]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/kern/posix_shm.c
xnu-3248.60.10.tar.gz
[apple/xnu.git] / bsd / kern / posix_shm.c
index d46eb0b3c1268ed385b2841d55824213b98190eb..38faf2939a0fcd7f369a7ac676b749c853da293a 100644 (file)
@@ -85,7 +85,7 @@
 #include <vm/vm_protos.h>
 
 #define f_flag f_fglob->fg_flag
-#define f_type f_fglob->fg_type
+#define f_type f_fglob->fg_ops->fo_type
 #define f_msgcount f_fglob->fg_msgcount
 #define f_cred f_fglob->fg_cred
 #define f_ops f_fglob->fg_ops
@@ -135,6 +135,10 @@ struct     pshmcache {
 };
 #define PSHMCACHE_NULL (struct pshmcache *)0
 
+#define PSHMCACHE_NOTFOUND (0)
+#define PSHMCACHE_FOUND    (-1)
+#define PSHMCACHE_NEGATIVE (ENOENT)
+
 struct pshmstats {
        long    goodhits;               /* hits that we can really use */
        long    neghits;                /* negative hits that we can use */
@@ -184,16 +188,24 @@ static int pshm_closefile (struct fileglob *fg, vfs_context_t ctx);
 static int pshm_kqfilter(struct fileproc *fp, struct knote *kn, vfs_context_t ctx);
 
 int pshm_access(struct pshminfo *pinfo, int mode, kauth_cred_t cred, proc_t p);
+int pshm_cache_purge_all(proc_t p);
+
 static int pshm_cache_add(struct pshminfo *pshmp, struct pshmname *pnp, struct pshmcache *pcp);
 static void pshm_cache_delete(struct pshmcache *pcp);
-#if NOT_USED
-static void pshm_cache_purge(void);
-#endif /* NOT_USED */
 static int pshm_cache_search(struct pshminfo **pshmp, struct pshmname *pnp,
        struct pshmcache **pcache, int addref);
-
-struct         fileops pshmops =
-       { pshm_read, pshm_write, pshm_ioctl, pshm_select, pshm_closefile, pshm_kqfilter, 0 };
+static int pshm_unlink_internal(struct pshminfo *pinfo, struct pshmcache *pcache);
+
+static const struct fileops pshmops = {
+       DTYPE_PSXSHM,
+       pshm_read,
+       pshm_write,
+       pshm_ioctl,
+       pshm_select,
+       pshm_closefile,
+       pshm_kqfilter,
+       0
+};
 
 static lck_grp_t       *psx_shm_subsys_lck_grp;
 static lck_grp_attr_t  *psx_shm_subsys_lck_grp_attr;
@@ -202,6 +214,7 @@ static lck_mtx_t        psx_shm_subsys_mutex;
 
 #define PSHM_SUBSYS_LOCK() lck_mtx_lock(& psx_shm_subsys_mutex)
 #define PSHM_SUBSYS_UNLOCK() lck_mtx_unlock(& psx_shm_subsys_mutex)
+#define PSHM_SUBSYS_ASSERT_HELD()  lck_mtx_assert(&psx_shm_subsys_mutex, LCK_MTX_ASSERT_OWNED)
 
 
 /* Initialize the mutex governing access to the posix shm subsystem */
@@ -236,7 +249,7 @@ pshm_cache_search(struct pshminfo **pshmp, struct pshmname *pnp,
 
        if (pnp->pshm_namelen > PSHMNAMLEN) {
                pshmstats.longnames++;
-               return (0);
+               return PSHMCACHE_NOTFOUND;
        }
 
        pcpp = PSHMHASH(pnp);
@@ -249,7 +262,7 @@ pshm_cache_search(struct pshminfo **pshmp, struct pshmname *pnp,
 
        if (pcp == 0) {
                pshmstats.miss++;
-               return (0);
+               return PSHMCACHE_NOTFOUND;
        }
 
        /* We found a "positive" match, return the vnode */
@@ -260,15 +273,14 @@ pshm_cache_search(struct pshminfo **pshmp, struct pshmname *pnp,
                *pcache = pcp;
                if (addref)
                        pcp->pshminfo->pshm_usecount++;
-               return (-1);
+               return PSHMCACHE_FOUND;
        }
 
        /*
         * We found a "negative" match, ENOENT notifies client of this match.
-        * The nc_vpid field records whether this is a whiteout.
         */
        pshmstats.neghits++;
-       return (ENOENT);
+       return PSHMCACHE_NEGATIVE;
 }
 
