file caps: always start with clear bprm->caps_*
[linux-2.6.git] / security / device_cgroup.c
index 15f2f80..5ba7870 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * dev_cgroup.c - device cgroup subsystem
+ * device_cgroup.c - device cgroup subsystem
  *
  * Copyright 2007 IBM Corp
  */
@@ -10,6 +10,7 @@
 #include <linux/list.h>
 #include <linux/uaccess.h>
 #include <linux/seq_file.h>
+#include <linux/rcupdate.h>
 
 #define ACC_MKNOD 1
 #define ACC_READ  2
 
 /*
  * whitelist locking rules:
- * cgroup_lock() cannot be taken under dev_cgroup->lock.
- * dev_cgroup->lock can be taken with or without cgroup_lock().
- *
- * modifications always require cgroup_lock
- * modifications to a list which is visible require the
- *   dev_cgroup->lock *and* cgroup_lock()
- * walking the list requires dev_cgroup->lock or cgroup_lock().
- *
- * reasoning: dev_whitelist_copy() needs to kmalloc, so needs
- *   a mutex, which the cgroup_lock() is.  Since modifying
- *   a visible list requires both locks, either lock can be
- *   taken for walking the list.
+ * hold cgroup_lock() for update/read.
+ * hold rcu_read_lock() for read.
  */
 
 struct dev_whitelist_item {
@@ -41,12 +32,12 @@ struct dev_whitelist_item {
        short type;
        short access;
        struct list_head list;
+       struct rcu_head rcu;
 };
 
 struct dev_cgroup {
        struct cgroup_subsys_state css;
        struct list_head whitelist;
-       spinlock_t lock;
 };
 
 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
@@ -59,6 +50,11 @@ static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
        return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
 }
 
+static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
+{
+       return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
+}
+
 struct cgroup_subsys devices_subsys;
 
 static int devcgroup_can_attach(struct cgroup_subsys *ss,
@@ -78,13 +74,9 @@ static int dev_whitelist_copy(struct list_head *dest, struct list_head *orig)
        struct dev_whitelist_item *wh, *tmp, *new;
 
        list_for_each_entry(wh, orig, list) {
-               new = kmalloc(sizeof(*wh), GFP_KERNEL);
+               new = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
                if (!new)
                        goto free_and_exit;
-               new->major = wh->major;
-               new->minor = wh->minor;
-               new->type = wh->type;
-               new->access = wh->access;
                list_add_tail(&new->list, dest);
        }
 
@@ -101,34 +93,50 @@ free_and_exit:
 /* Stupid prototype - don't bother combining existing entries */
 /*
  * called under cgroup_lock()
- * since the list is visible to other tasks, we need the spinlock also
  */
 static int dev_whitelist_add(struct dev_cgroup *dev_cgroup,
                        struct dev_whitelist_item *wh)
 {
-       struct dev_whitelist_item *whcopy;
+       struct dev_whitelist_item *whcopy, *walk;
 
-       whcopy = kmalloc(sizeof(*whcopy), GFP_KERNEL);
+       whcopy = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
        if (!whcopy)
                return -ENOMEM;
 
-       memcpy(whcopy, wh, sizeof(*whcopy));
-       spin_lock(&dev_cgroup->lock);
-       list_add_tail(&whcopy->list, &dev_cgroup->whitelist);
-       spin_unlock(&dev_cgroup->lock);
+       list_for_each_entry(walk, &dev_cgroup->whitelist, list) {
+               if (walk->type != wh->type)
+                       continue;
+               if (walk->major != wh->major)
+                       continue;
+               if (walk->minor != wh->minor)
+                       continue;
+
+               walk->access |= wh->access;
+               kfree(whcopy);
+               whcopy = NULL;
+       }
+
+       if (whcopy != NULL)
+               list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist);
        return 0;
 }
 
+static void whitelist_item_free(struct rcu_head *rcu)
+{
+       struct dev_whitelist_item *item;
+
+       item = container_of(rcu, struct dev_whitelist_item, rcu);
+       kfree(item);
+}
+
 /*
  * called under cgroup_lock()
- * since the list is visible to other tasks, we need the spinlock also
  */
 static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
                        struct dev_whitelist_item *wh)
 {
        struct dev_whitelist_item *walk, *tmp;
 
