Skip to contentSkip to conversation

fanotify: Disallow permission events for proc filesystem

PATCH Superseded Jan Kara 3 participants 6 messages May 15, 2019 – May 26, 2019

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Jan Kara

author maintainer
#0 lore raw

PATCH 1/1 fanotify: Disallow permission events for proc filesystem

Proc filesystem has special locking rules for various files. Thus fanotify which opens files on event delivery can easily deadlock against another process that waits for fanotify permission event to be handled. Since permission events on /proc have doubtful value anyway, just disallow them.

---
 fs/notify/fanotify/fanotify_user.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
fs/notify/fanotify/fanotify_user.c +20 -0
index a90bb19dcfa2..73719949faa6 100644
@@ -920,6 +920,20 @@ static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
 	return 0;
 }
 
+static int fanotify_events_supported(struct path *path, __u64 mask)
+{
+	/*
+	 * Proc is special and various files have special locking rules so
+	 * fanotify permission events have high chances of deadlocking the
+	 * system. Just disallow them.
+	 */
+	if (mask & FANOTIFY_PERM_EVENTS &&
+	    !strcmp(path->mnt->mnt_sb->s_type->name, "proc")) {
+		return -EOPNOTSUPP;
+	}
+	return 0;
+}
+
 static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
 			    int dfd, const char  __user *pathname)
 {
@@ -1018,6 +1032,12 @@ static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask,
 	if (ret)
 		goto fput_and_out;
 
+	if (flags & FAN_MARK_ADD) {
+		ret = fanotify_events_supported(&path, mask);
+		if (ret)
+			goto path_put_and_out;
+	}
+
 	if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
 		ret = fanotify_test_fid(&path, &__fsid);
 		if (ret)
Signature · 1 line
2.16.4
#1 lore raw
Quoted 7 lines · Jan Kara

Proc filesystem has special locking rules for various files. Thus fanotify which opens files on event delivery can easily deadlock against another process that waits for fanotify permission event to be handled. Since permission events on /proc have doubtful value anyway, just disallow them.

Let's add context:

Quoted 12 lines
---
 fs/notify/fanotify/fanotify_user.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
fs/notify/fanotify/fanotify_user.c +0 -0
index a90bb19dcfa2..73719949faa6 100644
@@ -920,6 +920,20 @@ static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
        return 0;
 }

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 931

Quoted 10 lines
fs/notify/fanotify/fanotify_user.c +9 -0
 
+static int fanotify_events_supported(struct path *path, __u64 mask)
+{
+       /*
+        * Proc is special and various files have special locking rules so
+        * fanotify permission events have high chances of deadlocking the
+        * system. Just disallow them.
+        */
+       if (mask & FANOTIFY_PERM_EVENTS &&
+           !strcmp(path->mnt->mnt_sb->s_type->name, "proc")) {

Better use an SB_I_ flag to forbid permission events on fs?

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 932

Quoted 1 line
fs/notify/fanotify/fanotify_user.c +1 -0
+               return -EOPNOTSUPP;

I would go with EINVAL following precedent of per filesystem flags check on rename(2), but not insisting.

Anyway, following Matthew's man page update for FAN_REPORT_FID, we should also add this as reason for EOPNOTSUPP/EINVAL.

Thanks,
Amir.

author maintainer
#2 lore raw
Quoted 11 lines · Amir Goldstein
Quoted 7 lines · Jan Kara

Proc filesystem has special locking rules for various files. Thus fanotify which opens files on event delivery can easily deadlock against another process that waits for fanotify permission event to be handled. Since permission events on /proc have doubtful value anyway, just disallow them.

Let's add context:

OK, will add.

