]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
21 years agoDon't track bio allocation in debug output.
Poul-Henning Kamp [Sun, 20 Oct 2002 08:45:17 +0000 (08:45 +0000)]
Don't track bio allocation in debug output.

Sponsored by: DARPA & NAI Labs.

21 years agoStyle(9) and english(9) fixes.
Poul-Henning Kamp [Sun, 20 Oct 2002 08:43:56 +0000 (08:43 +0000)]
Style(9) and english(9) fixes.

Submitted by: schweikh

21 years agoMake it possible to specify also via geom_t ID in the geom.ctl config ioctl.
Poul-Henning Kamp [Sun, 20 Oct 2002 08:42:18 +0000 (08:42 +0000)]
Make it possible to specify also via geom_t ID in the geom.ctl config ioctl.

Sponsored by: DARPA & NAI Labs.

21 years agoAfter much delay and anticipation, welcome RAIDFrame into the FreeBSD
Scott Long [Sun, 20 Oct 2002 08:17:39 +0000 (08:17 +0000)]
After much delay and anticipation, welcome RAIDFrame into the FreeBSD
world.  This should be considered highly experimental.

Approved-by: re

21 years agoWe HAVE_STPCPY now.
David E. O'Brien [Sun, 20 Oct 2002 07:50:20 +0000 (07:50 +0000)]
We HAVE_STPCPY now.

21 years ago - Lock page queue accesses in pmap_release().
Alan Cox [Sun, 20 Oct 2002 07:19:25 +0000 (07:19 +0000)]
 - Lock page queue accesses in pmap_release().

21 years agoRemove the special casing for IP addresses that are within the IVT
Marcel Moolenaar [Sun, 20 Oct 2002 06:03:05 +0000 (06:03 +0000)]
Remove the special casing for IP addresses that are within the IVT
or the do_syscall() function. We have unwind directives to stop the
unwinder.

21 years agoDefine IVT_ENTRY and IVT_END as special versions of ENTRY and END
Marcel Moolenaar [Sun, 20 Oct 2002 06:00:09 +0000 (06:00 +0000)]
Define IVT_ENTRY and IVT_END as special versions of ENTRY and END
for defining vectors. As a result, each vector will be a global
function with unwind directives to notify the unwinder that we're
in an interrupt handler. In the debugger this will show up something
like:

Debugger(0xe000000000a211d8, 0xe000000000748960) at Debugger+0x31
panic(0xe000000000a36858, 0xe0000000021d32d0, 0xe000000000ae42e8, ...
trap(0x14, 0x100000, 0xe0000000021d32d0, 0x0, 0xa0000000002095f0, ...
ivt_Data_TLB(0x14, 0x100000, 0xe0000000021d32d0) at ivt_Data_TLB+0x1f0

21 years agoCross-reference fmtcheck(3).
Tim J. Robbins [Sun, 20 Oct 2002 03:56:37 +0000 (03:56 +0000)]
Cross-reference fmtcheck(3).

21 years agoIf MAC_MAX_POLICIES isn't defined, don't try to define it, just let the
Robert Watson [Sun, 20 Oct 2002 03:41:09 +0000 (03:41 +0000)]
If MAC_MAX_POLICIES isn't defined, don't try to define it, just let the
compile fail.  MAC_MAX_POLICIES should always be defined, or we have
bigger problems at hand.

Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

21 years agoSort by date.
Greg Lehey [Sun, 20 Oct 2002 02:20:31 +0000 (02:20 +0000)]
Sort by date.

21 years agoClarify holidays which apply only to the USA
Greg Lehey [Sun, 20 Oct 2002 02:12:46 +0000 (02:12 +0000)]
Clarify holidays which apply only to the USA

Clarify that the USSR no longer exists, but some of the holidays are
celebrated anyway.

Reviewed in part by: ru

Remove Jewish and most Muslim holidays.  They're all wrong, since they
don't apply to the Western calendar.  The much more complete Jewish
holidays are in calendar.judaic.  The Muslim holidays need to be
collected into a file, but there's not much point in having the wrong
date.

Remove many Fiji holidays.  They change every year by Government
decree, and some were duplicated as a result.

Remove some duplicates.

There's still a lot to be done; in particular, I think the Japanese
and British holidays are very inaccurate.  This file needs checking by
people who know the details.

21 years agoNew sections translated
Sebastien Gioria [Sat, 19 Oct 2002 23:27:32 +0000 (23:27 +0000)]
New sections translated

21 years agoChange the symbol lookup order to search RTLD_GLOBAL objects
Alexander Kabaev [Sat, 19 Oct 2002 23:03:35 +0000 (23:03 +0000)]
Change the symbol lookup order to search RTLD_GLOBAL objects
before referencing object's DAG. This makes it possible for
C++ exceptions to work across shared libraries and brings
us closer to the search order used by Solaris/Linux.

