Make nfs pageout coherent with the dirty state of the buffers.
Write out the dirty pages using VOP_WRITE() instead of directly
calling ncl_writerpc(). The state of the buffers now reflects the
write, fixing some hard to diagnose consistency and write order
issues. The change also allowed to remove remapping of paged out
pages into kernel space and related allocation of the phys buffer.
Reviewed by: markj, rmacklem
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
Differential revision: https://reviews.freebsd.org/D10241
Handle possible vnode reclamation after ncl_vinvalbuf() call.
ncl_vinvalbuf() might need to upgrade vnode lock, allowing the vnode
to be reclaimed by other thread. Handle the situation, indicated by
the returned error zero and VI_DOOMED iflag set, converting it into
EBADF. Handle all calls, even where the vnode is exclusively locked
right now.
Reviewed by: markj, rmacklem
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
X-Differential revision: https://reviews.freebsd.org/D10241
Add V_VMIO flag for vinvalbuf(9) to indicate that the flush request
was issued during VM-initiated i/o (pageout), so that the function
does not try to flush or remove pages or wait for the vm object
paging-in-progress counter.
Reviewed by: markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
X-Differential revision: https://reviews.freebsd.org/D10241
r304436 attempted to optimize the handling of incoming UDP packet by only
making an expensive call to in_broadcast() if the mbuf was marked as an
broadcast packet. Unfortunately, this cannot work in the case of point-to-
point L2 protocols like PPP, which have no notion of "broadcast". The
optimization has been disabled for several months now with no progress
towards fixing it, so it needs to go.
Some style fixes for vnode_pager_generic_putpages(), in the local
declaration block.
Reviewed by: markj (as part of the larger patch)
Tested by: pho (as part of the larger patch)
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
X-Differential revision: https://reviews.freebsd.org/D10241
Use int instead of boolean_t for flags argument type in
vnode_pager_generic_putpages() prototype; change the argument name to
reflect that it is flags.
Reviewed by: markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
X-Differential revision: https://reviews.freebsd.org/D10241
hyperv/hn: Fixat RNDIS rxfilter after the successful RNDIS init.
Under certain conditions on certain versions of Hyper-V, the RNDIS
rxfilter is _not_ zero on the hypervisor side after the successful
RNDIS initialization, which breaks the assumption of any following
code (well, it breaks the RNDIS API contract actually). Clear the
RNDIS rxfilter explicitly, drain packets sneaking through, and drain
the interrupt taskqueues scheduled due to the stealth packets.
Reported by: dexuan@
MFC after: 3 days
Sponsored by: Microsoft
Differential Revision: https://reviews.freebsd.org/D10230
r316487 altered the defined values of rc_force from "yes" (for yes)
and NULL (for no) to "no" (for no) and no change to the definition
of yes. Two rc.d scripts, dhclient and bgfsck check rc_force for
yesi, using test -n, and no, using test -z. The redefinition of
yes and no by r316487 caused rc.d/dhclient, when invoked by devd
using a devd.conf rule, to assign DHCP assigned IP addresses for
interfaces with statically assigned interfaces, breaking boot.
Point of breakage was at line 25 of etc/rc.d/dhclient (r301068)
where $rc_force needs to be NULL.
Always pass the linker emulation via -m when linking modules and kernels.
Previously the linker emulation was only passed when building binary
objects for firmware modules. This change always passes the desired
output format for kernel modules and kernels rather than requiring the
toolchain's default output format to match the desired output format.
This in turn permits use of external toolchains whose default output
format does not match the desired output format.
Add an implementation of __ffssi2() derived from __ffsdi2().
Newer versions of GCC include an __ffssi2() symbol in libgcc and the
compiler can emit calls to it in generated code. This is true for at
least GCC 6.2 when compiling world for mips and mips64.
cem [Tue, 4 Apr 2017 17:05:37 +0000 (17:05 +0000)]
bsdgrep(1): Fix errors with invalid expressions
Invalid expressions with an ultimate compiled pattern length of 0 (e.g.,
"grep -E {") were not taken into account and caused a segfault while trying
to fill in the good suffix table.
Submitted by: Kyle Evans <kevans91 at ksu.edu>
Reviewed by: me
Differential Revision: https://reviews.freebsd.org/D10113
cem [Tue, 4 Apr 2017 16:08:51 +0000 (16:08 +0000)]
bsdgrep(1): Rip out "xmalloc" bits
xmalloc was a debug malloc implementation, but the x{malloc,calloc,free}
functions default to calling the malloc(3) equivalents.
