Forum Stats

  • 3,757,056 Users
  • 2,251,192 Discussions


[PATCH] The memory is not re-mapped after libdb extends the __db file

MIZUTA Takeshi
MIZUTA Takeshi Member Posts: 4 Green Ribbon
edited Sep 3, 2021 12:21AM in Berkeley DB

When the __db file size limit is reached, libdb calls lseek() and write() to extend the file.

 19 int

 20 __db_file_extend(env, fhp, size)


 51  pages = (db_pgno_t)((size - sizeof(buf)) / MEGABYTE);

 52  relative = (u_int32_t)((size - sizeof(buf)) % MEGABYTE);

 53  if ((ret = __os_seek(env, fhp, pages, MEGABYTE, relative)) == 0)

 54   ret = __os_write(env, fhp, &buf, sizeof(buf), &nw);

However, after expanding the __db file, it is not syncing with memory.

mmap has the following notes.

 From mmap(2) manpage:

 A file is mapped in multiples of the page size. For a file that is not

 a multiple of the page size, the remaining memory is zeroed when

 mapped, and writes to that region are not written out to the file. The

 effect of changing the size of the underlying file of a mapping on the

 pages that correspond to added or removed regions of the file is


 From mmap in IEEE Std 1003.1-2017:

 If the size of the mapped file changes after the call to mmap() as a result of

 some other operation on the mapped file, the effect of references to portions

 of the mapped region that correspond to added or removed portions of the file

 is unspecified.

Therefore, one of the following deals are required.

 - Call munmap()/mmap() to synchronize memory during lseek() and write()

 - Get the maximum size __db file in advance instead of gradually expanding the __db file.

The following is the latter patch:


diff --git a/db-18.1.40/src/os/os_map.c b/db-18.1.40/src/os/os_map.c

index dcf2c23..83a79a8 100644

--- a/db-18.1.40/src/os/os_map.c

+++ b/db-18.1.40/src/os/os_map.c

@@ -231,15 +231,7 @@ __os_attach(env, infop, rp)

    if (rp->max < rp->size)

        rp->max = rp->size;

    if (ret == 0 && F_ISSET(infop, REGION_CREATE)) {

-#ifdef HAVE_MLOCK

-        /*

-        * When locking the region in memory extend it fully so that it

-        * can all be mlock()'d now, and not later when paging could

-        * interfere with the application. [#21379]

-        */

-        if (F_ISSET(env, ENV_LOCKDOWN))

-            rp->size = rp->max;


+        rp->size = rp->max;

        if (F_ISSET(dbenv, DB_ENV_REGION_INIT))

            ret = __db_file_write(env, infop->fhp,

              rp->size / MEGABYTE, rp->size % MEGABYTE, 0x00);


Here's the patch:

Do you have any opinions about this patch?