Dimitry Andric [Tue, 2 Aug 2016 20:25:22 +0000 (20:25 +0000)]
Fix a segfault in bsdgrep when parsing the invalid extended regexps "?"
or "+" (these are invalid, because there is no preceding operand).
When bsdgrep attempts to emulate GNU grep in discarding and ignoring the
invalid ? or + operators, some later logic in tre_compile_fast() goes
beyond the end of the buffer, leading to a crash.
Fix this by bailing out, and reporting a bad pattern instead.
Conrad Meyer [Tue, 2 Aug 2016 18:13:50 +0000 (18:13 +0000)]
proc_init: Fix a few memory leaks of 'phdl'
In the normal case and correct failure cases, the 'phdl' pointer is passed to
callers to use or clean up as needed. However, some failure cases returned
early, failing to export the phdl pointer.
This was introduced in the restructuring of r303533.
sed(1): Revert r303047 "cleanup" and therefore r303572.
While big, the change was meant to have no effect on behavior and instead
so far we have found two regressions: one in the etcupdate tests and
another one in the games/openttd port[1].
Revert to a known working state. We will likely have to split the patch in
functional parts before bringing back the changes.
Conrad Meyer [Mon, 1 Aug 2016 23:07:31 +0000 (23:07 +0000)]
rtentry: Initialize rt_mtx with MTX_NEW
The "rtentry" zone does not use UMA_ZONE_ZINIT, so it is invalid to assume the
mutex's memory will be zero. Without MTX_NEW, garbage backing memory may
trigger the "re-initializing a mutex" assertion.
PR: 200991
Submitted by: Chang-Hsien Tsai <luke.tw AT gmail.com>
John Baldwin [Mon, 1 Aug 2016 22:39:51 +0000 (22:39 +0000)]
Store the offset of the KDOORBELL and GTS registers in the softc.
VF devices use a different register layout than PF devices. Storing
the offset in a value in the softc allows code to be shared between the
PF and VF drivers.
Mark Johnston [Mon, 1 Aug 2016 22:22:11 +0000 (22:22 +0000)]
ipoib: Bound the number of egress mbufs buffered during pathrec lookups.
In pathological situations where the master subnet manager becomes
unresponsive for an extended period, we may otherwise end up queuing all
of the system's mbufs while waiting for a response to a path record lookup.
This addresses the same issue as commit 1e85b806f9 in Linux.
John Baldwin [Mon, 1 Aug 2016 22:19:23 +0000 (22:19 +0000)]
Disable PCI hotplug support for slots with power controllers.
After further review of the spec, I do not think the current HotPlug
code handles slots with power controllers correctly. In particular,
the power state of the slot is to be inferred from other events, not
from examining the state of the power control bit in SLOT_CTL. For now,
disable PCI hotplug support on such slots.
PR: 211081
Tested by: Jeffrey E Pieper <jeffrey.e.pieper@intel.com>
MFC after: 3 days
Mateusz Guzik [Mon, 1 Aug 2016 21:48:37 +0000 (21:48 +0000)]
Implement trivial backoff for locking primitives.
All current spinning loops retry an atomic op the first chance they get,
which leads to performance degradation under load.
One classic solution to the problem consists of delaying the test to an
extent. This implementation has a trivial linear increment and a random
factor for each attempt.
For simplicity, this first thouch implementation only modifies spinning
loops where the lock owner is running. spin mutexes and thread lock were
not modified.
Current parameters are autotuned on boot based on mp_cpus.
Autotune factors are very conservative and are subject to change later.
Sean Bruno [Mon, 1 Aug 2016 21:19:51 +0000 (21:19 +0000)]
r293331 mistakingly failed to add an assignment of paddr to the rxbuf
but only in the NETMAP code. This lead to the NETMAP code paths
passing nothing up to userland.
