CVS log for src/usr.bin/kdump/kdump.c

Up to [cvs.NetBSD.org] / src / usr.bin / kdump

Request diff between arbitrary revisions

Keyword substitution: kv
Default branch: MAIN

Pull up following revision(s) (requested by kre in ticket #641):

	usr.bin/kdump/kdump.c: revision 1.145

Remove debug print

Remove debug print

Pull up following revision(s) (requested by kre in ticket #635):

	usr.bin/kdump/kdump.c: revision 1.141
	usr.bin/kdump/kdump.c: revision 1.142
	usr.bin/kdump/kdump.c: revision 1.143
	usr.bin/kdump/kdump.c: revision 1.144

PR/57918: Taylor R Campbell: Don't try to print symbolically ioctls that
have bits set on the high 32 bits of a 64 bit long.

PR bin/57918

Redo the previous correctly, the test should be whether or not only the
bottom 32 bits are set, rather than whether any of bits 32..63 are set.

This makes no difference if u_long (or unsigned long long) is 64 bits,
it is the same test in that case, but if u_long were 128 bits there is
a notable difference (though for the particular issue observed in the PR,
it is likely to work either way).
decode some {g,s}etsockopt*.

kdump: keep the comment with its corresponding code
No binary change.

kdump: keep the comment with its corresponding code

No binary change.

decode some {g,s}etsockopt*.

PR bin/57918

Redo the previous correctly, the test should be whether or not only the
bottom 32 bits are set, rather than whether any of bits 32..63 are set.
This makes no difference if u_long (or unsigned long long) is 64 bits,
it is the same test in that case, but if u_long were 128 bits there is
a notable difference (though for the particular issue observed in the PR,
it is likely to work either way).

PR/57918: Taylor R Campbell: Don't try to print symbolically ioctls that
have bits set on the high 32 bits of a 64 bit long.

fix printf format string for xattr names (the length of the xattr name
needs to be a precision rather than a width).

futexput: cast the mask to u_long (as the op already is)

Decode futex op codes.

Sync with HEAD

Understand xattr ktruser records

Mostly merge changes from HEAD upto 20200411

Merge changes from current as of 20200406

Update usage for ioctlprint.

Add the ability to:
1. list all ioctls (-l)
2. print the internal definition symbolically
3. search by name

$ ioctlprint -f "%n %E %e\n" TIOCGETA
TIOCGETA _IOR('t', 19, struct termios) _IOR('t',0x13,0x2c)

*** empty log message ***

Add ioctlprint - descriptive ioctl value printer

ioctlprint(1) is embedded in the kdump(1) program.

kdump(1): Symbolic printing for MALLOC_OPTIONS="U".

Sync with HEAD

You don't have to be _LP64 to run 32 bit binaries. You can be mips (64 bit
kernel/32 bit userland).

Synch with HEAD

print the mprotect perm argument symbolically.

Sync usage with man page.

Print negative time offsets correctly.  Provide an exegetical quote
from riastradh@, slightly edited to fit the metre.

Option -E to display elapsed timestamps (time since beginning of trace).

Option name from FreeBSD.  While here, make it possible to use a
combination of -T -E and -R to display timestamps in several formats.
Idea also from FreeBSD.

Drop support for KTRFACv0 (NetBSD 3) and KTRFACv1 (NetBSD 5).
Note that on 32-bit platforms it probably never worked anyway because
header size changed from v1 to v2 when time_t became 64 bit.

ok christos@
CVS: ----------------------------------------------------------------------
CVS: CVSROOT  cvs.NetBSD.org:/cvsroot
CVS: please use "PR category/123" to have the commitmsg appended to PR 123
CVS:
CVS: Please evaluate your changes and consider the following.
CVS: Abort checkin if you answer no.
CVS: => For all changes:
CVS: Do the changed files compile?
CVS: Has the change been tested?
CVS: => If you are not completely familiar with the changed components:
CVS: Has the change been posted for review?
CVS: Have you allowed enough time for feedback?
CVS: => If the change is major:
CVS: => If the change adds files to, or removes files from $DESTDIR:
CVS: => If you are changing a library or kernel interface:
CVS: Have you successfully run "./build.sh release"?

Drop main() prototype.

Symbolic printing for mmap

Don't enter infinite loop on big ktr_len values.

Fixes PR 49460.

Cast register_t to unsigned long before printf'ing it with %lx.

be a bit smarter about fcntl

more stray \n

fix some error handling.

Rebase to HEAD as of a few days ago.

