CVS log for src/sbin/savecore/savecore.c

Up to [cvs.NetBSD.org] / src / sbin / savecore

Request diff between arbitrary revisions

Keyword substitution: kv
Default branch: MAIN

Mostly merge changes from HEAD upto 20200411

Merge changes from current as of 20200406

Avoid overflows when reading strings.

elide recently introduced warnings that trigger in normal cases, eg:

savecore: dump_exists:471: kvm_read newdumpmag: _kvm_kvatop(ffffffff8180030c)

Sync with HEAD

Synch with HEAD

avoid infinite loop in kmem_check().  fixes bug introduced in previous.

Sync with HEAD, resolve a couple of conflicts

put a bunch of the kvm_read + warn on failure code into a macro that
describes more about what failed.  now errors tell you which actual
variable was being requested instead of simply saying "not yours".

tested on amd64 as working.  written for arm64 testing.

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")

resync from head

- avoid using globals where it is simple.
- simplify: only use stdio for output.

sync with head

factor out rawname() from dump, fsck, savecore.

print the missing flag if missing argument...

- always print corrupted dump messages instead of silently failing
- don't require /var/crash to be specified, default to it
- determine and print the kernel name
- print all messages to syslog

static + __dead

sync to netbsd-5

Pull up following revision(s) (requested by dogcow in ticket #1157):
	sbin/savecore/savecore.c: revision 1.81
Instead of exiting with an obscure error message if -N /kernelname isn't
specified, blithely assume the kernel will consume around 20 megs.

Instead of exiting with an obscure error message if -N /kernelname isn't
specified, blithely assume the kernel will consume around 20 megs.

Sync with HEAD.

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

fix sign-compare issues

Pull up following revision(s) (requested by tls in ticket #598):
	sbin/savecore/savecore.c: revision 1.79
Fix savecore so it uses /dev/ksyms when it can (doesn't override libkvm
default).  Fixes misc/38425.

Fix savecore so it uses /dev/ksyms when it can (doesn't override libkvm
default).  Fixes misc/38425.

Sync with HEAD.

fix dev_t format.

Enable the faked ELF kernel by default, now that ksyms output works with
gdb. Makes it possible to do post-mortem debugging on a modular kernel.

- Add code to produce a fake ELF kernel containing only symbol info. Done
  by extracting ksyms state from the dump, so we have symbols for modules
  loaded at the time of the crash. If ksyms extraction fails, this falls
  back to the current behaviour of copying /netbsd.

- Leave it disabled for the moment: (1) we are about to branch (2) most
  tools understand the artificial kernel but gdb has trouble with it for
  some reason.

Explicitly check that the dump device is not the console, a tty or pty.
While the list is adhoc, the problems reported are always with
/dev/console. Adresses PR 38425 and similiar issues with Xen.

Sync with HEAD.

Sync with wrstuden-revivesa-base-2.

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

Sync with HEAD.

sync with head.

sync() a couple of times to benefit diagnosis of sloppy kernel code that
crashes during boot to multiuser.

sync with HEAD

Sync with HEAD.

Don't use the block device to read crashdumps, if possible.

sync with HEAD

tyop: _time_seconf -> _time_second

from Anon Ymous: remove (void)&var;

c99 initializers

accept either "time_second" or "time" as kernel timevariable in
order to support common userland in enviroments where
timecounter kernels and non timecounter kernel co-exist
(e.g. xen & i386)

support timecounters (struct time gone, use time_second then)

Sync with head.

Free some allocated memory in the failure case when trying to
fetch the panic string.

Fixes Coverity CID 1856.

If we have timecounters, read time_second from the kernel for the
dump time instead of time.tv.sec.

Added -n option, which tests for a present core but doesn't write it
to disk or clear it (does 'nothing').

http://mail-index.netbsd.org/tech-userlevel/2004/10/09/0000.html

Apply patch (requested by dbj in ticket 1724):
Fix problems when building the netbsd-1-6 branch with GCC 3.3.3.

Pull up revision 1.64 (requested by tls in ticket #650):
Fix an integer overflow that prevented saving cores from machines with
more than 2GB of RAM.
"dumpsize" in the on-disk core header is in pages, but, unfortunately,
c.size in our kcore format is an unsigned 32-bit int, gross.
We sleaze it, using a 64-bit int internally here and handing it to
libkvm as a *signed* 32-bit int.  Won't always work, but shouldn't be
broken any worse than it is now, and sometimes work better.
We must fix the kcore format and libkvm.

Fix an integer overflow that prevented saving cores from machines with
more than 2GB of RAM.

"dumpsize" in the on-disk core header is in pages, but, unfortunately,
c.size in our kcore format is an unsigned 32-bit int, gross.

We sleaze it, using a 64-bit int internally here and handing it to
libkvm as a *signed* 32-bit int.  Won't always work, but shouldn't be
broken any worse than it is now, and sometimes work better.

We must fix the kcore format and libkvm.

revert accidental commit to branch

Fix an integer overflow that prevented saving cores from machines with
more than 2GB of RAM.

"dumpsize" in the on-disk core header is in pages, but, unfortunately,
c.size in our kcore format is an unsigned 32-bit int, gross.

We sleaze it, using a 64-bit int internally here and handing it to
libkvm as a *signed* 32-bit int.  Won't always work, but shouldn't be
broken any worse than it is now, and sometimes work better.

We must fix the kcore format and libkvm.

Replace the statfs() family of system calls with statvfs().
Retain binary compatibility.

Give a 60 day leeway instead of a week when judging if the dumptime is valid.
7 days is too short when you are debugging time problems.

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

Patches provided by Joel Baker in PR 22308, verified by myself.

simplify by strlcpy

Remove prototype for non-existent function log().

use humanize_number(3) for progress printout

PR/15502: Wolfgang Rupprecht: savecore fails for > 2GB files.

Change type of dumpmag and newdumpmag from int to u_int32_t
since they hold 32bit unsigned magic numbers. (sync with kernels)

fix a couple of -Wshadow warnings

explicitly set the default compression level to 1 (from the implicit 6),
because it's significantly faster and doesn't use that much more disk space.

Add "-Z level" to control the compression level that -z uses (which
defaults to -Z 6).  Depending on the relative speed of the CPU
versus disk, "-zZ1" might be faster than no compression at all.

Use getbootfile(3) instead of _PATH_UNIX by default.
Sprinkle some const.

If the position of the formatted panic string can be found in the
message buffer, use that instead of the raw panic string.

Suggested by Jason Thorpe eons ago.

Make option flag variables static. This avoids a conflict between
'int compress' in savecore.c and the function 'compress' in libz.
Gnu ld 2.10 (with BFD 2.10) used on sparc64 warns this conflict
(symbol "compress" changed size).

use gzip instead of compress

Bump the field width of `K to go' from 6 to 8; some systems have more
than 999999 KB of RAM... (this will suffice until it's common to have
~ 100 GB of RAM :)

Back out sysctl part for now, awaiting the result of further discussion,
on request of Andrew Doran.

Output proper warning if dumplo == -1 (as it can happen on macppc without
swap partition), instead of "kvm_dump_mkheader: Pread: Invalid argument".

de-__P(), and generally try to be more conforming to the new NetBSD style.
Use %m in syslog strings where appropriate.
Reduce warnings with WARNS=2.

Fix argument handling, which was pretty broken.
If sysctl supports it, try to get the kernel name with CPU_BOOTED_KERNEL.
Get current kernels version string in all cases.
Adapt some error messages to the correct kernel name.
Reviewed by Simon Burge.

initialize kernel to _PATH_UNIX and remove local variables in functions
which are essentially copies of kernel.

savecore was looking up dumpdev, etc, from the wrong kernel

pull up 1.41 (approved by jhawk):
>time is a `struct timeval' not a `time_t', so read it in properly.

time is a `struct timeval' not a `time_t', so read it in properly.

s/on device/in [dirname]/, to be a bit clearer...

pullup 1.38->1.39 (cgd)

Remove an extraneous fclose() call which would cause savecore to crash
if the bounds file was unwritable (e.g. because /var was read-only).
Fix supplied by Ronald Khoo in PR#7887.

Crash dumps should *not* be world readable.

Avoid truncating partition size in freespace calculation

Need <time.h> for time() and ctime() prototypes.

Avoid

resolve conflicts from lite-2 merge

imported from lite-2

make these compile on the alpha after WARNS=1.

* cleanup for WARNS=1
* fix .Nm usage
* deprecate register

Lseek(2) usage cleanup: the use of L_SET/L_INCR/L_XTND is deprecated,
use SEEK_SET/SEEK_CUR/SEEK_END instead.

More error checking.

set umask for savecore.

make datas item copied out from kernel (panicstr) and/or used as kvm pointers
(panicloc) be 'long's, not 'int's.

If kvm_dump_mkheader() returns 0, simply state 'no core dump' instead
of logging and error.

Omit kd_live argument to kvm_dump_mkheader().

New libkvm/savecore implementation. With the following differences:
     1) savecore will not access the dump or live-kernel directly. It
        will always use the kvm-functions. Allowing it to work on kernels
        that don't have a 1-1 PA-VA mapping.
     2) the kvm-lib has some additional functions to accomplish 1
          - kvm_dump_mkheader()
          - kvm_dump_wrtheader()
          - kvm_dump_inval()
     3) the file formats of the dump generated by the kernel and the dump
        generated by savecore have been changed. The file format now looks
        like the format produced for 'normal' core dumps.
Ports not yet supporting the new kvm-format will be using libkvm.old/
savecore.old for the time being.

Put the compression suffix *last*.

set dumpsize in dump_exists(), not save_core(), so that minfree check
works again.  solves pr 1257, but sets dumpsize earlier than suggested
in that pr, so that dumpsize is set if a core doesn't appear to be
present (because dumpsize would be used if the '-f' flag is given).

fix verbose printf of dumplo, clean up printfs slightly

convert to new RCS Id conventions; reduce my headache

Fix getopt string and test on return value.

 oops; get rid of previous (iffy) zopen declaration

make a proto for zopen (yech)

Eliminate uses of some obsolete functions.

prototype zopen()
NBPG --> getpagesize()

vmunix --> kernel (for variable names), and make the core file naming
convention match that of other executables.

Import 4.4-Lite version.

Fix up RCS ids.

Update to 4.4-lite, local changes to man page.

Initialize return value from save_core() in case dumpsize == 0.

Don't clear dump if writing it to the filesystem fails (eg. file system full)

queue/struct changes

kill two casts

serious perversion:
most of the 'off' arguments to Lseek() (the internal lseek() wrapper)
are unsigned longs.  we need them to be sign extended, though, so change
Lseek() to take a 'long' rather than an 'off_t', and let the compiler
do the dirty work via implicit casts.

lseek stuff, and clean up.  this is perverse.

correct ok() calculation for all architectures.

Check for dumpdev being NODEV. If so, exit silently, unless verbose is on.

Add RCS identifiers.

changed "Id" to "Header" for rcsids

added rcs ids to all files

initial import of 386bsd-0.1 sources

Initial revision