openldap-bugs

openldap-bugs@openldap.org

1 participants
21001 discussions

(ITS#5007) Compilation problems
by h.b.furuseth＠usit.uio.no 09 Jun '07

09 Jun '07

Full_Name: Hallvard B Furuseth Version: HEAD, RE23 OS: HP-UX, OSF1, non-gcc URL: Submission from: (NULL) (129.240.202.105) Submitted by: hallvard HEAD/RE23: liblber/sockbuf.c uses pointer arithmetic on void*, a gcc extension. HEAD: OSF1 pthread.h defines "leave" as an exception keyword, which breaks slapd and bdb. I'll rename "leave" labels to "done" or "finish". liblutil/getpeereid.c breaks if LDAP_PF_LOCAL_SENDMSG is defined but MSG_WAITALL is not. (Not sure why that's compiled at all after the ldapi trouble, but that's another matter.) I'll fix it by #ifdeffing MSG_WAITALL in getpeereid.c rather than where LDAP_PF_LOCAL_SENDMSG is defined, since we need LDAP_PF_LOCAL_SENDMSG to decide whether to send a dummy abandon to previous OpenLDAP server versions.

1 0

Re: (ITS#4936) default modulepath
by h.b.furuseth＠usit.uio.no 09 Jun '07

09 Jun '07

ando(a)sys-net.it writes: > It would be pretty straightforward to call lt_dlsetsearchpath() > (actually, the module_path() wrapper) from module_init() with a default > value. This could be, for example, LDAP_LIBEXECDIR, where slapd itself > gets installed by default. Use LDAP_LIBEXECDIR/openldap/, that's where modules get installed. Try with ./configure --enable-overlays=mod --enable-monitor=mod. -- Regards, Hallvard

1 0

Re: (ITS#4996) Use SRV records to locate local server for command-line clients
by ando＠sys-net.it 09 Jun '07

09 Jun '07

Committed to HEAD. Command-line clients now use DNS SRV to look up the host names related to a DN and build a list of URIs for subsequent use. To trigger, use -H "<proto>:///DN". The <host> portion must be empty, and a DN must be present. The DN must be in the "dc=<component>" form; it's turned into a domain and the host list is looked up using DNS SRV. After that, a list of URIs is built using the returned host/ports, and using <proto> as the LDAP scheme. Please test. p. Ing. Pierangelo Masarati OpenLDAP Core Team SysNet s.r.l. via Dossi, 8 - 27100 Pavia - ITALIA http://www.sys-net.it --------------------------------------- Office: +39 02 23998309 Mobile: +39 333 4963172 Email: pierangelo.masarati(a)sys-net.it ---------------------------------------

1 0

Re: (ITS#4936) default modulepath
by ando＠sys-net.it 09 Jun '07

09 Jun '07

It would be pretty straightforward to call lt_dlsetsearchpath() (actually, the module_path() wrapper) from module_init() with a default value. This could be, for example, LDAP_LIBEXECDIR, where slapd itself gets installed by default. Comments? p. Ing. Pierangelo Masarati OpenLDAP Core Team SysNet s.r.l. via Dossi, 8 - 27100 Pavia - ITALIA http://www.sys-net.it --------------------------------------- Office: +39 02 23998309 Mobile: +39 333 4963172 Email: pierangelo.masarati(a)sys-net.it ---------------------------------------

1 0

Re: (ITS#4943) tpool.c pause vs. finish
by h.b.furuseth＠usit.uio.no 09 Jun '07

09 Jun '07

I wrote: > I'll just make [the hash table] chained. (...) > Solves the multi-pool problem too. No it doesn't. pool_pause() would stilll only pause one pool, while purgekey() needs all pools to be paused. I guess I'll chain it anyway. I forgot one bug in my summary: Slapd does not always check if pool_submit() and pool_setkey() succeed, even though failure is not a bug. (They can run into some limits.) pool_submit(): daemon.c:slapd_daemon_task() with rtask->routine connection.c:connections_shutdown() with c_clientfunc connection.c:connection_read() with c_clientfunc back-bdb/tools.c:bdb_tool_entry_open() with trickle and index tasks pool_setkey(): connection.c:connection_fake_init2() for slapi object extensions backend.c:be_isroot_pw() for SLAPD_SPASSWD passwd.c:slap_passwd_check() for SLAPD_SPASSWD sl_malloc.c:slap_sl_mem_create() back-bdb/search.c:search_stack() back-meta/conn.c:meta_back_candidates_get() slapd-modules/smbk5pwd/smbk5pwd.c:smbk5pwd_op_bind() -- Regards, Hallvard

1 0

Re: (ITS#4986) manageDSAit might control cause problems in database selection
by ando＠sys-net.it 09 Jun '07

09 Jun '07

A tentative fix is in HEAD (servers/slapd/backend.c 1.377 -> 1.378). Please test. p. Ing. Pierangelo Masarati OpenLDAP Core Team SysNet s.r.l. via Dossi, 8 - 27100 Pavia - ITALIA http://www.sys-net.it --------------------------------------- Office: +39 02 23998309 Mobile: +39 333 4963172 Email: pierangelo.masarati(a)sys-net.it ---------------------------------------

