[XFS] Make the bulkstat_one compat ioctl handling more sane
[safe/jmp/linux-2.6] / fs / xfs / xfs_log.c
index 319b98e..aadaa14 100644 (file)
@@ -100,12 +100,11 @@ STATIC void xlog_ungrant_log_space(xlog_t  *log,
 
 
 /* local ticket functions */
-STATIC xlog_ticket_t   *xlog_ticket_get(xlog_t *log,
+STATIC xlog_ticket_t   *xlog_ticket_alloc(xlog_t *log,
                                         int    unit_bytes,
                                         int    count,
                                         char   clientid,
                                         uint   flags);
-STATIC void            xlog_ticket_put(xlog_t *log, xlog_ticket_t *ticket);
 
 #if defined(DEBUG)
 STATIC void    xlog_verify_dest_ptr(xlog_t *log, __psint_t ptr);
@@ -124,16 +123,27 @@ STATIC void       xlog_verify_tail_lsn(xlog_t *log, xlog_in_core_t *iclog,
 STATIC int     xlog_iclogs_empty(xlog_t *log);
 
 #if defined(XFS_LOG_TRACE)
+
+#define XLOG_TRACE_LOGGRANT_SIZE       2048
+#define XLOG_TRACE_ICLOG_SIZE          256
+
+void
+xlog_trace_loggrant_alloc(xlog_t *log)
+{
+       log->l_grant_trace = ktrace_alloc(XLOG_TRACE_LOGGRANT_SIZE, KM_NOFS);
+}
+
+void
+xlog_trace_loggrant_dealloc(xlog_t *log)
+{
+       ktrace_free(log->l_grant_trace);
+}
+
 void
 xlog_trace_loggrant(xlog_t *log, xlog_ticket_t *tic, xfs_caddr_t string)
 {
        unsigned long cnts;
 
-       if (!log->l_grant_trace) {
-               log->l_grant_trace = ktrace_alloc(2048, KM_NOSLEEP);
-               if (!log->l_grant_trace)
-                       return;
-       }
        /* ticket counts are 1 byte each */
        cnts = ((unsigned long)tic->t_ocnt) | ((unsigned long)tic->t_cnt) << 8;
 
@@ -157,10 +167,20 @@ xlog_trace_loggrant(xlog_t *log, xlog_ticket_t *tic, xfs_caddr_t string)
 }
 
 void
+xlog_trace_iclog_alloc(xlog_in_core_t *iclog)
+{
+       iclog->ic_trace = ktrace_alloc(XLOG_TRACE_ICLOG_SIZE, KM_NOFS);
+}
+
+void
+xlog_trace_iclog_dealloc(xlog_in_core_t *iclog)
+{
+       ktrace_free(iclog->ic_trace);
+}
+
+void
 xlog_trace_iclog(xlog_in_core_t *iclog, uint state)
 {
-       if (!iclog->ic_trace)
-               iclog->ic_trace = ktrace_alloc(256, KM_SLEEP);
        ktrace_enter(iclog->ic_trace,
                     (void *)((unsigned long)state),
                     (void *)((unsigned long)current_pid()),
@@ -170,8 +190,15 @@ xlog_trace_iclog(xlog_in_core_t *iclog, uint state)
                     (void *)NULL, (void *)NULL);
 }
 #else
+
+#define        xlog_trace_loggrant_alloc(log)
+#define        xlog_trace_loggrant_dealloc(log)
 #define        xlog_trace_loggrant(log,tic,string)
+
+#define        xlog_trace_iclog_alloc(iclog)
+#define        xlog_trace_iclog_dealloc(iclog)
 #define        xlog_trace_iclog(iclog,state)
+
 #endif /* XFS_LOG_TRACE */
 
 
@@ -226,20 +253,24 @@ xlog_grant_sub_space(struct log *log, int bytes)
 static void
 xlog_grant_add_space_write(struct log *log, int bytes)
 {
-       log->l_grant_write_bytes += bytes;
-       if (log->l_grant_write_bytes > log->l_logsize) {
-               log->l_grant_write_bytes -= log->l_logsize;
+       int tmp = log->l_logsize - log->l_grant_write_bytes;
+       if (tmp > bytes)
+               log->l_grant_write_bytes += bytes;
+       else {
                log->l_grant_write_cycle++;
+               log->l_grant_write_bytes = bytes - tmp;
        }
 }
 
 static void
 xlog_grant_add_space_reserve(struct log *log, int bytes)
 {
-       log->l_grant_reserve_bytes += bytes;
-       if (log->l_grant_reserve_bytes > log->l_logsize) {
-               log->l_grant_reserve_bytes -= log->l_logsize;
+       int tmp = log->l_logsize - log->l_grant_reserve_bytes;
+       if (tmp > bytes)
+               log->l_grant_reserve_bytes += bytes;
+       else {
                log->l_grant_reserve_cycle++;
+               log->l_grant_reserve_bytes = bytes - tmp;
        }
 }
 
