Tomohiro Kusumi [Tue, 1 Mar 2016 16:38:21 +0000 (01:38 +0900)]
sbin/hammer: Cleanup and squash elm format of hammer show
Fix possible trailing whitespace after object type field.
Record information was taking too many lines for no reason.
Squash three lines (ot=, dataoff=, fill=) into one.
Record type dependent part is printed from the next line.
Tomohiro Kusumi [Sat, 5 Mar 2016 05:44:21 +0000 (14:44 +0900)]
sbin/hammer: Add obfuscate option to hammer show
Add an option not to print directory entry name so that users
can send hammer show output to someone else if necessary
without making public their name. Most users never use hammer
show, but some users do and see errors in it.
By default hammer show printfs directory entry name like
dir-entry ino=
00000001007a933b lo=
00000000 name="hammer.c"
but by using this option it would be like
dir-entry ino=
00000001007a933b lo=
00000000
Note that hammer show command is a subcommand of /sbin/hammer,
so this option is not a part of getopt(3) options. It needs
to be used as follows if one does not need to filter btree.
# hammer -f /dev/da1:/dev/da2:/dev/da3 show none none obfuscate
Also note that this option is not written in hammer(8) manpage
just like filter/nofilter option is not written.
Tomohiro Kusumi [Wed, 2 Mar 2016 13:56:28 +0000 (22:56 +0900)]
sbin/hammer: Use sizeof(*elm)
Tomohiro Kusumi [Tue, 1 Mar 2016 13:28:45 +0000 (22:28 +0900)]
sbin/hammer: Use HAMMER_OFF_BAD
Tomohiro Kusumi [Wed, 2 Mar 2016 15:08:25 +0000 (00:08 +0900)]
sys/vfs/hammer: Add missing hammer_crc_set_btree()
set_btree is the only one missing in pairs of set/test crc functions,
though other set functions are not inlined.
Tomohiro Kusumi [Tue, 1 Mar 2016 17:29:09 +0000 (02:29 +0900)]
sys/vfs/hammer: Don't use HAMMER_CURSOR_GET_LEAF
HAMMER_CURSOR_GET_LEAF is not used by the current implementation
of hammer_btree_extract(), since this function will cause
cursor->leaf to point to the node element in question regardless
of the flag value. Then just don't use this flag when calling
hammer_btree_extract() when we know loading node element (leaf)
is the default behavior.
The cursor flag is already complicated enough, so simplifying
btree extract callers by either passing 0 or DATA but not LEAF
makes things a bit more clear.
Tomohiro Kusumi [Sun, 28 Feb 2016 15:29:59 +0000 (00:29 +0900)]
sys/vfs/hammer: Add hammer_btree_extract_data() [2/2]
This commit replaces
hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA)
with
hammer_btree_extract_data(&cursor)
which are the remaining ones from the previous commit.
From the way hammer_btree_extract_data() is implemented,
hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA)
is the same as
hammer_btree_extract(cursor, HAMMER_CURSOR_GET_LEAF | HAMMER_CURSOR_GET_DATA)
and it probably should have been (LEAF | DATA) instead of just DATA,
according to the way hammer_get_inode(), hammer_update_inode() and
hammer_update_itimes() set cursor flag. It's either LEAF or
(LEAF | DATA), but not just DATA that makes sense on calling
hammer_btree_extract().
Tomohiro Kusumi [Sun, 28 Feb 2016 15:25:03 +0000 (00:25 +0900)]
sys/vfs/hammer: Add hammer_btree_extract_data() [1/2]
hammer_btree_extract() reads data from block devices if flag
has HAMMER_CURSOR_GET_DATA. It doesn't really matter if the flag
is HAMMER_CURSOR_GET_DATA or (LEAF | DATA) as long as the flag
has HAMMER_CURSOR_GET_DATA. Calling this function will cause
cursor->leaf to point to the node element in question regardless
of the flag value.
This commit adds hammer_btree_extract_data() which is just a
wrapper over hammer_btree_extract(cursor, LEAF | DATA).
This commit replaces those with (LEAF | DATA).
Tomohiro Kusumi [Sun, 28 Feb 2016 15:05:10 +0000 (00:05 +0900)]
sys/vfs/hammer: Add hammer_btree_extract_leaf()
hammer_btree_extract() doesn't read data from block devices
unless flag has HAMMER_CURSOR_GET_DATA. It doesn't really matter
if the flag is HAMMER_CURSOR_GET_LEAF or not in order to just
extract the node element (without reading data) as long as the
flag doesn't have HAMMER_CURSOR_GET_DATA. Calling this function
will cause cursor->leaf to point to the node element in question
regardless of the flag value.
