IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
18.
when you're looking at a level 10, and it's all clutered with
tdb_pack/unpack, it's getting .... And anyway most of our code using
tdb_pack/unpack have DEBUG around the call if there is a problem.
J.F.
- tdb_open api changed so that you now pass an error handling
callback when opening the file, so that errors detected during
opening have somewhere to go. (All calls from the body of Samba to
this function go through a wrapper in tdbutil, which has been
updated.)
- Clean up logic for deciding how to open tdb. Emit log messages if
something goes wrong (e.g. bad magic.)
- tdbtool now logs errors to stderr.
database, but no underlying system call sets errno.
The particular case I had was a mangled .tdb, but there are others.
For this one, set EIO. It's a shame Unix messages aren't more
detailed -- "bad data format" would be better.
means that a read-write opener and a read-only opener are using different
locking mechanisms - this needs to be addressed, but it's hard as the
read-write opener using the spinlocks is usually first, so there's no
way to force them to change down to the fcntl method.
Read only access is less important anyway and can never corrupt the
tdb anyway, so errors in read-only record reads are more tolerable.
Jeremy
map_ptr not neccessarily set to NULL if no mmap, secondly, iterating
through a tdb was using SETLK, instead of SETLKW - would almost never
fail with mmap as the time holding the lock was so short, but was
easily seen with read/write. One finaly bug needs tracking down w.r.t.
traversal......
Jeremy.
Ok - now we're no longer trying to reach a silly 1k loc target,
change the formatting to be *readable* - eg.
change if (x) y else z to be :
if (x)
y
else
z
and other compact sillyness. Oh look - when I did this I found
some areas where we *WEREN'T CHECKING SYSTEM CALL ERROR RETURNS !!!!*
CompSci 101 guys....... :-).
Jeremy.
was the initial cause of the connections database becoming corrupt.
Note that this bug only happens when doing deletions within a
traversal, which is why it has only showed up now
- added delete within traversal testing to tdbtorture
- added a lot more logging to tdb
source/include/proto.h
- make proto
source/printing/nt_printing.c
source/rpc_server/srv_spoolss_nt.c
- Fix for the overwriting of printerdata entries when WinNT and
Win2k are modifying printer parameters on PCL printers. Turns out
that Win2k creates a printer with a NULL devmode entry and then
expects to set it on *OPEN* (yes this is insane). So we cannot
return a "default" devmode for a printer - and we must allow an
open to set it.
source/tdb/tdb.c
- Show freelist in an easier format. Show total free.
- When storing a new record, allocate memory for the key + data
before the tdb_allocate() as if the malloc fails a (sparse) hole
is left in the tdb.
source/tdb/tdbtool.c
- Show freelist in an easier format. Show total free.
source/tdb/Makefile
- cleaned up Makefile dependancies
source/smbd/lanman.c
- Fix for Win9x corrupting it's own parameter string.
source/printing/printfsp.c
source/printing/printing.c
source/rpc_server/srv_spoolss_nt.c
source/smbd/close.c
- Added normal close parameter into print_fsp_end() which treats an
abnormal close as error condition and deletes the spool file.
struct tdb_traverse_lock tl = { tdb->travlocks.next, 0, 0 };
The IRIX compiler complained that the first initialization value was not a
constant and errored out. Here's what it looks like now:
struct tdb_traverse_lock tl = { NULL, 0, 0 };
int ret, count = 0;
/* This was in the initializaton, above, but the IRIX compiler
* did not like it. crh
*/
tl.next = tdb->travlocks.next;
That worked.
Chris -)-----
When chaining together long lines of bloody "if" statements, which should
logically be separated, and one of them allocates memory, remember to
*free* it *WHETHER OR NOT THE IF STATEMENTS SUCCEEDED* !!!!
Yes I do consider this a bug in the coding style of Tridge, Rusty, Tim et al. :-).
I'm just pissed 'cos this took 4 hours to track down even with an insure error report
stating me in the face and also Ben Woodward looking over the code with me :-).
Jeremy.
dead records, else the record is just marked for deletion, not actually deleted.
Second, ensure allocated record is marked as "in use" before free list lock is
released, else other processes in the freelist merge code may try and merge it.
Jeremy.
a --with-spinlocks option to configure, this does mean the on-disk tdb
format has changed, so 2.2alphaX sites will need to re-create their
tdb's. The upside is no more tdb fragmentation and a +5% on netbench.
Swings and roundabouts....
Jeremy.