@@ -328,19 +359,16 @@ xfs_log_done(xfs_mount_t  *mp,
                 */
                xlog_trace_loggrant(log, ticket, "xfs_log_done: (non-permanent)");
                xlog_ungrant_log_space(log, ticket);
-               xlog_ticket_put(log, ticket);
+               xfs_log_ticket_put(ticket);
        } else {
                xlog_trace_loggrant(log, ticket, "xfs_log_done: (permanent)");
                xlog_regrant_reserve_log_space(log, ticket);
-       }
-
-       /* If this ticket was a permanent reservation and we aren't
-        * trying to release it, reset the inited flags; so next time
-        * we write, a start record will be written out.
-        */
-       if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) &&
-           (flags & XFS_LOG_REL_PERM_RESERV) == 0)
+               /* If this ticket was a permanent reservation and we aren't
+                * trying to release it, reset the inited flags; so next time
+                * we write, a start record will be written out.
+                */
                ticket->t_flags |= XLOG_TIC_INITED;
+       }
 
        return lsn;
 }      /* xfs_log_done */
@@ -353,11 +381,11 @@ xfs_log_done(xfs_mount_t  *mp,
  * Asynchronous forces are implemented by setting the WANT_SYNC
  * bit in the appropriate in-core log and then returning.
  *
- * Synchronous forces are implemented with a semaphore.  All callers
- * to force a given lsn to disk will wait on a semaphore attached to the
+ * Synchronous forces are implemented with a signal variable. All callers
+ * to force a given lsn to disk will wait on a the sv attached to the
  * specific in-core log.  When given in-core log finally completes its
  * write to disk, that thread will wake up all threads waiting on the
- * semaphore.
+ * sv.
  */
 int
 _xfs_log_force(
@@ -382,7 +410,27 @@ _xfs_log_force(
                return xlog_state_sync_all(log, flags, log_flushed);
        else
                return xlog_state_sync(log, lsn, flags, log_flushed);
-}      /* xfs_log_force */
+}      /* _xfs_log_force */
+
+/*
+ * Wrapper for _xfs_log_force(), to be used when caller doesn't care
+ * about errors or whether the log was flushed or not. This is the normal
+ * interface to use when trying to unpin items or move the log forward.
+ */
+void
+xfs_log_force(
+       xfs_mount_t     *mp,
+       xfs_lsn_t       lsn,
+       uint            flags)
+{
+       int     error;
+       error = _xfs_log_force(mp, lsn, flags, NULL);
+       if (error) {
+               xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: "
+                       "error %d returned.", error);
+       }
+}
+
 
 /*
  * Attaches a new iclog I/O completion callback routine during
@@ -465,7 +513,7 @@ xfs_log_reserve(xfs_mount_t  *mp,
                retval = xlog_regrant_write_log_space(log, internal_ticket);
        } else {
                /* may sleep if need to allocate more tickets */
-               internal_ticket = xlog_ticket_get(log, unit_bytes, cnt,
+               internal_ticket = xlog_ticket_alloc(log, unit_bytes, cnt,
                                                  client, flags);
                if (!internal_ticket)
                        return XFS_ERROR(ENOMEM);
@@ -514,16 +562,21 @@ xfs_log_mount(
        }
 
        mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
+       if (!mp->m_log) {
+               cmn_err(CE_WARN, "XFS: Log allocation failed: No memory!");
+               error = ENOMEM;
+               goto out;
+       }
 
        /*
         * Initialize the AIL now we have a log.
         */
-       spin_lock_init(&mp->m_ail_lock);
        error = xfs_trans_ail_init(mp);
        if (error) {
                cmn_err(CE_WARN, "XFS: AIL initialisation failed: error %d", error);
                goto error;
        }
+       mp->m_log->l_ailp = mp->m_ail;
 
        /*
         * skip log recovery on a norecovery mount.  pretend it all
@@ -552,6 +605,7 @@ xfs_log_mount(
        return 0;
 error:
        xfs_log_unmount_dealloc(mp);
+out:
        return error;
 }      /* xfs_log_mount */
 
