From: Sharyathi Nagesh Date: Fri, 10 Nov 2006 20:27:54 +0000 (-0800) Subject: [PATCH] fix Data Acess error in dup_fd X-Git-Tag: v2.6.19-rc6~36 X-Git-Url: https://err.no/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=0130b0b32ee53dc7add773fcea984f6a26ef1da3;p=linux-2.6 [PATCH] fix Data Acess error in dup_fd On running the Stress Test on machine for more than 72 hours following error message was observed. 0:mon> e cpu 0x0: Vector: 300 (Data Access) at [c00000007ce2f7f0] pc: c000000000060d90: .dup_fd+0x240/0x39c lr: c000000000060d6c: .dup_fd+0x21c/0x39c sp: c00000007ce2fa70 msr: 800000000000b032 dar: ffffffff00000028 dsisr: 40000000 current = 0xc000000074950980 paca = 0xc000000000454500 pid = 27330, comm = bash 0:mon> t [c00000007ce2fa70] c000000000060d28 .dup_fd+0x1d8/0x39c (unreliable) [c00000007ce2fb30] c000000000060f48 .copy_files+0x5c/0x88 [c00000007ce2fbd0] c000000000061f5c .copy_process+0x574/0x1520 [c00000007ce2fcd0] c000000000062f88 .do_fork+0x80/0x1c4 [c00000007ce2fdc0] c000000000011790 .sys_clone+0x5c/0x74 [c00000007ce2fe30] c000000000008950 .ppc_clone+0x8/0xc The problem is because of race window. When if(expand) block is executed in dup_fd unlocking of oldf->file_lock give a window for fdtable in oldf to be modified. So actual open_files in oldf may not match with open_files variable. Cc: Vadim Lobanov Cc: Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- diff --git a/kernel/fork.c b/kernel/fork.c index 3da978eec7..4b4eab2a31 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -687,6 +687,7 @@ static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp) * the latest pointer. */ spin_lock(&oldf->file_lock); + open_files = count_open_files(old_fdt); old_fdt = files_fdtable(oldf); }