This commit adds hammer_btree_extract_leaf() which is just a
wrapper over hammer_btree_extract(cursor, HAMMER_CURSOR_GET_LEAF).
This hides complexity of using HAMMER_CURSOR_GET_LEAF or 0 or
anything other than HAMMER_CURSOR_GET_DATA which are all the same.
Matthew Dillon [Sat, 5 Mar 2016 03:17:28 +0000 (19:17 -0800)]
hammer - Fix coredump bug which can stall processes
* Fix a coredump bug which can stall the process. If a process > 100MB is
in the process of being coredumped, a signal will cause the hammer_write()
path to hit tstop() due to the core-dump being in progress. Unfortunately
the write might be happening from the kernel core generator itself which
will stall the process forever with a number of locks held.
hammer only checks for signals on very large (> 100MB) read() and write()
calls so the issue does not effect coredumps of smaller processes.
* Use CURSIG_NOBLOCK() instead of CURSIG().
Sascha Wildner [Fri, 4 Mar 2016 18:27:44 +0000 (19:27 +0100)]
Clean up some manual pages (mainly .Dv related).
Sepherosa Ziehau [Fri, 4 Mar 2016 16:06:43 +0000 (00:06 +0800)]
kqueue: Add simple per-cpu cache for the knote
This moves the malloc(M_WAITOK) out of the kqueue token for most of
the time. This significantly reduces the kqueue token contention when
TCP connect rate is @395Kconns/s on connect side: contention rate
dropped from ~150K/s to ~50K/s.
Sepherosa Ziehau [Fri, 4 Mar 2016 14:15:54 +0000 (22:15 +0800)]
kqueue: Only allow one thread to register event to a kqueue
This prevents various races on the registration path, if it
ever blocks (kq token will be released):
- Kqueue hash table creation; the malloc(M_WAIT) could block.
- Install a new event. Holding the token protect the klist could block.
Since kqueue _should_ rarely be shared between threads, allowing one
thread to do the event registration to a kqueue can work well enough.
John Marino [Fri, 4 Mar 2016 09:35:13 +0000 (10:35 +0100)]
Rewrite bsd.cpu.gcc50.mk (new cpus types; i386 removed)
The bsd.cpu.gcc50.mk should have been reworked when gcc50 came into
base. The gcc47 version was such an unmaintainable mess (despite putting
a lot of time into it) that the gcc50 version was just copied from it as
a "good enough" solution that required work later. It's time to pay the
piper. I've almost rewritten it from scratch with a goal of comprehension
and maintainability.
- All 32-bit architectures were removed
- CPU types are formally defined per architecture. If somebody tries to
define CPUTYPE to an unrecognized value, the native march flag will be
set with generic features for that architecture. It will not propagate
values that gcc50 doesn't understand (as it did previously)
- Support for several new cpus have been added.
- For x86_64 (the only arch currently supported on DF), the recognized
CPUTYPE values are:
* athlon64 (aka k8, opteron, athlon-fx)
* athlon64-sse3 (aka k8-sse3, opteron-sse3)
* barcelona (aka amdfam10)
* bdver(1|2|3|4)
* bonnell
* broadwell
* btver(1|2)
* core2
* haswell
* ivybridge
* knl
* nocona
* nehalem
* sandybridge
* silvermont
* skylake
* skylake-avx512
* westmere
* znver1
- The method for defining CPU features was reworked. The new method
removes a ton of repetition by allowing CPU features to be build on
other CPUs. It's more readable and easy to add new ones. The trade-off
is that when CPUTYPE is defined to a legal value, the mk file will
define features for all CPUs internal (OTOH, condition chain is gone).
- CPU feature has been made fully complete based on GCC 5.3 documention
- Adding new architectures will be easy using x86_64 as a template
- General makefile improvements were made
When CPUTYPE is not defined (which is the default), this version of the
makefile is functionally identical to the previous versions. Most of the
improvements are only executed when CPUTYPE type is defined in arguments
or /etc/make.conf
Imre Vadasz [Wed, 2 Mar 2016 21:24:21 +0000 (22:24 +0100)]
libstand: Fix qdivrem.c on x86_64. Don't assume sizeof(int) == sizeof(long).
* Adds two _Static_asserts() to check if the assumptions of this code
actually hold.