@@ -564,12 +618,12 @@ error:
  * mp          - ubiquitous xfs mount point structure
  */
 int
-xfs_log_mount_finish(xfs_mount_t *mp, int mfsi_flags)
+xfs_log_mount_finish(xfs_mount_t *mp)
 {
        int     error;
 
        if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
-               error = xlog_recover_finish(mp->m_log, mfsi_flags);
+               error = xlog_recover_finish(mp->m_log);
        else {
                error = 0;
                ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
@@ -634,7 +688,8 @@ xfs_log_unmount_write(xfs_mount_t *mp)
        if (mp->m_flags & XFS_MOUNT_RDONLY)
                return 0;
 
-       xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC);
+       error = _xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC, NULL);
+       ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
 
 #ifdef DEBUG
        first_iclog = iclog = log->l_iclog;
@@ -676,13 +731,13 @@ xfs_log_unmount_write(xfs_mount_t *mp)
                atomic_inc(&iclog->ic_refcnt);
                spin_unlock(&log->l_icloglock);
                xlog_state_want_sync(log, iclog);
-               (void) xlog_state_release_iclog(log, iclog);
+               error = xlog_state_release_iclog(log, iclog);
 
                spin_lock(&log->l_icloglock);
                if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
                      iclog->ic_state == XLOG_STATE_DIRTY)) {
                        if (!XLOG_FORCED_SHUTDOWN(log)) {
-                               sv_wait(&iclog->ic_forcesema, PMEM,
+                               sv_wait(&iclog->ic_force_wait, PMEM,
                                        &log->l_icloglock, s);
                        } else {
                                spin_unlock(&log->l_icloglock);
@@ -693,7 +748,7 @@ xfs_log_unmount_write(xfs_mount_t *mp)
                if (tic) {
                        xlog_trace_loggrant(log, tic, "unmount rec");
                        xlog_ungrant_log_space(log, tic);
-                       xlog_ticket_put(log, tic);
+                       xfs_log_ticket_put(tic);
                }
        } else {
                /*
@@ -715,7 +770,7 @@ xfs_log_unmount_write(xfs_mount_t *mp)
                spin_unlock(&log->l_icloglock);
 
                xlog_state_want_sync(log, iclog);
-               (void) xlog_state_release_iclog(log, iclog);
+               error =  xlog_state_release_iclog(log, iclog);
 
                spin_lock(&log->l_icloglock);
 
@@ -723,14 +778,14 @@ xfs_log_unmount_write(xfs_mount_t *mp)
                        || iclog->ic_state == XLOG_STATE_DIRTY
                        || iclog->ic_state == XLOG_STATE_IOERROR) ) {
 
-                               sv_wait(&iclog->ic_forcesema, PMEM,
+                               sv_wait(&iclog->ic_force_wait, PMEM,
                                        &log->l_icloglock, s);
                } else {
                        spin_unlock(&log->l_icloglock);
                }
        }
 
-       return 0;
+       return error;
 }      /* xfs_log_unmount_write */
 
 /*
@@ -813,7 +868,7 @@ xfs_log_move_tail(xfs_mount_t       *mp,
                                break;
                        tail_lsn = 0;
                        free_bytes -= tic->t_unit_res;
-                       sv_signal(&tic->t_sema);
+                       sv_signal(&tic->t_wait);
                        tic = tic->t_next;
                } while (tic != log->l_write_headq);
        }
@@ -834,7 +889,7 @@ xfs_log_move_tail(xfs_mount_t       *mp,
                                break;
                        tail_lsn = 0;
                        free_bytes -= need_bytes;
-                       sv_signal(&tic->t_sema);
+                       sv_signal(&tic->t_wait);
                        tic = tic->t_next;
                } while (tic != log->l_reserve_headq);
        }
@@ -850,7 +905,7 @@ xfs_log_move_tail(xfs_mount_t       *mp,
 int
 xfs_log_need_covered(xfs_mount_t *mp)
 {
-       int             needed = 0, gen;
+       int             needed = 0;
        xlog_t          *log = mp->m_log;
 
        if (!xfs_fs_writable(mp))
@@ -859,7 +914,7 @@ xfs_log_need_covered(xfs_mount_t *mp)
        spin_lock(&log->l_icloglock);
        if (((log->l_covered_state == XLOG_STATE_COVER_NEED) ||
                (log->l_covered_state == XLOG_STATE_COVER_NEED2))
-                       && !xfs_trans_first_ail(mp, &gen)
+                       && !xfs_trans_ail_tail(log->l_ailp)
                        && xlog_iclogs_empty(log)) {
                if (log->l_covered_state == XLOG_STATE_COVER_NEED)
                        log->l_covered_state = XLOG_STATE_COVER_DONE;
@@ -896,7 +951,7 @@ xlog_assign_tail_lsn(xfs_mount_t *mp)
        xfs_lsn_t tail_lsn;
        xlog_t    *log = mp->m_log;
 
