fs: export kern_path_locked
Subsystems (4)
PATCH 1/1 fs: export kern_path_locked
the function is used outside and we have a prototype defined in namei.h
--- fs/namei.c | 1 + 1 file changed, 1 insertion(+)
fs/namei.c +1 -0
@@ -2450,6 +2450,7 @@ struct dentry *kern_path_locked(const char *name, struct path *path)putname (filename);return d; } +EXPORT_SYMBOL (kern_path_locked);int kern_path (const char *name,unsigned int flags,struct path *path) {
Signature · 1 line
2.16.4
Quoted 2 lines · Denis Kirjanov
the function is used outside and we have a prototype defined in
namei.h
Huh? It is only used in drivers/base/devtmpfs.c, and kernel/audit_*.c. Absolutely no need to export it for that.
Al Viro
maintainerQuoted 4 lines · Christoph Hellwig
Quoted 1 line · Denis Kirjanov
Do you mean just:
We'll still need to lock the parent inode.
Not just "lock", we wouldd need to have the lock _held_ across the entire sequence. Without that there's no warranty that it will refer to the same object we'd created.
In any case, unlink in any potentially public area is pretty much never the right approach. Once mknod has happened, that's it - too late to bail out.
IIRC, most of the PITA in that area is due to unix_autobind() iteractions. Basically, we try to bind() an unbound socket and another thread does sendmsg() on the same while we are in the middle of ->mknod(). Who should wait for whom?
->mknod() really should be a point of no return - any games with "so we unlink it" are unreliable in the best case, and that's only if we do _not_ unlock the parent through the entire sequence.
Seeing that we have separate bindlock and iolock now... How about this (completely untested) delta?
net/unix/af_unix.c +9 -10
@@ -1034,6 +1034,14 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) goto out; addr_len = err; + err =mutex_lock_interruptible (&u->bindlock); +if (err) + goto out; + + err = -EINVAL ; +if (u->addr) + goto out_up; +if (sun_path[0 ]) {umode_t mode =S_IFSOCK | (SOCK_INODE (sock)->i_mode & ~current_umask ());
@@ -1041,18 +1049,10 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (err) {if (err == -EEXIST ) err = -EADDRINUSE ; - goto out; + goto out_up; } } - err =mutex_lock_interruptible (&u->bindlock); -if (err) - goto out_put; - - err = -EINVAL ; -if (u->addr) - goto out_up; - err = -ENOMEM ; addr =kmalloc (sizeof (*addr)+addr_len,GFP_KERNEL );if (!addr)
@@ -1090,7 +1090,6 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)spin_unlock (&unix_table_lock); out_up:mutex_unlock (&u->bindlock); -out_put:if (err)path_put (&path); out:
Quoted 19 lines · Al Viro
Quoted 4 lines · Christoph Hellwig
Quoted 1 line · Denis Kirjanov
Do you mean just:
We'll still need to lock the parent inode.
Not just "lock", we wouldd need to have the lock _held_ across the entire sequence. Without that there's no warranty that it will refer to the same object we'd created.
In any case, unlink in any potentially public area is pretty much never the right approach. Once mknod has happened, that's it - too late to bail out.
IIRC, most of the PITA in that area is due to
unix_autobind()iteractions. Basically, we try tobind()an unbound socket and another thread doessendmsg()on the same while we are in the middle of ->mknod(). Who should wait for whom?
Quoted 6 lines
->mknod() really should be a point of no return - any games with "so we unlink it" are unreliable in the best case, and that's only if we do _not_ unlock the parent through the entire sequence.
Seeing that we have separate bindlock and iolock now... How about this (completely untested) delta?
We had a change like that:
Author: WANG Cong <[email protected]> Date: Mon Jan 23 11:17:35 2017 -0800
af_unix: move unix_mknod() out of bindlock
Dmitry reported a deadlock scenario:
unix_bind() path:
u->bindlock ==> sb_writer
do_splice() path:
sb_writer ==> pipe->mutex ==> u->bindlock
In the unix_bind() code path, unix_mknod() does not have to be done with u->bindlock held, since it is a pure fs operation, so we can just move unix_mknod() out.
Quoted 52 lines
net/unix/af_unix.c +8
-0@@ -1034,6 +1034,14 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) goto out; addr_len = err; + err =mutex_lock_interruptible (&u->bindlock); +if (err) + goto out; + + err = -EINVAL ; +if (u->addr) + goto out_up; +if (sun_path[0 ]) { umode_t mode = S_IFSOCK |net/unix/af_unix.c +1
-9(SOCK_INODE (sock)->i_mode & ~current_umask ());@@ -1041,18 +1049,10 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) if (err) {if (err == -EEXIST ) err = -EADDRINUSE ; - goto out; + goto out_up; } } - err =mutex_lock_interruptible (&u->bindlock); -if (err) - goto out_put; - - err = -EINVAL ; -if (u->addr) - goto out_up; - err = -ENOMEM ; addr =kmalloc (sizeof (*addr)+addr_len,GFP_KERNEL );net/unix/af_unix.c +0
-1if (!addr)@@ -1090,7 +1090,6 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) spin_unlock(&unix_table_lock); out_up:mutex_unlock (&u->bindlock); -out_put:if (err)path_put (&path);out:
Al Viro
maintainerQuoted 11 lines · Denis Kirjanov
We had a change like that:
Author: WANG Cong <[email protected]> Date: Mon Jan 23 11:17:35 2017 -0800
af_unix: moveunix_mknod()out of bindlockDmitry reported a deadlock scenario:
unix_bind()path:
u->bindlock==>sb_writer
Quoted 6 lines
do_splice()path:
sb_writer==>pipe->mutex==>u->bindlockIn the
unix_bind()code path,unix_mknod()does not have to be done withu->bindlockheld, since it is a pure fs operation, so we can just moveunix_mknod()out.
*cringe*
I remember now... Process set:
P1: bind() of AF_UNIX socket to /mnt/sock
P2: splice() from pipe to /mnt/foo
P3: freeze /mnt
P4: splice() from pipe to AF_UNIX socket
P1 grabs ->bindlock
P2 sb_start_write() for what's on /mnt
P2 grabs rwsem shared
P3 blocks in sb_wait_write() trying to grab the same rwsem exclusive
P1 sb_start_write() blocks trying to grab the same rwsem shared
P4 calls ->splice_write(), aka generic_splice_sendpage()
P4 grabs pipe->mutex
P4 calls ->sendpage(), aka sock_no_sendpage()
P4 calls ->sendmsg(), aka unix_dgram_sendmsg()
P4 calls unix_autobind()
P4 blocks trying to grab ->bindlock
P2 ->splice_write(), aka iter_file_splice_write()
P2 blocks trying to grab pipe->mutex
DEADLOCK
Sigh... OK, so we want something like
user_path_create() vfs_mknod() created = true grab bindlock .... drop bindlock if failed && created vfs_unlink() done_path_create()
in unix_bind()... That would push ->bindlock all way down in the hierarchy, so that should be deadlock-free, but it looks like that'll be fucking ugly ;-/
Let me try and play with that a bit, maybe it can be massaged to something relatively sane...
Al Viro
maintainerQuoted 10 lines · Al Viro
Sigh... OK, so we want something like
user_path_create() vfs_mknod() created = true grab bindlock .... drop bindlock if failed && created vfs_unlink() done_path_create()
Quoted 5 lines
in
unix_bind()... That would push ->bindlock all way down in the hierarchy, so that should be deadlock-free, but it looks like that'll be fucking ugly ;-/Let me try and play with that a bit, maybe it can be massaged to something relatively sane...
OK... Completely untested series follows. Preliminary massage in first 6 patches, then actual "add cleanup on failure", then minor followup cleanup.
af_unix: take address assignment/hash insertion into a new helper unix_bind(): allocate addr earlier
unix_bind(): separate BSD and abstract cases
unix_bind(): take BSD and abstract address cases into new helpers fold unix_mknod() into unix_bind_bsd()
unix_bind_bsd(): move done_path_create() call after dealing with ->bindlock unix_bind_bsd(): unlink if we fail after successful mknod __unix_find_socket_byname(): don't pass hash and type separately
Branch is in git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git #misc.af_unix, individual patches in followups
Al Viro
maintainerPATCH 1/8 af_unix: take address assignment/hash insertion into a new helper
Duplicated logics in all bind variants (autobind, bind-to-path, bind-to-abstract) gets taken into a common helper.
