Date: Sun, 28 May 2000 02:04:52 +0800 From: Trent Nelson <tpnelson@echidna.stu.cowan.edu.au> To: hackers@FreeBSD.org Cc: dillon@backplane.com Subject: Erroneous mmap() behaviour? Message-ID: <39300E44.558B632F@student.cowan.edu.au>
next in thread | raw e-mail | index | archive | help
Can someone explain to me why mmap() returns an address map you're
prohibited from accessing if the fd argument represents a file that has
just been created?
I have a function that calls the following, where name represents a
file that, if it exists, is intended to get written over, and if it
doesn't exist, is intended to get created:
open(name,
(O_RDWR | O_CREAT | O_TRUNC),
(S_IRUSR | S_IWUSR));
What I'm trying to do on return of this fd is, pass it through mmap()
and allocate a certain amount of memory to encapsulate some data I need
to put there - then calling msync() to have the memory buffer flushed
back to disk.
The mmap() call is:
248: mksd->msize_db = align(mksd->fsize_db);
249: mksd->mmap_db = mmap(NULL,
mksd->msize_db,
(PROT_READ | PROT_WRITE),
MAP_PRIVATE,
mksd->fd_db,
NULL);
If the file exists (size irrespective), then the following happens,
which is what we want:
bash-2.03$ ls -la database.dat
-rw------- 1 tnelson tnelson 0 May 27 23:59 database.dat
bash-2.03$ gcc -g -o mksigdata mksd-2.c file_sigs.o
bash-2.03$ gdb mksigdata
[...]
(gdb) break 249
Breakpoint 1 at 0x8048df7: file mksd-2.c, line 249.
(gdb) run files.dat database.dat
Breakpoint 1, write_database (mksd=0xbfbffb50) at mksd-2.c:249
249 mksd->mmap_db = mmap(NULL,
(gdb) n
255 if ((int)mksd->mmap_db == -1)
(gdb) display *mksd
1: *mksd = {file_list = {files = 0x280f3000, history = 0x804d030, size =
[...]
mmap_db = 0x280f4000}
^^^^^^^^^^ This is what we're after.
(Check the validity of the returned memory)
(gdb) x/4 0x280f4000
0x280f4000: 0x00000000 0x00000000 0x00000000 0x00000000
So, it all works fine. As soon as the file is removed and it has to be
created, though:
bash-2.03$ rm database.dat
bash-2.03$ gdb mksigdata
[...]
mmap_db = 0x280f4000}
(gdb) x/4 0x2980f4000
0x280f4000: Error accessing memory address 0x280f4000: Bad address.
Which is where my problem lies.
Is it me, or mmap()? If it's me, is there any better way of doing what
I want to do? (which is essentially allocating memory to be modified,
then sync'd to a most probably newly-created file).
From what I can see, the code works as it should in Linux.
Thanks in advance.
Regards,
Trent.
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?39300E44.558B632F>