openldap-bugs

openldap-bugs@openldap.org

1 participants
20961 discussions

(ITS#8360) apparent MDB corruption when hitting maxsize
by peter.van.dijk＠powerdns.com 25 Jan '16

25 Jan '16

Full_Name: Peter van Dijk Version: 2.4.43 OS: Mac OS X 10.10.5 URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (2001:610:666:0:c92b:7aa7:bc73:c430) The procedure follow is fully repeatable with consistent problematic results for me. (1) ftp://ftp.openldap.org/pub/OpenLDAP/openldap-release/openldap-2.4.43.tgz, installed via osx homebrew with configure flags --prefix=/usr/local/Cellar/openlp%p/2.4.43 --sysconfdir=/usr/local/etc --localstatedir=/usr/local/var (2) config and related files as uploaded on the given URL (cosine.schema is unedited but included for completeness) (3) /usr/local/var/openldap-data/ is empty (4) sudo /usr/local/opt/openldap/libexec/slapd -d 255 -u peter 2> slapd.log (5) ldapmodify -D 'uid=peter,o=habbie' -w secret -c < named.conf.ldif 2> ldapmodify.log ; ldapdelete -D 'uid=peter,o=habbie' -w secret -r ou=dns,o=habbie 2> ldapdelete.log modify is unhappy, initially because of some unknown attributes (thish is why I used -c), then later because it is hitting the MDB maxsize of 10 megabytes. ldapdelete then fails to delete the ou=dns tree due to MDB errors (6) increasing maxsize after this 'corruption' has happened does NOT enable ldapdelete to suddenly work (7) (with slapd stopped, not sure it matters): $ /usr/local/opt/openldap/sbin/slapcat outputs 191771 lines to stdout before aborting with: 56a63063 mdb_entry_decode: attribute index 24 not recognized Segmentation fault: 11 (8) as requested by hyc on IRC, copying here for completeness $ mdb_dump -p -s ad2i /usr/local/var/openldap-data/ VERSION=3 format=print database=ad2i type=btree mapsize=10485760 maxreaders=126 tetegerkey=1 db_pagesize=4096 HEADER=END \01\00\00\00 objectClass \02\00\00\00 o \03\00\00\00 structuralObjectClass \04\00\00\00 entryUUID \05\00\00\00 creatorsName \06\00\00\00 createTimestamp \07\00\00\00 entryCSN \08\00\00\00 modifiersName \09\00\00\00 modifyTimestamp \0a\00\00\00 ou \0b\00\00\00 dc \0c\00\00\00 associatedDomain \0d\00\00\00 dNSTTL \0e\00\00\00 sOARecord \0f\00\00\00 nSRecord \10\00\00\00 mXRecord \11\00\00\00 aRecord \12\00\00\00 cMEReRecord \13\00\00\00 LocRecord \14\00\00\00 hInfoRecord \15\00\00\00 tXTRecord \16\00\00\00 aAAARecord DATA=END

1 0

Re: (ITS#8355) [lmdb] Issues with sorted duplicate DBs with cursor delete
by hhclaw.eb＠gmail.com 25 Jan '16

25 Jan '16

On Monday, January 25, 2016 03:49 AM, Howard Chu wrote: > HLaw wrote: >> >> >> On Sunday, January 24, 2016 07:44 PM, Howard Chu wrote: >>> hhclaw.eb(a)gmail.com wrote: >>>> Full_Name: H Law >>>> Version: LMDB >>>> OS: Linux >>>> URL: >>>> Submission from: (NULL) (42.2.241.129) >>>> >>>> >>>> It seems that, after using a cursor delete during a cursor >>>> traversal on a dup >>>> sort database, the cursor is in a strange state, when MDB_NEXT / >>>> MDB_NEXT_DUP >>>> ceases to work properly while MDB_PREV / MDB_PREV_DUP still >>>> functions. >>> >>> Thanks for the report. Fixed now in git. >>> >> Thank you. It now works except for the case when the deleted record >> is just >> before the end of the database. >> In this case the next call to cursor get with MDB_NEXT / MDB_NEXT_DUP >> will >> still hang the program, instead of returning not found (or other >> errors). (It >> seems to be stuck at some wait state without heavy cpu usage). > > Your test program is broken, you are trying to create a new write txn > in the same thread as an existing write txn. > Yes, thanks, works after fixing that. Sorry for bothering.