1 0

Re: (ITS#4943) tpool.c pause vs. finish
by h.b.furuseth＠usit.uio.no 09 Jun '07

09 Jun '07

hyc(a)symas.com writes: > h.b.furuseth(a)usit.uio.no wrote: >> * When a thread is finishing, make it go active for context_reset(). >> Should make slapi module destructors safer, and I don't see that it >> can hurt. > > I don't understand this. Let me dig a bit back in the ITS... "context_reset() can call connection_fake_destroy(), which via slapi_int_free_object_extensions() calls slapi module destructors, which can do who-knows-what." >> * I'd like to rename thread contexts to tasks(?) and user contexts back >> to contexts (as in RE23). The current terminology is confusing. > > I don't care about this either way. Then I will. I still confuse myself sometimes, and I certainly would if I came back to this code next year. >> (...) > Never use malloc unless absolutely necessary. We have enough problems > with heap fragmentation already. Aha. OK. >> * Scheduling: If several threads call pool_pause(), then once there is a >> pause tpool does not schedule them all. They could get handled then, >> or another thread could undo the pause so tpool would wait to pause >> again. Is that deliberate? > > I don't understand the question. The point of the pause is to prevent > any other thread (in the pool) from running. Why should tpool schedule > any other threads at this point? Two threads call pool_pause(). Eventually the pool gets paused, and one pool_pause() call returns. When that thread calls pool_resume(), the other thread waiting in pool_pause() may or may not get scheduled. >> - pool_context() breaks if several ldap_pvt_thread_t bit patterns can >> represent the same thread ID: TID_HASH() would give different hashes >> for the same thread, and pool_context() stops searching when it hits >> a slot with ctx == NULL. (My previous bug report was a bit wrong.) > > How can a single thread ID be represented by multiple bit patterns? A struct/union with padding bytes seems the least unlikely possibility. A pointer in hardware where several address representations map to the same physical address, and the compiler handles that by normalizing pointers when they are compared/subtracted. Like DOS "huge" pointers would have been if the compiler normalized them when comparing them instead of when incrementing/decrementing them. 20-bit address bus, 32-bit pointers, physical address = 16 * <segment half of pointer> + <offset half of pointer>. Or, not that I take this too seriously: A pthread implementation which uses e.g. a 32-bit integer type for thread IDs but just ignores the top 22 bits. The Posix spec seems to allow that, and thread IDs are to be compared with pthread_equal(). >> The best fix would be to use use real thread-specific data instead. >> Just one key with the ctx for now, that minimizes the changes. OTOH >> it also means we'll do thread-specific key lookup twice - first in >> pthread to get the ctx, and then in ltu_key[] to find our data. >> Anyway, I said I'd do that later but might as well get on with it, >> at least for pthreads. Except I don't know if that's OS-dependent >> enough that it should wait for RE24? > > The best fix may be to just use the address of the thread's stack as the > thread ID. That's actually perfect for our purpose. How does a function called from somewhere inside the thread know that address? That's almost what the user context is, and thus what thread_keys[] maps the thread ID to. >> - thread_keys is a poor hash table implementation with a poor hash >> function. > > Ideally it would not be a hash table at all, but a direct lookup based > on the thread ID. If we can. See above. Anyway, I don't know why I keep kvetching about the hash table implementation. I'll just make it chained. As I finally figured out, that can be done without mallocs. Solves the multi-pool problem too. >> * I wonder if there are long-lived ldap_pvt_thread_cond_wait(), >> [r]mutex_lock() or rdwr_[rw]lock() calls which should make the thread >> go inactive first (after waiting for any pause to finish), so that >> (a) a pool_pause() will not block waiting for them, and >> (b) all threads in the pool can't be blocked in long waits without >> tpool knowing about it. > > Long-lived lock calls would be a quite noticeable bug. This seems to be > a non-issue. Well, it'd have to be with code which is very rarely executed, or unusual configurations. But yes, I feel relaxed enough about it. >> * back-bdb/tools.c uses pool_submit() to start long-lived threads: >> (...) > Tool threads are completely separate from slapd threads. This is not an > issue. Fine. I don't understand that in this context, but I'll take your word for it:-) >> * slapd/bconfig.c maintains its own idea of the max number of threads >> in connection_pool_max and slap_tool_thread_max, which does not match >> tpool's idea if the limit is <=0 or larger than LDAP_MAXTHR. >> >> Since slapd acts on its own count, I imagine that can cause breakage. >> Simplest fix might be to set max #threads and then read it back, >> I haven't checked. Would need to return LDAP_MAXTHR instead of 0 >> for "as many as allowed". Note however also the issue above: >> tpool's idea of actually available threads can be too high too. > > The practical limits are actually pretty low. Even if you're on a 1024 > processor machine with many terabytes of RAM, it's not a good idea to > have many thousands of threads... Scheduler overhead would overshadow > any useful work. OK. I forgot how much RAM a slapd thread requires. But then it won't hurt to give slapd a max limit lower than LDAP_MAXTHR either. >> * I said my new ltp_pause waits might be too aggressive. Maybe I should >> clarify: I don't see any instances where they can be dropped, but it's >> possible that slapd expects pauses to be less aggressive somewhere. > > I don't understand this comment. Clarifying the clarification: I don't know slapd well enough to know if the extra pauses do something bad to it. >> Oh, the comment /* no open threads at all?!? */: That's not strange. >> Happens in the first pool_submit() call(s) if thread_create() fails. >> Though in that case there was no danger of removing the wrong ctx >> from ltp_pending_list after all (fixed in rev 1.63). > > (Sure, but thread_create() should never fail if you have a sane > configuration.) And if the rest of the machine doesn't run wild. I imagine thread_create() can fail when fork() can fail on some OSes. -- Regards, Hallvard

