db53/db-5.3.28-condition_variable.patch

760 lines
26 KiB
Diff

diff -up db-5.3.28/dist/android/android_config.in.pthreads db-5.3.28/dist/android/android_config.in
--- db-5.3.28/dist/android/android_config.in.pthreads 2013-09-09 17:35:02.000000000 +0200
+++ db-5.3.28/dist/android/android_config.in 2017-06-13 11:15:15.323215161 +0200
@@ -123,6 +123,9 @@
/* Define to 1 if allocated filesystem blocks are not zeroed. */
/* #undef HAVE_FILESYSTEM_NOTZERO */
+/* Define to 1 if you have the `flock' function. */
+#undef HAVE_FLOCK
+
/* Define to 1 if you have the `fopen' function. */
#define HAVE_FOPEN 1
diff -up db-5.3.28/dist/config.hin.pthreads db-5.3.28/dist/config.hin
--- db-5.3.28/dist/config.hin.pthreads 2013-09-09 17:35:02.000000000 +0200
+++ db-5.3.28/dist/config.hin 2017-06-13 11:15:15.000000000 +0200
@@ -92,6 +92,9 @@
/* Define to 1 if you have the <dlfcn.h> header file. */
#undef HAVE_DLFCN_H
+/* Define to 1 if you have the `dl_iterate_phdr' function. */
+#undef HAVE_DL_ITERATE_PHDR
+
/* Define to 1 to use dtrace for performance monitoring. */
#undef HAVE_DTRACE
@@ -125,6 +128,9 @@
/* Define to 1 if allocated filesystem blocks are not zeroed. */
#undef HAVE_FILESYSTEM_NOTZERO
+/* Define to 1 if you have the `flock' function. */
+#undef HAVE_FLOCK
+
/* Define to 1 if you have the `fopen' function. */
#undef HAVE_FOPEN
@@ -360,6 +366,9 @@
/* Define to 1 if you have the `pstat_getdynamic' function. */
#undef HAVE_PSTAT_GETDYNAMIC
+/* Define to 1 if you have dl_iterate_phdr and use pthread-based mutexes. */
+#undef HAVE_PTHREADS_TIMESTAMP
+
/* Define to 1 if it is OK to initialize an already initialized
pthread_cond_t. */
#undef HAVE_PTHREAD_COND_REINIT_OKAY
@@ -620,6 +629,11 @@
/* Define to 1 to mask harmless uninitialized memory read/writes. */
#undef UMRW
+/* Enable large inode numbers on Mac OS X 10.5. */
+#ifndef _DARWIN_USE_64_BIT_INODE
+# define _DARWIN_USE_64_BIT_INODE 1
+#endif
+
/* Number of bits in a file offset, on hosts where this is settable. */
#undef _FILE_OFFSET_BITS
diff -up db-5.3.28/dist/configure.ac.pthreads db-5.3.28/dist/configure.ac
--- db-5.3.28/dist/configure.ac.pthreads 2017-06-13 11:15:15.309215434 +0200
+++ db-5.3.28/dist/configure.ac 2017-06-13 11:15:15.323215161 +0200
@@ -698,7 +698,7 @@ AC_REPLACE_FUNCS(\
# Check for system functions we optionally use.
AC_CHECK_FUNCS(\
_fstati64 backtrace backtrace_symbols directio fchmod fclose\
- fcntl fdatasync fgetc fgets fopen fwrite getgid\
+ fcntl fdatasync fgetc fgets flock fopen fwrite getgid\
getrusage getuid hstrerror mprotect pstat_getdynamic\
pthread_self pthread_yield random sched_yield select setgid setuid\
sigaction snprintf stat sysconf vsnprintf yield)
@@ -1042,6 +1042,34 @@ if test "$db_cv_localization" = "yes"; t
[Define to 1 if you have localization function to support globalization.])
fi
+# Check for dl_iterate_phdr; do the test explicitly instead of using
+# AC_CHECK_FUNCS because <netdb.h> isn't a standard include file.
+AC_CACHE_CHECK([for dl_iterate_phdr], db_cv_dl_iterate_phdr, [
+AC_TRY_LINK([
+#include <sys/types.h>
+#include <netdb.h>], [
+ dl_iterate_phdr(0, 0);
+], [db_cv_dl_iterate_phdr=yes], [db_cv_dl_iterate_phdr=no])])
+if test "$db_cv_dl_iterate_phdr" = "yes"; then
+ AC_DEFINE(HAVE_DL_ITERATE_PHDR)
+ AH_TEMPLATE(HAVE_DL_ITERATE_PHDR,
+ [Define to 1 if you have the `dl_iterate_phdr' function.])
