Further pseudofs improvements:

The pfs_info mutex is only needed to lock pi_unrhdr.  Everything else
in struct pfs_info is modified only while Giant is held (during
vfs_init() / vfs_uninit()); add assertions to that effect.

Simplify pfs_destroy somewhat.

Remove superfluous arguments from pfs_fileno_{alloc,free}(), and the
assertions which were added in the previous commit to ensure they were
consistent.

Assert that Giant is held while the vnode cache is initialized and
destroyed.  Also assert that the cache is empty when it is destroyed.

Rename the vnode cache mutex for consistency.

Fix a long-standing bug in pfs_getattr(): it would uncritically return
the node's pn_fileno as st_ino.  This would result in st_ino being 0
if the node had not previously been visited by readdir(), and also in
an incorrect st_ino for process directories and any files contained
therein.  Correct this by abstracting the fileno manipulations
previously done in pfs_readdir() into a new function, pfs_fileno(),
which is used by both pfs_getattr() and pfs_readdir().

6 files changed, 59 insertions, 62 deletions

diff --git a/sys/fs/pseudofs/pseudofs.c b/sys/fs/pseudofs/pseudofs.c
index 9464e9c..82ec971 100644
--- a/sys/fs/pseudofs/pseudofs.c
+++ b/sys/fs/pseudofs/pseudofs.c
@@ -73,7 +73,6 @@ _pfs_add_node(struct pfs_node *parent, struct pfs_node *node)
 
 	/* XXX should check for duplicate names etc. */
 
-	mtx_lock(&parent->pn_info->pi_mutex);
 	node->pn_info = parent->pn_info;
 	node->pn_parent = parent;
 	node->pn_next = parent->pn_nodes;
@@ -81,7 +80,6 @@ _pfs_add_node(struct pfs_node *parent, struct pfs_node *node)
 	/* Propagate flag to all child nodes (and thus their vnodes) */
 	if ((parent->pn_flags & PFS_PROCDEP) != 0)
 		node->pn_flags |= PFS_PROCDEP;
-	mtx_unlock(&parent->pn_info->pi_mutex);
 
 	return (0);
 }
@@ -210,8 +208,8 @@ pfs_find_node(struct pfs_node *parent, const char *name)
 
 	for (node = parent->pn_nodes; node != NULL; node = node->pn_next)
 		if (strcmp(node->pn_name, name) == 0)
-			return (node);
-	return (NULL);
+			break;
+	return (node);
 }
 
 /*
@@ -220,7 +218,7 @@ pfs_find_node(struct pfs_node *parent, const char *name)
 int
 pfs_destroy(struct pfs_node *node)
 {
-	struct pfs_node *parent, *rover;
+	struct pfs_node *parent, **rover;
 
 	KASSERT(node != NULL,
 	    ("%s(): node is NULL", __func__));
@@ -238,20 +236,14 @@ pfs_destroy(struct pfs_node *node)
 	if ((parent = node->pn_parent) != NULL) {
 		KASSERT(parent->pn_info == node->pn_info,
 		    ("%s(): parent has different pn_info", __func__));
-		mtx_lock(&node->pn_info->pi_mutex);
-		if (parent->pn_nodes == node) {
-			parent->pn_nodes = node->pn_next;
-		} else {
-			rover = parent->pn_nodes;
-			while (rover->pn_next != NULL) {
-				if (rover->pn_next == node) {
-					rover->pn_next = node->pn_next;
-					break;
-				}
-				rover = rover->pn_next;
+		rover = &parent->pn_nodes;
+		while (*rover != NULL) {
+			if (*rover == node) {
+				*rover = node->pn_next;
+				break;
 			}
+			rover = &(*rover)->pn_next;
 		}
-		mtx_unlock(&node->pn_info->pi_mutex);
 	}
 
 	/* callback to free any private resources */
@@ -260,7 +252,7 @@ pfs_destroy(struct pfs_node *node)
 
 	/* revoke fileno and vnodes and release memory */
 	if (node->pn_fileno)
-		pfs_fileno_free(node->pn_info, node);
+		pfs_fileno_free(node);
 	pfs_purge(node);
 	FREE(node, M_PFSNODES);
 
@@ -358,7 +350,7 @@ pfs_init(struct pfs_info *pi, struct vfsconf *vfc)
 	struct pfs_node *root;
 	int error;
 
-	mtx_init(&pi->pi_mutex, "pseudofs", NULL, MTX_DEF);
+	mtx_assert(&Giant, MA_OWNED);
 