Taken-From: FreeBSD
Sepherosa Ziehau [Wed, 2 Mar 2016 14:39:29 +0000 (22:39 +0800)]
kqueue: Use wakeup_one based on # of threads sleep on kqueue
Get rid of the sysctl to set wakeup_one optimization. This also
unbreaks the case when wakeup_one optimization is enabled, and
multiple threads share and sleep on one kqueue.
Sascha Wildner [Wed, 2 Mar 2016 09:33:30 +0000 (10:33 +0100)]
Local adjustments for OpenSSL 1.0.1s.
Sascha Wildner [Wed, 2 Mar 2016 09:32:33 +0000 (10:32 +0100)]
Merge branch 'vendor/OPENSSL'
Sascha Wildner [Wed, 2 Mar 2016 09:31:53 +0000 (10:31 +0100)]
Import OpenSSL 1.0.1s.
* Fix CVE-2016-0800, CVE-2016-0705, CVE-2016-0798, CVE-2016-0797,
CVE-2016-0799, CVE-2016-0702
* Disable weak ciphers in SSLv3 and up in default builds of OpenSSL.
For a more detailed list of changes, see crypto/openssl/CHANGES.
Imre Vadasz [Mon, 29 Feb 2016 20:14:44 +0000 (21:14 +0100)]
lib/libstand: Remove special lib/libstand/machine headers.
* stdarg.h was identical
* stdint.h stuff was already included in sys/cpu/x86_64/include/stdint.h.
* endian.h is obsoleted by the sys/cpu/x86_64/include/endian.h.
* Added #ifdef-s to sys/cpu/x86_64/include/asm.h to handle the __i386__
case.
* No need for CFLAGS+= -I${.CURDIR} in lib/libstand/Makefile anymore.
Sascha Wildner [Tue, 1 Mar 2016 20:59:10 +0000 (21:59 +0100)]
uname(1): The maximum length for a varsym value is MAXVARSYM_DATA.
Sascha Wildner [Tue, 1 Mar 2016 20:58:34 +0000 (21:58 +0100)]
uname(3): Make uname() honor varsym variables too.
Just like uname(1) does.
Sascha Wildner [Tue, 1 Mar 2016 20:28:08 +0000 (21:28 +0100)]
rtadv{d,ctl}(8): Resolve some __FreeBSD__/__FreeBSD_version checks.
Discussed-with: sephe
Tested-by: profmakx
Sascha Wildner [Sun, 28 Feb 2016 22:40:19 +0000 (23:40 +0100)]
vi(1): Raise WARNS to 1.
Anything should survive 1.
Sascha Wildner [Sun, 28 Feb 2016 22:39:39 +0000 (23:39 +0100)]
make(1): Raise WARNS to 6.
Sascha Wildner [Sun, 28 Feb 2016 22:39:34 +0000 (23:39 +0100)]
svc(8): Clean up the Makefile a bit.
* Remove redundant LDADD, we already add -pthread. Add DPADD instead for
properly dependencies.
* Remove commented out WARNS=2. It is 6 without specifying anything.
Sascha Wildner [Sun, 28 Feb 2016 22:36:44 +0000 (23:36 +0100)]
ping6(8): Fix a -Wcast-qual warning with a cast and raise WARNS to 6.
Sascha Wildner [Sun, 28 Feb 2016 20:50:56 +0000 (21:50 +0100)]
ldd(1): Fix some warnings, raise WARNS to 6, further cleanup.
* Remove remaining a.out support.
* Nuke some orphaned externs.
* Remove (void) casts.
* Use ssize_t for read(2)'s return value.
Sascha Wildner [Sun, 28 Feb 2016 20:16:57 +0000 (21:16 +0100)]
stat(1): Sync with FreeBSD, bump WARNS to 6.
* Add -H to stat.
* Add -f to readlink.
* Various improvements and bug fixes.
Taken-from: FreeBSD
Tomohiro Kusumi [Sun, 28 Feb 2016 10:32:58 +0000 (19:32 +0900)]
sys/vfs/hammer: Rename hammer_directory_namekey() to hammer_direntry_namekey()
The name of this function should be hammer_direntry_namekey()
since it's only used with directory entry rectype, but not
directories as an object type.
(Note that rectype for directory entry is HAMMER_RECTYPE_DIRENTRY)
Tomohiro Kusumi [Sun, 28 Feb 2016 10:30:55 +0000 (19:30 +0900)]
sys/vfs/hammer: Rename hammer_ip_del_directory() to hammer_ip_del_direntry()
The name "hammer_ip_del_directory" is confusing in the sense that
it sounds like the function deletes a directory as an object type.