Reviewed by: jdp
Approved by: obrien
MFC after: 1 month

21 years agoGive in on the __sF stuff. I have a better fix in mind that is future
Peter Wemm [Sat, 19 Oct 2002 22:28:07 +0000 (22:28 +0000)]
Give in on the __sF stuff.  I have a better fix in mind that is future
proof, but this should buy me some time for now.

21 years agoStake a claim on 418 (__xstat), 419 (__xfstat), 420 (__xlstat)
Peter Wemm [Sat, 19 Oct 2002 22:25:31 +0000 (22:25 +0000)]
Stake a claim on 418 (__xstat), 419 (__xfstat), 420 (__xlstat)

21 years agoPut back NORMAL_C_NOWERROR, it was actually used.
Maxime Henrion [Sat, 19 Oct 2002 22:24:43 +0000 (22:24 +0000)]
Put back NORMAL_C_NOWERROR, it was actually used.

Pointy hat to: mux

21 years agoumap_sync is empty and is identical to vfs_stdsync. Remove it and
Alexander Kabaev [Sat, 19 Oct 2002 22:22:42 +0000 (22:22 +0000)]
umap_sync is empty and is identical to vfs_stdsync. Remove it and
use generic function instead.

Approved by: obrien

21 years agostyle(9)
Alexander Kabaev [Sat, 19 Oct 2002 22:12:19 +0000 (22:12 +0000)]
style(9)

Approved by: obrien

21 years agoGrab 416/417 real estate before I get burned while testing again.
Peter Wemm [Sat, 19 Oct 2002 22:09:23 +0000 (22:09 +0000)]
Grab 416/417 real estate before I get burned while testing again.
This is for the not-quite-ready signal/fpu abi stuff.  It may not see
the light of day, but I'm certainly not going to be able to validate it
when getting shot in the foot due to syscall number conflicts.

21 years agoWhitespace fixes per style(9)
Alexander Kabaev [Sat, 19 Oct 2002 21:58:11 +0000 (21:58 +0000)]
Whitespace fixes per style(9)

Approved by: obrien

21 years agoNew translations
Sebastien Gioria [Sat, 19 Oct 2002 21:49:09 +0000 (21:49 +0000)]
New translations

21 years agoAdd the manual page which explains how gbde works and which bits it is
Poul-Henning Kamp [Sat, 19 Oct 2002 21:44:57 +0000 (21:44 +0000)]
Add the manual page which explains how gbde works and which bits it is
composed off.

Sponsored by: DARPA & NAI Labs.

21 years agoFix the path to the make.conf example file.
Marc Fonvieille [Sat, 19 Oct 2002 21:34:50 +0000 (21:34 +0000)]
Fix the path to the make.conf example file.

PR: docs/44278
Submitted by: Chris Pepper <pepper@rockefeller.edu>

21 years agoSet the NOMACCHECK flag for namei()'s generated by the NFS server code.
Robert Watson [Sat, 19 Oct 2002 21:27:40 +0000 (21:27 +0000)]
Set the NOMACCHECK flag for namei()'s generated by the NFS server code.
We currently don't enforce protections on NFS-originated VOP's.

Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

21 years agoNew translation
Sebastien Gioria [Sat, 19 Oct 2002 21:27:09 +0000 (21:27 +0000)]
New translation

21 years agoAdd a new 'NOMACCHECK' flag to namei() NDINIT flags, which permits the
Robert Watson [Sat, 19 Oct 2002 21:25:51 +0000 (21:25 +0000)]
Add a new 'NOMACCHECK' flag to namei() NDINIT flags, which permits the
caller to indicate that MAC checks are not required for the lookup.
Similar to IO_NOMACCHECK for vn_rdwr(), this indicates that the caller
has already performed all required protections and that this is an
internally generated operation.  This will be used by the NFS server
code, as we don't currently enforce MAC protections against requests
delivered via NFS.

While here, add NOCROSSMOUNT to PARAMASK; apparently this was used at
one point for name lookup flag checking, but isn't any longer or it
would have triggered from the NFS server code passing it to indicate
that mountpoints shouldn't be crossed in lookups.

Approved by: re
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

21 years agoRegen from addition of execve_mac placeholder.
Robert Watson [Sat, 19 Oct 2002 21:15:10 +0000 (21:15 +0000)]
Regen from addition of execve_mac placeholder.

21 years agoImplement the CDROMREADAUDIO ioctl.
Marcel Moolenaar [Sat, 19 Oct 2002 21:11:43 +0000 (21:11 +0000)]
Implement the CDROMREADAUDIO ioctl.

21 years agoAdd a placeholder for the execve_mac() system call, similar to SELinux's
Robert Watson [Sat, 19 Oct 2002 21:06:57 +0000 (21:06 +0000)]
Add a placeholder for the execve_mac() system call, similar to SELinux's
execve_secure() system call, which permits a process to pass in a label
for a label change during exec.  This permits SELinux to change the
label for the resulting exec without a race following a manual label
change on the process.  Because this interface uses our general purpose
MAC label abstraction, we call it execve_mac(), and wrap our port of
SELinux's execve_secure() around it with appropriate sid mappings.

Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

21 years agoDrop in the MAC check for file creation as part of open().
Robert Watson [Sat, 19 Oct 2002 20:56:44 +0000 (20:56 +0000)]
Drop in the MAC check for file creation as part of open().

Approved by: re
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

21 years agoMake sure to clear the 'registered' flag for MAC policies when they
Robert Watson [Sat, 19 Oct 2002 20:30:12 +0000 (20:30 +0000)]
Make sure to clear the 'registered' flag for MAC policies when they
unregister.  Under some obscure (perhaps demented) circumstances,
this can result in a panic if a policy is unregistered, and then someone
foolishly unregisters it again.

Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

21 years agoHook up most of the MAC entry points relating to file/directory/node
Robert Watson [Sat, 19 Oct 2002 20:25:57 +0000 (20:25 +0000)]
Hook up most of the MAC entry points relating to file/directory/node
creation, deletion, and rename.  There are one or two other stray
cases I'll catch in follow-up commits (such as unix domain socket
creation); this permits MAC policy modules to limit the ability to
perform these operations based on existing UNIX credential / vnode
attributes, extended attributes, and security labels.  In the rename
case using MAC, we now have to lock the from directory and file
vnodes for the MAC check, but this is done only in the MAC case,
and the locks are immediately released so that the remainder of the
rename implementation remains the same.  Because the create check
takes a vattr to know object type information, we now initialize
additional fields in the VATTR passed to VOP_SYMLINK() in the MAC
case.

Approved by: re
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

21 years agoSGML-ise accents
Sebastien Gioria [Sat, 19 Oct 2002 20:07:01 +0000 (20:07 +0000)]
SGML-ise accents

21 years agoRemove the _ia64_unwind_start and _ia64_unwind_end symbols. We now
Marcel Moolenaar [Sat, 19 Oct 2002 19:32:21 +0000 (19:32 +0000)]
Remove the _ia64_unwind_start and _ia64_unwind_end symbols. We now
find the unwind table through the ELF program headers.

21 years agoUpdate the unwind information when modules are loaded and unloaded
Marcel Moolenaar [Sat, 19 Oct 2002 19:30:38 +0000 (19:30 +0000)]
Update the unwind information when modules are loaded and unloaded
by using the linker hooks. Since these hooks are called for the
kernel as well, we don't need to deal with that with a special
SYSINIT. The initialization implicitly performed on the first
update of the unwind information is made explicit with a SYSINIT.
We now don't need the _ia64_unwind_{start|end} symbols.

21 years agoAdd two hooks to signal module load and module unload to MD code.
Marcel Moolenaar [Sat, 19 Oct 2002 19:16:03 +0000 (19:16 +0000)]
Add two hooks to signal module load and module unload to MD code.
The primary reason for this is to allow MD code to process machine
specific attributes, segments or sections in the ELF file and
update machine specific state accordingly. An immediate use of this
is in the ia64 port where unwind information is updated to allow
debugging and tracing in/across modules. Note that this commit
does not add the functionality to the ia64 port. See revision 1.9
of ia64/ia64/elf_machdep.c.

Validated on: alpha, i386, ia64

21 years agoReduce code duplication by moving the common actions in
Marcel Moolenaar [Sat, 19 Oct 2002 18:59:33 +0000 (18:59 +0000)]
Reduce code duplication by moving the common actions in
link_elf_init(), link_elf_link_preload_finish() and
link_elf_load_file() to link_elf_link_common_finish().
Since link_elf_init() did initializations as a side-effect
of doing the common actions, keep the initialization in
that function. Consequently, link_elf_add_gdb() is now also
called to insert the very first link_map() (ie the kernel).

21 years agoNon-functional change in preparation of the next commit:
Marcel Moolenaar [Sat, 19 Oct 2002 18:43:37 +0000 (18:43 +0000)]
Non-functional change in preparation of the next commit:
Move link_elf_add_gdb(), link_elf_delete_gdb() and link_elf_error()
near the top of the file. The *_gdb() functions are moved inside
the #ifdef DDB already present there.

21 years agoComplete the page queues locking needed for the page-based copy-
Alan Cox [Sat, 19 Oct 2002 18:34:39 +0000 (18:34 +0000)]
Complete the page queues locking needed for the page-based copy-
on-write (COW) mechanism.  (This mechanism is used by the zero-copy
TCP/IP implementation.)
 - Extend the scope of the page queues lock in vm_fault()
   to cover vm_page_cowfault().
 - Modify vm_page_cowfault() to release the page queues lock
   if it sleeps.

21 years agoFix a missing initialization.
Poul-Henning Kamp [Sat, 19 Oct 2002 17:19:41 +0000 (17:19 +0000)]
Fix a missing initialization.