 	/* set up the root diretory */
 	MALLOC(root, struct pfs_node *, sizeof *root,
@@ -377,7 +369,6 @@ pfs_init(struct pfs_info *pi, struct vfsconf *vfc)
 	if (error) {
 		pfs_destroy(root);
 		pi->pi_root = NULL;
-		mtx_destroy(&pi->pi_mutex);
 		return (error);
 	}
 
@@ -395,10 +386,11 @@ pfs_uninit(struct pfs_info *pi, struct vfsconf *vfc)
 {
 	int error;
 
+	mtx_assert(&Giant, MA_OWNED);
+
 	pfs_destroy(pi->pi_root);
 	pi->pi_root = NULL;
 	pfs_fileno_uninit(pi);
-	mtx_destroy(&pi->pi_mutex);
 	if (bootverbose)
 		printf("%s unregistered\n", pi->pi_name);
 	error = (pi->pi_uninit)(pi, vfc);
diff --git a/sys/fs/pseudofs/pseudofs.h b/sys/fs/pseudofs/pseudofs.h
index bb1efd1..4fa3311 100644
--- a/sys/fs/pseudofs/pseudofs.h
+++ b/sys/fs/pseudofs/pseudofs.h
@@ -165,14 +165,18 @@ typedef int (*pfs_destroy_t)(PFS_DESTROY_ARGS);
 
 /*
  * pfs_info: describes a pseudofs instance
+ *
+ * The pi_mutex is only used to avoid using the global subr_unit lock for
+ * unrhdr.  The rest of struct pfs_info is only modified while Giant is
+ * held (during vfs_init() and vfs_uninit()).
  */
 struct pfs_info {
 	char			 pi_name[PFS_FSNAMELEN];
 	pfs_init_t		 pi_init;
 	pfs_init_t		 pi_uninit;
-	/* members below this line aren't initialized */
+
+	/* members below this line are initialized at run time*/
 	struct pfs_node		*pi_root;
-	/* currently, the mutex is only used to protect the bitmap */
 	struct mtx		 pi_mutex;
 	struct unrhdr		*pi_unrhdr;
 };
diff --git a/sys/fs/pseudofs/pseudofs_fileno.c b/sys/fs/pseudofs/pseudofs_fileno.c
index 10c1617..155b0f5 100644
--- a/sys/fs/pseudofs/pseudofs_fileno.c
+++ b/sys/fs/pseudofs/pseudofs_fileno.c
@@ -38,6 +38,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/lock.h>
 #include <sys/malloc.h>
 #include <sys/mutex.h>
+#include <sys/proc.h>
 #include <sys/sysctl.h>
 #include <sys/systm.h>
 
@@ -50,12 +51,10 @@ __FBSDID("$FreeBSD$");
 void
 pfs_fileno_init(struct pfs_info *pi)
 {
-	struct unrhdr *up;
 
-	up = new_unrhdr(3, INT_MAX, &pi->pi_mutex);
-	mtx_lock(&pi->pi_mutex);
-	pi->pi_unrhdr = up;
-	mtx_unlock(&pi->pi_mutex);
+	mtx_assert(&Giant, MA_OWNED);
+	mtx_init(&pi->pi_mutex, "pfs_fileno", NULL, MTX_DEF);
+	pi->pi_unrhdr = new_unrhdr(3, INT_MAX / NO_PID, &pi->pi_mutex);
 }
 
 /*
@@ -64,30 +63,23 @@ pfs_fileno_init(struct pfs_info *pi)
 void
 pfs_fileno_uninit(struct pfs_info *pi)
 {
-	struct unrhdr *up;
 
-	mtx_lock(&pi->pi_mutex);
-
-	up = pi->pi_unrhdr;
+	mtx_assert(&Giant, MA_OWNED);
+	delete_unrhdr(pi->pi_unrhdr);
 	pi->pi_unrhdr = NULL;
-
-	mtx_unlock(&pi->pi_mutex);
-
-	delete_unrhdr(up);
+	mtx_destroy(&pi->pi_mutex);
 }
 
 /*
  * Allocate a file number
  */
 void
-pfs_fileno_alloc(struct pfs_info *pi, struct pfs_node *pn)
+pfs_fileno_alloc(struct pfs_node *pn)
 {
-	/* pi is not really necessary as it can be derived */
-	KASSERT(pi == pn->pn_info, ("pn / pi mismatch"));
 
