new_inode_autonum: add per-sb lastino counter and add new_inode_autonum function that guarantees i_ino uniqueness
PATCH 1/3 new_inode_autonum: add per-sb lastino counter and add new_inode_autonum function that guarantees i_ino uniqueness
- add a new per superblock s_nextino counter and change iunique to use it instead of its global inode counter
- make the size of the counter conditional on CONFIG_COMPAT. This is to try prevent userspace EOVERFLOWs when 32-bit programs not compiled with large offsets are run on 64-bit kernels
- add new_inode_autonum which guarantees that i_ino is assigned a unique val on the filesystem.
- Change new_inode to assign i_ino to 0 to catch filesystems that use it and don't reset it to a unique value.
fs/inode.c +41 -9
@@ -517,14 +517,14 @@ repeat: }/** -* new_inode - obtain an inode +* __new_inode - obtain an inode * @sb: superblock +* @autonum: if true, make sure that i_ino is unique * * Allocates a new inode for given superblock. */ -struct inode *new_inode (struct super_block *sb) +static struct inode *__new_inode (struct super_block *sb) { -static unsigned long last_ino;struct inode * inode;spin_lock_prefetch (&inode_lock);
@@ -535,13 +535,46 @@ struct inode *new_inode(struct super_blo inodes_stat.nr_inodes++;list_add (&inode->i_list, &inode_in_use);list_add (&inode->i_sb_list, &sb->s_inodes); - inode->i_ino = ++last_ino; inode->i_state =0 ;spin_unlock (&inode_lock); }return inode; } +/** +* new_inode_autonum - obtain an inode with a unique i_ino value +* @sb: superblock +* +* Allocates a new inode for given superblock. Ensures that i_ino is +* unique on the filesystem. +*/ +struct inode *new_inode_autonum (struct super_block *sb) +{ +struct inode *inode; + + inode =__new_inode (sb); + inode->i_ino =iunique (sb,0 ); +return inode; +} + +EXPORT_SYMBOL (new_inode_autonum); + +/** +* new_inode - obtain an inode -- i_ino not guaranteed unique +* @sb: superblock +* +* Allocates a new inode for given superblock. i_ino is not guaranteed to +* be unique. Should only be used when i_ino is going to be clobbered. +*/ +struct inode *new_inode (struct super_block *sb) +{ +struct inode *inode; + + inode =__new_inode (sb); + inode->i_ino =0 ;/* 0 to try to catch callers that don't reset it */ +return inode; +} +EXPORT_SYMBOL (new_inode);void unlock_new_inode (struct inode *inode)
@@ -683,22 +716,21 @@ static unsigned long hash(struct super_b*/ ino_t iunique (struct super_block *sb,ino_t max_reserved) { -static ino_t counter;struct inode *inode;struct hlist_head * head;ino_t res;spin_lock (&inode_lock); retry: -if (counter > max_reserved) { - head = inode_hashtable +hash (sb,counter); - res = counter++; +if (sb->s_nextino > max_reserved) { + head = inode_hashtable +hash (sb,sb->s_nextino); + res = sb->s_nextino++; inode =find_inode_fast (sb, head, res);if (!inode) {spin_unlock (&inode_lock);return res; } }else { - counter = max_reserved +1 ; + sb->s_nextino = max_reserved +1 ; } goto retry;
include/linux/fs.h +9 -0
@@ -961,6 +961,14 @@ #endif/* Granularity of c/m/atime in ns. Cannot be worse than a second */ u32 s_time_gran; + +/* per-sb inode counter for new_inode. Make it a 32-bit counter when +we have the possibility of dealing with 32-bit apps */ +#ifdef CONFIG_COMPAT +unsigned int s_nextino; +#else +unsigned long s_nextino; +#endif };extern struct timespec current_fs_time (struct super_block *sb);
@@ -1712,6 +1720,7 @@ extern void __iget(struct inode * inode)extern void clear_inode (struct inode *);extern void destroy_inode (struct inode *);extern struct inode *new_inode (struct super_block *); +extern struct inode *new_inode_autonum (struct super_block *);extern int __remove_suid (struct dentry *,int );extern int should_remove_suid (struct dentry *);extern int remove_suid (struct dentry *);
Quoted 9 lines · Jeff Layton
- add a new per superblock
s_nextinocounter and change iunique to use it instead of its global inode counter
- make the size of the counter conditional onCONFIG_COMPAT. This is to try prevent userspace EOVERFLOWs when 32-bit programs not compiled with large offsets are run on 64-bit kernels
- addnew_inode_autonumwhich guarantees thati_inois assigned a unique val on the filesystem.
- Changenew_inodeto assigni_inoto 0 to catch filesystems that use it and don't reset it to a unique value.
I like the general approach. A few tidbits...
- the comment for __new_inode isn't accurate, there is no @autonum parameter added.
- I'm not sure what the point of wrapping __new_inode() in new_inode() is. Why not just remove the i_ino setting from new_inode, (or initialize it to 0*), and let new_inode_autonum() call new_inode(); iunique() ? Less indirection IMHO....
- *does setting the new i_ino to 0 really help catch anything? And if it does, perhaps just setting it in alloc_inode() with all the other field initializations would be more consistent.
- the comment for the s_nextino field is a bit misleading, it is actually not used in new_inode(), only in iunique().
-Eric
Quoted 4 lines · Jeff Layton
fs/inode.c +4
-0+/** +* new_inode_autonum - obtain an inode with a unique i_ino value +* @sb: superblock +*
fs/inode.c struct inode *new_inode(struct super_blo · line 557
Quoted 10 lines
fs/inode.c +10
-0+* Allocates a new inode for given superblock. Ensures that i_ino is +* unique on the filesystem. +*/ +struct inode *new_inode_autonum (struct super_block *sb) +{ +struct inode *inode; + + inode =__new_inode (sb); + inode->i_ino =iunique (sb,0 ); +return inode;
Why do we need this wrapper? The callers could aswell just do the iunique call themselves. It's already exported aswell.
Quoted 4 lines
fs/inode.c +4
-0+/** +* new_inode - obtain an inode -- i_ino not guaranteed unique +* @sb: superblock +*
fs/inode.c struct inode *new_inode(struct super_blo · line 575
Quoted 10 lines
fs/inode.c +10
-0+* Allocates a new inode for given superblock. i_ino is not guaranteed to +* be unique. Should only be used when i_ino is going to be clobbered. +*/ +struct inode *new_inode (struct super_block *sb) +{ +struct inode *inode; + + inode =__new_inode (sb); + inode->i_ino =0 ;/* 0 to try to catch callers that don't reset it */ +return inode;
And this wrapper is rather pointless as inode_init_once already zeroes the whole inode. Just keep the name new_inode for the basic don't assigned inode number thing.
Quoted 4 lines
include/linux/fs.h +0
-0@@ -961,6 +961,14 @@ #endif/* Granularity of c/m/atime in ns.
include/linux/fs.h #endif · line 971
Quoted 10 lines
include/linux/fs.h +8
-0Cannot be worse than a second */ u32 s_time_gran; + +/* per-sb inode counter for new_inode. Make it a 32-bit counter when +we have the possibility of dealing with 32-bit apps */ +#ifdef CONFIG_COMPAT +unsigned int s_nextino; +#else +unsigned long s_nextino; +#endif
This is more thanb ugly. CONFIG_COMPAT should not modify core kernel behaviour. It's also short sightened as we once again plan to support 64bit inode numbers on 32bit hosts.