 /*
@@ -289,16 +301,13 @@ pshm_cache_add(struct pshminfo *pshmp, struct pshmname *pnp, struct pshmcache *p
 
 
        /*  if the entry has already been added by some one else return */
-       if (pshm_cache_search(&dpinfo, pnp, &dpcp, 0) == -1) {
-               return(EEXIST);
+       if (pshm_cache_search(&dpinfo, pnp, &dpcp, 0) == PSHMCACHE_FOUND) {
+               return EEXIST;
        }
        pshmnument++;
 
        /*
         * Fill in cache info, if vp is NULL this is a "negative" cache entry.
-        * For negative entries, we have to record whether it is a whiteout.
-        * the whiteout flag is stored in the nc_vpid field which is
-        * otherwise unused.
         */
        pcp->pshminfo = pshmp;
        pcp->pshm_nlen = pnp->pshm_namelen;
@@ -314,7 +323,7 @@ pshm_cache_add(struct pshminfo *pshmp, struct pshmname *pnp, struct pshmcache *p
        }
 #endif
        LIST_INSERT_HEAD(pcpp, pcp, pshm_hash);
-       return(0);
+       return 0;
 }
 
 /*
@@ -326,27 +335,44 @@ pshm_cache_init(void)
        pshmhashtbl = hashinit(desiredvnodes / 8, M_SHM, &pshmhash);
 }
 
-#if NOT_USED
 /*
- * Invalidate a all entries to particular vnode.
- * 
+ * Invalidate all entries and delete all objects associated with it. Entire
+ * non Kernel entries are going away. Just dump'em all
+ *
  * We actually just increment the v_id, that will do it. The entries will
  * be purged by lookup as they get found. If the v_id wraps around, we
  * need to ditch the entire cache, to avoid confusion. No valid vnode will
  * ever have (v_id == 0).
  */
-static void
-pshm_cache_purge(void)
+int
+pshm_cache_purge_all(__unused proc_t p)
 {
-       struct pshmcache *pcp;
+       struct pshmcache *pcp, *tmppcp;
        struct pshmhashhead *pcpp;
+       int error = 0;
+
+       if (kauth_cred_issuser(kauth_cred_get()) == 0)
+               return EPERM;
 
+       PSHM_SUBSYS_LOCK();
        for (pcpp = &pshmhashtbl[pshmhash]; pcpp >= pshmhashtbl; pcpp--) {
-               while ( (pcp = pcpp->lh_first) )
-                       pshm_cache_delete(pcp);
+               LIST_FOREACH_SAFE(pcp, pcpp, pshm_hash, tmppcp) {
+                       assert(pcp->pshm_nlen);
+                       error = pshm_unlink_internal(pcp->pshminfo, pcp);
+                       if (error)
+                               goto out;
+               }
        }
+       assert(pshmnument == 0);
+
+out:
+       PSHM_SUBSYS_UNLOCK();
+
+       if (error)
+               printf("%s: Error %d removing shm cache: %ld remain!\n",
+                      __func__, error, pshmnument);
+       return error;
 }
-#endif /* NOT_USED */
 
 static void
 pshm_cache_delete(struct pshmcache *pcp)
@@ -484,12 +510,12 @@ shm_open(proc_t p, struct shm_open_args *uap, int32_t *retval)
 
        PSHM_SUBSYS_UNLOCK();
 
-       if (error == ENOENT) {
+       if (error == PSHMCACHE_NEGATIVE) {
                error = EINVAL;
                goto bad;
        }
 
-       if (!error) {
+       if (error == PSHMCACHE_NOTFOUND) {
                incache = 0;
                if (fmode & O_CREAT) {
                        /*  create a new one (commit the allocation) */
@@ -606,7 +632,6 @@ shm_open(proc_t p, struct shm_open_args *uap, int32_t *retval)
 
        proc_fdlock(p);
        fp->f_flag = fmode & FMASK;
-       fp->f_type = DTYPE_PSXSHM;
        fp->f_ops = &pshmops;
        fp->f_data = (caddr_t)new_pnode;
        *fdflags(p, indx) |= UF_EXCLOSE;
@@ -664,10 +689,13 @@ pshm_truncate(__unused proc_t p, struct fileproc *fp, __unused int fd,
        mach_vm_size_t total_size, alloc_size;
        memory_object_size_t mosize;
        struct pshmobj *pshmobj, *pshmobj_next, **pshmobj_next_p;
+       vm_map_t        user_map;
 #if CONFIG_MACF
        int error;
 #endif
 