What this function actually does is delete a directory entry which
is different from a directory as an object type.
(Note that rectype for directory entry is HAMMER_RECTYPE_DIRENTRY)
Tomohiro Kusumi [Sun, 28 Feb 2016 10:28:46 +0000 (19:28 +0900)]
sys/vfs/hammer: Rename hammer_ip_add_directory() to hammer_ip_add_direntry()
The name "hammer_ip_add_directory" is confusing in the sense that
it sounds like the function adds a directory as an object type.
What this function actually does is add a directory entry which
is different from a directory as an object type.
(Note that rectype for directory entry is HAMMER_RECTYPE_DIRENTRY)
Tomohiro Kusumi [Sun, 28 Feb 2016 10:04:34 +0000 (19:04 +0900)]
sys/vfs/hammer: Rename struct hammer_entry_data to hammer_direntry_data
hammer_entry_data is a structure for dirctory entry, so it's
simply more understandable if the name was hammer_direntry_data.
(Note that rectype for directory entry is HAMMER_RECTYPE_DIRENTRY)
This change does not affect userspace since userspace applications
never use (and should never use) this structure, except for some
sbin/hammer commands. sbin/hammer/cmd_recover.c is the only file
that refers to the name of this structure.
Tomohiro Kusumi [Sat, 27 Feb 2016 19:11:19 +0000 (04:11 +0900)]
sys/vfs/hammer: Fix comment
This comment made by
66325755 in 2007 seems to be wrong today.
There is no code that indicates parent inode# == 0 means it's
deleted, or treats it as deleted.
Sascha Wildner [Sun, 28 Feb 2016 09:41:38 +0000 (10:41 +0100)]
vkernel.7: Improve formatting (also of usage(), while here).
Tomohiro Kusumi [Sat, 27 Feb 2016 16:07:28 +0000 (01:07 +0900)]
sbin/hammer: Reword "llid" to "pfs_id"
This file is the only one using "llid" instead of "pfs_id".
Not sure what llid stands for, but this is apparently PFS id,
so rename it to more common and obvious pfs_id.
Tomohiro Kusumi [Sat, 27 Feb 2016 03:29:28 +0000 (12:29 +0900)]
sbin/hammer: Use HAMMER_OBJID_ROOT for root inode
Tomohiro Kusumi [Thu, 25 Feb 2016 19:09:23 +0000 (04:09 +0900)]
sys/vfs/hammer: Add hammer_is_pfs_{master|slave|deleted}()
Makes code more readable, but introducing these macros could
make |= and &=~ unclear by hiding &.
There are a few |= and &=~ but not many.
Tomohiro Kusumi [Sat, 27 Feb 2016 02:13:36 +0000 (11:13 +0900)]
lib/libhammer: Add CFLAGS+= -I${.CURDIR}/../../sys
so libhammer includes headers under the working tree instead of
/usr/include/...
Tomohiro Kusumi [Sat, 27 Feb 2016 01:43:14 +0000 (10:43 +0900)]
sbin/hammer: Printf "PFS" instead of "pfs"
The existing printfs as well as hammer(5) use "PFS" unless
it's a part of command name like "pfs-master".
Imre Vadász [Sat, 27 Feb 2016 12:54:04 +0000 (13:54 +0100)]
sys/boot/pc32: Update smbios code a bit, needed for the UEFI bootloader.
* The void smbios_detect(const caddr_t addr) declaration is moved to a
separate smbios.h header.
* Since smbios_match() isn't needed yet, this change omits importing that
code from FreeBSD for now.
Taken-From: FreeBSD
Imre Vadasz [Sat, 27 Feb 2016 10:18:51 +0000 (11:18 +0100)]
sys/boot/common: Fix a pointer type mismatch when building for x86_64.
Imre Vadasz [Fri, 26 Feb 2016 21:19:23 +0000 (22:19 +0100)]
sys/boot/common: Add some casts to fix warnings when building for x86_64.
Inspired-By: FreeBSD
Bill Yuan [Fri, 26 Feb 2016 23:17:28 +0000 (23:17 +0000)]
ipfw3: update rule in func skipto
Bill Yuan [Fri, 26 Feb 2016 22:48:37 +0000 (22:48 +0000)]
ipfw3: new feature 'ipfwsync'
ipfwsync is a new feature to synchronize firewall states between machines
which is running ipfw3 firewall for high availability. ipfw3 can be configured
in centre or edge mode. the centre will automatically sync the states from
centre to the edge.
e.g.
ipfw3 sync edge 5000
ipfw3 sync start edge
ipfw3 sync centre 192.168.1.1:5000,192.168.1.2:5000
ipfw3 sync start centre
ipfw3 sync show config
ipfw3 sync show status
ipfw3 sync test centre 123
Imre Vadasz [Thu, 25 Feb 2016 20:16:14 +0000 (21:16 +0100)]
libstand: Fix some build warnings/errors when trying to build for 64bit.