--- net/unix/af_unix.c | 30 +++++++++++++++--------------- 1 file changed, 15 insertions(+), 15 deletions(-)
net/unix/af_unix.c +15 -15
@@ -262,6 +262,16 @@ static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)sk_add_node (sk, list); } +static void __unix_set_addr (struct sock *sk,struct unix_address *addr, +unsigned hash) + __releases(&unix_table_lock) +{ +__unix_remove_socket (sk); +smp_store_release (&unix_sk (sk)->addr, addr); +__unix_insert_socket (&unix_socket_table[hash], sk); +spin_unlock (&unix_table_lock); +} +static inline void unix_remove_socket (struct sock *sk) {spin_lock (&unix_table_lock);
@@ -912,10 +922,7 @@ static int unix_autobind(struct socket *sock) } addr->hash ^= sk->sk_type; -__unix_remove_socket (sk); -smp_store_release (&u->addr, addr); -__unix_insert_socket (&unix_socket_table[addr->hash], sk); -spin_unlock (&unix_table_lock); +__unix_set_addr (sk, addr, addr->hash); err =0 ; out:mutex_unlock (&u->bindlock);
@@ -1016,7 +1023,6 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)int err;unsigned int hash;struct unix_address *addr; -struct hlist_head *list;struct path path = { }; err = -EINVAL ;
@@ -1068,26 +1074,20 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 );spin_lock (&unix_table_lock); u->path = path; - list = &unix_socket_table[hash]; }else {spin_lock (&unix_table_lock); err = -EADDRINUSE ;if (__unix_find_socket_byname (net, sunaddr, addr_len, sk->sk_type, hash)) { +spin_unlock (&unix_table_lock);unix_release_addr (addr); - goto out_unlock; + goto out_up; } - - list = &unix_socket_table[addr->hash]; + hash = addr->hash; } +__unix_set_addr (sk, addr, hash); err =0 ; -__unix_remove_socket (sk); -smp_store_release (&u->addr, addr); -__unix_insert_socket (list, sk); - -out_unlock: -spin_unlock (&unix_table_lock); out_up:mutex_unlock (&u->bindlock); out_put:
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 2/8 unix_bind(): allocate addr earlier
In reply to Al ViroFrom 24439dbb7b78cb301c73254b364020e6a3f31902 Mon Sep 17 00:00:00 2001
From: Al Viro <[email protected]>
Date: Thu, 18 Feb 2021 15:52:53 -0500
Subject: [PATCH 2/8] unix_bind(): allocate addr earlier
makes it easier to massage; we do pay for that by extra work (kmalloc+memcpy+kfree) in some error cases, but those are not on the hot paths anyway.
--- net/unix/af_unix.c | 26 ++++++++++++++------------ 1 file changed, 14 insertions(+), 12 deletions(-)
net/unix/af_unix.c +14 -12
@@ -1039,6 +1039,15 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (err <0 ) goto out; addr_len = err; + err = -ENOMEM ; + addr =kmalloc (sizeof (*addr)+addr_len,GFP_KERNEL ); +if (!addr) + goto out; + +memcpy (addr->name, sunaddr, addr_len); + addr->len = addr_len; + addr->hash = hash ^ sk->sk_type; +refcount_set (&addr->refcnt,1 );if (sun_path[0 ]) { umode_t mode = S_IFSOCK |
@@ -1047,7 +1056,7 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (err) {if (err == -EEXIST ) err = -EADDRINUSE ; - goto out; + goto out_addr; } }
@@ -1059,16 +1068,6 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (u->addr) goto out_up; - err = -ENOMEM ; - addr =kmalloc (sizeof (*addr)+addr_len,GFP_KERNEL ); -if (!addr) - goto out_up; - -memcpy (addr->name, sunaddr, addr_len); - addr->len = addr_len; - addr->hash = hash ^ sk->sk_type; -refcount_set (&addr->refcnt,1 ); -if (sun_path[0 ]) { addr->hash =UNIX_HASH_SIZE ; hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 );
@@ -1080,19 +1079,22 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (__unix_find_socket_byname (net, sunaddr, addr_len, sk->sk_type, hash)) {spin_unlock (&unix_table_lock); -unix_release_addr (addr); goto out_up; } hash = addr->hash; }__unix_set_addr (sk, addr, hash); + addr =NULL ; err =0 ; out_up:mutex_unlock (&u->bindlock); out_put:if (err)path_put (&path); +out_addr: +if (addr) +unix_release_addr (addr); out:return err; }
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 3/8 unix_bind(): separate BSD and abstract cases
In reply to Al ViroWe do get some duplication that way, but it's minor compared to parts that are different. What we get is an ability to change locking in BSD case without making failure exits very hard to follow.
--- net/unix/af_unix.c | 52 ++++++++++++++++++++++++++++++++-------------------- 1 file changed, 32 insertions(+), 20 deletions(-)
net/unix/af_unix.c +32 -20
@@ -1023,7 +1023,6 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)int err;unsigned int hash;struct unix_address *addr; -struct path path = { }; err = -EINVAL ; if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
@@ -1050,6 +1049,7 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)refcount_set (&addr->refcnt,1 );if (sun_path[0 ]) { +struct path path = { };umode_t mode =S_IFSOCK | (SOCK_INODE (sock)->i_mode & ~current_umask ()); err =unix_mknod (sun_path, mode, &path);
@@ -1058,40 +1058,52 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) err = -EADDRINUSE ; goto out_addr; } - } - err =mutex_lock_interruptible (&u->bindlock); -if (err) - goto out_put; + err =mutex_lock_interruptible (&u->bindlock); +if (err) { +path_put (&path); + goto out_addr; + } - err = -EINVAL ; -if (u->addr) - goto out_up; + err = -EINVAL ; +if (u->addr) { +mutex_unlock (&u->bindlock); +path_put (&path); + goto out_addr; + } -if (sun_path[0 ]) { addr->hash =UNIX_HASH_SIZE ; hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 );spin_lock (&unix_table_lock); u->path = path; +__unix_set_addr (sk, addr, hash); +mutex_unlock (&u->bindlock); + addr =NULL ; + err =0 ; }else { + err =mutex_lock_interruptible (&u->bindlock); +if (err) + goto out_addr; + + err = -EINVAL ; +if (u->addr) { +mutex_unlock (&u->bindlock); + goto out_addr; + } +spin_lock (&unix_table_lock); err = -EADDRINUSE ;if (__unix_find_socket_byname (net, sunaddr, addr_len, sk->sk_type, hash)) {spin_unlock (&unix_table_lock); - goto out_up; +mutex_unlock (&u->bindlock); + goto out_addr; } - hash = addr->hash; +__unix_set_addr (sk, addr, addr->hash); +mutex_unlock (&u->bindlock); + addr =NULL ; + err =0 ; } - -__unix_set_addr (sk, addr, hash); - addr =NULL ; - err =0 ; -out_up: -mutex_unlock (&u->bindlock); -out_put: -if (err) -path_put (&path); out_addr:if (addr)unix_release_addr (addr);
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 4/8 unix_bind(): take BSD and abstract address cases into new helpers
In reply to Al Virounix_bind_bsd() and unix_bind_abstract() respectively.