Rebase.

sync with head.

for a reference, the tree before this commit was tagged
as yamt-pagecache-tag8.

this commit was splitted into small chunks to avoid
a limitation of cvs.  ("Protocol error: too many arguments")

Fix compat netbsd32 syscall return values display by calling
output_long() instead of printf().

PR/48413: Inappropriate sign extension in kdump output

resync with head

sync with head

Only expose the bulk of sys/proc.h and sys/lwp.h if _KERNEL or _KMEMUSER
is defined.
i386 and amd64 build ok.

don't limit sockets to sizeof(sizeof sockaddr_storage). We can create valid
sockaddr_un sockets that exceed that. Arguably either
sizeof(struct sockaddr_storage) should be amended, or we should disallow
unix domain sockaddrs that exceed that size.

Symbolic printing of some known MISC records.

- use PT_STRINGS
- use __arraycount()

Catchup with rmind-uvmplock merge.

kern/42030 - tracking of file descriptors by ktrace/kdump

Remove Mach specific trace points.

Remove Darwin, MACH and Mach-O support.

Fix string constant so it compiles again.

we don't have access to si_lid anymore.

print detailed info about other SI_ codes.

format SI_QUEUE properly.

fix emul state tracking for exit vs exit_group on linux.

Sync with HEAD.

Third (and last) commit. See http://mail-index.netbsd.org/source-changes/2009/05/13/msg221222.html

Allow kdump to be used as an ioctl decoder if invoked as ioctlname.

fix sign-compare issues

resolve conflict.

merge christos-time_t

Add minimal SI_LWP support.

update for version 2

Merge in changes between wrstuden-revivesa-base-2 and
wrstuden-revivesa-base-3.

Sync with wrstuden-revivesa-base-2.

file kdump.c was added on branch christos-time_t on 2008-12-29 01:23:13 +0000

handle SI_NOINFO

Remove the \n and tabs from the __COPYRIGHT() strings.
Tweak to use a consistent format.

Sync w/ -current. 34 merge conflicts to follow.

Output linux ptrace request name for processes under compat linux32
emulation too.

sync with head.

sync with HEAD

Add a few NULL entries to linux_ptrace_ops[], to make PTRACE_SYSCALL
string appear at the expected indice 24.

sync with HEAD

convert __attribute__s to applicable cdefs.h macros

Apply the -N (show syscalls by number) to the system return trace.
Change 'USER' to 'MISC' since they get used for misc kernel traces as well.

Apply fix from PR 35283 from David Holland:

ensure that kdump properly escapes newlines in argv

When displaying 'KTR_USER' data, exclude the header from the length, don't
enclose in ", and output using the hexdump function if -x (or -X) given.

fallout from si_sigval -> si_value

When decoding ioctl commands type; do not output non printable
characters, but the corresponding hexadecimal values.

ok by christos

Pull up following revision(s) (requested by chs in ticket #420):
	usr.bin/kdump/kdump.c: revision 1.88
skip over the "self" sa when printing upcalls.

Sync with kernel changes introduced by merging the newlock2 branch.

skip over the "self" sa when printing upcalls.

since register_t is sometimes unsigned, cast to (long) to check >= 0.

Add a -t+S flag to ktrace for tracing activity related to sysctl. MIB
names will be displayed, with data readen and written as well.

Use the standard macros to compute the ioctl length instead of rolling our
own. Thanks to Chuck for noticing.

Add PT_SYSCALL

Learn about v1 records and sa upcalls.

Print symbolic signal names for killpg *sigaction* and *sigvec*.

WARNS=3

appease gcc -Wuninitialized

Pull up revision 1.79 (requested by skrll in ticket #789):
Ensure a non-zero 'word_size' is passed to hexdump_buf for unknown
trace types.
Also fix an incorrect use of the global word_size variable

Ensure a non-zero 'word_size' is passed to hexdump_buf for unknown
trace types.
Also fix an incorrect use of the global word_size variable

Pull up revision 1.78 (requested by matt in ticket #278):
Allow kdump to print the names for machine dependent ptrace ops.
(This requires <machine/ptrace.h> to define PT_MACHDEP_STRINGS
which corresponds to the machdep ops).

Allow kdump to print the names for machine dependent ptrace ops.
(This requires <machine/ptrace.h> to define PT_MACHDEP_STRINGS
which corresponds to the machdep ops).

Suppress duplicate lines when hexdumping data

Fix whitespace usage and remove unnecessary semi-colon.

We can't access array[sizeof(array) / sizeof(array[0])].

