/*
* __curlog_search --
* WT_CURSOR.search method for the log cursor type.
*/
static int
__curlog_search(WT_CURSOR *cursor)
{
WT_CURSOR_LOG *cl;
WT_DECL_RET;
WT_LSN key;
WT_SESSION_IMPL *session;
uint32_t counter, key_file, key_offset;
cl = (WT_CURSOR_LOG *)cursor;
CURSOR_API_CALL(cursor, session, search, NULL);
/*
* !!! We are ignoring the counter and only searching based on the LSN.
*/
WT_ERR(__wt_cursor_get_key((WT_CURSOR *)cl,
&key_file, &key_offset, &counter));
WT_SET_LSN(&key, key_file, key_offset);
ret = __wt_log_scan(session, &key, WT_LOGSCAN_ONE,
__curlog_logrec, cl);
if (ret == ENOENT)
ret = WT_NOTFOUND;
WT_ERR(ret);
WT_ERR(__curlog_kv(session, cursor));
WT_STAT_FAST_CONN_INCR(session, cursor_search);
WT_STAT_FAST_DATA_INCR(session, cursor_search);
err: API_END_RET(session, ret);
}
/*
* __curlog_search --
* WT_CURSOR.search method for the log cursor type.
*/
static int
__curlog_search(WT_CURSOR *cursor)
{
WT_CURSOR_LOG *cl;
WT_DECL_RET;
WT_LSN key;
WT_SESSION_IMPL *session;
uint32_t counter, key_file, key_offset, raw;
cl = (WT_CURSOR_LOG *)cursor;
/* Temporarily turn off raw so we can do direct cursor operations. */
raw = F_MASK(cursor, WT_CURSTD_RAW);
F_CLR(cursor, WT_CURSTD_RAW);
CURSOR_API_CALL(cursor, session, search, NULL);
/*
* !!! We are ignoring the counter and only searching based on the LSN.
*/
WT_ERR(__wt_cursor_get_key(cursor, &key_file, &key_offset, &counter));
WT_SET_LSN(&key, key_file, key_offset);
ret = __wt_log_scan(session, &key, WT_LOGSCAN_ONE,
__curlog_logrec, cl);
if (ret == ENOENT)
ret = WT_NOTFOUND;
WT_ERR(ret);
WT_ERR(__curlog_kv(session, cursor));
WT_STAT_CONN_INCR(session, cursor_search);
WT_STAT_DATA_INCR(session, cursor_search);
err: F_SET(cursor, raw);
API_END_RET(session, ret);
}
/*
* __recovery_setup_file --
* Set up the recovery slot for a file.
*/
static int
__recovery_setup_file(WT_RECOVERY *r, const char *uri, const char *config)
{
WT_CONFIG_ITEM cval;
WT_LSN lsn;
uint32_t fileid, lsnfile, lsnoffset;
WT_RET(__wt_config_getones(r->session, config, "id", &cval));
fileid = (uint32_t)cval.val;
/* Track the largest file ID we have seen. */
if (fileid > r->max_fileid)
r->max_fileid = fileid;
if (r->nfiles <= fileid) {
WT_RET(__wt_realloc_def(
r->session, &r->file_alloc, fileid + 1, &r->files));
r->nfiles = fileid + 1;
}
WT_RET(__wt_strdup(r->session, uri, &r->files[fileid].uri));
WT_RET(
__wt_config_getones(r->session, config, "checkpoint_lsn", &cval));
/* If there is checkpoint logged for the file, apply everything. */
if (cval.type != WT_CONFIG_ITEM_STRUCT)
WT_INIT_LSN(&lsn);
/* NOLINTNEXTLINE(cert-err34-c) */
else if (sscanf(cval.str,
"(%" SCNu32 ",%" SCNu32 ")", &lsnfile, &lsnoffset) == 2)
WT_SET_LSN(&lsn, lsnfile, lsnoffset);
else
WT_RET_MSG(r->session, EINVAL,
"Failed to parse checkpoint LSN '%.*s'",
(int)cval.len, cval.str);
r->files[fileid].ckpt_lsn = lsn;
__wt_verbose(r->session, WT_VERB_RECOVERY,
"Recovering %s with id %" PRIu32 " @ (%" PRIu32 ", %" PRIu32 ")",
uri, fileid, lsn.l.file, lsn.l.offset);
if ((!WT_IS_MAX_LSN(&lsn) && !WT_IS_INIT_LSN(&lsn)) &&
(WT_IS_MAX_LSN(&r->max_ckpt_lsn) ||
__wt_log_cmp(&lsn, &r->max_ckpt_lsn) > 0))
r->max_ckpt_lsn = lsn;
return (0);
}
/*
* __wt_txn_checkpoint_logread --
* Read a log record for a checkpoint operation.