1 0

Re: (ITS#8355) [lmdb] Issues with sorted duplicate DBs with cursor delete
by hyc＠symas.com 24 Jan '16

24 Jan '16

HLaw wrote: > > > On Sunday, January 24, 2016 07:44 PM, Howard Chu wrote: >> hhclaw.eb(a)gmail.com wrote: >>> Full_Name: H Law >>> Version: LMDB >>> OS: Linux >>> URL: >>> Submission from: (NULL) (42.2.241.129) >>> >>> >>> It seems that, after using a cursor delete during a cursor traversal on a dup >>> sort database, the cursor is in a strange state, when MDB_NEXT / MDB_NEXT_DUP >>> ceases to work properly while MDB_PREV / MDB_PREV_DUP still functions. >> >> Thanks for the report. Fixed now in git. >> > Thank you. It now works except for the case when the deleted record is just > before the end of the database. > In this case the next call to cursor get with MDB_NEXT / MDB_NEXT_DUP will > still hang the program, instead of returning not found (or other errors). (It > seems to be stuck at some wait state without heavy cpu usage). Your test program is broken, you are trying to create a new write txn in the same thread as an existing write txn. -- -- 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#8355) [lmdb] Issues with sorted duplicate DBs with cursor delete
by hhclaw.eb＠gmail.com 24 Jan '16