Submitted by: Ad Schellevis <ad@opnsense.org>
Reported by: Franco Fichtner <franco@opnsense.org>
MFC after: 1 day
Do not invoke resize event if initial disk size is zero. Some disks
report the size only after first opening. And due to the events are
asynchronous, some consumers can receive this event too late and
this confuses them. This partially restores previous behaviour, and
at the same time this should fix the problem, when already opened
provider loses resize event.
This addresses a regression from an earlier upstream change which caused
cma_acquire_dev() to bypass the port GID cache and instead query the HCA
for each entry in its GID table. These queries can become extremely slow on
multiport devices, which has a negative impact on connection setup times.
Dimitry Andric [Mon, 1 Aug 2016 19:49:42 +0000 (19:49 +0000)]
Fix non-functional bsdinstall services dialog.
The most recent version of bsdinstall does not seem to respect any of
the checkboxes in the "Choose the services you would like to be started
at boot" dialog. None of the chosen services end up in the rc.conf file
that is installed onto the target system.
This is caused by the bsdinstall/scripts/hardening script, which
implements the new hardening options dialog. The script starts by
overwriting the previously written rc.conf.services file:
echo -n > $BSDINSTALL_TMPETC/rc.conf.services
which is obviously incorrect. It should clear out rc.conf.hardening
instead.
Allan Jude [Mon, 1 Aug 2016 19:37:43 +0000 (19:37 +0000)]
Make boot code and loader check for unsupported ZFS feature flags
OpenZFS uses feature flags instead of a zpool version number to track
features since the split from Oracle. In addition to avoiding confusion
on ZFS vs OpenZFS version numbers, this also allows features to be added
to different operating systems that use OpenZFS in different order.
The previous zfs boot code (gptzfsboot) and loader (zfsloader) blindly
tries to read the pool, and if failed provided only a vague error message.
With this change, both the boot code and loader check the MOS features
list in the ZFS label and compare it against the list of features that
the loader supports. If any unsupported feature is active, the pool is
not considered as a candidate for booting, and a helpful diagnostic
message is printed to the screen. Features that are merely enabled via
zpool upgrade, but not in use, do not block booting from the pool.
Submitted by: Toomas Soome <tsoome@me.com>
Reviewed by: delphij, mav
Relnotes: yes
Differential Revision: https://reviews.freebsd.org/D6857
Alan Cox [Mon, 1 Aug 2016 17:25:07 +0000 (17:25 +0000)]
Restore the historical behavior of "sysctl vm.swap_idle_enabled=1". Prior
to r254304, we had separate functions for reclamation and laundering
(vm_pageout_scan) versus updating usage information, i.e., "reference
bits", on active pages (vm_pageout_page_stats), and we only performed
vm_req_vmdaemon(VM_SWAP_IDLE) if vm_pages_needed was true. However, since
r254303, if vm_swap_idle_enabled was "1", we have performed
vm_req_vmdaemon(VM_SWAP_IDLE) regardless of whether we are short of free
pages. This was unintended and too aggressive, so I suspect no one uses
this feature. With this change, we restore the historical behavior and
only perform vm_req_vmdaemon(VM_SWAP_IDLE) when we are short of free
pages.
Andrew Gallatin [Mon, 1 Aug 2016 17:02:21 +0000 (17:02 +0000)]
Rework IPV6 TCP path MTU discovery to match IPv4
- Re-write tcp_ctlinput6() to closely mimic the IPv4 tcp_ctlinput()
- Now that tcp_ctlinput6() updates t_maxseg, we can allow ip6_output()
to send TCP packets without looking at the tcp host cache for every
single transmit.
- Make the icmp6 code mimic the IPv4 code & avoid returning
PRC_HOSTDEAD because it is so expensive.
Without these changes in place, every TCP6 pmtu discovery or host
unreachable ICMP resulted in a call to in6_pcbnotify() which walks the
tcbinfo table with the write lock held. Because the tcbinfo table is
shared between IPv4 and IPv6, this causes huge scalabilty issues on
servers with lots of (~100K) TCP connections, to the point where even
a small percent of IPv6 traffic had a disproportionate impact on
overall throughput.
