xnu-4903.231.4.tar.gz

[apple/xnu.git] / bsd / vfs / vfs_cache.c

diff --git a/bsd/vfs/vfs_cache.c b/bsd/vfs/vfs_cache.c
index b7876475d49f6bdd804fea31a7440d173cd9eaa5..56c69754c777d576f84a2cf98aee950068274545 100644 (file)
--- a/bsd/vfs/vfs_cache.c
+++ b/bsd/vfs/vfs_cache.c
@@ -82,6 +82,7 @@
 #include <sys/kauth.h>
 #include <sys/user.h>
 #include <sys/paths.h>
 #include <sys/kauth.h>
 #include <sys/user.h>
 #include <sys/paths.h>

+#include <os/overflow.h>

 
 #if CONFIG_MACF
 #include <security/mac_framework.h>
 
 #if CONFIG_MACF
 #include <security/mac_framework.h>


@@ -388,12 +389,12 @@ vnode_issubdir(vnode_t vp, vnode_t dvp, int *is_subdir, vfs_context_t ctx)
 }
 
 /*
 }
 
 /*

- * This function builds the path to a filename in "buff".  The
- * length of the buffer *INCLUDING* the trailing zero byte is
- * returned in outlen.  NOTE: the length includes the trailing
- * zero byte and thus the length is one greater than what strlen
- * would return.  This is important and lots of code elsewhere
- * in the kernel assumes this behavior.
+ * This function builds the path in "buff" from the supplied vnode.
+ * The length of the buffer *INCLUDING* the trailing zero byte is
+ * returned in outlen.  NOTE: the length includes the trailing zero
+ * byte and thus the length is one greater than what strlen would
+ * return.  This is important and lots of code elsewhere in the kernel
+ * assumes this behavior.

  * 
  * This function can call vnop in file system if the parent vnode 
  * does not exist or when called for hardlinks via volfs path.  
  * 
  * This function can call vnop in file system if the parent vnode 
  * does not exist or when called for hardlinks via volfs path.  


@@ -410,9 +411,19 @@ vnode_issubdir(vnode_t vp, vnode_t dvp, int *is_subdir, vfs_context_t ctx)
  * cross over mount points during building the path. 
  *
  * passed in vp must have a valid io_count reference
  * cross over mount points during building the path. 
  *
  * passed in vp must have a valid io_count reference

+ *
+ * If parent vnode is non-NULL it also must have an io count.  This
+ * allows build_path_with_parent to be safely called for operations
+ * unlink, rmdir and rename that already have io counts on the target
+ * and the directory. In this way build_path_with_parent does not have
+ * to try and obtain an additional io count on the parent.  Taking an
+ * io count ont the parent can lead to dead lock if a forced unmount
+ * occures at the right moment. For a fuller explaination on how this
+ * can occur see the comment for vn_getpath_with_parent.
+ *

  */
 int
  */
 int

-build_path(vnode_t first_vp, char *buff, int buflen, int *outlen, int flags, vfs_context_t ctx)
+build_path_with_parent(vnode_t first_vp, vnode_t parent_vp, char *buff, int buflen, int *outlen, int flags, vfs_context_t ctx)

 {
         vnode_t vp, tvp;
        vnode_t vp_with_iocount;
 {
         vnode_t vp, tvp;
        vnode_t vp_with_iocount;


@@ -587,7 +598,7 @@ again:
 
                        NAME_CACHE_UNLOCK();
 
 
                        NAME_CACHE_UNLOCK();
 

-                       if (vp != first_vp && vp != vp_with_iocount) {
+                       if (vp != first_vp && vp != parent_vp && vp != vp_with_iocount) {

                                if (vp_with_iocount) {
                                        vnode_put(vp_with_iocount);
                                        vp_with_iocount = NULLVP;
                                if (vp_with_iocount) {
                                        vnode_put(vp_with_iocount);
                                        vp_with_iocount = NULLVP;


@@ -678,7 +689,7 @@ bad_news:
 
                        NAME_CACHE_UNLOCK();
 
 
                        NAME_CACHE_UNLOCK();
 

-                       if (vp != first_vp && vp != vp_with_iocount) {
+                       if (vp != first_vp && vp != parent_vp && vp != vp_with_iocount) {

                                if (vp_with_iocount) {
                                        vnode_put(vp_with_iocount);
                                        vp_with_iocount = NULLVP;
                                if (vp_with_iocount) {
                                        vnode_put(vp_with_iocount);
                                        vp_with_iocount = NULLVP;


@@ -745,6 +756,11 @@ out:
        return (ret);
 }
 
        return (ret);
 }
 

+int
+build_path(vnode_t first_vp, char *buff, int buflen, int *outlen, int flags, vfs_context_t ctx)
+{
+       return (build_path_with_parent(first_vp, NULL, buff, buflen, outlen, flags, ctx));
+}

 
 /*
  * return NULLVP if vp's parent doesn't
 
 /*
  * return NULLVP if vp's parent doesn't


@@ -861,10 +877,8 @@ vnode_update_identity(vnode_t vp, vnode_t dvp, const char *name, int name_len, u
 {
        struct  namecache *ncp;
         vnode_t        old_parentvp = NULLVP;
 {
        struct  namecache *ncp;
         vnode_t        old_parentvp = NULLVP;

-#if NAMEDSTREAMS

        int isstream = (vp->v_flag & VISNAMEDSTREAM);
        int kusecountbumped = 0;
        int isstream = (vp->v_flag & VISNAMEDSTREAM);
        int kusecountbumped = 0;

-#endif

        kauth_cred_t tcred = NULL;
        const char *vname = NULL;
        const char *tname = NULL;
        kauth_cred_t tcred = NULL;
        const char *vname = NULL;
        const char *tname = NULL;


@@ -873,7 +887,6 @@ vnode_update_identity(vnode_t vp, vnode_t dvp, const char *name, int name_len, u
                if (dvp && vnode_ref(dvp) != 0) {
                        dvp = NULLVP;
                }
                if (dvp && vnode_ref(dvp) != 0) {
                        dvp = NULLVP;
                }

-#if NAMEDSTREAMS

                /* Don't count a stream's parent ref during unmounts */
                if (isstream && dvp && (dvp != vp) && (dvp != vp->v_parent) && (dvp->v_type == VREG)) {
                        vnode_lock_spin(dvp);
                /* Don't count a stream's parent ref during unmounts */
                if (isstream && dvp && (dvp != vp) && (dvp != vp->v_parent) && (dvp->v_type == VREG)) {
                        vnode_lock_spin(dvp);


@@ -881,7 +894,6 @@ vnode_update_identity(vnode_t vp, vnode_t dvp, const char *name, int name_len, u
                        kusecountbumped = 1;
                        vnode_unlock(dvp);
                }
                        kusecountbumped = 1;
                        vnode_unlock(dvp);
                }

-#endif

        } else {
                dvp = NULLVP;
        }
        } else {
                dvp = NULLVP;
        }


@@ -916,6 +928,7 @@ vnode_update_identity(vnode_t vp, vnode_t dvp, const char *name, int name_len, u
                        tcred = vp->v_cred;
                        vp->v_cred = NOCRED;
                        vp->v_authorized_actions = 0;
                        tcred = vp->v_cred;
                        vp->v_cred = NOCRED;
                        vp->v_authorized_actions = 0;

+                       vp->v_cred_timestamp = 0;

                }
                if ( (flags & VNODE_UPDATE_NAME) ) {
                        vname = vp->v_name;
                }
                if ( (flags & VNODE_UPDATE_NAME) ) {
                        vname = vp->v_name;


@@ -944,7 +957,6 @@ vnode_update_identity(vnode_t vp, vnode_t dvp, const char *name, int name_len, u
                        kauth_cred_unref(&tcred);
        }
        if (dvp != NULLVP) {
                        kauth_cred_unref(&tcred);
        }
        if (dvp != NULLVP) {

-#if NAMEDSTREAMS

                /* Back-out the ref we took if we lost a race for vp->v_parent. */
                if (kusecountbumped) {
                        vnode_lock_spin(dvp);
                /* Back-out the ref we took if we lost a race for vp->v_parent. */
                if (kusecountbumped) {
                        vnode_lock_spin(dvp);


@@ -952,20 +964,17 @@ vnode_update_identity(vnode_t vp, vnode_t dvp, const char *name, int name_len, u
                                --dvp->v_kusecount;  
                        vnode_unlock(dvp);
                }
                                --dvp->v_kusecount;  
                        vnode_unlock(dvp);
                }

-#endif

                vnode_rele(dvp);
        }
        if (old_parentvp) {
                struct  uthread *ut;
 
                vnode_rele(dvp);
        }
        if (old_parentvp) {
                struct  uthread *ut;
 

-#if NAMEDSTREAMS

                if (isstream) {
                        vnode_lock_spin(old_parentvp);
                        if ((old_parentvp->v_type != VDIR) && (old_parentvp->v_kusecount > 0))
                                --old_parentvp->v_kusecount;
                        vnode_unlock(old_parentvp);
                }
                if (isstream) {
                        vnode_lock_spin(old_parentvp);
                        if ((old_parentvp->v_type != VDIR) && (old_parentvp->v_kusecount > 0))
                                --old_parentvp->v_kusecount;
                        vnode_unlock(old_parentvp);
                }

-#endif

                ut = get_bsdthread_info(current_thread());
 
                /*
                ut = get_bsdthread_info(current_thread());
 
                /*


@@ -1361,7 +1370,7 @@ skiprsrcfork:
 
 #if CONFIG_MACF
 
 
 #if CONFIG_MACF
 

-               /* 
+               /*

                 * Name cache provides authorization caching (see below)
                 * that will short circuit MAC checks in lookup().
                 * We must perform MAC check here.  On denial
                 * Name cache provides authorization caching (see below)
                 * that will short circuit MAC checks in lookup().
                 * We must perform MAC check here.  On denial


@@ -1376,8 +1385,11 @@ skiprsrcfork:
                        }
                }
 #endif /* MAC */
                        }
                }
 #endif /* MAC */

-               if (ttl_enabled && ((tv.tv_sec - dp->v_cred_timestamp) > dp->v_mount->mnt_authcache_ttl))
+               if (ttl_enabled &&
+                   (dp->v_mount->mnt_authcache_ttl == 0 ||
+                   ((tv.tv_sec - dp->v_cred_timestamp) > dp->v_mount->mnt_authcache_ttl))) {

                        break;
                        break;

+               }

 
                /*
                 * NAME_CACHE_LOCK holds these fields stable
 
                /*
                 * NAME_CACHE_LOCK holds these fields stable


@@ -1386,11 +1398,17 @@ skiprsrcfork:
                 * so we make an ugly check for root here. root is always
                 * allowed and breaking out of here only to find out that is
                 * authorized by virtue of being root is very very expensive.
                 * so we make an ugly check for root here. root is always
                 * allowed and breaking out of here only to find out that is
                 * authorized by virtue of being root is very very expensive.

+                * However, the check for not root is valid only for filesystems
+                * which use local authorization.
+                *
+                * XXX: Remove the check for root when we can reliably set
+                * KAUTH_VNODE_SEARCHBYANYONE as root.

                 */
                if ((dp->v_cred != ucred || !(dp->v_authorized_actions & KAUTH_VNODE_SEARCH)) &&
                    !(dp->v_authorized_actions & KAUTH_VNODE_SEARCHBYANYONE) &&
                 */
                if ((dp->v_cred != ucred || !(dp->v_authorized_actions & KAUTH_VNODE_SEARCH)) &&
                    !(dp->v_authorized_actions & KAUTH_VNODE_SEARCHBYANYONE) &&

-                   !vfs_context_issuser(ctx))
+                   (ttl_enabled || !vfs_context_issuser(ctx)))  {

                        break;
                        break;

+               }

 
                /*
                 * indicate that we're allowed to traverse this directory...
 
                /*
                 * indicate that we're allowed to traverse this directory...


@@ -1412,7 +1430,7 @@ skiprsrcfork:
                                 * Force directory hardlinks to go to
                                 * file system for ".." requests.
                                 */
                                 * Force directory hardlinks to go to
                                 * file system for ".." requests.
                                 */

-                               if (dp && (dp->v_flag & VISHARDLINK)) {
+                               if ((dp->v_flag & VISHARDLINK)) {

                                        break;
                                }
                                /*
                                        break;
                                }
                                /*


@@ -1675,7 +1693,7 @@ cache_lookup_locked(vnode_t dvp, struct componentname *cnp)
        ncpp = NCHHASH(dvp, cnp->cn_hash);
        LIST_FOREACH(ncp, ncpp, nc_hash) {
                if ((ncp->nc_dvp == dvp) && (ncp->nc_hashval == hashval)) {
        ncpp = NCHHASH(dvp, cnp->cn_hash);
        LIST_FOREACH(ncp, ncpp, nc_hash) {
                if ((ncp->nc_dvp == dvp) && (ncp->nc_hashval == hashval)) {

-                       if (memcmp(ncp->nc_name, cnp->cn_nameptr, namelen) == 0 && ncp->nc_name[namelen] == 0)
+                       if (strncmp(ncp->nc_name, cnp->cn_nameptr, namelen) == 0 && ncp->nc_name[namelen] == 0)

                                break;
                }
        }
                                break;
                }
        }


@@ -1762,7 +1780,7 @@ relook:
        ncpp = NCHHASH(dvp, cnp->cn_hash);
        LIST_FOREACH(ncp, ncpp, nc_hash) {
                if ((ncp->nc_dvp == dvp) && (ncp->nc_hashval == hashval)) {
        ncpp = NCHHASH(dvp, cnp->cn_hash);
        LIST_FOREACH(ncp, ncpp, nc_hash) {
                if ((ncp->nc_dvp == dvp) && (ncp->nc_hashval == hashval)) {

-                       if (memcmp(ncp->nc_name, cnp->cn_nameptr, namelen) == 0 && ncp->nc_name[namelen] == 0)
+                       if (strncmp(ncp->nc_name, cnp->cn_nameptr, namelen) == 0 && ncp->nc_name[namelen] == 0)

                                break;
                }
        }
                                break;
                }
        }


@@ -2142,28 +2160,35 @@ name_cache_unlock(void)
 
 
 int
 
 
 int

-resize_namecache(u_int newsize)
+resize_namecache(int newsize)

 {
     struct nchashhead  *new_table;
     struct nchashhead  *old_table;
     struct nchashhead  *old_head, *head;
     struct namecache   *entry, *next;
     uint32_t           i, hashval;
 {
     struct nchashhead  *new_table;
     struct nchashhead  *old_table;
     struct nchashhead  *old_head, *head;
     struct namecache   *entry, *next;
     uint32_t           i, hashval;

-    int                        dNodes, dNegNodes;
+    int                        dNodes, dNegNodes, nelements;

     u_long             new_size, old_size;
 
     u_long             new_size, old_size;
 

+    if (newsize < 0)
+        return EINVAL;
+

     dNegNodes = (newsize / 10);
     dNodes = newsize + dNegNodes;
     dNegNodes = (newsize / 10);
     dNodes = newsize + dNegNodes;

-

     // we don't support shrinking yet
     if (dNodes <= desiredNodes) {
     // we don't support shrinking yet
     if (dNodes <= desiredNodes) {

-       return 0;
+        return 0;

     }
     }

-    new_table = hashinit(2 * dNodes, M_CACHE, &nchashmask);
+
+    if (os_mul_overflow(dNodes, 2, &nelements)) {
+        return EINVAL;
+    }
+
+    new_table = hashinit(nelements, M_CACHE, &nchashmask);

     new_size  = nchashmask + 1;
 
     if (new_table == NULL) {
     new_size  = nchashmask + 1;
 
     if (new_table == NULL) {

-       return ENOMEM;
+        return ENOMEM;

     }
 
     NAME_CACHE_LOCK();
     }
 
     NAME_CACHE_LOCK();


@@ -2201,7 +2226,7 @@ resize_namecache(u_int newsize)
 }
 
 static void
 }
 
 static void

-cache_delete(struct namecache *ncp, int age_entry)
+cache_delete(struct namecache *ncp, int free_entry)

 {
         NCHSTAT(ncs_deletes);
 
 {
         NCHSTAT(ncs_deletes);
 


@@ -2222,16 +2247,13 @@ cache_delete(struct namecache *ncp, int age_entry)
         */
        ncp->nc_hash.le_prev = NULL;
 
         */
        ncp->nc_hash.le_prev = NULL;
 

-       if (age_entry) {
-               /*
-                * make it the next one available
-                * for cache_enter's use
-                */
-               TAILQ_REMOVE(&nchead, ncp, nc_entry);
-               TAILQ_INSERT_HEAD(&nchead, ncp, nc_entry);
-       }

        vfs_removename(ncp->nc_name);
        ncp->nc_name = NULL;
        vfs_removename(ncp->nc_name);
        ncp->nc_name = NULL;

+       if (free_entry) {
+               TAILQ_REMOVE(&nchead, ncp, nc_entry);
+               FREE_ZONE(ncp, sizeof(*ncp), M_CACHE);
+               numcache--;
+       }

 }
 
 
 }
 
 


@@ -2465,7 +2487,7 @@ add_name_internal(const char *name, uint32_t len, u_int hashval, boolean_t need_
        lck_mtx_lock_spin(&strcache_mtx_locks[lock_index]);
 
        for (entry = head->lh_first; entry != NULL; chain_len++, entry = entry->hash_chain.le_next) {
        lck_mtx_lock_spin(&strcache_mtx_locks[lock_index]);
 
        for (entry = head->lh_first; entry != NULL; chain_len++, entry = entry->hash_chain.le_next) {

-               if (memcmp(entry->str, name, len) == 0 && entry->str[len] == 0) {
+               if (strncmp(entry->str, name, len) == 0 && entry->str[len] == 0) {

                        entry->refcount++;
                        break;
                }
                        entry->refcount++;
                        break;
                }