Quoted 13 lines
Quoted 13 lines
---
 fs/notify/fanotify/fanotify_user.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
fs/notify/fanotify/fanotify_user.c +0 -0
index a90bb19dcfa2..73719949faa6 100644
@@ -920,6 +920,20 @@ static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
        return 0;
 }
 
Quoted 11 lines

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 931

Quoted 9 lines
fs/notify/fanotify/fanotify_user.c +9 -0
+static int fanotify_events_supported(struct path *path, __u64 mask)
+{
+       /*
+        * Proc is special and various files have special locking rules so
+        * fanotify permission events have high chances of deadlocking the
+        * system. Just disallow them.
+        */
+       if (mask & FANOTIFY_PERM_EVENTS &&
+           !strcmp(path->mnt->mnt_sb->s_type->name, "proc")) {

Better use an SB_I_ flag to forbid permission events on fs?

So checking s_type->name indeed felt dirty. I don't think we need a superblock flag though. I'll probably just go with FS_XXX flag in file_system_type.

Quoted 5 lines

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 932

Quoted 2 lines
fs/notify/fanotify/fanotify_user.c +1 -0
+               return -EOPNOTSUPP;

I would go with EINVAL following precedent of per filesystem flags check on rename(2), but not insisting.

I was undecided between EOPNOTSUPP and EINVAL. So let's go with EINVAL.

Quoted 2 lines

Anyway, following Matthew's man page update for FAN_REPORT_FID, we should also add this as reason for EOPNOTSUPP/EINVAL.

Good point.

Thanks for review!

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#3 lore raw
Quoted 40 lines · Jan Kara
Quoted 11 lines · Amir Goldstein
Quoted 7 lines · Jan Kara

Proc filesystem has special locking rules for various files. Thus fanotify which opens files on event delivery can easily deadlock against another process that waits for fanotify permission event to be handled. Since permission events on /proc have doubtful value anyway, just disallow them.

Let's add context:

OK, will add.

Quoted 13 lines
Quoted 13 lines
---
 fs/notify/fanotify/fanotify_user.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
fs/notify/fanotify/fanotify_user.c +0 -0
index a90bb19dcfa2..73719949faa6 100644
@@ -920,6 +920,20 @@ static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
        return 0;
 }
 
Quoted 11 lines

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 931

Quoted 9 lines
fs/notify/fanotify/fanotify_user.c +9 -0
+static int fanotify_events_supported(struct path *path, __u64 mask)
+{
+       /*
+        * Proc is special and various files have special locking rules so
+        * fanotify permission events have high chances of deadlocking the
+        * system. Just disallow them.
+        */
+       if (mask & FANOTIFY_PERM_EVENTS &&
+           !strcmp(path->mnt->mnt_sb->s_type->name, "proc")) {

Better use an SB_I_ flag to forbid permission events on fs?

Quoted 3 lines

So checking s_type->name indeed felt dirty. I don't think we need a superblock flag though. I'll probably just go with FS_XXX flag in file_system_type.

Would the same apply for some files that backed by sysfs and reside in /sys?

Quoted 7 lines
Quoted 5 lines

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 932

Quoted 2 lines
fs/notify/fanotify/fanotify_user.c +1 -0
+               return -EOPNOTSUPP;

I would go with EINVAL following precedent of per filesystem flags check on rename(2), but not insisting.

I was undecided between EOPNOTSUPP and EINVAL. So let's go with EINVAL.

I was also thinking that EINVAL makes more sense in this particular case.

Quoted 4 lines
Quoted 2 lines

Anyway, following Matthew's man page update for FAN_REPORT_FID, we should also add this as reason for EOPNOTSUPP/EINVAL.

Good point.

I've followed up Michael in regards to the FAN_REPORT_FID patch series, but no response as of yet. I'm happy to write the changes for this one if you like?

Signature · 1 line
Matthew Bobrowski
author maintainer
#4 lore raw
Quoted 27 lines · Matthew Bobrowski
Quoted 26 lines · Jan Kara
Quoted 24 lines · Amir Goldstein
Quoted 12 lines
---
 fs/notify/fanotify/fanotify_user.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
fs/notify/fanotify/fanotify_user.c +0 -0
index a90bb19dcfa2..73719949faa6 100644
@@ -920,6 +920,20 @@ static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
        return 0;
 }

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 931

Quoted 10 lines
fs/notify/fanotify/fanotify_user.c +9 -0
 
+static int fanotify_events_supported(struct path *path, __u64 mask)
+{
+       /*
+        * Proc is special and various files have special locking rules so
+        * fanotify permission events have high chances of deadlocking the
+        * system. Just disallow them.
+        */
+       if (mask & FANOTIFY_PERM_EVENTS &&
+           !strcmp(path->mnt->mnt_sb->s_type->name, "proc")) {

Better use an SB_I_ flag to forbid permission events on fs?

