fs/devpts/inode.c: correctly check d_alloc_name() return code in devpts_pty_new()
[linux-2.6.git] / fs / devpts / inode.c
index a70d5d0..2f27e57 100644 (file)
 #include <linux/fs.h>
 #include <linux/sched.h>
 #include <linux/namei.h>
+#include <linux/slab.h>
 #include <linux/mount.h>
 #include <linux/tty.h>
 #include <linux/mutex.h>
+#include <linux/magic.h>
 #include <linux/idr.h>
 #include <linux/devpts_fs.h>
 #include <linux/parser.h>
 #include <linux/fsnotify.h>
 #include <linux/seq_file.h>
 
-#define DEVPTS_SUPER_MAGIC 0x1cd1
-
 #define DEVPTS_DEFAULT_MODE 0600
+/*
+ * ptmx is a new node in /dev/pts and will be unused in legacy (single-
+ * instance) mode. To prevent surprises in user space, set permissions of
+ * ptmx to 0. Use 'chmod' or remount with '-o ptmxmode' to set meaningful
+ * permissions.
+ */
+#define DEVPTS_DEFAULT_PTMX_MODE 0000
 #define PTMX_MINOR     2
 
 extern int pty_limit;                  /* Config limit on Unix98 ptys */
-static DEFINE_IDA(allocated_ptys);
 static DEFINE_MUTEX(allocated_ptys_lock);
 
 static struct vfsmount *devpts_mnt;
-static struct dentry *devpts_root;
 
-static struct {
+struct pts_mount_opts {
        int setuid;
        int setgid;
        uid_t   uid;
        gid_t   gid;
        umode_t mode;
-} config = {.mode = DEVPTS_DEFAULT_MODE};
+       umode_t ptmxmode;
+       int newinstance;
+};
 
 enum {
-       Opt_uid, Opt_gid, Opt_mode,
+       Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance,
        Opt_err
 };
 
-static match_table_t tokens = {
+static const match_table_t tokens = {
        {Opt_uid, "uid=%u"},
        {Opt_gid, "gid=%u"},
        {Opt_mode, "mode=%o"},
+#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
+       {Opt_ptmxmode, "ptmxmode=%o"},
+       {Opt_newinstance, "newinstance"},
+#endif
        {Opt_err, NULL}
 };
 
-static int devpts_remount(struct super_block *sb, int *flags, char *data)
+struct pts_fs_info {
+       struct ida allocated_ptys;
+       struct pts_mount_opts mount_opts;
+       struct dentry *ptmx_dentry;
+};
+
+static inline struct pts_fs_info *DEVPTS_SB(struct super_block *sb)
+{
+       return sb->s_fs_info;
+}
+
+static inline struct super_block *pts_sb_from_inode(struct inode *inode)
+{
+#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
+       if (inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
+               return inode->i_sb;
+#endif
+       return devpts_mnt->mnt_sb;
+}
+
+#define PARSE_MOUNT    0
+#define PARSE_REMOUNT  1
+
+/*
+ * parse_mount_options():
+ *     Set @opts to mount options specified in @data. If an option is not
+ *     specified in @data, set it to its default value. The exception is
+ *     'newinstance' option which can only be set/cleared on a mount (i.e.
+ *     cannot be changed during remount).
+ *
+ * Note: @data may be NULL (in which case all options are set to default).
+ */
+static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 {
        char *p;
 
-       config.setuid  = 0;
-       config.setgid  = 0;
-       config.uid     = 0;
-       config.gid     = 0;
-       config.mode    = DEVPTS_DEFAULT_MODE;
+       opts->setuid  = 0;
+       opts->setgid  = 0;
+       opts->uid     = 0;
+       opts->gid     = 0;
+       opts->mode    = DEVPTS_DEFAULT_MODE;
+       opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
+
+       /* newinstance makes sense only on initial mount */
+       if (op == PARSE_MOUNT)
+               opts->newinstance = 0;
 
