| Message ID | 20180731211045.5671-2-mfasheh@suse.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | vfs: map unique ino/dev pairs for user space | expand |
On Tue, Jul 31, 2018 at 02:10:42PM -0700, Mark Fasheh wrote: > diff --git a/fs/stat.c b/fs/stat.c > index f8e6fb2c3657..80ea42505219 100644 > --- a/fs/stat.c > +++ b/fs/stat.c > @@ -84,6 +84,29 @@ int vfs_getattr_nosec(const struct path *path, struct kstat *stat, > } > EXPORT_SYMBOL(vfs_getattr_nosec); > > +void vfs_map_unique_ino_dev(struct dentry *dentry, u64 *ino, dev_t *dev) > +{ > + int ret; > + struct path path; > + struct kstat stat; > + > + path.mnt = NULL; > + path.dentry = dentry; > + /* ->dev is always returned, so we only need to specify ino here */ > + ret = vfs_getattr_nosec(&path, &stat, STATX_INO, 0); > + if (ret) { > + struct inode *inode = d_inode(dentry); > + /* Fallback to old behavior in case of getattr error */ > + *ino = inode->i_ino; > + *dev = inode->i_sb->s_dev; > + return ret; Ooof, attached is a version of this patch which doesn't try to return a value from a void function. Apologies for the extra e-mail. From Mark Fasheh <mfasheh@suse.de> [PATCH 1/4] vfs: introduce function to map unique ino/dev pairs There are places in the VFS where we export an ino/dev pair to userspace. /proc/PID/maps is a good example - we directly expose inode->i_ino and inode->i_sb->s_dev to userspace there. Many filesystems don't put a unique value in inode->i_ino and instead rely on ->getattr to provide the real inode number to userspace. super->s_dev is similar - some filesystems expose a difference device from what's put in super->s_dev when queried via stat/statx. Ultimately this makes it impossible for a user (or software) to match one of those reported pairs to the right inode. We can fix this by adding a helper function, vfs_map_unique_ino_dev(), which will query the owning filesystem (via ->getattr) to get the correct ino/dev pair. Later patches will update those places which simply dump inode->i_ino and super->s_dev to use the helper. Signed-off-by: Mark Fasheh <mfasheh@suse.de> --- fs/stat.c | 22 ++++++++++++++++++++++ include/linux/fs.h | 2 ++ 2 files changed, 24 insertions(+) diff --git a/fs/stat.c b/fs/stat.c index f8e6fb2c3657..20c72d618ed5 100644 --- a/fs/stat.c +++ b/fs/stat.c @@ -84,6 +84,28 @@ int vfs_getattr_nosec(const struct path *path, struct kstat *stat, } EXPORT_SYMBOL(vfs_getattr_nosec); +void vfs_map_unique_ino_dev(struct dentry *dentry, u64 *ino, dev_t *dev) +{ + int ret; + struct path path; + struct kstat stat; + + path.mnt = NULL; + path.dentry = dentry; + /* ->dev is always returned, so we only need to specify ino here */ + ret = vfs_getattr_nosec(&path, &stat, STATX_INO, 0); + if (ret) { + struct inode *inode = d_inode(dentry); + /* Fallback to old behavior in case of getattr error */ + *ino = inode->i_ino; + *dev = inode->i_sb->s_dev; + return; + } + *ino = stat.ino; + *dev = stat.dev; +} +EXPORT_SYMBOL(vfs_map_unique_ino_dev); + /* * vfs_getattr - Get the enhanced basic attributes of a file * @path: The file of interest diff --git a/include/linux/fs.h b/include/linux/fs.h index d78d146a98da..b80789472438 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -3077,6 +3077,8 @@ extern void kfree_link(void *); extern void generic_fillattr(struct inode *, struct kstat *); extern int vfs_getattr_nosec(const struct path *, struct kstat *, u32, unsigned int); extern int vfs_getattr(const struct path *, struct kstat *, u32, unsigned int); +extern void vfs_map_unique_ino_dev(struct dentry *dentry, u64 *ino, dev_t *dev); + void __inode_add_bytes(struct inode *inode, loff_t bytes); void inode_add_bytes(struct inode *inode, loff_t bytes); void __inode_sub_bytes(struct inode *inode, loff_t bytes);
On Wed, Aug 1, 2018 at 2:21 AM, Mark Fasheh <mfasheh@suse.de> wrote: > > On Tue, Jul 31, 2018 at 02:10:42PM -0700, Mark Fasheh wrote: >> diff --git a/fs/stat.c b/fs/stat.c >> index f8e6fb2c3657..80ea42505219 100644 >> --- a/fs/stat.c >> +++ b/fs/stat.c >> @@ -84,6 +84,29 @@ int vfs_getattr_nosec(const struct path *path, struct kstat *stat, >> } >> EXPORT_SYMBOL(vfs_getattr_nosec); >> >> +void vfs_map_unique_ino_dev(struct dentry *dentry, u64 *ino, dev_t *dev) >> +{ >> + int ret; >> + struct path path; >> + struct kstat stat; >> + >> + path.mnt = NULL; >> + path.dentry = dentry; >> + /* ->dev is always returned, so we only need to specify ino here */ >> + ret = vfs_getattr_nosec(&path, &stat, STATX_INO, 0); >> + if (ret) { >> + struct inode *inode = d_inode(dentry); >> + /* Fallback to old behavior in case of getattr error */ >> + *ino = inode->i_ino; >> + *dev = inode->i_sb->s_dev; >> + return ret; > > Ooof, attached is a version of this