-       spin_lock(&dev_cgroup->lock);
        list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) {
                if (walk->type == DEV_ALL)
                        goto remove;
@@ -142,11 +150,10 @@ static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
 remove:
                walk->access &= ~wh->access;
                if (!walk->access) {
-                       list_del(&walk->list);
-                       kfree(walk);
+                       list_del_rcu(&walk->list);
+                       call_rcu(&walk->rcu, whitelist_item_free);
                }
        }
-       spin_unlock(&dev_cgroup->lock);
 }
 
 /*
@@ -174,7 +181,7 @@ static struct cgroup_subsys_state *devcgroup_create(struct cgroup_subsys *ss,
                }
                wh->minor = wh->major = ~0;
                wh->type = DEV_ALL;
-               wh->access = ACC_MKNOD | ACC_READ | ACC_WRITE;
+               wh->access = ACC_MASK;
                list_add(&wh->list, &dev_cgroup->whitelist);
        } else {
                parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
@@ -186,7 +193,6 @@ static struct cgroup_subsys_state *devcgroup_create(struct cgroup_subsys *ss,
                }
        }
 
-       spin_lock_init(&dev_cgroup->lock);
        return &dev_cgroup->css;
 }
 
@@ -208,7 +214,7 @@ static void devcgroup_destroy(struct cgroup_subsys *ss,
 #define DEVCG_DENY 2
 #define DEVCG_LIST 3
 
-#define MAJMINLEN 10
+#define MAJMINLEN 13
 #define ACCLEN 4
 
 static void set_access(char *acc, short access)
@@ -236,11 +242,10 @@ static char type_to_char(short type)
 
 static void set_majmin(char *str, unsigned m)
 {
-       memset(str, 0, MAJMINLEN);
        if (m == ~0)
-               sprintf(str, "*");
+               strcpy(str, "*");
        else
-               snprintf(str, MAJMINLEN, "%d", m);
+               sprintf(str, "%u", m);
 }
 
 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
@@ -250,15 +255,15 @@ static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
        struct dev_whitelist_item *wh;
        char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
 
-       spin_lock(&devcgroup->lock);
-       list_for_each_entry(wh, &devcgroup->whitelist, list) {
+       rcu_read_lock();
+       list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) {
                set_access(acc, wh->access);
                set_majmin(maj, wh->major);
                set_majmin(min, wh->minor);
                seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type),
                           maj, min, acc);
        }
-       spin_unlock(&devcgroup->lock);
+       rcu_read_unlock();
 
        return 0;
 }
@@ -286,7 +291,7 @@ static int may_access_whitelist(struct dev_cgroup *c,
                        continue;
                if (whitem->minor != ~0 && whitem->minor != refwh->minor)
                        continue;
-               if (refwh->access & (~(whitem->access | ACC_MASK)))
+               if (refwh->access & (~whitem->access))
                        continue;
                return 1;
        }
@@ -298,20 +303,16 @@ static int may_access_whitelist(struct dev_cgroup *c,
  * when adding a new allow rule to a device whitelist, the rule
  * must be allowed in the parent device
  */
-static int parent_has_perm(struct cgroup *childcg,
+static int parent_has_perm(struct dev_cgroup *childcg,
                                  struct dev_whitelist_item *wh)
 {
-       struct cgroup *pcg = childcg->parent;
+       struct cgroup *pcg = childcg->css.cgroup->parent;
        struct dev_cgroup *parent;
-       int ret;
 
        if (!pcg)
                return 1;
        parent = cgroup_to_devcgroup(pcg);
-       spin_lock(&parent->lock);
-       ret = may_access_whitelist(parent, wh);
-       spin_unlock(&parent->lock);
-       return ret;
+       return may_access_whitelist(parent, wh);
 }
 
