From 13f0a771fd659bc3e0752eee48439b497cf27dae Mon Sep 17 00:00:00 2001 From: Paul Mackerras Date: Sat, 13 Nov 2004 12:02:22 +0000 Subject: [PATCH] New version of TDB code, borrowed from samba. --- pppd/Makefile.linux | 8 +- pppd/main.c | 43 +- pppd/multilink.c | 14 +- pppd/pathnames.h | 8 +- pppd/pppd.8 | 8 +- pppd/pppd.h | 4 +- pppd/tdb.c | 2539 +++++++++++++++++++++++++++---------------- pppd/tdb.h | 187 +++- 8 files changed, 1826 insertions(+), 985 deletions(-) diff --git a/pppd/Makefile.linux b/pppd/Makefile.linux index aaf5c31..e7e52b8 100644 --- a/pppd/Makefile.linux +++ b/pppd/Makefile.linux @@ -1,6 +1,6 @@ # # pppd makefile for Linux -# $Id: Makefile.linux,v 1.65 2004/11/01 03:49:20 paulus Exp $ +# $Id: Makefile.linux,v 1.66 2004/11/13 12:02:22 paulus Exp $ # # Default installation locations @@ -153,9 +153,9 @@ endif # TDB ifdef USE_TDB CFLAGS += -DUSE_TDB=1 - PPPDSRCS += tdb.c - PPPDOBJS += tdb.o - HEADERS += tdb.h + PPPDSRCS += tdb.c spinlock.c + PPPDOBJS += tdb.o spinlock.o + HEADERS += tdb.h spinlock.h endif # Lock library binary for Linux is included in 'linux' subdirectory. diff --git a/pppd/main.c b/pppd/main.c index ffeaca0..862a2d4 100644 --- a/pppd/main.c +++ b/pppd/main.c @@ -66,7 +66,7 @@ * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */ -#define RCSID "$Id: main.c,v 1.146 2004/11/12 11:21:41 paulus Exp $" +#define RCSID "$Id: main.c,v 1.147 2004/11/13 12:02:22 paulus Exp $" #include #include @@ -481,14 +481,8 @@ main(argc, argv) /* * Open the loopback channel and set it up to be the ppp interface. */ -#ifdef USE_TDB - tdb_writelock(pppdb); -#endif fd_loop = open_ppp_loopback(); set_ifunit(1); -#ifdef USE_TDB - tdb_writeunlock(pppdb); -#endif /* * Configure the interface and mark it up, etc. */ @@ -1923,6 +1917,41 @@ script_unsetenv(var) #endif } +/* + * Any arbitrary string used as a key for locking the database. + * It doesn't matter what it is as long as all pppds use the same string. + */ +#define PPPD_LOCK_KEY "pppd lock" + +/* + * lock_db - get an exclusive lock on the TDB database. + * Used to ensure atomicity of various lookup/modify operations. + */ +void lock_db() +{ +#ifdef USE_TDB + TDB_DATA key; + + key.dptr = PPPD_LOCK_KEY; + key.dsize = strlen(key.dptr); + tdb_chainlock(pppdb, key); +#endif +} + +/* + * unlock_db - remove the exclusive lock obtained by lock_db. + */ +void unlock_db() +{ +#ifdef USE_TDB + TDB_DATA key; + + key.dptr = PPPD_LOCK_KEY; + key.dsize = strlen(key.dptr); + tdb_chainunlock(pppdb, key); +#endif +} + #ifdef USE_TDB /* * update_db_entry - update our entry in the database. diff --git a/pppd/multilink.c b/pppd/multilink.c index 7444162..b4748cd 100644 --- a/pppd/multilink.c +++ b/pppd/multilink.c @@ -193,7 +193,7 @@ mp_join_bundle() * Check if the bundle ID is already in the database. */ unit = -1; - tdb_writelock(pppdb); + lock_db(); key.dptr = bundle_id; key.dsize = p - bundle_id; pid = tdb_fetch(pppdb, key); @@ -221,7 +221,7 @@ mp_join_bundle() set_ifunit(0); script_setenv("BUNDLE", bundle_id + 7, 0); make_bundle_links(1); - tdb_writeunlock(pppdb); + unlock_db(); info("Link attached to %s", ifname); return 1; } @@ -234,7 +234,7 @@ mp_join_bundle() netif_set_mtu(0, mtu); script_setenv("BUNDLE", bundle_id + 7, 1); make_bundle_links(0); - tdb_writeunlock(pppdb); + unlock_db(); info("New bundle %s created", ifname); multilink_master = 1; return 0; @@ -242,9 +242,9 @@ mp_join_bundle() void mp_exit_bundle() { - tdb_writelock(pppdb); + lock_db(); remove_bundle_link(); - tdb_writeunlock(pppdb); + unlock_db(); } static void sendhup(char *str) @@ -271,13 +271,13 @@ void mp_bundle_terminated() script_unsetenv("IFNAME"); } - tdb_writelock(pppdb); + lock_db(); destroy_bundle(); iterate_bundle_links(sendhup); key.dptr = blinks_id; key.dsize = strlen(blinks_id); tdb_delete(pppdb, key); - tdb_writeunlock(pppdb); + unlock_db(); new_phase(PHASE_DEAD); } diff --git a/pppd/pathnames.h b/pppd/pathnames.h index dccf0fd..555f40f 100644 --- a/pppd/pathnames.h +++ b/pppd/pathnames.h @@ -1,7 +1,7 @@ /* * define path names * - * $Id: pathnames.h,v 1.15 2002/11/02 19:48:13 carlsonj Exp $ + * $Id: pathnames.h,v 1.16 2004/11/13 12:02:22 paulus Exp $ */ #ifdef HAVE_PATHS_H @@ -45,12 +45,12 @@ #endif /* IPX_CHANGE */ #ifdef __STDC__ -#define _PATH_PPPDB _ROOT_PATH _PATH_VARRUN "pppd.tdb" +#define _PATH_PPPDB _ROOT_PATH _PATH_VARRUN "pppd2.tdb" #else /* __STDC__ */ #ifdef HAVE_PATHS_H -#define _PATH_PPPDB "/var/run/pppd.tdb" +#define _PATH_PPPDB "/var/run/pppd2.tdb" #else -#define _PATH_PPPDB "/etc/ppp/pppd.tdb" +#define _PATH_PPPDB "/etc/ppp/pppd2.tdb" #endif #endif /* __STDC__ */ diff --git a/pppd/pppd.8 b/pppd/pppd.8 index ca44a39..cf6ab59 100644 --- a/pppd/pppd.8 +++ b/pppd/pppd.8 @@ -1,5 +1,5 @@ .\" manual page [] for pppd 2.4 -.\" $Id: pppd.8,v 1.80 2004/11/08 11:50:00 paulus Exp $ +.\" $Id: pppd.8,v 1.81 2004/11/13 12:02:22 paulus Exp $ .\" SH section heading .\" SS subsection heading .\" LP paragraph @@ -1653,6 +1653,12 @@ Process-ID for pppd process on ppp interface unit \fIn\fR. Process-ID for pppd process for logical link \fIname\fR (see the \fIlinkname\fR option). .TP +.B /var/run/pppd2.tdb +Database containing information about pppd processes, interfaces and +links, used for matching links to bundles in multilink operation. May +be examined by external programs to obtain information about running +pppd instances, the interfaces and devices they are using, IP address +assignments, etc. .B /etc/ppp/pap-secrets Usernames, passwords and IP addresses for PAP authentication. This file should be owned by root and not readable or writable by any other diff --git a/pppd/pppd.h b/pppd/pppd.h index de3a6a1..4c18d3d 100644 --- a/pppd/pppd.h +++ b/pppd/pppd.h @@ -39,7 +39,7 @@ * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. * - * $Id: pppd.h,v 1.87 2004/11/12 10:30:51 paulus Exp $ + * $Id: pppd.h,v 1.88 2004/11/13 12:02:22 paulus Exp $ */ /* @@ -493,6 +493,8 @@ void notify __P((struct notifier *, int)); int ppp_send_config __P((int, int, u_int32_t, int, int)); int ppp_recv_config __P((int, int, u_int32_t, int, int)); void remove_pidfiles __P((void)); +void lock_db __P((void)); +void unlock_db __P((void)); /* Procedures exported from tty.c. */ void tty_init __P((void)); diff --git a/pppd/tdb.c b/pppd/tdb.c index 024b502..bdc5828 100644 --- a/pppd/tdb.c +++ b/pppd/tdb.c @@ -1,21 +1,50 @@ -/* - * Database functions - * Copyright (C) Andrew Tridgell 1999 - * - * Redistribution and use in source and binary forms are permitted - * provided that the above copyright notice and this paragraph are - * duplicated in all such forms AND provided that this software or - * any derived work is only used as part of the PPP daemon (pppd) - * and related utilities. - * The name of the author may not be used to endorse or promote products - * derived from this software without specific prior written permission. - * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR - * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED - * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. + /* + Unix SMB/CIFS implementation. + + trivial database library + + Copyright (C) Andrew Tridgell 1999-2004 + Copyright (C) Paul `Rusty' Russell 2000 + Copyright (C) Jeremy Allison 2000-2003 + + ** NOTE! The following LGPL license applies to the tdb + ** library. This does NOT imply that all of Samba is released + ** under the LGPL + + This library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2 of the License, or (at your option) any later version. + + This library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with this library; if not, write to the Free Software + Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +*/ + + +/* NOTE: If you use tdbs under valgrind, and in particular if you run + * tdbtorture, you may get spurious "uninitialized value" warnings. I + * think this is because valgrind doesn't understand that the mmap'd + * area may be written to by other processes. Memory can, from the + * point of view of the grinded process, spontaneously become + * initialized. * - * Note: this software is also available under the Gnu Public License - * version 2 or later. - */ + * I can think of a few solutions. [mbp 20030311] + * + * 1 - Write suppressions for Valgrind so that it doesn't complain + * about this. Probably the most reasonable but people need to + * remember to use them. + * + * 2 - Use IO not mmap when running under valgrind. Not so nice. + * + * 3 - Use the special valgrind macros to mark memory as valid at the + * right time. Probably too hard -- the process just doesn't know. + */ #include #include @@ -26,668 +55,1045 @@ #include #include #include +#include #include "tdb.h" +#include "spinlock.h" -#define TDB_VERSION (0x26011967 + 1) +#define TDB_MAGIC_FOOD "TDB file\n" +#define TDB_VERSION (0x26011967 + 6) #define TDB_MAGIC (0x26011999U) #define TDB_FREE_MAGIC (~TDB_MAGIC) -#define TDB_ALIGN 4 -#define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGN) -#define DEFAULT_HASH_SIZE 128 +#define TDB_DEAD_MAGIC (0xFEE1DEAD) +#define TDB_ALIGNMENT 4 +#define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGNMENT) +#define DEFAULT_HASH_SIZE 131 #define TDB_PAGE_SIZE 0x2000 -#define TDB_LEN_MULTIPLIER 10 #define FREELIST_TOP (sizeof(struct tdb_header)) +#define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1)) +#define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24)) +#define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC) +#define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r)) +#define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off)) +#define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1) + TDB_SPINLOCK_SIZE(hash_size)) + -#define LOCK_SET 1 -#define LOCK_CLEAR 0 +/* NB assumes there is a local variable called "tdb" that is the + * current context, also takes doubly-parenthesized print-style + * argument. */ +#define TDB_LOG(x) (tdb->log_fn?((tdb->log_fn x),0) : 0) /* lock offsets */ #define GLOBAL_LOCK 0 #define ACTIVE_LOCK 4 -#define LIST_LOCK_BASE 1024 - -#define BUCKET(hash) ((hash) % tdb->header.hash_size) #ifndef MAP_FILE #define MAP_FILE 0 #endif +#ifndef MAP_FAILED +#define MAP_FAILED ((void *)-1) +#endif + +/* free memory if the pointer is valid and zero the pointer */ +#ifndef SAFE_FREE +#define SAFE_FREE(x) do { if ((x) != NULL) {free((x)); (x)=NULL;} } while(0) +#endif + +#define BUCKET(hash) ((hash) % tdb->header.hash_size) +TDB_DATA tdb_null; + +/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */ +static TDB_CONTEXT *tdbs = NULL; + +static int tdb_munmap(TDB_CONTEXT *tdb) +{ + if (tdb->flags & TDB_INTERNAL) + return 0; + +#ifdef HAVE_MMAP + if (tdb->map_ptr) { + int ret = munmap(tdb->map_ptr, tdb->map_size); + if (ret != 0) + return ret; + } +#endif + tdb->map_ptr = NULL; + return 0; +} + +static void tdb_mmap(TDB_CONTEXT *tdb) +{ + if (tdb->flags & TDB_INTERNAL) + return; + +#ifdef HAVE_MMAP + if (!