Merge branch 'for-v5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...

[linux-2.6-microblaze.git] / fs / xfs / xfs_log.c

diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
index 1c328ef..60ac5fd 100644 (file)
--- a/fs/xfs/xfs_log.c
+++ b/fs/xfs/xfs_log.c
@@ -78,13 +78,12 @@ xlog_verify_iclog(
 STATIC void
 xlog_verify_tail_lsn(
        struct xlog             *log,
-       struct xlog_in_core     *iclog,
-       xfs_lsn_t               tail_lsn);
+       struct xlog_in_core     *iclog);
 #else
 #define xlog_verify_dest_ptr(a,b)
 #define xlog_verify_grant_tail(a)
 #define xlog_verify_iclog(a,b,c)
-#define xlog_verify_tail_lsn(a,b,c)
+#define xlog_verify_tail_lsn(a,b)
 #endif
 
 STATIC int
@@ -489,12 +488,31 @@ out_error:
 
 /*
  * Flush iclog to disk if this is the last reference to the given iclog and the
- * it is in the WANT_SYNC state.  If the caller passes in a non-zero
- * @old_tail_lsn and the current log tail does not match, there may be metadata
- * on disk that must be persisted before this iclog is written.  To satisfy that
- * requirement, set the XLOG_ICL_NEED_FLUSH flag as a condition for writing this
- * iclog with the new log tail value.
+ * it is in the WANT_SYNC state.
+ *
+ * If the caller passes in a non-zero @old_tail_lsn and the current log tail
+ * does not match, there may be metadata on disk that must be persisted before
+ * this iclog is written.  To satisfy that requirement, set the
+ * XLOG_ICL_NEED_FLUSH flag as a condition for writing this iclog with the new
+ * log tail value.
+ *
+ * If XLOG_ICL_NEED_FUA is already set on the iclog, we need to ensure that the
+ * log tail is updated correctly. NEED_FUA indicates that the iclog will be
+ * written to stable storage, and implies that a commit record is contained
+ * within the iclog. We need to ensure that the log tail does not move beyond
+ * the tail that the first commit record in the iclog ordered against, otherwise
+ * correct recovery of that checkpoint becomes dependent on future operations
+ * performed on this iclog.
+ *
+ * Hence if NEED_FUA is set and the current iclog tail lsn is empty, write the
+ * current tail into iclog. Once the iclog tail is set, future operations must
+ * not modify it, otherwise they potentially violate ordering constraints for
+ * the checkpoint commit that wrote the initial tail lsn value. The tail lsn in
+ * the iclog will get zeroed on activation of the iclog after sync, so we
+ * always capture the tail lsn on the iclog on the first NEED_FUA release
+ * regardless of the number of active reference counts on this iclog.
  */
+
 int
 xlog_state_release_iclog(
        struct xlog             *log,
@@ -519,6 +537,10 @@ xlog_state_release_iclog(
 
                if (old_tail_lsn && tail_lsn != old_tail_lsn)
                        iclog->ic_flags |= XLOG_ICL_NEED_FLUSH;
+
+               if ((iclog->ic_flags & XLOG_ICL_NEED_FUA) &&
+                   !iclog->ic_header.h_tail_lsn)
+                       iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
        }
 
        if (!atomic_dec_and_test(&iclog->ic_refcnt))
@@ -530,8 +552,9 @@ xlog_state_release_iclog(
        }
 
        iclog->ic_state = XLOG_STATE_SYNCING;
-       iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
-       xlog_verify_tail_lsn(log, iclog, tail_lsn);
+       if (!iclog->ic_header.h_tail_lsn)
+               iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
+       xlog_verify_tail_lsn(log, iclog);
        trace_xlog_iclog_syncing(iclog, _RET_IP_);
 
        spin_unlock(&log->l_icloglock);
@@ -2579,6 +2602,7 @@ xlog_state_activate_iclog(
        memset(iclog->ic_header.h_cycle_data, 0,
                sizeof(iclog->ic_header.h_cycle_data));
        iclog->ic_header.h_lsn = 0;
+       iclog->ic_header.h_tail_lsn = 0;
 }
 
 /*
@@ -3614,10 +3638,10 @@ xlog_verify_grant_tail(
 STATIC void
 xlog_verify_tail_lsn(
        struct xlog             *log,
-       struct xlog_in_core     *iclog,
-       xfs_lsn_t               tail_lsn)
+       struct xlog_in_core     *iclog)
 {
-    int blocks;
+       xfs_lsn_t       tail_lsn = be64_to_cpu(iclog->ic_header.h_tail_lsn);
+       int             blocks;
 
     if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
        blocks =