 /*
@@ -327,40 +328,17 @@ static int parent_has_perm(struct cgroup *childcg,
  * new access is only allowed if you're in the top-level cgroup, or your
  * parent cgroup has the access you're asking for.
  */
-static ssize_t devcgroup_access_write(struct cgroup *cgroup, struct cftype *cft,
-                               struct file *file, const char __user *userbuf,
-                               size_t nbytes, loff_t *ppos)
+static int devcgroup_update_access(struct dev_cgroup *devcgroup,
+                                  int filetype, const char *buffer)
 {
-       struct cgroup *cur_cgroup;
-       struct dev_cgroup *devcgroup, *cur_devcgroup;
-       int filetype = cft->private;
-       char *buffer, *b;
-       int retval = 0, count;
+       const char *b;
+       char *endp;
+       int count;
        struct dev_whitelist_item wh;
 
        if (!capable(CAP_SYS_ADMIN))
                return -EPERM;
 
-       devcgroup = cgroup_to_devcgroup(cgroup);
-       cur_cgroup = task_cgroup(current, devices_subsys.subsys_id);
-       cur_devcgroup = cgroup_to_devcgroup(cur_cgroup);
-
-       buffer = kmalloc(nbytes+1, GFP_KERNEL);
-       if (!buffer)
-               return -ENOMEM;
-
-       if (copy_from_user(buffer, userbuf, nbytes)) {
-               retval = -EFAULT;
-               goto out1;
-       }
-       buffer[nbytes] = 0;     /* nul-terminate */
-
-       cgroup_lock();
-       if (cgroup_is_removed(cgroup)) {
-               retval = -ENODEV;
-               goto out2;
-       }
-
        memset(&wh, 0, sizeof(wh));
        b = buffer;
 
@@ -368,6 +346,8 @@ static ssize_t devcgroup_access_write(struct cgroup *cgroup, struct cftype *cft,
        case 'a':
                wh.type = DEV_ALL;
                wh.access = ACC_MASK;
+               wh.major = ~0;
+               wh.minor = ~0;
                goto handle;
        case 'b':
                wh.type = DEV_BLOCK;
@@ -376,32 +356,23 @@ static ssize_t devcgroup_access_write(struct cgroup *cgroup, struct cftype *cft,
                wh.type = DEV_CHAR;
                break;
        default:
-               retval = -EINVAL;
-               goto out2;
+               return -EINVAL;
        }
        b++;
-       if (!isspace(*b)) {
-               retval = -EINVAL;
-               goto out2;
-       }
+       if (!isspace(*b))
+               return -EINVAL;
        b++;
        if (*b == '*') {
                wh.major = ~0;
                b++;
        } else if (isdigit(*b)) {
-               wh.major = 0;
-               while (isdigit(*b)) {
-                       wh.major = wh.major*10+(*b-'0');
-                       b++;
-               }
+               wh.major = simple_strtoul(b, &endp, 10);
+               b = endp;
        } else {
-               retval = -EINVAL;
-               goto out2;
-       }
-       if (*b != ':') {
-               retval = -EINVAL;
-               goto out2;
+               return -EINVAL;
        }
+       if (*b != ':')
+               return -EINVAL;
        b++;
 
        /* read minor */
@@ -409,19 +380,13 @@ static ssize_t devcgroup_access_write(struct cgroup *cgroup, struct cftype *cft,
                wh.minor = ~0;
                b++;
        } else if (isdigit(*b)) {
-               wh.minor = 0;
-               while (isdigit(*b)) {
-                       wh.minor = wh.minor*10+(*b-'0');
-                       b++;
-               }
+               wh.minor = simple_strtoul(b, &endp, 10);
+               b = endp;
        } else {
-               retval = -EINVAL;
-               goto out2;
-       }
-       if (!isspace(*b)) {
-               retval = -EINVAL;
-               goto out2;
+               return -EINVAL;
        }
+       if (!isspace(*b))
+               return -EINVAL;
        for (b++, count = 0; count < 3; count++, b++) {
                switch (*b) {
                case 'r':
@@ -438,47 +403,46 @@ static ssize_t devcgroup_access_write(struct cgroup *cgroup, struct cftype *cft,
                        count = 3;
                        break;
                default:
-                       retval = -EINVAL;
-                       goto out2;
+                       return -EINVAL;
                }
        }
 