-       tail_lsn = xfs_trans_tail_ail(mp);
+       tail_lsn = xfs_trans_ail_tail(mp->m_ail);
        spin_lock(&log->l_grant_lock);
        if (tail_lsn != 0) {
                log->l_tail_lsn = tail_lsn;
@@ -974,20 +1029,15 @@ xlog_iodone(xfs_buf_t *bp)
        ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long) 2);
        XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
        aborted = 0;
-
-       /*
-        * Some versions of cpp barf on the recursive definition of
-        * ic_log -> hic_fields.ic_log and expand ic_log twice when
-        * it is passed through two macros.  Workaround broken cpp.
-        */
        l = iclog->ic_log;
 
        /*
-        * If the ordered flag has been removed by a lower
-        * layer, it means the underlyin device no longer supports
+        * If the _XFS_BARRIER_FAILED flag was set by a lower
+        * layer, it means the underlying device no longer supports
         * barrier I/O. Warn loudly and turn off barriers.
         */
-       if ((l->l_mp->m_flags & XFS_MOUNT_BARRIER) && !XFS_BUF_ORDERED(bp)) {
+       if (bp->b_flags & _XFS_BARRIER_FAILED) {
+               bp->b_flags &= ~_XFS_BARRIER_FAILED;
                l->l_mp->m_flags &= ~XFS_MOUNT_BARRIER;
                xfs_fs_cmn_err(CE_WARN, l->l_mp,
                                "xlog_iodone: Barriers are no longer supported"
@@ -1166,7 +1216,9 @@ xlog_alloc_log(xfs_mount_t        *mp,
        int                     i;
        int                     iclogsize;
 
-       log = (xlog_t *)kmem_zalloc(sizeof(xlog_t), KM_SLEEP);
+       log = kmem_zalloc(sizeof(xlog_t), KM_MAYFAIL);
+       if (!log)
+               return NULL;
 
        log->l_mp          = mp;
        log->l_targ        = log_target;
@@ -1198,6 +1250,8 @@ xlog_alloc_log(xfs_mount_t        *mp,
        xlog_get_iclog_buffer_size(mp, log);
 
        bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp);
+       if (!bp)
+               goto out_free_log;
        XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
        XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
        XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
@@ -1207,8 +1261,9 @@ xlog_alloc_log(xfs_mount_t        *mp,
 
        spin_lock_init(&log->l_icloglock);
        spin_lock_init(&log->l_grant_lock);
-       initnsema(&log->l_flushsema, 0, "ic-flush");
+       sv_init(&log->l_flush_wait, 0, "flush_wait");
 
+       xlog_trace_loggrant_alloc(log);
        /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
        ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0);
 
@@ -1223,20 +1278,24 @@ xlog_alloc_log(xfs_mount_t      *mp,
        iclogsize = log->l_iclog_size;
        ASSERT(log->l_iclog_size >= 4096);
        for (i=0; i < log->l_iclog_bufs; i++) {
-               *iclogp = (xlog_in_core_t *)
-                         kmem_zalloc(sizeof(xlog_in_core_t), KM_SLEEP);
+               *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
+               if (!*iclogp)
+                       goto out_free_iclog;
+
                iclog = *iclogp;
                iclog->ic_prev = prev_iclog;
                prev_iclog = iclog;
 
                bp = xfs_buf_get_noaddr(log->l_iclog_size, mp->m_logdev_targp);
+               if (!bp)
+                       goto out_free_iclog;
                if (!XFS_BUF_CPSEMA(bp))
                        ASSERT(0);
                XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
                XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
                XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
                iclog->ic_bp = bp;
-               iclog->hic_data = bp->b_addr;
+               iclog->ic_data = bp->b_addr;
 #ifdef DEBUG
                log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
 #endif
@@ -1256,12 +1315,14 @@ xlog_alloc_log(xfs_mount_t      *mp,
                atomic_set(&iclog->ic_refcnt, 0);
                spin_lock_init(&iclog->ic_callback_lock);
                iclog->ic_callback_tail = &(iclog->ic_callback);
-               iclog->ic_datap = (char *)iclog->hic_data + log->l_iclog_hsize;
+               iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
 
                ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp));
                ASSERT(XFS_BUF_VALUSEMA(iclog->ic_bp) <= 0);
-               sv_init(&iclog->ic_forcesema, SV_DEFAULT, "iclog-force");
-               sv_init(&iclog->ic_writesema, SV_DEFAULT, "iclog-write");
+               sv_init(&iclog->ic_force_wait, SV_DEFAULT, "iclog-force");
+               sv_init(&iclog->ic_write_wait, SV_DEFAULT, "iclog-write");
+
+               xlog_trace_iclog_alloc(iclog);
 
                iclogp = &iclog->ic_next;
        }
@@ -1269,6 +1330,25 @@ xlog_alloc_log(xfs_mount_t       *mp,
        log->l_iclog->ic_prev = prev_iclog;     /* re-write 1st prev ptr */
 
        return log;
+
+out_free_iclog:
+       for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
+               prev_iclog = iclog->ic_next;
+               if (iclog->ic_bp) {
+                       sv_destroy(&iclog->ic_force_wait);
+                       sv_destroy(&iclog->ic_write_wait);
+                       xfs_buf_free(iclog->ic_bp);
+                       xlog_trace_iclog_dealloc(iclog);
+               }
+               kmem_free(iclog);
+       }
+       spinlock_destroy(&log->l_icloglock);
+       spinlock_destroy(&log->l_grant_lock);
+       xlog_trace_loggrant_dealloc(log);
+       xfs_buf_free(log->l_xbuf);
+out_free_log:
+       kmem_free(log);
+       return NULL;
 }      /* xlog_alloc_log */
 
 
@@ -1359,7 +1439,7 @@ xlog_grant_push_ail(xfs_mount_t   *mp,
      */
     if (threshold_lsn &&
        !XLOG_FORCED_SHUTDOWN(log))
-           xfs_trans_push_ail(mp, threshold_lsn);
+           xfs_trans_ail_push(log->l_ailp, threshold_lsn);
 }      /* xlog_grant_push_ail */
 
 
@@ -1540,33 +1620,21 @@ xlog_dealloc_log(xlog_t *log)
 
        iclog = log->l_iclog;
        for (i=0; i<log->l_iclog_bufs; i++) {
-               sv_destroy(&iclog->ic_forcesema);
-               sv_destroy(&iclog->ic_writesema);
+               sv_destroy(&iclog->ic_force_wait);
+               sv_destroy(&iclog->ic_write_wait);
                xfs_buf_free(iclog->ic_bp);
-#ifdef XFS_LOG_TRACE
-               if (iclog->ic_trace != NULL) {
-                       ktrace_free(iclog->ic_trace);
-               }
-#endif
+               xlog_trace_iclog_dealloc(iclog);
                next_iclog = iclog->ic_next;
-               kmem_free(iclog, sizeof(xlog_in_core_t));
+               kmem_free(iclog);
                iclog = next_iclog;
        }
-       freesema(&log->l_flushsema);
        spinlock_destroy(&log->l_icloglock);
        spinlock_destroy(&log->l_grant_lock);
 
        xfs_buf_free(log->l_xbuf);
-#ifdef XFS_LOG_TRACE
-       if (log->l_trace != NULL) {
-               ktrace_free(log->l_trace);
-       }
-       if (log->l_grant_trace != NULL) {
-               ktrace_free(log->l_grant_trace);
-       }
-#endif
+       xlog_trace_loggrant_dealloc(log);
        log->l_mp->m_log = NULL;
-       kmem_free(log, sizeof(xlog_t));
+       kmem_free(log);
 }      /* xlog_dealloc_log */
 
 /*
@@ -1952,7 +2020,7 @@ xlog_write(xfs_mount_t *  mp,
 /* Clean iclogs starting from the head.  This ordering must be
  * maintained, so an iclog doesn't become ACTIVE beyond one that
  * is SYNCING.  This is also required to maintain the notion that we use
- * a counting semaphore to hold off would be writers to the log when every
+ * a ordered wait queue to hold off would be writers to the log when every
  * iclog is trying to sync to disk.
  *
  * State Change: DIRTY -> ACTIVE
@@ -2076,6 +2144,7 @@ xlog_state_do_callback(
        int                funcdidcallbacks; /* flag: function did callbacks */
        int                repeats;     /* for issuing console warnings if
                                         * looping too many times */
+       int                wake = 0;
 
        spin_lock(&log->l_icloglock);
        first_iclog = iclog = log->l_iclog;
@@ -2215,7 +2284,7 @@ xlog_state_do_callback(
                        xlog_state_clean_log(log);
 
