audit: dynamically allocate audit_names when not enough space is in the names array
authorEric Paris <eparis@redhat.com>
Tue, 3 Jan 2012 19:23:05 +0000 (14:23 -0500)
committerAl Viro <viro@zeniv.linux.org.uk>
Tue, 17 Jan 2012 21:16:54 +0000 (16:16 -0500)
This patch does 2 things.  First it reduces the number of audit_names
allocated in every audit context from 20 to 5.  5 should be enough for all
'normal' syscalls (rename being the worst).  Some syscalls can still touch
more the 5 inodes such as mount.  When rpc filesystem is mounted it will
create inodes and those can exceed 5.  To handle that problem this patch will
dynamically allocate audit_names if it needs more than 5.  This should
decrease the typicall memory usage while still supporting all the possible
kernel operations.

Signed-off-by: Eric Paris <eparis@redhat.com>
kernel/auditsc.c

index a09c503170597e9624a0491e85abb8b15e6798f8..1a92d61ddd276587b63262a780a16872ade664cd 100644 (file)
@@ -71,8 +71,9 @@
 #include "audit.h"
 
 /* AUDIT_NAMES is the number of slots we reserve in the audit_context
- * for saving names from getname(). */
-#define AUDIT_NAMES    20
+ * for saving names from getname().  If we get more names we will allocate
+ * a name dynamically and also add those to the list anchored by names_list. */
+#define AUDIT_NAMES    5
 
 /* Indicates that audit should log the full pathname. */
 #define AUDIT_NAME_FULL -1
@@ -101,9 +102,8 @@ struct audit_cap_data {
  *
  * Further, in fs/namei.c:path_lookup() we store the inode and device. */
 struct audit_names {
+       struct list_head list;          /* audit_context->names_list */
        const char      *name;
-       int             name_len;       /* number of name's characters to log */
-       unsigned        name_put;       /* call __putname() for this name */
        unsigned long   ino;
        dev_t           dev;
        umode_t         mode;
@@ -113,6 +113,14 @@ struct audit_names {
        u32             osid;
        struct audit_cap_data fcap;
        unsigned int    fcap_ver;
+       int             name_len;       /* number of name's characters to log */
+       bool            name_put;       /* call __putname() for this name */
+       /*
+        * This was an allocated audit_names and not from the array of
+        * names allocated in the task audit context.  Thus this name
+        * should be freed on syscall exit
+        */
+       bool            should_free;
 };
 
 struct audit_aux_data {
@@ -174,8 +182,17 @@ struct audit_context {
        long                return_code;/* syscall return code */
        u64                 prio;
        int                 return_valid; /* return code is valid */
-       int                 name_count;
-       struct audit_names  names[AUDIT_NAMES];
+       /*
+        * The names_list is the list of all audit_names collected during this
+        * syscall.  The first AUDIT_NAMES entries in the names_list will
+        * actually be from the preallocated_names array for performance
+        * reasons.  Except during allocation they should never be referenced
+        * through the preallocated_names array and should only be found/used
+        * by running the names_list.
+        */
+       struct audit_names  preallocated_names[AUDIT_NAMES];
+       int                 name_count; /* total records in names_list */
+       struct list_head    names_list; /* anchor for struct audit_names->list */
        char *              filterkey;  /* key for rule that triggered record */
        struct path         pwd;
        struct audit_context *previous; /* For nested syscalls */
@@ -307,17 +324,18 @@ static int audit_match_perm(struct audit_context *ctx, int mask)
 
 static int audit_match_filetype(struct audit_context *ctx, int val)
 {
-       int index;
+       struct audit_names *n;
        umode_t mode = (umode_t)val;
 
        if (unlikely(!ctx))
                return 0;
 
-       for (index = 0; index < ctx->name_count; index++) {
-               if ((ctx->names[index].ino != -1) &&
-                   ((ctx->names[index].mode & S_IFMT) == mode))
+       list_for_each_entry(n, &ctx->names_list, list) {
+               if ((n->ino != -1) &&
+                   ((n->mode & S_IFMT) == mode))
                        return 1;
        }
+
        return 0;
 }
 