--- net/unix/af_unix.c | 144 +++++++++++++++++++++++++++-------------------------- 1 file changed, 74 insertions(+), 70 deletions(-)
net/unix/af_unix.c +74 -70
@@ -1013,101 +1013,105 @@ static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)return err; } +static int unix_bind_bsd (struct sock *sk,struct unix_address *addr) +{ +struct unix_sock *u =unix_sk (sk); +struct path path = { }; +umode_t mode =S_IFSOCK | + (SOCK_INODE (sk->sk_socket)->i_mode & ~current_umask ()); +unsigned int hash; +int err; + + err =unix_mknod (addr->name->sun_path, mode, &path); +if (err) { +if (err == -EEXIST ) + err = -EADDRINUSE ; +return err; + } + + err =mutex_lock_interruptible (&u->bindlock); +if (err) { +path_put (&path); +return err; + } + +if (u->addr) { +mutex_unlock (&u->bindlock); +path_put (&path); +return -EINVAL ; + } + + addr->hash =UNIX_HASH_SIZE ; + hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 ); +spin_lock (&unix_table_lock); + u->path = path; +__unix_set_addr (sk, addr, hash); +mutex_unlock (&u->bindlock); +return 0 ; +} + +static int unix_bind_abstract (struct sock *sk,unsigned hash, +struct unix_address *addr) +{ +struct unix_sock *u =unix_sk (sk); +int err; + + err =mutex_lock_interruptible (&u->bindlock); +if (err) +return err; + +if (u->addr) { +mutex_unlock (&u->bindlock); +return -EINVAL ; + } + +spin_lock (&unix_table_lock); +if (__unix_find_socket_byname (sock_net (sk), addr->name, addr->len, + sk->sk_type, hash)) { +spin_unlock (&unix_table_lock); +mutex_unlock (&u->bindlock); +return -EADDRINUSE ; + } +__unix_set_addr (sk, addr, addr->hash); +mutex_unlock (&u->bindlock); +return 0 ; +} +static int unix_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len) {struct sock *sk = sock->sk; -struct net *net =sock_net (sk); -struct unix_sock *u =unix_sk (sk);struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;char *sun_path = sunaddr->sun_path;int err;unsigned int hash;struct unix_address *addr; - err = -EINVAL ; if (addr_len < offsetofend(struct sockaddr_un, sun_family) || sunaddr->sun_family !=AF_UNIX ) - goto out; +return -EINVAL ; -if (addr_len ==sizeof (short )) { - err =unix_autobind (sock); - goto out; - } +if (addr_len ==sizeof (short )) +return unix_autobind (sock); err =unix_mkname (sunaddr, addr_len, &hash);if (err <0 ) - goto out; +return err; addr_len = err; - err = -ENOMEM ; addr =kmalloc (sizeof (*addr)+addr_len,GFP_KERNEL );if (!addr) - goto out; +return -ENOMEM ;memcpy (addr->name, sunaddr, addr_len); addr->len = addr_len; addr->hash = hash ^ sk->sk_type;refcount_set (&addr->refcnt,1 ); -if (sun_path[0 ]) { -struct path path = { }; -umode_t mode =S_IFSOCK | - (SOCK_INODE (sock)->i_mode & ~current_umask ()); - err =unix_mknod (sun_path, mode, &path); -if (err) { -if (err == -EEXIST ) - err = -EADDRINUSE ; - goto out_addr; - } - - err =mutex_lock_interruptible (&u->bindlock); -if (err) { -path_put (&path); - goto out_addr; - } - - err = -EINVAL ; -if (u->addr) { -mutex_unlock (&u->bindlock); -path_put (&path); - goto out_addr; - } - - addr->hash =UNIX_HASH_SIZE ; - hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 ); -spin_lock (&unix_table_lock); - u->path = path; -__unix_set_addr (sk, addr, hash); -mutex_unlock (&u->bindlock); - addr =NULL ; - err =0 ; - }else { - err =mutex_lock_interruptible (&u->bindlock); -if (err) - goto out_addr; - - err = -EINVAL ; -if (u->addr) { -mutex_unlock (&u->bindlock); - goto out_addr; - } - -spin_lock (&unix_table_lock); - err = -EADDRINUSE ; -if (__unix_find_socket_byname (net, sunaddr, addr_len, - sk->sk_type, hash)) { -spin_unlock (&unix_table_lock); -mutex_unlock (&u->bindlock); - goto out_addr; - } -__unix_set_addr (sk, addr, addr->hash); -mutex_unlock (&u->bindlock); - addr =NULL ; - err =0 ; - } -out_addr: -if (addr) +if (sun_path[0 ]) + err =unix_bind_bsd (sk, addr); +else + err =unix_bind_abstract (sk, hash, addr); +if (err)unix_release_addr (addr); -out:return err; }
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 5/8 fold unix_mknod() into unix_bind_bsd()
In reply to Al Viro--- net/unix/af_unix.c | 39 +++++++++++++++------------------------ 1 file changed, 15 insertions(+), 24 deletions(-)
net/unix/af_unix.c +15 -24
@@ -984,45 +984,36 @@ static struct sock *unix_find_other(struct net *net,return NULL ; } -static int unix_mknod (const char *sun_path,umode_t mode,struct path *res) +static int unix_bind_bsd (struct sock *sk,struct unix_address *addr) { +struct unix_sock *u =unix_sk (sk); +umode_t mode =S_IFSOCK | + (SOCK_INODE (sk->sk_socket)->i_mode & ~current_umask ()); +struct path parent, path;struct dentry *dentry; -struct path path; -int err =0 ; +unsigned int hash; +int err; +/* * Get the parent directory, calculate the hash for last * component. */ - dentry =kern_path_create (AT_FDCWD , sun_path, &path,0 ); - err =PTR_ERR (dentry); + dentry =kern_path_create (AT_FDCWD , addr->name->sun_path, &parent,0 );if (IS_ERR (dentry)) -return err; +return PTR_ERR (dentry);/* * All right, let's create it. */ - err =security_path_mknod (&path, dentry, mode,0 ); + err =security_path_mknod (&parent, dentry, mode,0 );if (!err) { - err =vfs_mknod (d_inode (path.dentry), dentry, mode,0 ); + err =vfs_mknod (d_inode (parent.dentry), dentry, mode,0 );if (!err) { - res->mnt =mntget (path.mnt); - res->dentry =dget (dentry); + path.mnt =mntget (parent.mnt); + path.dentry =dget (dentry); } } -done_path_create (&path, dentry); -return err; -} - -static int unix_bind_bsd (struct sock *sk,struct unix_address *addr) -{ -struct unix_sock *u =unix_sk (sk); -struct path path = { }; -umode_t mode =S_IFSOCK | - (SOCK_INODE (sk->sk_socket)->i_mode & ~current_umask ()); -unsigned int hash; -int err; - - err =unix_mknod (addr->name->sun_path, mode, &path); +done_path_create (&parent, dentry);if (err) {if (err == -EEXIST ) err = -EADDRINUSE ;
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 6/8 unix_bind_bsd(): move done_path_create() call after dealing with ->bindlock
In reply to Al ViroFinal preparations for doing unlink on failure past the successful mknod. We can't hold ->bindlock over ->mknod() or ->unlink(), since either might do sb_start_write() (e.g. on overlayfs). However, we can do it while holding filesystem and VFS locks - doing
kern_path_create() vfs_mknod() grab ->bindlock if u->addr had been set drop ->bindlock done_path_create return -EINVAL else assign the address to socket drop ->bindlock done_path_create return 0
would be deadlock-free. Here we massage unix_bind_bsd() to that form. We are still doing equivalent transformations.
Next commit will *not* be an equivalent transformation - it will add a call of vfs_unlink() before done_path_create() in "alread bound" case.