        while ((p = strsep(&data, ",")) != NULL) {
                substring_t args[MAX_OPT_ARGS];
@@ -79,20 +127,32 @@ static int devpts_remount(struct super_block *sb, int *flags, char *data)
                case Opt_uid:
                        if (match_int(&args[0], &option))
                                return -EINVAL;
-                       config.uid = option;
-                       config.setuid = 1;
+                       opts->uid = option;
+                       opts->setuid = 1;
                        break;
                case Opt_gid:
                        if (match_int(&args[0], &option))
                                return -EINVAL;
-                       config.gid = option;
-                       config.setgid = 1;
+                       opts->gid = option;
+                       opts->setgid = 1;
                        break;
                case Opt_mode:
                        if (match_octal(&args[0], &option))
                                return -EINVAL;
-                       config.mode = option & S_IALLUGO;
+                       opts->mode = option & S_IALLUGO;
+                       break;
+#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
+               case Opt_ptmxmode:
+                       if (match_octal(&args[0], &option))
+                               return -EINVAL;
+                       opts->ptmxmode = option & S_IALLUGO;
                        break;
+               case Opt_newinstance:
+                       /* newinstance makes sense only on initial mount */
+                       if (op == PARSE_MOUNT)
+                               opts->newinstance = 1;
+                       break;
+#endif
                default:
                        printk(KERN_ERR "devpts: called with bogus options\n");
                        return -EINVAL;
@@ -102,13 +162,103 @@ static int devpts_remount(struct super_block *sb, int *flags, char *data)
        return 0;
 }
 
+#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
+static int mknod_ptmx(struct super_block *sb)
+{
+       int mode;
+       int rc = -ENOMEM;
+       struct dentry *dentry;
+       struct inode *inode;
+       struct dentry *root = sb->s_root;
+       struct pts_fs_info *fsi = DEVPTS_SB(sb);
+       struct pts_mount_opts *opts = &fsi->mount_opts;
+
+       mutex_lock(&root->d_inode->i_mutex);
+
+       /* If we have already created ptmx node, return */
+       if (fsi->ptmx_dentry) {
+               rc = 0;
+               goto out;
+       }
+
+       dentry = d_alloc_name(root, "ptmx");
+       if (!dentry) {
+               printk(KERN_NOTICE "Unable to alloc dentry for ptmx node\n");
+               goto out;
+       }
+
+       /*
+        * Create a new 'ptmx' node in this mount of devpts.
+        */
+       inode = new_inode(sb);
+       if (!inode) {
+               printk(KERN_ERR "Unable to alloc inode for ptmx node\n");
+               dput(dentry);
+               goto out;
+       }
+
+       inode->i_ino = 2;
+       inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
+
+       mode = S_IFCHR|opts->ptmxmode;
+       init_special_inode(inode, mode, MKDEV(TTYAUX_MAJOR, 2));
+
+       d_add(dentry, inode);
+
+       fsi->ptmx_dentry = dentry;
+       rc = 0;
+out:
+       mutex_unlock(&root->d_inode->i_mutex);
+       return rc;
+}
+
+static void update_ptmx_mode(struct pts_fs_info *fsi)
+{
+       struct inode *inode;
+       if (fsi->ptmx_dentry) {
+               inode = fsi->ptmx_dentry->d_inode;
+               inode->i_mode = S_IFCHR|fsi->mount_opts.ptmxmode;
+       }
+}
+#else
+static inline void update_ptmx_mode(struct pts_fs_info *fsi)
+{
+       return;
+}
+#endif
+
+static int devpts_remount(struct super_block *sb, int *flags, char *data)
+{
+       int err;
+       struct pts_fs_info *fsi = DEVPTS_SB(sb);
+       struct pts_mount_opts *opts = &fsi->mount_opts;
+
+       err = parse_mount_options(data, PARSE_REMOUNT, opts);
+
+       /*
+        * parse_mount_options() restores options to default values
+        * before parsing and may have changed ptmxmode. So, update the
+        * mode in the inode too. Bogus options don't fail the remount,
+        * so do this even on error return.
+        */
+       update_ptmx_mode(fsi);
+
+       return err;
+}
+
 static int devpts_show_options(struct seq_file *seq, struct vfsmount *vfs)
 {
-       if (config.setuid)
-               seq_printf(seq, ",uid=%u", config.uid);
-       if (config.setgid)
-               seq_printf(seq, ",gid=%u", config.gid);
-       seq_printf(seq, ",mode=%03o", config.mode);
+       struct pts_fs_info *fsi = DEVPTS_SB(vfs->mnt_sb);
+       struct pts_mount_opts *opts = &fsi->mount_opts;
+
+       if (opts->setuid)
+               seq_printf(seq, ",uid=%u", opts->uid);
+       if (opts->setgid)
+               seq_printf(seq, ",gid=%u", opts->gid);
+       seq_printf(seq, ",mode=%03o", opts->mode);
+#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
+       seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
+#endif
 
        return 0;
 }
@@ -119,10 +269,25 @@ static const struct super_operations devpts_sops = {
        .show_options   = devpts_show_options,
 };
 
