Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Drop slot's LWLock before returning from SaveSlotToPath()
authorPeter Eisentraut <peter@eisentraut.org>
Thu, 26 Mar 2020 10:51:39 +0000 (11:51 +0100)
committerPeter Eisentraut <peter@eisentraut.org>
Thu, 26 Mar 2020 12:31:07 +0000 (13:31 +0100)
When SaveSlotToPath() is called with elevel=LOG, the early exits didn't
release the slot's io_in_progress_lock.

This could result in a walsender being stuck on the lock forever.  A
possible way to get into this situation is if the offending code paths
are triggered in a low disk space situation.

Author: Pavan Deolasee <pavan.deolasee@2ndquadrant.com>
Reported-by: Craig Ringer <craig@2ndquadrant.com>
Discussion: https://www.postgresql.org/message-id/flat/56a138c5-de61-f553-7e8f-6789296de785%402ndquadrant.com

src/backend/replication/slot.c

index f2fd8f336ed5e1a031782aa9c2b51d2518a5b09f..409db4b2eff2f52a91d2ee7d74cb35f18f4ead27 100644 (file)
@@ -1256,6 +1256,12 @@ SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
    fd = OpenTransientFile(tmppath, O_CREAT | O_EXCL | O_WRONLY | PG_BINARY);
    if (fd < 0)
    {
+       /*
+        * If not an ERROR, then release the lock before returning.  In case
+        * of an ERROR, the error recovery path automatically releases the
+        * lock, but no harm in explicitly releasing even in that case.
+        */
+       LWLockRelease(&slot->io_in_progress_lock);
        ereport(elevel,
                (errcode_for_file_access(),
                 errmsg("could not create file \"%s\": %m",
@@ -1287,6 +1293,7 @@ SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
 
        pgstat_report_wait_end();
        CloseTransientFile(fd);
+       LWLockRelease(&slot->io_in_progress_lock);
 
        /* if write didn't set errno, assume problem is no disk space */
        errno = save_errno ? save_errno : ENOSPC;
@@ -1306,6 +1313,7 @@ SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
 
        pgstat_report_wait_end();
        CloseTransientFile(fd);
+       LWLockRelease(&slot->io_in_progress_lock);
        errno = save_errno;
        ereport(elevel,
                (errcode_for_file_access(),
@@ -1317,6 +1325,7 @@ SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
 
    if (CloseTransientFile(fd))
    {
+       LWLockRelease(&slot->io_in_progress_lock);
        ereport(elevel,
                (errcode_for_file_access(),
                 errmsg("could not close file \"%s\": %m",
@@ -1327,6 +1336,7 @@ SaveSlotToPath(ReplicationSlot *slot, const char *dir, int elevel)
    /* rename to permanent file, fsync file and directory */
    if (rename(tmppath, path) != 0)
    {
+       LWLockRelease(&slot->io_in_progress_lock);
        ereport(elevel,
                (errcode_for_file_access(),
                 errmsg("could not rename file \"%s\" to \"%s\": %m",