- 2 patches from upstream
authorArkadiusz Miśkiewicz <arekm@maven.pl>
Mon, 5 Jan 2009 20:45:32 +0000 (20:45 +0000)
committercvs2git <feedback@pld-linux.org>
Sun, 24 Jun 2012 12:13:13 +0000 (12:13 +0000)
Changed files:
    patch.4.7.25.2 -> 1.1
    patch.4.7.25.3 -> 1.1

patch.4.7.25.2 [new file with mode: 0644]
patch.4.7.25.3 [new file with mode: 0644]

diff --git a/patch.4.7.25.2 b/patch.4.7.25.2
new file mode 100644 (file)
index 0000000..2e4a7ec
--- /dev/null
@@ -0,0 +1,71 @@
+Index: lock/lock.c
+===================================================================
+RCS file: /a/CVSROOT/db/lock/lock.c,v
+retrieving revision 12.61
+diff -c -r12.61 lock.c
+*** lock/lock.c        22 Jul 2008 12:08:53 -0000      12.61
+--- lock/lock.c        19 Aug 2008 17:28:24 -0000
+***************
+*** 1278,1287 ****
+               SH_TAILQ_REMOVE(
+                   &lt->obj_tab[obj_ndx], sh_obj, links, __db_lockobj);
+               if (sh_obj->lockobj.size > sizeof(sh_obj->objdata)) {
+!                      LOCK_REGION_LOCK(env);
+                       __env_alloc_free(&lt->reginfo,
+                           SH_DBT_PTR(&sh_obj->lockobj));
+!                      LOCK_REGION_UNLOCK(env);
+               }
+               SH_TAILQ_INSERT_HEAD(
+                   &FREE_OBJS(lt, part_id), sh_obj, links, __db_lockobj);
+--- 1278,1289 ----
+               SH_TAILQ_REMOVE(
+                   &lt->obj_tab[obj_ndx], sh_obj, links, __db_lockobj);
+               if (sh_obj->lockobj.size > sizeof(sh_obj->objdata)) {
+!                      if (region->part_t_size != 1)
+!                              LOCK_REGION_LOCK(env);
+                       __env_alloc_free(&lt->reginfo,
+                           SH_DBT_PTR(&sh_obj->lockobj));
+!                      if (region->part_t_size != 1)
+!                              LOCK_REGION_UNLOCK(env);
+               }
+               SH_TAILQ_INSERT_HEAD(
+                   &FREE_OBJS(lt, part_id), sh_obj, links, __db_lockobj);
+***************
+*** 1470,1484 ****
+               if (obj->size <= sizeof(sh_obj->objdata))
+                       p = sh_obj->objdata;
+               else {
+!                      LOCK_REGION_LOCK(env);
+                       if ((ret =
+                           __env_alloc(&lt->reginfo, obj->size, &p)) != 0) {
+                               __db_errx(env,
+                                   "No space for lock object storage");
+!                              LOCK_REGION_UNLOCK(env);
+                               goto err;
+                       }
+!                      LOCK_REGION_UNLOCK(env);
+               }
+  
+               memcpy(p, obj->data, obj->size);
+--- 1472,1492 ----
+               if (obj->size <= sizeof(sh_obj->objdata))
+                       p = sh_obj->objdata;
+               else {
+!                      /*
+!                       * If we have only one partition, the region is locked.
