--- a/usr/src/cmd/mdb/common/modules/genunix/genunix.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/cmd/mdb/common/modules/genunix/genunix.c Thu Aug 12 17:34:26 2010 -0700
@@ -4171,7 +4171,7 @@
pfiles_help },
/* from zone.c */
- { "zone", "?", "display kernel zone(s)", zoneprt },
+ { "zone", "?[-r [-v]]", "display kernel zone(s)", zoneprt },
{ "zsd", ":[-v] [zsd_key]", "display zone-specific-data entries for "
"selected zones", zsd },
--- a/usr/src/cmd/mdb/common/modules/genunix/zone.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/cmd/mdb/common/modules/genunix/zone.c Thu Aug 12 17:34:26 2010 -0700
@@ -19,8 +19,7 @@
* CDDL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
*/
#include <mdb/mdb_param.h>
@@ -39,6 +38,22 @@
#define ZONE_PATHLEN 40
#endif
+/*
+ * Names corresponding to zone_status_t values in sys/zone.h
+ */
+char *zone_status_names[] = {
+ "uninitialized", /* ZONE_IS_UNINITIALIZED */
+ "initialized", /* ZONE_IS_INITIALIZED */
+ "ready", /* ZONE_IS_READY */
+ "booting", /* ZONE_IS_BOOTING */
+ "running", /* ZONE_IS_RUNNING */
+ "shutting_down", /* ZONE_IS_SHUTTING_DOWN */
+ "empty", /* ZONE_IS_EMPTY */
+ "down", /* ZONE_IS_DOWN */
+ "dying", /* ZONE_IS_DYING */
+ "dead" /* ZONE_IS_DEAD */
+};
+
int
zoneprt(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
{
@@ -46,8 +61,10 @@
char name[ZONE_NAMELEN];
char path[ZONE_PATHLEN];
int len;
+ uint_t vopt_given;
+ uint_t ropt_given;
- if (argc != 0)
+ if (argc > 2)
return (DCMD_USAGE);
if (!(flags & DCMD_ADDRSPEC)) {
@@ -57,10 +74,38 @@
}
return (DCMD_OK);
}
+
+ /*
+ * Get the optional -r (reference counts) and -v (verbose output)
+ * arguments.
+ */
+ vopt_given = FALSE;
+ ropt_given = FALSE;
+ if (argc > 0 && mdb_getopts(argc, argv, 'v', MDB_OPT_SETBITS, TRUE,
+ &vopt_given, 'r', MDB_OPT_SETBITS, TRUE, &ropt_given, NULL) != argc)
+ return (DCMD_USAGE);
+
+ /*
+ * -v can only be specified with -r.
+ */
+ if (vopt_given == TRUE && ropt_given == FALSE)
+ return (DCMD_USAGE);
+
+ /*
+ * Print a table header, if necessary.
+ */
if (DCMD_HDRSPEC(flags)) {
- mdb_printf("%<u>%?s %6s %-20s %-s%</u>\n",
- "ADDR", "ID", "NAME", "PATH");
+ if (ropt_given == FALSE)
+ mdb_printf("%<u>%?s %6s %-13s %-20s %-s%</u>\n",
+ "ADDR", "ID", "STATUS", "NAME", "PATH");
+ else
+ mdb_printf("%<u>%?s %6s %10s %10s %-20s%</u>\n",
+ "ADDR", "ID", "REFS", "CREFS", "NAME");
}
+
+ /*
+ * Read the zone_t structure at the given address and read its name.
+ */
if (mdb_vread(&zn, sizeof (zone_t), addr) == -1) {
mdb_warn("can't read zone_t structure at %p", addr);
return (DCMD_ERR);
@@ -72,14 +117,110 @@
} else {
(void) strcpy(name, "??");
}
- len = mdb_readstr(path, ZONE_PATHLEN, (uintptr_t)zn.zone_rootpath);
- if (len > 0) {
- if (len == ZONE_PATHLEN)
- (void) strcpy(&path[len - 4], "...");
+
+ if (ropt_given == FALSE) {
+ char *statusp;
+
+ /*
+ * Default display
+ * Fetch the zone's path and print the results.