 	/* make sure our parent has a file number */
 	if (pn->pn_parent && !pn->pn_parent->pn_fileno)
-		pfs_fileno_alloc(pi, pn->pn_parent);
+		pfs_fileno_alloc(pn->pn_parent);
 
 	switch (pn->pn_type) {
 	case pfstype_root:
@@ -98,7 +90,7 @@ pfs_fileno_alloc(struct pfs_info *pi, struct pfs_node *pn)
 	case pfstype_file:
 	case pfstype_symlink:
 	case pfstype_procdir:
-		pn->pn_fileno = alloc_unr(pi->pi_unrhdr);
+		pn->pn_fileno = alloc_unr(pn->pn_info->pi_unrhdr);
 		break;
 	case pfstype_this:
 		KASSERT(pn->pn_parent != NULL,
@@ -108,7 +100,7 @@ pfs_fileno_alloc(struct pfs_info *pi, struct pfs_node *pn)
 	case pfstype_parent:
 		KASSERT(pn->pn_parent != NULL,
 		    ("pfstype_parent node has no parent"));
-		if (pn->pn_parent == pi->pi_root) {
+		if (pn->pn_parent == pn->pn_info->pi_root) {
 			pn->pn_fileno = pn->pn_parent->pn_fileno;
 			break;
 		}
@@ -123,7 +115,7 @@ pfs_fileno_alloc(struct pfs_info *pi, struct pfs_node *pn)
 	}
 
 #if 0
-	printf("pfs_fileno_alloc(): %s: ", pi->pi_name);
+	printf("pfs_fileno_alloc(): %s: ", pn->pn_info->pi_name);
 	if (pn->pn_parent) {
 		if (pn->pn_parent->pn_parent) {
 			printf("%s/", pn->pn_parent->pn_parent->pn_name);
@@ -138,10 +130,8 @@ pfs_fileno_alloc(struct pfs_info *pi, struct pfs_node *pn)
  * Release a file number
  */
 void
-pfs_fileno_free(struct pfs_info *pi, struct pfs_node *pn)
+pfs_fileno_free(struct pfs_node *pn)
 {
-	/* pi is not really necessary as it can be derived */
-	KASSERT(pi == pn->pn_info, ("pn / pi mismatch"));
 
 	switch (pn->pn_type) {
 	case pfstype_root:
@@ -151,7 +141,7 @@ pfs_fileno_free(struct pfs_info *pi, struct pfs_node *pn)
 	case pfstype_file:
 	case pfstype_symlink:
 	case pfstype_procdir:
-		free_unr(pi->pi_unrhdr, pn->pn_fileno);
+		free_unr(pn->pn_info->pi_unrhdr, pn->pn_fileno);
 		break;
 	case pfstype_this:
 	case pfstype_parent:
diff --git a/sys/fs/pseudofs/pseudofs_internal.h b/sys/fs/pseudofs/pseudofs_internal.h
index 6abd704..8f213ed 100644
--- a/sys/fs/pseudofs/pseudofs_internal.h
+++ b/sys/fs/pseudofs/pseudofs_internal.h
@@ -61,7 +61,7 @@ int	 pfs_vncache_free	(struct vnode *);
  */
 void	 pfs_fileno_init	(struct pfs_info *);
 void	 pfs_fileno_uninit	(struct pfs_info *);
-void	 pfs_fileno_alloc	(struct pfs_info *, struct pfs_node *);
-void	 pfs_fileno_free	(struct pfs_info *, struct pfs_node *);
+void	 pfs_fileno_alloc	(struct pfs_node *);
+void	 pfs_fileno_free	(struct pfs_node *);
 
 #endif
diff --git a/sys/fs/pseudofs/pseudofs_vncache.c b/sys/fs/pseudofs/pseudofs_vncache.c
index e4d8db8..7544cb1 100644
--- a/sys/fs/pseudofs/pseudofs_vncache.c
+++ b/sys/fs/pseudofs/pseudofs_vncache.c
@@ -83,7 +83,9 @@ extern struct vop_vector pfs_vnodeops;	/* XXX -> .h file */
 void
 pfs_vncache_load(void)
 {
-	mtx_init(&pfs_vncache_mutex, "pseudofs_vncache", NULL, MTX_DEF);
+
+	mtx_assert(&Giant, MA_OWNED);
+	mtx_init(&pfs_vncache_mutex, "pfs_vncache", NULL, MTX_DEF);
 	pfs_exit_tag = EVENTHANDLER_REGISTER(process_exit, pfs_exit, NULL,
 	    EVENTHANDLER_PRI_ANY);
 }
@@ -94,10 +96,11 @@ pfs_vncache_load(void)
 void
 pfs_vncache_unload(void)
 {
+
+	mtx_assert(&Giant, MA_OWNED);
 	EVENTHANDLER_DEREGISTER(process_exit, pfs_exit_tag);
-	if (pfs_vncache_entries != 0)
-		printf("pfs_vncache_unload(): %d entries remaining\n",
-		    pfs_vncache_entries);
+	KASSERT(pfs_vncache_entries == 0,
+	    ("%d vncache entries remaining", pfs_vncache_entries));
 	mtx_destroy(&pfs_vncache_mutex);
 }
 