(tdb->flags & TDB_NOMMAP)) { + tdb->map_ptr = mmap(NULL, tdb->map_size, + PROT_READ|(tdb->read_only? 0:PROT_WRITE), + MAP_SHARED|MAP_FILE, tdb->fd, 0); + + /* + * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!! + */ + + if (tdb->map_ptr == MAP_FAILED) { + tdb->map_ptr = NULL; + TDB_LOG((tdb, 2, "tdb_mmap failed for size %d (%s)\n", + tdb->map_size, strerror(errno))); + } + } else { + tdb->map_ptr = NULL; + } +#else + tdb->map_ptr = NULL; +#endif +} + +/* Endian conversion: we only ever deal with 4 byte quantities */ +static void *convert(void *buf, u32 size) +{ + u32 i, *p = buf; + for (i = 0; i < size / 4; i++) + p[i] = TDB_BYTEREV(p[i]); + return buf; +} +#define DOCONV() (tdb->flags & TDB_CONVERT) +#define CONVERT(x) (DOCONV() ? convert(&x, sizeof(x)) : &x) + /* the body of the database is made of one list_struct for the free space plus a separate data list for each hash value */ struct list_struct { - tdb_len rec_len; /* total byte length of record */ tdb_off next; /* offset of the next record in the list */ + tdb_len rec_len; /* total byte length of record */ tdb_len key_len; /* byte length of key */ tdb_len data_len; /* byte length of data */ - unsigned full_hash; /* the full 32 bit hash of the key */ - unsigned magic; /* try to catch errors */ - /* - the following union is implied - union { - char record[rec_len]; - struct { - char key[key_len]; - char data[data_len]; - } - } + u32 full_hash; /* the full 32 bit hash of the key */ + u32 magic; /* try to catch errors */ + /* the following union is implied: + union { + char record[rec_len]; + struct { + char key[key_len]; + char data[data_len]; + } + u32 totalsize; (tailer) + } */ }; -/* a null data record - useful for error returns */ -static TDB_DATA null_data; +/*************************************************************** + Allow a caller to set a "alarm" flag that tdb can check to abort + a blocking lock on SIGALRM. +***************************************************************/ + +static sig_atomic_t *palarm_fired; + +void tdb_set_lock_alarm(sig_atomic_t *palarm) +{ + palarm_fired = palarm; +} /* a byte range locking function - return 0 on success - this functions locks/unlocks 1 byte at the specified offset */ + this functions locks/unlocks 1 byte at the specified offset. + + On error, errno is also set so that errors are passed back properly + through tdb_open(). */ static int tdb_brlock(TDB_CONTEXT *tdb, tdb_off offset, - int set, int rw_type, int lck_type) + int rw_type, int lck_type, int probe) { -#if NOLOCK - return 0; -#else struct flock fl; + int ret; - if (tdb->fd == -1) return 0; /* for in memory tdb */ - - if (tdb->read_only) return -1; + if (tdb->flags & TDB_NOLOCK) + return 0; + if ((rw_type == F_WRLCK) && (tdb->read_only)) { + errno = EACCES; + return -1; + } - fl.l_type = set==LOCK_SET?rw_type:F_UNLCK; + fl.l_type = rw_type; fl.l_whence = SEEK_SET; fl.l_start = offset; fl.l_len = 1; fl.l_pid = 0; - if (fcntl(tdb->fd, lck_type, &fl) != 0) { -#if TDB_DEBUG - if (lck_type == F_SETLKW) { - printf("lock %d failed at %d (%s)\n", - set, offset, strerror(errno)); + do { + ret = fcntl(tdb->fd,lck_type,&fl); + if (ret == -1 && errno == EINTR && palarm_fired && *palarm_fired) + break; + } while (ret == -1 && errno == EINTR); + + if (ret == -1) { + if (!probe && lck_type != F_SETLK) { + /* Ensure error code is set for log fun to examine. */ + if (errno == EINTR && palarm_fired && *palarm_fired) + tdb->ecode = TDB_ERR_LOCK_TIMEOUT; + else + tdb->ecode = TDB_ERR_LOCK; + TDB_LOG((tdb, 5,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d\n", + tdb->fd, offset, rw_type, lck_type)); } -#endif - tdb->ecode = TDB_ERR_LOCK; - return -1; + /* Was it an alarm timeout ? */ + if (errno == EINTR && palarm_fired && *palarm_fired) { + TDB_LOG((tdb, 5, "tdb_brlock timed out (fd=%d) at offset %d rw_type=%d lck_type=%d\n", + tdb->fd, offset, rw_type, lck_type)); + return TDB_ERRCODE(TDB_ERR_LOCK_TIMEOUT, -1); + } + /* Otherwise - generic lock error. errno set by fcntl. + * EAGAIN is an expected return from non-blocking + * locks. */ + if (errno != EAGAIN) { + TDB_LOG((tdb, 5, "tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d: %s\n", + tdb->fd, offset, rw_type, lck_type, + strerror(errno))); + } + return TDB_ERRCODE(TDB_ERR_LOCK, -1); } return 0; -#endif } /* lock a list in the database. list -1 is the alloc list */ -static int tdb_lock(TDB_CONTEXT *tdb, int list) +static int tdb_lock(TDB_CONTEXT *tdb, int list, int ltype) { if (list < -1 || list >= (int)tdb->header.hash_size) { -#if TDB_DEBUG - printf("bad list %d\n", list); -#endif + TDB_LOG((tdb, 0,"tdb_lock: invalid list %d for ltype=%d\n", + list, ltype)); return -1; } - if (tdb->locked[list+1] == 0) { - if (tdb_brlock(tdb, LIST_LOCK_BASE + 4*list, LOCK_SET, - F_WRLCK, F_SETLKW) != 0) { + if (tdb->flags & TDB_NOLOCK) + return 0; + + /* Since fcntl locks don't nest, we do a lock for the first one, + and simply bump the count for future ones */ + if (tdb->locked[list+1].count == 0) { + if (!tdb->read_only && tdb->header.rwlocks) { + if (tdb_spinlock(tdb, list, ltype)) { + TDB_LOG((tdb, 0, "tdb_lock spinlock failed on list %d ltype=%d\n", + list, ltype)); + return -1; + } + } else if (tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0)) { + TDB_LOG((tdb, 0,"tdb_lock failed on list %d ltype=%d (%s)\n", + list, ltype, strerror(errno))); return -1; } + tdb->locked[list+1].ltype = ltype; } - tdb->locked[list+1]++; + tdb->locked[list+1].count++; return 0; } -/* unlock the database. */ -static int tdb_unlock(TDB_CONTEXT *tdb, int list) +/* unlock the database: returns void because it's too late for errors. */ + /* changed to return int it may be interesting to know there + has been an error --simo */ +static int tdb_unlock(TDB_CONTEXT *tdb, int list, int ltype) { + int ret = -1; + + if (tdb->flags & TDB_NOLOCK) + return 0; + + /* Sanity checks */ if (list < -1 || list >= (int)tdb->header.hash_size) { -#if TDB_DEBUG - printf("bad unlock list %d\n", list); -#endif - return -1; + TDB_LOG((tdb, 0, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size)); + return ret; } - if (tdb->locked[list+1] == 0) { -#if TDB_DEBUG - printf("not locked %d\n", list); -#endif - tdb->ecode = TDB_ERR_LOCK; - return -1; - } - if (tdb->locked[list+1] == 1) { - if (tdb_brlock(tdb, LIST_LOCK_BASE + 4*list, LOCK_CLEAR, - F_WRLCK, F_SETLKW) != 0) { - return -1; - } + if (tdb->locked[list+1].count==0) { + TDB_LOG((tdb, 0, "tdb_unlock: count is 0\n")); + return ret; } - tdb->locked[list+1]--; - return 0; -} - -/* the hash algorithm - turn a key into an integer - This is based on the hash agorithm from gdbm */ -static unsigned tdb_hash(TDB_DATA *key) -{ - unsigned value; /* Used to compute the hash value. */ - unsigned i; /* Used to cycle through random values. */ - /* Set the initial value from the key size. */ - value = 0x238F13AF * key->dsize; - for (i=0; i < key->dsize; i++) { - value = (value + (key->dptr[i] << (i*5 % 24))); + if (tdb->locked[list+1].count == 1) { + /* Down to last nested lock: unlock underneath */ + if (!tdb->read_only && tdb->header.rwlocks) { + ret = tdb_spinunlock(tdb, list, ltype); + } else { + ret = tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0); + } + } else { + ret = 0; } + tdb->locked[list+1].count--; - value = (1103515243 * value + 12345); - - return value; -} - -/* find the top of the hash chain for an open database */ -static tdb_off tdb_hash_top(TDB_CONTEXT *tdb, unsigned hash) -{ - tdb_off ret; - hash = BUCKET(hash); - ret = FREELIST_TOP + (hash+1)*sizeof(tdb_off); + if (ret) + TDB_LOG((tdb, 0,"tdb_unlock: An error occurred unlocking!\n")); return ret; } - /* check for an out of bounds access - if it is out of bounds then see if the database has been expanded by someone else and expand - if necessary */ -static int tdb_oob(TDB_CONTEXT *tdb, tdb_off offset) + if necessary + note that "len" is the minimum length needed for the db +*/ +static int tdb_oob(TDB_CONTEXT *tdb, tdb_off len, int probe) { struct stat st; - if ((offset <= tdb->map_size) || (tdb->fd == -1)) return 0; - - fstat(tdb->fd, &st); - if (st.st_size <= (ssize_t)offset) { - tdb->ecode = TDB_ERR_IO; - return -1; + if (len <= tdb->map_size) + return 0; + if (tdb->flags & TDB_INTERNAL) { + if (!probe) { + /* Ensure ecode is set for log fn. */ + tdb->ecode = TDB_ERR_IO; + TDB_LOG((tdb, 0,"tdb_oob len %d beyond internal malloc size %d\n", + (int)len, (int)tdb->map_size)); + } + return TDB_ERRCODE(TDB_ERR_IO, -1); } -#if HAVE_MMAP - if (tdb->map_ptr) { - munmap(tdb->map_ptr, tdb->map_size); - tdb->map_ptr = NULL; + if (fstat(tdb->fd, &st) == -1) + return TDB_ERRCODE(TDB_ERR_IO, -1); + + if (st.st_size < (size_t)len) { + if (!probe) { + /* Ensure ecode is set for log fn. */ + tdb->ecode = TDB_ERR_IO; + TDB_LOG((tdb, 0,"tdb_oob len %d beyond eof at %d\n", + (int)len, (int)st.st_size)); + } + return TDB_ERRCODE(TDB_ERR_IO, -1); } -#endif + /* Unmap, update size, remap */ + if (tdb_munmap(tdb) == -1) + return TDB_ERRCODE(TDB_ERR_IO, -1); tdb->map_size = st.st_size; -#if HAVE_MMAP - tdb->map_ptr = (void *)mmap(NULL, tdb->map_size, - tdb->read_only?PROT_READ:PROT_READ|PROT_WRITE, - MAP_SHARED | MAP_FILE, tdb->fd, 0); - if (tdb->map_ptr == MAP_FAILED) { - tdb->map_ptr = NULL; - } -#endif + tdb_mmap(tdb); return 0; } - /* write a lump of data at a specified offset */ -static int tdb_write(TDB_CONTEXT *tdb, tdb_off offset, const char *buf, tdb_len len) +static int tdb_write(TDB_CONTEXT *tdb, tdb_off off, void *buf, tdb_len len) { - if (tdb_oob(tdb, offset + len) != 0) { - /* oops - trying to write beyond the end of the database! */ + if (tdb_oob(tdb, off + len, 0) != 0) return -1; - } - if (tdb->map_ptr) { - memcpy(offset + (char *)tdb->map_ptr, buf, len); - } else { - if (lseek(tdb->fd, offset, SEEK_SET) != offset || - write(tdb->fd, buf, len) != (ssize_t)len) { - tdb->ecode = TDB_ERR_IO; - return -1; - } + if (tdb->map_ptr) + memcpy(off + (char *)tdb->map_ptr, buf, len); +#ifdef HAVE_PWRITE + else if (pwrite(tdb->fd, buf, len, off) != (ssize_t)len) { +#else + else if (lseek(tdb->fd, off, SEEK_SET) != off + || write(tdb->fd, buf, len) != (ssize_t)len) { +#endif + /* Ensure ecode is set for log fn. */ + tdb->ecode = TDB_ERR_IO; + TDB_LOG((tdb, 0,"tdb_write failed at %d len=%d (%s)\n", + off, len, strerror(errno))); + return TDB_ERRCODE(TDB_ERR_IO, -1); } return 0; } -/* read a lump of data at a specified offset */ -static int tdb_read(TDB_CONTEXT *tdb, tdb_off offset, char *buf, tdb_len len) +/* read a lump of data at a specified offset, maybe convert */ +static int tdb_read(TDB_CONTEXT *tdb,tdb_off off,void *buf,tdb_len len,int cv) { - if (tdb_oob(tdb, offset + len) != 0) { - /* oops - trying to read beyond the end of the database! */ + if (tdb_oob(tdb, off + len, 0) != 0) return -1; - } - if (tdb->map_ptr) { - memcpy(buf, offset + (char *)tdb->map_ptr, len); - } else { - if (lseek(tdb->fd, offset, SEEK_SET) != offset || - read(tdb->fd, buf, len) != (ssize_t)len) { - tdb->ecode = TDB_ERR_IO; - return -1; - } + if (tdb->map_ptr) + memcpy(buf, off + (char *)tdb->map_ptr, len); +#ifdef HAVE_PREAD + else if (pread(tdb->fd, buf, len, off) != (ssize_t)len) { +#else + else if (lseek(tdb->fd, off, SEEK_SET) != off + || read(tdb->fd, buf, len) != (ssize_t)len) { +#endif + /* Ensure ecode is set for log fn. */ + tdb->ecode = TDB_ERR_IO; + TDB_LOG((tdb, 0,"tdb_read failed at %d len=%d (%s)\n", + off, len, strerror(errno))); + return TDB_ERRCODE(TDB_ERR_IO, -1); } + if (cv) + convert(buf, len); return 0; } - /* read a lump of data, allocating the space for it */ static char *tdb_alloc_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_len len) { char *buf; - buf = (char *)malloc(len); - - if (!buf) { + if (!