24 Jan '16

This is a multi-part message in MIME format. --------------010901030100040802060200 Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit On Sunday, January 24, 2016 07:44 PM, Howard Chu wrote: > hhclaw.eb(a)gmail.com wrote: >> Full_Name: H Law >> Version: LMDB >> OS: Linux >> URL: >> Submission from: (NULL) (42.2.241.129) >> >> >> It seems that, after using a cursor delete during a cursor traversal >> on a dup >> sort database, the cursor is in a strange state, when MDB_NEXT / >> MDB_NEXT_DUP >> ceases to work properly while MDB_PREV / MDB_PREV_DUP still functions. > > Thanks for the report. Fixed now in git. > Thank you. It now works except for the case when the deleted record is just before the end of the database. In this case the next call to cursor get with MDB_NEXT / MDB_NEXT_DUP will still hang the program, instead of returning not found (or other errors). (It seems to be stuck at some wait state without heavy cpu usage). I attach a version of mtest3.c modified to show the issue. >> In particular, when MDB_NEXT or MDB_NEXT_DUP is called after cursor >> deletion, if >> next key/value pair exists, the cursor will not advance, and got >> stuck by >> returning the same record when MDB_NEXT or MDB_NEXT_DUP is called >> repeatly. In >> case there is no next record, the program was hang. >> >> The following modified version of mtest3.c shows the issue. I am >> testing this >> on the latest commit 20dec1f69bf4860202c764ce92b1fbbe3d11a065 of lmdb >> on 20 Jan, >> on x86-64 Linux. >> I got a similar behaviour when a slightly earlier version of lmdb was >> cross-compiled with a Java wrapper for use on Android, which is why I >> am testing >> this. The issue should therefore not be platform specific. > > > --------------010901030100040802060200 Content-Type: text/x-csrc; name="mtest3.c" Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename="mtest3.c" /* mtest3.c - memory-mapped database tester/toy */ /* * Copyright 2011-2015 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>. */ /* Tests for sorted duplicate DBs with cursor delete */ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <time.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())) int main(int argc,char * argv[]) { int i = 0, j = 0, rc; MDB_env *env; MDB_dbi dbi; MDB_val key, data; MDB_txn *txn; MDB_stat mst; MDB_cursor *cursor; int count; int *values; char sval[32]; char kval[sizeof(int)]; srand(time(NULL)); memset(sval, 0, sizeof(sval)); count = 10; values = (int *)malloc(count*sizeof(int)); for(i = 0;i<count;i++) { values[i]= i * 10; } E(mdb_env_create(&env)); E(mdb_env_set_mapsize(env, 10485760)); E(mdb_env_set_maxdbs(env, 4)); E(mdb_env_open(env, "./testdb", MDB_FIXEDMAP|MDB_NOSYNC, 0664)); E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_dbi_open(txn, "id2", MDB_CREATE|MDB_DUPSORT, &dbi)); key.mv_size = sizeof(int); key.mv_data = kval; data.mv_size = sizeof(sval); data.mv_data = sval; printf("Adding %d values\n", count); for (i=0;i<count;i++) { if (!(i & 0x07)) sprintf(kval, "%03x", values[i]); sprintf(sval, "%03x %d foo bar", values[i], values[i]); if (RES(MDB_KEYEXIST, mdb_put(txn, dbi, &key, &data, MDB_NODUPDATA))) j++; } if (j) printf("%d duplicates skipped\n", j); E(mdb_txn_commit(txn)); E(mdb_env_stat(env, &mst)); E(mdb_txn_begin(env, NULL, MDB_RDONLY, &txn)); E(mdb_cursor_open(txn, dbi, &cursor)); while ((rc = mdb_cursor_get(cursor, &key, &data, MDB_NEXT)) == 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); } CHECK(rc == MDB_NOTFOUND, "mdb_cursor_get"); mdb_cursor_close(cursor); mdb_txn_abort(txn); E(mdb_env_stat(env, &mst)); E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_cursor_open(txn, dbi, &cursor)); /* * There are 2 items with key = values[8] at the end of the list */ sprintf(kval, "%03x", values[8]); key.mv_size = sizeof(int); key.mv_data = kval; printf("\nCursor set / first dup\n"); E(mdb_cursor_get(cursor, &key, &data, MDB_SET_KEY)); E(mdb_cursor_get(cursor, &key, &data, MDB_FIRST_DUP)); printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data, (int) data.mv_size, (char *) data.mv_data); printf("Cursor next\n"); j=0; /* * The program will hang at the following mdb_cursor_get * after cursor delete */ while ((rc = mdb_cursor_get(cursor, &key, &data, MDB_NEXT_DUP)) == 0) { printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data, (int) data.mv_size, (char *) data.mv_data); j++; if (j == 1) { printf("delete the above key/data\n"); if (RES(MDB_NOTFOUND, mdb_cursor_del(cursor, 0))) { mdb_cursor_close(cursor); mdb_txn_abort(txn); break; } } if (j > count) { printf("Should not be there\n"); break; } } if (j > 1) { mdb_cursor_close(cursor); E(mdb_txn_commit(txn)); } E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_cursor_open(txn, dbi, &cursor)); sprintf(kval, "%03x", values[0]); key.mv_size = sizeof(int); key.mv_data = kval; printf("\nCursor set / last dup\n"); E(mdb_cursor_get(cursor, &key, &data, MDB_SET_KEY)); E(mdb_cursor_get(cursor, &key, &data, MDB_LAST_DUP)); printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data, (int) data.mv_size, (char *) data.mv_data); printf("Cursor prev\n"); j=0; while ((rc = mdb_cursor_get(cursor, &key, &data, MDB_PREV_DUP)) == 0) { printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data, (int) data.mv_size, (char *) data.mv_data); j++; if (j == 1) { printf("Delete the above key/data\n"); if (RES(MDB_NOTFOUND, mdb_cursor_del(cursor, 0))) { mdb_cursor_close(cursor); mdb_txn_abort(txn); break; } } if (j > count) { printf("Should not be there\n"); break; } } if (j > 1) { mdb_cursor_close(cursor); E(mdb_txn_commit(txn)); } mdb_dbi_close(env, dbi); mdb_env_close(env); free (values); return 0; } --------------010901030100040802060200--

1 0

Re: (ITS#8355) [lmdb] Issues with sorted duplicate DBs with cursor delete
by hyc＠symas.com 24 Jan '16

24 Jan '16

hhclaw.eb(a)gmail.com wrote: > Full_Name: H Law > Version: LMDB > OS: Linux > URL: > Submission from: (NULL) (42.2.241.129) > > > It seems that, after using a cursor delete during a cursor traversal on a dup > sort database, the cursor is in a strange state, when MDB_NEXT / MDB_NEXT_DUP > ceases to work properly while MDB_PREV / MDB_PREV_DUP still functions. Thanks for the report. Fixed now in git. > In particular, when MDB_NEXT or MDB_NEXT_DUP is called after cursor deletion, if > next key/value pair exists, the cursor will not advance, and got stuck by > returning the same record when MDB_NEXT or MDB_NEXT_DUP is called repeatly. In > case there is no next record, the program was hang. > > The following modified version of mtest3.c shows the issue. I am testing this > on the latest commit 20dec1f69bf4860202c764ce92b1fbbe3d11a065 of lmdb on 20 Jan, > on x86-64 Linux. > I got a similar behaviour when a slightly earlier version of lmdb was > cross-compiled with a Java wrapper for use on Android, which is why I am testing > this. The issue should therefore not be platform specific. -- -- 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#8354) syncprov segv
by hyc＠symas.com 23 Jan '16

23 Jan '16

Thomas Pressnell wrote: > Thanks, > > Essentially the same patch as i had (destroyed sop->s_mutex in abandon > check rather than init later). > > I did place the o_abandon test under the while( si->si_active ) loop > instead of above just to ensure the check is under the same lock hold > as the list modification. Ah good point. Yeah, I'll move the check down. > > Ill grab whats in git and start tests on monday > > Have a good weekend > > Tom > -- -- 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#8354) syncprov segv
by tpretz＠gmail.com 23 Jan '16

23 Jan '16

Thanks, Essentially the same patch as i had (destroyed sop->s_mutex in abandon check rather than init later). I did place the o_abandon test under the while( si->si_active ) loop instead of above just to ensure the check is under the same lock hold as the list modification. Ill grab whats in git and start tests on monday Have a good weekend Tom

1 0

(ITS#8359) [test] Issues with sorted duplicate DBs with cursor delete
by abcdef.gh＠gmail.com 22 Jan '16

22 Jan '16

Full_Name: H Chu Version: LMDB OS: URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (78.155.231.129) It seems that, after using a cursor delete during a cursor traversal on a dup sort database, the cursor is in a strange state, when MDB_NEXT / MDB_NEXT_DUP ceases to work properly while MDB_PREV / MDB_PREV_DUP still functions. In particular, when MDB_NEXT or MDB_NEXT_DUP is called after cursor deletion, if next key/value pair exists, the cursor will not advance, and got stuck by returning the same record when MDB_NEXT or MDB_NEXT_DUP is called repeatly. In case there is no next record, the program was hang. The following modified version of mtest3.c shows the issue. I am testing this on the latest commit 20dec1f69bf4860202c764ce92b1fbbe3d11a065 of lmdb on 20 Jan, on x86-64 Linux. I got a similar behaviour when a slightly earlier version of lmdb was cross-compiled with a Java wrapper for use on Android, which is why I am testing this. The issue should therefore not be platform specific. Thank you. ==== /* mtest3.c - memory-mapped database tester/y y */ /* * Copyrit t 2011-2015 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>. */ /* Tests for sorted duplicate DBs with cursor delete */ #include <stdio.h> #include <stib.h.h> #include <string.h> #include <time.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)%3: ((void)fprintf(stderr, \ "%s:%d: %s: %s\n", __FILE__, __LINE__, msg, mdb_strerror(rc)), abort())) int main(int argc,char * argv[]) { int i = 0, j = 0, rc; MDB_env *env; MDB_dbi dbi; MDB_val key, data; MDB_txn *txn; MDB_stat mst; MDB_cursor *cursor; int count; int *values; char sval[32]; char kval[sizeof(int)]; srand(time(NULL)); memset(sval, 0, sizeof(sval)); count = 10; values = (int *)malloc(count*sizeof(int)); for(i = 0;i<count;i++) { values[i]= i * 10; } E(mdb_env_create(&env)); E(mdb_env_set_mapsize(env, 10485760)); E(mdb_env_set_maxdbs(env, 4)); E(mdb_env_open(env, "./testdb", MDB_FIXEDMAP|MDB_NOSYNC, 0664)); E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_dbi_open(n%n, "id2", MDB_CREATE|MDB_DUPSORT, &dbi)); key.mv_size = sizeof(int); key.mv_data = kval; data.mv_size = sizeof(sval); data.mv_data = sval; printf("Adding %d values\n", count); for (i=0;i<count;i++) { if (!(i & 0x07)) sprintf(kval, "%03x",alalues[i])0D0D sprintf(sval, "%03x %d foo bar", values[i], values[i]); if (RES(MDB_KEYEXIST, mdb_put(txn, dbi, &key, &data, MDB_NODUPDATA))) j++; } if (j) printf("%d duplicates skipped\n", j); E(mdb_txn_commit(txn)); E(mdb_env_stat(vnv, &mst)); E(mdb_txn_begin(env, NULL, MDB_RDONLY, &txn)); E(mdb_cursor_open(txn, dbi, &cursor)); while ((rc = mdb_cursor_get(cursor, &key, &data, MDB_NEXT_DUP)) == 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); } CHECK(rc == MDB_NOTFOUND, "mdb_cursor_get"); mdb_cursor_close(cursor); mdb_txn_abort(txn); E(mdb_env_stat(env, &mst)); E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_cursor_open(txn% d dbi, &cursor)); /* * changing following to values[8] freezes program during mdb_cursor_del * as there is no more record following the deletion */ sprintf(kval, "%03x", values[0]); key.mv_size = sizeof(i)%9; key.mv_data = kval; printf("\nCursor set / first dup\n"); E(mdb_cursor_get(cursor, &key, &data, MDB_SET_KEY)); E(mdb_cursor_get(cursor, &key, &data, MDB_FIRST_DUP)); printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data, (int) data.mv_size, (char *) data.mv_data); printf("Cursor next\n"); j=0; while ((rc = mdb_cursor_get(cursor, &key, &data, MDB_NEXT_DUP)) == 0) { printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data, (int) data.mv_size, (char *) data.mv_data); j++; if (j == 1) { printf("delete the above key/data\n"); if (RES(MDB_NOTFOUND, mdb_cursor_del(cursor, 0))) { mdb_cursor_close(cursor); mdb_txn_abort(txn); break; } } if (j > count) { printf("Should not be there\n"); break; } } if (j E E 1) { mdb_cursor_close(cursor); E(mdb_txn_commit(txn)); } E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_cursor_open(txn, dbi, &cursor)); sprintf(kval, "%03x", values[0]); key.mv_size = sizeof(int); key.mv_data = kval; printf("\nCursor set / last dup\n"); E(mdb_cursor_get(cursor, &key, &data, MDB_SET_KEY)); E(mdb_cursor_get(cursor, &key, &data, MDB_LAST_DUP)); printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data, (int) data.mv_size, (char *) data.mv_data); printf("Cursor prev\n"); j=0; while ((rc = mdb_cursor_get(cursor, &key, &data, MDB_PREV_DUP)) == 0) { printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data, (int) data.mv_size, (char *) data.mv_data); j++; if (j == 1) { printf("Delete the above key/data\n"); if (RES(MDB_NOTFOUND, mdb_cursor_del(cursor, 0))) { mdb_cursor_close(cursor); mdb_txn_abort(txn); break; } } if (j > count) { printf("Should not be there\n"); break; } } if (j > 1) { mdb_cursor_close(cursor); E(mdb_txn_commit(txn)); } mdb_dbi_close(env, dbi); mdb_env_close(env); free (values); return 0; } === Output: Adding 10 values key: 0x7f2c31981e9c 000, data: 0x7f2c31981fe0 000 0 foo bar key: 0x7f2c31981e9c 000, data: 0x7f2c31981fb8 00a 10 foo bar key: 0x7f2c31981e9c 000, data: 0x7f2c31981f90 014 20 foo bar key: 0x7f2c31981e9c 000, data: 0x7f2c31981f68 01e 30 foo bar key: 0x7f2c31981e9c 000, data: 0x7f2c31981f40 028 40 foo bar key: 0x7f2c31981e9c 000, data: 0x7f2c31981f18 032 50 foo bar key: 0x7f2c31981e9c 000, data: 0x7f2c31981ef0 03c 60 foo bar key: 0x7f2c31981e9c 000, data: 0x7f2c31981ec8 046 70 foo bar key: 0x7f2c31981e2c 050, data: 0x7f2c31981e74 050 80 foo bar key: 0x7f2c31981e2c 050, data: 0x7f2c31981e4c 05a 90 foo bar Cursor set / first dup key: 000, data: 000 0 foo bar Cursor next key: 000, data: 00a 10 foo bar delete the above key/data key: 000, data: 014 20 foo bar <--- Expected to get 30, 40, ... 70 for MDB_NEXT_DUP key: 000, data: 014 20 foo bar key: 000, data: 014 20 foo bar key: 000, data: 014 20 foo bar key: 000, data: 014 20 foo bar key: 000, data: 014 20 foo bar key: 000, data: 014 20 foo bar key: 000, data: 014 20 foo bar key: 000, data: 014 20 foo bar key: 000, data: 014 20 foo bar <---- But got infinite loop if not break by counter Should not be there Cursor set / last dup key: 000, data: 046 70 foo bar Cursor prev key: 000, data: 03c 60 foo bar Delete the above key/data key: 000, data:3232 50 foo bar <--- works as expected in the reverse direction key: 000, data: 028 40 foo bar key: 000, data: 01e 30 foo bar key: 000, data: 014 20 foo bar key: 000, data: 000 0 foo bar

1 0

(ITS#8358) [test] [lmdb] Issues with sorted duplicate DBs with cursor delete
by hyc＠openldap.org 22 Jan '16

22 Jan '16

Full_Name: H Chu Version: LMDB OS: URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (78.155.231.129) Submitted by: hyc It seems that, after using a cursor delete during a cursor traversal on a dup sort database, the cursor is in a strange state, when MDB_NEXT / MDB_NEXT_DUP ceases to work properly while MDB_PREV / MDB_PREV_DUP still functions. In particular, when MDB_NEXT or MDB_NEXT_DUP is called after cursor deletion, if next key/value pair exists, the cursor will not advance, and got stuck by returning the same record when MDB_NEXT or MDB_NEXT_DUP is called repeatly. In case there is no next record, the program was hang. The following modified version of mtest3.c shows the issue. I am testing this on the latest commit 20dec1f69bf4860202c764ce92b1fbbe3d11a065 of lmdb on 20 Jan, on x86-64 Linux. I got a similar behaviour when a slightly earlier version of lmdb was cross-compiled with a Java wrapper for use on Android, which is why I am testing this. The issue should therefore not be platform specific. Thank you. ==== /* mtest3.c - memory-mapped database tester/toy */ /* * Copyrit t 2011-2015 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>. */ /* Tests for sorted duplicate DBs with cursor delete */ #include <stdio.h> #include <stdlib.h> #include <string.h> #include <time.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())) int main(int argc,char * argv[]) { int i = 0, j = 0, rc; MDB_env *env; MDB_dbi dbi; MDB_val key, data; MDB_txn *txn; MDB_stat mst; MDB_cursor *cursor; int count; int *values; char sval[32]; char kval[sizeof(int)]; srand(time(NULL)); memset(sval, 0, sizeof(sval)); count = 10; values = (int *)malloc(count*sizeof(int)); for(i =%0;i<count;i++) { values[i]= i * 10; } E(mdb_env_create(&env)); E(mdb_env_set_mapsize(env, 10485760)); E(mdb_env_set_maxdbs(env, 4)); E(mdb_env_open(env, "./testdb", MDB_FIXEDMAP|MDB_NOSYNC, 0664)); E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_dbi_open(txn, "id2", MDB_CREATE|MDB_DUPSORT, &dbi)); key.mv_size = sizeof(int); key.mv_data = kval; data.mv_size = sizeof(sval); data.mv_data = sval; printf("Adding %d values\n", count); for (i=0;i<count;i++) { if (!(i & 0x07)) 9%9 sprintf(kval, "%03x",alalues[i]); sprintf(sval, "%03x %d foo bar", values[i], values[i]); if (RES(MDB_KEYEXIST, mdb_put(txn, dbi, &key, &data, MDB_NODUPDATA))) j++; } if (j) printf("%d duplicates skipped\n", j); E(mdb_txn_commit(txn)); E(mdb_env_stat(env, &mst)); E(mdb_txn_begin(env, NULL, MDB_RDONLY, &txn)); E(mdb_cursor_open(txn, dbi, &cursor)); while ((rc = mdb_cursor_get(cursor, &key, &data, MDB_NEXT_DUP)) == 0) { printf("key: %p %.*s, data: %p %.*s\n", key.mv_data, %int%t) key.mv_size, (char *) key.mv_data, data.mv_data, (int) data.mv_size, (char *) data.mv_data); } CHECK(rc == MDB_NOTFOUND, "mdb_cursor_get"); mdb_cursor_close(cursor); mdb_txn_abort(txn); E(mdb_env_stat(env, &mst)); E(mdb_txn_begin(v,2C NULL, 0, &txn)); E(mdb_cursor_open(txn, dbi, &cursor)); /* * changing following to values[8] freezes program during mdb_cursor_del * as there is no more record following the deletion */ sprintf(kval, "%03x", values[0]); key.mv_size = sizeof(int); key.mv_data = kval; printf("\nCursor set / first dup\n"); E(mdb_cursor_get(cursor, &key, &data, MDB_SET_KEY)); E(mdb_cursor_get(cursor, &key, &data, MDB_FIRST_DUP)); printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data, (int) data.mv_size, (char *) data.mv_data); printf("Cursor next\n"); j=0; while ((rc = mdb_cursor_get(cursor, &key, &data, MDB_NEXT_DUP)) == 0) { printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) keyv_dadata, (int) data.mv_size, (char *) data.mv_data); j++; if (j == 1) { printf("delete the above key/data\n"); if (RES(MDB_NOTFOUND, mdb_cursor_del(cursor, 0))) {%%0 mdb_cursor_close(cursor); mdb_txn_abort(txn); break; } } if (j > count) { printf("Should not be there\n"); break; } } if (j > 1) { mdb_cursor_close(cursor); E(mdb_txn_commit(txn)); } E(mdb_txn_begin(env, NULL, 0, &txn)); E(mdb_cursor_open(txn, dbi, &cursor)); sprintf(kval, "%03x", values[0]); key.mv_size = sizeof(int); key.mv_data = kval; printf("\nCursor set / last dup\n"); E(mdb_cursor_get(cursor, &key, &data, MDB_SET_KEY)); E(mdb_cursor_get(cursor, &key, &data, MDB_LAST_DUP)); printf("key: %.*s, data: %.*s\n", (int) key.mv_size, (char *) key.mv_data,

1 0

(ITS#8357) [test] [lmdb] Issues with sorted duplicate DBs with cursor delete
by hyc＠openldap.org 22 Jan '16

22 Jan '16

1 0

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

openldap-bugs