An old tables implementation had all tables preallocated,
so when user did `ipfw table N flush` it always worked, but now
when table N doesn't exist the kernel returns ESRCH error.
This isn't fatal error for flush and destroy commands. Do not
call err(3) when errno is equal to ESRCH. Also warn only when
quiet mode isn't enabled. This fixes a regression in behavior,
when old rules are loaded from file.
Also use correct value for switch in the table_swap().
Andrew Turner [Mon, 1 Aug 2016 12:17:44 +0000 (12:17 +0000)]
Add a kernel variable to let the user to select their preferred order
between ACPI and FDT. This will be needed on machines with both, e.g. the
SoftIron Overdrive 3000. The kernel will accept one or more comma separated
values of either 'acpi' or 'fdt'. Any other values are skipped.
To set it the user can either set it on the loader command line, or
in loader.conf e.g. in loader.conf:
kern.cfg.order=acpi,fdt
This will try using ACPI then FDT. If none of the selected options work the
kernel tries to use one to get the serial console, then panics.
Reviewed by: emaste (earlier version)
Obtained from: ABT Systems Ltd
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D7274
Julian Elischer [Mon, 1 Aug 2016 11:34:12 +0000 (11:34 +0000)]
slite style changes. There is an incoming patch that rewrites a
lot of this module and I want to get the style and whitespace changes in
a separate commit (or maybe more).
Pedro F. Giffuni [Sun, 31 Jul 2016 21:43:43 +0000 (21:43 +0000)]
indent(1): Rearrange option parsing code to squelch clang's static analyzer.
clang-analyzer complained that eqin() sets file-scoped pointer param_start
to point into char buffer defined in scan_profile(), and once
scan_profile() exits, param_start is a "dangling reference". param_start
was never used afterwards, but it's cleaner to move it to set_option()
which is the only branch where param_start is needed.
Pedro F. Giffuni [Sun, 31 Jul 2016 21:09:22 +0000 (21:09 +0000)]
indent(1): Untangle the connection between pr_comment.c and io.c.
It's pr_comment.c that should decide whether to put a "star comment
continuation" or not. This duplicates code a bit, but it simplifies
pr_comment() at the same time since pr_comment() no longer has to "signal"
whether a star continuation is needed or not.
This change requires indent(1) to not wrap comment lines that lack a blank
character, but I think it's for the better if you look at cases when that
happens (mostly long URIs and file system paths, which arguably shouldn't
be wrapped).
It also fixes two bugs:
1. Cases where asterisk is a part of the comment's content (like in "*we*
are the champions") and happens to appear at the beginning of the line,
misleading dump_line() into thinking that this is part of the star comment
continuation, leading to misalignment.
2. Cases where blank starred lines had three too many characters on the
line when wrapped.
Pedro F. Giffuni [Sun, 31 Jul 2016 20:13:00 +0000 (20:13 +0000)]
indent(1): Fix wrapping of some lines in comments.
After a blank line was printed (to separate paragraphs in comments), the
next line was sometimes wrapped to the column at which the previous
non-empty line ended. The fix is to reset the last blank pointer (last_bl)
on newline.
Pedro F. Giffuni [Sun, 31 Jul 2016 20:04:18 +0000 (20:04 +0000)]
indent(1): Simplify pr_comment().
Modify count_spaces() to take a third parameter "end" that will make the
function return when the end is reached. This lets the caller pass a
pointer to non nul-terminated sequence of characters. Rename
count_spaces() to count_spaces_until() and reinstate count_spaces(), this
time based on count_spaces_until().