+ */
+ len = mdb_readstr(path, ZONE_PATHLEN,
+ (uintptr_t)zn.zone_rootpath);
+ if (len > 0) {
+ if (len == ZONE_PATHLEN)
+ (void) strcpy(&path[len - 4], "...");
+ } else {
+ (void) strcpy(path, "??");
+ }
+ if (zn.zone_status >= ZONE_IS_UNINITIALIZED && zn.zone_status <=
+ ZONE_IS_DEAD)
+ statusp = zone_status_names[zn.zone_status];
+ else
+ statusp = "???";
+ mdb_printf("%0?p %6d %-13s %-20s %s\n", addr, zn.zone_id,
+ statusp, name, path);
} else {
- (void) strcpy(path, "??");
+ /*
+ * Display the zone's reference counts.
+ * Display the zone's subsystem-specific reference counts if
+ * the user specified the '-v' option.
+ */
+ mdb_printf("%0?p %6d %10u %10u %-20s\n", addr, zn.zone_id,
+ zn.zone_ref, zn.zone_cred_ref, name);
+ if (vopt_given == TRUE) {
+ GElf_Sym subsys_names_sym;
+ uintptr_t **zone_ref_subsys_names;
+ uint_t num_subsys;
+ uint_t n;
+
+ /*
+ * Read zone_ref_subsys_names from the kernel image.
+ */
+ if (mdb_lookup_by_name("zone_ref_subsys_names",
+ &subsys_names_sym) != 0) {
+ mdb_warn("can't find zone_ref_subsys_names");
+ return (DCMD_ERR);
+ }
+ if (subsys_names_sym.st_size != ZONE_REF_NUM_SUBSYS *
+ sizeof (char *)) {
+ mdb_warn("number of subsystems in target "
+ "differs from what mdb expects (mismatched"
+ " kernel versions?)");
+ if (subsys_names_sym.st_size <
+ ZONE_REF_NUM_SUBSYS * sizeof (char *))
+ num_subsys = subsys_names_sym.st_size /
+ sizeof (char *);
+ else
+ num_subsys = ZONE_REF_NUM_SUBSYS;
+ } else {
+ num_subsys = ZONE_REF_NUM_SUBSYS;
+ }
+ if ((zone_ref_subsys_names = mdb_alloc(
+ subsys_names_sym.st_size, UM_GC)) == NULL) {
+ mdb_warn("out of memory");
+ return (DCMD_ERR);
+ }
+ if (mdb_readvar(zone_ref_subsys_names,
+ "zone_ref_subsys_names") == -1) {
+ mdb_warn("can't find zone_ref_subsys_names");
+ return (DCMD_ERR);
+ }
+
+ /*
+ * Display each subsystem's reference count if it's
+ * nonzero.
+ */
+ mdb_inc_indent(7);
+ for (n = 0; n < num_subsys; ++n) {
+ char subsys_name[16];
+
+ /*
+ * Skip subsystems lacking outstanding
+ * references.
+ */
+ if (zn.zone_subsys_ref[n] == 0)
+ continue;
+
+ /*
+ * Each subsystem's name must be read from
+ * the target's image.