--- net/unix/af_unix.c | 24 ++++++++++-------------- 1 file changed, 10 insertions(+), 14 deletions(-)
net/unix/af_unix.c +10 -14
@@ -989,7 +989,7 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)struct unix_sock *u =unix_sk (sk);umode_t mode =S_IFSOCK | (SOCK_INODE (sk->sk_socket)->i_mode & ~current_umask ()); -struct path parent, path; +struct path parent;struct dentry *dentry;unsigned int hash;int err;
@@ -1006,38 +1006,34 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)* All right, let's create it. */ err =security_path_mknod (&parent, dentry, mode,0 ); -if (!err) { +if (!err) err =vfs_mknod (d_inode (parent.dentry), dentry, mode,0 ); -if (!err) { - path.mnt =mntget (parent.mnt); - path.dentry =dget (dentry); - } - } -done_path_create (&parent, dentry); +if (err) {if (err == -EEXIST ) err = -EADDRINUSE ; +done_path_create (&parent, dentry);return err; } - err =mutex_lock_interruptible (&u->bindlock);if (err) { -path_put (&path); +done_path_create (&parent, dentry);return err; } -if (u->addr) {mutex_unlock (&u->bindlock); -path_put (&path); +done_path_create (&parent, dentry);return -EINVAL ; } addr->hash =UNIX_HASH_SIZE ; - hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 ); + hash =d_backing_inode (dentry)->i_ino & (UNIX_HASH_SIZE -1 );spin_lock (&unix_table_lock); - u->path = path; + u->path.mnt =mntget (parent.mnt); + u->path.dentry =dget (dentry);__unix_set_addr (sk, addr, hash);mutex_unlock (&u->bindlock); +done_path_create (&parent, dentry);return 0 ; }
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 7/8 unix_bind_bsd(): unlink if we fail after successful mknod
In reply to Al ViroWe can do that more or less safely, since the parent is held locked all along. Yes, somebody might observe the object via dcache, only to have it disappear afterwards, but there's really no good way to prevent that. It won't race with other bind(2) or attempts to move the sucker elsewhere, or put something else in its place - locked parent prevents that.
--- net/unix/af_unix.c | 33 ++++++++++++++++----------------- 1 file changed, 16 insertions(+), 17 deletions(-)
net/unix/af_unix.c +16 -17
@@ -1001,30 +1001,19 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr) dentry =kern_path_create (AT_FDCWD , addr->name->sun_path, &parent,0 );if (IS_ERR (dentry))return PTR_ERR (dentry); -/* * All right, let's create it. */ err =security_path_mknod (&parent, dentry, mode,0 );if (!err) err =vfs_mknod (d_inode (parent.dentry), dentry, mode,0 ); - -if (err) { -if (err == -EEXIST ) - err = -EADDRINUSE ; -done_path_create (&parent, dentry); -return err; - } +if (err) + goto out; err =mutex_lock_interruptible (&u->bindlock); -if (err) { -done_path_create (&parent, dentry); -return err; - } -if (u->addr) { -mutex_unlock (&u->bindlock); -done_path_create (&parent, dentry); -return -EINVAL ; - } +if (err) + goto out_unlink; +if (u->addr) + goto out_unlock; addr->hash =UNIX_HASH_SIZE ; hash =d_backing_inode (dentry)->i_ino & (UNIX_HASH_SIZE -1 );
@@ -1035,6 +1024,16 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)mutex_unlock (&u->bindlock);done_path_create (&parent, dentry);return 0 ; + +out_unlock: +mutex_unlock (&u->bindlock); + err = -EINVAL ; +out_unlink: +/* failed after successful mknod? unlink what we'd created... */ +vfs_unlink (d_inode (parent.dentry), dentry,NULL ); +out: +done_path_create (&parent, dentry); +return err == -EEXIST ? -EADDRINUSE : err; } static int unix_bind_abstract(struct sock *sk, unsigned hash,
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 8/8 __unix_find_socket_byname(): don't pass hash and type separately
In reply to Al ViroWe only care about exclusive or of those, so pass that directly. Makes life simpler for callers as well...
--- net/unix/af_unix.c | 23 ++++++++++------------- 1 file changed, 10 insertions(+), 13 deletions(-)
net/unix/af_unix.c +10 -13
@@ -288,11 +288,11 @@ static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)static struct sock *__unix_find_socket_byname (struct net *net,struct sockaddr_un *sunname, -int len,int type,unsigned int hash) +int len,unsigned int hash) {struct sock *s; -sk_for_each (s, &unix_socket_table[hash ^ type]) { +sk_for_each (s, &unix_socket_table[hash]) {struct unix_sock *u =unix_sk (s);if (!net_eq (sock_net (s), net))
@@ -307,13 +307,12 @@ static struct sock *__unix_find_socket_byname(struct net *net,static inline struct sock *unix_find_socket_byname (struct net *net,struct sockaddr_un *sunname, -int len,int type, -unsigned int hash) +int len,unsigned int hash) {struct sock *s;spin_lock (&unix_table_lock); - s =__unix_find_socket_byname (net, sunname, len, type, hash); + s =__unix_find_socket_byname (net, sunname, len, hash);if (s)sock_hold (s);spin_unlock (&unix_table_lock);
@@ -900,12 +899,12 @@ static int unix_autobind(struct socket *sock) retry: addr->len =sprintf (addr->name->sun_path+1 ,"%05x" , ordernum) +1 +sizeof (short ); addr->hash =unix_hash_fold (csum_partial (addr->name, addr->len,0 )); + addr->hash ^= sk->sk_type;spin_lock (&unix_table_lock); ordernum = (ordernum+1 )&0xFFFFF ; -if (__unix_find_socket_byname (net, addr->name, addr->len, sock->type, - addr->hash)) { +if (__unix_find_socket_byname (net, addr->name, addr->len, addr->hash)) {spin_unlock (&unix_table_lock);/* * __unix_find_socket_byname() may take long time if many names
@@ -920,7 +919,6 @@ static int unix_autobind(struct socket *sock) } goto retry; } - addr->hash ^= sk->sk_type;__unix_set_addr (sk, addr, addr->hash); err =0 ;
@@ -966,7 +964,7 @@ static struct sock *unix_find_other(struct net *net, } }else { err = -ECONNREFUSED ; - u =unix_find_socket_byname (net, sunname, len, type, hash); + u =unix_find_socket_byname (net, sunname, len, type ^ hash);if (u) {struct dentry *dentry; dentry =unix_sk (u)->path.dentry;
@@ -1036,8 +1034,7 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)return err == -EEXIST ? -EADDRINUSE : err; } -static int unix_bind_abstract (struct sock *sk,unsigned hash, -struct unix_address *addr) +static int unix_bind_abstract (struct sock *sk,struct unix_address *addr) {struct unix_sock *u =unix_sk (sk);int err;
@@ -1053,7 +1050,7 @@ static int unix_bind_abstract(struct sock *sk, unsigned hash,spin_lock (&unix_table_lock);if (__unix_find_socket_byname (sock_net (sk), addr->name, addr->len, - sk->sk_type, hash)) { + addr->hash)) {spin_unlock (&unix_table_lock);mutex_unlock (&u->bindlock);return -EADDRINUSE ;
@@ -1095,7 +1092,7 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (sun_path[0 ]) err =unix_bind_bsd (sk, addr);else - err =unix_bind_abstract (sk, hash, addr); + err =unix_bind_abstract (sk, addr);if (err)unix_release_addr (addr);return err;
Signature · 1 line
2.11.0
Re: [PATCH 1/8] af_unix: take address assignment/hash insertion into a new helper
In reply to Al ViroQuoted 15 lines · Al Viro
Duplicated logics in all bind variants (autobind, bind-to-path, bind-to-abstract) gets taken into a common helper.
Signed-off-by Al Viro <[email protected]>--- net/unix/af_unix.c | 30 +++++++++++++++--------------- 1 file changed, 15 insertions(+), 15 deletions(-)net/unix/af_unix.c +0
-0@@ -262,6 +262,16 @@ static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)sk_add_node (sk, list);
net/unix/af_unix.c static void __unix_insert_socket(struct hlist_head *list, struct sock *sk) · line 272
Quoted 10 lines
net/unix/af_unix.c +8
-0} +static void __unix_set_addr (struct sock *sk,struct unix_address *addr, +unsigned hash) + __releases(&unix_table_lock) +{ +__unix_remove_socket (sk); +smp_store_release (&unix_sk (sk)->addr, addr); +__unix_insert_socket (&unix_socket_table[hash], sk); +spin_unlock (&unix_table_lock);
Please take the unlock out, it is clearly an anti-pattern.
And please Cc netdev for networking changes.
Thanks.
Al Viro
maintainerRe: [PATCH 1/8] af_unix: take address assignment/hash insertion into a new helper
Quoted 18 lines · Cong Wang
Quoted 17 lines · Al Viro
Duplicated logics in all bind variants (autobind, bind-to-path, bind-to-abstract) gets taken into a common helper.
