Skip to contentSkip to conversation

Prevent btrfsck to run on mounted filesystems

PATCH In review Andi Drebes 3 participants 5 messages Oct 29, 2009 – Nov 9, 2009 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Prevent btrfsck to run on mounted filesystems

As recently discussed on the list, btrfsck should only be run on unmounted filesystems. This patch adds a short check for the mount status at the beginning of btrfsck. If the FS is mounted, the program aborts showing an error message.

---

btrfsck.c +11 -0
index 73f1836..87ae776 100644
@@ -28,6 +28,7 @@
 #include "transaction.h"
 #include "list.h"
 #include "version.h"
+#include "utils.h"
 
 static u64 bytes_used = 0;
 static u64 total_csum_bytes = 0;
@@ -2819,6 +2820,16 @@ int main(int ac, char **av)
 	if (ac < 2)
 		print_usage();
 
+	ret = check_mounted(av[1]);
+	if (ret < 0) {
+		fprintf(stderr, "error checking %s mount status\n", av[1]);
+		return 1;
+	}
+	if (ret == 1) {
+		fprintf(stderr, "%s is mounted. btrfsck can only be run on an unmounted filesystem.\n", av[1]);
+		return 1;
+	}
+
 	radix_tree_init();
 	cache_tree_init(&root_cache);
 	root = open_ctree(av[1], 0, 0);
#1 lore raw
Quoted 1 line · Andi Drebes

As recently discussed on the list, btrfsck should only be run on unmounted filesystems. This patch adds a short check for the mount status at the beginning of btrfsck. If the FS is mounted, the program aborts showing an error message.

Just open the nodes with O_EXCL and you'll get all the checking for free. Also make sure that for a pure, read-only checks instead of a repair to allow running on at least a read-only mounted filesystem.

#2 lore raw
Quoted 7 lines · Christoph Hellwig
Quoted 1 line · Andi Drebes

As recently discussed on the list, btrfsck should only be run on unmounted filesystems. This patch adds a short check for the mount status at the beginning of btrfsck. If the FS is mounted, the program aborts showing an error message.

Just open the nodes with O_EXCL and you'll get all the checking for free. Also make sure that for a pure, read-only checks instead of a repair to allow running on at least a read-only mounted filesystem.

Thanks for working on this patch Andi.

In this case O_EXCL is going to be more accurate just because the mounted check doesn't cover every disk in the FS. For now btrfsck doesn't really give consistent results even readonly on a mounted filesystem. We should prevent it with a message just to prevent confusion.

-chris

#3 lore raw
Quoted 11 lines
Quoted 3 lines

Just open the nodes with O_EXCL and you'll get all the checking for free. Also make sure that for a pure, read-only checks instead of a repair to allow running on at least a read-only mounted filesystem.

Thanks for working on this patch Andi.

In this case O_EXCL is going to be more accurate just because the mounted check doesn't cover every disk in the FS. For now btrfsck doesn't really give consistent results even readonly on a mounted filesystem. We should prevent it with a message just to prevent confusion.

Thanks for the reply. I'll do this as soon as I understand the code that is affected by those changes (I'm still in the learning phase and stepping through the code).

Cheers,
Andi

#4 lore raw

Hi!

Quoted 8 lines
Quoted 5 lines

In this case O_EXCL is going to be more accurate just because the mounted check doesn't cover every disk in the FS. For now btrfsck doesn't really give consistent results even readonly on a mounted filesystem. We should prevent it with a message just to prevent confusion.

Thanks for the reply. I'll do this as soon as I understand the code that is affected by those changes (I'm still in the learning phase and stepping through the code).

OK, I've stepped through the code so far and I think I now have an idea of what should be changed. However, concerning the solution, I was confusing two things.

Quoted 2 lines · Christoph Hellwig earlier

Just open the nodes with O_EXCL and you'll get all the checking for free.

I thought that open() with O_EXCL would fail if the file is already opened RW somewhere else. But the man page for open() says something else:

"O_EXCL: Ensure that this call creates the file: if this flag is specified in conjunction with O_CREAT, and pathname already exists, then open() will fail."

And especially:

"The behavior of O_EXCL is undefined if O_CREAT is not specified."

After having read this, I don't see how this can help to prevent btrfsck to open a device that is mounted. What am I getting wrong? Currently, I would do the same check as in the original patch, but in btrfs_open_devices() instead of main().

Cheers,
Andi