Quoted 6 lines
Quoted 3 lines

So checking s_type->name indeed felt dirty. I don't think we need a superblock flag though. I'll probably just go with FS_XXX flag in file_system_type.

Would the same apply for some files that backed by sysfs and reside in /sys?

So far I'm not aware of similar easy to trigger deadlocks with sysfs. So I opted for a cautious path and disabled permission events only for proc. We'll see how that fares.

Quoted 9 lines
Quoted 6 lines
Quoted 4 lines

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 932

Quoted 1 line
fs/notify/fanotify/fanotify_user.c +1 -0
+               return -EOPNOTSUPP;

I would go with EINVAL following precedent of per filesystem flags check on rename(2), but not insisting.

I was undecided between EOPNOTSUPP and EINVAL. So let's go with EINVAL.

I was also thinking that EINVAL makes more sense in this particular case.

Good, that's what I used in v2.

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

Anyway, following Matthew's man page update for FAN_REPORT_FID, we should also add this as reason for EOPNOTSUPP/EINVAL.

Good point.

I've followed up Michael in regards to the FAN_REPORT_FID patch series, but no response as of yet. I'm happy to write the changes for this one if you like?

If you had time for that, that would be nice. Thanks!

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#5 lore raw
Quoted 54 lines · Jan Kara
Quoted 27 lines · Matthew Bobrowski
Quoted 26 lines · Jan Kara
Quoted 24 lines · Amir Goldstein
Quoted 12 lines
---
 fs/notify/fanotify/fanotify_user.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
fs/notify/fanotify/fanotify_user.c +0 -0
index a90bb19dcfa2..73719949faa6 100644
@@ -920,6 +920,20 @@ static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid)
        return 0;
 }

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 931

Quoted 10 lines
fs/notify/fanotify/fanotify_user.c +9 -0
 
+static int fanotify_events_supported(struct path *path, __u64 mask)
+{
+       /*
+        * Proc is special and various files have special locking rules so
+        * fanotify permission events have high chances of deadlocking the
+        * system. Just disallow them.
+        */
+       if (mask & FANOTIFY_PERM_EVENTS &&
+           !strcmp(path->mnt->mnt_sb->s_type->name, "proc")) {

Better use an SB_I_ flag to forbid permission events on fs?

Quoted 6 lines
Quoted 3 lines

So checking s_type->name indeed felt dirty. I don't think we need a superblock flag though. I'll probably just go with FS_XXX flag in file_system_type.

Would the same apply for some files that backed by sysfs and reside in /sys?

So far I'm not aware of similar easy to trigger deadlocks with sysfs. So I opted for a cautious path and disabled permission events only for proc. We'll see how that fares.

Quoted 9 lines
Quoted 6 lines
Quoted 4 lines

fs/notify/fanotify/fanotify_user.c static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) · line 932

Quoted 1 line
fs/notify/fanotify/fanotify_user.c +1 -0
+               return -EOPNOTSUPP;

I would go with EINVAL following precedent of per filesystem flags check on rename(2), but not insisting.

I was undecided between EOPNOTSUPP and EINVAL. So let's go with EINVAL.

I was also thinking that EINVAL makes more sense in this particular case.

Good, that's what I used in v2.

Quoted 3 lines
Quoted 3 lines
Quoted 2 lines

Anyway, following Matthew's man page update for FAN_REPORT_FID, we should also add this as reason for EOPNOTSUPP/EINVAL.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

Good point.

I've followed up Michael in regards to the FAN_REPORT_FID patch series, but no response as of yet. I'm happy to write the changes for this one if you like?

If you had time for that, that would be nice. Thanks!

Simple. I will write the changes and send it through for review.

Signature · 1 line
Matthew Bobrowski