                        /* wake up threads waiting in xfs_log_force() */
-                       sv_broadcast(&iclog->ic_forcesema);
+                       sv_broadcast(&iclog->ic_force_wait);
 
                        iclog = iclog->ic_next;
                } while (first_iclog != iclog);
@@ -2225,7 +2294,7 @@ xlog_state_do_callback(
                        repeats = 0;
                        xfs_fs_cmn_err(CE_WARN, log->l_mp,
                                "%s: possible infinite loop (%d iterations)",
-                               __FUNCTION__, flushcnt);
+                               __func__, flushcnt);
                }
        } while (!ioerrors && loopdidcallbacks);
 
@@ -2257,15 +2326,13 @@ xlog_state_do_callback(
        }
 #endif
 
-       flushcnt = 0;
-       if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) {
-               flushcnt = log->l_flushcnt;
-               log->l_flushcnt = 0;
-       }
+       if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
+               wake = 1;
        spin_unlock(&log->l_icloglock);
-       while (flushcnt--)
-               vsema(&log->l_flushsema);
-}      /* xlog_state_do_callback */
+
+       if (wake)
+               sv_broadcast(&log->l_flush_wait);
+}
 
 
 /*
@@ -2279,8 +2346,7 @@ xlog_state_do_callback(
  * the second completion goes through.
  *
  * Callbacks could take time, so they are done outside the scope of the
- * global state machine log lock.  Assume that the calls to cvsema won't
- * take a long time.  At least we know it won't sleep.
+ * global state machine log lock.
  */
 STATIC void
 xlog_state_done_syncing(
@@ -2316,7 +2382,7 @@ xlog_state_done_syncing(
         * iclog buffer, we wake them all, one will get to do the
         * I/O, the others get to wait for the result.
         */
-       sv_broadcast(&iclog->ic_writesema);
+       sv_broadcast(&iclog->ic_write_wait);
        spin_unlock(&log->l_icloglock);
        xlog_state_do_callback(log, aborted, iclog);    /* also cleans log */
 }      /* xlog_state_done_syncing */
@@ -2324,11 +2390,9 @@ xlog_state_done_syncing(
 
 /*
  * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
- * sleep.  The flush semaphore is set to the number of in-core buffers and
- * decremented around disk syncing.  Therefore, if all buffers are syncing,
- * this semaphore will cause new writes to sleep until a sync completes.
- * Otherwise, this code just does p() followed by v().  This approximates
- * a sleep/wakeup except we can't race.
+ * sleep.  We wait on the flush queue on the head iclog as that should be
+ * the first iclog to complete flushing. Hence if all iclogs are syncing,
+ * we will wait here and all new writes will sleep until a sync completes.
  *
  * The in-core logs are used in a circular fashion. They are not used
  * out-of-order even when an iclog past the head is free.
@@ -2363,16 +2427,15 @@ restart:
        }
 
        iclog = log->l_iclog;
-       if (! (iclog->ic_state == XLOG_STATE_ACTIVE)) {
-               log->l_flushcnt++;
-               spin_unlock(&log->l_icloglock);
+       if (iclog->ic_state != XLOG_STATE_ACTIVE) {
                xlog_trace_iclog(iclog, XLOG_TRACE_SLEEP_FLUSH);
                XFS_STATS_INC(xs_log_noiclogs);
-               /* Ensure that log writes happen */
-               psema(&log->l_flushsema, PINOD);
+
+               /* Wait for log writes to have flushed */
+               sv_wait(&log->l_flush_wait, 0, &log->l_icloglock, 0);
                goto restart;
        }
-       ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
+
        head = &iclog->ic_header;
 
        atomic_inc(&iclog->ic_refcnt);  /* prevents sync */
@@ -2406,13 +2469,20 @@ restart:
        if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
                xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
 