+fi
+
+# If we are using pthread mutex or condition variables, and dl_iterate_phdr() is
+# available, then we try to detect when libpthread is updated -- which can
+# render existing environment invalid. DB_ENV->open() tries to rebuild such
+# environments when they are idle.
+case "$db_cv_mutex" in
+ *pthreads*)
+ if test "$db_cv_dl_iterate_phdr" = "yes" ; then
+ AC_DEFINE(HAVE_PTHREADS_TIMESTAMP)
+ AH_TEMPLATE(HAVE_PTHREADS_TIMESTAMP,
+ [Define to 1 if you have dl_iterate_phdr and use pthread-based mutexes.])
+ fi
+ ;;
+esac
+
# We need to add the additional object files into the Makefile with the correct
# suffix. We can't use $LTLIBOBJS itself, because that variable has $U encoded
# in it for automake, and that's not what we want. See SR #7227 for additional
diff -up db-5.3.28/src/db/db_meta.c.pthreads db-5.3.28/src/db/db_meta.c
--- db-5.3.28/src/db/db_meta.c.pthreads 2013-09-09 17:35:07.000000000 +0200
+++ db-5.3.28/src/db/db_meta.c 2017-06-13 11:15:15.323215161 +0200
@@ -1330,8 +1330,9 @@ __db_haslock(env, locker, dbmfp, pgno, m
}
/*
* __db_has_pagelock --
- * Determine if this locker holds a particular page lock.
- * Returns 0 if lock is held, non-zero otherwise.
+ * Determine if this locker holds a particular page lock, and return an
+ * error if it is missing a page lock that it should have.
+ * Otherwise (TDS with the page locked, or DS or CDS) return 0.
*
* PUBLIC: #ifdef DIAGNOSTIC
* PUBLIC: int __db_has_pagelock __P((ENV *, DB_LOCKER *,
@@ -1348,6 +1349,9 @@ __db_has_pagelock(env, locker, dbmfp, pa
{
int ret;
+ if (!FLD_ISSET(env->open_flags, DB_INIT_TXN))
+ return (0);
+
switch (pagep->type) {
case P_OVERFLOW:
case P_INVALID:
diff -up db-5.3.28/src/dbinc_auto/int_def.in.pthreads db-5.3.28/src/dbinc_auto/int_def.in
--- db-5.3.28/src/dbinc_auto/int_def.in.pthreads 2017-06-13 11:15:15.317215278 +0200
+++ db-5.3.28/src/dbinc_auto/int_def.in 2017-06-13 11:15:15.324215141 +0200
@@ -1545,6 +1545,7 @@
#if defined(HAVE_REPLICATION_THREADS)
#define __os_freeaddrinfo __os_freeaddrinfo@DB_VERSION_UNIQUE_NAME@
#endif
+#define __os_pthreads_timestamp __os_pthreads_timestamp@DB_VERSION_UNIQUE_NAME@
#define __os_umalloc __os_umalloc@DB_VERSION_UNIQUE_NAME@
#define __os_urealloc __os_urealloc@DB_VERSION_UNIQUE_NAME@
#define __os_ufree __os_ufree@DB_VERSION_UNIQUE_NAME@
diff -up db-5.3.28/src/dbinc_auto/os_ext.h.pthreads db-5.3.28/src/dbinc_auto/os_ext.h
--- db-5.3.28/src/dbinc_auto/os_ext.h.pthreads 2013-09-09 17:35:08.000000000 +0200
+++ db-5.3.28/src/dbinc_auto/os_ext.h 2017-06-13 11:15:15.324215141 +0200
@@ -14,6 +14,7 @@ int __os_getaddrinfo __P((ENV *, const c
#if defined(HAVE_REPLICATION_THREADS)
void __os_freeaddrinfo __P((ENV *, ADDRINFO *));
#endif
+time_t __os_pthreads_timestamp __P((ENV *));
int __os_umalloc __P((ENV *, size_t, void *));
int __os_urealloc __P((ENV *, size_t, void *));
void __os_ufree __P((ENV *, void *));
@@ -40,7 +41,7 @@ void __os_set_errno __P((int));
char *__os_strerror __P((int, char *, size_t));
int __os_posix_err __P((int));
int __os_fileid __P((ENV *, const char *, int, u_int8_t *));
-int __os_fdlock __P((ENV *, DB_FH *, off_t, int, int));