*/
int
__wt_txn_checkpoint_logread(WT_SESSION_IMPL *session,
const uint8_t **pp, const uint8_t *end, WT_LSN *ckpt_lsn)
{
WT_DECL_RET;
WT_ITEM ckpt_snapshot_unused;
uint32_t ckpt_file, ckpt_offset;
u_int ckpt_nsnapshot_unused;
const char *fmt = WT_UNCHECKED_STRING(IIIu);
if ((ret = __wt_struct_unpack(session, *pp, WT_PTRDIFF(end, *pp), fmt,
&ckpt_file, &ckpt_offset,
&ckpt_nsnapshot_unused, &ckpt_snapshot_unused)) != 0)
WT_RET_MSG(session,
ret, "txn_checkpoint_logread: unpack failure");
WT_SET_LSN(ckpt_lsn, ckpt_file, ckpt_offset);
*pp = end;
return (0);
}
/*
* __log_file_server --
* The log file server thread. This worker thread manages
* log file operations such as closing and syncing.
*/
static WT_THREAD_RET
__log_file_server(void *arg)
{
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
WT_FH *close_fh;
WT_LOG *log;
WT_LSN close_end_lsn, min_lsn;
WT_SESSION_IMPL *session;
uint32_t filenum;
bool locked;
session = arg;
conn = S2C(session);
log = conn->log;
locked = false;
while (F_ISSET(conn, WT_CONN_LOG_SERVER_RUN)) {
/*
* If there is a log file to close, make sure any outstanding
* write operations have completed, then fsync and close it.
*/
if ((close_fh = log->log_close_fh) != NULL) {
WT_ERR(__wt_log_extract_lognum(session, close_fh->name,
&filenum));
/*
* We update the close file handle before updating the
* close LSN when changing files. It is possible we
* could see mismatched settings. If we do, yield
* until it is set. This should rarely happen.
*/
while (log->log_close_lsn.l.file < filenum)
__wt_yield();
if (__wt_log_cmp(
&log->write_lsn, &log->log_close_lsn) >= 0) {
/*
* We've copied the file handle, clear out the
* one in the log structure to allow it to be
* set again. Copy the LSN before clearing
* the file handle.
* Use a barrier to make sure the compiler does
* not reorder the following two statements.
*/
close_end_lsn = log->log_close_lsn;
WT_FULL_BARRIER();
log->log_close_fh = NULL;
/*
* Set the close_end_lsn to the LSN immediately
* after ours. That is, the beginning of the
* next log file. We need to know the LSN
* file number of our own close in case earlier
* calls are still in progress and the next one
* to move the sync_lsn into the next file for
* later syncs.
*/
WT_ERR(__wt_fsync(session, close_fh));
/*
* We want to make sure the file size reflects
* actual data and has minimal pre-allocated
* zeroed space.
*/
WT_ERR(__wt_ftruncate(session,
close_fh, close_end_lsn.l.offset));
WT_SET_LSN(&close_end_lsn,
close_end_lsn.l.file + 1, 0);
__wt_spin_lock(session, &log->log_sync_lock);
locked = true;
WT_ERR(__wt_close(session, &close_fh));
WT_ASSERT(session, __wt_log_cmp(
&close_end_lsn, &log->sync_lsn) >= 0);
log->sync_lsn = close_end_lsn;
WT_ERR(__wt_cond_signal(
session, log->log_sync_cond));
locked = false;
__wt_spin_unlock(session, &log->log_sync_lock);
}
}
/*
* If a later thread asked for a background sync, do it now.
*/
if (__wt_log_cmp(&log->bg_sync_lsn, &log->sync_lsn) > 0) {
/*
* Save the latest write LSN which is the minimum
* we will have written to disk.
*/
min_lsn = log->write_lsn;
/*
* We have to wait until the LSN we asked for is
* written. If it isn't signal the wrlsn thread
* to get it written.
*
* We also have to wait for the written LSN and the
* sync LSN to be in the same file so that we know we
* have synchronized all earlier log files.
*/
if (__wt_log_cmp(&log->bg_sync_lsn, &min_lsn) <= 0) {
/*
* If the sync file is behind either the one
* wanted for a background sync or the write LSN
* has moved to another file continue to let
* this worker thread process that older file
* immediately.
