The following series implements read-only bind mounts. This set does not take the previously tried approach of pushing down the vfsmount structure deeply into call paths, such that it might be checked in functions like permission(), may_create() and may_open(). Instead, it does checks near the entry points in the kernel, bumping a reference count in the vfsmount structure and effetively holding a kind of a lock on the vfsmount. This approach will come in handy in the future, allowing transitions from rw to ro mounts while guaranteeing that no writers are currently active. I apologize for so many small patches, but breaking them up like this really did help me find bugs. A rollup is here: http://sr71.net/~dave/patches/ro-bind-mounts-06-07-2006.patch This series passes a test I got from Herbert Poetzl, checking to see that these r/o bind mounts have the same properties as regular r/o device mounts for a number of syscalls. http://vserver.13thfloor.at/Experimental/BME/fstest-0.03.tar.bz2 Here is the script I used to generate and compare the output from that test: #!/bin/sh { umount ro-bind-mount ro-mount; rm -rf ro-mount-source ro-bind-mount-source ro-mount ro-bind-mount; dd if=/dev/zero of=ro-mount-source count=10000; mkfs.ext3 -F ro-mount-source; mkdir ro-mount ro-bind-mount; mkdir ro-bind-mount-source } > /dev/null 2>&1 mount --bind -o ro ro-bind-mount-source ro-bind-mount mount -t ext3 -o loop,ro ro-mount-source ro-mount function fstest_and_diff { ARG1="$1" shift ARG2="$1" ./fstest -d "$ARG1" > "$ARG1".log; ./fstest -d "$ARG2" > "$ARG2".log; diff -ru "$ARG1".log "$ARG2".log } fstest_and_diff ro-mount ro-bind-mount umount ro-bind-mount ro-mount Signed-off-by: Dave Hansen <haveblue@xxxxxxxxxx> - To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