+       user_map = current_map();
+
        if (fp->f_type != DTYPE_PSXSHM) {
                return(EINVAL);
        }
@@ -695,7 +723,8 @@ pshm_truncate(__unused proc_t p, struct fileproc *fp, __unused int fd,
 #endif
 
        pinfo->pshm_flags |= PSHM_ALLOCATING;
-       total_size = round_page_64(length);
+       total_size = vm_map_round_page(length,
+                                      vm_map_page_mask(user_map));
        pshmobj_next_p = &pinfo->pshm_memobjects;
 
        for (alloc_size = 0;
@@ -735,7 +764,8 @@ pshm_truncate(__unused proc_t p, struct fileproc *fp, __unused int fd,
                pshmobj_next_p = &pshmobj->pshmo_next;
        }
        
-       pinfo->pshm_flags = PSHM_ALLOCATED;
+       pinfo->pshm_flags |= PSHM_ALLOCATED;
+       pinfo->pshm_flags &= ~(PSHM_ALLOCATING);
        pinfo->pshm_length = total_size;
        PSHM_SUBSYS_UNLOCK();
        return(0);
@@ -898,9 +928,11 @@ pshm_mmap(__unused proc_t p, struct mmap_args *uap, user_addr_t *retval, struct
 
        if ((flags & MAP_FIXED) == 0) {
                alloc_flags = VM_FLAGS_ANYWHERE;
-               user_addr = vm_map_round_page(user_addr); 
+               user_addr = vm_map_round_page(user_addr,
+                                             vm_map_page_mask(user_map)); 
        } else {
-               if (user_addr != vm_map_round_page(user_addr))
+               if (user_addr != vm_map_round_page(user_addr,
+                                                  vm_map_page_mask(user_map)))
                        return (EINVAL);
                /*
                 * We do not get rid of the existing mappings here because
@@ -996,24 +1028,72 @@ out:
 
 }
 
+static int
+pshm_unlink_internal(struct pshminfo *pinfo, struct pshmcache *pcache)
+{
+       struct pshmobj *pshmobj, *pshmobj_next;
+
+       PSHM_SUBSYS_ASSERT_HELD();
+
+       if (!pinfo || !pcache)
+               return EINVAL;
+
+       if ((pinfo->pshm_flags & (PSHM_DEFINED | PSHM_ALLOCATED)) == 0)
+               return EINVAL;
+
+       if (pinfo->pshm_flags & PSHM_INDELETE)
+               return 0;
+
+       pinfo->pshm_flags |= PSHM_INDELETE;
+       pinfo->pshm_usecount--;
+
+       pshm_cache_delete(pcache);
+       pinfo->pshm_flags |= PSHM_REMOVED;
+
+       /* release the existence reference */
+       if (!pinfo->pshm_usecount) {
+#if CONFIG_MACF
+               mac_posixshm_label_destroy(pinfo);
+#endif
+               /*
+                * If this is the last reference going away on the object,
+                * then we need to destroy the backing object.  The name
+                * has an implied but uncounted reference on the object,
+                * once it's created, since it's used as a rendezvous, and
+                * therefore may be subsequently reopened.
+                */
+               for (pshmobj = pinfo->pshm_memobjects;
+                    pshmobj != NULL;
+                    pshmobj = pshmobj_next) {
+                       mach_memory_entry_port_release(pshmobj->pshmo_memobject);
+                       pshmobj_next = pshmobj->pshmo_next;
+                       FREE(pshmobj, M_SHM);
+               }
+               FREE(pinfo,M_SHM);
+       }
+
+       FREE(pcache, M_SHM);
+
+       return 0;
+}
+
 int
-shm_unlink(__unused proc_t p, struct shm_unlink_args *uap, 
-                       __unused int32_t *retval)
+shm_unlink(proc_t p, struct shm_unlink_args *uap, __unused int32_t *retval)
 {
        size_t i;
-       int error=0;
+       char * pnbuf;
+       size_t pathlen;
+       int error = 0;
+
        struct pshmname nd;
        struct pshminfo *pinfo;
-       char * pnbuf;
        char * nameptr;
        char * cp;
-       size_t pathlen, plen;
-       int incache = 0;
        struct pshmcache *pcache = PSHMCACHE_NULL;
-       struct pshmobj *pshmobj, *pshmobj_next;
 
        pinfo = PSHMINFO_NULL;
 