@@ -456,13 +474,14 @@ static int audit_filter_rules(struct task_struct *tsk,
                              bool task_creation)
 {
        const struct cred *cred;
-       int i, j, need_sid = 1;
+       int i, need_sid = 1;
        u32 sid;
 
        cred = rcu_dereference_check(tsk->cred, tsk == current || task_creation);
 
        for (i = 0; i < rule->field_count; i++) {
                struct audit_field *f = &rule->fields[i];
+               struct audit_names *n;
                int result = 0;
 
                switch (f->type) {
@@ -525,8 +544,8 @@ static int audit_filter_rules(struct task_struct *tsk,
                                result = audit_comparator(MAJOR(name->dev),
                                                          f->op, f->val);
                        else if (ctx) {
-                               for (j = 0; j < ctx->name_count; j++) {
-                                       if (audit_comparator(MAJOR(ctx->names[j].dev),  f->op, f->val)) {
+                               list_for_each_entry(n, &ctx->names_list, list) {
+                                       if (audit_comparator(MAJOR(n->dev), f->op, f->val)) {
                                                ++result;
                                                break;
                                        }
@@ -538,8 +557,8 @@ static int audit_filter_rules(struct task_struct *tsk,
                                result = audit_comparator(MINOR(name->dev),
                                                          f->op, f->val);
                        else if (ctx) {
-                               for (j = 0; j < ctx->name_count; j++) {
-                                       if (audit_comparator(MINOR(ctx->names[j].dev), f->op, f->val)) {
+                               list_for_each_entry(n, &ctx->names_list, list) {
+                                       if (audit_comparator(MINOR(n->dev), f->op, f->val)) {
                                                ++result;
                                                break;
                                        }
@@ -550,8 +569,8 @@ static int audit_filter_rules(struct task_struct *tsk,
                        if (name)
                                result = (name->ino == f->val);
                        else if (ctx) {
-                               for (j = 0; j < ctx->name_count; j++) {
-                                       if (audit_comparator(ctx->names[j].ino, f->op, f->val)) {
+                               list_for_each_entry(n, &ctx->names_list, list) {
+                                       if (audit_comparator(n->ino, f->op, f->val)) {
                                                ++result;
                                                break;
                                        }
@@ -606,11 +625,10 @@ static int audit_filter_rules(struct task_struct *tsk,
                                                   name->osid, f->type, f->op,
                                                   f->lsm_rule, ctx);
                                } else if (ctx) {
-                                       for (j = 0; j < ctx->name_count; j++) {
-                                               if (security_audit_rule_match(
-                                                     ctx->names[j].osid,
-                                                     f->type, f->op,
-                                                     f->lsm_rule, ctx)) {
+                                       list_for_each_entry(n, &ctx->names_list, list) {
+                                               if (security_audit_rule_match(n->osid, f->type,
+                                                                             f->op, f->lsm_rule,
+                                                                             ctx)) {
                                                        ++result;
                                                        break;
                                                }
@@ -721,40 +739,53 @@ static enum audit_state audit_filter_syscall(struct task_struct *tsk,
        return AUDIT_BUILD_CONTEXT;
 }
 
-/* At syscall exit time, this filter is called if any audit_names[] have been
+/*
+ * Given an audit_name check the inode hash table to see if they match.
+ * Called holding the rcu read lock to protect the use of audit_inode_hash
+ */
+static int audit_filter_inode_name(struct task_struct *tsk,
+                                  struct audit_names *n,
+                                  struct audit_context *ctx) {
+       int word, bit;
+       int h = audit_hash_ino((u32)n->ino);
+       struct list_head *list = &audit_inode_hash[h];
+       struct audit_entry *e;
+       enum audit_state state;
+
+       word = AUDIT_WORD(ctx->major);
+       bit  = AUDIT_BIT(ctx->major);
+
+       if (list_empty(list))
+               return 0;
+
+       list_for_each_entry_rcu(e, list, list) {
+               if ((e->rule.mask[word] & bit) == bit &&
+                   audit_filter_rules(tsk, &e->rule, ctx, n, &state, false)) {
+                       ctx->current_state = state;
+                       return 1;
+               }
+       }
+
+       return 0;
+}
+
+/* At syscall exit time, this filter is called if any audit_names have been
  * collected during syscall processing.  We only check rules in sublists at hash
- * buckets applicable to the inode numbers in audit_names[].
+ * buckets applicable to the inode numbers in audit_names.
  * Regarding audit_state, same rules apply as for audit_filter_syscall().
  */
 void audit_filter_inodes(struct task_struct *tsk, struct audit_context *ctx)
 {
-       int i;
-       struct audit_entry *e;
-       enum audit_state state;
+       struct audit_names *n;
 
        if (audit_pid && tsk->tgid == audit_pid)
                return;
 
        rcu_read_lock();
-       for (i = 0; i < ctx->name_count; i++) {
-               int word = AUDIT_WORD(ctx->major);
-               int bit  = AUDIT_BIT(ctx->major);
-               struct audit_names *n = &ctx->names[i];
-               int h = audit_hash_ino((u32)n->ino);
-               struct list_head *list = &audit_inode_hash[h];
 
-               if (list_empty(list))
-                       continue;
-
-               list_for_each_entry_rcu(e, list, list) {
-                       if ((e->rule.mask[word] & bit) == bit &&
-                           audit_filter_rules(tsk, &e->rule, ctx, n,
-                                              &state, false)) {
-                               rcu_read_unlock();
-                               ctx->current_state = state;
-                               return;
-                       }
-               }
+       list_for_each_entry(n, &ctx->names_list, list) {
+               if (audit_filter_inode_name(tsk, n, ctx))
+                       break;
        }
        rcu_read_unlock();
 }
@@ -798,7 +829,7 @@ static inline struct audit_context *audit_get_context(struct task_struct *tsk,
 
 static inline void audit_free_names(struct audit_context *context)
 {
-       int i;
+       struct audit_names *n, *next;
 
 #if AUDIT_DEBUG == 2
        if (context->put_count + context->ino_count != context->name_count) {
@@ -809,10 +840,9 @@ static inline void audit_free_names(struct audit_context *context)
                       context->serial, context->major, context->in_syscall,
                       context->name_count, context->put_count,
                       context->ino_count);
-               for (i = 0; i < context->name_count; i++) {
+               list_for_each_entry(n, &context->names_list, list) {
                        printk(KERN_ERR "names[%d] = %p = %s\n", i,
-                              context->names[i].name,
-                              context->names[i].name ?: "(null)");
+                              n->name, n->name ?: "(null)");
                }
                dump_stack();
                return;
@@ -823,9 +853,12 @@ static inline void audit_free_names(struct audit_context *context)
        context->ino_count  = 0;
 #endif
 
-       for (i = 0; i < context->name_count; i++) {
-               if (context->names[i].name && context->names[i].name_put)
-                       __putname(context->names[i].name);
+       list_for_each_entry_safe(n, next, &context->names_list, list) {
+               list_del(&n->list);
+               if (n->name && n->name_put)
+                       __putname(n->name);
+               if (n->should_free)
+                       kfree(n);
        }
        context->name_count = 0;
        path_put(&context->pwd);
@@ -863,6 +896,7 @@ static inline struct audit_context *audit_alloc_context(enum audit_state state)
                return NULL;
        audit_zero_context(context, state);
        INIT_LIST_HEAD(&context->killed_trees);
+       INIT_LIST_HEAD(&context->names_list);
        return context;
 }
 
@@ -1323,6 +1357,68 @@ static void show_special(struct audit_context *context, int *call_panic)
        audit_log_end(ab);
 }
 