-               /* If I'm the only one writing to this iclog, sync it to disk */
-               if (atomic_read(&iclog->ic_refcnt) == 1) {
+               /*
+                * If I'm the only one writing to this iclog, sync it to disk.
+                * We need to do an atomic compare and decrement here to avoid
+                * racing with concurrent atomic_dec_and_lock() calls in
+                * xlog_state_release_iclog() when there is more than one
+                * reference to the iclog.
+                */
+               if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
+                       /* we are the only one */
                        spin_unlock(&log->l_icloglock);
-                       if ((error = xlog_state_release_iclog(log, iclog)))
+                       error = xlog_state_release_iclog(log, iclog);
+                       if (error)
                                return error;
                } else {
-                       atomic_dec(&iclog->ic_refcnt);
                        spin_unlock(&log->l_icloglock);
                }
                goto restart;
@@ -2479,7 +2549,7 @@ xlog_grant_log_space(xlog_t          *log,
                        goto error_return;
 
                XFS_STATS_INC(xs_sleep_logspace);
-               sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
+               sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
                /*
                 * If we got an error, and the filesystem is shutting down,
                 * we'll catch it down below. So just continue...
@@ -2505,7 +2575,7 @@ redo:
                xlog_trace_loggrant(log, tic,
                                    "xlog_grant_log_space: sleep 2");
                XFS_STATS_INC(xs_sleep_logspace);
-               sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
+               sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
 
                if (XLOG_FORCED_SHUTDOWN(log)) {
                        spin_lock(&log->l_grant_lock);
@@ -2604,7 +2674,7 @@ xlog_regrant_write_log_space(xlog_t          *log,
                        if (free_bytes < ntic->t_unit_res)
                                break;
                        free_bytes -= ntic->t_unit_res;
-                       sv_signal(&ntic->t_sema);
+                       sv_signal(&ntic->t_wait);
                        ntic = ntic->t_next;
                } while (ntic != log->l_write_headq);
 
@@ -2615,7 +2685,7 @@ xlog_regrant_write_log_space(xlog_t          *log,
                        xlog_trace_loggrant(log, tic,
                                    "xlog_regrant_write_log_space: sleep 1");
                        XFS_STATS_INC(xs_sleep_logspace);
-                       sv_wait(&tic->t_sema, PINOD|PLTWAIT,
+                       sv_wait(&tic->t_wait, PINOD|PLTWAIT,
                                &log->l_grant_lock, s);
 
                        /* If we're shutting down, this tic is already
@@ -2644,7 +2714,7 @@ redo:
                if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
                        xlog_ins_ticketq(&log->l_write_headq, tic);
                XFS_STATS_INC(xs_sleep_logspace);
-               sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
+               sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
 
                /* If we're shutting down, this tic is already off the queue */
                if (XLOG_FORCED_SHUTDOWN(log)) {
@@ -2887,7 +2957,7 @@ xlog_state_switch_iclogs(xlog_t           *log,
  *     2. the current iclog is drity, and the previous iclog is in the
  *             active or dirty state.
  *
- * We may sleep (call psema) if:
+ * We may sleep if:
  *
  *     1. the current iclog is not in the active nor dirty state.
  *     2. the current iclog dirty, and the previous iclog is not in the
@@ -2984,7 +3054,7 @@ maybe_sleep:
                        return XFS_ERROR(EIO);
                }
                XFS_STATS_INC(xs_log_force_sleep);
-               sv_wait(&iclog->ic_forcesema, PINOD, &log->l_icloglock, s);
+               sv_wait(&iclog->ic_force_wait, PINOD, &log->l_icloglock, s);
                /*
                 * No need to grab the log lock here since we're
                 * only deciding whether or not to return EIO
@@ -3067,7 +3137,7 @@ try_again:
                                                 XLOG_STATE_SYNCING))) {
                        ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
                        XFS_STATS_INC(xs_log_force_sleep);
-                       sv_wait(&iclog->ic_prev->ic_writesema, PSWP,
+                       sv_wait(&iclog->ic_prev->ic_write_wait, PSWP,
                                &log->l_icloglock, s);
                        *log_flushed = 1;
                        already_slept = 1;
@@ -3087,7 +3157,7 @@ try_again:
            !(iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
 
                /*
-                * Don't wait on the forcesema if we know that we've
+                * Don't wait on completion if we know that we've
                 * gotten a log write error.
                 */
                if (iclog->ic_state & XLOG_STATE_IOERROR) {
@@ -3095,7 +3165,7 @@ try_again:
                        return XFS_ERROR(EIO);
                }
                XFS_STATS_INC(xs_log_force_sleep);
-               sv_wait(&iclog->ic_forcesema, PSWP, &log->l_icloglock, s);
+               sv_wait(&iclog->ic_force_wait, PSWP, &log->l_icloglock, s);
                /*
                 * No need to grab the log lock here since we're
                 * only deciding whether or not to return EIO
@@ -3145,22 +3215,33 @@ xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog)
  */
 
 /*
- * Free a used ticket.
+ * Free a used ticket when it's refcount falls to zero.
  */
-STATIC void
-xlog_ticket_put(xlog_t         *log,
-               xlog_ticket_t   *ticket)
+void
+xfs_log_ticket_put(
+       xlog_ticket_t   *ticket)
 {
-       sv_destroy(&ticket->t_sema);
-       kmem_zone_free(xfs_log_ticket_zone, ticket);
-}      /* xlog_ticket_put */
+       ASSERT(atomic_read(&ticket->t_ref) > 0);
+       if (atomic_dec_and_test(&ticket->t_ref)) {
+               sv_destroy(&ticket->t_wait);
+               kmem_zone_free(xfs_log_ticket_zone, ticket);
+       }
+}
 
+xlog_ticket_t *
+xfs_log_ticket_get(
+       xlog_ticket_t   *ticket)
+{
+       ASSERT(atomic_read(&ticket->t_ref) > 0);
+       atomic_inc(&ticket->t_ref);
+       return ticket;
+}
 
 /*
  * Allocate and initialise a new log ticket.
  */
 STATIC xlog_ticket_t *
-xlog_ticket_get(xlog_t         *log,
+xlog_ticket_alloc(xlog_t               *log,
                int             unit_bytes,
                int             cnt,
                char            client,
@@ -3231,6 +3312,7 @@ xlog_ticket_get(xlog_t            *log,
                unit_bytes += 2*BBSIZE;
         }
 
+       atomic_set(&tic->t_ref, 1);
        tic->t_unit_res         = unit_bytes;
        tic->t_curr_res         = unit_bytes;
        tic->t_cnt              = cnt;
@@ -3241,12 +3323,12 @@ xlog_ticket_get(xlog_t          *log,
        tic->t_trans_type       = 0;
        if (xflags & XFS_LOG_PERM_RESERV)
                tic->t_flags |= XLOG_TIC_PERM_RESERV;
-       sv_init(&(tic->t_sema), SV_DEFAULT, "logtick");
+       sv_init(&(tic->t_wait), SV_DEFAULT, "logtick");
 
        xlog_tic_reset_res(tic);
 
        return tic;
-}      /* xlog_ticket_get */
+}
 
 
 /******************************************************************************
@@ -3375,7 +3457,7 @@ xlog_verify_iclog(xlog_t   *log,
        ptr = iclog->ic_datap;
        base_ptr = ptr;
        ophead = (xlog_op_header_t *)ptr;
-       xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
+       xhdr = iclog->ic_data;
        for (i = 0; i < len; i++) {
                ophead = (xlog_op_header_t *)ptr;
 
@@ -3481,7 +3563,8 @@ xfs_log_force_umount(
        if (!log ||
            log->l_flags & XLOG_ACTIVE_RECOVERY) {
                mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
-               XFS_BUF_DONE(mp->m_sb_bp);
+               if (mp->m_sb_bp)
+                       XFS_BUF_DONE(mp->m_sb_bp);
                return 0;
        }
 
@@ -3499,10 +3582,12 @@ xfs_log_force_umount(
         * before we mark the filesystem SHUTDOWN and wake
         * everybody up to tell the bad news.
         */
-       spin_lock(&log->l_grant_lock);
        spin_lock(&log->l_icloglock);
+       spin_lock(&log->l_grant_lock);
        mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
-       XFS_BUF_DONE(mp->m_sb_bp);
+       if (mp->m_sb_bp)
+               XFS_BUF_DONE(mp->m_sb_bp);
+
        /*
         * This flag is sort of redundant because of the mount flag, but
         * it's good to maintain the separation between the log and the rest
@@ -3528,14 +3613,14 @@ xfs_log_force_umount(
         */
        if ((tic = log->l_reserve_headq)) {
                do {
-                       sv_signal(&tic->t_sema);
+                       sv_signal(&tic->t_wait);
                        tic = tic->t_next;
                } while (tic != log->l_reserve_headq);
        }
 
        if ((tic = log->l_write_headq)) {
                do {
-                       sv_signal(&tic->t_sema);
+                       sv_signal(&tic->t_wait);
                        tic = tic->t_next;
                } while (tic != log->l_write_headq);
        }