Use count_spaces_until() to recalculate current column when going through
a comment just before the fragment which decides if current line of the
comment should be wrapped. This move simplifies this code by eliminating
the need for keeping the column counter up to date every time e_com is
advanced and also reduces spread of code that has to know how many columns
a tab will produce.
Deduplicate code that decided if a comment needs a blank line at the top.
Pedro F. Giffuni [Sun, 31 Jul 2016 15:02:53 +0000 (15:02 +0000)]
indent(1): Remove dead code relating to unix-style comments.
The original indent(1) described unix-style comments as similar to box
comments, except the first non-blank character on each line is lined up
with the '*' of the "/*" which appears on a line by itself.
The code has been turned off for ages and -sc/-nsc make it even
less relevant.
The current implementation uses non-temporal writes. This turns out to
be detrimental to performance if the page is used shortly after, which
is the typical case with page faults.
Ed Schouten [Sun, 31 Jul 2016 08:05:15 +0000 (08:05 +0000)]
Fix up setgrent(3) to have a POSIX-compliant prototype.
Just like with freelocale(3), I haven't been able to find any piece of
code that actually makes use of this function's return value, both in
base and in ports. The reason for this is that FreeBSD seems to be the
only operating system to have such a prototype. This is why I'm deciding
to not use symbol versioning for this.
It does seem that the pw(8) utility depends on the function's typing and
already had a switch in place to toggle between the FreeBSD and POSIX
variant of this function. Clean this up by always expecting the POSIX
variant.
There is also a single port that has a couple of local declarations of
setgrent(3) that need to be patched up. This is in the process of being
fixed.
Pedro F. Giffuni [Sun, 31 Jul 2016 04:58:06 +0000 (04:58 +0000)]
indent(1): Bail out if there's no more space on the parser stack.
Also increase the stack size still keeping a conservative value of 256.
This is based on a similar changes done for PostgreSQL which instead
uses a stack size of 1000.
Differential Revision: https://reviews.freebsd.org/D6966 (Partial)
Submitted by: Piotr Stefaniak (with changes)
Pedro F. Giffuni [Sun, 31 Jul 2016 04:14:20 +0000 (04:14 +0000)]
indent(1): Consistently indent declarations.
This fixes a very visible issue that may be hidden by some indent.pro
settings as in the example from FreeBSD's /usr/share.
From Piotr's log:
____
To prevent losing tabs from indentation in declarations, FreeBSD indent's
r125624 added code for the most common case when it's an identifier that
is indented, but didn't do anything with the original code that did the
same for any other cases. The other cases are: lparens (function pointer
declaration), asterisks (pointer declaration), stray semicolons, and
commas leading identifiers instead of trailing them.
Use the code added in r125624 (and improved in later commits) to write a
new function indent_declaration() and use it in all places that meant to
indent declarations. In order to indent only once per line, reuse existing
ps.dumped_decl_indent variable that was only used when formatting for
troff (-troff) until now.
____
Both C99 and POSIX directly prohibits any standard function to set errno
to 0. Breaking this rule in 2001 NetBSD hack was imported which attempts
to workaround very limited glob() return codes amount. Use POSIX-compatible
workaround now with E2BIG which can't comes from other functions used
instead of prohibited 0.
Allan Jude [Sat, 30 Jul 2016 17:53:37 +0000 (17:53 +0000)]
Improve boot loader quote parsing
parse() is the boot loader's interp_parse.c is too naive about quotes
both single and double quotes were allowed to be mixed, and single
quotes did not follow the usual semantics (re variable expansion).
The old code did not check for terminating quotes
This update implements:
* distinguishing single and double quote
* variable expansion will not be done inside single quote protected area
* will preserve inner quote for values like "value 'some list'"
* ending quote check.
this diff does not implement ending quote order check, it shouldn't
be too hard, needs some improvements on parser state machine.
PR: 204602
Submitted by: Toomas Soome <tsoome@me.com>
Relnotes: yes
Differential Revision: https://reviews.freebsd.org/D6000