Signed-off-by Al Viro <[email protected]>--- net/unix/af_unix.c | 30 +++++++++++++++--------------- 1 file changed, 15 insertions(+), 15 deletions(-)net/unix/af_unix.c +0
-0@@ -262,6 +262,16 @@ static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)sk_add_node (sk, list); }
Quoted 10 lines
net/unix/af_unix.c static void __unix_insert_socket(struct hlist_head *list, struct sock *sk) · line 272
Quoted 8 lines
net/unix/af_unix.c +8
-0+static void __unix_set_addr (struct sock *sk,struct unix_address *addr, +unsigned hash) + __releases(&unix_table_lock) +{ +__unix_remove_socket (sk); +smp_store_release (&unix_sk (sk)->addr, addr); +__unix_insert_socket (&unix_socket_table[hash], sk); +spin_unlock (&unix_table_lock);Please take the unlock out, it is clearly an anti-pattern.
Why? "Insert into locked and unlock" is fairly common...
Quoted 1 line
And please Cc netdev for networking changes.
Sorry about that - I'd ended up emulating git send-email by hand and missed the Cc. Sorted out by now...
Re: [PATCH 1/8] af_unix: take address assignment/hash insertion into a new helper
Quoted 31 lines · Al Viro
Quoted 18 lines · Cong Wang
Quoted 17 lines · Al Viro
Duplicated logics in all bind variants (autobind, bind-to-path, bind-to-abstract) gets taken into a common helper.
Signed-off-by Al Viro <[email protected]>--- net/unix/af_unix.c | 30 +++++++++++++++--------------- 1 file changed, 15 insertions(+), 15 deletions(-)net/unix/af_unix.c +0
-0@@ -262,6 +262,16 @@ static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)sk_add_node (sk, list); }Quoted 10 lines
net/unix/af_unix.c static void __unix_insert_socket(struct hlist_head *list, struct sock *sk) · line 272
Quoted 8 lines
net/unix/af_unix.c +8
-0+static void __unix_set_addr (struct sock *sk,struct unix_address *addr, +unsigned hash) + __releases(&unix_table_lock) +{ +__unix_remove_socket (sk); +smp_store_release (&unix_sk (sk)->addr, addr); +__unix_insert_socket (&unix_socket_table[hash], sk); +spin_unlock (&unix_table_lock);Please take the unlock out, it is clearly an anti-pattern.
Quoted 1 line
Why? "Insert into locked and unlock" is fairly common...
Because it does not lock the lock, just compare:
lock();__unix_set_addr();unlock();
to:
lock();__unix_set_addr();
Clearly the former is more readable and less error-prone. Even if you really want to do unlock, pick a name which explicitly says it, for example, __unix_set_addr_unlock().
Thanks.
Al Viro
maintainerRe: [PATCH 1/8] af_unix: take address assignment/hash insertion into a new helper
Quoted 6 lines · Cong Wang
Because it does not lock the lock, just compare:
lock();__unix_set_addr();unlock();
Quoted 8 lines
to:
lock();__unix_set_addr();Clearly the former is more readable and less error-prone. Even if you really want to do unlock, pick a name which explicitly says it, for example,
__unix_set_addr_unlock().
*shrug*
If anything, __unix_complete_bind() might make a better name for that, with dropping ->bindlock also pulled in, but TBH I don't have sufficiently strong preferences - might as well leave dropping the lock to caller.
I'll post that series to netdev tonight.
Al Viro
maintainerPATCH 1/1 making unix_bind() undo mknod on failure
Quoted 7 lines · Al Viro
*shrug*
If anything,
__unix_complete_bind()might make a better name for that, with dropping ->bindlock also pulled in, but TBH I don't have sufficiently strong preferences - might as well leave dropping the lock to caller.I'll post that series to netdev tonight.
Took longer than I hoped... Anyway, here's the current variant;
it's 5.11-based, lives ingit://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git misc.af_unix
Shortlog:
Al Viro (8):
af_unix: take address assignment/hash insertion into a new helper unix_bind(): allocate addr earlier
unix_bind(): separate BSD and abstract cases
unix_bind(): take BSD and abstract address cases into new helpers fold unix_mknod() into unix_bind_bsd()
unix_bind_bsd(): move done_path_create() call after dealing with ->bindlock unix_bind_bsd(): unlink if we fail after successful mknod __unix_find_socket_byname(): don't pass hash and type separately
Diffstat:
net/unix/af_unix.c | 186 +++++++++++++++++++++++++++-------------------------- 1 file changed, 94 insertions(+), 92 deletions(-)
The actual fix is in #7/8, the first 6 are massage in preparation to that and #8/8 is a minor followup cleanup. Individual patches in followups. Please, review.
Al Viro
maintainerPATCH 3/8 unix_bind(): separate BSD and abstract cases
In reply to Al ViroWe do get some duplication that way, but it's minor compared to parts that are different. What we get is an ability to change locking in BSD case without making failure exits very hard to follow.
--- net/unix/af_unix.c | 55 +++++++++++++++++++++++++++++++++--------------------- 1 file changed, 34 insertions(+), 21 deletions(-)
net/unix/af_unix.c +34 -21
@@ -1022,7 +1022,6 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)int err;unsigned int hash;struct unix_address *addr; -struct path path = { }; err = -EINVAL ; if (addr_len < offsetofend(struct sockaddr_un, sun_family) ||
@@ -1049,6 +1048,7 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)refcount_set (&addr->refcnt,1 );if (sun_path[0 ]) { +struct path path = { };umode_t mode =S_IFSOCK | (SOCK_INODE (sock)->i_mode & ~current_umask ()); err =unix_mknod (sun_path, mode, &path);
@@ -1057,41 +1057,54 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) err = -EADDRINUSE ; goto out_addr; } - } - err =mutex_lock_interruptible (&u->bindlock); -if (err) - goto out_put; + err =mutex_lock_interruptible (&u->bindlock); +if (err) { +path_put (&path); + goto out_addr; + } - err = -EINVAL ; -if (u->addr) - goto out_up; + err = -EINVAL ; +if (u->addr) { +mutex_unlock (&u->bindlock); +path_put (&path); + goto out_addr; + } -if (sun_path[0 ]) { addr->hash =UNIX_HASH_SIZE ; hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 );spin_lock (&unix_table_lock); u->path = path; +__unix_set_addr (sk, addr, hash); +spin_unlock (&unix_table_lock); +mutex_unlock (&u->bindlock); + addr =NULL ; + err =0 ; }else { + err =mutex_lock_interruptible (&u->bindlock); +if (err) + goto out_addr; + + err = -EINVAL ; +if (u->addr) { +mutex_unlock (&u->bindlock); + goto out_addr; + } +spin_lock (&unix_table_lock); err = -EADDRINUSE ;if (__unix_find_socket_byname (net, sunaddr, addr_len, sk->sk_type, hash)) {spin_unlock (&unix_table_lock); - goto out_up; +mutex_unlock (&u->bindlock); + goto out_addr; } - hash = addr->hash; +__unix_set_addr (sk, addr, addr->hash); +spin_unlock (&unix_table_lock); +mutex_unlock (&u->bindlock); + addr =NULL ; + err =0 ; } - -__unix_set_addr (sk, addr, hash); -spin_unlock (&unix_table_lock); - addr =NULL ; - err =0 ; -out_up: -mutex_unlock (&u->bindlock); -out_put: -if (err) -path_put (&path); out_addr:if (addr)unix_release_addr (addr);
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 2/8 unix_bind(): allocate addr earlier
In reply to Al Viromakes it easier to massage; we do pay for that by extra work (kmalloc+memcpy+kfree) in some error cases, but those are not on the hot paths anyway.
