- openldap-bugs - openldap.org

Re: (ITS#4033) Enhancement requests : Need LDIF parsing API's
by rmeggins＠redhat.com 04 Jan '10

04 Jan '10

See also http://www.openldap.org/its/index.cgi/Software%20Enhancements?id=6194;selec… An API that returned LDAPMod ** and LDAPControl ** would be sufficient for all cases, including the LDAP ADD and any LDAP modify cases. The current discussion revolves around exposing the high level LDAPMod and LDAPControl APIs to the lower level libldif, or just passing the parsed values as lists of struct berval *

1 0

Re: (ITS#6437) sl_malloc issues
by h.b.furuseth＠usit.uio.no 04 Jan '10

04 Jan '10

hyc(a)symas.com writes: > Nobody uses the pool code. I would suggest you profile it against the stack > code before spending any time fixing it. The pool version crashes if I divide SLAP_SLAB_SIZE by 16 or multiply the allocated amounts by 17. The stack version stays happy. I also tried RE23 and rev 1.48 in case of code rot. (Should have tried OpenLDAP 2.2, but would need to install an old Berkeley DB.) Adding a sl_malloc fallback to ch_malloc instead of returning NULL didn't help. Anyway, I tried this because I assume the pool version is intended to reduce the number of sl_malloc fallbacks to ch_malloc, but that almost never happens in 'make test' with the current SLAP_SLAB_SIZE. Which made that aspect hard to test. Also the pool version does a number of ch_mallocs for administration and ch_frees them at reset. If the idea is to reduce ch_mallocs, it should keep these blocks around. (Or allocate them with the slab.) Except the 'slab_object's It only needs cleanup if the slab size changes, which slapd never seems to do. -- Hallvard

1 0

Re: (ITS#6437) sl_malloc issues
by h.b.furuseth＠usit.uio.no 02 Jan '10

02 Jan '10

hyc(a)symas.com writes: > h.b.furuseth(a)usit.uio.no wrote: > Go ahead and commit. In working pieces if necessary. We'll pick and choose... Committed one batch of fixes/tweaks, hopefully independent after the first one or two. Next batch will have to walk all over some of these. >> * Why is there a slap_sl_mem_detach() when there is no way to reattach >> it? As far as I can tell all the caller can do is to inspect/clean up >> the returned slab_heap itself. > > It was originally implemented in 2.2 to support psearch for sync replication. > It was obsoleted when the syncprov overlay was written in 2.3, we just never > deleted it. Go ahead and delete it. Actually I just realized the context can be used thread-less just fine, even if it's only relevant if any 3rd party modules do it. Might as well keep it in RE24, just in case. But I'll move structs slap_object and slap_heap back to sl_malloc.c, so such modules cannot mess with them. > (And remember, the point of sl_malloc is that a heap here doesn't need > any special cleanup - just free it and forget it.) Not quite, due to the ch_malloc fallbacks. Anyway, I've committed some comments with my understanding of the package. >> * Some failures assert(0)/exit() like ch_malloc.c, but without giving a >> Debug() message first. > > don't really care... I do, I've had silent exits from slapd before and hated it. -- Hallvard

1 0

Re: (ITS#6437) sl_malloc issues
by h.b.furuseth＠usit.uio.no 02 Jan '10

02 Jan '10

hyc(a)symas.com writes: > Go ahead and commit. In working pieces if necessary. We'll pick and choose... Thanks. Will do. > Nobody uses the pool code. I would suggest you profile it against the stack > code before spending any time fixing it. Right, that and detach was what I wondered most about. Beyond just "with enough changes, some will be bugs":-) Well, the current pool code blocks the "ber_len_t -> 32-bit" space optimization, which would import the alignment problem to 64-bit hosts. Will profile first. When I get time. >> * Regarding code duplication, the #ifdef SLAP_NO_SL_MALLOC code >> duplicates the !ctx code. Make that if(<enum No_sl_malloc> && !ctx) >> and the #ifdefs can be killed. > > I would personally prefer no runtime checks, and all compile-time checks. Right, but if(compile-time constant) is a compile-time check on a decent compiler. Otherwise we shouldn't be using '#define foobar do { ... } while(0)' trick to allow ';' after a macro. -- Hallvard