+static void audit_log_name(struct audit_context *context, struct audit_names *n,
+                          int record_num, int *call_panic)
+{
+       struct audit_buffer *ab;
+       ab = audit_log_start(context, GFP_KERNEL, AUDIT_PATH);
+       if (!ab)
+               return; /* audit_panic has been called */
+
+       audit_log_format(ab, "item=%d", record_num);
+
+       if (n->name) {
+               switch (n->name_len) {
+               case AUDIT_NAME_FULL:
+                       /* log the full path */
+                       audit_log_format(ab, " name=");
+                       audit_log_untrustedstring(ab, n->name);
+                       break;
+               case 0:
+                       /* name was specified as a relative path and the
+                        * directory component is the cwd */
+                       audit_log_d_path(ab, "name=", &context->pwd);
+                       break;
+               default:
+                       /* log the name's directory component */
+                       audit_log_format(ab, " name=");
+                       audit_log_n_untrustedstring(ab, n->name,
+                                                   n->name_len);
+               }
+       } else
+               audit_log_format(ab, " name=(null)");
+
+       if (n->ino != (unsigned long)-1) {
+               audit_log_format(ab, " inode=%lu"
+                                " dev=%02x:%02x mode=%#ho"
+                                " ouid=%u ogid=%u rdev=%02x:%02x",
+                                n->ino,
+                                MAJOR(n->dev),
+                                MINOR(n->dev),
+                                n->mode,
+                                n->uid,
+                                n->gid,
+                                MAJOR(n->rdev),
+                                MINOR(n->rdev));
+       }
+       if (n->osid != 0) {
+               char *ctx = NULL;
+               u32 len;
+               if (security_secid_to_secctx(
+                       n->osid, &ctx, &len)) {
+                       audit_log_format(ab, " osid=%u", n->osid);
+                       *call_panic = 2;
+               } else {
+                       audit_log_format(ab, " obj=%s", ctx);
+                       security_release_secctx(ctx, len);
+               }
+       }
+
+       audit_log_fcaps(ab, n);
+
+       audit_log_end(ab);
+}
+
 static void audit_log_exit(struct audit_context *context, struct task_struct *tsk)
 {
        const struct cred *cred;
@@ -1330,6 +1426,7 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
        struct audit_buffer *ab;
        struct audit_aux_data *aux;
        const char *tty;
+       struct audit_names *n;
 
        /* tsk == current */
        context->pid = tsk->pid;
@@ -1469,66 +1566,10 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
                        audit_log_end(ab);
                }
        }
-       for (i = 0; i < context->name_count; i++) {
-               struct audit_names *n = &context->names[i];
-
-               ab = audit_log_start(context, GFP_KERNEL, AUDIT_PATH);
-               if (!ab)
-                       continue; /* audit_panic has been called */
-
-               audit_log_format(ab, "item=%d", i);
-
-               if (n->name) {
-                       switch(n->name_len) {
-                       case AUDIT_NAME_FULL:
-                               /* log the full path */
-                               audit_log_format(ab, " name=");
-                               audit_log_untrustedstring(ab, n->name);
-                               break;
-                       case 0:
-                               /* name was specified as a relative path and the
-                                * directory component is the cwd */
-                               audit_log_d_path(ab, "name=", &context->pwd);
-                               break;
-                       default:
-                               /* log the name's directory component */
-                               audit_log_format(ab, " name=");
-                               audit_log_n_untrustedstring(ab, n->name,
-                                                           n->name_len);
-                       }
-               } else
-                       audit_log_format(ab, " name=(null)");
-
-               if (n->ino != (unsigned long)-1) {
-                       audit_log_format(ab, " inode=%lu"
-                                        " dev=%02x:%02x mode=%#ho"
-                                        " ouid=%u ogid=%u rdev=%02x:%02x",
-                                        n->ino,
-                                        MAJOR(n->dev),
-                                        MINOR(n->dev),
-                                        n->mode,
-                                        n->uid,
-                                        n->gid,
-                                        MAJOR(n->rdev),
-                                        MINOR(n->rdev));
-               }
-               if (n->osid != 0) {
-                       char *ctx = NULL;
-                       u32 len;
-                       if (security_secid_to_secctx(
-                               n->osid, &ctx, &len)) {
-                               audit_log_format(ab, " osid=%u", n->osid);
-                               call_panic = 2;
-                       } else {
-                               audit_log_format(ab, " obj=%s", ctx);
-                               security_release_secctx(ctx, len);
-                       }
-               }
-
-               audit_log_fcaps(ab, n);
 
-               audit_log_end(ab);
-       }
+       i = 0;
+       list_for_each_entry(n, &context->names_list, list)
+               audit_log_name(context, n, i++, &call_panic);
 