+!                       */
+!                      if (region->part_t_size != 1)
+!                              LOCK_REGION_LOCK(env);
+                       if ((ret =
+                           __env_alloc(&lt->reginfo, obj->size, &p)) != 0) {
+                               __db_errx(env,
+                                   "No space for lock object storage");
+!                              if (region->part_t_size != 1)
+!                                      LOCK_REGION_UNLOCK(env);
+                               goto err;
+                       }
+!                      if (region->part_t_size != 1)
+!                              LOCK_REGION_UNLOCK(env);
+               }
+  
+               memcpy(p, obj->data, obj->size);
diff --git a/patch.4.7.25.3 b/patch.4.7.25.3
new file mode 100644 (file)
index 0000000..316d615
--- /dev/null
@@ -0,0 +1,314 @@
+*** lock/lock_deadlock.c       2008-03-11 00:31:33.000000000 +1100
+--- lock/lock_deadlock.c       2008-12-16 21:54:18.000000000 +1100
+***************
+*** 121,127 ****
+       DB_LOCKTAB *lt;
+       db_timespec now;
+       locker_info *idmap;
+!      u_int32_t *bitmap, *copymap, **deadp, **free_me, *tmpmap;
+       u_int32_t i, cid, keeper, killid, limit, nalloc, nlockers;
+       u_int32_t lock_max, txn_max;
+       int ret, status;
+--- 121,127 ----
+       DB_LOCKTAB *lt;
+       db_timespec now;
+       locker_info *idmap;
+!      u_int32_t *bitmap, *copymap, **deadp, **deadlist, *tmpmap;
+       u_int32_t i, cid, keeper, killid, limit, nalloc, nlockers;
+       u_int32_t lock_max, txn_max;
+       int ret, status;
+***************
+*** 133,139 ****
+       if (IS_REP_CLIENT(env))
+               atype = DB_LOCK_MINWRITE;
+  
+!      free_me = NULL;
+  
+       lt = env->lk_handle;
+       if (rejectp != NULL)
+--- 133,140 ----
+       if (IS_REP_CLIENT(env))
+               atype = DB_LOCK_MINWRITE;
+  
+!      copymap = tmpmap = NULL;
+!      deadlist = NULL;
+  
+       lt = env->lk_handle;
+       if (rejectp != NULL)
+***************
+*** 179,189 ****
+       memcpy(copymap, bitmap, nlockers * sizeof(u_int32_t) * nalloc);
+  
+       if ((ret = __os_calloc(env, sizeof(u_int32_t), nalloc, &tmpmap)) != 0)
+!              goto err1;
+  
+       /* Find a deadlock. */
+       if ((ret =
+!          __dd_find(env, bitmap, idmap, nlockers, nalloc, &deadp)) != 0)
+               return (ret);
+  
+       /*
+--- 180,190 ----
+       memcpy(copymap, bitmap, nlockers * sizeof(u_int32_t) * nalloc);
+  
+       if ((ret = __os_calloc(env, sizeof(u_int32_t), nalloc, &tmpmap)) != 0)
+!              goto err;
+  
+       /* Find a deadlock. */
+       if ((ret =
+!          __dd_find(env, bitmap, idmap, nlockers, nalloc, &deadlist)) != 0)
+               return (ret);
+  
+       /*
+***************
+*** 204,211 ****
+               txn_max = TXN_MAXIMUM;
+  
+       killid = BAD_KILLID;
+!      free_me = deadp;
+!      for (; *deadp != NULL; deadp++) {
+               if (rejectp != NULL)
+                       ++*rejectp;
+               killid = (u_int32_t)(*deadp - bitmap) / nalloc;
+--- 205,211 ----
+               txn_max = TXN_MAXIMUM;
+  
+       killid = BAD_KILLID;
+!      for (deadp = deadlist; *deadp != NULL; deadp++) {
+               if (rejectp != NULL)
+                       ++*rejectp;
+               killid = (u_int32_t)(*deadp - bitmap) / nalloc;
+***************
+*** 342,352 ****
+                       __db_msg(env,
+                           "Aborting locker %lx", (u_long)idmap[killid].id);
+       }
+!      __os_free(env, tmpmap);
+! err1:        __os_free(env, copymap);
+! 
+! err: if (free_me != NULL)
+!              __os_free(env, free_me);
+       __os_free(env, bitmap);
+       __os_free(env, idmap);
+  
+--- 342,353 ----
+                       __db_msg(env,
+                           "Aborting locker %lx", (u_long)idmap[killid].id);
+       }
+! err: if(copymap != NULL)
+!              __os_free(env, copymap);
+!      if (deadlist != NULL)
+!              __os_free(env, deadlist);
+!      if(tmpmap != NULL)
+!              __os_free(env, tmpmap);
+       __os_free(env, bitmap);
+       __os_free(env, idmap);
+  
+***************
+*** 360,365 ****
+--- 361,377 ----
+  
+  #define      DD_INVALID_ID   ((u_int32_t) -1)
+  
++ /*
++  * __dd_build --
++  *   Build the lock dependency bit maps.
++  * Notes on syncronization:  
++  *   LOCK_SYSTEM_LOCK is used to hold objects locked when we have
++  *           a single partition.
++  *   LOCK_LOCKERS is held while we are walking the lockers list and
++  *           to single thread the use of lockerp->dd_id.
++  *   LOCK_DD protects the DD list of objects.
++  */
++ 
+  static int
+  __dd_build(env, atype, bmp, nlockers, allocp, idmap, rejectp)
+       ENV *env;
+***************
+*** 393,398 ****
+--- 405,411 ----
+        * In particular we do not build the conflict array and our caller
+        * needs to expect this.
+        */
++      LOCK_SYSTEM_LOCK(lt, region);
+       if (atype == DB_LOCK_EXPIRE) {
+  skip:                LOCK_DD(env, region);
+               op = SH_TAILQ_FIRST(&region->dd_objs, __db_lockobj);
+***************
+*** 430,446 ****
+                       OBJECT_UNLOCK(lt, region, indx);
+               }
+               UNLOCK_DD(env, region);
+               goto done;
+       }
+  
+       /*
+!       * We'll check how many lockers there are, add a few more in for
+!       * good measure and then allocate all the structures.  Then we'll
+!       * verify that we have enough room when we go back in and get the
+!       * mutex the second time.
+        */
+! retry:       count = region->stat.st_nlockers;
+       if (count == 0) {
+               *nlockers = 0;
+               return (0);
+       }
+--- 443,460 ----
+                       OBJECT_UNLOCK(lt, region, indx);
+               }
+               UNLOCK_DD(env, region);
++              LOCK_SYSTEM_UNLOCK(lt, region);
+               goto done;
+       }
+  
+       /*
+!       * Allocate after locking the region
+!       * to make sure the structures are large enough.
+        */
+!      LOCK_LOCKERS(env, region);
+!      count = region->stat.st_nlockers;
+       if (count == 0) {
++              UNLOCK_LOCKERS(env, region);
+               *nlockers = 0;
+               return (0);
+       }
+***************
+*** 448,497 ****
+       if (FLD_ISSET(env->dbenv->verbose, DB_VERB_DEADLOCK))
+               __db_msg(env, "%lu lockers", (u_long)count);
+  
+-      count += 20;
+       nentries = (u_int32_t)DB_ALIGN(count, 32) / 32;
+  
+!      /*
+!       * Allocate enough space for a count by count bitmap matrix.
+!       *
+!       * XXX
+!       * We can probably save the malloc's between iterations just
+!       * reallocing if necessary because count grew by too much.
+!       */
+       if ((ret = __os_calloc(env, (size_t)count,
+!          sizeof(u_int32_t) * nentries, &bitmap)) != 0)
+               return (ret);
+  
+       if ((ret = __os_calloc(env,
+           sizeof(u_int32_t), nentries, &tmpmap)) != 0) {
+               __os_free(env, bitmap);
+               return (ret);
+       }
+  
+       if ((ret = __os_calloc(env,
+           (size_t)count, sizeof(locker_info), &id_array)) != 0) {
+               __os_free(env, bitmap);
+               __os_free(env, tmpmap);
+               return (ret);
+       }
+  
+       /*
+-       * Now go back in and actually fill in the matrix.