* Just casts uintptr_t values to uintmax_t, since libstand doesn't have
<inttypes.h> and hence no PRIuPTR for formatting those.
* Use %zd and %zu for ssize_t and size_t values, instead of %d formats.
Partly Submitted-by: zrj (casting uintptr_t to uintmax_t as easiest fix).
Tomohiro Kusumi [Wed, 24 Feb 2016 17:26:10 +0000 (02:26 +0900)]
sys/vfs/hammer: Redo commit
283f6970
283f6970 was reverted by
8a94c3eb.
This commit redo the same sanity check in the right place.
Mirror write code that isn't trying to create a PFS still
calls hammer_mkroot_pseudofs() (which is kind of confusing),
so this sanity check must be done only after hammer_get_inode()
failed to find the existing inode by returning NULL.
Bill Yuan [Fri, 26 Feb 2016 00:26:11 +0000 (00:26 +0000)]
ipfw3: fix func lookup_host
Imre Vadasz [Thu, 25 Feb 2016 15:49:39 +0000 (16:49 +0100)]
libstand: Use the +Q register constraint in the __byte_swap16_var() macro.
* Using +Q makes sure that the register also has a 'high' 8-bit
sub-register. On amd64, the +q constraint would also contain the
SIL and DIL registers, which don't have a 'high' 8-bit sub-register.
Sepherosa Ziehau [Thu, 25 Feb 2016 14:15:08 +0000 (22:15 +0800)]
kqueue: Avoid code duplication for knote searching in kqueue_register
Imre Vadász [Thu, 11 Feb 2016 22:36:49 +0000 (23:36 +0100)]
kernel/syscons: Add EFI Framebuffer support.
* Uses the SC_DFLT_FONT kernel configuration option by default, which is
required for a usable EFI framebuffer console (otherwise no font would
be available there, resulting in invisible text).
Sepherosa Ziehau [Thu, 25 Feb 2016 13:43:43 +0000 (21:43 +0800)]
kqueue: Use lwkt_{get,rel}pooltoken in kqueue_register
Sepherosa Ziehau [Thu, 25 Feb 2016 12:41:43 +0000 (20:41 +0800)]
kqueue: Move fp holding out of kqueue token
Imre Vadász [Mon, 22 Feb 2016 21:29:03 +0000 (22:29 +0100)]
kernel/pc64: Make have_efi_framebuffer static in pc64/x86_64/machdep.c.
probe_efi_fb() should be enough for everyone else.
François Tigeot [Tue, 23 Feb 2016 20:21:06 +0000 (21:21 +0100)]
drm/i915: Update to Linux 4.2
* Work in progress atomic modesetting support
* Broadwell support improvements
* Skylake support improvements, including runtime power management.
Starting from Linux 4.2, a separate firmware blob is required to save
and restore the state of display engines in some low-power modes. These
low-power modes have been forcibly disabled in the DragonFly version of
this driver in order to keep it blob-free.
* Basic Broxton (BXT) support
* Various Displayport improvements including optimized link training,
increased robustness, and automated support for compliance testing
* Execlist overhead reduced
* Frequency boost tuning
* Plenty of bugfixes all around
Sepherosa Ziehau [Tue, 23 Feb 2016 13:58:44 +0000 (21:58 +0800)]
kqueue: Nuke unused code
Sepherosa Ziehau [Tue, 23 Feb 2016 13:44:23 +0000 (21:44 +0800)]
kqueue: Return value of knote_release is no longer useful.
Tomohiro Kusumi [Tue, 23 Feb 2016 13:36:26 +0000 (22:36 +0900)]
Revert
283f6970e848724e9bd2c20d5311e412dea5e67c
There seems to be an issue when hammer_ioc_set_pseudofs() is
called by mirror code (that doesn't attemp to create a PFS).
Sepherosa Ziehau [Mon, 22 Feb 2016 14:08:36 +0000 (22:08 +0800)]
kqueue: Avoid reprocessing processed knotes in KNOTE.
Put a marker before the currently processing knote, so that even if
we have to reprocess this knote for non-zero hint, we don't have to
reprocess already-processed knotes, i.e. knotes before the marker.