Instead of relying on this malloc shim, we can devise better ways to debug
malloc issues that aren't misleading upon initial inspection. (I.e., using
jemalloc's various built-in debugging capabilities.)
Submitted by: Kyle Evans <kevans91 at ksu.edu>
Reviewed by: emaste, cem
Differential Revision: https://reviews.freebsd.org/D10269
When we are doing SA lookup for TCP-MD5, check both source and
destination addresses. Previous code has used only destination address
for lookup. But for inbound packets the source address was used as SA
destination address. Thus only outbound SA were used for both directions.
Now we use addresses from a packet as is, thus SAs for both directions are
needed.
Don't calltsec_receive_intr_locked() from the error interrupt handler.
The tsec_error_intr_locked() is called with the global lock owned (e.g.
the transmit and the receive lock are both owned). We must not call
tsec_receive_intr_locked() while owning the transmit lock. The normal
receive interrupt takes care that frames are received, this is none of
the business of the error interrupt.
Submitted by: Sebastian Huber <sebastian.huber_AT_embedded-brains.de>
Timeout is now effectively a boolean rather than a time-remaining. This was
missed in r316478, but included in the original patch (mis-merged with a manual
merge).
- Set REG_NOTBOL if we've already matched beginning of line and we're
examining later parts
- For each pattern we examine, apply it to the remaining bits of the
line rather than (potentially) smaller subsets
- Check for REG_NOSUB after we've looked at all patterns initially
matching the line
- Keep track of the last match we made to later determine if we're
simply not matching any longer or if we need to proceed another byte
because we hit a zero-length match
- Match the earliest and longest bit of each line before moving the
beginning of what we match to further in the line, past the end of the
longest match; this generally matches how gnugrep(1) seems to behave,
and seems like pretty good behavior to me
- Finally, bail out of printing any matches if we were set to print all
(empty pattern) but -o (output matches) was set
Remove excessive horizontal whitespace from hier(7) by correctly
using "-width". The http://mdocml.bsd.lv/mdoc/details/width.html
says: "Do not use macros in the argument specifying the width,
that's not portable. While GNU troff can handle it, mandoc cannot."
The same problem seems to exist in many other man pages.
Fix expandsz 16.0E vals and vdev_min_asize of RAIDZ children
When a member of a RAIDZ has been replaced with a device smaller than the
original, then the top level vdev can report its expand size as 16.0E.
The reduced child asize causes the RAIDZ to have a vdev_asize lower than its
vdev_max_asize which then results in an underflow during the calculation of
the parents expand size.
Fix this by updating the vdev_asize if it shrinks, which is already
protected by a check against vdev_min_asize so should always be safe.
Also for RAIDZ vdevs, ensure that the sum of their child vdev_min_asize is
always greater than the parents vdev_min_size.
Show high (blinking foreground/bright background) background colors. Format
output and source better for this. Format output for 40 columns if there
are less than 80 columns.
Vendor changes (FreeBSD-related):
Report which extended attributes could not be restored
Update archive_read_disk.3 and archive_write_disk.3 manual pages
Plug memory leaks in xattr tests.
Remove the global variable 'info' and fix associated bugs and style bugs.
This variable was used 4 times in 1 function and all uses were wrong.
The 4 uses were in he test_frame() (show) function, to try to restore
4 colors, 2 unnecessarily and these 2 now broken. This was wrong
because it is the previous colors that must be restored, but the global
holds the original colors. Excessive setting of colors at the end
restored the previous colors correctly in most cases, but I removed
this a couple of revisions ago.
Originally, this variable had 1 correct use, to test for being on a vty
as a side effect of initializing it. This is now down in init(), and
init() also leaves a better-named global with the same contents.
Fix this by reading the current console info into a local variable in
test_frame(), as is done for several other functions. Fix style bugs
in this reading for all callers:
- extra blank lines
- all error messages different. The first one now in init() is not as
specific as the old one, but it is after a different specific one for
another ioctl and is unlikely to be reached when the first ioctl
succeeds. Ones after the first are to repeat the ioctl, so are even
more likely to be reached. The correctness of full removal of the
old global depends on the error handling for failure to initialize
it being unreachable.