        /* Send end of event record to help user space know we are finished */
        ab = audit_log_start(context, GFP_KERNEL, AUDIT_EOE);
@@ -1820,6 +1861,30 @@ retry:
 #endif
 }
 
+static struct audit_names *audit_alloc_name(struct audit_context *context)
+{
+       struct audit_names *aname;
+
+       if (context->name_count < AUDIT_NAMES) {
+               aname = &context->preallocated_names[context->name_count];
+               memset(aname, 0, sizeof(*aname));
+       } else {
+               aname = kzalloc(sizeof(*aname), GFP_NOFS);
+               if (!aname)
+                       return NULL;
+               aname->should_free = true;
+       }
+
+       aname->ino = (unsigned long)-1;
+       list_add_tail(&aname->list, &context->names_list);
+
+       context->name_count++;
+#if AUDIT_DEBUG
+       context->ino_count++;
+#endif
+       return aname;
+}
+
 /**
  * audit_getname - add a name to the list
  * @name: name to add
@@ -1830,6 +1895,7 @@ retry:
 void __audit_getname(const char *name)
 {
        struct audit_context *context = current->audit_context;
+       struct audit_names *n;
 
        if (IS_ERR(name) || !name)
                return;
@@ -1842,13 +1908,15 @@ void __audit_getname(const char *name)
 #endif
                return;
        }
-       BUG_ON(context->name_count >= AUDIT_NAMES);
-       context->names[context->name_count].name = name;
-       context->names[context->name_count].name_len = AUDIT_NAME_FULL;
-       context->names[context->name_count].name_put = 1;
-       context->names[context->name_count].ino  = (unsigned long)-1;
-       context->names[context->name_count].osid = 0;
-       ++context->name_count;
+
+       n = audit_alloc_name(context);
+       if (!n)
+               return;
+
+       n->name = name;
+       n->name_len = AUDIT_NAME_FULL;
+       n->name_put = true;
+
        if (!context->pwd.dentry)
                get_fs_pwd(current->fs, &context->pwd);
 }
@@ -1870,12 +1938,13 @@ void audit_putname(const char *name)
                printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
                       __FILE__, __LINE__, context->serial, name);
                if (context->name_count) {
+                       struct audit_names *n;
                        int i;
-                       for (i = 0; i < context->name_count; i++)
+
+                       list_for_each_entry(n, &context->names_list, list)
                                printk(KERN_ERR "name[%d] = %p = %s\n", i,
-                                      context->names[i].name,
-                                      context->names[i].name ?: "(null)");
-               }
+                                      n->name, n->name ?: "(null)");
+                       }
 #endif
                __putname(name);
        }
@@ -1896,39 +1965,11 @@ void audit_putname(const char *name)
 #endif
 }
 
-static int audit_inc_name_count(struct audit_context *context,
-                               const struct inode *inode)
-{
-       if (context->name_count >= AUDIT_NAMES) {
-               if (inode)
-                       printk(KERN_DEBUG "audit: name_count maxed, losing inode data: "
-                              "dev=%02x:%02x, inode=%lu\n",
-                              MAJOR(inode->i_sb->s_dev),
-                              MINOR(inode->i_sb->s_dev),
-                              inode->i_ino);
-
-               else
-                       printk(KERN_DEBUG "name_count maxed, losing inode data\n");
-               return 1;
-       }
-       context->name_count++;
-#if AUDIT_DEBUG
-       context->ino_count++;
-#endif
-       return 0;
-}
-
-
 static inline int audit_copy_fcaps(struct audit_names *name, const struct dentry *dentry)
 {
        struct cpu_vfs_cap_data caps;
        int rc;
 
-       memset(&name->fcap.permitted, 0, sizeof(kernel_cap_t));
-       memset(&name->fcap.inheritable, 0, sizeof(kernel_cap_t));
-       name->fcap.fE = 0;
-       name->fcap_ver = 0;
-
        if (!dentry)
                return 0;
 