Calculate the width again if we start newline at tab character, since
start column changes.

Fix reversed logic (decimal argument passed in output_long, where it expects
hex). Change tests to <= 9 to a small(v) macro, so that negative numbers fail
too consistently.

redo the former in light of making the kernel output normal 64 bit
records for 32 bit & 64 bit processes.  we still make some adjustments
when dealing with 32-bit processes, but largely things are handled the
same now.  make the code to match "netbsd32_<syscall>" in place of
plain "<syscall>" more general.

- add a new flags field to the emulation vector, with one user so far
EMUL_FLAG_NETBSD32.  set EMUL_FLAG_NETBSD32 in any 32 bit on 64 bit
kernel emulation layers.
- if EMUL_FLAG_NETBSD32 is set, calculate the proper argument count.
(XXX: this should use register32_t, but that's not visible to all
builders of kdump...)

now netbsd32_ioctl(2) actually reports sane values, and other arguments
are mostly correctly printed (there are still some signed extension
issues with 32 bit numbers being displayed as "0xffffffff8xxxxxxx".)

Fix cross-build problem on sparc caused by a cast size_t -> int

Enable tracing of out of line data sent with Mach messages

For Mach messages, always display the message Id in decimal. The information
is indeed in the message, but in hex. It's a bit inconvenient.

Stop core dump if an ioctl trace has fewer than expected arguments.
- Check names of system calls (for special treatment) because emulations
  might use different numbers.
- Report an error if arguments to -p or -m non-numeric.
- Just take last of -x and -Xvalue, stop -X0x80000000 being valid (core dumps).
- Keep 'last used' emulation_ctx at top of list, create on lookup (usually
  EMUL anyway), delete when exit called (doesn't return).
- Slightly improve hack to get correct system call name in execve return
  when emulation has changed.
- Rename global variables 'current'/'previous' to 'cur_emul'/'prev_emul'.
(TODO: save system call type (and maybe an argument) per pid so GIO trace
format can depend on the actual system call.)
Fixes part of PR sparc64/23473 - but system call arguments will still not be
displayed correctly.

Following mycroft's comment, restore -x to its original behavior: without
an argument. Introduce -Xsize to do the job. -x is equivalent to -X1
Of course -x and -Xsize are mutually exclusive.

Validate the numeric argument to -x, support -x1|2|4|8|16.
Produce sensible output on little-endian systems.

Add an argument to the -x flag to choose between byte per byte display
and word per word display in hex dumps.
-x1 gives you byte per byte, -x4 word per word, and -x defaults to -x1

kdump now displays Mach services names itself, including a table of
id/names in sys/compat/mach/mach_services_names.c

Remove ports and flags displays, the information is already in the message.

Fix printf botch

improve formatting in the non-siginfo case, by printing the code. This is
for compatibility with the old ktrace files.

Print siginfo_t information when available.

Move UCB-licensed code from 4-clause to 3-clause licence.

Patches provided by Joel Baker in PR 22365, verified by myself.

Grok KTR_EXEC_ENV and KTR_EXEC_ARG.
Make ktrace -c default to removing all trace points (not just the default ones).
Make kdump default to dumping everything in the trace file.
Add 'A' (all) and '-' (remove following) to valid -t arguments.
Dump data block of UNKNOWN trace points in hex + ascii.
Make first time output with -R 0 (instead of time since epoch).
Use svis() instead of vis() to get " escaped (as \") (needs fixed libc)
Correctly pass unsigned values to svis()
Update man pages.

Change back to print hex bytes. The 'word' print was completely broken
on little endian systems.  It is also much easier to read misaligned
words in a straight byte hexdump.
Cut offset to 3 hex digits so that it fixs in 80 columns and still
leaves a tab at the start of 'continuation' lines.
The ktrace buffer is limited to 4k, so 3 digits is ample.

On some ports, pointer differences are not ints, so do a cast
before using such a result as a field width.

Make hexdump output fix in 80 char-wide terminal

Use the same code to display hexdump of GIO and MMSG (Mach Messages).
While we are there, display offsets in the hexdump, it can be useful.
Remove size display in Mach Messages as user processes don't always
fill it correctly.

Add -x to usage output.

Add a '-x' option to display GIO data blocks in hex and ascii (instead of
vis(3) format).
Maybe there could be a heuristic to decide whether the data is printable...

Back out previous.

Since a caddr_t is (mis-)used as the lwp-id, cast it to intptr_t and use
apropriate format string.

add appropriate hack to display ktr_buf as the lwpid if -L is supplied on
the command line.