+static void *new_pts_fs_info(void)
+{
+       struct pts_fs_info *fsi;
+
+       fsi = kzalloc(sizeof(struct pts_fs_info), GFP_KERNEL);
+       if (!fsi)
+               return NULL;
+
+       ida_init(&fsi->allocated_ptys);
+       fsi->mount_opts.mode = DEVPTS_DEFAULT_MODE;
+       fsi->mount_opts.ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
+
+       return fsi;
+}
+
 static int
 devpts_fill_super(struct super_block *s, void *data, int silent)
 {
-       struct inode * inode;
+       struct inode *inode;
 
        s->s_blocksize = 1024;
        s->s_blocksize_bits = 10;
@@ -130,39 +295,132 @@ devpts_fill_super(struct super_block *s, void *data, int silent)
        s->s_op = &devpts_sops;
        s->s_time_gran = 1;
 
+       s->s_fs_info = new_pts_fs_info();
+       if (!s->s_fs_info)
+               goto fail;
+
        inode = new_inode(s);
        if (!inode)
-               goto fail;
+               goto free_fsi;
        inode->i_ino = 1;
        inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
-       inode->i_blocks = 0;
-       inode->i_uid = inode->i_gid = 0;
        inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
        inode->i_op = &simple_dir_inode_operations;
        inode->i_fop = &simple_dir_operations;
        inode->i_nlink = 2;
 
-       devpts_root = s->s_root = d_alloc_root(inode);
+       s->s_root = d_alloc_root(inode);
        if (s->s_root)
                return 0;
-       
-       printk("devpts: get root dentry failed\n");
+
+       printk(KERN_ERR "devpts: get root dentry failed\n");
        iput(inode);
+
+free_fsi:
+       kfree(s->s_fs_info);
 fail:
        return -ENOMEM;
 }
 
-static int devpts_get_sb(struct file_system_type *fs_type,
-       int flags, const char *dev_name, void *data, struct vfsmount *mnt)
+#ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
+static int compare_init_pts_sb(struct super_block *s, void *p)
 {
-       return get_sb_single(fs_type, flags, data, devpts_fill_super, mnt);
+       if (devpts_mnt)
+               return devpts_mnt->mnt_sb == s;
+       return 0;
+}
+
+/*
+ * devpts_mount()
+ *
+ *     If the '-o newinstance' mount option was specified, mount a new
+ *     (private) instance of devpts.  PTYs created in this instance are
+ *     independent of the PTYs in other devpts instances.
+ *
+ *     If the '-o newinstance' option was not specified, mount/remount the
+ *     initial kernel mount of devpts.  This type of mount gives the
+ *     legacy, single-instance semantics.
+ *
+ *     The 'newinstance' option is needed to support multiple namespace
+ *     semantics in devpts while preserving backward compatibility of the
+ *     current 'single-namespace' semantics. i.e all mounts of devpts
+ *     without the 'newinstance' mount option should bind to the initial
+ *     kernel mount, like mount_single().
+ *
+ *     Mounts with 'newinstance' option create a new, private namespace.
+ *
+ *     NOTE:
+ *
+ *     For single-mount semantics, devpts cannot use mount_single(),
+ *     because mount_single()/sget() find and use the super-block from
+ *     the most recent mount of devpts. But that recent mount may be a
+ *     'newinstance' mount and mount_single() would pick the newinstance
+ *     super-block instead of the initial super-block.
+ */
+static struct dentry *devpts_mount(struct file_system_type *fs_type,
+       int flags, const char *dev_name, void *data)
+{
+       int error;
+       struct pts_mount_opts opts;
+       struct super_block *s;
+
+       error = parse_mount_options(data, PARSE_MOUNT, &opts);
+       if (error)
+               return ERR_PTR(error);
+
+       if (opts.newinstance)
+               s = sget(fs_type, NULL, set_anon_super, NULL);
+       else
+               s = sget(fs_type, compare_init_pts_sb, set_anon_super, NULL);
+
+       if (IS_ERR(s))
+               return ERR_CAST(s);
+
+       if (!s->s_root) {
+               s->s_flags = flags;
+               error = devpts_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
+               if (error)
+                       goto out_undo_sget;
+               s->s_flags |= MS_ACTIVE;
+       }
+
+       memcpy(&(DEVPTS_SB(s))->mount_opts, &opts, sizeof(opts));
+
+       error = mknod_ptmx(s);
+       if (error)
+               goto out_undo_sget;
+
+       return dget(s->s_root);
+
+out_undo_sget:
+       deactivate_locked_super(s);
+       return ERR_PTR(error);
+}
+
+#else
+/*
+ * This supports only the legacy single-instance semantics (no
+ * multiple-instance semantics)
+ */
+static struct dentry *devpts_mount(struct file_system_type *fs_type, int flags,
+               const char *dev_name, void *data)
+{
+       return mount_single(fs_type, flags, data, devpts_fill_super);
+}
+#endif
+
+static void devpts_kill_sb(struct super_block *sb)
+{
+       struct pts_fs_info *fsi = DEVPTS_SB(sb);
+
+       kfree(fsi);
+       kill_litter_super(sb);
 }
 