*/
if ((log->sync_lsn.l.file <
log->bg_sync_lsn.l.file) ||
(log->sync_lsn.l.file < min_lsn.l.file))
continue;
WT_ERR(__wt_fsync(session, log->log_fh));
__wt_spin_lock(session, &log->log_sync_lock);
locked = true;
/*
* The sync LSN could have advanced while we
* were writing to disk.
*/
if (__wt_log_cmp(
&log->sync_lsn, &min_lsn) <= 0) {
WT_ASSERT(session,
min_lsn.l.file ==
log->sync_lsn.l.file);
log->sync_lsn = min_lsn;
WT_ERR(__wt_cond_signal(
session, log->log_sync_cond));
}
locked = false;
__wt_spin_unlock(session, &log->log_sync_lock);
} else {
WT_ERR(__wt_cond_auto_signal(
session, conn->log_wrlsn_cond));
/*
* We do not want to wait potentially a second
* to process this. Yield to give the wrlsn
* thread a chance to run and try again in
* this case.
*/
__wt_yield();
continue;
}
}
/* Wait until the next event. */
WT_ERR(__wt_cond_wait(
session, conn->log_file_cond, WT_MILLION / 10));
}
if (0) {
err: __wt_err(session, ret, "log close server error");
}
if (locked)
__wt_spin_unlock(session, &log->log_sync_lock);
return (WT_THREAD_RET_VALUE);
}
/*
* __log_archive_once --
* Perform one iteration of log archiving. Must be called with the
* log archive lock held.
*/
static int
__log_archive_once(WT_SESSION_IMPL *session, uint32_t backup_file)
{
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
WT_LOG *log;
uint32_t lognum, min_lognum;
u_int i, logcount;
bool locked;
char **logfiles;
conn = S2C(session);
log = conn->log;
logcount = 0;
logfiles = NULL;
/*
* If we're coming from a backup cursor we want the smaller of
* the last full log file copied in backup or the checkpoint LSN.
* Otherwise we want the minimum of the last log file written to
* disk and the checkpoint LSN.
*/
if (backup_file != 0)
min_lognum = WT_MIN(log->ckpt_lsn.l.file, backup_file);
else
min_lognum = WT_MIN(
log->ckpt_lsn.l.file, log->sync_lsn.l.file);
WT_RET(__wt_verbose(session, WT_VERB_LOG,
"log_archive: archive to log number %" PRIu32, min_lognum));
/*
* Main archive code. Get the list of all log files and
* remove any earlier than the minimum log number.
*/
WT_RET(__wt_dirlist(session, conn->log_path,
WT_LOG_FILENAME, WT_DIRLIST_INCLUDE, &logfiles, &logcount));
/*
* We can only archive files if a hot backup is not in progress or
* if we are the backup.
*/
WT_RET(__wt_readlock(session, conn->hot_backup_lock));
locked = true;
if (!conn->hot_backup || backup_file != 0) {
for (i = 0; i < logcount; i++) {
WT_ERR(__wt_log_extract_lognum(
session, logfiles[i], &lognum));
if (lognum < min_lognum)
WT_ERR(__wt_log_remove(
session, WT_LOG_FILENAME, lognum));
}
}
WT_ERR(__wt_readunlock(session, conn->hot_backup_lock));
locked = false;
__wt_log_files_free(session, logfiles, logcount);
logfiles = NULL;
logcount = 0;
/*
* Indicate what is our new earliest LSN. It is the start
* of the log file containing the last checkpoint.
*/
WT_SET_LSN(&log->first_lsn, min_lognum, 0);
if (0)
err: __wt_err(session, ret, "log archive server error");
if (locked)
WT_TRET(__wt_readunlock(session, conn->hot_backup_lock));
if (logfiles != NULL)
__wt_log_files_free(session, logfiles, logcount);
return (ret);
}
/*
* __log_file_server --
* The log file server thread. This worker thread manages
* log file operations such as closing and syncing.
*/
static WT_THREAD_RET
__log_file_server(void *arg)
{
WT_CONNECTION_IMPL *conn;
WT_DECL_RET;
WT_FH *close_fh;
WT_LOG *log;
WT_LSN close_end_lsn, min_lsn;
WT_SESSION_IMPL *session;
uint64_t yield_count;
uint32_t filenum;
bool locked;
session = arg;
conn = S2C(session);
log = conn->log;
locked = false;
yield_count = 0;
while (F_ISSET(conn, WT_CONN_SERVER_LOG)) {
/*
* If there is a log file to close, make sure any outstanding
* write operations have completed, then fsync and close it.