21 years agoAdd Geom Based Disk Encryption to the tree.
Poul-Henning Kamp [Sat, 19 Oct 2002 17:02:17 +0000 (17:02 +0000)]
Add Geom Based Disk Encryption to the tree.

This is an encryption module designed for to secure denial of access
to the contents of "cold disks" with or without destruction activation.

Major features:

   * Based on AES, MD5 and ARC4 algorithms.
   * Four cryptographic barriers:
        1) Pass-phrase encrypts the master key.
        2) Pass-phrase + Lock data locates master key.
        3) 128 bit key derived from 2048 bit master key protects sector key.
        3) 128 bit random single-use sector keys protect data payload.
   * Up to four different changeable pass-phrases.
   * Blackening feature for provable destruction of master key material.
   * Isotropic disk contents offers no information about sector contents.
   * Configurable destination sector range allows steganographic deployment.

This commit adds the kernel part, separate commits will follow for the
userland utility and documentation.

This software was developed for the FreeBSD Project by Poul-Henning Kamp and
NAI Labs, the Security Research Division of Network Associates, Inc.  under
DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
research program.

Many thanks to Robert Watson, CBOSS Principal Investigator for making this
possible.

Sponsored by:   DARPA & NAI Labs.

21 years agoPermits UFS ACLs to be used with the GENERIC kernel. Due to recent
Robert Watson [Sat, 19 Oct 2002 16:54:15 +0000 (16:54 +0000)]
Permits UFS ACLs to be used with the GENERIC kernel.  Due to recent
ACL configuration changes, this shouldn't result in different code paths
for file systems not explicitly configured for ACLs by the system
administrator.  For UFS1, administrators must still recompile their
kernel to add support for extended attributes; for UFS2, it's sufficient
to enable ACLs using tunefs or at mount-time (tunefs preferred for
reliability reasons).  UFS2, for a variety of reasons, including
performance and reliability, is the preferred file system for use with
ACLs.

Approved by: re

21 years ago- Remove ${NORMAL_C_NOWERROR}, it is not used anymore. To build
Maxime Henrion [Sat, 19 Oct 2002 16:47:13 +0000 (16:47 +0000)]
- Remove ${NORMAL_C_NOWERROR}, it is not used anymore.  To build
  without -Werror, we do "make WERROR=", which doesn't need this
  variable.
- Use ${.IMPSRC} instead of $< in ${NORMAL_M} for consistency with
  the rest of the file.
- Add ${WERROR} for the ${NORMAL_M} case.

Tested on: i386, sparc64

21 years agoFixups to get make release going again for i386:
Sam Leffler [Sat, 19 Oct 2002 16:44:16 +0000 (16:44 +0000)]
Fixups to get make release going again for i386:

o mount (and unmount) devfs in the chroot'd filesystem
o handle umounting devfs in case of a keyboard interrupt
o remove MAKEDEV from the fixit floppy so things fit again (mount_devfs was
  added in a separate commit)
o explicitly remove mfsroot.gz and loader.conf when building the cd-rom area
  as otherwice cp -Rp aborts when a previous run left identical files around
o increase the number of inodes on the drivers floppy; moving drivers from
  the kernel there (to shrink it's size) caused things to overflow
o while we're increasing the driver's floppy inode config, separate out all
  parameters from the boot floppy

Approved by: re

21 years agoo since you can't use DIOCGDINFO and DIOCGSLICEINFO on drive nodes with geom,
Sam Leffler [Sat, 19 Oct 2002 16:39:28 +0000 (16:39 +0000)]
o since you can't use DIOCGDINFO and DIOCGSLICEINFO on drive nodes with geom,
  get the xml configuration for the devices and "parse" the information to
  get what's needed
o replace #ifdef DEBUG constructs with DPRINT/DPRINTX to make the code more
  readable

Note the xml "parser" is very very hackish and should be replaced with a
real one.  This one was done to be very small and special-purpose; don't
think about copying it elsewhere.

Approved by: phk

21 years agohardware/common/dev.sgml : MFen 1.106
Sebastien Gioria [Sat, 19 Oct 2002 16:26:33 +0000 (16:26 +0000)]
hardware/common/dev.sgml : MFen 1.106
Makefile relnotes/Makefile  : Bump Original revision
relnotes/common/new.sgml relnotes/common/relnotes.ent : Bump Original revision
relnotes/i386/Makefile share/sgml/release.dsl : Bump Original revision

21 years agoNew release note: Wide character C library functions.
Bruce A. Mah [Sat, 19 Oct 2002 16:16:55 +0000 (16:16 +0000)]
New release note:  Wide character C library functions.

Modified release note:  Use manpage entity for pst(4).

21 years agoUse manpage entity for pst(4).
Bruce A. Mah [Sat, 19 Oct 2002 16:16:04 +0000 (16:16 +0000)]
Use manpage entity for pst(4).

