlx4ib(4). The mlx5ib(4) kernel
module will only be built when WITH_OFED=YES is specified.
Ed Maste [Tue, 22 Aug 2017 17:57:34 +0000 (17:57 +0000)]
newvers.sh: accommodate `git worktree`
newvers.sh looks for a .vcs subdirectory (e.g. .git, .svn) to determine
which vcs info tool to run (e.g., git rev-parse, svn info).
(As of r308789 if a .vcs subdirectory is not found at ${TOPDIR} then
newvers.sh walks up successive parent directories, testing for the .vcs
subdirectory at each step. This is done in case the FreeBSD source is
built in a subdirectory as part of some larger project, but either way
newvers.sh still tests for the .vcs subdirectory.)
However, when using git worktree there is no .git subdirectory but
rather a plain text .git file which contains a reference to the main
working tree.
Change findvcs() to test that the .vcs entry exists, regardless of type.
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Conrad Meyer [Tue, 22 Aug 2017 00:10:15 +0000 (00:10 +0000)]
subr_smp: Clean up topology analysis, add additional layers
Rather than repeatedly nesting loops, separate concerns with a single loop
per call stack level. Use a table to drive the recursive routine. Handle
missing topology layers more gracefully (infer a single unit).
Analyze some additional optional layers which may be present on e.g. AMD Zen
systems (groups, aka dies, per package; and cachegroups, aka CCXes, per
group).
Display that additional information in the boot-time topology information,
when it is relevent (non-one).
Mark Johnston [Mon, 21 Aug 2017 21:56:02 +0000 (21:56 +0000)]
Fix an off-by-two in the llquantize() action parameter validation.
The aggregation created by llquantize() partitions values into buckets; the
lower bound of the bucket containing the largest values is b^{m+1}, where
b and m are the second and fourth parameters to the action, respectively.
Bucket bounds are stored in a 64-bit integer, and so the llquantize()
validation checks need to verify that b^{m+1} fits in 64 bits. However, it
was only verifying that b^{m-1} fits in 64 bits, so certain parameter
combinations could trigger assertion failures in libdtrace.
Upgrade FW to 5.4.66
sysctls to display stats, stats polled every 2 seconds
Modify QLA_LOCK()/QLA_UNLOCK() to not sleep after acquiring mtx_lock
Add support to turn OFF/ON error recovery following heartbeat failure for
debug purposes.
Set default max values to 32 Tx/Rx/SDS rings
Glen Barber [Mon, 21 Aug 2017 20:23:05 +0000 (20:23 +0000)]
Apply changes from bin/chmod/tests/chmod_test.sh (r321949, r321950,
and r322101), adding atf_expect_fail() before chflags(8) is invoked
if the filesystem is ZFS, which does not support UF_IMMUTABLE.
MFC after: 3 days
Sponsored by: The FreeBSD Foundation
Andrew Turner [Mon, 21 Aug 2017 18:12:32 +0000 (18:12 +0000)]
Improve the performance of the arm64 thread switching code.
The full system memory barrier around a TLB invalidation is stricter than
required. It needs to wait on accesses to main memory, with just the weaker
store variant before the invalidate. As such use the dsb istst, tlbi, dlb
ish sequence already used in pmap.
The tlbi instruction in this sequence is also unnecessarily using a
broadcast invalidate when it just needs to invalidate the local CPUs TLB.
Switch to a non-broadcast variant of this instruction.
Make WRFSBASE and WRGSBASE instructions functional.
Right now, we enable the CR4.FSGSBASE bit on CPUs which support the
facility (Ivy and later), to allow usermode to read fs and gs bases
without syscalls. This bit also controls the write access to bases
from userspace, but WRFSBASE and WRGSBASE instructions currently
cannot be used, because return path from both exceptions or interrupts
overrides bases with the values from pcb.
Supporting the instructions is useful because this means that usermode
can implement green-threads completely in userspace without issuing
syscalls to change all of the machine context.
Support is implemented by saving the fs base and user gs base when
PCB_FULL_IRET flag is set. The flag is set on the context switch,
which potentially causes clobber of the bases due to activation of
another context, and when explicit modification of the user context by
a syscall or exception handler is performed. In particular, the patch
moves setting of the flag before syscalls change context.
The changes to doreti_exit and PUSH_FRAME to clear PCB_FULL_IRET on
entry from userspace can be considered a bug fixes on its own.
Avoid dereferencing potentially freed workitem in
softdep_count_dependencies().
Buffer's b_dep list is protected by the SU mount lock. Owning the
buffer lock is not enough to guarantee the stability of the list.
Calculation of the UFS mount owning the workitems from the buffer must
be much more careful to not dereference the work item which might be
freed meantime. To get to ump, use the pointers chain which does not
involve workitems at all.
Reported and tested by: pho
Reviewed by: mckusick
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
When a security policy should match TCP connection with specific ports,
the SYN+ACK segment send by syncache_respond() is considered as forwarded
packet, because at this moment TCP connection does not have PCB structure,
and ip_output() is called without inpcb pointer. In this case SPIDX filled
for SP lookup will not contain TCP ports and security policy will not
be found. This can lead to unencrypted SYN+ACK on the wire.
This patch restores the old behavior, when ports will not be filled only
for forwarded packets.
Fix for deadlock situation in the LinuxKPI's RCU synchronize API.
Deadlock condition:
The return value of TDQ_LOCKPTR(td) is the same for two threads.
1) The first thread signals a wakeup while keeping the rcu_read_lock().
This invokes sched_add() which in turn will try to lock TDQ_LOCK().
2) The second thread is calling synchronize_rcu() calling mi_switch() over
and over again trying to yield(). This prevents the first thread from running
and releasing the RCU reader lock.
Solution:
Release the thread lock while yielding to allow other threads to acquire the
lock pointed to by TDQ_LOCKPTR(td).