- openldap-bugs - openldap.org

(ITS#7862) slapd crash on modifications after host hard reset (in mdb)
by vadius＠vadius.ru 28 May '14

28 May '14

Full_Name: Vadim Prozorov Version: 2.4.38 OS: RHEL 6.3 URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (91.210.4.1) Hello, I'm using slapd (2.4.38 release) built from sources on RHEL 6.3 64bit. Build params: LD_LIBRARY_PATH="/usr/lib:/usr/local/lib:/opt/local/BerkeleyDB.5.3/lib" LDFLAGS="-L/usr/local/lib -L/opt/local/BerkeleyDB.5.3/lib" CPPFLAGS="-I/usr/local/include -I/opt/local/BerkeleyDB.5.3/include" ./configure --prefix=/opt/local/OpenLDAP --enable-dynacl --enable-ldap enable-overlays --disable-static --enable-dynamic slapd works in master-master sync mode on host with big amount of RAM (whole DB can fit into RAM). One time host was hardly restarted. After restart slapd has started, but crash on any modification (at least, from ext application; may be form replica, haven't check yet). There is always the same crash stack: (gdb) bt #0 mdb_xcursor_init1 (mc=0x7f3e1d6bbd40, node=0x7f3f99f46f82) at ./../../../libraries/liblmdb/mdb.c:6585 #1 0x000000000049f8eb in mdb_cursor_set (mc=0x7f3e1d6bbd40, key=0x7f3e1d6bbef0, data=0x0, op=MDB_SET_RANGE, exactp=<value optimized out>) at ./../../../libraries/liblmdb/mdb.c:5329 #2 0x00000000004a019f in mdb_cursor_get (mc=0x7f3e1d6bbd40, key=0x7f3e1d6bbef0, data=0x0, op=<value optimized out>) at ./../../../libraries/liblmdb/mdb.c:5529 #3 0x000000000049e0ed in mdb_page_alloc (mc=<value optimized out>, num=1, mp=0x7f3e1d6bbf68) at ./../../../libraries/liblmdb/mdb.c:1727 #4 0x000000000049e469 in mdb_page_touch (mc=0x7f3e1d6bc1d0) at ./../../../libraries/liblmdb/mdb.c:1911 #5 0x000000000049eb3a in mdb_page_search (mc=0x7f3e1d6bc1d0, key=0x0, flags=5) at ./../../../libraries/liblmdb/mdb.c:4830 #6 0x00000000004a7404 in mdb_freelist_save (txn=0x7f3e10115130) at ./../../../libraries/liblmdb/mdb.c:2522 #7 mdb_txn_commit (txn=0x7f3e10115130) at ./../../../libraries/liblmdb/mdb.c:2972 #8 0x00000000004a9e4e in mdb_modify (op=0x7f3e100028f0, rs=0x7f3e1d6bd910) at modify.c:664 #9 0x000000000043b4cb in fe_op_modify (op=0x7f3e100028f0, rs=0x7f3e1d6bd910) at modify.c:303 #10 0x000000000043bdf6 in do_modify (op=0x7f3e100028f0, rs=0x7f3e1d6bd910) at modify.c:177 #11 0x0000000000423c99 in connection_operation (ctx=0x7f3e1d6bda70, arg_v=0x7f3e100028f0) at connection.c:1155 #12 0x0000000000424475 in connection_read_thread (ctx=0x7f3e1d6bda70, argv=<value optimized out>) at connection.c:1291 #13 0x00007f3f9b6f04e8 in ldap_int_thread_pool_wrapper (xpool=0x16f36d0) at tpool.c:688 #14 0x0000003b6f2079d1 in start_thread () from /lib64/libpthread.so.0 #15 0x0000003b6eee8b6d in clone () from /lib64/libc.so.6 Crash can be reproduced by using data.mdb file on any host with the same slapd binaries. Also note contextCSN of DB became 'old' - it was actual on May 22 for both instances, when reset occured, but after start context CSN shows May 21 and 20. Also mdb_stat utility crashes on attempt to show FreeDB data: $ mdb_stat -f ./data Freelist Status Tree depth: 3 Branch pages: 7 Leaf pages: 817 Overflow pages: 0 Entries: 16995 Segmentation fault (core dumped)

1 0

(ITS#7861) Wrong entry count with MDB_DUPSORT databases
by h.b.furuseth＠usit.uio.no 27 May '14

27 May '14

Full_Name: Hallvard B Furuseth Version: mdb.master, 4b9aed26a5b1c0027b10d211c6e7270dbb1ade1c OS: Linux x86_64 URL: Submission from: (NULL) (2001:700:100:556::233) Submitted by: hallvard mdb_cursor_put() does not always know if it'll be inserting a new item or not, e.g. when inserting in a sub-database. This was always wrong, but recent commits changed how it was wrong. Fix: Split 'insert' in 'insert_key' + 'insert_data'. Compute insert_data from mx_db.md_entries before and after the insert. In my UiO repo, branch mdb/todo-3: 02dd3865d5 [Rewritten] ITS#xxxx Fix MDB_db.md_entries. c6c78f71d3 [New] Tweak varname 'insert'. (Cosmetic patch which does not apply cleanly, I guess I should make an mdb/todo-4 branch where it is applied earlier.) Affects ITS#7834. mdb_cursor_del(, MDB_NODUPDATA) only subtracts 1 from md_entries when deleting a sub-page, instead of no of DUP entries. Fix: fa8e827026 [New,Squash] Fix md_entries in mdb_cursor_del:MDB_NODUPDATA.

1 0

(ITS#7860) [PATCH] correction to ldap_ava in ldap_get_dn.3
by ryan＠nardis.ca 27 May '14

27 May '14

Full_Name: Ryan Tandy Version: master/b22a614, RE24/d961650 OS: Debian unstable URL: https://bugs.debian.org/cgi-bin/bugreport.cgi?msg=38;bug=465024 Submission from: (NULL) (24.68.121.206) Hi, In Debian bug #465024, Philipp Hahn has provided the following patch, supplementing the change from ITS#5366: Index: openldap-2.4.39/doc/man/man3/ldap_get_dn.3 =================================================================== --- openldap-2.4.39.orig/doc/man/man3/ldap_get_dn.3 2014-02-21 00:12:24.000000000 +0100 +++ openldap-2.4.39/doc/man/man3/ldap_get_dn.3 2014-04-17 11:28:10.468925942 +0200 @@ -82,8 +82,8 @@ .ft B typedef struct ldap_ava { - char *la_attr; - struct berval *la_value; + struct berval la_attr; + struct berval la_value; unsigned la_flags; } LDAPAVA; It looks correct to me. (I assume the omission of la_private is intentional.) thanks, Ryan

1 0

Re: (ITS#7859) ldif processing issue with long lines
by hyc＠symas.com 27 May '14

27 May '14

mwarren(a)symas.com wrote: > Also, another issue has been reported where the length of LDIF lines is greater > by two than a specified 'ldif-wrap' definition when using slapcat. This appears to be the result of an intentional backward-compatiblility kludge. While the spec may have defaulted to 76 chars, the UMich code has always output 78 columns by default, so there's an offset built in to the LDIF output functions. This offset wasn't taken into account when the variable output width support was written. Probably we should just remove the kludge and define our default output width to 78 instead of pretending it's still 76. -- -- 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#7859) ldif processing issue with long lines
by mwarren＠symas.com 26 May '14

26 May '14

Full_Name: Mark Warren Version: 2.4.35 OS: Linux URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (72.35.133.119) A customer of ours has identified an issue with slapadd and lines right at the maximum length(4096 characters). Howard has taken a look and reports the following: I took a quick look at the slapadd issue; libldap/ldif.c is using an input buffer of size [LDIF_MAXLINE] which is 4096 characters. It uses fgets() to read the input. The problem is that fgets() always stores a terminating NUL character, which means the buffer can only accept an input of up to 4095 characters. The input line in question is "only" 4095 characters long, but in reality it also has a terminating newline, making it 4096. The end result, fgets() reads the data but leaves the newline in the input buffer since there's no room for it in the user buffer. Then the next call to fgets() returns just this newline all by itself, which signals to ldif_read_record that it's come to the end of the entry, and so it treats the next input line as a new entry even though it's supposed to be a continuation of the previous line. The simple fix here is to bump up the input buffer size by 1, to accommodate the terminating newline. Actually the buffer needs to be upped by 2, to allow for CR/LF line terminators. -- End Howard Quote -- Also, another issue has been reported where the length of LDIF lines is greater by two than a specified 'ldif-wrap' definition when using slapcat. Regards, Mark Warren