21 years agoClarify that the UFS1 extended attribute configuration steps do not apply
Robert Watson [Sat, 19 Oct 2002 16:09:16 +0000 (16:09 +0000)]
Clarify that the UFS1 extended attribute configuration steps do not apply
to UFS2 file systems.

Submitted by: jedgar
Obtained from: TrustedBSD Project

21 years agoExplicitely specify an alignment for struct pcb. While all regular pcb's
Thomas Moestl [Sat, 19 Oct 2002 15:54:34 +0000 (15:54 +0000)]
Explicitely specify an alignment for struct pcb. While all regular pcb's
are positioned and aligned by md code, dumppcb is just a static
variable and requires this.

21 years agoThe argument to the DIOCGMEDIASIZE ioctl() is an off_t, not an u_int.
Thomas Moestl [Sat, 19 Oct 2002 15:40:39 +0000 (15:40 +0000)]
The argument to the DIOCGMEDIASIZE ioctl() is an off_t, not an u_int.

Reviewed by: phk

21 years agoSince NEWCARD is the default pccard subsystem, pccard[cd] is no
Makoto Matsushita [Sat, 19 Oct 2002 14:23:18 +0000 (14:23 +0000)]
Since NEWCARD is the default pccard subsystem, pccard[cd] is no
longer needed.  Note that pc98 uses OLDCARD so this is only for i386
(thanks nyan-san).

OKed by: imp

21 years agoIndent code example with one tab, not two, for consistency with the rest.
Tim J. Robbins [Sat, 19 Oct 2002 13:48:45 +0000 (13:48 +0000)]
Indent code example with one tab, not two, for consistency with the rest.

21 years agoC89 does not specifiy strsep(), so our strsep() implementation cannot
Tim J. Robbins [Sat, 19 Oct 2002 13:41:22 +0000 (13:41 +0000)]
C89 does not specifiy strsep(), so our strsep() implementation cannot
conform to it.

Obtained from: OpenBSD

21 years agoThe ftok() function has not been in libcompat for quite a while.
Tim J. Robbins [Sat, 19 Oct 2002 13:33:12 +0000 (13:33 +0000)]
The ftok() function has not been in libcompat for quite a while.

21 years agoAdd sr_YU locales
Andrey A. Chernov [Sat, 19 Oct 2002 13:04:05 +0000 (13:04 +0000)]
Add sr_YU locales

Submitted by:    Toni Andjelkovic <toni@soth.at>

21 years agoAdd sr_YU locales
Andrey A. Chernov [Sat, 19 Oct 2002 12:59:19 +0000 (12:59 +0000)]
Add sr_YU locales
Submitted by: Toni Andjelkovic <toni@soth.at>

21 years agoAdd sr_YU locale directories
Andrey A. Chernov [Sat, 19 Oct 2002 12:56:56 +0000 (12:56 +0000)]
Add sr_YU locale directories
Submitted by: Toni Andjelkovic <toni@soth.at>

21 years agoStyle(9). Make some function declarations consistent with the rest,
Mark Murray [Sat, 19 Oct 2002 11:57:38 +0000 (11:57 +0000)]
Style(9). Make some function declarations consistent with the rest,
and remove some nearby extraneous {}'s.

21 years agoSeveral malloc() calls were passing the M_DONTWAIT flag
Maxime Henrion [Sat, 19 Oct 2002 11:31:50 +0000 (11:31 +0000)]
Several malloc() calls were passing the M_DONTWAIT flag
which is an mbuf allocation flag.  Use the correct
M_NOWAIT malloc() flag.  Fortunately, both were defined
to 1, so this commit is a no-op.

21 years agoFix a problem with RTLD_TRACE flag to dlopen(3), which sometimes can return
Maxim Sobolev [Sat, 19 Oct 2002 10:18:29 +0000 (10:18 +0000)]
Fix a problem with RTLD_TRACE flag to dlopen(3), which sometimes can return
even if there was no error occured (when trying to dlopen(3) object that
already linked into executable which does dlopen(3) call). This is more
proper fix for `ldd /usr/lib/libc.so' problem, because the new behaviour
conforms to documentation.

Remove workaround from ldd.c (rev.1.32).

PR: 35099
Submitted by: Nathan Hawkins <utsl@quic.net>
MFC after: 1 week

21 years agoFix security bug in contains_dot_dot routine.
Maxim Sobolev [Sat, 19 Oct 2002 09:32:03 +0000 (09:32 +0000)]
Fix security bug in contains_dot_dot routine.

PR:             43575
Submitted by:   Brett Glass <brett@lariat.org>

X-MFC after: immediately

21 years agoClear the pending counts in the superblock after a successful run
Kirk McKusick [Sat, 19 Oct 2002 05:36:48 +0000 (05:36 +0000)]
Clear the pending counts in the superblock after a successful run
of fsck so that the kernel does not complain about them being
non-zero when the filesystem is mounted.

Sponsored by: DARPA & NAI Labs.

21 years agoIn link_elf_load_file(), when SPARSE_MAPPING is defined and we
Marcel Moolenaar [Sat, 19 Oct 2002 05:01:54 +0000 (05:01 +0000)]
In link_elf_load_file(), when SPARSE_MAPPING is defined and we
cannot allocate ef->object, we freed ef before bailing out with
an error. This is wrong because ef=lf and when we have an error
and lf is non-NULL (which holds if we try to alloc ef->object),
we free lf and thus ef as part of the bailing-out.

21 years agoSo a positive time zone offset is west, right? or is it east? Um.
Tony Finch [Sat, 19 Oct 2002 04:49:10 +0000 (04:49 +0000)]
So a positive time zone offset is west, right? or is it east? Um.
What is the standard for this, anyway? Ah, we get to choose. I see.

This commit is brought to you by the numbers 2001 and 2822 and the
letters P, O, S, I, X and R, F, C.

21 years agoMake this compile when DDB is not defined by conditionally compiling
Marcel Moolenaar [Sat, 19 Oct 2002 04:14:08 +0000 (04:14 +0000)]
Make this compile when DDB is not defined by conditionally compiling
all references to ksym_start and ksym_end.

21 years agoMake the unwind functions standard and not optional on ddb. They
Marcel Moolenaar [Sat, 19 Oct 2002 04:02:16 +0000 (04:02 +0000)]
Make the unwind functions standard and not optional on ddb. They
will eventually be used for ktrace(2) too.

21 years agoAdd the libz derived files, added in the previous commit, to
Marcel Moolenaar [Sat, 19 Oct 2002 02:23:09 +0000 (02:23 +0000)]
Add the libz derived files, added in the previous commit, to
CLEANFILES. We were not cleaning up after ourselves.

21 years agoDon't leak memory in semop(2). (Fix a bug I introduced in rev 1.55.)
Alfred Perlstein [Sat, 19 Oct 2002 02:07:35 +0000 (02:07 +0000)]
Don't leak memory in semop(2). (Fix a bug I introduced in rev 1.55.)

Detective work by: jake

21 years agoUpdate the documentation for kthread_create to include the pages argument.
Benno Rice [Sat, 19 Oct 2002 01:39:44 +0000 (01:39 +0000)]
Update the documentation for kthread_create to include the pages argument.

Reviewed by: sheldonh
Forgotten by: scottl

21 years agoPass the right number of tlb slots to the kernel. The allocation scheme
Thomas Moestl [Fri, 18 Oct 2002 23:49:18 +0000 (23:49 +0000)]
Pass the right number of tlb slots to the kernel. The allocation scheme
was changed in r1.4, but I neglected to update most of the code in
metadata.c.

Pointy hat to: tmm

21 years agoAdd sab driver to serial interfaces section.
Bruce A. Mah [Fri, 18 Oct 2002 23:24:20 +0000 (23:24 +0000)]
Add sab driver to serial interfaces section.

21 years agoNew release notes: P1003_1B gone, CPU_DISABLE_CMPXCHG, FAST_IPSEC,
Bruce A. Mah [Fri, 18 Oct 2002 23:23:32 +0000 (23:23 +0000)]
New release notes:  P1003_1B gone, CPU_DISABLE_CMPXCHG, FAST_IPSEC,
ifconfig(8) monitor, ofwdump(8), ping(8) -o, quota(1) -l, tunefs(8)
-a/-l, ugidfw(8).

Modified release notes:  se driver replaced by sab driver.

21 years agoBound the size of the superblock to SBLOCKSIZE.
Kirk McKusick [Fri, 18 Oct 2002 23:17:30 +0000 (23:17 +0000)]
Bound the size of the superblock to SBLOCKSIZE.

Submitted by: BOUWSMA Beery <freebsd-misuser@netscum.dyndns.dk>
Sponsored by: DARPA & NAI Labs.

21 years agoFix a file-rewrite performance case for UFS[2]. When rewriting portions
Matthew Dillon [Fri, 18 Oct 2002 22:52:41 +0000 (22:52 +0000)]
Fix a file-rewrite performance case for UFS[2].  When rewriting portions
of a file in chunks that are less then the filesystem block size, if the
data is not already cached the system will perform a read-before-write.
The problem is that it does this on a block-by-block basis, breaking up the
I/Os and making clustering impossible for the writes.  Programs such
as INN using cyclic file buffers suffer greatly.  This problem is only going
to get worse as we use larger and larger filesystem block sizes.

The solution is to extend the sequential heuristic so UFS[2] can perform
a far larger read and readahead when dealing with this case.

(note: maximum disk write bandwidth is 27MB/sec thru filesystem)
(note: filesystem blocksize in test is 8K (1K frag))
dd if=/dev/zero of=test.dat bs=1k count=2m conv=notrunc

Before:  (note half of these are reads)
      tty             da0              da1             acd0             cpu
 tin tout  KB/t tps  MB/s   KB/t tps  MB/s   KB/t tps  MB/s  us ni sy in id
   0   76 14.21 598  8.30   0.00   0  0.00   0.00   0  0.00   0  0  7  1 92
   0   76 14.09 813 11.19   0.00   0  0.00   0.00   0  0.00   0  0  9  5 86
   0   76 14.28 821 11.45   0.00   0  0.00   0.00   0  0.00   0  0  8  1 91

After: (note half of these are reads)
      tty             da0              da1             acd0             cpu
 tin tout  KB/t tps  MB/s   KB/t tps  MB/s   KB/t tps  MB/s  us ni sy in id
   0   76 63.62 434 26.99   0.00   0  0.00   0.00   0  0.00   0  0 18  1 80
   0   76 63.58 424 26.30   0.00   0  0.00   0.00   0  0.00   0  0 17  2 82
   0   76 63.82 438 27.32   0.00   0  0.00   0.00   0  0.00   1  0 19  2 79

Reviewed by: mckusick
Approved by: re
X-MFC after: immediately (was heavily tested in -stable for 4 months)

21 years ago* Add CDRIOC{READ,WRITE}SPEED ioctls to cd(4). Units are in KB/sec.
Nate Lawson [Fri, 18 Oct 2002 22:03:39 +0000 (22:03 +0000)]
* Add CDRIOC{READ,WRITE}SPEED ioctls to cd(4).  Units are in KB/sec.
* Change atapi-cd ioctls to use the same units.
* Change burncd, cdcontrol to convert CDROM speed to KB/sec before
calling the ioctl.  Add a "max" speed option for their command lines.

This change does not break ABI but does change the units passed through
the ioctl so 3rd party software that uses cdrio.h will have to convert
(most likely by multiplying CDROM speed by 177 to get KB/s).

PR: kern/36845
Submitted by: Philipp Mergenthaler <p@i609a.hadiko.de> (CAM ioctls)
Reviewed by: sos, ken
MFC after: 1 month

21 years agoMFufs 1.33:
Bruce Evans [Fri, 18 Oct 2002 21:41:41 +0000 (21:41 +0000)]
MFufs 1.33:
    In the 'found' case for ext2_lookup() the underlying bp's data was
    being accessed after the bp had been releaed.  A simple move of the
    brelse() solves the problem.

The PR reports that this caused panics running the GDB testsuite unless
NO_GEOM is configured.

PR: 44060
Reported by: Mark Kettenis <kettenis@chello.nl>
MFC after: 3 days

21 years ago(1) added LSI Logic copyright, and legal line 3 in license, and string
Eric Moore [Fri, 18 Oct 2002 21:29:14 +0000 (21:29 +0000)]
(1) added LSI Logic copyright, and legal line 3 in license, and string
changes for "LSILogic"
(2) enabled non-disk support through CAM interface
(3) HA_INQ (a) enabled tagged queuing (b) disable reset during
driver loading (b) renamed BSDi string to LSI
(4) disabled detecting disk devices during SCSI INQUIRY
(5) changed dcdb single element sglist to send one entire buffer chunk
(6) nsgelem not set in sglist
(7) ap_data_transfer_length not set for dcdb
(8) changed "struct thread" to "d_thread_t" for compatibliity { xxx_open,
xxx_close, xxx_ioctl }
(9) miscellaneous compatiblity fixes
(10) bug fix for 0x0409/0x1000 card
(11) added compiling amr_cam.c in sys/conf/files
(12) added compiling amr_cam.c in sys/modules/amr/Makefile

Reviewed by:ps
MFC after:1 week
1 week

21 years agoUpdate extended attribute readme file to note that no special configuration
Robert Watson [Fri, 18 Oct 2002 21:11:36 +0000 (21:11 +0000)]
Update extended attribute readme file to note that no special configuration
is required to use EAs with UFS2, and that UFS2 is recommend for EA use
for a variety of reasons.

Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

21 years agoUpdate instructions for ACLs given recent tunefs, mount changes. Also
Robert Watson [Fri, 18 Oct 2002 21:09:57 +0000 (21:09 +0000)]
Update instructions for ACLs given recent tunefs, mount changes.  Also
note that UFS2 doesn't require explicit extended attribute configuration,
and is recommends for this and other reasons if you plan to use ACLs.

Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

21 years agoUse 'size_t' instead of 'int' for the result of sizeof().
Robert Watson [Fri, 18 Oct 2002 21:03:30 +0000 (21:03 +0000)]
Use 'size_t' instead of 'int' for the result of sizeof().

21 years agoModified release note: New commands for lpc(1).
Bruce A. Mah [Fri, 18 Oct 2002 19:54:52 +0000 (19:54 +0000)]
Modified release note:  New commands for lpc(1).