+int __os_fdlock __P((ENV *, DB_FH *, off_t, db_lockmode_t, int));
int __os_fsync __P((ENV *, DB_FH *));
int __os_getenv __P((ENV *, const char *, char **, size_t));
int __os_openhandle __P((ENV *, const char *, int, int, DB_FH **));
diff -up db-5.3.28/src/dbinc/region.h.pthreads db-5.3.28/src/dbinc/region.h
--- db-5.3.28/src/dbinc/region.h.pthreads 2013-09-09 17:35:08.000000000 +0200
+++ db-5.3.28/src/dbinc/region.h 2017-06-13 11:15:15.324215141 +0200
@@ -178,6 +178,16 @@ typedef struct __db_reg_env { /* SHARED
u_int32_t envid; /* Unique environment ID. */
u_int32_t signature; /* Structure signatures. */
+#if defined(HAVE_PTHREADS_TIMESTAMP) && defined(HAVE_MUTEX_PTHREADS)
+ /*
+ * Updates to glibc/libpthread can change its pthreads implementation
+ * and invalidate on-disk environments, even without changing the
+ * version number. If using POSIX mutexes and a change in this
+ * timestamp is detecting when opening an environment with DB_CREATE,
+ * __env_attach recreates any existing on-disk environment.
+ */
+ time_t pthreads_timestamp;
+#endif
time_t timestamp; /* Creation time. */
diff -up db-5.3.28/src/env/env_open.c.pthreads db-5.3.28/src/env/env_open.c
--- db-5.3.28/src/env/env_open.c.pthreads 2013-09-09 17:35:08.000000000 +0200
+++ db-5.3.28/src/env/env_open.c 2017-06-13 11:15:15.324215141 +0200
@@ -1031,11 +1031,11 @@ __env_attach_regions(dbenv, flags, orig_
goto err;
/*
- * __env_attach will return the saved init_flags field, which contains
+ * __env_attach has returned the saved init_flags field, which contains
* the DB_INIT_* flags used when the environment was created.
*
- * We may be joining an environment -- reset our flags to match the
- * ones in the environment.
+ * We may be joining an existing environment -- reset our flags to match
+ * the ones in the environment.
*/
if (FLD_ISSET(init_flags, DB_INITENV_CDB))
LF_SET(DB_INIT_CDB);
diff -up db-5.3.28/src/env/env_region.c.pthreads db-5.3.28/src/env/env_region.c
--- db-5.3.28/src/env/env_region.c.pthreads 2013-09-09 17:35:08.000000000 +0200
+++ db-5.3.28/src/env/env_region.c 2017-06-13 11:16:02.159299222 +0200
@@ -18,13 +18,49 @@ static int __env_des_get __P((ENV *, RE
static int __env_faultmem __P((ENV *, void *, size_t, int));
static int __env_sys_attach __P((ENV *, REGINFO *, REGION *));
static int __env_sys_detach __P((ENV *, REGINFO *, int));
+static int __env_check_recreate __P((ENV *, REGENV *, u_int32_t));
static void __env_des_destroy __P((ENV *, REGION *));
static void __env_remove_file __P((ENV *));
+
+/*
+ * If the system supports flock()-like file locking, then the primary region
+ * file __db.001 is exclusively locked during creation, and is read-locked while
+ * the environment is open. Most Unix-like systems have flock(), with the
+ * notable exception of Solaris.
+ * Note: fcntl cannot be used for this locking because of the unfortunate
+ * definition of its interaction with close(2). A process's fcntl locks are
+ * released whenever it closes any file descriptor for that file. So, if an
+ * environment is opened more than once, closing one of the DB_ENV handles would
+ * release the read lock that protects the other handle.