1 0

Re: (ITS#7858) [PATCH] nssov/Makefile wrong rpath
by hyc＠symas.com 25 May '14

25 May '14

ryan(a)nardis.ca wrote: > Full_Name: Ryan Tandy > Version: master/932a11a, RE24/d961650 > OS: Debian unstable > URL: > Submission from: (NULL) (24.68.121.206) > > > Hi, > > nssov can't be installed: Thanks, fixed now in master > > $ cd contrib/slapd-modules/nssov > $ make > [...] > $ sudo make install > mkdir -p /usr/local/libexec/openldap > ../../../libtool --mode=install cp nssov.la /usr/local/libexec/openldap > libtool: install: error: cannot install `nssov.la' to a directory not ending in > /usr/local/lib > Makefile:67: recipe for target 'install' failed > make: *** [install] Error 1 > > The Makefile says "-rpath $(libdir)" where every other contrib module uses > "-rpath $(moduledir)". Changing it to that lets it build, and the resulting > module seems to work. > > diff --git a/contrib/slapd-modules/nssov/Makefile > b/contrib/slapd-modules/nssov/Makefile > index 8527ee3..28c8003 100644 > --- a/contrib/slapd-modules/nssov/Makefile > +++ b/contrib/slapd-modules/nssov/Makefile > @@ -61,7 +61,7 @@ $(OBJS): nssov.h > > nssov.la: $(OBJS) $(XOBJS) > $(LIBTOOL) --mode=link $(CC) $(OPT) -version-info 0:0:0 \ > - -rpath $(libdir) -module -o $@ $(OBJS) $(XOBJS) $(LIBS) > + -rpath $(moduledir) -module -o $@ $(OBJS) $(XOBJS) $(LIBS) > > install: nssov.la > mkdir -p $(DESTDIR)$(moduledir) > > thanks, > Ryan > > -- -- 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#7858) [PATCH] nssov/Makefile wrong rpath
by ryan＠nardis.ca 25 May '14

25 May '14

Full_Name: Ryan Tandy Version: master/932a11a, RE24/d961650 OS: Debian unstable URL: Submission from: (NULL) (24.68.121.206) Hi, nssov can't be installed: $ cd contrib/slapd-modules/nssov $ make [...] $ sudo make install mkdir -p /usr/local/libexec/openldap ../../../libtool --mode=install cp nssov.la /usr/local/libexec/openldap libtool: install: error: cannot install `nssov.la' to a directory not ending in /usr/local/lib Makefile:67: recipe for target 'install' failed make: *** [install] Error 1 The Makefile says "-rpath $(libdir)" where every other contrib module uses "-rpath $(moduledir)". Changing it to that lets it build, and the resulting module seems to work. diff --git a/contrib/slapd-modules/nssov/Makefile b/contrib/slapd-modules/nssov/Makefile index 8527ee3..28c8003 100644 --- a/contrib/slapd-modules/nssov/Makefile +++ b/contrib/slapd-modules/nssov/Makefile @@ -61,7 +61,7 @@ $(OBJS): nssov.h nssov.la: $(OBJS) $(XOBJS) $(LIBTOOL) --mode=link $(CC) $(OPT) -version-info 0:0:0 \ - -rpath $(libdir) -module -o $@ $(OBJS) $(XOBJS) $(LIBS) + -rpath $(moduledir) -module -o $@ $(OBJS) $(XOBJS) $(LIBS) install: nssov.la mkdir -p $(DESTDIR)$(moduledir) thanks, Ryan

1 0

Re: (ITS#7765) Crash when adding syncrepl configuration in cn=config without search base
by hyc＠symas.com 23 May '14

23 May '14

coudot(a)linagora.com wrote: > Full_Name: Clément OUDOT > Version: 2.4.38 > OS: GNU/Linux > URL: ftp://ftp.openldap.org/incoming/ > Submission from: (NULL) (88.173.78.196) > > > When adding an olcSyncrepl attribute in my cn=config, OpenLDAP crashed. > > The syslog output is: > > Dec 16 11:56:42 ader slapd[8707]: conn=1096 op=0 BIND dn="cn=config" method=128 > Dec 16 11:56:42 ader slapd[8707]: conn=1096 op=0 BIND dn="cn=config" mech=SIMPLE > ssf=0 > Dec 16 11:56:42 ader slapd[8707]: conn=1096 op=0 RESULT tag=97 err=0 text= > Dec 16 11:56:42 ader slapd[8707]: conn=1096 op=1 MOD > dn="olcDatabase={2}mdb,cn=config" > Dec 16 11:56:42 ader slapd[8707]: conn=1096 op=1 MOD attr=olcaccess olcdbindex > olcsyncrepl > Dec 16 11:56:42 ader slapd[8707]: olcSyncrepl: value #0: Error: Malformed > "syncrepl" line in slapd config file, missing searchbase. > Dec 16 11:56:42 ader slapd[8707]: failed to add syncinfo > Dec 16 11:56:42 ader slapd[8707]: ch_calloc of 1 elems of 0 bytes failed > > > I have no gdb trace to give for now, if you think it is required, I will try to > find time to do it. But it should be easy to reproduce the issue on your side. Please provide your exact ldapmodify request and the database config before the modify attempt, thanks. -- -- 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#7857) let openldap build against the bionic libc
by hyc＠symas.com 19 May '14

19 May '14

tux_mind(a)monumentmail.com wrote: > Full_Name: tux_mind > Version: lastest from git > OS: gentoo GNU/Linux > URL: https://github.com/tux-mind/platform_external_openldap > Submission from: (NULL) (188.153.8.167) > > > hello guys. > i've modified a little your project to let it build against the bionic libc ( > the one used in android ). > > please let me known if you can merge this into the upstream project. > > modified sources can be found here: > https://github.com/tux-mind/platform_external_openldap > > my changes are the last 3 of these commits: > https://github.com/tux-mind/platform_external_openldap/commits/master "added generated files" - No. Don't be silly. The last time I compiled OpenLDAP for Android, I don't recall needing any #include patches. (That was for Froyo, a few years ago now.) I don't have a current dev system to check these changes. The reason those generated files are generated is because those platform-specific features are known to change. E.g., 64 bit Android devices are coming out soon, and hardcoding things for 32 bit would be a mistake. Including an Android.mk file in our distribution would require us to maintain it in parallel with our regular Makefiles, forever after. We will not do redundant support work, nor is it necessary. The standard OpenLDAP Makefiles work fine to create Android binaries. Rejecting these patches. > > thanks for your attention > > -- -- 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#7857) let openldap build against the bionic libc
by tux_mind＠monumentmail.com 19 May '14

19 May '14

Full_Name: tux_mind Version: lastest from git OS: gentoo GNU/Linux URL: https://github.com/tux-mind/platform_external_openldap Submission from: (NULL) (188.153.8.167) hello guys. i've modified a little your project to let it build against the bionic libc ( the one used in android ). please let me known if you can merge this into the upstream project. modified sources can be found here: https://github.com/tux-mind/platform_external_openldap my changes are the last 3 of these commits: https://github.com/tux-mind/platform_external_openldap/commits/master thanks for your attention

1 0

Re: (ITS#7856) TLS_REQCERT try is same as TLS_REQCERT hard?
by jsynacek＠redhat.com 19 May '14

19 May '14

On 05/16/2014 09:11 AM, pguenther(a)proofpoint.com wrote: > Full_Name: Philip Guenther > Version: 2.4.39 > OS: OpenBSD > URL: ftp://ftp.openldap.org/incoming/ > Submission from: (NULL) (76.253.0.176) > > > The ldap.conf(5) manpage says this about TLS_REQCERT > TLS_REQCERT <level> > Specifies what checks to perform on server certificates in a TLS > session, if any. The <level> can be specified as one of the > following keywords: > ... > > try The server certificate is requested. If no certificate is > provided, the session proceeds normally. If a bad > certificate is provided, the session is immediately > terminated. > > demand | hard > These keywords are equivalent. The server certificate is > requested. If no certificate is provided, or a bad > certificate is provided, the session is immediately > terminated. This is the default setting. > > > In testing, I can find no difference in behavior between the 'try' and 'hard' > keywords. For the ldap* tools, both 'try' and 'hard' seem to place the same > requirements on the server. What does "if no certificate is provided" *mean* in > terms of server and/or client configuration? > See ITS#7744. -- Jan Synacek Software Engineer, Red Hat

1 0

Re: (ITS#7842) mdb readers/writer exclusion protocol, as implemented, is racy.
by rsbx＠acm.org 17 May '14