diff --git a/sys/fs/pseudofs/pseudofs_vnops.c b/sys/fs/pseudofs/pseudofs_vnops.c
index 347513f..e36e102 100644
--- a/sys/fs/pseudofs/pseudofs_vnops.c
+++ b/sys/fs/pseudofs/pseudofs_vnops.c
@@ -81,6 +81,19 @@ SYSCTL_INT(_vfs_pfs, OID_AUTO, trace, CTLFLAG_RW, &pfs_trace, 0,
 #endif
 
 /*
+ *
+ */
+static uint32_t
+pfs_fileno(struct pfs_node *pn, pid_t pid)
+{
+	if (!pn->pn_fileno)
+		pfs_fileno_alloc(pn);
+	if (pid != NO_PID)
+		return (pn->pn_fileno * NO_PID + pid);
+	return (pn->pn_fileno);
+}
+
+/*
  * Returns non-zero if given file is visible to given process.  If the 'p'
  * parameter is non-NULL, then it will hold a pointer to the process the
  * given file belongs to on return and the process will be locked.
@@ -193,7 +206,7 @@ pfs_getattr(struct vop_getattr_args *va)
 
 	VATTR_NULL(vap);
 	vap->va_type = vn->v_type;
-	vap->va_fileid = pn->pn_fileno;
+	vap->va_fileid = pfs_fileno(pn, pvd->pvd_pid);
 	vap->va_flags = 0;
 	vap->va_blocksize = PAGE_SIZE;
 	vap->va_bytes = vap->va_size = 0;
@@ -591,7 +604,6 @@ static int
 pfs_readdir(struct vop_readdir_args *va)
 {
 	struct vnode *vn = va->a_vp;
-	struct pfs_info *pi = (struct pfs_info *)vn->v_mount->mnt_data;
 	struct pfs_vdata *pvd = (struct pfs_vdata *)vn->v_data;
 	struct pfs_node *pd = pvd->pvd_pn;
 	pid_t pid = pvd->pvd_pid;
@@ -603,6 +615,8 @@ pfs_readdir(struct vop_readdir_args *va)
 	int error, i, resid;
 	char *buf, *ent;
 
+	KASSERT(pd->pn_info == vn->v_mount->mnt_data,
+	    ("directory's pn_info does not match mountpoint's mnt_data"));
 	PFS_TRACE((pd->pn_name));
 
 	if (vn->v_type != VDIR)
@@ -639,12 +653,7 @@ pfs_readdir(struct vop_readdir_args *va)
 		entry->d_reclen = PFS_DELEN;
 		if (!pn->pn_parent)
 			pn->pn_parent = pd;
-		if (!pn->pn_fileno)
-			pfs_fileno_alloc(pi, pn);
-		if (pid != NO_PID)
-			entry->d_fileno = pn->pn_fileno * NO_PID + pid;
-		else
-			entry->d_fileno = pn->pn_fileno;
+		entry->d_fileno = pfs_fileno(pn, pid);
 		/* PFS_DELEN was picked to fit PFS_NAMLEN */
 		for (i = 0; i < PFS_NAMELEN - 1 && pn->pn_name[i] != '\0'; ++i)
 			entry->d_name[i] = pn->pn_name[i];
@@ -654,7 +663,6 @@ pfs_readdir(struct vop_readdir_args *va)
 		case pfstype_procdir:
 			KASSERT(p != NULL,
 			    ("reached procdir node with p == NULL"));
-			entry->d_fileno = pn->pn_fileno * NO_PID + p->p_pid;
 			entry->d_namlen = snprintf(entry->d_name,
 			    PFS_NAMELEN, "%d", p->p_pid);
 			/* fall through */