 static struct file_system_type devpts_fs_type = {
-       .owner          = THIS_MODULE,
        .name           = "devpts",
-       .get_sb         = devpts_get_sb,
-       .kill_sb        = kill_anon_super,
+       .mount          = devpts_mount,
+       .kill_sb        = devpts_kill_sb,
 };
 
 /*
@@ -172,16 +430,17 @@ static struct file_system_type devpts_fs_type = {
 
 int devpts_new_index(struct inode *ptmx_inode)
 {
+       struct super_block *sb = pts_sb_from_inode(ptmx_inode);
+       struct pts_fs_info *fsi = DEVPTS_SB(sb);
        int index;
        int ida_ret;
 
 retry:
-       if (!ida_pre_get(&allocated_ptys, GFP_KERNEL)) {
+       if (!ida_pre_get(&fsi->allocated_ptys, GFP_KERNEL))
                return -ENOMEM;
-       }
 
        mutex_lock(&allocated_ptys_lock);
-       ida_ret = ida_get_new(&allocated_ptys, &index);
+       ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
        if (ida_ret < 0) {
                mutex_unlock(&allocated_ptys_lock);
                if (ida_ret == -EAGAIN)
@@ -190,7 +449,7 @@ retry:
        }
 
        if (index >= pty_limit) {
-               ida_remove(&allocated_ptys, index);
+               ida_remove(&fsi->allocated_ptys, index);
                mutex_unlock(&allocated_ptys_lock);
                return -EIO;
        }
@@ -200,18 +459,27 @@ retry:
 
 void devpts_kill_index(struct inode *ptmx_inode, int idx)
 {
+       struct super_block *sb = pts_sb_from_inode(ptmx_inode);
+       struct pts_fs_info *fsi = DEVPTS_SB(sb);
+
        mutex_lock(&allocated_ptys_lock);
-       ida_remove(&allocated_ptys, idx);
+       ida_remove(&fsi->allocated_ptys, idx);
        mutex_unlock(&allocated_ptys_lock);
 }
 
 int devpts_pty_new(struct inode *ptmx_inode, struct tty_struct *tty)
 {
-       int number = tty->index; /* tty layer puts index from devpts_new_index() in here */
+       /* tty layer puts index from devpts_new_index() in here */
+       int number = tty->index;
        struct tty_driver *driver = tty->driver;
        dev_t device = MKDEV(driver->major, driver->minor_start+number);
        struct dentry *dentry;
-       struct inode *inode = new_inode(devpts_mnt->mnt_sb);
+       struct super_block *sb = pts_sb_from_inode(ptmx_inode);
+       struct inode *inode = new_inode(sb);
+       struct dentry *root = sb->s_root;
+       struct pts_fs_info *fsi = DEVPTS_SB(sb);
+       struct pts_mount_opts *opts = &fsi->mount_opts;
+       int ret = 0;
        char s[12];
 