+ */
+ if (mdb_readstr(subsys_name,
+ sizeof (subsys_name),
+ (uintptr_t)zone_ref_subsys_names[n]) ==
+ -1) {
+ mdb_warn("unable to read subsystem name"
+ " from zone_ref_subsys_names[%u]",
+ n);
+ return (DCMD_ERR);
+ }
+ mdb_printf("%15s: %10u\n", subsys_name,
+ zn.zone_subsys_ref[n]);
+ }
+ mdb_dec_indent(7);
+ }
}
- mdb_printf("%0?p %6d %-20s %s\n", addr, zn.zone_id, name, path);
return (DCMD_OK);
}
--- a/usr/src/uts/common/fs/mntfs/mntvfsops.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/mntfs/mntvfsops.c Thu Aug 12 17:34:26 2010 -0700
@@ -213,7 +213,8 @@
}
mutex_exit(&mvp->v_lock);
- zone_hold(mnt->mnt_zone = zone);
+ zone_init_ref(&mnt->mnt_zone_ref);
+ zone_hold_ref(zone, &mnt->mnt_zone_ref, ZONE_REF_MNTFS);
mnp = &mnt->mnt_node;
vfsp->vfs_fstype = mntfstype;
@@ -256,7 +257,7 @@
}
mutex_exit(&vp->v_lock);
- zone_rele(mnt->mnt_zone);
+ zone_rele_ref(&mnt->mnt_zone_ref, ZONE_REF_MNTFS);
vn_invalid(vp);
vn_free(vp);
kmem_free(mnt, sizeof (*mnt));
--- a/usr/src/uts/common/fs/mntfs/mntvnops.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/mntfs/mntvnops.c Thu Aug 12 17:34:26 2010 -0700
@@ -398,7 +398,7 @@
static void
mntfs_freesnap(mntnode_t *mnp, mntsnap_t *snapp)
{
- zone_t *zonep = MTOD(mnp)->mnt_zone;
+ zone_t *zonep = MTOD(mnp)->mnt_zone_ref.zref_zone;
krwlock_t *dblockp = &zonep->zone_mntfs_db_lock;
mntelem_t **elempp = &zonep->zone_mntfs_db;
mntelem_t *elemp;
@@ -500,7 +500,7 @@
mntfs_snapshot(mntnode_t *mnp, mntsnap_t *snapp)
{
mntdata_t *mnd = MTOD(mnp);
- zone_t *zonep = mnd->mnt_zone;
+ zone_t *zonep = mnd->mnt_zone_ref.zref_zone;
int is_global_zone = (zonep == global_zone);
int show_hidden = mnp->mnt_flags & MNT_SHOWHIDDEN;
vfs_t *vfsp, *firstvfsp, *lastvfsp;
@@ -885,7 +885,7 @@
mntread(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cred, caller_context_t *ct)
{
mntnode_t *mnp = VTOM(vp);
- zone_t *zonep = MTOD(mnp)->mnt_zone;
+ zone_t *zonep = MTOD(mnp)->mnt_zone_ref.zref_zone;
mntsnap_t *snapp = &mnp->mnt_read;
off_t off = uio->uio_offset;
size_t len = uio->uio_resid;
@@ -1472,7 +1472,7 @@
mntnode_t *mnp = VTOM(vp);
mntsnap_t *snapp = &mnp->mnt_ioctl;
int error = 0;
- zone_t *zonep = MTOD(mnp)->mnt_zone;
+ zone_t *zonep = MTOD(mnp)->mnt_zone_ref.zref_zone;
krwlock_t *dblockp = &zonep->zone_mntfs_db_lock;
model_t datamodel = flag & DATAMODEL_MASK;
@@ -1554,7 +1554,7 @@
size_t len;
uint_t start = 0;
mntdata_t *mntdata = MTOD(mnp);
- zone_t *zone = mntdata->mnt_zone;
+ zone_t *zone = mntdata->mnt_zone_ref.zref_zone;
STRUCT_INIT(tagdesc, flag & DATAMODEL_MASK);
if (copyin(dp, STRUCT_BUF(tagdesc), STRUCT_SIZE(tagdesc))) {
--- a/usr/src/uts/common/fs/nfs/nfs3_vfsops.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/nfs/nfs3_vfsops.c Thu Aug 12 17:34:26 2010 -0700
@@ -19,8 +19,7 @@
* CDDL HEADER END
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
*/
/*
@@ -1122,7 +1121,9 @@
cv_init(&mi->mi_async_cv, NULL, CV_DEFAULT, NULL);
mi->mi_vfsp = vfsp;
- zone_hold(mi->mi_zone = zone);
+ mi->mi_zone = zone;
+ zone_init_ref(&mi->mi_zone_ref);
+ zone_hold_ref(zone, &mi->mi_zone_ref, ZONE_REF_NFS);
nfs_mi_zonelist_add(mi);
/*
--- a/usr/src/uts/common/fs/nfs/nfs4_client.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/nfs/nfs4_client.c Thu Aug 12 17:34:26 2010 -0700
@@ -19,8 +19,7 @@
* CDDL HEADER END
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
*/
/*
@@ -2932,7 +2931,8 @@
*/
list_remove(&mig->mig_list, mi);
mutex_exit(&mig->mig_lock);
- zone_rele(mi->mi_zone);
+ zone_rele_ref(&mi->mi_zone_ref, ZONE_REF_NFSV4);
+
/*
* Release hold on vfs and mi done to prevent race with zone
* shutdown. This releases the hold in nfs4_mi_zonelist_add.