--- net/unix/af_unix.c | 28 +++++++++++++++------------- 1 file changed, 15 insertions(+), 13 deletions(-)
net/unix/af_unix.c +15 -13
@@ -1038,6 +1038,15 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (err <0 ) goto out; addr_len = err; + err = -ENOMEM ; + addr =kmalloc (sizeof (*addr)+addr_len,GFP_KERNEL ); +if (!addr) + goto out; + +memcpy (addr->name, sunaddr, addr_len); + addr->len = addr_len; + addr->hash = hash ^ sk->sk_type; +refcount_set (&addr->refcnt,1 );if (sun_path[0 ]) { umode_t mode = S_IFSOCK |
@@ -1046,7 +1055,7 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (err) {if (err == -EEXIST ) err = -EADDRINUSE ; - goto out; + goto out_addr; } }
@@ -1058,16 +1067,6 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (u->addr) goto out_up; - err = -ENOMEM ; - addr =kmalloc (sizeof (*addr)+addr_len,GFP_KERNEL ); -if (!addr) - goto out_up; - -memcpy (addr->name, sunaddr, addr_len); - addr->len = addr_len; - addr->hash = hash ^ sk->sk_type; -refcount_set (&addr->refcnt,1 ); -if (sun_path[0 ]) { addr->hash =UNIX_HASH_SIZE ; hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 );
@@ -1079,20 +1078,23 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (__unix_find_socket_byname (net, sunaddr, addr_len, sk->sk_type, hash)) {spin_unlock (&unix_table_lock); -unix_release_addr (addr); goto out_up; } hash = addr->hash; } - err =0 ;__unix_set_addr (sk, addr, hash);spin_unlock (&unix_table_lock); + addr =NULL ; + err =0 ; out_up:mutex_unlock (&u->bindlock); out_put:if (err)path_put (&path); +out_addr: +if (addr) +unix_release_addr (addr); out:return err; }
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 7/8 unix_bind_bsd(): unlink if we fail after successful mknod
In reply to Al ViroWe can do that more or less safely, since the parent is held locked all along. Yes, somebody might observe the object via dcache, only to have it disappear afterwards, but there's really no good way to prevent that. It won't race with other bind(2) or attempts to move the sucker elsewhere, or put something else in its place - locked parent prevents that.
--- net/unix/af_unix.c | 30 ++++++++++++++++-------------- 1 file changed, 16 insertions(+), 14 deletions(-)
net/unix/af_unix.c +16 -14
@@ -1000,27 +1000,19 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr) dentry =kern_path_create (AT_FDCWD , addr->name->sun_path, &parent,0 );if (IS_ERR (dentry))return PTR_ERR (dentry); -/* * All right, let's create it. */ err =security_path_mknod (&parent, dentry, mode,0 );if (!err) err =vfs_mknod (d_inode (parent.dentry), dentry, mode,0 ); -if (err) { -done_path_create (&parent, dentry); -return err; - } +if (err) + goto out; err =mutex_lock_interruptible (&u->bindlock); -if (err) { -done_path_create (&parent, dentry); -return err; - } -if (u->addr) { -mutex_unlock (&u->bindlock); -done_path_create (&parent, dentry); -return -EINVAL ; - } +if (err) + goto out_unlink; +if (u->addr) + goto out_unlock; addr->hash =UNIX_HASH_SIZE ; hash =d_backing_inode (dentry)->i_ino & (UNIX_HASH_SIZE -1 );
@@ -1032,6 +1024,16 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)mutex_unlock (&u->bindlock);done_path_create (&parent, dentry);return 0 ; + +out_unlock: +mutex_unlock (&u->bindlock); + err = -EINVAL ; +out_unlink: +/* failed after successful mknod? unlink what we'd created... */ +vfs_unlink (d_inode (parent.dentry), dentry,NULL ); +out: +done_path_create (&parent, dentry); +return err; } static int unix_bind_abstract(struct sock *sk, unsigned hash,
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 6/8 unix_bind_bsd(): move done_path_create() call after dealing with ->bindlock
In reply to Al ViroFinal preparations for doing unlink on failure past the successful mknod. We can't hold ->bindlock over ->mknod() or ->unlink(), since either might do sb_start_write() (e.g. on overlayfs). However, we can do it while holding filesystem and VFS locks - doing
kern_path_create() vfs_mknod() grab ->bindlock if u->addr had been set drop ->bindlock done_path_create return -EINVAL else assign the address to socket drop ->bindlock done_path_create return 0
would be deadlock-free. Here we massage unix_bind_bsd() to that form. We are still doing equivalent transformations.
Next commit will *not* be an equivalent transformation - it will add a call of vfs_unlink() before done_path_create() in "alread bound" case.
--- net/unix/af_unix.c | 26 +++++++++++--------------- 1 file changed, 11 insertions(+), 15 deletions(-)
net/unix/af_unix.c +11 -15
@@ -988,7 +988,7 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)struct unix_sock *u =unix_sk (sk);umode_t mode =S_IFSOCK | (SOCK_INODE (sk->sk_socket)->i_mode & ~current_umask ()); -struct path parent, path; +struct path parent;struct dentry *dentry;unsigned int hash;int err;
@@ -1005,36 +1005,32 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)* All right, let's create it. */ err =security_path_mknod (&parent, dentry, mode,0 ); -if (!err) { +if (!err) err =vfs_mknod (d_inode (parent.dentry), dentry, mode,0 ); -if (!err) { - path.mnt =mntget (parent.mnt); - path.dentry =dget (dentry); - } - } -done_path_create (&parent, dentry); -if (err) +if (err) { +done_path_create (&parent, dentry);return err; - + } err =mutex_lock_interruptible (&u->bindlock);if (err) { -path_put (&path); +done_path_create (&parent, dentry);return err; } -if (u->addr) {mutex_unlock (&u->bindlock); -path_put (&path); +done_path_create (&parent, dentry);return -EINVAL ; } addr->hash =UNIX_HASH_SIZE ; - hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 ); + hash =d_backing_inode (dentry)->i_ino & (UNIX_HASH_SIZE -1 );spin_lock (&unix_table_lock); - u->path = path; + u->path.mnt =mntget (parent.mnt); + u->path.dentry =dget (dentry);__unix_set_addr (sk, addr, hash);spin_unlock (&unix_table_lock);mutex_unlock (&u->bindlock); +done_path_create (&parent, dentry);return 0 ; }
Signature · 1 line
2.11.0
Al Viro
maintainerRe: [PATCHSET] making unix_bind() undo mknod on failure
In reply to Al ViroQuoted 25 lines · Al Viro
Quoted 7 lines · Al Viro
*shrug*
If anything,
__unix_complete_bind()might make a better name for that, with dropping ->bindlock also pulled in, but TBH I don't have sufficiently strong preferences - might as well leave dropping the lock to caller.I'll post that series to netdev tonight.
Took longer than I hoped... Anyway, here's the current variant;it's 5.11-based, lives in
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.gitmisc.af_unixShortlog:
Al Viro (8):
af_unix: take address assignment/hash insertion into a new helperunix_bind(): allocate addr earlier
unix_bind(): separate BSD and abstract cases
unix_bind(): take BSD and abstract address cases into new helpers foldunix_mknod()intounix_bind_bsd()
unix_bind_bsd(): movedone_path_create()call after dealing with ->bindlockunix_bind_bsd(): unlink if we fail after successful mknod__unix_find_socket_byname(): don't pass hash and type separately
Quoted 7 lines
Diffstat:
net/unix/af_unix.c | 186 +++++++++++++++++++++++++++-------------------------- 1 file changed, 94 insertions(+), 92 deletions(-)The actual fix is in #7/8, the first 6 are massage in preparation to that and #8/8 is a minor followup cleanup. Individual patches in followups. Please, review.
Argh... git send-email is playing silly buggers again ;-/
Al Viro
maintainerPATCH 1/8 af_unix: take address assignment/hash insertion into a new helper
In reply to Al ViroDuplicated logics in all bind variants (autobind, bind-to-path, bind-to-abstract) gets taken into a common helper.