+
        MALLOC_ZONE(pnbuf, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK);
        if (pnbuf == NULL) {
                return(ENOSPC);         /* XXX non-standard */
@@ -1029,12 +1109,12 @@ shm_unlink(__unused proc_t p, struct shm_unlink_args *uap,
                goto bad;
        }
 
+       nameptr = pnbuf;
 
 #ifdef PSXSHM_NAME_RESTRICT
-       nameptr = pnbuf;
        if (*nameptr == '/') {
                while (*(nameptr++) == '/') {
-                       plen--;
+                       pathlen--;
                        error = EINVAL;
                        goto bad;
                }
@@ -1044,31 +1124,24 @@ shm_unlink(__unused proc_t p, struct shm_unlink_args *uap,
        }
 #endif /* PSXSHM_NAME_RESTRICT */
 
-       plen = pathlen;
-       nameptr = pnbuf;
        nd.pshm_nameptr = nameptr;
-       nd.pshm_namelen = plen;
-       nd. pshm_hash =0;
+       nd.pshm_namelen = pathlen;
+       nd.pshm_hash = 0;
 
-        for (cp = nameptr, i=1; *cp != 0 && i <= plen; i++, cp++) {
+        for (cp = nameptr, i=1; *cp != 0 && i <= pathlen; i++, cp++) {
                nd.pshm_hash += (unsigned char)*cp * i;
        }
 
        PSHM_SUBSYS_LOCK();
        error = pshm_cache_search(&pinfo, &nd, &pcache, 0);
 
-       if (error == ENOENT) {
+       /* During unlink lookup failure also implies ENOENT */ 
+       if (error != PSHMCACHE_FOUND) {
                PSHM_SUBSYS_UNLOCK();
-               error = EINVAL;
+               error = ENOENT;
                goto bad;
 
        }
-       if (!error) {
-               PSHM_SUBSYS_UNLOCK();
-               error = EINVAL;
-               goto bad;
-       } else
-               incache = 1;
 
        if ((pinfo->pshm_flags & (PSHM_DEFINED | PSHM_ALLOCATED))==0) {
                PSHM_SUBSYS_UNLOCK();
@@ -1088,6 +1161,7 @@ shm_unlink(__unused proc_t p, struct shm_unlink_args *uap,
                error = 0;
                goto bad;
        }
+
 #if CONFIG_MACF
        error = mac_posixshm_check_unlink(kauth_cred_get(), pinfo, nameptr);
        if (error) {
@@ -1100,46 +1174,20 @@ shm_unlink(__unused proc_t p, struct shm_unlink_args *uap,
                  pinfo->pshm_mode);
 
        /* 
-        * following file semantics, unlink should be allowed 
-        * for users with write permission only. 
+        * following file semantics, unlink should be allowed
+        * for users with write permission only.
         */
        if ( (error = pshm_access(pinfo, FWRITE, kauth_cred_get(), p)) ) {
                PSHM_SUBSYS_UNLOCK();
                goto bad;
        }
 
-       pinfo->pshm_flags |= PSHM_INDELETE;
-       pshm_cache_delete(pcache);
-       pinfo->pshm_flags |= PSHM_REMOVED;
-       /* release the existence reference */
-       if (!--pinfo->pshm_usecount) {
-#if CONFIG_MACF
-               mac_posixshm_label_destroy(pinfo);
-#endif
-               PSHM_SUBSYS_UNLOCK();
-               /*
-                * If this is the last reference going away on the object,
-                * then we need to destroy the backing object.  The name
-                * has an implied but uncounted reference on the object,
-                * once it's created, since it's used as a rendezvous, and
-                * therefore may be subsequently reopened.
-                */
-               for (pshmobj = pinfo->pshm_memobjects;
-                    pshmobj != NULL;
-                    pshmobj = pshmobj_next) {
-                       mach_memory_entry_port_release(pshmobj->pshmo_memobject);
-                       pshmobj_next = pshmobj->pshmo_next;
-                       FREE(pshmobj, M_SHM);
-               }
-               FREE(pinfo,M_SHM);
-       } else {
-               PSHM_SUBSYS_UNLOCK();
-       }
-       FREE(pcache, M_SHM);
-       error = 0;
+       error = pshm_unlink_internal(pinfo, pcache);
+       PSHM_SUBSYS_UNLOCK();
+
 bad:
        FREE_ZONE(pnbuf, MAXPATHLEN, M_NAMEI);
-       return (error);
+       return error;
 }
 
 /* already called locked */