Print rval[1] if non-zero and present in the trace entry.
To get both fd numbers for pipe().

Make "-m0" print no data, so that we agree better with the man page.

Added support for dumping mach messages in ktrace/kdump. While we are
there, KNFify a few functions.

Sync usage with man page (and actual options).

signame(): if sig == 0, return name " 0", so that the result would be
printed as:
   CALL  kill(0x1e4, SIG 0)
rathern than previous
   CALL  kill(0x1e4, SIG*unknown 0*)

Add -N which inhibits translation of system calls numbers into names.

Now properly display Mach traps

Allow passing of the ktrace file as the last argument without a -f.

Fix formatting of ktr_user entries so that values of individual bytes
of ktrace record are treated as unsigned char, and printed using "%02x".
Change suggested in bin/17832 by Paul Goyette.

It's easy for ktrace can generate trace files that contain data from
multiple processes, so let's add a "-p pid" option to kdump so that it
can pull out the data for one process easily.

Fix translation in the native emulation. Noted by veego.

Provide hooks for signal number to name translation, and use them for kill
and PSIG.

XXX[1]: This shows that signal emulation ktrace is slightly busted: posted
	signals should really be translated to the emulated ones instead of
	producing ktrace records with the signal numbers of the native
	emulation.
XXX[2]: There are other places where signal names can be displayed, but this
	is not done yet.

reverse the notion of fancy, use plain, and increment it so that we can
have multiple levels of plainness. I want to be able to display syscalls
numerically, for that use -nn. This is not documented as it is an wip.

Cosmetic fix so that RET record after CALL execve get the appropriate syscall
even if the emulation changed

adapt to KTR_USER structure changes

avoid using "errno" as struct member name, to avoid conflict with errno.h.
ok by jdolecek

add support for KTR_USER records

map linux ptrace op (number) to it's appropriate name - this is kind of hack,
but turned out to be pretty usable for me

Apply patch (requested by he):
  Fix format / arg mismatch.

Pullup usr.bin string format fixes [is].
See "cvs log" for explicit revision numbers per file, from sommerfeld.

Format string cleanups

Prevent "kdump" from causing a segmentation fault if the io buffer that
it's printing to stdout ends on a page boundary. Patch supplied by
Andrew Brown in PR bin/9983.

put setemul(), ectx_* & appropriate structures to separate file, to make it
more easily sharable with ktruss(1)

Keep track of what pid uses which emulation and ensure proper
structures are used when decoding the ktrace records. This makes
kdump of ktrace -i output of a process executing binaries of various
emulations actually useful - before, the code used emulation structures
appropriate for last recently executed process, which is bogus.

Cast timeval members to types we know the printf conversions of.

Add COMPAT_NETBSD32 support.

Don't truncate syscall return values on LP64 platforms.

Adjust to Eric's recent errno renaming.

Fix emulation errno translation.

Update for signal handling changes.

use static int instead of static

add support for freebsd emulation. from Dave Huang <khym@bga.com> [bin/4293]

WARNSify mkioctls output, fix .Nm usage, deprecate register

add explicit return types and prototypes for local functions
use __COPYRIGHT and __RCSID macros, fix other gcc -Wall warnings

* NULL => 0 (Arne Juul; PR#3629)
* Nuke spacoids

fix decode of _IOR/_IOW/_IOWR: 'R' and 'W' were switched (i.e. _IOW ioctls
would print out as '_IOR'), and _IOWR ioctls were being printed as
_IORW (which doesn't exist).

eliminate multiple inclusion of <sys/errno.h>.

Decode unknown ioctl commands to _IO{R,W,}('c',x[,y]), instead of 0xabcdef00

Sync with 4.4lite2

imported from 44lite2

Add support for displaying and switching between
different OS system call emulations

only print MAXCOMLEN chars of the process command name.  fixes pr 1080.

 don't get syscalls.c with an absolute path

KERNEL -> _KERNEL

update for the new data stored in ktrace files:
	(1) use argsize rather than nargs
	(2) print long-sized (assumed == register_t-sized; XXX) items
		rather than int-sized ones.

Make two variables non-static.

Update to use timer{add,sub}().

Some KNF, POSIX, and err*(3)/warn*(3) cleanup.  Also, use sys_signame[]
rather than a private, home-grown list.

Add RCS ids.

netbsd does not define PT_STEP for architectures that lack
in-kernel single stepping capabilities.

updated kdump

Initial revision