- err() instead of warn() for failure in load_font(). This is almost
unreachable, and it makes no sense to continue after undoing previous
changes with revert().
- unreachable return after err() for failure in dump_screen().
Undo large renaming of local variables from the good name 'info' to the
bad name _info, which was done to protect the buggy global's bad name
from -Wshadow warnings.
Break audit_bsm_klib.c into two files: one (audit_bsm_klib.c)
retaining various utility functions used during BSM generation,
and a second (audit_bsm_db.c) that contains the various in-kernel
databases supporting various audit activities (the class and
event-name tables).
Change 25 uses of errc() to err(). 2 of these were correct. 23 used
errno for the arg so were just a verbose spelling of err(). 5 of these
were just style bugs, and 18 depended on revert() saving errno.
1 warn() also depended on revert() saving errno.
Remove 2 warnx()'s that duplicate the message from a later errx().
These used to be before returns, and should have reported errno in
some cases. errno is also not reported for for openguess() failures.
Only "restore" the video mode in revert() if the mode was just set
(necessarily partially). Setting the mode to its current setting
doesn't "restore" it, but resets it, with the least harmful change
being to clear the screen. revert() is called even for some harmless
syntax errors and usually did nothing except reset the mode. Now it
usually does nothing. The only things that it tries to restore apart
from the mode are the active vty number, the screen map, 2 colors that
only need to be restored (only after a mode change) to fix kernel bugs,
and 3 colors that can't be restored due to kernel bugs. (This is
mostly for sc, since vt doesn't support mode changes.)
revert() is not called for syntax and setting errors after [mode], so
the mode reset is only done for failures to set raster mode after
setting graphics mode. normal colors can only be set after [mode],
and that is why reverting them should be unnecessary.
r146736 added an undocumented syntax and many bugs handling it. The
documented syntax is "... [mode] [fg [bg]] [show]", where it is critical
for reducing ambiguity and keeping things simple that the mode is
parsed first. r146736 added buggy support for "... [mode] [fg [bg]]
[show] [mode] [fg [bg]]". One error was that after for failing to set
a partially-supported graphics mode, argv[optind] remains pointing to
the mode so doesn't match the first [fg [bg]], so the setting is
attempted again, with slightly worse error handling.
Fix this by removing it (support for the trailing '[mode] [fg [bg]]')
and cleaning up. The cleanups are mostly to remove convolutions and
bugs that didn't work to handle the ambiguous syntax '[fg [bg]] [fg [bg]]'
when [mode] and [show] are not present. Globals were set to allow
repeating the color settings at the end. The functions that set the
colors earlier were misnamed from set* to get*. All that they "got" is
is settings from argv. They applied the settings to the kernel and
the globals.
Fix restoration of colors in revert() by restoring 2 after the mode
change. Colors should not need to be restored, but a bug in scteken
clobbers them on any mode change, including ones for restoration. Don't
move the restoration of the other 3. Teken doesn't clobber them on
mode changes because it doesn't support them at all (sc still supports
the border color, but only using a non-teken ioctl).
Add restoration of colors after a successful mode change to work around
the scteken bug there too. The bug was previously masked by the general
setting of colors at the end.
Fix a longstanding parsing/error handling bug by exiting almost immediately
after matching the [mode] arg but failing to set the mode. Just revert
if necessary. Don't return to continue parsing but do it wrong. This
bug caused spamming the output with a usage() message and exiting with
status 1 whenever [mode] is not present bug [fg [bg]] or [show]. The
exit code 1 was actualy an ambiguous internal code for failure to match
[mode] or failure to set [mode]. This 1 was obfuscated by spelling it
EXIT_FAILURE, but actual exit codes spell EXIT_FAILURE as 1. Remove
another global which could have been used to disambiguate this but was
only used to micro-optimize the (unnecessary except for other bugs)
setting of colors at the end.
The module is designed for modification of a packets of any protocols.
For now it implements only TCP MSS modification. It adds the external
action handler for "tcp-setmss" action.
A rule with tcp-setmss action does additional check for protocol and
TCP flags. If SYN flag is present, it parses TCP options and modifies
MSS option if its value is greater than configured value in the rule.
Then it adjustes TCP checksum if needed. After handling the search
continues with the next rule.
This opcode can be used to attach some data to external action opcode.
And unlike to O_EXTERNAL_INSTANCE opcode, this opcode does not require
creating of named instance to pass configuration arguments to external
action handler. The data is coming just next to O_EXTERNAL_ACTION opcode.