1 0

Re: (ITS#5003) issue about upper_needs_cast
by ando＠sys-net.it 09 Jun '07

09 Jun '07

caojunliang(a)huawei.com wrote: > when I set the option upper_needs_cast as Yes.It only takes effect on id_query > when adding an entry,but does not take effect on child id query when deleting an > entry. > > I found that in the src file back-sql/init.c > > line 374 --379 > > there are not operation about upper cast. > > I want to know if what I think is right. Not clear what's your point. "upper_needs_cast" is a hack to work around the fact that some RDBMSes need an explicit cast before passing a literal into whatever "uppercase" function they provide. Can you point out what's the problem and what's the expected behavior? I don't see any literal passed to the query that looks up the parent's and that needs any special treatment. A log of the offending operation, and an example of what you'd consider correct, would help. p. Ing. Pierangelo Masarati OpenLDAP Core Team SysNet s.r.l. via Dossi, 8 - 27100 Pavia - ITALIA http://www.sys-net.it --------------------------------------- Office: +39 02 23998309 Mobile: +39 333 4963172 Email: pierangelo.masarati(a)sys-net.it ---------------------------------------

1 0

Re: (ITS#4965) slapd stops if access to cn=monitor is restricted
by ando＠sys-net.it 09 Jun '07

09 Jun '07

This is now made consistent in HEAD: the inability to register custom monitor stuff always results in a failure. Note that the software provides different means to prevent this feature: a) allow anonymous read access to the monitor database b) define a rootdn (even without any possibility to auth) for the monitor database c) disable monitoring for all bdb/hdb databases p. Ing. Pierangelo Masarati OpenLDAP Core Team SysNet s.r.l. via Dossi, 8 - 27100 Pavia - ITALIA http://www.sys-net.it --------------------------------------- Office: +39 02 23998309 Mobile: +39 333 4963172 Email: pierangelo.masarati(a)sys-net.it ---------------------------------------

1 0

Re: (ITS#5005) test017 fails
by ando＠sys-net.it 09 Jun '07

09 Jun '07

dieter(a)dkluenter.de wrote: The problem is in slap_parse_sync_cookie(); when no cookie is passed, a string containing "rid=001" is parsed. The parser expects it to end with a comma. If passing "rid=001" is correct, the fix is trivial: instead of checking for (*next != ','), check for (*next && *next != ','). Index: servers/slapd/ldapsync.c =================================================================== RCS file: /repo/OpenLDAP/pkg/ldap/servers/slapd/ldapsync.c,v retrieving revision 1.42 diff -u -r1.42 ldapsync.c --- servers/slapd/ldapsync.c 18 May 2007 12:46:52 -0000 1.42 +++ servers/slapd/ldapsync.c 9 Jun 2007 09:42:24 -0000 @@ -180,7 +180,10 @@ if ( !strncmp( next, "rid=", STRLENOF("rid=") )) { rid_ptr = next; cookie->rid = strtoul( &rid_ptr[ STRLENOF( "rid=" ) ], &next, 10 ); - if ( next == rid_ptr || next > end || *next != ',' ) { + if ( next == rid_ptr + || next > end + || ( *next && *next != ',' ) ) + { return -1; } if ( *next == ',' ) { @@ -194,7 +197,10 @@ if ( !strncmp( next, "sid=", STRLENOF("sid=") )) { rid_ptr = next; cookie->sid = strtoul( &rid_ptr[ STRLENOF( "sid=" ) ], &next, 16 ); - if ( next == rid_ptr || next > end || *next != ',' ) { + if ( next == rid_ptr + || next > end + || ( *next && *next != ',' ) ) + { return -1; } if ( *next == ',' ) { I'm not committing this fix because I'm not sure it doesn't break anything else. p. Ing. Pierangelo Masarati OpenLDAP Core Team SysNet s.r.l. via Dossi, 8 - 27100 Pavia - ITALIA http://www.sys-net.it --------------------------------------- Office: +39 02 23998309 Mobile: +39 333 4963172 Email: pierangelo.masarati(a)sys-net.it ---------------------------------------

1 0

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

openldap-bugs