+ */
+#ifdef HAVE_FLOCK
+#define ENV_PRIMARY_LOCK(env, lockmode, async) \
+ ((env)->lockfhp == NULL ? 0 : \
+ __os_fdlock((env), (env)->lockfhp, -1, lockmode, async))
+#define ENV_PRIMARY_UNLOCK(env) \
+ ((env)->lockfhp == NULL ? 0 : \
+ __os_fdlock((env), (env)->lockfhp, -1, DB_LOCK_NG, 0))
+#else
+#define ENV_PRIMARY_LOCK(env, lockmode, async) (0)
+#define ENV_PRIMARY_UNLOCK(env) (0)
+#endif
+
/*
* __env_attach
* Join/create the environment
*
+ * Safely detecting and managing multiple processes' environment handles:
+ * BDB uses a shared or exclusive fcntl()-style lock on the first byte
+ * of the primary region file (__db.001) to detect whether other processes
+ * have the environment open, and to single-thread attempts to create the
+ * environment. If the open includes DB_CREATE, an exclusive lock is
+ * obtained during the open call. After the creation is finished, and
+ * anytime during a non-DB_CREATE env open, the process holds a shared
+ * lock.
+ * - single-thread creation of the environment
+ * - detect whether any other processes are currently attached to it.
+ *
* PUBLIC: int __env_attach __P((ENV *, u_int32_t *, int, int));
*/
int
@@ -122,7 +158,11 @@ loop: renv = NULL;
if ((ret = __os_open(
env, infop->name, 0, DB_OSO_REGION, 0, &env->lockfhp)) != 0)
goto err;
-
+ /* Wait to get shared access to the primary region. */
+ if ((ret = ENV_PRIMARY_LOCK(env, DB_LOCK_READ, 0)) != 0) {
+ __db_err(env, ret, "__env_attach: existing: shared lock error");
+ goto err;
+ }
/*
* !!!
* The region may be in system memory not backed by the filesystem
@@ -218,11 +258,10 @@ loop: renv = NULL;
segid = ref.segid;
}
-#ifndef HAVE_MUTEX_FCNTL
+#if !defined(HAVE_FCNTL) && !defined(HAVE_PTHREADS_TIMESTAMP)
/*
- * If we're not doing fcntl locking, we can close the file handle. We
- * no longer need it and the less contact between the buffer cache and
- * the VM, the better.
+ * Without fcntl-like support, we no longer need the file handle. Close
+ * it to limit the interaction between the buffer cache and the VM.
*/
(void)__os_closehandle(env, env->lockfhp);
env->lockfhp = NULL;
@@ -233,6 +272,8 @@ loop: renv = NULL;
tregion.size = (roff_t)size;
tregion.max = (roff_t)max;
tregion.segid = segid;
+ /* Attach to the existing primary region. */
+ /* The leaking db.001 gets open inside of here, in __os_attach(). */
if ((ret = __env_sys_attach(env, infop, &tregion)) != 0)
goto err;
@@ -245,21 +286,48 @@ user_map_functions:
infop->primary = infop->addr;
infop->head = (u_int8_t *)infop->addr + sizeof(REGENV);
renv = infop->primary;
+ ret = __env_check_recreate(env, renv, signature);
+
+ if (create_ok &&
+ ret == DB_OLD_VERSION &&
+ ENV_PRIMARY_LOCK(env, DB_LOCK_WRITE, 1) == 0) {
+ if (FLD_ISSET(dbenv->verbose, DB_VERB_RECOVERY))
+ __db_msg(env, "Recreating idle environment");
+ F_SET(infop, REGION_CREATE_OK);
+
+ /*
+ * Detach from the environment region; we need to unmap it (and
+ * close any file handle) so that we don't leak memory or files.
+ */
+ DB_ASSERT(env, infop->rp == NULL);
+ infop->rp = &tregion;
+ (void)__env_sys_detach(env, infop, 0);
+ goto creation;
+ }
+
+ /* We have an old environment but cannot rebuild it safely. */
+ if (ret == DB_OLD_VERSION) {
+ __db_errx(env, DB_STR("1539",
+ "Build signature doesn't match environment"));
+ ret = DB_VERSION_MISMATCH;
+ goto err;
+ }
- /*
- * Make sure the region matches our build. Special case a region
- * that's all nul bytes, just treat it like any other corruption.
- */
if (renv->majver != DB_VERSION_MAJOR ||
renv->minver != DB_VERSION_MINOR) {
- if (renv->majver != 0 || renv->minver != 0) {
+ /*
+ * Special case a region that's all nul bytes, just treat it
+ * like any other corruption.
+ */
+ if (renv->majver == 0 && renv->minver == 0)
+ ret = EINVAL;
+ else {
__db_errx(env, DB_STR_A("1538",
- "Program version %d.%d doesn't match environment version %d.%d",
+ "Program version %d.%d doesn't match in-use environment version %d.%d",
"%d %d %d %d"), DB_VERSION_MAJOR, DB_VERSION_MINOR,
renv->majver, renv->minver);
ret = DB_VERSION_MISMATCH;
- } else
- ret = EINVAL;
+ }
goto err;
}
if (renv->signature != signature) {
@@ -289,6 +357,18 @@ user_map_functions:
}
if (renv->magic != DB_REGION_MAGIC)
goto retry;
+ /*
+ * A bad magic number means that the env is new and not yet available:
+ * wait a while and try again. If the magic number says recovery is in
+ * process, remember the env creation time to record that recovery was
+ * the reason that the open failed.
+ */
+ if (renv->magic != DB_REGION_MAGIC) {
+ __db_msg(env, "attach sees bad region magic 0x%lx",
+ (u_long)renv->magic);
+ goto retry;
+ }
+
/*
* Get a reference to the underlying REGION information for this
@@ -346,6 +426,12 @@ user_map_functions:
return (0);
creation:
+ /* Should this wait for the lock (passing 0 instead of 1)? */
+ if ((ret = ENV_PRIMARY_LOCK(env, DB_LOCK_WRITE, 1)) != 0) {
+ __db_err(env, ret, "__env_attach: creation could not lock %s",
+ env->lockfhp->name);
+ goto err;
+ }
/* Create the environment region. */
F_SET(infop, REGION_CREATE);
@@ -437,7 +523,14 @@ creation:
renv->minver = (u_int32_t)minver;
renv->patchver = (u_int32_t)patchver;
renv->signature = signature;
-
+#ifdef HAVE_PTHREADS_TIMESTAMP
+ renv->pthreads_timestamp = __os_pthreads_timestamp(env);
+ {
+ char *s = getenv("TS_ADJUST");
+ if (s != NULL)
+ renv->pthreads_timestamp -= atoi(s);
+ }
+#endif
(void)time(&renv->timestamp);
__os_unique_id(env, &renv->envid);
@@ -513,16 +606,24 @@ find_err: __db_errx(env, DB_STR_A("1544"
}
}
-#ifndef HAVE_MUTEX_FCNTL
- /*
- * If we're not doing fcntl locking, we can close the file handle. We
- * no longer need it and the less contact between the buffer cache and
- * the VM, the better.
- */
+#ifdef HAVE_FCNTL
+ if ((ret = ENV_PRIMARY_UNLOCK(env)) != 0) {
+ __db_err(env, ret, "__env_attach: release exclusive lock");
+ goto err;
+ }
+ if ((ret = ENV_PRIMARY_LOCK(env, DB_LOCK_READ, 0)) != 0) {
+ __db_err(env, ret, "__env_attach: new: acquire shared lock");
+ goto err;
+ }
+#else
+ /*
+ * We no longer need the primary region file's handle and the less
+ * contact between the buffer cache and the VM, the better.
+ */
if (env->lockfhp != NULL) {
(void)__os_closehandle(env, env->lockfhp);
env->lockfhp = NULL;
- }
+ }
#endif
/* Everything looks good, we're done. */
@@ -562,9 +663,9 @@ retry: /* Close any open file handle. */
/* If we had a temporary error, wait awhile and try again. */
if (ret == 0) {
if (!retry_ok || ++retry_cnt > 3) {
+ ret = EAGAIN;
__db_errx(env, DB_STR("1546",
"unable to join the environment"));
- ret = EAGAIN;
} else {
__os_yield(env, retry_cnt * 3, 0);
goto loop;
@@ -575,6 +676,59 @@ retry: /* Close any open file handle. */
}
/*
+ * __env_check_recreate --
+ * Determine whether an existing on-disk environment should be recreated
+ * because it is not compatible with this compiled BDB library.
+ *
+ * Returns:
+ * 0 -
+ * The env was generated by this library. No recreation needed.
+ * DB_OLD_VERSION -
+ * It was created by an earlier BDB version, or by an earlier
+ * version of libpthreads (on certain Linux systems). The caller
+ * will try to recreate it with the currently configured settings.
+ * DB_VERSION_MISMATCH -
+ * It was created by a newer version of BDB. Do not attempt to
+ * fix it, something is probably wrong with the application setup.
+ */
+static int
+__env_check_recreate(env, renv, signature)
+ ENV *env;
+ REGENV *renv;
+ u_int32_t signature;
+{
+#ifdef HAVE_PTHREADS_TIMESTAMP
+ time_t pthreads_time;
+ char envtime[CTIME_BUFLEN], libtime[CTIME_BUFLEN];
+#endif
+
+ /* First, bail out if the env is too new for this code to handle. */
+ if (renv->majver > DB_VERSION_MAJOR ||
+ (renv->majver == DB_VERSION_MAJOR &&
+ renv->minver > DB_VERSION_MINOR))
+ return (DB_VERSION_MISMATCH);
+
+#ifdef HAVE_PTHREADS_TIMESTAMP
+ pthreads_time = __os_pthreads_timestamp(env);
+ if (pthreads_time != renv->pthreads_timestamp) {
+ if (FLD_ISSET(env->dbenv->verbose, DB_VERB_RECOVERY))
+ __db_msg(env,
+ "Pthreads timestamp changed: env %.24s current %.24s",
+ __os_ctime(&renv->pthreads_timestamp, envtime),
+ __os_ctime(&pthreads_time, libtime));
+ return (DB_OLD_VERSION);
+ }
+#endif
+ if (renv->signature != signature || renv->majver != DB_VERSION_MAJOR ||
+ renv->minver != DB_VERSION_MINOR) {
+ if (FLD_ISSET(env->dbenv->verbose, DB_VERB_RECOVERY))
+ __db_msg(env, "Signature or version changed");
+ return (DB_OLD_VERSION);
+ }
+ return (0);
+}
+
+/*
* __env_turn_on --
* Turn on the created environment.
*
diff -up db-5.3.28/src/env/env_register.c.pthreads db-5.3.28/src/env/env_register.c
--- db-5.3.28/src/env/env_register.c.pthreads 2013-09-09 17:35:08.000000000 +0200
+++ db-5.3.28/src/env/env_register.c 2017-06-13 11:15:15.325215121 +0200
@@ -19,9 +19,9 @@
#define PID_LEN (25) /* PID entry length */
#define REGISTRY_LOCK(env, pos, nowait) \
- __os_fdlock(env, (env)->dbenv->registry, (off_t)(pos), 1, nowait)
+ __os_fdlock(env, (env)->dbenv->registry, (off_t)(pos), DB_LOCK_WRITE, nowait)
#define REGISTRY_UNLOCK(env, pos) \
- __os_fdlock(env, (env)->dbenv->registry, (off_t)(pos), 0, 0)
+ __os_fdlock(env, (env)->dbenv->registry, (off_t)(pos), DB_LOCK_NG, 0)
#define REGISTRY_EXCL_LOCK(env, nowait) \
REGISTRY_LOCK(env, 1, nowait)
#define REGISTRY_EXCL_UNLOCK(env) \
diff -up db-5.3.28/src/env/env_stat.c.pthreads db-5.3.28/src/env/env_stat.c
--- db-5.3.28/src/env/env_stat.c.pthreads 2013-09-09 17:35:08.000000000 +0200
+++ db-5.3.28/src/env/env_stat.c 2017-06-13 11:15:15.325215121 +0200
@@ -177,6 +177,10 @@ __env_print_stats(env, flags)
STAT_LONG("Txn version", DB_TXNVERSION);
__db_msg(env,
"%.24s\tCreation time", __os_ctime(&renv->timestamp, time_buf));
+#if defined(HAVE_PTHREADS_TIMESTAMP) && defined(HAVE_MUTEX_PTHREADS)
+ __db_msg(env,
+ "%.24s\tlibpthread timestamp", __os_ctime(&renv->pthreads_timestamp, time_buf));
+#endif
STAT_HEX("Environment ID", renv->envid);
__mutex_print_debug_single(env,
"Primary region allocation and reference count mutex",
diff -up db-5.3.28/src/os/os_addrinfo.c.pthreads db-5.3.28/src/os/os_addrinfo.c
--- db-5.3.28/src/os/os_addrinfo.c.pthreads 2013-09-09 17:35:09.000000000 +0200
+++ db-5.3.28/src/os/os_addrinfo.c 2017-06-13 11:15:15.325215121 +0200
@@ -10,6 +10,10 @@
#include "db_int.h"
+#ifdef HAVE_PTHREADS_TIMESTAMP
+#include <link.h>
+#endif
+
/*
* __os_getaddrinfo and __os_freeaddrinfo wrap the getaddrinfo and freeaddrinfo
* calls, as well as the associated platform dependent error handling, mapping
@@ -177,3 +181,48 @@ __os_freeaddrinfo(env, ai)
}
#endif
}
+
+#ifdef HAVE_PTHREADS_TIMESTAMP
+/*
+ * callback_find_pthreads --
+ * dl_iterate_phdr() calls this once for each loaded library.
+ *
+ * Returns:
+ * 0 - the library does not appear to be libpthreads.
+ * 1 - the library *does* seem to be libpthreads. Its modification time is
+ * stored into into the last argument's location.
+ */
+static int
+callback_find_pthreads(struct dl_phdr_info *info, size_t size, void *data)
+{
+ struct stat stbuf;
+
+ /* Stop (return non-zero) when libc is found. */
+ if (strstr(info->dlpi_name, "libpthread") != NULL &&
+ stat(info->dlpi_name, &stbuf) == 0) {
+ *(time_t *)data = stbuf.st_mtime;
+ return (1);
+ }
+ COMPQUIET(size, 0);
+ COMPQUIET(data, NULL);
+ return (0);
+}
+
+/*
+ * __os_pthreads_timestamp --
+ *
+ * PUBLIC: time_t __os_pthreads_timestamp __P((ENV *));
+ */
+time_t
+__os_pthreads_timestamp(env)
+ ENV *env;
+{
+ time_t timestamp;
+
+ timestamp = 0;
+ dl_iterate_phdr(callback_find_pthreads, &timestamp);
+
+ COMPQUIET(env, 0);
+ return (timestamp);
+}
+#endif
diff -up db-5.3.28/src/os/os_flock.c.pthreads db-5.3.28/src/os/os_flock.c
--- db-5.3.28/src/os/os_flock.c.pthreads 2013-09-09 17:35:09.000000000 +0200
+++ db-5.3.28/src/os/os_flock.c 2017-06-13 11:15:15.325215121 +0200
@@ -10,41 +10,98 @@
#include "db_int.h"
+#if !defined(HAVE_FCNTL) || !defined(HAVE_FLOCK)
+static int __os_filelocking_notsup __P((ENV *));
+#endif
+
/*
* __os_fdlock --
* Acquire/release a lock on a byte in a file.
*
- * PUBLIC: int __os_fdlock __P((ENV *, DB_FH *, off_t, int, int));
+ * The lock modes supported here are:
+ * DB_LOCK_NG - release the lock
+ * DB_LOCK_READ - get shared access
+ * DB_LOCK_WRITE - get exclusive access
+ *
+ * Use fcntl()-like semantics most of the time (DB_REGISTER support). Fcntl
+ * supports range locking, but has the additional broken semantics that
+ * closing any of the file's descriptors releases any locks, even if its
+ * other file descriptors remain open. Thanks SYSV & POSIX.
+ * However, if the offset is negative (which is allowed, because POSIX
+ * off_t a signed integer) then use flock() instead. It has only whole-
+ * file locks, but they persist until explicitly unlocked or the process
+ * exits.
+ * PUBLIC: int __os_fdlock __P((ENV *, DB_FH *, off_t, db_lockmode_t, int));
*/
int
-__os_fdlock(env, fhp, offset, acquire, nowait)
+__os_fdlock(env, fhp, offset, lockmode, nowait)
ENV *env;
DB_FH *fhp;
- int acquire, nowait;
off_t offset;
+ db_lockmode_t lockmode;
+ int nowait;
{
#ifdef HAVE_FCNTL
DB_ENV *dbenv;
struct flock fl;
int ret, t_ret;
+ static char *mode_string[DB_LOCK_WRITE + 1] = {
+ "unlock",
+ "read",
+ "write"
+ };
+ short mode_fcntl[DB_LOCK_WRITE + 1] = {
+ F_UNLCK,
+ F_RDLCK,
+ F_WRLCK
+ };
+#ifdef HAVE_FLOCK
+ short mode_flock[DB_LOCK_WRITE + 1] = {
+ LOCK_UN,
+ LOCK_SH,
+ LOCK_EX
+ };
+#endif
dbenv = env == NULL ? NULL : env->dbenv;
DB_ASSERT(env, F_ISSET(fhp, DB_FH_OPENED) && fhp->fd != -1);
+ DB_ASSERT(env, lockmode <= DB_LOCK_WRITE);
- if (dbenv != NULL && FLD_ISSET(dbenv->verbose, DB_VERB_FILEOPS_ALL))
- __db_msg(env, DB_STR_A("0138",
- "fileops: flock %s %s offset %lu", "%s %s %lu"), fhp->name,
- acquire ? DB_STR_P("acquire"): DB_STR_P("release"),
- (u_long)offset);
-
- fl.l_start = offset;
- fl.l_len = 1;
- fl.l_type = acquire ? F_WRLCK : F_UNLCK;
- fl.l_whence = SEEK_SET;
-
- RETRY_CHK_EINTR_ONLY(
- (fcntl(fhp->fd, nowait ? F_SETLK : F_SETLKW, &fl)), ret);
+ if (dbenv != NULL && FLD_ISSET(dbenv->verbose, DB_VERB_FILEOPS_ALL)) {
+ if (offset < 0)
+ __db_msg(env, DB_STR_A("####",
+ "fileops: flock %s %s %s", "%s %s %s"),
+ fhp->name, mode_string[lockmode],
+ nowait ? "nowait" : "");
+ else
+ __db_msg(env, DB_STR_A("0020",
+ "fileops: fcntls %s %s offset %lu", "%s %s %lu"),
+ fhp->name, mode_string[lockmode], (u_long)offset);
+ }
+
+ if (offset < 0) {
+#ifdef HAVE_FLOCK
+ RETRY_CHK_EINTR_ONLY(flock(fhp->fd,
+ mode_flock[lockmode] | (nowait ? LOCK_NB : 0)), ret);
+#else
+ ret = __os_filelocking_notsup(env);
+#endif
+ } else {
+ fl.l_start = offset;
+ fl.l_len = 1;
+ fl.l_whence = SEEK_SET;
+ fl.l_type = mode_fcntl[lockmode];
+ RETRY_CHK_EINTR_ONLY(
+ fcntl(fhp->fd, nowait ? F_SETLK : F_SETLKW, &fl), ret);
+ }
+
+ if (offset < 0 && dbenv != NULL &&
+ FLD_ISSET(dbenv->verbose, DB_VERB_FILEOPS_ALL))
+ __db_msg(env, DB_STR_A("####",
+ "fileops: flock %s %s %s returns %s", "%s %s %s"),
+ fhp->name, mode_string[lockmode],
+ nowait ? "nowait" : "", db_strerror(ret));
if (ret == 0)
return (0);
@@ -53,12 +110,29 @@ __os_fdlock(env, fhp, offset, acquire, n
__db_syserr(env, ret, DB_STR("0139", "fcntl"));
return (t_ret);
#else
+ ret = __os_filelocking_notsup(env);
COMPQUIET(fhp, NULL);
- COMPQUIET(acquire, 0);
+ COMPQUIET(lockmode, 0);
COMPQUIET(nowait, 0);
COMPQUIET(offset, 0);
+ return (ret)
+#endif
+}
+
+
+#if !defined(HAVE_FCNTL) || !defined(HAVE_FLOCK)
+/*
+ * __os_filelocking_notsup --
+ * Generate an error message if fcntl() or flock() is requested on a
+ * platform that does not support it.
+ *
+ */
+static int
+__os_filelocking_notsup(env)
+ ENV *env;
+{
__db_syserr(env, DB_OPNOTSUP, DB_STR("0140",
"advisory file locking unavailable"));
return (DB_OPNOTSUP);
-#endif
}
+#endif
diff -up db-5.3.28/src/os/os_map.c.pthreads db-5.3.28/src/os/os_map.c
--- db-5.3.28/src/os/os_map.c.pthreads 2013-09-09 17:35:09.000000000 +0200
+++ db-5.3.28/src/os/os_map.c 2017-06-13 11:15:15.325215121 +0200
@@ -32,7 +32,7 @@ static int __no_system_mem __P((ENV *));
/*
* __os_attach --
- * Create/join a shared memory region.
+ * Create/join a 'shared' region of Berkeley DB memory.
*
* PUBLIC: int __os_attach __P((ENV *, REGINFO *, REGION *));
*/
@@ -50,6 +50,7 @@ __os_attach(env, infop, rp)
* so there must be a valid handle.
*/
DB_ASSERT(env, env != NULL && env->dbenv != NULL);
+ DB_ASSERT(env, infop->fhp == NULL);
dbenv = env->dbenv;
if (DB_GLOBAL(j_region_map) != NULL) {