17 May '14

On 05/16/2014 08:09 PM, Howard Chu wrote: > rsbx(a)acm.org wrote: >> Full_Name: Raymond S Brand >> Version: mdb.master >> OS: linux >> URL: ftp://ftp.openldap.org/incoming/ >> Submission from: (NULL) (107.145.137.13) >> >> >> [I would have uploaded this to the ftp.openldap.org but there was no >> space.] >> >> As implemented, the mdb readers/writer exclusion protocol has a race >> condition >> that could result in a writer reclaiming and over-writing pages still >> in use >> by a reader. > > Yes, you're basically correct. But this is unlikely in practice because > the reader thread has no blocking calls in its codepath, while the > writer must acquire locks etc. Generally the only way for two write txns > to complete while a reader thread is stalled is if you explicitly send a > SIGSTOP to the reader thread while it's in its critical section. Or just unlucky process scheduling. > >> >> Pseudo code of the snapshot locking protocols of reader and writer >> transactions. The labeled sections, for the purposes of this analysis, >> can be >> assumed to execute atomically. >> >> READER >> ====== >> >> *R1* t0 = meta[0].txid; t1 = meta[1].txid >> if (t1 > t0) >> t0 = t1 >> >> *R2* readers[me].txid = t0 >> >> *R3* snapshot_root = meta[t0&1].snapshot >> >> *R4* /* lookup data */ >> >> *R5* readers[me].txid = -1 // Release snapshot >> >> WRITER >> ====== >> >> *W1* lock(writer, EXCLUSIVE) >> curr_txid = meta[0].txid >> if (meta[1].txid > curr_txid) >> curr_txid = meta[1].txid >> >> *W2* oldest = curr_txid >> for (i=0; i<reader_slots; i++) >> t = readers[i].txid >> if (t != -1 && t < oldest) >> oldest = t >> >> *W3* reclaim_old_pages(oldest) >> >> /* >> ** Commit new transaction >> */ >> >> *W4* unlock(writer) >> >> --------------------------------------------------------------------------- >> >> >> Adversarial scheduling analysis: >> >> The following timeline demonstrates that a writer can reclaim pages in >> use by a >> reader. >> >> T0 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 >> R1 R2 R3 R4 >> W1 W2 W3 W4 * W1 W2 W3 >> >> T0 Reader has found the latest txid, Xn. >> >> T1->T4 Reader is not scheduled to run and a write transaction >> commits, Xn+1. >> >> T5 Reader is still not scheduled and another write transaction >> starts. >> >> T6 Writer finds that the oldest referenced transaction is the last >> committed transaction, Xn+1. >> >> T7 Reader records Xn in the reader table. >> >> T8 Reader gets the snapshot root page for transaction Xn. >> >> T9 Writer, believing that Xn+1 is the oldest reference transaction, >> reclaims the pages of transaction Xn. >> >> >> T10 Reader attempts to navigate pages of a transaction, Xn, that >> has been >> reclaimed and "Bad Things Happen". >> >> In fact, bad things will happen if an even number of write >> transactions commit >> between W4 and W1, represented by the '*', in the timeline above. >> >> The fix is to search for the highest txid in R1 and between R2 and R3. >> Optionally, followed by recording the txid, of the snapshot root found >> in R3, >> in the reader's reader table slot to, possibly, increase the number of >> reclaimable transactions. >> >> The lack of compiler and memory barriers in the implementation of the >> locking >> protocol is also of concern. >> >> Beyond the above, the code in mdb_txn_renew0() after the >> "/* Copy the DB info and flags */" >> comment appears to have a number of data races. >> >> --- >> libraries/liblmdb/mdb.c | 24 +++++++++++++++--------- >> 1 file changed, 15 insertions(+), 9 deletions(-) >> >> diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c >> index e0f551e..908417c 100644 >> --- a/libraries/liblmdb/mdb.c >> +++ b/libraries/liblmdb/mdb.c >> @@ -533,11 +533,11 @@ typedef struct MDB_rxbody { >> * started from so we can avoid overwriting any data >> used in that >> * particular version. >> */ >> - txnid_t mrb_txnid; >> + volatile txnid_t mrb_txnid; >> /** The process ID of the process owning this reader txn. */ >> - MDB_PID_T mrb_pid; >> + volatile MDB_PID_T mrb_pid; >> /** The thread ID of the thread owning this txn. */ >> - pthread_t mrb_tid; >> + volatile pthread_t mrb_tid; >> } MDB_rxbody; >> >> /** The actual reader record, with cacheline padding. */ >> @@ -585,12 +585,12 @@ typedef struct MDB_txbody { >> * This is recorded here only for convenience; >> the value >> can always >> * be determined by reading the main database >> meta pages. >> */ >> - txnid_t mtb_txnid; >> + volatile txnid_t mtb_txnid; >> /** The number of slots that have been used in the >> reader >> table. >> * This always records the maximum count, it is not >> decremented >> * when readers release their slots. >> */ >> - unsigned mtb_numreaders; >> + volatile unsigned mtb_numreaders; >> } MDB_txbody; >> >> /** The actual reader table definition. */ >> @@ -854,7 +854,7 @@ typedef struct MDB_meta { >> /** Any persistent environment flags. @ref mdb_env */ >> #define mm_flags mm_dbs[0].md_flags >> pgno_t mm_last_pg; /**< last >> used page in >> file */ >> - txnid_t mm_txnid; /**< txnid that >> committed this page */ >> + volatile txnid_t mm_txnid; /**< txnid that >> committed this >> page */ >> } MDB_meta; >> >> /** Buffer for a stack-allocated meta page. >> @@ -2303,10 +2303,12 @@ mdb_txn_renew0(MDB_txn *txn) >> >> if (txn->mt_flags & MDB_TXN_RDONLY) { >> if (!ti) { >> + /* No readers table; app responsible for >> locking */ >> meta = env->me_metas[ mdb_env_pick_meta(env) ]; >> txn->mt_txnid = meta->mm_txnid; >> txn->mt_u.reader = NULL; >> } else { >> + /* Has readers table */ >> MDB_reader *r = (env->me_flags & MDB_NOTLS) ? >> txn->mt_u.reader : >> pthread_getspecific(env->me_txkey); >> if (r) { >> @@ -2347,8 +2349,12 @@ mdb_txn_renew0(MDB_txn *txn) >> return rc; >> } >> } >> - txn->mt_txnid = r->mr_txnid = ti->mti_txnid; >> txn->mt_u.reader = r; >> + r->mr_txnid = ti->mti_txnid; >> + >> + /* Really need a memory barrier here */ >> + >> + txn->mt_txnid = r->mr_txnid = ti->mti_txnid; >> meta = env->me_metas[txn->mt_txnid & 1]; >> } >> } else { >> @@ -2376,10 +2382,10 @@ mdb_txn_renew0(MDB_txn *txn) >> } >> >> /* Copy the DB info and flags */ >> - memcpy(txn->mt_dbs, meta->mm_dbs, 2 * sizeof(MDB_db)); >> + memcpy(txn->mt_dbs, meta->mm_dbs, 2 * sizeof(MDB_db)); /* >> Racy */ >> >> /* Moved to here to avoid a data race in read TXNs */ >> - txn->mt_next_pgno = meta->mm_last_pg+1; >> + txn->mt_next_pgno = meta->mm_last_pg+1; /* >> Racy */ >> >> for (i=2; i<txn->mt_numdbs; i++) { >> x = env->me_dbflags[i]; >> > >

1 0

Re: (ITS#7842) mdb readers/writer exclusion protocol, as implemented, is racy.
by hyc＠symas.com 17 May '14