+-       */
+-      if (region->stat.st_nlockers > count) {
+-              __os_free(env, bitmap);
+-              __os_free(env, tmpmap);
+-              __os_free(env, id_array);
+-              goto retry;
+-      }
+- 
+-      /*
+        * First we go through and assign each locker a deadlock detector id.
+        */
+       id = 0;
+-      LOCK_LOCKERS(env, region);
+       SH_TAILQ_FOREACH(lip, &region->lockers, ulinks, __db_locker) {
+               if (lip->master_locker == INVALID_ROFF) {
+                       lip->dd_id = id++;
+                       id_array[lip->dd_id].id = lip->id;
+                       switch (atype) {
+--- 462,498 ----
+       if (FLD_ISSET(env->dbenv->verbose, DB_VERB_DEADLOCK))
+               __db_msg(env, "%lu lockers", (u_long)count);
+  
+       nentries = (u_int32_t)DB_ALIGN(count, 32) / 32;
+  
+!      /* Allocate enough space for a count by count bitmap matrix. */
+       if ((ret = __os_calloc(env, (size_t)count,
+!          sizeof(u_int32_t) * nentries, &bitmap)) != 0) {
+!              UNLOCK_LOCKERS(env, region);
+               return (ret);
++      }
+  
+       if ((ret = __os_calloc(env,
+           sizeof(u_int32_t), nentries, &tmpmap)) != 0) {
++              UNLOCK_LOCKERS(env, region);
+               __os_free(env, bitmap);
+               return (ret);
+       }
+  
+       if ((ret = __os_calloc(env,
+           (size_t)count, sizeof(locker_info), &id_array)) != 0) {
++              UNLOCK_LOCKERS(env, region);
+               __os_free(env, bitmap);
+               __os_free(env, tmpmap);
+               return (ret);
+       }
+  
+       /*
+        * First we go through and assign each locker a deadlock detector id.
+        */
+       id = 0;
+       SH_TAILQ_FOREACH(lip, &region->lockers, ulinks, __db_locker) {
+               if (lip->master_locker == INVALID_ROFF) {
++                      DB_ASSERT(env, id < count);
+                       lip->dd_id = id++;
+                       id_array[lip->dd_id].id = lip->id;
+                       switch (atype) {
+***************
+*** 510,516 ****
+                       lip->dd_id = DD_INVALID_ID;
+  
+       }
+-      UNLOCK_LOCKERS(env, region);
+  
+       /*
+        * We only need consider objects that have waiters, so we use
+--- 511,516 ----
+***************
+*** 669,675 ****
+        * status after building the bit maps so that we will not detect
+        * a blocked transaction without noting that it is already aborting.
+        */
+-      LOCK_LOCKERS(env, region);
+       for (id = 0; id < count; id++) {
+               if (!id_array[id].valid)
+                       continue;
+--- 669,674 ----
+***************
+*** 738,743 ****
+--- 737,743 ----
+                       id_array[id].in_abort = 1;
+       }
+       UNLOCK_LOCKERS(env, region);
++      LOCK_SYSTEM_UNLOCK(lt, region);
+  
+       /*
+        * Now we can release everything except the bitmap matrix that we
+***************
+*** 839,844 ****
+--- 839,845 ----
+       ret = 0;
+  
+       /* We must lock so this locker cannot go away while we abort it. */
++      LOCK_SYSTEM_LOCK(lt, region);
+       LOCK_LOCKERS(env, region);
+  
+       /*
+***************
+*** 895,900 ****
+--- 896,902 ----
+  done:        OBJECT_UNLOCK(lt, region, info->last_ndx);
+  err:
+  out: UNLOCK_LOCKERS(env, region);
++      LOCK_SYSTEM_UNLOCK(lt, region);
+       return (ret);
+  }
+  
This page took 0.10203 seconds and 4 git commands to generate.