The userlevel part currenly supports formatting for opcode with ipfw_insn
size, by default it expects u16 numeric value in the arg1.
Remove checks that background (bg) colors are not bright and buggy
attempts to keep them that way. The bg brightness bit is interpreted
as blinking in some modes, but it would barely be useful to disallow
setting it when it would give blinking in code which knew when that
is. The old code mostly knew this wrong, and added handling errors.
It is in fact impossible to know, since future mode switches may
change the meaning of the bit many times on the screen and in history.
Old versions of vidcontrol disallowed bg color numbers >= 8 in all
cases. This is very VGA/syscons-centric. Syscons uses the VGA defaults
of blinking fg instead of bright bg in text mode and bright bg in
graphics mode. On VGA, this is very easy to toggle at any time, and
vt blows away the VGA text mode default at boot time.
r146736 changed this to try to allow bg color numbers in graphics mode
only. This is even more VGA/syscons-centric, and there are many bugs
in this, and many nearby bugs in the parser. These are increased or
decreased by differences and bugs in vt and teken.
Perhaps the most obvious bug was that almost any vidcontrol command
which changes any color or the mode causes an error if the initial fg
color is bright. E.g., in syscons text mode, after "vidcontrol
lightwhite" to make the fg bright, another "vidcontrol lightwhite" is
rejected and buggy fixup code changes the fg to white. This is because
the bright fg color creates a bright bg color for the phantom reverse
video attribute, so was rejected. (The reverse video attribute is
phantom because teken ignores the user's setting of it and simply
reverses the fg attributes to create the bg attributes. Sometimes
some layer masks off the brightness/blinking bit, but not here.)
Perhaps the next most obvious one was that "vidcontrol lightgreen
lightblue" was misparsed as 2 settings of the fg instead of 1 setting
of the fg and 1 invalid setting of the bg. This is because the
parser supports an undocumented syntax with many parsing bugs (an
ambiguity gives this one).
I recently fix bugs in teken that broke setting of bright fg's and
bg's in the normal way. This gave more settings of then, so the old
bugs showed up more often.
Reset the DAC to 6-bit mode before calling the BIOS to set the screen
mode. This works around bugs in at least 2 Intel BIOSes for our
subsequent setting of the DAC back to 8-bit mode. The bug caused dark
(mostly 1/4-intensity) colors for all except the first setting to a
VESA graphics mode (including for settings to the current mode).
Remove restoration (with less bits) of the palette in vesa_unload()
after resetting the DAC to 6-bit mode. Depend on the BIOS to keep
the palette consistent with the DAC for the simpler reset case like
we do everywhere else in places that are actually important.
Setting the video mode should reset everything to defaults, although
we usually don't want that. Even the buggy BIOSes set the DAC to the
default 6-bit mode, and set the palette to a default that matches the
DAC. We don't undo the reset for most things, but we do undo it for
the DAC (more precisely, we change to an 8-bit DAC if possible, and
this is the only way that we set to an 8-bit DAC; it is accidental
that if the DAC was in 8-bit mode from a previous mode switch then
setting it to 8-bit mode is an undo). The buggy BIOSes are confused
by our setting of the DAC to 8-bit mode in the "undo" case. They
should multiply palette entries by 4 to match, but they actually leave
all palette entries except #2 (green) and #248-255 (unused) untouched.
Green is mysteriously scaled from 0x2a to 0x6a, and #248-255 are scaled
correctly.
Our support for the 8-bit DAC had almost no effect except to enable
bugs. Syscons barely supports 16 colors, so it doesn't benefit much
from having a palette with 16 million colors instead of only 256K.
Applications can manage the palette using FBIO_{GET,SET}PALETTE, but
the palette managed by this is only used in the less interesting modes
(text and non-truecolor graphics modes up to 8 bits wide), and the
kernel loses the changes on any mode switch (including to another vt
in a different mode).
As noted by bde@ negative tv_sec values are not checked for overflow,
so overflow can still occur. Fix that. Also remove the extra check for
tv_sec size as under COMPAT_LINUX32 it is always true.
Add a helper function to get system reference clock
Many devices are clocked from the SoC's platform clock / 2. Some device nodes
include their own clock-frequency property, while others are dependent on the
SoC's bus-frequency property instead. To simplify, add a helper function to get
this clock.