--- net/unix/af_unix.c | 26 +++++++++++++------------- 1 file changed, 13 insertions(+), 13 deletions(-)
net/unix/af_unix.c +13 -13
@@ -262,6 +262,14 @@ static void __unix_insert_socket(struct hlist_head *list, struct sock *sk)sk_add_node (sk, list); } +static void __unix_set_addr (struct sock *sk,struct unix_address *addr, +unsigned hash) +{ +__unix_remove_socket (sk); +smp_store_release (&unix_sk (sk)->addr, addr); +__unix_insert_socket (&unix_socket_table[hash], sk); +} +static inline void unix_remove_socket (struct sock *sk) {spin_lock (&unix_table_lock);
@@ -912,9 +920,7 @@ static int unix_autobind(struct socket *sock) } addr->hash ^= sk->sk_type; -__unix_remove_socket (sk); -smp_store_release (&u->addr, addr); -__unix_insert_socket (&unix_socket_table[addr->hash], sk); +__unix_set_addr (sk, addr, addr->hash);spin_unlock (&unix_table_lock); err =0 ;
@@ -1016,7 +1022,6 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)int err;unsigned int hash;struct unix_address *addr; -struct hlist_head *list;struct path path = { }; err = -EINVAL ;
@@ -1068,25 +1073,20 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 );spin_lock (&unix_table_lock); u->path = path; - list = &unix_socket_table[hash]; }else {spin_lock (&unix_table_lock); err = -EADDRINUSE ;if (__unix_find_socket_byname (net, sunaddr, addr_len, sk->sk_type, hash)) { +spin_unlock (&unix_table_lock);unix_release_addr (addr); - goto out_unlock; + goto out_up; } - - list = &unix_socket_table[addr->hash]; + hash = addr->hash; } err =0 ; -__unix_remove_socket (sk); -smp_store_release (&u->addr, addr); -__unix_insert_socket (list, sk); - -out_unlock: +__unix_set_addr (sk, addr, hash);spin_unlock (&unix_table_lock); out_up:mutex_unlock (&u->bindlock);
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 8/8 __unix_find_socket_byname(): don't pass hash and type separately
We only care about exclusive or of those, so pass that directly. Makes life simpler for callers as well...
--- net/unix/af_unix.c | 23 ++++++++++------------- 1 file changed, 10 insertions(+), 13 deletions(-)
net/unix/af_unix.c +10 -13
@@ -286,11 +286,11 @@ static inline void unix_insert_socket(struct hlist_head *list, struct sock *sk)static struct sock *__unix_find_socket_byname (struct net *net,struct sockaddr_un *sunname, -int len,int type,unsigned int hash) +int len,unsigned int hash) {struct sock *s; -sk_for_each (s, &unix_socket_table[hash ^ type]) { +sk_for_each (s, &unix_socket_table[hash]) {struct unix_sock *u =unix_sk (s);if (!net_eq (sock_net (s), net))
@@ -305,13 +305,12 @@ static struct sock *__unix_find_socket_byname(struct net *net,static inline struct sock *unix_find_socket_byname (struct net *net,struct sockaddr_un *sunname, -int len,int type, -unsigned int hash) +int len,unsigned int hash) {struct sock *s;spin_lock (&unix_table_lock); - s =__unix_find_socket_byname (net, sunname, len, type, hash); + s =__unix_find_socket_byname (net, sunname, len, hash);if (s)sock_hold (s);spin_unlock (&unix_table_lock);
@@ -898,12 +897,12 @@ static int unix_autobind(struct socket *sock) retry: addr->len =sprintf (addr->name->sun_path+1 ,"%05x" , ordernum) +1 +sizeof (short ); addr->hash =unix_hash_fold (csum_partial (addr->name, addr->len,0 )); + addr->hash ^= sk->sk_type;spin_lock (&unix_table_lock); ordernum = (ordernum+1 )&0xFFFFF ; -if (__unix_find_socket_byname (net, addr->name, addr->len, sock->type, - addr->hash)) { +if (__unix_find_socket_byname (net, addr->name, addr->len, addr->hash)) {spin_unlock (&unix_table_lock);/* * __unix_find_socket_byname() may take long time if many names
@@ -918,7 +917,6 @@ static int unix_autobind(struct socket *sock) } goto retry; } - addr->hash ^= sk->sk_type;__unix_set_addr (sk, addr, addr->hash);spin_unlock (&unix_table_lock);
@@ -965,7 +963,7 @@ static struct sock *unix_find_other(struct net *net, } }else { err = -ECONNREFUSED ; - u =unix_find_socket_byname (net, sunname, len, type, hash); + u =unix_find_socket_byname (net, sunname, len, type ^ hash);if (u) {struct dentry *dentry; dentry =unix_sk (u)->path.dentry;
@@ -1036,8 +1034,7 @@ static int unix_bind_bsd(struct sock *sk, struct unix_address *addr)return err; } -static int unix_bind_abstract (struct sock *sk,unsigned hash, -struct unix_address *addr) +static int unix_bind_abstract (struct sock *sk,struct unix_address *addr) {struct unix_sock *u =unix_sk (sk);int err;
@@ -1053,7 +1050,7 @@ static int unix_bind_abstract(struct sock *sk, unsigned hash,spin_lock (&unix_table_lock);if (__unix_find_socket_byname (sock_net (sk), addr->name, addr->len, - sk->sk_type, hash)) { + addr->hash)) {spin_unlock (&unix_table_lock);mutex_unlock (&u->bindlock);return -EADDRINUSE ;
@@ -1096,7 +1093,7 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)if (sun_path[0 ]) err =unix_bind_bsd (sk, addr);else - err =unix_bind_abstract (sk, hash, addr); + err =unix_bind_abstract (sk, addr);if (err)unix_release_addr (addr);return err == -EEXIST ? -EADDRINUSE : err;
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 4/8 unix_bind(): take BSD and abstract address cases into new helpers
In reply to Al Virounix_bind_bsd() and unix_bind_abstract() respectively.
--- net/unix/af_unix.c | 147 +++++++++++++++++++++++++++-------------------------- 1 file changed, 74 insertions(+), 73 deletions(-)
net/unix/af_unix.c +74 -73
@@ -1012,104 +1012,105 @@ static int unix_mknod(const char *sun_path, umode_t mode, struct path *res)return err; } +static int unix_bind_bsd (struct sock *sk,struct unix_address *addr) +{ +struct unix_sock *u =unix_sk (sk); +struct path path = { }; +umode_t mode =S_IFSOCK | + (SOCK_INODE (sk->sk_socket)->i_mode & ~current_umask ()); +unsigned int hash; +int err; + + err =unix_mknod (addr->name->sun_path, mode, &path); +if (err) +return err; + + err =mutex_lock_interruptible (&u->bindlock); +if (err) { +path_put (&path); +return err; + } + +if (u->addr) { +mutex_unlock (&u->bindlock); +path_put (&path); +return -EINVAL ; + } + + addr->hash =UNIX_HASH_SIZE ; + hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 ); +spin_lock (&unix_table_lock); + u->path = path; +__unix_set_addr (sk, addr, hash); +spin_unlock (&unix_table_lock); +mutex_unlock (&u->bindlock); +return 0 ; +} + +static int unix_bind_abstract (struct sock *sk,unsigned hash, +struct unix_address *addr) +{ +struct unix_sock *u =unix_sk (sk); +int err; + + err =mutex_lock_interruptible (&u->bindlock); +if (err) +return err; + +if (u->addr) { +mutex_unlock (&u->bindlock); +return -EINVAL ; + } + +spin_lock (&unix_table_lock); +if (__unix_find_socket_byname (sock_net (sk), addr->name, addr->len, + sk->sk_type, hash)) { +spin_unlock (&unix_table_lock); +mutex_unlock (&u->bindlock); +return -EADDRINUSE ; + } +__unix_set_addr (sk, addr, addr->hash); +spin_unlock (&unix_table_lock); +mutex_unlock (&u->bindlock); +return 0 ; +} +static int unix_bind (struct socket *sock,struct sockaddr *uaddr,int addr_len) {struct sock *sk = sock->sk; -struct net *net =sock_net (sk); -struct unix_sock *u =unix_sk (sk);struct sockaddr_un *sunaddr = (struct sockaddr_un *)uaddr;char *sun_path = sunaddr->sun_path;int err;unsigned int hash;struct unix_address *addr; - err = -EINVAL ; if (addr_len < offsetofend(struct sockaddr_un, sun_family) || sunaddr->sun_family !