patch which doesn't try to return a value > from a void function. Apologies for the extra e-mail. > > From Mark Fasheh <mfasheh@suse.de> > > [PATCH 1/4] vfs: introduce function to map unique ino/dev pairs > > There are places in the VFS where we export an ino/dev pair to userspace. > /proc/PID/maps is a good example - we directly expose inode->i_ino and > inode->i_sb->s_dev to userspace there. Many filesystems don't put a unique > value in inode->i_ino and instead rely on ->getattr to provide the real > inode number to userspace. super->s_dev is similar - some filesystems > expose a difference device from what's put in super->s_dev when queried via > stat/statx. > > Ultimately this makes it impossible for a user (or software) to match one of > those reported pairs to the right inode. > > We can fix this by adding a helper function, vfs_map_unique_ino_dev(), which > will query the owning filesystem (via ->getattr) to get the correct ino/dev > pair. Later patches will update those places which simply dump inode->i_ino > and super->s_dev to use the helper. > > Signed-off-by: Mark Fasheh <mfasheh@suse.de> > --- > fs/stat.c | 22 ++++++++++++++++++++++ > include/linux/fs.h | 2 ++ > 2 files changed, 24 insertions(+) > > diff --git a/fs/stat.c b/fs/stat.c > index f8e6fb2c3657..20c72d618ed5 100644 > --- a/fs/stat.c > +++ b/fs/stat.c > @@ -84,6 +84,28 @@ int vfs_getattr_nosec(const struct path *path, struct kstat *stat, > } > EXPORT_SYMBOL(vfs_getattr_nosec); > > +void vfs_map_unique_ino_dev(struct dentry *dentry, u64 *ino, dev_t *dev) I find this function name a bit more than function can guaranty. It's just a fancy wrapper around ->getattr() How about vfs_get_ino_dev() ? Thanks, Amir.
Hi Amir, On Wed, Aug 01, 2018 at 08:41:14AM +0300, Amir Goldstein wrote: > > +void vfs_map_unique_ino_dev(struct dentry *dentry, u64 *ino, dev_t *dev) > > I find this function name a bit more than function can guaranty. > It's just a fancy wrapper around ->getattr() > How about vfs_get_ino_dev() ? Yeah I agree with that. An early version actually had the name vfs_get_ino_dev(), I don't mind going back to it. Thanks for the review! --Mark
diff --git a/fs/stat.c b/fs/stat.c index f8e6fb2c3657..80ea42505219 100644 --- a/fs/stat.c +++ b/fs/stat.c @@ -84,6 +84,29 @@ int vfs_getattr_nosec(const struct path *path, struct kstat *stat, } EXPORT_SYMBOL(vfs_getattr_nosec); +void vfs_map_unique_ino_dev(struct dentry *dentry, u64 *ino, dev_t *dev) +{ + int ret; + struct path path; + struct kstat stat; + + path.mnt = NULL; + path.dentry = dentry; + /* ->dev is always returned, so we only need to specify ino here */ + ret = vfs_getattr_nosec(&path, &stat, STATX_INO, 0); + if (ret) { + struct inode *inode = d_inode(dentry); + /* Fallback to old behavior in case of getattr error */ + *ino = inode->i_ino; + *dev = inode->i_sb->s_dev; + return ret; + } + *ino = stat.ino; + *dev = stat.dev; + return 0; +} +EXPORT_SYMBOL(vfs_map_unique_ino_dev); + /* * vfs_getattr - Get the enhanced basic attributes of a file * @path: The file of interest diff --git a/include/linux/fs.h b/include/linux/fs.h index d78d146a98da..b80789472438 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -3077,6 +3077,8 @@ extern void kfree_link(void *); extern void generic_fillattr(struct inode *, struct kstat *); extern int vfs_getattr_nosec(const struct path *, struct kstat *, u32, unsigned int); extern int vfs_getattr(const struct path *, struct kstat *, u32, unsigned int); +extern void vfs_map_unique_ino_dev(struct dentry *dentry, u64 *ino, dev_t *dev); + void __inode_add_bytes(struct inode *inode, loff_t bytes); void inode_add_bytes(struct inode *inode, loff_t bytes); void __inode_sub_bytes(struct inode *inode, loff_t bytes);
There are places in the VFS where we export an ino/dev pair to userspace. /proc/PID/maps is a good example - we directly expose inode->i_ino and inode->i_sb->s_dev to userspace there. Many filesystems don't put a unique value in inode->i_ino and instead rely on ->getattr to provide the real inode number to userspace. super->s_dev is similar - some filesystems expose a difference device from what's put in super->s_dev when queried via stat/statx. Ultimately this makes it impossible for a user (or software) to match one of those reported pairs to the right inode. We can fix this by adding a helper function, vfs_map_unique_ino_dev(), which will query the owning filesystem (via ->getattr) to get the correct ino/dev pair. Later patches will update those places which simply dump inode->i_ino and super->s_dev to use the helper. Signed-off-by: Mark Fasheh <mfasheh@suse.de> --- fs/stat.c | 23 +++++++++++++++++++++++ include/linux/fs.h | 2 ++ 2 files changed, 25 insertions(+)