Reviewed-by: dillon@
Bill Yuan [Tue, 23 Feb 2016 15:30:54 +0000 (15:30 +0000)]
test commit
Imre Vadász [Mon, 22 Feb 2016 21:03:15 +0000 (22:03 +0100)]
sys/boot/common: Simpler fix for "ls -l", which also works on cd9660 fs.
* Blindly using "." as path instead of "" when no explicit path is given
for ls, doesn't work on the cd9660 filesystem at the moment.
* Just omitting the "/" at the beginning of the buffer, when building the
path for calling rel_stat() seems better.
Imre Vadász [Sun, 21 Feb 2016 15:28:39 +0000 (16:28 +0100)]
pc64: Introduce machine/framebuffer.h header. Probe EFI fb.
* Move struct fb_info definition from dev/drm/include/linux/fb.h to
platform/pc64/include/framebuffer.h.
* Implement EFI framebuffer probing in platform/pc64/x86_64/machdep.c.
Imre Vadász [Sun, 21 Feb 2016 16:40:48 +0000 (17:40 +0100)]
kernel/syscons: Add NULL checks for struct fb_info's restore callback.
François Tigeot [Sun, 21 Feb 2016 07:58:43 +0000 (08:58 +0100)]
drm: Use drm_legacy_xxx functions
Reducing differences with Linux
Imre Vadász [Sat, 20 Feb 2016 14:38:04 +0000 (15:38 +0100)]
syscons: Check if vaddr == 0 in sckmsrndr.c. Improve IOCTLs with KMS.
* Just disable syscons drawing when the vaddr in struct fb_info is 0.
This allows e.g. to initialize syscons with a framebuffer console,
before the framebuffer has been mapped into virtual memory.
* Some more IOCTLs should check if the KMS framebuffer is active and
avoid VGA accesses in that case.
Imre Vadász [Fri, 19 Feb 2016 19:37:41 +0000 (20:37 +0100)]
kernel/syscons: Make the sc_update_render() function a bit more readable.
Just get rid of an unnecessary indentation level.
Imre Vadász [Fri, 19 Feb 2016 19:29:23 +0000 (20:29 +0100)]
kernel/syscons: More accurate IOCTL behaviour when KMS fb is used.
* Some IOCTLs should fail for now, when the KMS framebuffer is active,
rather than crashing the machine or producing surprising results.
* Implement the CONS_MODEINFO/FBIO_MODEINFO IOCTLs for the KMS framebuffer,
by just fabricating the video_info_t data.
Imre Vadász [Thu, 18 Feb 2016 21:25:37 +0000 (22:25 +0100)]
kernel/syscons: Disable a bit more VGA console code, when scp->fbi != NULL.
Avoids calling load_palette() during vty switching when we have already
switched to the KMS framebuffer.
Matthew Dillon [Thu, 18 Feb 2016 22:08:04 +0000 (14:08 -0800)]
kernel - Add kqueue support to NFS (fix firefox issues w/nfs)
* Firefox appears to get semi-random memory corruption and otherwise
implodes if one or more filesystems it accesses does not support
kqueue. This appears to be due to some interaction between
firefox, glib, and the kernel when kqueue support is missing
from a filesystem.
* Add host-local kqueue support to NFS. As with locks, the support
is host-local only and will not work across multiple clients
sharing the same files.
* Appears to stabilize firefox when file(s) it accesses are on
NFS.
Imre Vadász [Thu, 18 Feb 2016 19:55:13 +0000 (20:55 +0100)]
kernel/syscons: Fixup previous commit, readd scp->fbi = sc->fbi in init_scp
Imre Vadász [Wed, 17 Feb 2016 22:01:23 +0000 (23:01 +0100)]
kernel/syscons: Factor out font-scaling calculation to sc_font_scale().
Avoid setting desired_cols after calculating the font scaling factor,
otherwise multiple calls to sc_font_scale() could give surprising results.
Also use sc_font_scale in init_scp, and limit the terminal size on the
first vty to 80x25 to deal with the statically allocated sc_buffer for the
kernel console.
Sepherosa Ziehau [Thu, 18 Feb 2016 14:16:00 +0000 (22:16 +0800)]
kqueue: Knote should not be accessed once the KN_PROCESSING is cleared
Though the original code is safe, since knote_release() is called w/
kq token, it is better that we clear KN_PROCESSING as the last step.
Matthew Dillon [Wed, 17 Feb 2016 23:27:29 +0000 (15:27 -0800)]
etc/rc.d - diskless adjustments
* Try to 'mount -u /' before testing for a R/W /var and /tmp. This fixes
an issue where diskless would always create a MFS /var and MFS /tmp even
when the NFS root is nominally writable.
* Test for a R/W /var and /tmp with a echo | dd command instead of mkdir,
which should work nicely even with multiple clients booting at the same
time.
Imre Vadász [Wed, 17 Feb 2016 22:34:30 +0000 (23:34 +0100)]
kernel/syscons: Implement vr_draw_border_t function in sckmsrndr.c
Matthew Dillon [Wed, 17 Feb 2016 04:40:44 +0000 (20:40 -0800)]
libc - Improve realloc() performance.
* Oversized free()'s <= 1MB will be cached in a 16-entry array. Any oversized
allocation can make use of the cache, even if available cache entries are
significantly larger than desired, in order to accomodate application
buffer-growing behavior. This means that applications which realloc() a
lot (at least as long as the buffer is <= 1MB) will wind up being handed
a ~1MB buffer quite frequently, eliminating new mmap() operations.
* Keeps track of the excess in oversized allocations and will garbage-collect
it if it grows too large (typically > 16MB). This is primarily what allows
us to be generous.
* realloc() will attempt to use MAP_TRYFIXED to extend an oversized allocation
before giving up and remapping it.
* Improves programs which realloc() or malloc()/free() large numbers of
oversized blocks by reducing or eliminating mmap()/munmap() calls.
Sepherosa Ziehau [Tue, 16 Feb 2016 13:18:46 +0000 (21:18 +0800)]
tools/tcp_stream: Fix test name
John Marino [Tue, 16 Feb 2016 11:26:51 +0000 (12:26 +0100)]
uname(3): Add env override capability
The uname(1) program has honored UNAME_[x] overrides since DragonFly's
beginning, but libc's uname never did. Make it recognize the following
environment variables:
UNAME_m
UNAME_r
UNAME_s
UNAME_v
These variables cause uname() to return the specified values instead of
what was built into libc. This capability, which has been in FreeBSD for
more than a decade, is required for package builders that need to define
the jail environment different from the host (prime example: Python)
Imre Vadász [Mon, 15 Feb 2016 21:24:39 +0000 (22:24 +0100)]
sys/boot/common: Make stat output in "ls -l" (without path argument) work
It was accidentally using an absolute path for stat-ing the files, when
the "ls -l" was used without explicitly specifying a path.
Imre Vadász [Mon, 15 Feb 2016 20:32:37 +0000 (21:32 +0100)]
libstand: Clean up dos_readdir() a bit.
Imre Vadász [Mon, 15 Feb 2016 20:44:20 +0000 (21:44 +0100)]
libstand: Fix readdir for the root directory on a FAT32 file system.
Taken-From: FreeBSD
Sascha Wildner [Mon, 15 Feb 2016 09:08:18 +0000 (10:08 +0100)]
Update the pciconf(8) database.
February 6, 2016 snapshot from http://pciids.sourceforge.net/
François Tigeot [Sun, 14 Feb 2016 16:53:48 +0000 (17:53 +0100)]
drm/linux: Add two ktime_xxx functions
Sepherosa Ziehau [Sun, 14 Feb 2016 12:38:55 +0000 (20:38 +0800)]
tools/tcp_stream: Fix building on linux
And white space clean up.
Imre Vadász [Thu, 11 Feb 2016 21:58:09 +0000 (22:58 +0100)]
kernel/acpi: Use hint.acpi.0.rsdp to get the RSDP table physical address.
hint.acpi.0.rsdp isn't set by the legacy BIOS bootloader at the moment,
but it will be be used with the EFI bootloader.
Imre Vadasz [Wed, 10 Feb 2016 22:01:03 +0000 (23:01 +0100)]
kernel/pc64: Add support for efi memory map parsing. Add sys/sys/efi.h.
Taken-From: FreeBSD
Sascha Wildner [Sat, 13 Feb 2016 21:31:03 +0000 (22:31 +0100)]
Sync ACPICA with Intel's version
20160212.
* ACPI specification 6.1 support.
* Ensure \_SB_._INI is executed before any _REG methods are executed.
* Reverted a change made in version
20151218 which enabled method
invocations to be targets of various ASL operators (SuperName and
Target grammar elements). Also, disallow this in iASL (flagged as
errors now) in preparation for ACPI 6.2.
* Completed full support for the ACPI 6.0 External() AML opcode.
* In iASL, implemented support for an optional cross-reference
output file.
For a more detailed list, please see sys/contrib/dev/acpica/changes.txt.
Sascha Wildner [Sat, 13 Feb 2016 14:01:14 +0000 (15:01 +0100)]
<sys/un.h>: Use a standard type for sockaddr_un's sun_len.
Sascha Wildner [Sat, 13 Feb 2016 14:00:07 +0000 (15:00 +0100)]
<sys/un.h>: Whitespace adjustments (no functional change).
Sascha Wildner [Sat, 13 Feb 2016 13:59:16 +0000 (14:59 +0100)]
libedit: Restore installation of the editrc(5) manual page.
Removal of manual pages for the library privatization affects only
section 3 manual pages (for the functions in the library which is
now private). editrc.5 describes the ~/.editrc file that has an
effect on all things linked against the privatized edit library
too, so we want to ship documentation for it.
Sascha Wildner [Sat, 13 Feb 2016 13:45:29 +0000 (14:45 +0100)]
libssh: Restore installation of the moduli(5) manual page.
Removal of manual pages for the library privatization affects only
section 3 manual pages (for the functions in the library which is
now private). moduli.5 describes the /etc/ssh/moduli file as we
deliver it in base so its documentation has to be delivered too.
Sascha Wildner [Sat, 13 Feb 2016 13:38:14 +0000 (14:38 +0100)]
<netinet/in.h>: Use standard types.
Sascha Wildner [Sat, 13 Feb 2016 13:37:19 +0000 (14:37 +0100)]
<netinet/in.h>: Whitespace adjustments (no functional change).
Sascha Wildner [Sat, 13 Feb 2016 09:02:08 +0000 (10:02 +0100)]
<unistd.h>: Use types which we actually have.
This is dead code at the moment because <unistd.h> includes
<sys/types.h>, leading to these types being already defined
at this point. This commit prepares for the day when we
remove <sys/types.h> from all headers which do not actually
need it.
We should probably have __gid_t and __uid_t.
Sepherosa Ziehau [Sat, 13 Feb 2016 08:25:33 +0000 (16:25 +0800)]
kqueue/signal: Send kevent even if the signal is masked
This kinda follows FreeBSD's signal kqueue filter semantic, though
FreeBSD seems to send kevent even if no pending signal will be
rendered.
Reported-by: ftigeot@
Tested-by: ftigeot@
Sepherosa Ziehau [Fri, 12 Feb 2016 12:46:59 +0000 (20:46 +0800)]
kqueue: Rework timer expiration
So that a being processed timer knote will not be activated.
While I'm here, factor out function to reset timer knote callout.
Sepherosa Ziehau [Fri, 12 Feb 2016 11:46:30 +0000 (19:46 +0800)]
kqueue: Remove unused global kqueue token
zrj [Fri, 12 Feb 2016 10:53:06 +0000 (12:53 +0200)]
Unbreak VKERNEL64 build.
After a05cac6 removal of MAKEBOOTDEV and B_ flags from sys/reboot.h
(that is not the place for such things) I broken the VKERNEL64 build.
This is a quick fix before someone with basic motoric functions
and a heart beat will take a better look (I lack the latter).
Not sure if disabling FFS_ROOT breaks anything, but it is better
than having build errors on vkernel/x86_64/autoconf.c
Also it gave me an oportunity to finally go through procedure
described in vkernel(7) to test it out for the first time!
So far no side effects on both X86_64_NON_GENERIC and VKERNEL64
except for no longer printed "no B_DEVMAGIC (bootdev=0)" in dmesg.
I should have guarded MAKEBOOTDEV and B_ flags with __i386__ in
sys/platform/pc64/include/bootinfo.h cause there is no platform/pc32
and I'm pretty sure thats a i386 thing only.
Sepherosa Ziehau [Fri, 12 Feb 2016 10:54:52 +0000 (18:54 +0800)]
ifconfig: Fix inet6 address deletion
Submitted-by: ygrossel
Dragonfly-bug: http://bugs.dragonflybsd.org/issues/2888
Sascha Wildner [Wed, 10 Feb 2016 20:58:11 +0000 (21:58 +0100)]
libstand: Change sbrk() in libstand to take an intptr_t like libc's.
zrj [Wed, 10 Feb 2016 13:44:40 +0000 (15:44 +0200)]
sys/boot: Simplify MAKEBOOTDEV macro
Treat slice as a single part of bootdev.
Previous implementation artificially slit it into
adaptor and controller fields.
No functional change intended.
Taken-from: FreeBSD