        /* We're supposed to be given the slave end of a pty */
@@ -221,55 +489,72 @@ int devpts_pty_new(struct inode *ptmx_inode, struct tty_struct *tty)
        if (!inode)
                return -ENOMEM;
 
-       inode->i_ino = number+2;
-       inode->i_uid = config.setuid ? config.uid : current->fsuid;
-       inode->i_gid = config.setgid ? config.gid : current->fsgid;
+       inode->i_ino = number + 3;
+       inode->i_uid = opts->setuid ? opts->uid : current_fsuid();
+       inode->i_gid = opts->setgid ? opts->gid : current_fsgid();
        inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
-       init_special_inode(inode, S_IFCHR|config.mode, device);
+       init_special_inode(inode, S_IFCHR|opts->mode, device);
        inode->i_private = tty;
        tty->driver_data = inode;
 
        sprintf(s, "%d", number);
 
-       mutex_lock(&devpts_root->d_inode->i_mutex);
+       mutex_lock(&root->d_inode->i_mutex);
 
-       dentry = d_alloc_name(devpts_root, s);
-       if (!IS_ERR(dentry)) {
+       dentry = d_alloc_name(root, s);
+       if (dentry) {
                d_add(dentry, inode);
-               fsnotify_create(devpts_root->d_inode, dentry);
+               fsnotify_create(root->d_inode, dentry);
+       } else {
+               iput(inode);
+               ret = -ENOMEM;
        }
 
-       mutex_unlock(&devpts_root->d_inode->i_mutex);
+       mutex_unlock(&root->d_inode->i_mutex);
 
-       return 0;
+       return ret;
 }
 
 struct tty_struct *devpts_get_tty(struct inode *pts_inode, int number)
 {
+       struct dentry *dentry;
+       struct tty_struct *tty;
+
        BUG_ON(pts_inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 
+       /* Ensure dentry has not been deleted by devpts_pty_kill() */
+       dentry = d_find_alias(pts_inode);
+       if (!dentry)
+               return NULL;
+
+       tty = NULL;
        if (pts_inode->i_sb->s_magic == DEVPTS_SUPER_MAGIC)
-               return (struct tty_struct *)pts_inode->i_private;
-       return NULL;
+               tty = (struct tty_struct *)pts_inode->i_private;
+
+       dput(dentry);
+
+       return tty;
 }
 
 void devpts_pty_kill(struct tty_struct *tty)
 {
        struct inode *inode = tty->driver_data;
+       struct super_block *sb = pts_sb_from_inode(inode);
+       struct dentry *root = sb->s_root;
        struct dentry *dentry;
 
        BUG_ON(inode->i_rdev == MKDEV(TTYAUX_MAJOR, PTMX_MINOR));
 
-       mutex_lock(&devpts_root->d_inode->i_mutex);
+       mutex_lock(&root->d_inode->i_mutex);
 
        dentry = d_find_alias(inode);
-       if (dentry && !IS_ERR(dentry)) {
-               inode->i_nlink--;
-               d_delete(dentry);
-               dput(dentry);
-       }
 
-       mutex_unlock(&devpts_root->d_inode->i_mutex);
+       inode->i_nlink--;
+       d_delete(dentry);
+       dput(dentry);   /* d_alloc_name() in devpts_pty_new() */
+       dput(dentry);           /* d_find_alias above */
+
+       mutex_unlock(&root->d_inode->i_mutex);
 }
 
 static int __init init_devpts_fs(void)
@@ -277,18 +562,11 @@ static int __init init_devpts_fs(void)
        int err = register_filesystem(&devpts_fs_type);
        if (!err) {
                devpts_mnt = kern_mount(&devpts_fs_type);
-               if (IS_ERR(devpts_mnt))
+               if (IS_ERR(devpts_mnt)) {
                        err = PTR_ERR(devpts_mnt);
+                       unregister_filesystem(&devpts_fs_type);
+               }
        }
        return err;
 }
-
-static void __exit exit_devpts_fs(void)
-{
-       unregister_filesystem(&devpts_fs_type);
-       mntput(devpts_mnt);
-}
-
 module_init(init_devpts_fs)
-module_exit(exit_devpts_fs)
-MODULE_LICENSE("GPL");