17 May '14

rsbx(a)acm.org wrote: > Full_Name: Raymond S Brand > Version: mdb.master > OS: linux > URL: ftp://ftp.openldap.org/incoming/ > Submission from: (NULL) (107.145.137.13) > > > [I would have uploaded this to the ftp.openldap.org but there was no space.] > > As implemented, the mdb readers/writer exclusion protocol has a race condition > that could result in a writer reclaiming and over-writing pages still in use > by a reader. Yes, you're basically correct. But this is unlikely in practice because the reader thread has no blocking calls in its codepath, while the writer must acquire locks etc. Generally the only way for two write txns to complete while a reader thread is stalled is if you explicitly send a SIGSTOP to the reader thread while it's in its critical section. > > Pseudo code of the snapshot locking protocols of reader and writer > transactions. The labeled sections, for the purposes of this analysis, can be > assumed to execute atomically. > > READER > ====== > > *R1* t0 = meta[0].txid; t1 = meta[1].txid > if (t1 > t0) > t0 = t1 > > *R2* readers[me].txid = t0 > > *R3* snapshot_root = meta[t0&1].snapshot > > *R4* /* lookup data */ > > *R5* readers[me].txid = -1 // Release snapshot > > WRITER > ====== > > *W1* lock(writer, EXCLUSIVE) > curr_txid = meta[0].txid > if (meta[1].txid > curr_txid) > curr_txid = meta[1].txid > > *W2* oldest = curr_txid > for (i=0; i<reader_slots; i++) > t = readers[i].txid > if (t != -1 && t < oldest) > oldest = t > > *W3* reclaim_old_pages(oldest) > > /* > ** Commit new transaction > */ > > *W4* unlock(writer) > > --------------------------------------------------------------------------- > > Adversarial scheduling analysis: > > The following timeline demonstrates that a writer can reclaim pages in use by a > reader. > > T0 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 > R1 R2 R3 R4 > W1 W2 W3 W4 * W1 W2 W3 > > T0 Reader has found the latest txid, Xn. > > T1->T4 Reader is not scheduled to run and a write transaction commits, Xn+1. > > T5 Reader is still not scheduled and another write transaction starts. > > T6 Writer finds that the oldest referenced transaction is the last > committed transaction, Xn+1. > > T7 Reader records Xn in the reader table. > > T8 Reader gets the snapshot root page for transaction Xn. > > T9 Writer, believing that Xn+1 is the oldest reference transaction, > reclaims the pages of transaction Xn. > > > T10 Reader attempts to navigate pages of a transaction, Xn, that has been > reclaimed and "Bad Things Happen". > > In fact, bad things will happen if an even number of write transactions commit > between W4 and W1, represented by the '*', in the timeline above. > > The fix is to search for the highest txid in R1 and between R2 and R3. > Optionally, followed by recording the txid, of the snapshot root found in R3, > in the reader's reader table slot to, possibly, increase the number of > reclaimable transactions. > > The lack of compiler and memory barriers in the implementation of the locking > protocol is also of concern. > > Beyond the above, the code in mdb_txn_renew0() after the > "/* Copy the DB info and flags */" > comment appears to have a number of data races. > > --- > libraries/liblmdb/mdb.c | 24 +++++++++++++++--------- > 1 file changed, 15 insertions(+), 9 deletions(-) > > diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c > index e0f551e..908417c 100644 > --- a/libraries/liblmdb/mdb.c > +++ b/libraries/liblmdb/mdb.c > @@ -533,11 +533,11 @@ typedef struct MDB_rxbody { > * started from so we can avoid overwriting any data used in that > * particular version. > */ > - txnid_t mrb_txnid; > + volatile txnid_t mrb_txnid; > /** The process ID of the process owning this reader txn. */ > - MDB_PID_T mrb_pid; > + volatile MDB_PID_T mrb_pid; > /** The thread ID of the thread owning this txn. */ > - pthread_t mrb_tid; > + volatile pthread_t mrb_tid; > } MDB_rxbody; > > /** The actual reader record, with cacheline padding. */ > @@ -585,12 +585,12 @@ typedef struct MDB_txbody { > * This is recorded here only for convenience; the value > can always > * be determined by reading the main database meta pages. > */ > - txnid_t mtb_txnid; > + volatile txnid_t mtb_txnid; > /** The number of slots that have been used in the reader > table. > * This always records the maximum count, it is not > decremented > * when readers release their slots. > */ > - unsigned mtb_numreaders; > + volatile unsigned mtb_numreaders; > } MDB_txbody; > > /** The actual reader table definition. */ > @@ -854,7 +854,7 @@ typedef struct MDB_meta { > /** Any persistent environment flags. @ref mdb_env */ > #define mm_flags mm_dbs[0].md_flags > pgno_t mm_last_pg; /**< last used page in > file */ > - txnid_t mm_txnid; /**< txnid that > committed this page */ > + volatile txnid_t mm_txnid; /**< txnid that committed this > page */ > } MDB_meta; > > /** Buffer for a stack-allocated meta page. > @@ -2303,10 +2303,12 @@ mdb_txn_renew0(MDB_txn *txn) > > if (txn->mt_flags & MDB_TXN_RDONLY) { > if (!ti) { > + /* No readers table; app responsible for locking */ > meta = env->me_metas[ mdb_env_pick_meta(env) ]; > txn->mt_txnid = meta->mm_txnid; > txn->mt_u.reader = NULL; > } else { > + /* Has readers table */ > MDB_reader *r = (env->me_flags & MDB_NOTLS) ? > txn->mt_u.reader : > pthread_getspecific(env->me_txkey); > if (r) { > @@ -2347,8 +2349,12 @@ mdb_txn_renew0(MDB_txn *txn) > return rc; > } > } > - txn->mt_txnid = r->mr_txnid = ti->mti_txnid; > txn->mt_u.reader = r; > + r->mr_txnid = ti->mti_txnid; > + > + /* Really need a memory barrier here */ > + > + txn->mt_txnid = r->mr_txnid = ti->mti_txnid; > meta = env->me_metas[txn->mt_txnid & 1]; > } > } else { > @@ -2376,10 +2382,10 @@ mdb_txn_renew0(MDB_txn *txn) > } > > /* Copy the DB info and flags */ > - memcpy(txn->mt_dbs, meta->mm_dbs, 2 * sizeof(MDB_db)); > + memcpy(txn->mt_dbs, meta->mm_dbs, 2 * sizeof(MDB_db)); /* Racy */ > > /* Moved to here to avoid a data race in read TXNs */ > - txn->mt_next_pgno = meta->mm_last_pg+1; > + txn->mt_next_pgno = meta->mm_last_pg+1; /* Racy */ > > for (i=2; i<txn->mt_numdbs; i++) { > x = env->me_dbflags[i]; > -- -- 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#7844) LMDB Delete Cursor inconsistencies
by armon.dadgar＠gmail.com 16 May '14

16 May '14

--53767cd5_238e1f29_1704c Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline I checked on gitorious only hours previously and fca18d2 was the latest v= ersion. However, I now see=C2=A02764360. I=E2=80=99ve updated to the latest versi= on and can no longer reproduce the issue. Thanks=21 Best Regards, Armon Dadgar =46rom:=C2=A0Howard Chu hyc=40symas.com Reply:=C2=A0Howard Chu hyc=40symas.com Date:=C2=A0May 16, 2014 at 1:54:18 PM To:=C2=A0Armon Dadgar armon.dadgar=40gmail.com, openldap-its=40openldap.o= rg openldap-its=40openldap.org Subject:=C2=A0 Re: (ITS=237844) LMDB Delete Cursor inconsistencies =20 Armon Dadgar wrote: =20 > When I run the attached test case on my machine, I=E2=80=99m hitting th= e failing case. =20 > =20 > Here is the test output: https://gist.github.com/armon/e529d7909fe30112= 6fc6 =20 You seem to be running obsolete code. I checked your gomdb github repo, y= ou're =20 using mdb.c at rev fca18d2. Current mdb.master is 4844a72. =20 > My steps: =20 > =24 clang 7844.c mdb.c midl.c =20 > =24 mkdir testdb =20 > =24 ./a.out =20 > =20 > =24 clang -v =20 > Apple LLVM version 5.1 (clang-503.0.40) (based on LLVM 3.4svn) =20 > Target: x86=5F64-apple-darwin13.1.0 =20 > Thread model: posix =20 > =20 > =24 uname -a =20 > Darwin Armons-MacBook-Air.local 13.1.0 Darwin Kernel Version 13.1.0: Th= u Jan =20 > 16 19:40:37 PST 2014; root:xnu-2422.90.20=7E2/RELEASE=5FX86=5F64 x86=5F= 64 =20 > =20 > Best Regards, =20 > Armon Dadgar =20 > =20 > =46rom: Howard Chu hyc=40symas.com <mailto:hyc=40symas.com> =20 > Reply: Howard Chu hyc=40symas.com <mailto:hyc=40symas.com> =20 > Date: May 16, 2014 at 11:06:49 AM =20 > To: armon.dadgar=40gmail.com armon.dadgar=40gmail.com =20 > <mailto:armon.dadgar=40gmail.com>, openldap-its=40openldap.org =20 > openldap-its=40openldap.org <mailto:openldap-its=40openldap.org> =20 > Subject: Re: (ITS=237844) LMDB Delete Cursor inconsistencies =20 > =20 >> armon.dadgar=40gmail.com wrote: =20 >> > --5372ac85=5F8edbdab=5F1271 =20 >> > Content-Type: text/plain; charset=3D=22utf-8=22 =20 >> > Content-Transfer-Encoding: quoted-printable =20 >> > Content-Disposition: inline =20 >> > =20 >> > =3D46or now, we have application code to retry the delete until no f= urther =3D =20 >> > rows are removed. =20 >> > Still, it would be nice to have this resolved (and tested) in master= =3D21 =20 >> =20 >> Unable to reproduce the issue. I've attached my test program based on = your =20 >> description. =20 >> =20 >> -- =20 >> -- Howard Chu =20 >> CTO, Symas Corp. http://www.symas.com =20 >> Director, Highland Sun http://highlandsun.com/hyc/ =20 >> Chief Architect, OpenLDAP http://www.openldap.org/project/ =20 >> ----------------------------------------------------------------------= -------- =20 -- =20 -- Howard Chu =20 CTO, Symas Corp. http://www.symas.com =20 Director, Highland Sun http://highlandsun.com/hyc/ =20 Chief Architect, OpenLDAP http://www.openldap.org/project/ =20 --53767cd5_238e1f29_1704c Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline <html><head><style>body=7Bfont-family:Helvetica,Arial;font-size:13px=7D</= style></head><body style=3D=22word-wrap: break-word; -webkit-nbsp-mode: s= pace; -webkit-line-break: after-white-space;=22><div id=3D=22bloop=5Fcust= omfont=22 style=3D=22font-family:Helvetica,Arial;font-size:13px; color: r= gba(0,0,0,1.0); margin: 0px; line-height: auto;=22>I checked on gitorious= only hours previously and fca18d2 was the latest version.</div><div id=3D= =22bloop=5Fcustomfont=22 style=3D=22font-family:Helvetica,Arial;font-size= :13px; color: rgba(0,0,0,1.0); margin: 0px; line-height: auto;=22>However= , I now see 2764360. I=E2=80=99ve updated to the latest version and = can no longer</div><div id=3D=22bloop=5Fcustomfont=22 style=3D=22font-fam= ily:Helvetica,Arial;font-size:13px; color: rgba(0,0,0,1.0); margin: 0px; = line-height: auto;=22>reproduce the issue.</div><div id=3D=22bloop=5Fcust= omfont=22 style=3D=22font-family:Helvetica,Arial;font-size:13px; color: r= gba(0,0,0,1.0); margin: 0px; line-height: auto;=22><br></div><div id=3D=22= bloop=5Fcustomfont=22 style=3D=22font-family:Helvetica,Arial;font-size:13= px; color: rgba(0,0,0,1.0); margin: 0px; line-height: auto;=22>Thanks=21<= /div><div id=3D=22bloop=5Fcustomfont=22 style=3D=22font-family:Helvetica,= Arial;font-size:13px; color: rgba(0,0,0,1.0); margin: 0px; line-height: a= uto;=22><br></div> <div id=3D=22bloop=5Fsign=5F1400274014470972928=22 cla= ss=3D=22bloop=5Fsign=22><div style=3D=22font-family:helvetica,arial;font-= size:13px=22>Best Regards,</div><div style=3D=22font-family:helvetica,ari= al;font-size:13px=22>Armon Dadgar<br></div></div> <div style=3D=22color:b= lack=22><br>=46rom: <span style=3D=22color:black=22>Howard Chu</span= > <a href=3D=22mailto:hyc=40symas.com=22>hyc=40symas.com</a><br>Reply:&nb= sp;<span style=3D=22color:black=22>Howard Chu</span> <a href=3D=22mailto:= hyc=40symas.com=22>hyc=40symas.com</a><br>Date: <span style=3D=22col= or:black=22>May 16, 2014 at 1:54:18 PM</span><br>To: <span style=3D=22= color:black=22>Armon Dadgar</span> <a href=3D=22mailto:armon.dadgar=40gma= il.com=22>armon.dadgar=40gmail.com</a>, <span style=3D=22color:black=22>o= penldap-its=40openldap.org</span> <a href=3D=22mailto:openldap-its=40open= ldap.org=22>openldap-its=40openldap.org</a><br>Subject: <span style=3D= =22color:black=22> Re: (ITS=237844) LMDB Delete Cursor inconsistencies <b= r></span></div><br> <blockquote type=3D=22cite=22 class=3D=22clean=5Fbq=22= ><span><div><div></div><div>Armon Dadgar wrote: <br>> When I run the attached test case on my machine, I=E2=80=99m hit= ting the failing case. <br>> <br>> Here is the test output: https://gist.github.com/armon/e529d7909= fe301126fc6 <br> <br>You seem to be running obsolete code. I checked your gomdb github rep= o, you're =20 <br>using mdb.c at rev fca18d2. Current mdb.master is 4844a72. <br> <br>> My steps: <br>> =24 clang 7844.c mdb.c midl.c <br>> =24 mkdir testdb <br>> =24 ./a.out <br>> <br>> =24 clang -v <br>> Apple LLVM version 5.1 (clang-503.0.40) (based on LLVM 3.4svn) <br>> Target: x86=5F64-apple-darwin13.1.0 <br>> Thread model: posix <br>> <br>> =24 uname -a <br>> Darwin Armons-MacBook-Air.local 13.1.0 Darwin Kernel Version 13.= 1.0: Thu Jan <br>> 16 19:40:37 PST 2014; root:xnu-2422.90.20=7E2/RELEASE=5FX86=5F64= x86=5F64 <br>> <br>> Best Regards, <br>> Armon Dadgar <br>> <br>> =46rom: Howard Chu hyc=40symas.com <mailto:hyc=40symas.com&gt= ; <br>> Reply: Howard Chu hyc=40symas.com <mailto:hyc=40symas.com>= <br>> Date: May 16, 2014 at 11:06:49 AM <br>> To: armon.dadgar=40gmail.com armon.dadgar=40gmail.com <br>> <mailto:armon.dadgar=40gmail.com>, openldap-its=40openldap= .org <br>> openldap-its=40openldap.org <mailto:openldap-its=40openldap.o= rg> <br>> Subject: Re: (ITS=237844) LMDB Delete Cursor inconsistencies <br>> <br>>> armon.dadgar=40gmail.com wrote: <br>>> > --5372ac85=5F8edbdab=5F1271 <br>>> > Content-Type: text/plain; charset=3D=22utf-8=22 <br>>> > Content-Transfer-Encoding: quoted-printable <br>>> > Content-Disposition: inline <br>>> > <br>>> > =3D46or now, we have application code to retry the dele= te until no further =3D <br>>> > rows are removed. <br>>> > Still, it would be nice to have this resolved (and test= ed) in master=3D21 <br>>> <br>>> Unable to reproduce the issue. I've attached my test program= based on your <br>>> description. <br>>> <br>>> -- <br>>> -- Howard Chu <br>>> CTO, Symas Corp. http://www.symas.com <br>>> Director, Highland Sun http://highlandsun.com/hyc/ <br>>> Chief Architect, OpenLDAP http://www.openldap.org/project/ <br>>> ------------------------------------------------------------= ------------------ <br> <br> <br>-- =20 <br> -- Howard Chu <br> CTO, Symas Corp. http://www.symas.com <br> Director, Highland Sun http://highlandsun.com/hyc/ <br> Chief Architect, OpenLDAP http://www.openldap.org/project/ <br></div></div></span></blockquote></body></html> --53767cd5_238e1f29_1704c--

1 0

Re: (ITS#7844) LMDB Delete Cursor inconsistencies
by hyc＠symas.com 16 May '14

16 May '14

Armon Dadgar wrote: > When I run the attached test case on my machine, Im hitting the failing case. > > Here is the test output: https://gist.github.com/armon/e529d7909fe301126fc6 You seem to be running obsolete code. I checked your gomdb github repo, you're using mdb.c at rev fca18d2. Current mdb.master is 4844a72. > My steps: > $ clang 7844.c mdb.c midl.c > $ mkdir testdb > $ ./a.out > > $ clang -v > Apple LLVM version 5.1 (clang-503.0.40) (based on LLVM 3.4svn) > Target: x86_64-apple-darwin13.1.0 > Thread model: posix > > $ uname -a > Darwin Armons-MacBook-Air.local 13.1.0 Darwin Kernel Version 13.1.0: Thu Jan > 16 19:40:37 PST 2014; root:xnu-2422.90.20~2/RELEASE_X86_64 x86_64 > > Best Regards, > Armon Dadgar > > From: Howard Chu hyc(a)symas.com <mailto:hyc@symas.com> > Reply: Howard Chu hyc(a)symas.com <mailto:hyc@symas.com> > Date: May 16, 2014 at 11:06:49 AM > To: armon.dadgar(a)gmail.com armon.dadgar(a)gmail.com > <mailto:armon.dadgar@gmail.com>, openldap-its(a)openldap.org > openldap-its(a)openldap.org <mailto:openldap-its@openldap.org> > Subject: Re: (ITS#7844) LMDB Delete Cursor inconsistencies > >> armon.dadgar(a)gmail.com wrote: >> > --5372ac85_8edbdab_1271 >> > Content-Type: text/plain; charset="utf-8" >> > Content-Transfer-Encoding: quoted-printable >> > Content-Disposition: inline >> > >> > =46or now, we have application code to retry the delete until no further = >> > rows are removed. >> > Still, it would be nice to have this resolved (and tested) in master=21 >> >> Unable to reproduce the issue. I've attached my test program based on your >> description. >> >> -- >> -- Howard Chu >> CTO, Symas Corp. http://www.symas.com >> Director, Highland Sun http://highlandsun.com/hyc/ >> Chief Architect, OpenLDAP http://www.openldap.org/project/ >> ------------------------------------------------------------------------------ -- -- 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#7844) LMDB Delete Cursor inconsistencies
by armon.dadgar＠gmail.com 16 May '14

16 May '14

--537675cf_2ae8944a_1704c Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline When I run the attached test case on my machine, I=E2=80=99m hitting the = failing case. Here is the test output:=C2=A0https://gist.github.com/armon/e529d7909fe30= 1126fc6 My steps: =24 clang 7844.c mdb.c midl.c =24 mkdir testdb =24 ./a.out =24 clang -v Apple LLVM version 5.1 (clang-503.0.40) (based on LLVM 3.4svn) Target: x86=5F64-apple-darwin13.1.0 Thread model: posix =24 uname -a Darwin Armons-MacBook-Air.local 13.1.0 Darwin Kernel Version 13.1.0: Thu = Jan 16 19:40:37 PST 2014; root:xnu-2422.90.20=7E2/RELEASE=5FX86=5F64 x86=5F= 64 Best Regards, Armon Dadgar =46rom:=C2=A0Howard Chu hyc=40symas.com Reply:=C2=A0Howard Chu hyc=40symas.com Date:=C2=A0May 16, 2014 at 11:06:49 AM To:=C2=A0armon.dadgar=40gmail.com armon.dadgar=40gmail.com, openldap-its=40= openldap.org openldap-its=40openldap.org Subject:=C2=A0 Re: (ITS=237844) LMDB Delete Cursor inconsistencies =20 armon.dadgar=40gmail.com wrote: =20 > --5372ac85=5F8edbdab=5F1271 =20 > Content-Type: text/plain; charset=3D=22utf-8=22 =20 > Content-Transfer-Encoding: quoted-printable =20 > Content-Disposition: inline =20 > =20 > =3D46or now, we have application code to retry the delete until no furt= her =3D =20 > rows are removed. =20 > Still, it would be nice to have this resolved (and tested) in master=3D= 21 =20 Unable to reproduce the issue. I've attached my test program based on you= r =20 description. =20 -- =20 -- Howard Chu =20 CTO, Symas Corp. http://www.symas.com =20 Director, Highland Sun http://highlandsun.com/hyc/ =20 Chief Architect, OpenLDAP http://www.openldap.org/project/ =20 --537675cf_2ae8944a_1704c Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline <html><head><style>body=7Bfont-family:Helvetica,Arial;font-size:13px=7D</= style></head><body style=3D=22word-wrap: break-word; -webkit-nbsp-mode: s= pace; -webkit-line-break: after-white-space;=22><div id=3D=22bloop=5Fcust= omfont=22 style=3D=22font-family:Helvetica,Arial;font-size:13px; color: r= gba(0,0,0,1.0); margin: 0px; line-height: auto;=22>When I run the attache= d test case on my machine, I=E2=80=99m hitting the failing case.</div><di= v id=3D=22bloop=5Fcustomfont=22 style=3D=22font-family:Helvetica,Arial;fo= nt-size:13px; color: rgba(0,0,0,1.0); margin: 0px; line-height: auto;=22>= <br></div><div id=3D=22bloop=5Fcustomfont=22 style=3D=22font-family:Helve= tica,Arial;font-size:13px; color: rgba(0,0,0,1.0); margin: 0px; line-heig= ht: auto;=22>Here is the test output: https://gist.github.com/armon/= e529d7909fe301126fc6</div><div id=3D=22bloop=5Fcustomfont=22 style=3D=22f= ont-family:Helvetica,Arial;font-size:13px; color: rgba(0,0,0,1.0); margin= : 0px; line-height: auto;=22><br></div><div id=3D=22bloop=5Fcustomfont=22= style=3D=22font-family:Helvetica,Arial;font-size:13px; color: rgba(0,0,0= ,1.0); margin: 0px; line-height: auto;=22>My steps:</div><div id=3D=22blo= op=5Fcustomfont=22 style=3D=22font-family:Helvetica,Arial;font-size:13px;= color: rgba(0,0,0,1.0); margin: 0px; line-height: auto;=22>=24 clang 784= 4.c mdb.c midl.c</div><div id=3D=22bloop=5Fcustomfont=22 style=3D=22font-= family:Helvetica,Arial;font-size:13px; color: rgba(0,0,0,1.0); margin: 0p= x; line-height: auto;=22>=24 mkdir testdb</div><div id=3D=22bloop=5Fcusto= mfont=22 style=3D=22font-family:Helvetica,Arial;font-size:13px; color: rg= ba(0,0,0,1.0); margin: 0px; line-height: auto;=22>=24 ./a.out</div><div i= d=3D=22bloop=5Fcustomfont=22 style=3D=22font-family:Helvetica,Arial;font-= size:13px; color: rgba(0,0,0,1.0); margin: 0px; line-height: auto;=22><br= ></div><div id=3D=22bloop=5Fcustomfont=22 style=3D=22font-family:Helvetic= a,Arial;font-size:13px; color: rgba(0,0,0,1.0); margin: 0px; line-height:= auto;=22><div id=3D=22bloop=5Fcustomfont=22 style=3D=22margin: 0px;=22>=24= clang -v</div><div id=3D=22bloop=5Fcustomfont=22 style=3D=22margin: 0px;= =22>Apple LLVM version 5.1 (clang-503.0.40) (based on LLVM 3.4svn)</div><= div id=3D=22bloop=5Fcustomfont=22 style=3D=22margin: 0px;=22>Target: x86=5F= 64-apple-darwin13.1.0</div><div id=3D=22bloop=5Fcustomfont=22 style=3D=22= margin: 0px;=22>Thread model: posix</div><div id=3D=22bloop=5Fcustomfont=22= style=3D=22margin: 0px;=22><br></div><div id=3D=22bloop=5Fcustomfont=22 = style=3D=22margin: 0px;=22><div id=3D=22bloop=5Fcustomfont=22 style=3D=22= margin: 0px;=22>=24 uname -a</div><div id=3D=22bloop=5Fcustomfont=22 styl= e=3D=22margin: 0px;=22>Darwin Armons-MacBook-Air.local 13.1.0 Darwin Kern= el Version 13.1.0: Thu Jan 16 19:40:37 PST 2014; root:xnu-2422.90.20=7E2/= RELEASE=5FX86=5F64 x86=5F64</div></div></div><div id=3D=22bloop=5Fcustomf= ont=22 style=3D=22font-family:Helvetica,Arial;font-size:13px; color: rgba= (0,0,0,1.0); margin: 0px; line-height: auto;=22><br></div> <div id=3D=22b= loop=5Fsign=5F1400272071967389952=22 class=3D=22bloop=5Fsign=22><div styl= e=3D=22font-family:helvetica,arial;font-size:13px=22>Best Regards,</div><= div style=3D=22font-family:helvetica,arial;font-size:13px=22>Armon Dadgar= <br></div></div> <div style=3D=22color:black=22><br>=46rom: <span st= yle=3D=22color:black=22>Howard Chu</span> <a href=3D=22mailto:hyc=40symas= .com=22>hyc=40symas.com</a><br>Reply: <span style=3D=22color:black=22= >Howard Chu</span> <a href=3D=22mailto:hyc=40symas.com=22>hyc=40symas.com= </a><br>Date: <span style=3D=22color:black=22>May 16, 2014 at 11:06:= 49 AM</span><br>To: <span style=3D=22color:black=22>armon.dadgar=40g= mail.com</span> <a href=3D=22mailto:armon.dadgar=40gmail.com=22>armon.dad= gar=40gmail.com</a>, <span style=3D=22color:black=22>openldap-its=40openl= dap.org</span> <a href=3D=22mailto:openldap-its=40openldap.org=22>openlda= p-its=40openldap.org</a><br>Subject: <span style=3D=22color:black=22= > Re: (ITS=237844) LMDB Delete Cursor inconsistencies <br></span></div><b= r> <blockquote type=3D=22cite=22 class=3D=22clean=5Fbq=22><span><div><div= ></div><div>armon.dadgar=40gmail.com wrote: <br>> --5372ac85=5F8edbdab=5F1271 <br>> Content-Type: text/plain; charset=3D=22utf-8=22 <br>> Content-Transfer-Encoding: quoted-printable <br>> Content-Disposition: inline <br>> <br>> =3D46or now, we have application code to retry the delete until = no further =3D <br>> rows are removed. <br>> Still, it would be nice to have this resolved (and tested) in ma= ster=3D21 <br> <br>Unable to reproduce the issue. I've attached my test program based on= your =20 <br>description. <br> <br>-- =20 <br> -- Howard Chu <br> CTO, Symas Corp. http://www.symas.com <br> Director, Highland Sun http://highlandsun.com/hyc/ <br> Chief Architect, OpenLDAP http://www.openldap.org/project/ <br><hr></div></div></span></blockquote></body></html> --537675cf_2ae8944a_1704c--

1 0

Re: (ITS#7844) LMDB Delete Cursor inconsistencies
by hyc＠symas.com 16 May '14

16 May '14

This is a multi-part message in MIME format. --------------010403060500020806080002 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit armon.dadgar(a)gmail.com wrote: > --5372ac85_8edbdab_1271 > Content-Type: text/plain; charset="utf-8" > Content-Transfer-Encoding: quoted-printable > Content-Disposition: inline > > =46or now, we have application code to retry the delete until no further = > rows are removed. > Still, it would be nice to have this resolved (and tested) in master=21 Unable to reproduce the issue. I've attached my test program based on your description. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/ --------------010403060500020806080002 Content-Type: text/plain; charset=UTF-8; name="7844.c.txt" Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename="7844.c.txt" /* 7844.c - memory-mapped database tester/toy */ /* * Copyright 2014 Howard Chu, Symas Corp. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted only as authorized by the OpenLDAP * Public License. * * A copy of this license is available in the file LICENSE in the * top-level directory of the distribution or, alternatively, at * <http://www.OpenLDAP.org/license.html>. */ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <uuid/uuid.h> #include "lmdb.h" #define E(expr) CHECK((rc = (expr)) == MDB_SUCCESS, #expr) #define RES(err, expr) ((rc = expr) == (err) || (CHECK(!rc, #expr), 0)) #define CHECK(test, msg) ((test) ? (void)0 : ((void)fprintf(stderr, \ "%s:%d: %s: %s\n", __FILE__, __LINE__, msg, mdb_strerror(rc)), abort())) #define UUID_LEN 36 typedef char kval[UUID_LEN*2+2]; char sval[100]; int main(int argc,char * argv[]) { int i = 0, j = 0, rc; MDB_env *env; MDB_dbi kdbi, ddbi; MDB_val key, ikey, data; MDB_txn *txn; MDB_stat mst; MDB_cursor *cursor, *cur2; int count; unsigned long lastid = 0; kval *keys; char prefix[UUID_LEN+1], *ptr; uuid_t uuid; count = 128; keys = (kval *)malloc(count*sizeof(kval)); E(mdb_env_create(&env)); E(mdb_env_set_mapsize(env, 128*1024*1024)); E(mdb_env_set_maxdbs(env, 4)); E(mdb_env_open(env, "./testdb", MDB_NOSYNC, 0664)); E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_open(txn, "kvs", MDB_CREATE|MDB_INTEGERKEY, &ddbi)); E(mdb_open(txn, "kvs_id_idx", MDB_CREATE, &kdbi)); key.mv_size = UUID_LEN*2+1; data.mv_size = sizeof(sval); data.mv_data = sval; ikey.mv_size = sizeof(lastid); ikey.mv_data = &lastid; uuid_generate(uuid); uuid_unparse(uuid, prefix); printf("Adding %d values\n", count); for (i=0;i<count;i++) { uuid_generate(uuid); ptr = (char *)&keys[i]; memcpy(ptr, prefix, UUID_LEN); ptr[UUID_LEN] = '/'; uuid_unparse(uuid, ptr+UUID_LEN+1); ptr[UUID_LEN*2+1] = '\0'; printf("\t%s\n", ptr); key.mv_data = ptr; sprintf(sval, "Some stuff for key %s", ptr); lastid++; if (RES(MDB_KEYEXIST, mdb_put(txn, kdbi, &key, &ikey, MDB_NOOVERWRITE))) { j++; ikey.mv_size = sizeof(lastid); ikey.mv_data = &lastid; } E(mdb_put(txn, ddbi, &ikey, &data, MDB_NOOVERWRITE)); } if (j) printf("%d duplicates skipped\n", j); E(mdb_txn_commit(txn)); E(mdb_env_stat(env, &mst)); uuid_generate(uuid); uuid_unparse(uuid, prefix); printf("Adding %d values\n", count); E(mdb_txn_begin(env, NULL, 0, &txn)); for (i=0;i<count;i++) { uuid_generate(uuid); ptr = (char *)&keys[i]; memcpy(ptr, prefix, UUID_LEN); ptr[UUID_LEN] = '/'; uuid_unparse(uuid, ptr+UUID_LEN+1); ptr[UUID_LEN*2+1] = '\0'; printf("\t%s\n", ptr); key.mv_data = ptr; sprintf(sval, "Some stuff for key %s", ptr); lastid++; if (RES(MDB_KEYEXIST, mdb_put(txn, kdbi, &key, &ikey, MDB_NOOVERWRITE))) { j++; ikey.mv_size = sizeof(lastid); ikey.mv_data = &lastid; } E(mdb_put(txn, ddbi, &ikey, &data, MDB_NOOVERWRITE)); } if (j) printf("%d duplicates skipped\n", j); E(mdb_txn_commit(txn)); E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_cursor_open(txn, kdbi, &cursor)); E(mdb_cursor_open(txn, ddbi, &cur2)); key.mv_size = UUID_LEN; key.mv_data = prefix; j = 0; i = count; rc = mdb_cursor_get(cursor, &key, &ikey, MDB_SET_RANGE); while (rc == 0) { #if 0 printf("key: %p %.*s, data: %p %.*s\n", key.mv_data, (int) key.mv_size, (char *) key.mv_data, data.mv_data, (int) data.mv_size, (char *) data.mv_data); #endif E(mdb_cursor_get(cur2, &ikey, &data, MDB_SET)); E(mdb_cursor_del(cur2, 0)); E(mdb_cursor_del(cursor, 0)); j++; i--; if (i == 0) break; rc = mdb_cursor_get(cursor, &key, &ikey, MDB_NEXT); } mdb_cursor_close(cursor); mdb_cursor_close(cur2); E(mdb_txn_commit(txn)); printf("Deleted %d values\n", j); mdb_env_close(env); return 0; } --------------010403060500020806080002--