@@ -1968,30 +2009,25 @@ static void audit_copy_inode(struct audit_names *name, const struct dentry *dent
  */
 void __audit_inode(const char *name, const struct dentry *dentry)
 {
-       int idx;
        struct audit_context *context = current->audit_context;
        const struct inode *inode = dentry->d_inode;
+       struct audit_names *n;
 
        if (!context->in_syscall)
                return;
-       if (context->name_count
-           && context->names[context->name_count-1].name
-           && context->names[context->name_count-1].name == name)
-               idx = context->name_count - 1;
-       else if (context->name_count > 1
-                && context->names[context->name_count-2].name
-                && context->names[context->name_count-2].name == name)
-               idx = context->name_count - 2;
-       else {
-               /* FIXME: how much do we care about inodes that have no
-                * associated name? */
-               if (audit_inc_name_count(context, inode))
-                       return;
-               idx = context->name_count - 1;
-               context->names[idx].name = NULL;
+
+       list_for_each_entry_reverse(n, &context->names_list, list) {
+               if (n->name && (n->name == name))
+                       goto out;
        }
+
+       /* unable to find the name from a previous getname() */
+       n = audit_alloc_name(context);
+       if (!n)
+               return;
+out:
        handle_path(dentry);
-       audit_copy_inode(&context->names[idx], dentry, inode);
+       audit_copy_inode(n, dentry, inode);
 }
 
 /**
@@ -2010,11 +2046,11 @@ void __audit_inode(const char *name, const struct dentry *dentry)
 void __audit_inode_child(const struct dentry *dentry,
                         const struct inode *parent)
 {
-       int idx;
        struct audit_context *context = current->audit_context;
        const char *found_parent = NULL, *found_child = NULL;
        const struct inode *inode = dentry->d_inode;
        const char *dname = dentry->d_name.name;
+       struct audit_names *n;
        int dirlen = 0;
 
        if (!context->in_syscall)
@@ -2024,9 +2060,7 @@ void __audit_inode_child(const struct dentry *dentry,
                handle_one(inode);
 
        /* parent is more likely, look for it first */
-       for (idx = 0; idx < context->name_count; idx++) {
-               struct audit_names *n = &context->names[idx];
-
+       list_for_each_entry(n, &context->names_list, list) {
                if (!n->name)
                        continue;
 
@@ -2039,9 +2073,7 @@ void __audit_inode_child(const struct dentry *dentry,
        }
 
        /* no matching parent, look for matching child */
-       for (idx = 0; idx < context->name_count; idx++) {
-               struct audit_names *n = &context->names[idx];
-
+       list_for_each_entry(n, &context->names_list, list) {
                if (!n->name)
                        continue;
 
@@ -2059,34 +2091,29 @@ void __audit_inode_child(const struct dentry *dentry,
 
 add_names:
        if (!found_parent) {
-               if (audit_inc_name_count(context, parent))
+               n = audit_alloc_name(context);
+               if (!n)
                        return;
-               idx = context->name_count - 1;
-               context->names[idx].name = NULL;
-               audit_copy_inode(&context->names[idx], NULL, parent);
+               audit_copy_inode(n, NULL, parent);
        }
 
        if (!found_child) {
-               if (audit_inc_name_count(context, inode))
+               n = audit_alloc_name(context);
+               if (!n)
                        return;
-               idx = context->name_count - 1;
 
                /* Re-use the name belonging to the slot for a matching parent
                 * directory. All names for this context are relinquished in
                 * audit_free_names() */
                if (found_parent) {
-                       context->names[idx].name = found_parent;
-                       context->names[idx].name_len = AUDIT_NAME_FULL;
+                       n->name = found_parent;
+                       n->name_len = AUDIT_NAME_FULL;
                        /* don't call __putname() */
-                       context->names[idx].name_put = 0;
-               } else {
-                       context->names[idx].name = NULL;
+                       n->name_put = false;
                }
 
                if (inode)
-                       audit_copy_inode(&context->names[idx], NULL, inode);
-               else
-                       context->names[idx].ino = (unsigned long)-1;
+                       audit_copy_inode(n, NULL, inode);
        }
 }
 EXPORT_SYMBOL_GPL(__audit_inode_child);