1 0

Re: (ITS#6437) sl_malloc issues
by hyc＠symas.com 01 Jan '10

01 Jan '10

h.b.furuseth(a)usit.uio.no wrote: > Full_Name: Hallvard B Furuseth > Version: HEAD > OS: Linux x86_64 > URL: > Submission from: (NULL) (129.240.6.233) > Submitted by: hallvard > > > Here are a number of bugs and other issues with slapd/sl_malloc.c, and > one for ber_memcalloc. Some for RE24, likely some for RE25, but I don't > know where to cut off for RE24? (Waiting for someone to say...) Go ahead and commit. In working pieces if necessary. We'll pick and choose... > > * It claims to align on 2*sizeof(int) boundaries, but it looks like > stack:realloc and pool:mem_create can un-align to (2n+1)*sizeof(int) > boundaries on 32-bit hosts, where sizeof(ber_len_t) == sizeof(int). > > A slap_sl_realloc() fix for the stack implementation is to replace > size += pad + sizeof( ber_len_t ); > size&= ~pad; > with > size = ((size + pad)& ~pad) + sizeof( ber_len_t ); > however see the optimizations and tweaks below. Fine. > For the pool implementation, maybe slap_sl_mem_create() can just set > so->so_ptr = (char *) sh->sh_base + sizeof(ber_len_t) % (pad+1); > Unless that messes up its expectations of how much space is left, I > haven't checked. Should mem_create size be increased accordingly? Nobody uses the pool code. I would suggest you profile it against the stack code before spending any time fixing it. > Unfortunately sl_malloc.c exports its internals, so I don't know what > kind of fix would affect it: > > * Why is there a slap_sl_mem_detach() when there is no way to reattach > it? As far as I can tell all the caller can do is to inspect/clean up > the returned slab_heap itself. It was originally implemented in 2.2 to support psearch for sync replication. It was obsoleted when the syncprov overlay was written in 2.3, we just never deleted it. Go ahead and delete it. (And remember, the point of sl_malloc is that a heap here doesn't need any special cleanup - just free it and forget it.) > * slap_sl_mem_create() switching implementation stack<->pool on an old > memctx slab can crash or leak, it runs the wrong cleanup code. > > Fix: Remove code duplication, rearrange. The pool cleanup duplicates > slap_sl_mem_destroy(), except the final frees. Let destroy(key=NULL,) > mean don't free and call that. Also the create/destroy pool branches > duplicate the stack branches. > > * However, will the pool (i.e. non-stack) implementation ever be used? > If not, a simpler fix to some of this is to delete it. OTOH if it > someday may be used, hopefully some of these cleanups will be leaving > it a bit more correct, easier to read, and faster. Though I haven't > gone to the drastic step of studying it to see how it works. > > Or we could #ifdef it out by default. It's unclear to me why this is > a run-time argument to the create function instead of instead of a > -D<compiler flag>, why would the caller care? Dunno. I guess it was for quicker turnaround during testing. But afaik only Jong ever tested/used it, and no one else has spent any time on it. #if'ing it all out is fine with me. Of course, if you find that it's actually better performing than the stack version, then we can #if out the stack code instead. > > #define SLAP_SLAB_MODE<undef or 1:stack, 0:pool: -1:support both>. > #define SLAP_SLAB_ISSTACK(stack)<stack if both, else SLAP_SLAB_MODE> > > * Regarding code duplication, the #ifdef SLAP_NO_SL_MALLOC code > duplicates the !ctx code. Make that if(<enum No_sl_malloc> && !ctx) > and the #ifdefs can be killed. I would personally prefer no runtime checks, and all compile-time checks. > * Also we can throw the #ifdef NO_THREADS/#else/#endif code out to two > macros #define GET_MEMCTX()/SET_MEMCTX(). right. > * The "enough room?" test (ptr + size< endptr) is invalid, can e.g. > wrap around, when there is not enough room. Use (size< endptr-ptr). ok. > * Some failures assert(0)/exit() like ch_malloc.c, but without giving a > Debug() message first. don't really care... > * The slap_sl_malloc() pool implementation can return NULL on failure > at comment "FIXME: missing return; guessing...". For consistency it > should exit on failure like the rest of sl_malloc.c and ch_malloc.c. don't care, unless we decide to keep the pool code. Go ahead and commit. > * Realloc(last block) when the following free space is not big enough, > does not move sh_last and reclaim free space properly. > > * slap_sl_calloc()/ber_memcalloc() do not catch<count>*<size> overflow: > ber_len_t total = n * size; > if (!n&& total/n == size)<allocate...>; > > That can be optimized a bit, if anyone cares: > > #define LIM_SQRT(t) /* some value< sqrt(max value of unsigned type t) */ \ > ((0UL|(t)-1)>>31>>31> 1 ? ((t)1<<32) - 1 : \ > (0UL|(t)-1)>>31 ? 65535U : (0UL|(t)-1)>>15 ? 255U : 15U) > ... > /* sqrt test = slight optimization: often avoids the division */ > if ((n | size)<= LIM_SQRT(ber_len_t) || !n || total/n == size) ... > > * slap_sl_<malloc/realloc> increase size before proceeding, they should > reduce it before falling back to ch_malloc and for Debug messages. > > Space optimizations/cleanup: > > * To save space on 64-bit hosts: We can store store sizes (head, tail) > in 4-byte typedef ber_uint_t slap_sl_size_t instead of a ber_len_t. > > After fixing the misalignment issue above, however. > > * Also the stack implementation does not need to allocate the tail. > It can store the "block is freed" bit in the head of the _next_ block, > and only store a tail in free blocks (i.e. if that bit is set). > > I think that does add a bit of code, though it is convenient for the > "reclaim freed blocks at sh_last" code in slap_sl_free(). > > * Ignoring last point, the "reclaim" code can in any case be simplified: > /* mark it free */ > p[-1] = size |= 1; > /* reclaim free space off tail */ > if (sh->sh_last == p) { > do { > p = (slap_sllen_t*) ((char*) p - size + 1) - 1; > size = p[-1]; > } while (size& 1); > sh->sh_last = p; > } > > * Realloc can check if previous or next block is free before giving up > and doing malloc/copy/free. That does multiply its code size though. > Dunno if it's a good idea. Needs testing/timing, I guess. > > * Aligning with (size& ~pad) is wrong for non-two's complement signed > pad, it should be (size& -(pad + 1)) aka (size& -Alignment). > > All right, so I'm nitpicking. I just dislike unportable code which > isn't even simpler than the portable variant. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/

1 0

(ITS#6437) sl_malloc issues
by h.b.furuseth＠usit.uio.no 01 Jan '10

01 Jan '10

Full_Name: Hallvard B Furuseth Version: HEAD OS: Linux x86_64 URL: Submission from: (NULL) (129.240.6.233) Submitted by: hallvard Here are a number of bugs and other issues with slapd/sl_malloc.c, and one for ber_memcalloc. Some for RE24, likely some for RE25, but I don't know where to cut off for RE24? (Waiting for someone to say...) * It claims to align on 2*sizeof(int) boundaries, but it looks like stack:realloc and pool:mem_create can un-align to (2n+1)*sizeof(int) boundaries on 32-bit hosts, where sizeof(ber_len_t) == sizeof(int). A slap_sl_realloc() fix for the stack implementation is to replace size += pad + sizeof( ber_len_t ); size &= ~pad; with size = ((size + pad) & ~pad) + sizeof( ber_len_t ); however see the optimizations and tweaks below. For the pool implementation, maybe slap_sl_mem_create() can just set so->so_ptr = (char *) sh->sh_base + sizeof(ber_len_t) % (pad+1); Unless that messes up its expectations of how much space is left, I haven't checked. Should mem_create size be increased accordingly? Unfortunately sl_malloc.c exports its internals, so I don't know what kind of fix would affect it: * Why is there a slap_sl_mem_detach() when there is no way to reattach it? As far as I can tell all the caller can do is to inspect/clean up the returned slab_heap itself. * slap_sl_mem_create() switching implementation stack<->pool on an old memctx slab can crash or leak, it runs the wrong cleanup code. Fix: Remove code duplication, rearrange. The pool cleanup duplicates slap_sl_mem_destroy(), except the final frees. Let destroy(key=NULL,) mean don't free and call that. Also the create/destroy pool branches duplicate the stack branches. * However, will the pool (i.e. non-stack) implementation ever be used? If not, a simpler fix to some of this is to delete it. OTOH if it someday may be used, hopefully some of these cleanups will be leaving it a bit more correct, easier to read, and faster. Though I haven't gone to the drastic step of studying it to see how it works. Or we could #ifdef it out by default. It's unclear to me why this is a run-time argument to the create function instead of instead of a -D<compiler flag>, why would the caller care? #define SLAP_SLAB_MODE <undef or 1:stack, 0:pool: -1:support both>. #define SLAP_SLAB_ISSTACK(stack) <stack if both, else SLAP_SLAB_MODE> * Regarding code duplication, the #ifdef SLAP_NO_SL_MALLOC code duplicates the !ctx code. Make that if(<enum No_sl_malloc> && !ctx) and the #ifdefs can be killed. * Also we can throw the #ifdef NO_THREADS/#else/#endif code out to two macros #define GET_MEMCTX()/SET_MEMCTX(). * The "enough room?" test (ptr + size < endptr) is invalid, can e.g. wrap around, when there is not enough room. Use (size < endptr-ptr). * Some failures assert(0)/exit() like ch_malloc.c, but without giving a Debug() message first. * The slap_sl_malloc() pool implementation can return NULL on failure at comment "FIXME: missing return; guessing...". For consistency it should exit on failure like the rest of sl_malloc.c and ch_malloc.c. * Realloc(last block) when the following free space is not big enough, does not move sh_last and reclaim free space properly. * slap_sl_calloc()/ber_memcalloc() do not catch <count>*<size> overflow: ber_len_t total = n * size; if (!n && total/n == size) <allocate...>; That can be optimized a bit, if anyone cares: #define LIM_SQRT(t) /* some value < sqrt(max value of unsigned type t) */ \ ((0UL|(t)-1) >>31>>31 > 1 ? ((t)1 <<32) - 1 : \ (0UL|(t)-1) >>31 ? 65535U : (0UL|(t)-1) >>15 ? 255U : 15U) ... /* sqrt test = slight optimization: often avoids the division */ if ((n | size) <= LIM_SQRT(ber_len_t) || !n || total/n == size) ... * slap_sl_<malloc/realloc> increase size before proceeding, they should reduce it before falling back to ch_malloc and for Debug messages. Space optimizations/cleanup: * To save space on 64-bit hosts: We can store store sizes (head, tail) in 4-byte typedef ber_uint_t slap_sl_size_t instead of a ber_len_t. After fixing the misalignment issue above, however. * Also the stack implementation does not need to allocate the tail. It can store the "block is freed" bit in the head of the _next_ block, and only store a tail in free blocks (i.e. if that bit is set). I think that does add a bit of code, though it is convenient for the "reclaim freed blocks at sh_last" code in slap_sl_free(). * Ignoring last point, the "reclaim" code can in any case be simplified: /* mark it free */ p[-1] = size |= 1; /* reclaim free space off tail */ if (sh->sh_last == p) { do { p = (slap_sllen_t*) ((char*) p - size + 1) - 1; size = p[-1]; } while (size & 1); sh->sh_last = p; } * Realloc can check if previous or next block is free before giving up and doing malloc/copy/free. That does multiply its code size though. Dunno if it's a good idea. Needs testing/timing, I guess. * Aligning with (size & ~pad) is wrong for non-two's complement signed pad, it should be (size & -(pad + 1)) aka (size & -Alignment). All right, so I'm nitpicking. I just dislike unportable code which isn't even simpler than the portable variant. * pad and order_start can be compile-time enums instead of computed at run time (if any compilers do not optimize it away already): enum { Align = 2*sizeof(int), Align_log2 = 1 + (Align>2) + (Align>4) + (Align>8) + (Align>16), pad = Align - 1, /* but drop that, it's just as easy to use Align */ order_start = Align_log2 /* which also can be dropped */ }; ... assert(Align == 1 << Align_log2); /* Adding head+tail preserves alignment */ assert(2*sizeof(ber_len_t) % Align == 0); * The pool code can be cleaned up by basing 'order' on Align-sized chunks instead of byte-sized chunks, i.e. subtract order_start from sh->sh_maxorder, order/i/j variables. Also subtract 1 from mem_create:order, so that 'order' means the same as 'order' elsewhere. (Straightforward replacement, no need to invoke the brain and understand the code:-) * The pool implementation sometimes does 1<<foo where it should do (ber_len_t)1<<foo or 1U<<foo.

1 0

Re: (ITS#4868) Binary Attribute Patch(es)
by hyc＠symas.com 30 Dec '09

30 Dec '09

vargok(a)yahoo.com wrote: > (I realize this is ridiculously late, but... well, better late than never, and this is the first I've gotten back to it.) > > Yes! It's working now in 2.4.x as expected, via the new userCertificate > stub data provided by example. FWIW, it appears to be working fine with the PostgreSQL example files as well. > > Unfortunately, it was never "back" ported to 2.3. We're currently using 2.3.43 as part of RHEL-5. Any chance it could get back-ported into the 2.3 branch? Development on 2.3 ended about a year ago. > Thanks, > Kevin > > > > ----- Original Message ---- >> From: Pierangelo Masarati<ando(a)sys-net.it> >> To: vargok(a)yahoo.com >> Cc: openldap-its(a)openldap.org >> Sent: Wed, April 11, 2007 7:06:57 PM >> Subject: Re: (ITS#4868) Binary Attribute Patch(es) >> >> ando(a)sys-net.it wrote: >> >>> Right. Time permitting, I'll check why back-sql behaves differently >>> and, in case, how to fix it. >> >> HEAD should be working as expected, now. I've also updated mysql and >> psql sql-test000 (the latter is untested, though). >> >> Please test and report. 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 >> --------------------------------------- -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/

1 0

Re: (ITS#4868) Binary Attribute Patch(es)
by vargok＠yahoo.com 30 Dec '09

30 Dec '09

(I realize this is ridiculously late, but... well, better late than never, and this is the first I've gotten back to it.) Yes! It's working now in 2.4.x as expected, via the new userCertificate stub data provided by example. FWIW, it appears to be working fine with the PostgreSQL example files as well. Unfortunately, it was never "back" ported to 2.3. We're currently using 2.3.43 as part of RHEL-5. Any chance it could get back-ported into the 2.3 branch? Thanks, Kevin ----- Original Message ---- > From: Pierangelo Masarati <ando(a)sys-net.it> > To: vargok(a)yahoo.com > Cc: openldap-its(a)openldap.org > Sent: Wed, April 11, 2007 7:06:57 PM > Subject: Re: (ITS#4868) Binary Attribute Patch(es) > > ando(a)sys-net.it wrote: > > > Right. Time permitting, I'll check why back-sql behaves differently > > and, in case, how to fix it. > > HEAD should be working as expected, now. I've also updated mysql and > psql sql-test000 (the latter is untested, though). > > Please test and report. 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#6322) slapd suddenly stops working, and starts using 100% CPU
by helge＠monsternett.no 30 Dec '09

30 Dec '09

On Wed, Dec 30, 2009 at 09:04:12AM +0100, Helge Milde wrote: >I just enabled trace loglevel, perhaps that might give us a clue on what's going on. Slapd just stopped working again. Here's the (annotated) log for viewing and/or downloading: http://www.pastie.org/761198.txt Does this give you any clues? Tell me if I can do anything else to help. -- Helge Milde, 69701808 www.monsternett.no

1 0

Re: (ITS#6322) slapd suddenly stops working, and starts using 100% CPU
by helge＠monsternett.no 30 Dec '09

30 Dec '09

On Tue, Dec 29, 2009 at 11:00:25AM -0800, Howard Chu wrote: >Sorry, your issue was closed since we hadn't received any feedback >from you in a couple of months. > >The last trace you sent shows nothing unusual, just a bunch of >threads all waiting to write on their respective connections. If the core dump backtrace wasn't helpful, then please tell me so; I thought I had given you sufficient feedback, and that you were working on it. >Have you tried the .21 release yet? No.. I might do that next week. Is there anything else that I can do? I just enabled trace loglevel, perhaps that might give us a clue on what's going on. >-- > -- Howard Chu > CTO, Symas Corp. http://www.symas.com > Director, Highland Sun http://highlandsun.com/hyc/ > Chief Architect, OpenLDAP http://www.openldap.org/project/ -- Helge Milde, 69701808 www.monsternett.no

1 0

(ITS#6247)
by daniel＠pluta.biz 29 Dec '09

29 Dec '09

Hi, I did just some small cleanups, but added no new functionality. I've extracted the minimalistic slapd-api modifications in an extra file called "now_slapd.patch" also contained in the tgz. The code compiles and runs against todays version of "HEAD". The new tgz-file can be downloaded from here: ftp://ftp.openldap.org/incoming/daniel-pluta-now-091229-1.tgz Cheers Daniel

1 0

Re: (ITS#6322) slapd suddenly stops working, and starts using 100% CPU
by hyc＠symas.com 29 Dec '09

29 Dec '09

helge(a)monsternett.no wrote: > Would be nice to get some feedback on this one.. > Were using a core dump sufficient? If so, was the information helpful at all? > If there's anything more I could do, please say so. > Sorry, your issue was closed since we hadn't received any feedback from you in a couple of months. The last trace you sent shows nothing unusual, just a bunch of threads all waiting to write on their respective connections. Have you tried the .21 release yet? -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/

1 0

Re: (ITS#6322) slapd suddenly stops working, and starts using 100% CPU
by helge＠monsternett.no 29 Dec '09

29 Dec '09

Would be nice to get some feedback on this one.. Were using a core dump sufficient? If so, was the information helpful at all? If there's anything more I could do, please say so. -- Helge Milde, 69701808 www.monsternett.no

1 0

Re: (ITS#6194) Patch - Enhancement - provide LDIF support as libldif
by hyc＠symas.com 22 Dec '09

22 Dec '09

rmeggins(a)redhat.com wrote: > The ldif code does not currently have knowledge of any of the higher > level LDAP data types such as LDAPMod or LDAPControl - some of the > proposals would have the ldif API return these objects - Do we really > want to introduce this higher level knowledge into the relatively low > level libldif, along with the dependency on libldap? I think in order > to do this properly, we would have to introduce a new LDAP API e.g. > ldap_mods_from_ldif(const char *ldifstr, LDAPMod***, LDAPControl***) or > something like that - something that would parse the given ldif record > into those higher level data types. Should have mentioned this sooner; this ITS needs to be merged with ITS#4033. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/

1 0

Re: Re : Re : ITS#6434
by masarati＠aero.polimi.it 22 Dec '09

22 Dec '09

KISTER RAPHAEL wrote: > I re-test with servers/slapd/overlays/rwm.c in version 1.125 and it works fine. I don't test with the next version. Fine. The other modifications are not strictly related to your issue, but rather to modifications to olcRwmMap using back-config. p.

1 0

Re : Re : ITS#6434
by kraph＠yahoo.com 22 Dec '09

22 Dec '09

Hello,=0A=0AI re-test with servers/slapd/overlays/rwm.c in version 1.125 an= d it works fine. I don't test with the next=A0version.=0A=0ARaphael KISTER= =0A=0A=0A

1 0

Re: (ITS#6435) Hidden schema elements
by michael＠stroeder.com 22 Dec '09

22 Dec '09

Kurt Zeilenga wrote: > On Dec 21, 2009, at 12:35 PM, Michael Ströder wrote: > >> Kurt(a)OpenLDAP.org wrote: >>> Yes, it has long been our practice not to publish schema elements which = >>> are not yet well standardized. This would include any element which = >>> carries a OpenLDAP.666 OID. >>> >>> The idea being that use of such attributes should be limited to early = >>> adopters and such. >> Sorry, but this practice is inconsequent. > > That's your opinion. My opinion is that we should avoid publishing > ''works in progress'' in production systems. Works in progress, by their > very nature, are subject to change without notice. If you'd really like to avoid issues with attributes being falsely handled by the client because the semantics changed you would especially also have to change the attribute type's NAME. Because that's what regular naive LDAP clients look for. A fully schema-aware client like web2ldap discovers OID changes and handles things correctly. E.g. in web2ldap plugins are preferrably registered by OID (besides some exceptions in case of missing schema descriptions). If the semantics changes and you assign a new OID web2ldap would not naively just treat it the same way. One has to explicitly register a plugin class for the new OID, off course after checking that semantics are the same then. Opposite to that if e.g. the semantics of entryCSN changes and you assign a new OID of a hidden schema description there's no chance to discover that. That's simply bad and therefore I won't follow your arguments at all. BTW: OpenLDAP also prefers NAME handling over OIDs. So with an OpenLDAP deployment changing the OID of a schema descriptions has no effect. But changing the NAME does. That's what I think is inconsequent: The current practice does not avoid issues. It only makes things worse. Ciao, Michael.

1 0

Re: (ITS#6435) Hidden schema elements
by Kurt＠OpenLDAP.org 21 Dec '09

21 Dec '09

On Dec 21, 2009, at 12:35 PM, Michael Str=F6der wrote: > Kurt(a)OpenLDAP.org wrote: >> Yes, it has long been our practice not to publish schema elements = which =3D >> are not yet well standardized. This would include any element which = =3D >> carries a OpenLDAP.666 OID. >>=20 >> The idea being that use of such attributes should be limited to early = =3D >> adopters and such. >=20 > Sorry, but this practice is inconsequent. That's your opinion. My opinion is that we should avoid publishing = ''works in progress'' in production systems. Works in progress, by = their very nature, are subject to change without notice. > The attributes are returned in LDAP > responses Generally because the client asked for more than it should have. = Clients really shouldn't generally be asking for * and/or +. The two exceptions are clients which are servers attempting to = server-to-server replication and clients operated by directory = administrators. Clients which are intended to support various user applications really = should ask only for what they designed to consume. > and therefore a client should be able to look up the attribute type > description in the schema e.g. to determine the syntax. I disagree. Schema may not be made available to the client for any = number of reasons. The X.500 and LDAP specifications allow for this. >=20 > Some of the OIDs are now with .666 for years. Each time this topic is = raised > nothing happens... That's because (in general) the ''works in progress'' have not be = finalized. If something hasn't changed in years, then maybe it's time = to finalize it and, in doing so, assign it final OID. -- Kurt=

1 0

Re: (ITS#6435) Hidden schema elements
by michael＠stroeder.com 21 Dec '09

21 Dec '09

Kurt(a)OpenLDAP.org wrote: > Yes, it has long been our practice not to publish schema elements which = > are not yet well standardized. This would include any element which = > carries a OpenLDAP.666 OID. > > The idea being that use of such attributes should be limited to early = > adopters and such. Sorry, but this practice is inconsequent. The attributes are returned in LDAP responses and therefore a client should be able to look up the attribute type description in the schema e.g. to determine the syntax. Some of the OIDs are now with .666 for years. Each time this topic is raised nothing happens... Ciao, Michael.

1 0

Re: (ITS#6436) olcRwmMap atomic modification
by masarati＠aero.polimi.it 21 Dec '09

21 Dec '09

masarati(a)aero.polimi.it wrote: > olcRwmMap modification via back-config is not atomic; direct insert/delete not > allowed. A patch that implements support for this is proposed. p. The patch makes modification of this attribute consistent (all changes succeeed, or none); if modifications to olcRwmMap are followed by modifications to other attrs within the same write operation, the whole operation is not atomic. This less than ideal behavior is common to many operations allowed by back-config, though. Please test. p.