(buf = malloc(len))) { + /* Ensure ecode is set for log fn. */ tdb->ecode = TDB_ERR_OOM; - return NULL; + TDB_LOG((tdb, 0,"tdb_alloc_read malloc failed len=%d (%s)\n", + len, strerror(errno))); + return TDB_ERRCODE(TDB_ERR_OOM, buf); } - - if (tdb_read(tdb, offset, buf, len) == -1) { - free(buf); + if (tdb_read(tdb, offset, buf, len, 0) == -1) { + SAFE_FREE(buf); return NULL; } - return buf; } -/* convenience routine for writing a record */ -static int rec_write(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec) +/* read/write a tdb_off */ +static int ofs_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d) { - return tdb_write(tdb, offset, (char *)rec, sizeof(*rec)); + return tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV()); } - -/* convenience routine for writing a tdb_off */ static int ofs_write(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d) { - return tdb_write(tdb, offset, (char *)d, sizeof(*d)); + tdb_off off = *d; + return tdb_write(tdb, offset, CONVERT(off), sizeof(*d)); } -/* read a tdb_off from the store */ -static int ofs_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d) +/* read/write a record */ +static int rec_read(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec) +{ + if (tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1) + return -1; + if (TDB_BAD_MAGIC(rec)) { + /* Ensure ecode is set for log fn. */ + tdb->ecode = TDB_ERR_CORRUPT; + TDB_LOG((tdb, 0,"rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset)); + return TDB_ERRCODE(TDB_ERR_CORRUPT, -1); + } + return tdb_oob(tdb, rec->next+sizeof(*rec), 0); +} +static int rec_write(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec) { - return tdb_read(tdb, offset, (char *)d, sizeof(*d)); + struct list_struct r = *rec; + return tdb_write(tdb, offset, CONVERT(r), sizeof(r)); } -/* read a record and check for simple errors */ -static int rec_read(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec) +/* read a freelist record and check for simple errors */ +static int rec_free_read(TDB_CONTEXT *tdb, tdb_off off, struct list_struct *rec) { - if (tdb_read(tdb, offset, (char *)rec, sizeof(*rec)) == -1) return -1; - if (rec->magic != TDB_MAGIC) { -#if TDB_DEBUG - printf("bad magic 0x%08x at offset %d\n", - rec->magic, offset); -#endif - tdb->ecode = TDB_ERR_CORRUPT; + if (tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1) return -1; + + if (rec->magic == TDB_MAGIC) { + /* this happens when a app is showdown while deleting a record - we should + not completely fail when this happens */ + TDB_LOG((tdb, 0,"rec_free_read non-free magic 0x%x at offset=%d - fixing\n", + rec->magic, off)); + rec->magic = TDB_FREE_MAGIC; + if (tdb_write(tdb, off, rec, sizeof(*rec)) == -1) + return -1; } - if (tdb_oob(tdb, rec->next) != 0) { - return -1; + + if (rec->magic != TDB_FREE_MAGIC) { + /* Ensure ecode is set for log fn. */ + tdb->ecode = TDB_ERR_CORRUPT; + TDB_LOG((tdb, 0,"rec_free_read bad magic 0x%x at offset=%d\n", + rec->magic, off)); + return TDB_ERRCODE(TDB_ERR_CORRUPT, -1); } + if (tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0) + return -1; return 0; } -/* expand the database at least length bytes by expanding the - underlying file and doing the mmap again if necessary */ -static int tdb_expand(TDB_CONTEXT *tdb, tdb_off length) +/* update a record tailer (must hold allocation lock) */ +static int update_tailer(TDB_CONTEXT *tdb, tdb_off offset, + const struct list_struct *rec) { - struct list_struct rec; - tdb_off offset, ptr; - char b = 0; - - tdb_lock(tdb,-1); + tdb_off totalsize; - /* make sure we know about any previous expansions by another - process */ - tdb_oob(tdb,tdb->map_size + 1); + /* Offset of tailer from record header */ + totalsize = sizeof(*rec) + rec->rec_len; + return ofs_write(tdb, offset + totalsize - sizeof(tdb_off), + &totalsize); +} - /* always make room for at least 10 more records */ - length *= TDB_LEN_MULTIPLIER; +static tdb_off tdb_dump_record(TDB_CONTEXT *tdb, tdb_off offset) +{ + struct list_struct rec; + tdb_off tailer_ofs, tailer; - /* and round the database up to a multiple of TDB_PAGE_SIZE */ - length = ((tdb->map_size + length + TDB_PAGE_SIZE) & ~(TDB_PAGE_SIZE - 1)) - tdb->map_size; + if (tdb_read(tdb, offset, (char *)&rec, sizeof(rec), DOCONV()) == -1) { + printf("ERROR: failed to read record at %u\n", offset); + return 0; + } - /* expand the file itself */ - if (tdb->fd != -1) { - lseek(tdb->fd, tdb->map_size + length - 1, SEEK_SET); - if (write(tdb->fd, &b, 1) != 1) goto fail; - } + printf(" rec: offset=%u next=%d rec_len=%d key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n", + offset, rec.next, rec.rec_len, rec.key_len, rec.data_len, rec.full_hash, rec.magic); - /* form a new freelist record */ - offset = FREELIST_TOP; - rec.rec_len = length - sizeof(rec); - rec.magic = TDB_FREE_MAGIC; - if (ofs_read(tdb, offset, &rec.next) == -1) { - goto fail; + tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off); + if (ofs_read(tdb, tailer_ofs, &tailer) == -1) { + printf("ERROR: failed to read tailer at %u\n", tailer_ofs); + return rec.next; } -#if HAVE_MMAP - if (tdb->fd != -1 && tdb->map_ptr) { - munmap(tdb->map_ptr, tdb->map_size); - tdb->map_ptr = NULL; + if (tailer != rec.rec_len + sizeof(rec)) { + printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n", + (unsigned)tailer, (unsigned)(rec.rec_len + sizeof(rec))); } -#endif + return rec.next; +} - tdb->map_size += length; +static int tdb_dump_chain(TDB_CONTEXT *tdb, int i) +{ + tdb_off rec_ptr, top; - if (tdb->fd == -1) { - tdb->map_ptr = realloc(tdb->map_ptr, tdb->map_size); - } + top = TDB_HASH_TOP(i); - /* write it out */ - if (rec_write(tdb, tdb->map_size - length, &rec) == -1) { - goto fail; - } + if (tdb_lock(tdb, i, F_WRLCK) != 0) + return -1; - /* link it into the free list */ - ptr = tdb->map_size - length; - if (ofs_write(tdb, offset, &ptr) == -1) goto fail; - -#if HAVE_MMAP - if (tdb->fd != -1) { - tdb->map_ptr = (void *)mmap(NULL, tdb->map_size, - PROT_READ|PROT_WRITE, - MAP_SHARED | MAP_FILE, tdb->fd, 0); - if (tdb->map_ptr == MAP_FAILED) { - tdb->map_ptr = NULL; - } - } -#endif + if (ofs_read(tdb, top, &rec_ptr) == -1) + return tdb_unlock(tdb, i, F_WRLCK); - tdb_unlock(tdb, -1); - return 0; + if (rec_ptr) + printf("hash=%d\n", i); - fail: - tdb_unlock(tdb,-1); - return -1; + while (rec_ptr) { + rec_ptr = tdb_dump_record(tdb, rec_ptr); + } + + return tdb_unlock(tdb, i, F_WRLCK); } -/* allocate some space from the free list. The offset returned points - to a unconnected list_struct within the database with room for at - least length bytes of total data +void tdb_dump_all(TDB_CONTEXT *tdb) +{ + int i; + for (i=0;iheader.hash_size;i++) { + tdb_dump_chain(tdb, i); + } + printf("freelist:\n"); + tdb_dump_chain(tdb, -1); +} - 0 is returned if the space could not be allocated - */ -static tdb_off tdb_allocate(TDB_CONTEXT *tdb, tdb_len length) +int tdb_printfreelist(TDB_CONTEXT *tdb) { - tdb_off offset, rec_ptr, last_ptr; - struct list_struct rec, lastrec, newrec; + int ret; + long total_free = 0; + tdb_off offset, rec_ptr; + struct list_struct rec; - tdb_lock(tdb, -1); + if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0) + return ret; - again: - last_ptr = 0; offset = FREELIST_TOP; /* read in the freelist top */ if (ofs_read(tdb, offset, &rec_ptr) == -1) { - goto fail; + tdb_unlock(tdb, -1, F_WRLCK); + return 0; } - /* keep looking until we find a freelist record that is big - enough */ + printf("freelist top=[0x%08x]\n", rec_ptr ); while (rec_ptr) { - if (tdb_read(tdb, rec_ptr, (char *)&rec, sizeof(rec)) == -1) { - goto fail; + if (tdb_read(tdb, rec_ptr, (char *)&rec, sizeof(rec), DOCONV()) == -1) { + tdb_unlock(tdb, -1, F_WRLCK); + return -1; } if (rec.magic != TDB_FREE_MAGIC) { -#if TDB_DEBUG printf("bad magic 0x%08x in free list\n", rec.magic); -#endif - goto fail; + tdb_unlock(tdb, -1, F_WRLCK); + return -1; } - if (rec.rec_len >= length) { - /* found it - now possibly split it up */ - if (rec.rec_len > length + MIN_REC_SIZE) { - length = (length + TDB_ALIGN) & ~(TDB_ALIGN-1); - - newrec.rec_len = rec.rec_len - (sizeof(rec) + length); - newrec.next = rec.next; - newrec.magic = TDB_FREE_MAGIC; - - rec.rec_len = length; - rec.next = rec_ptr + sizeof(rec) + rec.rec_len; - - if (rec_write(tdb, rec.next, &newrec) == -1) { - goto fail; - } - - if (rec_write(tdb, rec_ptr, &rec) == -1) { - goto fail; - } - } - - /* remove it from the list */ - if (last_ptr == 0) { - offset = FREELIST_TOP; - - if (ofs_write(tdb, offset, &rec.next) == -1) { - goto fail; - } - } else { - lastrec.next = rec.next; - if (rec_write(tdb, last_ptr, &lastrec) == -1) { - goto fail; - } - } - - /* all done - return the new record offset */ - tdb_unlock(tdb, -1); - return rec_ptr; - } + printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)]\n", rec.next, rec.rec_len, rec.rec_len ); + total_free += rec.rec_len; /* move to the next record */ - lastrec = rec; - last_ptr = rec_ptr; rec_ptr = rec.next; } + printf("total rec_len = [0x%08x (%d)]\n", (int)total_free, + (int)total_free); - /* we didn't find enough space. See if we can expand the - database and if we can then try again */ - if (tdb_expand(tdb, length + sizeof(rec)) == 0) goto again; - - fail: -#if TDB_DEBUG - printf("tdb_allocate failed for size %u\n", length); -#endif - tdb_unlock(tdb, -1); - return 0; + return tdb_unlock(tdb, -1, F_WRLCK); } -/* initialise a new database with a specified hash size */ -static int tdb_new_database(TDB_CONTEXT *tdb, int hash_size) +/* Remove an element from the freelist. Must have alloc lock. */ +static int remove_from_freelist(TDB_CONTEXT *tdb, tdb_off off, tdb_off next) { - struct tdb_header header; - tdb_off offset; - int i, size = 0; - tdb_off buf[16]; - - /* create the header */ - memset(&header, 0, sizeof(header)); - memcpy(header.magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1); - header.version = TDB_VERSION; - header.hash_size = hash_size; - lseek(tdb->fd, 0, SEEK_SET); - ftruncate(tdb->fd, 0); - - if (tdb->fd != -1 && write(tdb->fd, &header, sizeof(header)) != - sizeof(header)) { - tdb->ecode = TDB_ERR_IO; - return -1; - } else size += sizeof(header); - - /* the freelist and hash pointers */ - offset = 0; - memset(buf, 0, sizeof(buf)); - - for (i=0;(hash_size+1)-i >= 16; i += 16) { - if (tdb->fd != -1 && write(tdb->fd, buf, sizeof(buf)) != - sizeof(buf)) { - tdb->ecode = TDB_ERR_IO; - return -1; - } else size += sizeof(buf); - } - - for (;ifd != -1 && write(tdb->fd, buf, sizeof(tdb_off)) != - sizeof(tdb_off)) { - tdb->ecode = TDB_ERR_IO; - return -1; - } else size += sizeof(tdb_off); - } - - if (tdb->fd == -1) { - tdb->map_ptr = calloc(size, 1); - tdb->map_size = size; - if (tdb->map_ptr == NULL) { - tdb->ecode = TDB_ERR_IO; - return -1; - } - memcpy(&tdb->header, &header, sizeof(header)); - } - -#if TDB_DEBUG - printf("initialised database of hash_size %u\n", - hash_size); -#endif - return 0; + tdb_off last_ptr, i; + + /* read in the freelist top */ + last_ptr = FREELIST_TOP; + while (ofs_read(tdb, last_ptr, &i) != -1 && i != 0) { + if (i == off) { + /* We've found it! */ + return ofs_write(tdb, last_ptr, &next); + } + /* Follow chain (next offset is at start of record) */ + last_ptr = i; + } + TDB_LOG((tdb, 0,"remove_from_freelist: not on list at off=%d\n", off)); + return TDB_ERRCODE(TDB_ERR_CORRUPT, -1); } -/* Returns 0 on fail. On success, return offset of record, and fills - in rec */ -static tdb_off tdb_find(TDB_CONTEXT *tdb, TDB_DATA key, unsigned int hash, - struct list_struct *rec) +/* Add an element into the freelist. Merge adjacent records if + neccessary. */ +static int tdb_free(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec) { - tdb_off offset, rec_ptr; - - /* find the top of the hash chain */ - offset = tdb_hash_top(tdb, hash); + tdb_off right, left; - /* read in the hash top */ - if (ofs_read(tdb, offset, &rec_ptr) == -1) - return 0; + /* Allocation and tailer lock */ + if (tdb_lock(tdb, -1, F_WRLCK) != 0) + return -1; - /* keep looking until we find the right record */ - while (rec_ptr) { - if (rec_read(tdb, rec_ptr, rec) == -1) - return 0; + /* set an initial tailer, so if we fail we don't leave a bogus record */ + if (update_tailer(tdb, offset, rec) != 0) { + TDB_LOG((tdb, 0, "tdb_free: upfate_tailer failed!\n")); + goto fail; + } - if (hash == rec->full_hash && key.dsize == rec->key_len) { - char *k; - /* a very likely hit - read the key */ - k = tdb_alloc_read(tdb, rec_ptr + sizeof(*rec), - rec->key_len); + /* Look right first (I'm an Australian, dammit) */ + right = offset + sizeof(*rec) + rec->rec_len; + if (right + sizeof(*rec) <= tdb->map_size) { + struct list_struct r; - if (!k) - return 0; + if (tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) { + TDB_LOG((tdb, 0, "tdb_free: right read failed at %u\n", right)); + goto left; + } - if (memcmp(key.dptr, k, key.dsize) == 0) { - free(k); - return rec_ptr; + /* If it's free, expand to include it. */ + if (r.magic == TDB_FREE_MAGIC) { + if (remove_from_freelist(tdb, right, r.next) == -1) { + TDB_LOG((tdb, 0, "tdb_free: right free failed at %u\n", right)); + goto left; } - free(k); + rec->rec_len += sizeof(r) + r.rec_len; } - - /* move to the next record */ - rec_ptr = rec->next; } - return 0; -} -/* - return an error string for the last tdb error -*/ -char *tdb_error(TDB_CONTEXT *tdb) -{ - int i; - static struct { - enum TDB_ERROR ecode; - char *estring; - } emap[] = { - {TDB_SUCCESS, "Success"}, - {TDB_ERR_CORRUPT, "Corrupt database"}, - {TDB_ERR_IO, "IO Error"}, - {TDB_ERR_LOCK, "Locking error"}, - {TDB_ERR_OOM, "Out of memory"}, - {TDB_ERR_EXISTS, "Record exists"}, - {-1, NULL}}; - if (tdb != NULL) { - for (i=0;emap[i].estring;i++) { - if (tdb->ecode == emap[i].ecode) return emap[i].estring; - } - } else { - return "Invalid tdb context"; - } - return "Invalid error code"; +left: + /* Look left */ + left = offset - sizeof(tdb_off); + if (left > TDB_DATA_START(tdb->header.hash_size)) { + struct list_struct l; + tdb_off leftsize; + + /* Read in tailer and jump back to header */ + if (ofs_read(tdb, left, &leftsize) == -1) { + TDB_LOG((tdb, 0, "tdb_free: left offset read failed at %u\n", left)); + goto update; + } + left = offset - leftsize; + + /* Now read in record */ + if (tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) { + TDB_LOG((tdb, 0, "tdb_free: left read failed at %u (%u)\n", left, leftsize)); + goto update; + } + + /* If it's free, expand to include it. */ + if (l.magic == TDB_FREE_MAGIC) { + if (remove_from_freelist(tdb, left, l.next) == -1) { + TDB_LOG((tdb, 0, "tdb_free: left free failed at %u\n", left)); + goto update; + } else { + offset = left; + rec->rec_len += leftsize; + } + } + } + +update: + if (update_tailer(tdb, offset, rec) == -1) { + TDB_LOG((tdb, 0, "tdb_free: update_tailer failed at %u\n", offset)); + goto fail; + } + + /* Now, prepend to free list */ + rec->magic = TDB_FREE_MAGIC; + + if (ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 || + rec_write(tdb, offset, rec) == -1 || + ofs_write(tdb, FREELIST_TOP, &offset) == -1) { + TDB_LOG((tdb, 0, "tdb_free record write failed at offset=%d\n", offset)); + goto fail; + } + + /* And we're done. */ + tdb_unlock(tdb, -1, F_WRLCK); + return 0; + + fail: + tdb_unlock(tdb, -1, F_WRLCK); + return -1; } -/* update an entry in place - this only works if the new data size - is <= the old data size and the key exists. - on failure return -1 -*/ -int tdb_update(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf) +/* expand a file. we prefer to use ftruncate, as that is what posix + says to use for mmap expansion */ +static int expand_file(TDB_CONTEXT *tdb, tdb_off size, tdb_off addition) +{ + char buf[1024]; +#if HAVE_FTRUNCATE_EXTEND + if (ftruncate(tdb->fd, size+addition) != 0) { + TDB_LOG((tdb, 0, "expand_file ftruncate to %d failed (%s)\n", + size+addition, strerror(errno))); + return -1; + } +#else + char b = 0; + +#ifdef HAVE_PWRITE + if (pwrite(tdb->fd, &b, 1, (size+addition) - 1) != 1) { +#else + if (lseek(tdb->fd, (size+addition) - 1, SEEK_SET) != (size+addition) - 1 || + write(tdb->fd, &b, 1) != 1) { +#endif + TDB_LOG((tdb, 0, "expand_file to %d failed (%s)\n", + size+addition, strerror(errno))); + return -1; + } +#endif + + /* now fill the file with something. This ensures that the file isn't sparse, which would be + very bad if we ran out of disk. This must be done with write, not via mmap */ + memset(buf, 0x42, sizeof(buf)); + while (addition) { + int n = addition>sizeof(buf)?sizeof(buf):addition; +#ifdef HAVE_PWRITE + int ret = pwrite(tdb->fd, buf, n, size); +#else + int ret; + if (lseek(tdb->fd, size, SEEK_SET) != size) + return -1; + ret = write(tdb->fd, buf, n); +#endif + if (ret != n) { + TDB_LOG((tdb, 0, "expand_file write of %d failed (%s)\n", + n, strerror(errno))); + return -1; + } + addition -= n; + size += n; + } + return 0; +} + + +/* expand the database at least size bytes by expanding the underlying + file and doing the mmap again if necessary */ +static int tdb_expand(TDB_CONTEXT *tdb, tdb_off size) { - unsigned hash; struct list_struct rec; + tdb_off offset; + + if (tdb_lock(tdb, -1, F_WRLCK) == -1) { + TDB_LOG((tdb, 0, "lock failed in tdb_expand\n")); + return -1; + } + + /* must know about any previous expansions by another process */ + tdb_oob(tdb, tdb->map_size + 1, 1); + + /* always make room for at least 10 more records, and round + the database up to a multiple of TDB_PAGE_SIZE */ + size = TDB_ALIGN(tdb->map_size + size*10, TDB_PAGE_SIZE) - tdb->map_size; + + if (!(tdb->flags & TDB_INTERNAL)) + tdb_munmap(tdb); + + /* + * We must ensure the file is unmapped before doing this + * to ensure consistency with systems like OpenBSD where + * writes and mmaps are not consistent. + */ + + /* expand the file itself */ + if (!(tdb->flags & TDB_INTERNAL)) { + if (expand_file(tdb, tdb->map_size, size) != 0) + goto fail; + } + + tdb->map_size += size; + + if (tdb->flags & TDB_INTERNAL) + tdb->map_ptr = realloc(tdb->map_ptr, tdb->map_size); + else { + /* + * We must ensure the file is remapped before adding the space + * to ensure consistency with systems like OpenBSD where + * writes and mmaps are not consistent. + */ + + /* We're ok if the mmap fails as we'll fallback to read/write */ + tdb_mmap(tdb); + } + + /* form a new freelist record */ + memset(&rec,'\0',sizeof(rec)); + rec.rec_len = size - sizeof(rec); + + /* link it into the free list */ + offset = tdb->map_size - size; + if (tdb_free(tdb, offset, &rec) == -1) + goto fail; + + tdb_unlock(tdb, -1, F_WRLCK); + return 0; + fail: + tdb_unlock(tdb, -1, F_WRLCK); + return -1; +} + +/* allocate some space from the free list. The offset returned points + to a unconnected list_struct within the database with room for at + least length bytes of total data + + 0 is returned if the space could not be allocated + */ +static tdb_off tdb_allocate(TDB_CONTEXT *tdb, tdb_len length, + struct list_struct *rec) +{ + tdb_off rec_ptr, last_ptr, newrec_ptr; + struct list_struct newrec; + + memset(&newrec, '\0', sizeof(newrec)); + + if (tdb_lock(tdb, -1, F_WRLCK) == -1) + return 0; + + /* Extra bytes required for tailer */ + length += sizeof(tdb_off); + + again: + last_ptr = FREELIST_TOP; + + /* read in the freelist top */ + if (ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1) + goto fail; + + /* keep looking until we find a freelist record big enough */ + while (rec_ptr) { + if (rec_free_read(tdb, rec_ptr, rec) == -1) + goto fail; + + if (rec->rec_len >= length) { + /* found it - now possibly split it up */ + if (rec->rec_len > length + MIN_REC_SIZE) { + /* Length of left piece */ + length = TDB_ALIGN(length, TDB_ALIGNMENT); + + /* Right piece to go on free list */ + newrec.rec_len = rec->rec_len + - (sizeof(*rec) + length); + newrec_ptr = rec_ptr + sizeof(*rec) + length; + + /* And left record is shortened */ + rec->rec_len = length; + } else + newrec_ptr = 0; + + /* Remove allocated record from the free list */ + if (ofs_write(tdb, last_ptr, &rec->next) == -1) + goto fail; + + /* Update header: do this before we drop alloc + lock, otherwise tdb_free() might try to + merge with us, thinking we're free. + (Thanks Jeremy Allison). */ + rec->magic = TDB_MAGIC; + if (rec_write(tdb, rec_ptr, rec) == -1) + goto fail; + + /* Did we create new block? */ + if (newrec_ptr) { + /* Update allocated record tailer (we + shortened it). */ + if (update_tailer(tdb, rec_ptr, rec) == -1) + goto fail; + + /* Free new record */ + if (tdb_free(tdb, newrec_ptr, &newrec) == -1) + goto fail; + } + + /* all done - return the new record offset */ + tdb_unlock(tdb, -1, F_WRLCK); + return rec_ptr; + } + /* move to the next record */ + last_ptr = rec_ptr; + rec_ptr = rec->next; + } + /* we didn't find enough space. See if we can expand the + database and if we can then try again */ + if (tdb_expand(tdb, length + sizeof(*rec)) == 0) + goto again; + fail: + tdb_unlock(tdb, -1, F_WRLCK); + return 0; +} + +/* initialise a new database with a specified hash size */ +static int tdb_new_database(TDB_CONTEXT *tdb, int hash_size) +{ + struct tdb_header *newdb; + int size, ret = -1; + + /* We make it up in memory, then write it out if not internal */ + size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off); + if (!(newdb = calloc(size, 1))) + return TDB_ERRCODE(TDB_ERR_OOM, -1); + + /* Fill in the header */ + newdb->version = TDB_VERSION; + newdb->hash_size = hash_size; + if (tdb->flags & TDB_INTERNAL) { + tdb->map_size = size; + tdb->map_ptr = (char *)newdb; + memcpy(&tdb->header, newdb, sizeof(tdb->header)); + /* Convert the `ondisk' version if asked. */ + CONVERT(*newdb); + return 0; + } + if (lseek(tdb->fd, 0, SEEK_SET) == -1) + goto fail; + + if (ftruncate(tdb->fd, 0) == -1) + goto fail; + + /* This creates an endian-converted header, as if read from disk */ + CONVERT(*newdb); + memcpy(&tdb->header, newdb, sizeof(tdb->header)); + /* Don't endian-convert the magic food! */ + memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1); + if (write(tdb->fd, newdb, size) != size) + ret = -1; + else + ret = tdb_create_rwlocks(tdb->fd, hash_size); + + fail: + SAFE_FREE(newdb); + return ret; +} + +/* Returns 0 on fail. On success, return offset of record, and fills + in rec */ +static tdb_off tdb_find(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, + struct list_struct *r) +{ tdb_off rec_ptr; - int ret = -1; + + /* read in the hash top */ + if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) + return 0; - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_update() called with null context\n"); -#endif - return -1; - } + /* keep looking until we find the right record */ + while (rec_ptr) { + if (rec_read(tdb, rec_ptr, r) == -1) + return 0; - /* find which hash bucket it is in */ - hash = tdb_hash(&key); + if (!TDB_DEAD(r) && hash==r->full_hash && key.dsize==r->key_len) { + char *k; + /* a very likely hit - read the key */ + k = tdb_alloc_read(tdb, rec_ptr + sizeof(*r), + r->key_len); + if (!k) + return 0; - tdb_lock(tdb, BUCKET(hash)); - rec_ptr = tdb_find(tdb, key, hash, &rec); + if (memcmp(key.dptr, k, key.dsize) == 0) { + SAFE_FREE(k); + return rec_ptr; + } + SAFE_FREE(k); + } + rec_ptr = r->next; + } + return TDB_ERRCODE(TDB_ERR_NOEXIST, 0); +} - if (!rec_ptr) - goto out; +/* As tdb_find, but if you succeed, keep the lock */ +static tdb_off tdb_find_lock_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, int locktype, + struct list_struct *rec) +{ + u32 rec_ptr; - /* must be long enough */ - if (rec.rec_len < key.dsize + dbuf.dsize) - goto out; + if (tdb_lock(tdb, BUCKET(hash), locktype) == -1) + return 0; + if (!(rec_ptr = tdb_find(tdb, key, hash, rec))) + tdb_unlock(tdb, BUCKET(hash), locktype); + return rec_ptr; +} + +enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb) +{ + return tdb->ecode; +} + +static struct tdb_errname { + enum TDB_ERROR ecode; const char *estring; +} emap[] = { {TDB_SUCCESS, "Success"}, + {TDB_ERR_CORRUPT, "Corrupt database"}, + {TDB_ERR_IO, "IO Error"}, + {TDB_ERR_LOCK, "Locking error"}, + {TDB_ERR_OOM, "Out of memory"}, + {TDB_ERR_EXISTS, "Record exists"}, + {TDB_ERR_NOLOCK, "Lock exists on other keys"}, + {TDB_ERR_NOEXIST, "Record does not exist"} }; + +/* Error string for the last tdb error */ +const char *tdb_errorstr(TDB_CONTEXT *tdb) +{ + u32 i; + for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++) + if (tdb->ecode == emap[i].ecode) + return emap[i].estring; + return "Invalid error code"; +} + +/* update an entry in place - this only works if the new data size + is <= the old data size and the key exists. + on failure return -1. +*/ + +static int tdb_update_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, TDB_DATA dbuf) +{ + struct list_struct rec; + tdb_off rec_ptr; + + /* find entry */ + if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) + return -1; + + /* must be long enough key, data and tailer */ + if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off)) { + tdb->ecode = TDB_SUCCESS; /* Not really an error */ + return -1; + } if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len, dbuf.dptr, dbuf.dsize) == -1) - goto out; + return -1; if (dbuf.dsize != rec.data_len) { /* update size */ rec.data_len = dbuf.dsize; - ret = rec_write(tdb, rec_ptr, &rec); - } else - ret = 0; - - out: - tdb_unlock(tdb, BUCKET(hash)); - return ret; + return rec_write(tdb, rec_ptr, &rec); + } + + return 0; } /* find an entry in the database given a key */ +/* If an entry doesn't exist tdb_err will be set to + * TDB_ERR_NOEXIST. If a key has no data attached + * tdb_err will not be set. Both will return a + * zero pptr and zero dsize. + */ + TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key) { - unsigned hash; tdb_off rec_ptr; struct list_struct rec; - TDB_DATA ret = null_data; - - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_fetch() called with null context\n"); -#endif - return null_data; - } + TDB_DATA ret; + u32 hash; /* find which hash bucket it is in */ - hash = tdb_hash(&key); + hash = tdb->hash_fn(&key); + if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec))) + return tdb_null; - tdb_lock(tdb, BUCKET(hash)); - rec_ptr = tdb_find(tdb, key, hash, &rec); - - if (rec_ptr) { - ret.dptr = tdb_alloc_read(tdb, - rec_ptr + sizeof(rec) + rec.key_len, + if (rec.data_len) + ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len, rec.data_len); - ret.dsize = rec.data_len; - } - - tdb_unlock(tdb, BUCKET(hash)); + else + ret.dptr = NULL; + ret.dsize = rec.data_len; + tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK); return ret; } @@ -697,596 +1103,907 @@ TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key) this doesn't match the conventions in the rest of this module, but is compatible with gdbm */ -int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key) +static int tdb_exists_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash) { - unsigned hash; - tdb_off rec_ptr; struct list_struct rec; - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_exists() called with null context\n"); -#endif - return 0; - } + if (tdb_find_lock_hash(tdb, key, hash, F_RDLCK, &rec) == 0) + return 0; + tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK); + return 1; +} - /* find which hash bucket it is in */ - hash = tdb_hash(&key); +int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key) +{ + u32 hash = tdb->hash_fn(&key); + return tdb_exists_hash(tdb, key, hash); +} + +/* record lock stops delete underneath */ +static int lock_record(TDB_CONTEXT *tdb, tdb_off off) +{ + return off ? tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0) : 0; +} +/* + Write locks override our own fcntl readlocks, so check it here. + Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not + an error to fail to get the lock here. +*/ + +static int write_lock_record(TDB_CONTEXT *tdb, tdb_off off) +{ + struct tdb_traverse_lock *i; + for (i = &tdb->travlocks; i; i = i->next) + if (i->off == off) + return -1; + return tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1); +} + +/* + Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not + an error to fail to get the lock here. +*/ - tdb_lock(tdb, BUCKET(hash)); - rec_ptr = tdb_find(tdb, key, hash, &rec); - tdb_unlock(tdb, BUCKET(hash)); +static int write_unlock_record(TDB_CONTEXT *tdb, tdb_off off) +{ + return tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0); +} +/* fcntl locks don't stack: avoid unlocking someone else's */ +static int unlock_record(TDB_CONTEXT *tdb, tdb_off off) +{ + struct tdb_traverse_lock *i; + u32 count = 0; - return rec_ptr != 0; + if (off == 0) + return 0; + for (i = &tdb->travlocks; i; i = i->next) + if (i->off == off) + count++; + return (count == 1 ? tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0) : 0); } -/* traverse the entire database - calling fn(tdb, key, data) on each element. - return -1 on error or the record count traversed - if fn is NULL then it is not called - a non-zero return value from fn() indicates that the traversal should stop - */ -int tdb_traverse(TDB_CONTEXT *tdb, int (*fn)(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, void* state), void* state) +/* actually delete an entry in the database given the offset */ +static int do_delete(TDB_CONTEXT *tdb, tdb_off rec_ptr, struct list_struct*rec) { - int count = 0; - unsigned h; - tdb_off offset, rec_ptr; - struct list_struct rec; - char *data; - TDB_DATA key, dbuf; + tdb_off last_ptr, i; + struct list_struct lastrec; - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_traverse() called with null context\n"); -#endif - return -1; - } + if (tdb->read_only) return -1; - /* loop over all hash chains */ - for (h = 0; h < tdb->header.hash_size; h++) { - tdb_lock(tdb, BUCKET(h)); + if (write_lock_record(tdb, rec_ptr) == -1) { + /* Someone traversing here: mark it as dead */ + rec->magic = TDB_DEAD_MAGIC; + return rec_write(tdb, rec_ptr, rec); + } + if (write_unlock_record(tdb, rec_ptr) != 0) + return -1; - /* read in the hash top */ - offset = tdb_hash_top(tdb, h); - if (ofs_read(tdb, offset, &rec_ptr) == -1) { - goto fail; - } + /* find previous record in hash chain */ + if (ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1) + return -1; + for (last_ptr = 0; i != rec_ptr; last_ptr = i, i = lastrec.next) + if (rec_read(tdb, i, &lastrec) == -1) + return -1; - /* traverse all records for this hash */ - while (rec_ptr) { - if (rec_read(tdb, rec_ptr, &rec) == -1) { - goto fail; - } + /* unlink it: next ptr is at start of record. */ + if (last_ptr == 0) + last_ptr = TDB_HASH_TOP(rec->full_hash); + if (ofs_write(tdb, last_ptr, &rec->next) == -1) + return -1; - /* now read the full record */ - data = tdb_alloc_read(tdb, rec_ptr + sizeof(rec), - rec.key_len + rec.data_len); - if (!data) { - goto fail; - } + /* recover the space */ + if (tdb_free(tdb, rec_ptr, rec) == -1) + return -1; + return 0; +} - key.dptr = data; - key.dsize = rec.key_len; - dbuf.dptr = data + rec.key_len; - dbuf.dsize = rec.data_len; - count++; +/* Uses traverse lock: 0 = finish, -1 = error, other = record offset */ +static int tdb_next_lock(TDB_CONTEXT *tdb, struct tdb_traverse_lock *tlock, + struct list_struct *rec) +{ + int want_next = (tlock->off != 0); - if (fn && fn(tdb, key, dbuf, state) != 0) { - /* they want us to stop traversing */ - free(data); - tdb_unlock(tdb, BUCKET(h)); - return count; - } + /* Lock each chain from the start one. */ + for (; tlock->hash < tdb->header.hash_size; tlock->hash++) { + if (tdb_lock(tdb, tlock->hash, F_WRLCK) == -1) + return -1; - /* a miss - drat */ - free(data); + /* No previous record? Start at top of chain. */ + if (!tlock->off) { + if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash), + &tlock->off) == -1) + goto fail; + } else { + /* Otherwise unlock the previous record. */ + if (unlock_record(tdb, tlock->off) != 0) + goto fail; + } - /* move to the next record */ - rec_ptr = rec.next; + if (want_next) { + /* We have offset of old record: grab next */ + if (rec_read(tdb, tlock->off, rec) == -1) + goto fail; + tlock->off = rec->next; } - tdb_unlock(tdb, BUCKET(h)); - } - /* return the number traversed */ - return count; + /* Iterate through chain */ + while( tlock->off) { + tdb_off current; + if (rec_read(tdb, tlock->off, rec) == -1) + goto fail; + if (!TDB_DEAD(rec)) { + /* Woohoo: we found one! */ + if (lock_record(tdb, tlock->off) != 0) + goto fail; + return tlock->off; + } + /* Try to clean dead ones from old traverses */ + current = tlock->off; + tlock->off = rec->next; + if (!tdb->read_only && + do_delete(tdb, current, rec) != 0) + goto fail; + } + tdb_unlock(tdb, tlock->hash, F_WRLCK); + want_next = 0; + } + /* We finished iteration without finding anything */ + return TDB_ERRCODE(TDB_SUCCESS, 0); fail: - tdb_unlock(tdb, BUCKET(h)); + tlock->off = 0; + if (tdb_unlock(tdb, tlock->hash, F_WRLCK) != 0) + TDB_LOG((tdb, 0, "tdb_next_lock: On error unlock failed!\n")); return -1; } +/* traverse the entire database - calling fn(tdb, key, data) on each element. + return -1 on error or the record count traversed + if fn is NULL then it is not called + a non-zero return value from fn() indicates that the traversal should stop + */ +int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *private) +{ + TDB_DATA key, dbuf; + struct list_struct rec; + struct tdb_traverse_lock tl = { NULL, 0, 0 }; + int ret, count = 0; + + /* This was in the initializaton, above, but the IRIX compiler + * did not like it. crh + */ + tl.next = tdb->travlocks.next; + + /* fcntl locks don't stack: beware traverse inside traverse */ + tdb->travlocks.next = &tl; + + /* tdb_next_lock places locks on the record returned, and its chain */ + while ((ret = tdb_next_lock(tdb, &tl, &rec)) > 0) { + count++; + /* now read the full record */ + key.dptr = tdb_alloc_read(tdb, tl.off + sizeof(rec), + rec.key_len + rec.data_len); + if (!key.dptr) { + ret = -1; + if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0) + goto out; + if (unlock_record(tdb, tl.off) != 0) + TDB_LOG((tdb, 0, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n")); + goto out; + } + key.dsize = rec.key_len; + dbuf.dptr = key.dptr + rec.key_len; + dbuf.dsize = rec.data_len; + + /* Drop chain lock, call out */ + if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0) { + ret = -1; + goto out; + } + if (fn && fn(tdb, key, dbuf, private)) { + /* They want us to terminate traversal */ + ret = count; + if (unlock_record(tdb, tl.off) != 0) { + TDB_LOG((tdb, 0, "tdb_traverse: unlock_record failed!\n"));; + ret = -1; + } + tdb->travlocks.next = tl.next; + SAFE_FREE(key.dptr); + return count; + } + SAFE_FREE(key.dptr); + } +out: + tdb->travlocks.next = tl.next; + if (ret < 0) + return -1; + else + return count; +} /* find the first entry in the database and return its key */ TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb) { - tdb_off offset, rec_ptr; + TDB_DATA key; struct list_struct rec; - unsigned hash; - TDB_DATA ret; - - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_firstkey() called with null context\n"); -#endif - return null_data; - } - /* look for a non-empty hash chain */ - for (hash = 0, rec_ptr = 0; - hash < tdb->header.hash_size; - hash++) { - /* find the top of the hash chain */ - offset = tdb_hash_top(tdb, hash); - - tdb_lock(tdb, BUCKET(hash)); + /* release any old lock */ + if (unlock_record(tdb, tdb->travlocks.off) != 0) + return tdb_null; + tdb->travlocks.off = tdb->travlocks.hash = 0; + + if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0) + return tdb_null; + /* now read the key */ + key.dsize = rec.key_len; + key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize); + if (tdb_unlock(tdb, BUCKET(tdb->travlocks.hash), F_WRLCK) != 0) + TDB_LOG((tdb, 0, "tdb_firstkey: error occurred while tdb_unlocking!\n")); + return key; +} - /* read in the hash top */ - if (ofs_read(tdb, offset, &rec_ptr) == -1) { - goto fail; +/* find the next entry in the database, returning its key */ +TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA oldkey) +{ + u32 oldhash; + TDB_DATA key = tdb_null; + struct list_struct rec; + char *k = NULL; + + /* Is locked key the old key? If so, traverse will be reliable. */ + if (tdb->travlocks.off) { + if (tdb_lock(tdb,tdb->travlocks.hash,F_WRLCK)) + return tdb_null; + if (rec_read(tdb, tdb->travlocks.off, &rec) == -1 + || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec), + rec.key_len)) + || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) { + /* No, it wasn't: unlock it and start from scratch */ + if (unlock_record(tdb, tdb->travlocks.off) != 0) + return tdb_null; + if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0) + return tdb_null; + tdb->travlocks.off = 0; } - if (rec_ptr) break; - - tdb_unlock(tdb, BUCKET(hash)); + SAFE_FREE(k); } - if (rec_ptr == 0) return null_data; - - /* we've found a non-empty chain, now read the record */ - if (rec_read(tdb, rec_ptr, &rec) == -1) { - goto fail; + if (!tdb->travlocks.off) { + /* No previous element: do normal find, and lock record */ + tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), F_WRLCK, &rec); + if (!tdb->travlocks.off) + return tdb_null; + tdb->travlocks.hash = BUCKET(rec.full_hash); + if (lock_record(tdb, tdb->travlocks.off) != 0) { + TDB_LOG((tdb, 0, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno))); + return tdb_null; + } } + oldhash = tdb->travlocks.hash; + + /* Grab next record: locks chain and returned record, + unlocks old record */ + if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) { + key.dsize = rec.key_len; + key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec), + key.dsize); + /* Unlock the chain of this new record */ + if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0) + TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n")); + } + /* Unlock the chain of old record */ + if (tdb_unlock(tdb, BUCKET(oldhash), F_WRLCK) != 0) + TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n")); + return key; +} + +/* delete an entry in the database given a key */ +static int tdb_delete_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash) +{ + tdb_off rec_ptr; + struct list_struct rec; + int ret; - /* allocate and read the key space */ - ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec), rec.key_len); - ret.dsize = rec.key_len; - tdb_unlock(tdb, BUCKET(hash)); + if (!(rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_WRLCK, &rec))) + return -1; + ret = do_delete(tdb, rec_ptr, &rec); + if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0) + TDB_LOG((tdb, 0, "tdb_delete: WARNING tdb_unlock failed!\n")); return ret; +} - fail: - tdb_unlock(tdb, BUCKET(hash)); - return null_data; +int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key) +{ + u32 hash = tdb->hash_fn(&key); + return tdb_delete_hash(tdb, key, hash); } -/* find the next entry in the database, returning its key */ -TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA key) +/* store an element in the database, replacing any existing element + with the same key + + return 0 on success, -1 on failure +*/ +int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag) { - unsigned hash, hbucket; - tdb_off rec_ptr, offset; struct list_struct rec; - TDB_DATA ret; - - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_nextkey() called with null context\n"); -#endif - return null_data; - } + u32 hash; + tdb_off rec_ptr; + char *p = NULL; + int ret = 0; /* find which hash bucket it is in */ - hash = tdb_hash(&key); - hbucket = BUCKET(hash); - - tdb_lock(tdb, hbucket); - rec_ptr = tdb_find(tdb, key, hash, &rec); - if (rec_ptr) { - /* we want the next record after this one */ - rec_ptr = rec.next; - } - - /* not found or last in hash: look for next non-empty hash chain */ - while (rec_ptr == 0) { - tdb_unlock(tdb, hbucket); - - if (++hbucket >= tdb->header.hash_size - 1) - return null_data; + hash = tdb->hash_fn(&key); + if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1) + return -1; - offset = tdb_hash_top(tdb, hbucket); - tdb_lock(tdb, hbucket); - /* read in the hash top */ - if (ofs_read(tdb, offset, &rec_ptr) == -1) { - tdb_unlock(tdb, hbucket); - return null_data; + /* check for it existing, on insert. */ + if (flag == TDB_INSERT) { + if (tdb_exists_hash(tdb, key, hash)) { + tdb->ecode = TDB_ERR_EXISTS; + goto fail; } + } else { + /* first try in-place update, on modify or replace. */ + if (tdb_update_hash(tdb, key, hash, dbuf) == 0) + goto out; + if (tdb->ecode == TDB_ERR_NOEXIST && + flag == TDB_MODIFY) { + /* if the record doesn't exist and we are in TDB_MODIFY mode then + we should fail the store */ + goto fail; } - - /* Read the record. */ - if (rec_read(tdb, rec_ptr, &rec) == -1) { - tdb_unlock(tdb, hbucket); - return null_data; } - /* allocate and read the key */ - ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec), rec.key_len); - ret.dsize = rec.key_len; - tdb_unlock(tdb, hbucket); + /* reset the error code potentially set by the tdb_update() */ + tdb->ecode = TDB_SUCCESS; - return ret; -} + /* delete any existing record - if it doesn't exist we don't + care. Doing this first reduces fragmentation, and avoids + coalescing with `allocated' block before it's updated. */ + if (flag != TDB_INSERT) + tdb_delete_hash(tdb, key, hash); -/* delete an entry in the database given a key */ -int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key) -{ - unsigned hash; - tdb_off offset, rec_ptr, last_ptr; - struct list_struct rec, lastrec; - char *data = NULL; - - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_delete() called with null context\n"); -#endif - return -1; - } + /* Copy key+value *before* allocating free space in case malloc + fails and we are left with a dead spot in the tdb. */ - /* find which hash bucket it is in */ - hash = tdb_hash(&key); + if (!(p = (char *)malloc(key.dsize + dbuf.dsize))) { + tdb->ecode = TDB_ERR_OOM; + goto fail; + } - tdb_lock(tdb, BUCKET(hash)); + memcpy(p, key.dptr, key.dsize); + if (dbuf.dsize) + memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize); - /* find the top of the hash chain */ - offset = tdb_hash_top(tdb, hash); + /* we have to allocate some space */ + if (!(rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec))) + goto fail; - /* read in the hash top */ - if (ofs_read(tdb, offset, &rec_ptr) == -1) { + /* Read hash top into next ptr */ + if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1) goto fail; - } - last_ptr = 0; + rec.key_len = key.dsize; + rec.data_len = dbuf.dsize; + rec.full_hash = hash; + rec.magic = TDB_MAGIC; - /* keep looking until we find the right record */ - while (rec_ptr) { - if (rec_read(tdb, rec_ptr, &rec) == -1) { - goto fail; - } + /* write out and point the top of the hash chain at it */ + if (rec_write(tdb, rec_ptr, &rec) == -1 + || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1 + || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) { + /* Need to tdb_unallocate() here */ + goto fail; + } + out: + SAFE_FREE(p); + tdb_unlock(tdb, BUCKET(hash), F_WRLCK); + return ret; +fail: + ret = -1; + goto out; +} - if (hash == rec.full_hash && key.dsize == rec.key_len) { - /* a very likely hit - read the record and full key */ - data = tdb_alloc_read(tdb, rec_ptr + sizeof(rec), - rec.key_len); - if (!data) { - goto fail; - } +/* Attempt to append data to an entry in place - this only works if the new data size + is <= the old data size and the key exists. + on failure return -1. Record must be locked before calling. +*/ +static int tdb_append_inplace(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, TDB_DATA new_dbuf) +{ + struct list_struct rec; + tdb_off rec_ptr; - if (memcmp(key.dptr, data, key.dsize) == 0) { - /* a definite match - delete it */ - if (last_ptr == 0) { - offset = tdb_hash_top(tdb, hash); - if (ofs_write(tdb, offset, &rec.next) == -1) { - goto fail; - } - } else { - lastrec.next = rec.next; - if (rec_write(tdb, last_ptr, &lastrec) == -1) { - goto fail; - } - } - tdb_unlock(tdb, BUCKET(hash)); - tdb_lock(tdb, -1); - /* and recover the space */ - offset = FREELIST_TOP; - if (ofs_read(tdb, offset, &rec.next) == -1) { - goto fail2; - } - rec.magic = TDB_FREE_MAGIC; - if (rec_write(tdb, rec_ptr, &rec) == -1) { - goto fail2; - } - if (ofs_write(tdb, offset, &rec_ptr) == -1) { - goto fail2; - } - - /* yipee - all done */ - free(data); - tdb_unlock(tdb, -1); - return 0; - } + /* find entry */ + if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) + return -1; - /* a miss - drat */ - free(data); - data = NULL; - } + /* Append of 0 is always ok. */ + if (new_dbuf.dsize == 0) + return 0; - /* move to the next record */ - last_ptr = rec_ptr; - lastrec = rec; - rec_ptr = rec.next; + /* must be long enough for key, old data + new data and tailer */ + if (rec.rec_len < key.dsize + rec.data_len + new_dbuf.dsize + sizeof(tdb_off)) { + /* No room. */ + tdb->ecode = TDB_SUCCESS; /* Not really an error */ + return -1; } - fail: - if (data) free(data); - tdb_unlock(tdb, BUCKET(hash)); - return -1; + if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len + rec.data_len, + new_dbuf.dptr, new_dbuf.dsize) == -1) + return -1; - fail2: - if (data) free(data); - tdb_unlock(tdb, -1); - return -1; + /* update size */ + rec.data_len += new_dbuf.dsize; + return rec_write(tdb, rec_ptr, &rec); } +/* Append to an entry. Create if not exist. */ -/* store an element in the database, replacing any existing element - with the same key - - return 0 on success, -1 on failure -*/ -int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag) +int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf) { struct list_struct rec; - unsigned hash; - tdb_off rec_ptr, offset; + u32 hash; + tdb_off rec_ptr; char *p = NULL; - - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_store() called with null context\n"); -#endif - return -1; - } + int ret = 0; + size_t new_data_size = 0; /* find which hash bucket it is in */ - hash = tdb_hash(&key); - - /* check for it existing */ - if (flag == TDB_INSERT && tdb_exists(tdb, key)) { - tdb->ecode = TDB_ERR_EXISTS; + hash = tdb->hash_fn(&key); + if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1) return -1; - } - /* first try in-place update */ - if (flag != TDB_INSERT && tdb_update(tdb, key, dbuf) == 0) { - return 0; - } + /* first try in-place. */ + if (tdb_append_inplace(tdb, key, hash, new_dbuf) == 0) + goto out; - rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize); - if (rec_ptr == 0) { - return -1; - } + /* reset the error code potentially set by the tdb_append_inplace() */ + tdb->ecode = TDB_SUCCESS; + + /* find entry */ + if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) { + if (tdb->ecode != TDB_ERR_NOEXIST) + goto fail; - tdb_lock(tdb, BUCKET(hash)); + /* Not found - create. */ - /* delete any existing record - if it doesn't exist we don't care */ - if (flag != TDB_INSERT) { - tdb_delete(tdb, key); + ret = tdb_store(tdb, key, new_dbuf, TDB_INSERT); + goto out; } - /* read the newly created record */ - if (tdb_read(tdb, rec_ptr, (char *)&rec, sizeof(rec)) == -1) { + new_data_size = rec.data_len + new_dbuf.dsize; + + /* Copy key+old_value+value *before* allocating free space in case malloc + fails and we are left with a dead spot in the tdb. */ + + if (!(p = (char *)malloc(key.dsize + new_data_size))) { + tdb->ecode = TDB_ERR_OOM; goto fail; } - if (rec.magic != TDB_FREE_MAGIC) goto fail; + /* Copy the key in place. */ + memcpy(p, key.dptr, key.dsize); + + /* Now read the old data into place. */ + if (rec.data_len && + tdb_read(tdb, rec_ptr + sizeof(rec) + rec.key_len, p + key.dsize, rec.data_len, 0) == -1) + goto fail; + + /* Finally append the new data. */ + if (new_dbuf.dsize) + memcpy(p+key.dsize+rec.data_len, new_dbuf.dptr, new_dbuf.dsize); - /* find the top of the hash chain */ - offset = tdb_hash_top(tdb, hash); + /* delete any existing record - if it doesn't exist we don't + care. Doing this first reduces fragmentation, and avoids + coalescing with `allocated' block before it's updated. */ - /* read in the hash top diretcly into our next pointer */ - if (ofs_read(tdb, offset, &rec.next) == -1) { + tdb_delete_hash(tdb, key, hash); + + if (!(rec_ptr = tdb_allocate(tdb, key.dsize + new_data_size, &rec))) + goto fail; + + /* Read hash top into next ptr */ + if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1) goto fail; - } rec.key_len = key.dsize; - rec.data_len = dbuf.dsize; + rec.data_len = new_data_size; rec.full_hash = hash; rec.magic = TDB_MAGIC; - p = (char *)malloc(sizeof(rec) + key.dsize + dbuf.dsize); - if (!p) { - tdb->ecode = TDB_ERR_OOM; + /* write out and point the top of the hash chain at it */ + if (rec_write(tdb, rec_ptr, &rec) == -1 + || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+new_data_size)==-1 + || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) { + /* Need to tdb_unallocate() here */ goto fail; } - memcpy(p, &rec, sizeof(rec)); - memcpy(p+sizeof(rec), key.dptr, key.dsize); - memcpy(p+sizeof(rec)+key.dsize, dbuf.dptr, dbuf.dsize); - - if (tdb_write(tdb, rec_ptr, p, sizeof(rec)+key.dsize+dbuf.dsize) == -1) - goto fail; + out: + SAFE_FREE(p); + tdb_unlock(tdb, BUCKET(hash), F_WRLCK); + return ret; - free(p); - p = NULL; +fail: + ret = -1; + goto out; +} - /* and point the top of the hash chain at it */ - if (ofs_write(tdb, offset, &rec_ptr) == -1) goto fail; +static int tdb_already_open(dev_t device, + ino_t ino) +{ + TDB_CONTEXT *i; + + for (i = tdbs; i; i = i->next) { + if (i->device == device && i->inode == ino) { + return 1; + } + } - tdb_unlock(tdb, BUCKET(hash)); return 0; - - fail: -#if TDB_DEBUG - printf("store failed for hash 0x%08x in bucket %u\n", hash, BUCKET(hash)); -#endif - if (p) free(p); - tdb_unlock(tdb, BUCKET(hash)); - return -1; } +/* This is based on the hash algorithm from gdbm */ +static u32 default_tdb_hash(TDB_DATA *key) +{ + u32 value; /* Used to compute the hash value. */ + u32 i; /* Used to cycle through random values. */ + + /* Set the initial value from the key size. */ + for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++) + value = (value + (key->dptr[i] << (i*5 % 24))); + + return (1103515243 * value + 12345); +} /* open the database, creating it if necessary - The open_flags and mode are passed straight to the open call on the database - file. A flags value of O_WRONLY is invalid + The open_flags and mode are passed straight to the open call on the + database file. A flags value of O_WRONLY is invalid. The hash size + is advisory, use zero for a default value. - The hash size is advisory, use zero for a default value. + Return is NULL on error, in which case errno is also set. Don't + try to call tdb_error or tdb_errname, just do strerror(errno). - return is NULL on error -*/ -TDB_CONTEXT *tdb_open(char *name, int hash_size, int tdb_flags, + @param name may be NULL for internal databases. */ +TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags, int open_flags, mode_t mode) { - TDB_CONTEXT tdb, *ret; - struct stat st; + return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL); +} - memset(&tdb, 0, sizeof(tdb)); - tdb.fd = -1; - tdb.name = NULL; - tdb.map_ptr = NULL; +TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags, + int open_flags, mode_t mode, + tdb_log_func log_fn, + tdb_hash_func hash_fn) +{ + TDB_CONTEXT *tdb; + struct stat st; + int rev = 0, locked = 0; + unsigned char *vp; + u32 vertest; - if ((open_flags & O_ACCMODE) == O_WRONLY) { + if (!(tdb = calloc(1, sizeof *tdb))) { + /* Can't log this */ + errno = ENOMEM; goto fail; } + tdb->fd = -1; + tdb->name = NULL; + tdb->map_ptr = NULL; + tdb->flags = tdb_flags; + tdb->open_flags = open_flags; + tdb->log_fn = log_fn; + tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash; - if (hash_size == 0) hash_size = DEFAULT_HASH_SIZE; + if ((open_flags & O_ACCMODE) == O_WRONLY) { + TDB_LOG((tdb, 0, "tdb_open_ex: can't open tdb %s write-only\n", + name)); + errno = EINVAL; + goto fail; + } + + if (hash_size == 0) + hash_size = DEFAULT_HASH_SIZE; + if ((open_flags & O_ACCMODE) == O_RDONLY) { + tdb->read_only = 1; + /* read only databases don't do locking or clear if first */ + tdb->flags |= TDB_NOLOCK; + tdb->flags &= ~TDB_CLEAR_IF_FIRST; + } - tdb.read_only = ((open_flags & O_ACCMODE) == O_RDONLY); + /* internal databases don't mmap or lock, and start off cleared */ + if (tdb->flags & TDB_INTERNAL) { + tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP); + tdb->flags &= ~TDB_CLEAR_IF_FIRST; + if (tdb_new_database(tdb, hash_size) != 0) { + TDB_LOG((tdb, 0, "tdb_open_ex: tdb_new_database failed!")); + goto fail; + } + goto internal; + } - if (name != NULL) { - tdb.fd = open(name, open_flags, mode); - if (tdb.fd == -1) { - goto fail; - } - (void) fcntl(tdb.fd, F_SETFD, FD_CLOEXEC); - } + if ((tdb->fd = open(name, open_flags, mode)) == -1) { + TDB_LOG((tdb, 5, "tdb_open_ex: could not open file %s: %s\n", + name, strerror(errno))); + goto fail; /* errno set by open(2) */ + } /* ensure there is only one process initialising at once */ - tdb_brlock(&tdb, GLOBAL_LOCK, LOCK_SET, F_WRLCK, F_SETLKW); - - if (tdb_flags & TDB_CLEAR_IF_FIRST) { - /* we need to zero the database if we are the only - one with it open */ - if (tdb_brlock(&tdb, ACTIVE_LOCK, LOCK_SET, F_WRLCK, F_SETLK) == 0) { - ftruncate(tdb.fd, 0); - tdb_brlock(&tdb, ACTIVE_LOCK, LOCK_CLEAR, F_WRLCK, F_SETLK); - } + if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1) { + TDB_LOG((tdb, 0, "tdb_open_ex: failed to get global lock on %s: %s\n", + name, strerror(errno))); + goto fail; /* errno set by tdb_brlock */ } - /* leave this lock in place */ - tdb_brlock(&tdb, ACTIVE_LOCK, LOCK_SET, F_RDLCK, F_SETLKW); + /* we need to zero database if we are the only one with it open */ + if ((tdb_flags & TDB_CLEAR_IF_FIRST) && + (locked = (tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 0))) { + open_flags |= O_CREAT; + if (ftruncate(tdb->fd, 0) == -1) { + TDB_LOG((tdb, 0, "tdb_open_ex: " + "failed to truncate %s: %s\n", + name, strerror(errno))); + goto fail; /* errno set by ftruncate */ + } + } - if (read(tdb.fd, &tdb.header, sizeof(tdb.header)) != sizeof(tdb.header) || - strcmp(tdb.header.magic_food, TDB_MAGIC_FOOD) != 0 || - tdb.header.version != TDB_VERSION) { + if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header) + || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0 + || (tdb->header.version != TDB_VERSION + && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) { /* its not a valid database - possibly initialise it */ - if (!(open_flags & O_CREAT)) { + if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) { + errno = EIO; /* ie bad format or something */ goto fail; } - if (tdb_new_database(&tdb, hash_size) == -1) goto fail; - - lseek(tdb.fd, 0, SEEK_SET); - if (tdb.fd != -1 && read(tdb.fd, &tdb.header, - sizeof(tdb.header)) != - sizeof(tdb.header)) - goto fail; - } - - if (tdb.fd != -1) { - fstat(tdb.fd, &st); - - /* map the database and fill in the return structure */ - tdb.name = (char *)strdup(name); - tdb.map_size = st.st_size; - } - - tdb.locked = (int *)calloc(tdb.header.hash_size+1, - sizeof(tdb.locked[0])); - if (!tdb.locked) { - goto fail; - } - -#if HAVE_MMAP - if (tdb.fd != -1) { - tdb.map_ptr = (void *)mmap(NULL, st.st_size, - tdb.read_only? PROT_READ : PROT_READ|PROT_WRITE, - MAP_SHARED | MAP_FILE, tdb.fd, 0); - if (tdb.map_ptr == MAP_FAILED) { - tdb.map_ptr = NULL; - } - } -#endif + rev = (tdb->flags & TDB_CONVERT); + } + vp = (unsigned char *)&tdb->header.version; + vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) | + (((u32)vp[2]) << 8) | (u32)vp[3]; + tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0; + if (!rev) + tdb->flags &= ~TDB_CONVERT; + else { + tdb->flags |= TDB_CONVERT; + convert(&tdb->header, sizeof(tdb->header)); + } + if (fstat(tdb->fd, &st) == -1) + goto fail; - ret = (TDB_CONTEXT *)malloc(sizeof(tdb)); - if (!ret) goto fail; + /* Is it already in the open list? If so, fail. */ + if (tdb_already_open(st.st_dev, st.st_ino)) { + TDB_LOG((tdb, 2, "tdb_open_ex: " + "%s (%d,%d) is already open in this process\n", + name, (int)st.st_dev, (int)st.st_ino)); + errno = EBUSY; + goto fail; + } - *ret = tdb; + if (!(tdb->name = (char *)strdup(name))) { + errno = ENOMEM; + goto fail; + } -#if TDB_DEBUG - printf("mapped database of hash_size %u map_size=%u\n", - hash_size, tdb.map_size); -#endif + tdb->map_size = st.st_size; + tdb->device = st.st_dev; + tdb->inode = st.st_ino; + tdb->locked = calloc(tdb->header.hash_size+1, sizeof(tdb->locked[0])); + if (!tdb->locked) { + TDB_LOG((tdb, 2, "tdb_open_ex: " + "failed to allocate lock structure for %s\n", + name)); + errno = ENOMEM; + goto fail; + } + tdb_mmap(tdb); + if (locked) { + if (!tdb->read_only) + if (tdb_clear_spinlocks(tdb) != 0) { + TDB_LOG((tdb, 0, "tdb_open_ex: " + "failed to clear spinlock\n")); + goto fail; + } + if (tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1) { + TDB_LOG((tdb, 0, "tdb_open_ex: " + "failed to take ACTIVE_LOCK on %s: %s\n", + name, strerror(errno))); + goto fail; + } - tdb_brlock(&tdb, GLOBAL_LOCK, LOCK_CLEAR, F_WRLCK, F_SETLKW); - return ret; + } + + /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if + we didn't get the initial exclusive lock as we need to let all other + users know we're using it. */ + + if (tdb_flags & TDB_CLEAR_IF_FIRST) { + /* leave this lock in place to indicate it's in use */ + if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1) + goto fail; + } + + + internal: + /* Internal (memory-only) databases skip all the code above to + * do with disk files, and resume here by releasing their + * global lock and hooking into the active list. */ + if (tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1) + goto fail; + tdb->next = tdbs; + tdbs = tdb; + return tdb; fail: - if (tdb.name) free(tdb.name); - if (tdb.fd != -1) close(tdb.fd); - if (tdb.map_ptr) munmap(tdb.map_ptr, tdb.map_size); + { int save_errno = errno; + if (!tdb) + return NULL; + + if (tdb->map_ptr) { + if (tdb->flags & TDB_INTERNAL) + SAFE_FREE(tdb->map_ptr); + else + tdb_munmap(tdb); + } + SAFE_FREE(tdb->name); + if (tdb->fd != -1) + if (close(tdb->fd) != 0) + TDB_LOG((tdb, 5, "tdb_open_ex: failed to close tdb->fd on error!\n")); + SAFE_FREE(tdb->locked); + SAFE_FREE(tdb); + errno = save_errno; return NULL; + } } -/* close a database */ +/** + * Close a database. + * + * @returns -1 for error; 0 for success. + **/ int tdb_close(TDB_CONTEXT *tdb) { - if (!tdb) return -1; - - if (tdb->name) free(tdb->name); - if (tdb->fd != -1) close(tdb->fd); - if (tdb->locked) free(tdb->locked); + TDB_CONTEXT **i; + int ret = 0; if (tdb->map_ptr) { - if (tdb->fd != -1) { - munmap(tdb->map_ptr, tdb->map_size); - } else { - free(tdb->map_ptr); - } - } + if (tdb->flags & TDB_INTERNAL) + SAFE_FREE(tdb->map_ptr); + else + tdb_munmap(tdb); + } + SAFE_FREE(tdb->name); + if (tdb->fd != -1) + ret = close(tdb->fd); + SAFE_FREE(tdb->locked); + + /* Remove from contexts list */ + for (i = &tdbs; *i; i = &(*i)->next) { + if (*i == tdb) { + *i = tdb->next; + break; + } + } memset(tdb, 0, sizeof(*tdb)); - free(tdb); + SAFE_FREE(tdb); - return 0; + return ret; } -/* lock the database. If we already have it locked then don't do anything */ -int tdb_writelock(TDB_CONTEXT *tdb) +/* lock/unlock entire database */ +int tdb_lockall(TDB_CONTEXT *tdb) { - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_writelock() called with null context\n"); -#endif - return -1; - } + u32 i; + + /* There are no locks on read-only dbs */ + if (tdb->read_only) + return TDB_ERRCODE(TDB_ERR_LOCK, -1); + for (i = 0; i < tdb->header.hash_size; i++) + if (tdb_lock(tdb, i, F_WRLCK)) + break; + + /* If error, release locks we have... */ + if (i < tdb->header.hash_size) { + u32 j; + + for ( j = 0; j < i; j++) + tdb_unlock(tdb, j, F_WRLCK); + return TDB_ERRCODE(TDB_ERR_NOLOCK, -1); + } - return tdb_lock(tdb, -1); + return 0; +} +void tdb_unlockall(TDB_CONTEXT *tdb) +{ + u32 i; + for (i=0; i < tdb->header.hash_size; i++) + tdb_unlock(tdb, i, F_WRLCK); } -/* unlock the database. */ -int tdb_writeunlock(TDB_CONTEXT *tdb) +/* lock/unlock one hash chain. This is meant to be used to reduce + contention - it cannot guarantee how many records will be locked */ +int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key) { - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_writeunlock() called with null context\n"); -#endif - return -1; - } + return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK); +} - return tdb_unlock(tdb, -1); +int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key) +{ + return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK); } -/* lock one hash chain. This is meant to be used to reduce locking - contention - it cannot guarantee how many records will be locked */ -int tdb_lockchain(TDB_CONTEXT *tdb, TDB_DATA key) +int tdb_chainlock_read(TDB_CONTEXT *tdb, TDB_DATA key) { - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_lockchain() called with null context\n"); -#endif - return -1; - } + return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK); +} - return tdb_lock(tdb, BUCKET(tdb_hash(&key))); +int tdb_chainunlock_read(TDB_CONTEXT *tdb, TDB_DATA key) +{ + return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK); } -/* unlock one hash chain */ -int tdb_unlockchain(TDB_CONTEXT *tdb, TDB_DATA key) +/* register a loging function */ +void tdb_logging_function(TDB_CONTEXT *tdb, void (*fn)(TDB_CONTEXT *, int , const char *, ...)) { - if (tdb == NULL) { -#ifdef TDB_DEBUG - printf("tdb_unlockchain() called with null context\n"); -#endif - return -1; - } + tdb->log_fn = fn; +} + +/* reopen a tdb - this can be used after a fork to ensure that we have an independent + seek pointer from our parent and to re-establish locks */ +int tdb_reopen(TDB_CONTEXT *tdb) +{ + struct stat st; + + if (tdb->flags & TDB_INTERNAL) + return 0; /* Nothing to do. */ + if (tdb_munmap(tdb) != 0) { + TDB_LOG((tdb, 0, "tdb_reopen: munmap failed (%s)\n", strerror(errno))); + goto fail; + } + if (close(tdb->fd) != 0) + TDB_LOG((tdb, 0, "tdb_reopen: WARNING closing tdb->fd failed!\n")); + tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0); + if (tdb->fd == -1) { + TDB_LOG((tdb, 0, "tdb_reopen: open failed (%s)\n", strerror(errno))); + goto fail; + } + if (fstat(tdb->fd, &st) != 0) { + TDB_LOG((tdb, 0, "tdb_reopen: fstat failed (%s)\n", strerror(errno))); + goto fail; + } + if (st.st_ino != tdb->inode || st.st_dev != tdb->device) { + TDB_LOG((tdb, 0, "tdb_reopen: file dev/inode has changed!\n")); + goto fail; + } + tdb_mmap(tdb); + if ((tdb->flags & TDB_CLEAR_IF_FIRST) && (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)) { + TDB_LOG((tdb, 0, "tdb_reopen: failed to obtain active lock\n")); + goto fail; + } + + return 0; + +fail: + tdb_close(tdb); + return -1; +} + +/* reopen all tdb's */ +int tdb_reopen_all(void) +{ + TDB_CONTEXT *tdb; - return tdb_unlock(tdb, BUCKET(tdb_hash(&key))); + for (tdb=tdbs; tdb; tdb = tdb->next) { + /* Ensure no clear-if-first. */ + tdb->flags &= ~TDB_CLEAR_IF_FIRST; + if (tdb_reopen(tdb) != 0) + return -1; + } + + return 0; } diff --git a/pppd/tdb.h b/pppd/tdb.h index 56ae0ac..153b6e9 100644 --- a/pppd/tdb.h +++ b/pppd/tdb.h @@ -1,77 +1,164 @@ -#define STANDALONE 1 +#ifndef __TDB_H__ +#define __TDB_H__ + /* - * Database functions - * Copyright (C) Andrew Tridgell 1999 - * - * Redistribution and use in source and binary forms are permitted - * provided that the above copyright notice and this paragraph are - * duplicated in all such forms AND provided that this software or - * any derived work is only used as part of the PPP daemon (pppd) - * and related utilities. - * The name of the author may not be used to endorse or promote products - * derived from this software without specific prior written permission. - * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR - * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED - * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. - * - * Note: this software is also available under the Gnu Public License - * version 2 or later. - */ - -typedef unsigned tdb_len; -typedef unsigned tdb_off; - -#define TDB_MAGIC_FOOD "TDB file\n" + Unix SMB/CIFS implementation. + + trivial database library + + Copyright (C) Andrew Tridgell 1999-2004 + + ** NOTE! The following LGPL license applies to the tdb + ** library. This does NOT imply that all of Samba is released + ** under the LGPL + + This library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2 of the License, or (at your option) any later version. -/* this is stored at the front of every database */ -struct tdb_header { - char magic_food[32]; /* for /etc/magic */ - unsigned version; /* version of the code */ - unsigned hash_size; /* number of hash entries */ -}; + This library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with this library; if not, write to the Free Software + Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA +*/ + +#ifdef __cplusplus +extern "C" { +#endif + +#ifndef PRINTF_ATTRIBUTE +/** Use gcc attribute to check printf fns. a1 is the 1-based index of + * the parameter containing the format, and a2 the index of the first + * argument. Note that some gcc 2.x versions don't handle this + * properly **/ +#if (__GNUC__ >= 3) +#define PRINTF_ATTRIBUTE(a1, a2) __attribute__ ((format (__printf__, a1, a2))) +#else +#define PRINTF_ATTRIBUTE(a1, a2) +#endif +#endif + +/* flags to tdb_store() */ +#define TDB_REPLACE 1 +#define TDB_INSERT 2 +#define TDB_MODIFY 3 + +/* flags for tdb_open() */ +#define TDB_DEFAULT 0 /* just a readability place holder */ +#define TDB_CLEAR_IF_FIRST 1 +#define TDB_INTERNAL 2 /* don't store on disk */ +#define TDB_NOLOCK 4 /* don't do any locking */ +#define TDB_NOMMAP 8 /* don't use mmap */ +#define TDB_CONVERT 16 /* convert endian (internal use) */ +#define TDB_BIGENDIAN 32 /* header is big-endian (internal use) */ + +#define TDB_ERRCODE(code, ret) ((tdb->ecode = (code)), ret) + +/* error codes */ +enum TDB_ERROR {TDB_SUCCESS=0, TDB_ERR_CORRUPT, TDB_ERR_IO, TDB_ERR_LOCK, + TDB_ERR_OOM, TDB_ERR_EXISTS, TDB_ERR_NOLOCK, TDB_ERR_LOCK_TIMEOUT, + TDB_ERR_NOEXIST}; + +#ifndef u32 +#define u32 unsigned +#endif typedef struct { char *dptr; size_t dsize; } TDB_DATA; +typedef u32 tdb_len; +typedef u32 tdb_off; + +/* this is stored at the front of every database */ +struct tdb_header { + char magic_food[32]; /* for /etc/magic */ + u32 version; /* version of the code */ + u32 hash_size; /* number of hash entries */ + tdb_off rwlocks; + tdb_off reserved[31]; +}; + +struct tdb_lock_type { + u32 count; + u32 ltype; +}; + +struct tdb_traverse_lock { + struct tdb_traverse_lock *next; + u32 off; + u32 hash; +}; + /* this is the context structure that is returned from a db open */ -typedef struct { +typedef struct tdb_context { char *name; /* the name of the database */ void *map_ptr; /* where it is currently mapped */ int fd; /* open file descriptor for the database */ tdb_len map_size; /* how much space has been mapped */ int read_only; /* opened read-only */ - int *locked; /* set if we have a chain locked */ - int ecode; /* error code for last tdb error */ + struct tdb_lock_type *locked; /* array of chain locks */ + enum TDB_ERROR ecode; /* error code for last tdb error */ struct tdb_header header; /* a cached copy of the header */ + u32 flags; /* the flags passed to tdb_open */ + struct tdb_traverse_lock travlocks; /* current traversal locks */ + struct tdb_context *next; /* all tdbs to avoid multiple opens */ + dev_t device; /* uniquely identifies this tdb */ + ino_t inode; /* uniquely identifies this tdb */ + void (*log_fn)(struct tdb_context *tdb, int level, const char *, ...) PRINTF_ATTRIBUTE(3,4); /* logging function */ + u32 (*hash_fn)(TDB_DATA *key); + int open_flags; /* flags used in the open - needed by reopen */ } TDB_CONTEXT; -/* flags to tdb_store() */ -#define TDB_REPLACE 1 -#define TDB_INSERT 2 - -/* flags for tdb_open() */ -#define TDB_CLEAR_IF_FIRST 1 - -/* error codes */ -enum TDB_ERROR {TDB_SUCCESS=0, TDB_ERR_CORRUPT, TDB_ERR_IO, TDB_ERR_LOCK, - TDB_ERR_OOM, TDB_ERR_EXISTS}; +typedef int (*tdb_traverse_func)(TDB_CONTEXT *, TDB_DATA, TDB_DATA, void *); +typedef void (*tdb_log_func)(TDB_CONTEXT *, int , const char *, ...); +typedef u32 (*tdb_hash_func)(TDB_DATA *key); -#if STANDALONE -TDB_CONTEXT *tdb_open(char *name, int hash_size, int tdb_flags, +TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags, int open_flags, mode_t mode); -char *tdb_error(TDB_CONTEXT *tdb); -int tdb_writelock(TDB_CONTEXT *tdb); -int tdb_writeunlock(TDB_CONTEXT *tdb); +TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags, + int open_flags, mode_t mode, + tdb_log_func log_fn, + tdb_hash_func hash_fn); + +int tdb_reopen(TDB_CONTEXT *tdb); +int tdb_reopen_all(void); +void tdb_logging_function(TDB_CONTEXT *tdb, tdb_log_func); +enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb); +const char *tdb_errorstr(TDB_CONTEXT *tdb); TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key); int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key); int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag); +int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf); int tdb_close(TDB_CONTEXT *tdb); TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb); TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA key); -int tdb_traverse(TDB_CONTEXT *tdb, - int (*fn)(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, void *state), - void *state); +int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *); int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key); +int tdb_lockkeys(TDB_CONTEXT *tdb, u32 number, TDB_DATA keys[]); +void tdb_unlockkeys(TDB_CONTEXT *tdb); +int tdb_lockall(TDB_CONTEXT *tdb); +void tdb_unlockall(TDB_CONTEXT *tdb); + +/* Low level locking functions: use with care */ +void tdb_set_lock_alarm(sig_atomic_t *palarm); +int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key); +int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key); + +/* Debug functions. Not used in production. */ +void tdb_dump_all(TDB_CONTEXT *tdb); +int tdb_printfreelist(TDB_CONTEXT *tdb); + +extern TDB_DATA tdb_null; + +#ifdef __cplusplus +} #endif + +#endif /* tdb.h */ -- 2.39.2