 handle:
-       retval = 0;
        switch (filetype) {
        case DEVCG_ALLOW:
-               if (!parent_has_perm(cgroup, &wh))
-                       retval = -EPERM;
-               else
-                       retval = dev_whitelist_add(devcgroup, &wh);
-               break;
+               if (!parent_has_perm(devcgroup, &wh))
+                       return -EPERM;
+               return dev_whitelist_add(devcgroup, &wh);
        case DEVCG_DENY:
                dev_whitelist_rm(devcgroup, &wh);
                break;
        default:
-               retval = -EINVAL;
-               goto out2;
+               return -EINVAL;
        }
+       return 0;
+}
 
-       if (retval == 0)
-               retval = nbytes;
-
-out2:
+static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
+                                 const char *buffer)
+{
+       int retval;
+       if (!cgroup_lock_live_group(cgrp))
+               return -ENODEV;
+       retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
+                                        cft->private, buffer);
        cgroup_unlock();
-out1:
-       kfree(buffer);
        return retval;
 }
 
 static struct cftype dev_cgroup_files[] = {
        {
                .name = "allow",
-               .write  = devcgroup_access_write,
+               .write_string  = devcgroup_access_write,
                .private = DEVCG_ALLOW,
        },
        {
                .name = "deny",
-               .write = devcgroup_access_write,
+               .write_string = devcgroup_access_write,
                .private = DEVCG_DENY,
        },
        {
@@ -506,7 +470,6 @@ struct cgroup_subsys devices_subsys = {
 
 int devcgroup_inode_permission(struct inode *inode, int mask)
 {
-       struct cgroup *cgroup;
        struct dev_cgroup *dev_cgroup;
        struct dev_whitelist_item *wh;
 
@@ -515,13 +478,12 @@ int devcgroup_inode_permission(struct inode *inode, int mask)
                return 0;
        if (!S_ISBLK(inode->i_mode) && !S_ISCHR(inode->i_mode))
                return 0;
-       cgroup = task_cgroup(current, devices_subsys.subsys_id);
-       dev_cgroup = cgroup_to_devcgroup(cgroup);
-       if (!dev_cgroup)
-               return 0;
 
-       spin_lock(&dev_cgroup->lock);
-       list_for_each_entry(wh, &dev_cgroup->whitelist, list) {
+       rcu_read_lock();
+
+       dev_cgroup = task_devcgroup(current);
+
+       list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
                if (wh->type & DEV_ALL)
                        goto acc_check;
                if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode))
@@ -537,26 +499,24 @@ acc_check:
                        continue;
                if ((mask & MAY_READ) && !(wh->access & ACC_READ))
                        continue;
-               spin_unlock(&dev_cgroup->lock);
+               rcu_read_unlock();
                return 0;
        }
-       spin_unlock(&dev_cgroup->lock);
+
+       rcu_read_unlock();
 
        return -EPERM;
 }
 
 int devcgroup_inode_mknod(int mode, dev_t dev)
 {
-       struct cgroup *cgroup;
        struct dev_cgroup *dev_cgroup;
        struct dev_whitelist_item *wh;
 
-       cgroup = task_cgroup(current, devices_subsys.subsys_id);
-       dev_cgroup = cgroup_to_devcgroup(cgroup);
-       if (!dev_cgroup)
-               return 0;
+       rcu_read_lock();
+
+       dev_cgroup = task_devcgroup(current);
 
-       spin_lock(&dev_cgroup->lock);
        list_for_each_entry(wh, &dev_cgroup->whitelist, list) {
                if (wh->type & DEV_ALL)
                        goto acc_check;
@@ -571,9 +531,11 @@ int devcgroup_inode_mknod(int mode, dev_t dev)
 acc_check:
                if (!(wh->access & ACC_MKNOD))
                        continue;
-               spin_unlock(&dev_cgroup->lock);
+               rcu_read_unlock();
                return 0;
        }
-       spin_unlock(&dev_cgroup->lock);
+
+       rcu_read_unlock();
+
        return -EPERM;
 }