*/
if ((close_fh = log->log_close_fh) != NULL) {
WT_ERR(__wt_log_extract_lognum(session, close_fh->name,
&filenum));
/*
* The closing file handle should have a correct close
* LSN.
*/
WT_ASSERT(session,
log->log_close_lsn.l.file == filenum);
if (__wt_log_cmp(
&log->write_lsn, &log->log_close_lsn) >= 0) {
/*
* We've copied the file handle, clear out the
* one in the log structure to allow it to be
* set again. Copy the LSN before clearing
* the file handle.
* Use a barrier to make sure the compiler does
* not reorder the following two statements.
*/
close_end_lsn = log->log_close_lsn;
WT_FULL_BARRIER();
log->log_close_fh = NULL;
/*
* Set the close_end_lsn to the LSN immediately
* after ours. That is, the beginning of the
* next log file. We need to know the LSN
* file number of our own close in case earlier
* calls are still in progress and the next one
* to move the sync_lsn into the next file for
* later syncs.
*/
WT_ERR(__wt_fsync(session, close_fh, true));
/*
* We want to have the file size reflect actual
* data with minimal pre-allocated zeroed space.
* We can't truncate the file during hot backup,
* or the underlying file system may not support
* truncate: both are OK, it's just more work
* during cursor traversal.
*/
if (!conn->hot_backup) {
__wt_readlock(
session, &conn->hot_backup_lock);
if (!conn->hot_backup)
WT_ERR_ERROR_OK(
__wt_ftruncate(session,
close_fh,
close_end_lsn.l.offset),
ENOTSUP);
__wt_readunlock(
session, &conn->hot_backup_lock);
}
WT_SET_LSN(&close_end_lsn,
close_end_lsn.l.file + 1, 0);
__wt_spin_lock(session, &log->log_sync_lock);
locked = true;
WT_ERR(__wt_close(session, &close_fh));
WT_ASSERT(session, __wt_log_cmp(
&close_end_lsn, &log->sync_lsn) >= 0);
log->sync_lsn = close_end_lsn;
__wt_cond_signal(session, log->log_sync_cond);
locked = false;
__wt_spin_unlock(session, &log->log_sync_lock);
}
}
/*
* If a later thread asked for a background sync, do it now.
*/
if (__wt_log_cmp(&log->bg_sync_lsn, &log->sync_lsn) > 0) {
/*
* Save the latest write LSN which is the minimum
* we will have written to disk.
*/
min_lsn = log->write_lsn;
/*
* We have to wait until the LSN we asked for is
* written. If it isn't signal the wrlsn thread
* to get it written.
*
* We also have to wait for the written LSN and the
* sync LSN to be in the same file so that we know we
* have synchronized all earlier log files.
*/
if (__wt_log_cmp(&log->bg_sync_lsn, &min_lsn) <= 0) {
/*
* If the sync file is behind either the one
* wanted for a background sync or the write LSN
* has moved to another file continue to let
* this worker thread process that older file
* immediately.
*/
if ((log->sync_lsn.l.file <
log->bg_sync_lsn.l.file) ||
(log->sync_lsn.l.file < min_lsn.l.file))
continue;
WT_ERR(__wt_fsync(session, log->log_fh, true));
__wt_spin_lock(session, &log->log_sync_lock);
locked = true;
/*
* The sync LSN could have advanced while we
* were writing to disk.
*/
if (__wt_log_cmp(
&log->sync_lsn, &min_lsn) <= 0) {
WT_ASSERT(session,
min_lsn.l.file ==
log->sync_lsn.l.file);
log->sync_lsn = min_lsn;
__wt_cond_signal(
session, log->log_sync_cond);
}
locked = false;
__wt_spin_unlock(session, &log->log_sync_lock);
} else {
__wt_cond_signal(session, conn->log_wrlsn_cond);
/*
* We do not want to wait potentially a second
* to process this. Yield to give the wrlsn
* thread a chance to run and try again in
* this case.
*/
yield_count++;
__wt_yield();
continue;
}
}
/* Wait until the next event. */
__wt_cond_wait(session, conn->log_file_cond, 100000, NULL);
}
if (0) {
err: WT_PANIC_MSG(session, ret, "log close server error");
}
WT_STAT_CONN_INCRV(session, log_server_sync_blocked, yield_count);
if (locked)
__wt_spin_unlock(session, &log->log_sync_lock);
return (WT_THREAD_RET_VALUE);
}