=AF_UNIX ) - goto out; +return -EINVAL ; -if (addr_len ==sizeof (short )) { - err =unix_autobind (sock); - goto out; - } +if (addr_len ==sizeof (short )) +return unix_autobind (sock); err =unix_mkname (sunaddr, addr_len, &hash);if (err <0 ) - goto out; +return err; addr_len = err; - err = -ENOMEM ; addr =kmalloc (sizeof (*addr)+addr_len,GFP_KERNEL );if (!addr) - goto out; +return -ENOMEM ;memcpy (addr->name, sunaddr, addr_len); addr->len = addr_len; addr->hash = hash ^ sk->sk_type;refcount_set (&addr->refcnt,1 ); -if (sun_path[0 ]) { -struct path path = { }; -umode_t mode =S_IFSOCK | - (SOCK_INODE (sock)->i_mode & ~current_umask ()); - err =unix_mknod (sun_path, mode, &path); -if (err) { -if (err == -EEXIST ) - err = -EADDRINUSE ; - goto out_addr; - } - - err =mutex_lock_interruptible (&u->bindlock); -if (err) { -path_put (&path); - goto out_addr; - } - - err = -EINVAL ; -if (u->addr) { -mutex_unlock (&u->bindlock); -path_put (&path); - goto out_addr; - } - - addr->hash =UNIX_HASH_SIZE ; - hash =d_backing_inode (path.dentry)->i_ino & (UNIX_HASH_SIZE -1 ); -spin_lock (&unix_table_lock); - u->path = path; -__unix_set_addr (sk, addr, hash); -spin_unlock (&unix_table_lock); -mutex_unlock (&u->bindlock); - addr =NULL ; - err =0 ; - }else { - err =mutex_lock_interruptible (&u->bindlock); -if (err) - goto out_addr; - - err = -EINVAL ; -if (u->addr) { -mutex_unlock (&u->bindlock); - goto out_addr; - } - -spin_lock (&unix_table_lock); - err = -EADDRINUSE ; -if (__unix_find_socket_byname (net, sunaddr, addr_len, - sk->sk_type, hash)) { -spin_unlock (&unix_table_lock); -mutex_unlock (&u->bindlock); - goto out_addr; - } -__unix_set_addr (sk, addr, addr->hash); -spin_unlock (&unix_table_lock); -mutex_unlock (&u->bindlock); - addr =NULL ; - err =0 ; - } -out_addr: -if (addr) +if (sun_path[0 ]) + err =unix_bind_bsd (sk, addr); +else + err =unix_bind_abstract (sk, hash, addr); +if (err)unix_release_addr (addr); -out: -return err; +return err == -EEXIST ? -EADDRINUSE : err; }static void unix_state_double_lock (struct sock *sk1,struct sock *sk2)
Signature · 1 line
2.11.0
Al Viro
maintainerPATCH 5/8 fold unix_mknod() into unix_bind_bsd()
In reply to Al Viro--- net/unix/af_unix.c | 39 +++++++++++++++------------------------ 1 file changed, 15 insertions(+), 24 deletions(-)
net/unix/af_unix.c +15 -24
@@ -983,45 +983,36 @@ static struct sock *unix_find_other(struct net *net,return NULL ; } -static int unix_mknod (const char *sun_path,umode_t mode,struct path *res) +static int unix_bind_bsd (struct sock *sk,struct unix_address *addr) { +struct unix_sock *u =unix_sk (sk); +umode_t mode =S_IFSOCK | + (SOCK_INODE (sk->sk_socket)->i_mode & ~current_umask ()); +struct path parent, path;struct dentry *dentry; -struct path path; -int err =0 ; +unsigned int hash; +int err; +/* * Get the parent directory, calculate the hash for last * component. */ - dentry =kern_path_create (AT_FDCWD , sun_path, &path,0 ); - err =PTR_ERR (dentry); + dentry =kern_path_create (AT_FDCWD , addr->name->sun_path, &parent,0 );if (IS_ERR (dentry)) -return err; +return PTR_ERR (dentry);/* * All right, let's create it. */ - err =security_path_mknod (&path, dentry, mode,0 ); + err =security_path_mknod (&parent, dentry, mode,0 );if (!err) { - err =vfs_mknod (d_inode (path.dentry), dentry, mode,0 ); + err =vfs_mknod (d_inode (parent.dentry), dentry, mode,0 );if (!err) { - res->mnt =mntget (path.mnt); - res->dentry =dget (dentry); + path.mnt =mntget (parent.mnt); + path.dentry =dget (dentry); } } -done_path_create (&path, dentry); -return err; -} - -static int unix_bind_bsd (struct sock *sk,struct unix_address *addr) -{ -struct unix_sock *u =unix_sk (sk); -struct path path = { }; -umode_t mode =S_IFSOCK | - (SOCK_INODE (sk->sk_socket)->i_mode & ~current_umask ()); -unsigned int hash; -int err; - - err =unix_mknod (addr->name->sun_path, mode, &path); +done_path_create (&parent, dentry);if (err)return err;
Signature · 1 line
2.11.0
Al Viro
maintainerRe: [PATCHSET] making unix_bind() undo mknod on failure
In reply to Al ViroQuoted 30 lines · Al Viro
Quoted 29 lines · Al Viro
Quoted 7 lines · Al Viro
*shrug*
If anything,
__unix_complete_bind()might make a better name for that, with dropping ->bindlock also pulled in, but TBH I don't have sufficiently strong preferences - might as well leave dropping the lock to caller.I'll post that series to netdev tonight.
Took longer than I hoped... Anyway, here's the current variant;it's 5.11-based, lives in
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.gitmisc.af_unixShortlog:
Al Viro (8):
af_unix: take address assignment/hash insertion into a new helperunix_bind(): allocate addr earlier
unix_bind(): separate BSD and abstract cases
unix_bind(): take BSD and abstract address cases into new helpers foldunix_mknod()intounix_bind_bsd()
unix_bind_bsd(): movedone_path_create()call after dealing with ->bindlockunix_bind_bsd(): unlink if we fail after successful mknod__unix_find_socket_byname(): don't pass hash and type separatelyDiffstat:
net/unix/af_unix.c | 186 +++++++++++++++++++++++++++-------------------------- 1 file changed, 94 insertions(+), 92 deletions(-)
Quoted 5 lines
Quoted 3 lines
The actual fix is in #7/8, the first 6 are massage in preparation to that and #8/8 is a minor followup cleanup. Individual patches in followups. Please, review.
Argh... git send-email is playing silly buggers again ;-/
Cute - looks like having EMAIL in environment confuses the living hell out git-send-email. Oh, well...
Re: [PATCHSET] making unix_bind() undo mknod on failure
In reply to Al ViroQuoted 25 lines · On Mon, 22 Feb 2021 19:06:00 +0000 Al Viro wrote:
Quoted 7 lines · Al Viro
*shrug*
If anything,
__unix_complete_bind()might make a better name for that, with dropping ->bindlock also pulled in, but TBH I don't have sufficiently strong preferences - might as well leave dropping the lock to caller.I'll post that series to netdev tonight.
Took longer than I hoped... Anyway, here's the current variant;it's 5.11-based, lives in
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.gitmisc.af_unixShortlog:
Al Viro (8):
af_unix: take address assignment/hash insertion into a new helperunix_bind(): allocate addr earlier
unix_bind(): separate BSD and abstract cases
unix_bind(): take BSD and abstract address cases into new helpers foldunix_mknod()intounix_bind_bsd()
unix_bind_bsd(): movedone_path_create()call after dealing with ->bindlockunix_bind_bsd(): unlink if we fail after successful mknod__unix_find_socket_byname(): don't pass hash and type separately
Quoted 6 lines
Diffstat:
net/unix/af_unix.c | 186 +++++++++++++++++++++++++++-------------------------- 1 file changed, 94 insertions(+), 92 deletions(-)The actual fix is in #7/8, the first 6 are massage in preparation to that and #8/8 is a minor followup cleanup. Individual patches in followups.
Dave is out this week, but this looks good to me. You said "please review" - I'm assuming you'll send these to Linus yourself, so: