src/patches/suse-2.6.27.25/patches.suse/dlm-clear-defunct-cancel-state.patch

   1 From: David Teigland <teigland@redhat.com>
   2 commit 43279e5376017c40b4be9af5bc79cbb4ef6f53d7
   3 Author: David Teigland <teigland@redhat.com>
   4 Date:   Wed Jan 28 14:37:54 2009 -0600
   5 Subject: dlm: clear defunct cancel state
   6
   7     When a conversion completes successfully and finds that a cancel
   8     of the convert is still in progress (which is now a moot point),
   9     preemptively clear the state associated with outstanding cancel.
  10     That state could cause a subsequent conversion to be ignored.
  11
  12     Also, improve the consistency and content of error and debug
  13     messages in this area.
  14
  15 Signed-off-by: David Teigland <teigland@redhat.com>
  16 Signed-off-by: Coly Li <coly.li@suse.de>
  17
  18 diff --git a/fs/dlm/lock.c b/fs/dlm/lock.c
  19 index 01e7d39..8cb9204 100644
  20 --- a/fs/dlm/lock.c
  21 +++ b/fs/dlm/lock.c
  22 @@ -835,7 +835,7 @@ static int add_to_waiters(struct dlm_lkb *lkb, int mstype)
  23                 lkb->lkb_wait_count++;
  24                 hold_lkb(lkb);
  25
  26 -               log_debug(ls, "add overlap %x cur %d new %d count %d flags %x",
  27 +               log_debug(ls, "addwait %x cur %d overlap %d count %d f %x",
  28                           lkb->lkb_id, lkb->lkb_wait_type, mstype,
  29                           lkb->lkb_wait_count, lkb->lkb_flags);
  30                 goto out;
  31 @@ -851,7 +851,7 @@ static int add_to_waiters(struct dlm_lkb *lkb, int mstype)
  32         list_add(&lkb->lkb_wait_reply, &ls->ls_waiters);
  33   out:
  34         if (error)
  35 -               log_error(ls, "add_to_waiters %x error %d flags %x %d %d %s",
  36 +               log_error(ls, "addwait error %x %d flags %x %d %d %s",
  37                           lkb->lkb_id, error, lkb->lkb_flags, mstype,
  38                           lkb->lkb_wait_type, lkb->lkb_resource->res_name);
  39         mutex_unlock(&ls->ls_waiters_mutex);
  40 @@ -863,23 +863,55 @@ static int add_to_waiters(struct dlm_lkb *lkb, int mstype)
  41     request reply on the requestqueue) between dlm_recover_waiters_pre() which
  42     set RESEND and dlm_recover_waiters_post() */
  43
  44 -static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype)
  45 +static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype,
  46 +                               struct dlm_message *ms)
  47  {
  48         struct dlm_ls *ls = lkb->lkb_resource->res_ls;
  49         int overlap_done = 0;
  50
  51         if (is_overlap_unlock(lkb) && (mstype == DLM_MSG_UNLOCK_REPLY)) {
  52 +               log_debug(ls, "remwait %x unlock_reply overlap", lkb->lkb_id);
  53                 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_UNLOCK;
  54                 overlap_done = 1;
  55                 goto out_del;
  56         }
  57
  58         if (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL_REPLY)) {
  59 +               log_debug(ls, "remwait %x cancel_reply overlap", lkb->lkb_id);
  60                 lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
  61                 overlap_done = 1;
  62                 goto out_del;
  63         }
  64
  65 +       /* Cancel state was preemptively cleared by a successful convert,
  66 +          see next comment, nothing to do. */
  67 +
  68 +       if ((mstype == DLM_MSG_CANCEL_REPLY) &&
  69 +           (lkb->lkb_wait_type != DLM_MSG_CANCEL)) {
  70 +               log_debug(ls, "remwait %x cancel_reply wait_type %d",
  71 +                         lkb->lkb_id, lkb->lkb_wait_type);
  72 +               return -1;
  73 +       }
  74 +
  75 +       /* Remove for the convert reply, and premptively remove for the
  76 +          cancel reply.  A convert has been granted while there's still
  77 +          an outstanding cancel on it (the cancel is moot and the result
  78 +          in the cancel reply should be 0).  We preempt the cancel reply
  79 +          because the app gets the convert result and then can follow up
  80 +          with another op, like convert.  This subsequent op would see the
  81 +          lingering state of the cancel and fail with -EBUSY. */
  82 +
  83 +       if ((mstype == DLM_MSG_CONVERT_REPLY) &&
  84 +           (lkb->lkb_wait_type == DLM_MSG_CONVERT) &&
  85 +           is_overlap_cancel(lkb) && ms && !ms->m_result) {
  86 +               log_debug(ls, "remwait %x convert_reply zap overlap_cancel",
  87 +                         lkb->lkb_id);
  88 +               lkb->lkb_wait_type = 0;
  89 +               lkb->lkb_flags &= ~DLM_IFL_OVERLAP_CANCEL;
  90 +               lkb->lkb_wait_count--;
  91 +               goto out_del;
  92 +       }
  93 +
  94         /* N.B. type of reply may not always correspond to type of original
  95            msg due to lookup->request optimization, verify others? */
  96
  97 @@ -888,8 +920,8 @@ static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype)
  98                 goto out_del;
  99         }
 100
 101 -       log_error(ls, "remove_from_waiters lkid %x flags %x types %d %d",
 102 -                 lkb->lkb_id, lkb->lkb_flags, mstype, lkb->lkb_wait_type);
 103 +       log_error(ls, "remwait error %x reply %d flags %x no wait_type",
 104 +                 lkb->lkb_id, mstype, lkb->lkb_flags);
 105         return -1;
 106
 107   out_del:
 108 @@ -899,7 +931,7 @@ static int _remove_from_waiters(struct dlm_lkb *lkb, int mstype)
 109            this would happen */
 110
 111         if (overlap_done && lkb->lkb_wait_type) {
 112 -               log_error(ls, "remove_from_waiters %x reply %d give up on %d",
 113 +               log_error(ls, "remwait error %x reply %d wait_type %d overlap",
 114                           lkb->lkb_id, mstype, lkb->lkb_wait_type);
 115                 lkb->lkb_wait_count--;
 116                 lkb->lkb_wait_type = 0;
 117 @@ -921,7 +953,7 @@ static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
 118         int error;
 119
 120         mutex_lock(&ls->ls_waiters_mutex);
 121 -       error = _remove_from_waiters(lkb, mstype);
 122 +       error = _remove_from_waiters(lkb, mstype, NULL);
 123         mutex_unlock(&ls->ls_waiters_mutex);
 124         return error;
 125  }
 126 @@ -936,7 +968,7 @@ static int remove_from_waiters_ms(struct dlm_lkb *lkb, struct dlm_message *ms)
 127
 128         if (ms != &ls->ls_stub_ms)
 129                 mutex_lock(&ls->ls_waiters_mutex);
 130 -       error = _remove_from_waiters(lkb, ms->m_type);
 131 +       error = _remove_from_waiters(lkb, ms->m_type, ms);
 132         if (ms != &ls->ls_stub_ms)
 133                 mutex_unlock(&ls->ls_waiters_mutex);
 134         return error;
 135 @@ -2083,6 +2115,11 @@ static int validate_lock_args(struct dlm_ls *ls, struct dlm_lkb *lkb,
 136         lkb->lkb_timeout_cs = args->timeout;
 137         rv = 0;
 138   out:
 139 +       if (rv)
 140 +               log_debug(ls, "validate_lock_args %d %x %x %x %d %d %s",
 141 +                         rv, lkb->lkb_id, lkb->lkb_flags, args->flags,
 142 +                         lkb->lkb_status, lkb->lkb_wait_type,
 143 +                         lkb->lkb_resource->res_name);
 144         return rv;
 145  }
 146