1 0

(ITS#6436) olcRwmMap atomic modification
by masarati＠aero.polimi.it 21 Dec '09

21 Dec '09

Full_Name: Pierangelo Masarati Version: HEAD/re24 OS: irrelevant URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (131.175.154.146) Submitted by: ando olcRwmMap modification via back-config is not atomic; direct insert/delete not allowed. A patch that implements support for this is proposed. p.

1 0

Re: (ITS#6435) Hidden schema elements
by michael＠stroeder.com 21 Dec '09

21 Dec '09

rein(a)OpenLDAP.org wrote: > Full_Name: Rein Tollevik > Version: CVS HEAD > OS: Irrelevant > URL: > Submission from: (NULL) (2a01:600:0:1:21c:23ff:feab:61cd) > Submitted by: rein > > Some of the operational attributes defined in the slapd source are hidden from > the clients unless slapd is compiled with LDAP_DEVEL enabled. Still, some of > these elements are used in the database (as in the authz* and monitor related > attributes). Yes, see also: http://www.openldap.org/its/index.cgi?findid=5573 http://www.openldap.org/its/index.cgi?findid=5574 http://www.openldap.org/its/index.cgi?findid=5576 IIRC also a bunch of attribute types used in back-config which makes back-config almost unusable with a stock schema aware client. The standard answer by Kurt and others is that as along as an experimental OID with .666 is used a schema description should be hidden. I strongly disagree with that though. > This causes my schema-aware application to complain when it sees > (or worse tries to modify) these attributes. (Sigh!) I also had to add several work-arounds to web2ldap regarding this. > A patch that adds a new optional define that can be used to disable the schema > hiding without enabling LDAP_DEVEL is coming. Looking forward to this being committed. > A better fix would be to not hide those schema elements that is actually > being used.. Amen. Ciao, Michael.

1 0

Re: (ITS#6435) Hidden schema elements
by Kurt＠OpenLDAP.org 21 Dec '09

21 Dec '09

On Dec 21, 2009, at 8:09 AM, rein(a)OpenLDAP.org wrote: > Full_Name: Rein Tollevik > Version: CVS HEAD > OS: Irrelevant > URL:=20 > Submission from: (NULL) (2a01:600:0:1:21c:23ff:feab:61cd) > Submitted by: rein >=20 >=20 > Some of the operational attributes defined in the slapd source are = hidden from > the clients unless slapd is compiled with LDAP_DEVEL enabled. Yes, it has long been our practice not to publish schema elements which = are not yet well standardized. This would include any element which = carries a OpenLDAP.666 OID. The idea being that use of such attributes should be limited to early = adopters and such. > Still, some of > these elements are used in the database (as in the authz* and monitor = related > attributes). This causes my schema-aware application to complain when = it sees > (or worse tries to modify) these attributes. >=20 > A patch that adds a new optional define that can be used to disable = the schema > hiding without enabling LDAP_DEVEL is coming. A better fix would be = to not hide > those schema elements that is actually being used.. It may be that the particular schema elements simply shouldn't be behind = LDAP_DEVEL any longer, possibly because a permanent OID has yet to be = assigned to now production code. >=20 > Rein >=20

1 0

Re: Re : ITS#6434
by masarati＠aero.polimi.it 21 Dec '09

21 Dec '09

KISTER RAPHAEL wrote: > Hello, > > I make a test with servers/slapd/overlays/rwm.c version 1.124. It works fine. Thank you very much for your help. I've re-fixed the problem in a different way; please re-test. Thanks, p.

1 0

Re: (ITS#6435) Hidden schema elements
by masarati＠aero.polimi.it 21 Dec '09

21 Dec '09

rein(a)OpenLDAP.org wrote: > A better fix would be to not hide > those schema elements that is actually being used.. As far as I can tell, OpenLDAP's policy with respect to hiding schema definitions consists in un-hiding only those that are documented in standards track documents, or are for internal use and thus documented internally (e.g. back-config related). So the "right" fix would be to work at formalizing and submitting documents to stardards track. p.

1 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

openldap-bugs