PR: 44070
Submitted by: gad

21 years agoOops. xargs -J is in fact not deprecated.
Bruce A. Mah [Fri, 18 Oct 2002 19:05:47 +0000 (19:05 +0000)]
Oops.  xargs -J is in fact not deprecated.

PR: 44070
Submitted by: gad
Reviewed by: jmallett

21 years agoAdd myself as da(4) maintainer. This is mostly so I can review and commit
Nate Lawson [Fri, 18 Oct 2002 18:27:56 +0000 (18:27 +0000)]
Add myself as da(4) maintainer.  This is mostly so I can review and commit
quirks.

21 years agoDo not lock the process when calling fdfree() (this would have recursed on
John Baldwin [Fri, 18 Oct 2002 17:45:41 +0000 (17:45 +0000)]
Do not lock the process when calling fdfree() (this would have recursed on
a non-recursive lock, the proc lock, before) since we don't need it to
change p_fd.

21 years agofdfree() clears p_fd for us, no need to do it again.
John Baldwin [Fri, 18 Oct 2002 17:44:39 +0000 (17:44 +0000)]
fdfree() clears p_fd for us, no need to do it again.

21 years agoDon't lock the proc lock to clear p_fd. p_fd isn't protected by the proc
John Baldwin [Fri, 18 Oct 2002 17:42:28 +0000 (17:42 +0000)]
Don't lock the proc lock to clear p_fd.  p_fd isn't protected by the proc
lock.

21 years agoReplace the vm_page hash table with a per-vmobject splay tree. There should
Matthew Dillon [Fri, 18 Oct 2002 17:24:30 +0000 (17:24 +0000)]
Replace the vm_page hash table with a per-vmobject splay tree.  There should
be no major change in performance from this change at this time but this
will allow other work to progress:  Giant lock removal around VM system
in favor of per-object mutexes, ranged fsyncs, more optimal COMMIT rpc's for
NFS, partial filesystem syncs by the syncer, more optimal object flushing,
etc.  Note that the buffer cache is already using a similar splay tree
mechanism.

Note that a good chunk of the old hash table code is still in the tree.
Alan or I will remove it prior to the release if the new code does not
introduce unsolvable bugs, else we can revert more easily.

Submitted by: alc (this is Alan's code)
Approved by: re

21 years agoOops. Also provide a lint-compatible unused argument warning killer.
Mark Murray [Fri, 18 Oct 2002 16:24:20 +0000 (16:24 +0000)]
Oops. Also provide a lint-compatible unused argument warning killer.

21 years agoISOfy functions, sort headers and mark unused arguments.
Mark Murray [Fri, 18 Oct 2002 16:22:13 +0000 (16:22 +0000)]
ISOfy functions, sort headers and mark unused arguments.

21 years agoCorrect the headers needed to use dbopen(3) and friends.
Mark Murray [Fri, 18 Oct 2002 16:20:08 +0000 (16:20 +0000)]
Correct the headers needed to use dbopen(3) and friends.

21 years agoConnect ofwdump to the sparc64 build.
Thomas Moestl [Fri, 18 Oct 2002 15:38:39 +0000 (15:38 +0000)]
Connect ofwdump to the sparc64 build.

21 years agoAdd a utility to examine the OpenFirmware device tree (on sparc64). This
Thomas Moestl [Fri, 18 Oct 2002 15:37:15 +0000 (15:37 +0000)]
Add a utility to examine the OpenFirmware device tree (on sparc64). This
allows access to detailed machine configuration information, and should
be especially useful to gather information for driver-related bug
reports.

21 years agoInstall the include files in sys/dev/ofw.
Thomas Moestl [Fri, 18 Oct 2002 15:30:50 +0000 (15:30 +0000)]
Install the include files in sys/dev/ofw.

21 years agoBuild openfirmio on sparc64.
Thomas Moestl [Fri, 18 Oct 2002 15:27:02 +0000 (15:27 +0000)]
Build openfirmio on sparc64.

21 years agoAdd a pseudo device which allows to access the OpenFirmware device tree
Thomas Moestl [Fri, 18 Oct 2002 15:23:43 +0000 (15:23 +0000)]
Add a pseudo device which allows to access the OpenFirmware device tree
via ioctl()s. This was ported from NetBSD and adapted a bit to better
match our OpenFirmware support code.

21 years agoAdd an #ifdef _KERNEL to make it possible to include this file from
Thomas Moestl [Fri, 18 Oct 2002 15:21:09 +0000 (15:21 +0000)]
Add an #ifdef _KERNEL to make it possible to include this file from
userland (to get the typedefs).

21 years agoStaticify for lint.
Mark Murray [Fri, 18 Oct 2002 14:48:48 +0000 (14:48 +0000)]
Staticify for lint.

21 years agoConstify and staticify for lint.
Mark Murray [Fri, 18 Oct 2002 14:45:00 +0000 (14:45 +0000)]
Constify and staticify for lint.