1 0

(ITS#7856) TLS_REQCERT try is same as TLS_REQCERT hard?
by pguenther＠proofpoint.com 16 May '14

16 May '14

Full_Name: Philip Guenther Version: 2.4.39 OS: OpenBSD URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (76.253.0.176) The ldap.conf(5) manpage says this about TLS_REQCERT TLS_REQCERT <level> Specifies what checks to perform on server certificates in a TLS session, if any. The <level> can be specified as one of the following keywords: ... try The server certificate is requested. If no certificate is provided, the session proceeds normally. If a bad certificate is provided, the session is immediately terminated. demand | hard These keywords are equivalent. The server certificate is requested. If no certificate is provided, or a bad certificate is provided, the session is immediately terminated. This is the default setting. In testing, I can find no difference in behavior between the 'try' and 'hard' keywords. For the ldap* tools, both 'try' and 'hard' seem to place the same requirements on the server. What does "if no certificate is provided" *mean* in terms of server and/or client configuration?

1 0

Re: (ITS#7705) mdb back-end segfaults sporadically with paged searches
by hyc＠symas.com 16 May '14

16 May '14

openldap(a)semyon.org wrote: > On 13-09-23 10:16 PM, Howard Chu wrote: > >> If you backup the DB with slapcat and reload it on another server with >> slapadd, can you still reproduce the fault on the copy? > > Unfortunately, yes. It breaks trying to retrieve the same record as > before, on the same instruction in back_mdb: > > yesterday: > [126393.233615] slapd[19244]: segfault at 7f499f38d000 ip > 00007f4da0e9b6a1 sp 00007f499f1fa540 error 6 in > back_mdb-2.4.so.2.9.2[7f4da0e80000+34000] > > today: > [517860.953779] slapd[24871]: segfault at 7fdeb50a0000 ip > 00007fe2b63ad6a1 sp 00007fdeb4f0d540 error 6 in > back_mdb-2.4.so.2.9.2[7fe2b6392000+34000] > > Semyon A fix is now in git master. -- -- 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#7855) Update to latest autotools to support new architecture ppc64le
by hyc＠symas.com 15 May '14

15 May '14

rajesh(a)linux.vnet.ibm.com wrote: > Full_Name: Rajeshkumar S > Version: 2.4.39 > OS: Ubuntu 14.04 > URL: ftp://ftp.openldap.org/incoming/ > Submission from: (NULL) (122.248.161.59) > > > > I am trying to build the openldap package from the source following the > release tarball from > ftp://ftp.openldap.org/pub/OpenLDAP/openldap-release/openldap-2.4.39.tgz. in a > new architecture ppc64le ( IBM PowerPC Little endian ). As the config.guess and > libtool did not have the required patches to identify this new architecture, I > did autoreconf -f -i in my build system whose latest automake and libtool has > the patches > of ppc64le. > autoreconf fails with automake errors as shown below > > automake: error: no 'Makefile.am' found for any configure output > autoreconf: automake failed with exit status: 1 That's to be expected, since there certainly are no Makefile.am files in our source tree. We don't use automake. This is not a bug. > As discussed in the mailing list, Quanah Gibson-Mount suggested me to request an > update for auto-tools to support new architectures. > > The automake with version >= 1.13.4 has the correct config.guess with the > required ppc64le fixes. > Regarding the libtool, the last release is more than 2 years ago I believe, we > managed to get an alpha source release which has all the latest patches to > support the new architecture ppc64le > > An alpha version of libtool with ppc64le support is available at > ftp://alpha.gnu.org/gnu/libtool/libtool-2.4.2.418.tar.gz. Libtool is so perpetually problematic that we will certainly not integrate an alpha version into our source tree. > Hence I request for an update of the auto tools in your build system to support > these new architectures. > > Thanks and Regards > Rajeshkumar S > Linux Technology Center, IBM > > -- -- 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#7855) Update to latest autotools to support new architecture ppc64le
by rajesh＠linux.vnet.ibm.com 15 May '14

15 May '14

Full_Name: Rajeshkumar S Version: 2.4.39 OS: Ubuntu 14.04 URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (122.248.161.59) I am trying to build the openldap package from the source following the release tarball from ftp://ftp.openldap.org/pub/OpenLDAP/openldap-release/openldap-2.4.39.tgz. in a new architecture ppc64le ( IBM PowerPC Little endian ). As the config.guess and libtool did not have the required patches to identify this new architecture, I did autoreconf -f -i in my build system whose latest automake and libtool has the patches of ppc64le. autoreconf fails with automake errors as shown below automake: error: no 'Makefile.am' found for any configure output autoreconf: automake failed with exit status: 1 As discussed in the mailing list, Quanah Gibson-Mount suggested me to request an update for auto-tools to support new architectures. The automake with version >= 1.13.4 has the correct config.guess with the required ppc64le fixes. Regarding the libtool, the last release is more than 2 years ago I believe, we managed to get an alpha source release which has all the latest patches to support the new architecture ppc64le An alpha version of libtool with ppc64le support is available at ftp://alpha.gnu.org/gnu/libtool/libtool-2.4.2.418.tar.gz. Hence I request for an update of the auto tools in your build system to support these new architectures. Thanks and Regards Rajeshkumar S Linux Technology Center, IBM

1 0

Re: (ITS#7853) Improve safety of mdb_env_close
by hbf＠mail.uio.no 14 May '14

14 May '14

On 05/14/2014 01:18 AM, hyc(a)symas.com wrote: > dw(a)botanicus.net wrote: >> Currently if a user (wilfully or accidentally, say, through composition of >> third party libs) opens the same LMDB environment twice within a process, and >> maintains active read transactions at the time one mdb_env_close() is called, >> all reader slots will be deallocated in all environments due to the logic >> around line 4253 that unconditionally clears reader slots based on PID. >> >> I'd like to avoid this in py-lmdb, and seems there are a few alternatives to >> fix it: >> >> 1) Tell users not to do that (doc fix, nobody reads docs) > > This is already in the doc - don't open the same env twice within one process. 1.5) Use something else than fcntl locks on the lock-file descriptor to control whether to re-init the lockfile and its mutexes/semaphores. Like MDB_NOLOCK. Needed to get any other suggestions off the ground, see Caveats in lmdb.h. Once two MDB_env's in a process has the lockfile open, both lose their fcntl locks when either closes it - and the next mdb_env_open() will reset the lockfile. Then both readers and writers are screwed, and more care with reader slots won't help. Well, one MDB_env could leak its lockfile descriptor instead of closing it, or it could give the descriptor to the other MDB_env for a deferred close if it could find it using some global state. (Could work until someone wrote an open-close loop which would run into the max-open-FD limit.) Either way, no fun. Another way would be #ifdef ... more unportable code ... #endif:-( Surely some OSes have come up with sane way to address this problem, there's just no portable way. >> 2) Externally maintain a set of (st_dev, st_ino) keys representing already >> opened environments, and check any attempt to open an environment against this >> list, failing if data.mdb's key already appears. > > That would require global library state, which is definitely unwanted. Though in one sense we already have that, which is the problem. Global (well, per-process) lock state on open files and mutexes. >> 3) Disable the mdb_env_close() cleanup loop when MDB_NOTLS is enabled. Since no >> reader slot will ever exist in TLS in this mode, the existing mdb_env_close() >> doc "All transactions, databases, and cursors must already be closed before >> calling this function." ensures no readers exist at close time, and the loop >> need not run. > > Could do this, but since it's an incomplete solution, not sure there's any point. That's MDB_NOLOCK anyway. MDB_NOTLS does have reader table entries. > (...) -- Hallvard

1 0

(ITS#7854) minor patches to allow compile with VS2012
by louis＠steelbytes.com 14 May '14

14 May '14

Full_Name: Louis Solomon Version: 2.4.39 OS: Win32 & Win64 URL: Submission from: (NULL) (59.167.196.195) 1) in libraries/libldap/ldif.c move int ldif_debug = 0; to after _all_ the includes. Why? because the linker gets confused with it being declared as declspec(dllexport). Order matters. 2) in include/ac/socket.h EWOULDBLOCK, EINPROGRESS, ETIMEDOUT (lines 101-103) conflict with the windows header files. simply wrap with any rational choice of ifdef/ifndef 3) in include/ldap_utf8.h wrap prototypes for ldap_x_utf8_to_wc (line 61) through to ldap_x_mbs_utf8s (line 99) with #if SIZEOF_WCHAR>=4 to match the ifdef at the top of utf-8-conv.c. Why? because the linker tries to export them as they are defined in the header with declspec(dllexport) but are not in any .o/.obj 4) in include/ldap_cdefs.h the usage of __declspec(dllexport/dllimport) is not quite right ... should be like #if defined(_WIN32) ... # define LDAP_F(type) __declspec(dllimport) type # define LDAP_V(type) extern __declspec(dllimport) type #else # define LDAP_F(type) __declspec(dllexport) type # define LDAP_V(type) extern __declspec(dllexport) type #endif and of course the same for each of the other sets (LBER, LDAP, LDAP_AVL, LDAP_LDIF, etc) With the above patches and making the appropriate .h files from .hin files manually and then creating VS2012 project files, I was successfully able to create and use liblber, libldap, libldap_r as .dlls

1 0

Re: (ITS#7844) LMDB Delete Cursor inconsistencies
by armon.dadgar＠gmail.com 14 May '14

14 May '14

--5372ac85_8edbdab_1271 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline =46or now, we have application code to retry the delete until no further = rows are removed. Still, it would be nice to have this resolved (and tested) in master=21 Best Regards, Armon Dadgar =46rom:=C2=A0David Wilson dw=40hmmz.org Reply:=C2=A0David Wilson dw=40hmmz.org Date:=C2=A0May 13, 2014 at 3:41:08 PM To:=C2=A0armon.dadgar=40gmail.com armon.dadgar=40gmail.com Cc:=C2=A0openldap-its=40openldap.org openldap-its=40openldap.org Subject:=C2=A0 Re: (ITS=237844) LMDB Delete Cursor inconsistencies =20 This sounds very much like a known regression in mdb.master in the past =20 few months. If you are still affected, please retest against LMDB 0.9.11 = =20 which at least was not affected by the regression. =20 David =20 --5372ac85_8edbdab_1271 Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline <html><head><style>body=7Bfont-family:Helvetica,Arial;font-size:13px=7D</= style></head><body style=3D=22word-wrap: break-word; -webkit-nbsp-mode: s= pace; -webkit-line-break: after-white-space;=22><div id=3D=22bloop=5Fcust= omfont=22 style=3D=22font-family:Helvetica,Arial;font-size:13px; color: r= gba(0,0,0,1.0); margin: 0px; line-height: auto;=22>=46or now, we have app= lication code to retry the delete until no further rows are removed.</div= ><div id=3D=22bloop=5Fcustomfont=22 style=3D=22font-family:Helvetica,Aria= l;font-size:13px; color: rgba(0,0,0,1.0); margin: 0px; line-height: auto;= =22>Still, it would be nice to have this resolved (and tested) in master=21= </div><div id=3D=22bloop=5Fcustomfont=22 style=3D=22font-family:Helvetica= ,Arial;font-size:13px; color: rgba(0,0,0,1.0); margin: 0px; line-height: = auto;=22><br></div> <div id=3D=22bloop=5Fsign=5F1400024102388026112=22 cl= ass=3D=22bloop=5Fsign=22><div style=3D=22font-family:helvetica,arial;font= -size:13px=22>Best Regards,</div><div style=3D=22font-family:helvetica,ar= ial;font-size:13px=22>Armon Dadgar<br></div></div> <div style=3D=22color:= black=22><br>=46rom: <span style=3D=22color:black=22>David Wilson</s= pan> <a href=3D=22mailto:dw=40hmmz.org=22>dw=40hmmz.org</a><br>Reply:&nbs= p;<span style=3D=22color:black=22>David Wilson</span> <a href=3D=22mailto= :dw=40hmmz.org=22>dw=40hmmz.org</a><br>Date: <span style=3D=22color:= black=22>May 13, 2014 at 3:41:08 PM</span><br>To: <span style=3D=22c= olor:black=22>armon.dadgar=40gmail.com</span> <a href=3D=22mailto:armon.d= adgar=40gmail.com=22>armon.dadgar=40gmail.com</a><br>Cc: <span style= =3D=22color:black=22>openldap-its=40openldap.org</span> <a href=3D=22mail= to:openldap-its=40openldap.org=22>openldap-its=40openldap.org</a><br>Subj= ect: <span style=3D=22color:black=22> Re: (ITS=237844) LMDB Delete C= ursor inconsistencies <br></span></div><br> <blockquote type=3D=22cite=22= class=3D=22clean=5Fbq=22><span><div><div></div><div>This sounds very muc= h like a known regression in mdb.master in the past <br>few months. If you are still affected, please retest against LMDB 0.9= .11 <br>which at least was not affected by the regression. <br> <br> <br>David <br></div></div></span></blockquote></body></html> --5372ac85_8edbdab_1271--

1 0

Re: (ITS#7853) Improve safety of mdb_env_close
by hyc＠symas.com 14 May '14

14 May '14

dw(a)botanicus.net wrote: > Full_Name: David Wilson > Version: LMDB 0.9.11 > OS: > URL: https://raw.githubusercontent.com/dw/py-lmdb/master/misc/readers_mrb_env.pa… > Submission from: (NULL) (178.238.153.20) > > > Currently if a user (wilfully or accidentally, say, through composition of > third party libs) opens the same LMDB environment twice within a process, and > maintains active read transactions at the time one mdb_env_close() is called, > all reader slots will be deallocated in all environments due to the logic > around line 4253 that unconditionally clears reader slots based on PID. > > I'd like to avoid this in py-lmdb, and seems there are a few alternatives to > fix it: > > > 1) Tell users not to do that (doc fix, nobody reads docs) This is already in the doc - don't open the same env twice within one process. > 2) Externally maintain a set of (st_dev, st_ino) keys representing already > opened environments, and check any attempt to open an environment against this > list, failing if data.mdb's key already appears. That would require global library state, which is definitely unwanted. > 3) Disable the mdb_env_close() cleanup loop when MDB_NOTLS is enabled. Since no > reader slot will ever exist in TLS in this mode, the existing mdb_env_close() > doc "All transactions, databases, and cursors must already be closed before > calling this function." ensures no readers exist at close time, and the loop > need not run. Could do this, but since it's an incomplete solution, not sure there's any point. > 4) Modify lock.mdb to include MDB_env* address within the process, and update > mdb_env_close() to invalidate only readers associated with the environment > being closed. I dislike using the MDB_env* as an opaque publicly visible > cookie, but perhaps storing this field might be reused for other things in > future. Hmm. I guess this could work. I think storing the MDB_env* is harmless since the pointer is meaningless to any other process. > Option 3 lets the user wilfully mix py-lmdb Environment objects within the > process, since py-lmdb always enables NOTLS, but it does not fix the case where > the user is integrating a Python application with a C library with some opaque > interface, and both the application and the library independently have need to > access the environment. Allowing multiple independent libraries to open the same environment would consume redundant chunks of address space. For large enough DBs this may simply fail. Unfortunately the semantics of fcntl locks prevents us from even detecting that this problem has occurred. > This kind of scenario regularly crops up in huge software projects, the user > may not even be aware the library uses LMDB internally. In that case, the user > is again exposed to having their reader slots silently become deallocated > through no fault of their own. > > > Option 4 is the least aesthetic, but has all the benefits of option 3 in > addition to preventing the "accidental integration" scenario. I've attached a > patch for it (against 0.9.11), although I completely understand if it does not > get applied. :) > > One alternative would be instead of using MDB_env* address, using a > monotonically increasing counter based on a static global, say, "mrb_token", > but this might break in the case two copies of LMDB were somehow linked in the > application. (It's easy to argue the user is an idiot in this case ;) Global state is verboten. > > > David > > -- -- 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

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

openldap-bugs