--- a/usr/src/uts/common/fs/nfs/nfs4_vfsops.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/nfs/nfs4_vfsops.c Thu Aug 12 17:34:26 2010 -0700
@@ -1179,7 +1179,8 @@
/* need to remove it from the zone */
removed = nfs4_mi_zonelist_remove(mi);
if (removed)
- zone_rele(mi->mi_zone);
+ zone_rele_ref(&mi->mi_zone_ref,
+ ZONE_REF_NFSV4);
MI4_RELE(mi);
if (!(uap->flags & MS_SYSSPACE) && args) {
nfs4_free_args(args);
@@ -2383,7 +2384,9 @@
cv_init(&mi->mi_inact_req_cv, NULL, CV_DEFAULT, NULL);
mi->mi_vfsp = vfsp;
- zone_hold(mi->mi_zone = zone);
+ mi->mi_zone = zone;
+ zone_init_ref(&mi->mi_zone_ref);
+ zone_hold_ref(zone, &mi->mi_zone_ref, ZONE_REF_NFSV4);
nfs4_mi_zonelist_add(mi);
/*
@@ -2646,7 +2649,7 @@
nfs4_async_manager_stop(vfsp);
removed = nfs4_mi_zonelist_remove(mi);
if (removed)
- zone_rele(mi->mi_zone);
+ zone_rele_ref(&mi->mi_zone_ref, ZONE_REF_NFSV4);
/*
* This releases the initial "hold" of the mi since it will never
@@ -2771,7 +2774,7 @@
nfs4_remove_mi_from_server(mi, NULL);
removed = nfs4_mi_zonelist_remove(mi);
if (removed)
- zone_rele(mi->mi_zone);
+ zone_rele_ref(&mi->mi_zone_ref, ZONE_REF_NFSV4);
return (0);
}
@@ -4402,7 +4405,7 @@
removed = nfs4_mi_zonelist_remove(mi);
if (removed)
- zone_rele(mi->mi_zone);
+ zone_rele_ref(&mi->mi_zone_ref, ZONE_REF_NFSV4);
}
/* Referral related sub-routines */
--- a/usr/src/uts/common/fs/nfs/nfs_client.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/nfs/nfs_client.c Thu Aug 12 17:34:26 2010 -0700
@@ -19,8 +19,7 @@
* CDDL HEADER END
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
*
* Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T.
* All rights reserved.
@@ -3229,7 +3228,7 @@
cv_destroy(&mi->mi_async_work_cv[NFS_ASYNC_PGOPS_QUEUE]);
cv_destroy(&mi->mi_async_reqs_cv);
cv_destroy(&mi->mi_async_cv);
- zone_rele(mi->mi_zone);
+ zone_rele_ref(&mi->mi_zone_ref, ZONE_REF_NFS);
kmem_free(mi, sizeof (*mi));
}
--- a/usr/src/uts/common/fs/nfs/nfs_vfsops.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/nfs/nfs_vfsops.c Thu Aug 12 17:34:26 2010 -0700
@@ -19,8 +19,7 @@
* CDDL HEADER END
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
*
* Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T.
* All rights reserved.
@@ -1241,7 +1240,9 @@
cv_init(&mi->mi_async_cv, NULL, CV_DEFAULT, NULL);
mi->mi_vfsp = vfsp;
- zone_hold(mi->mi_zone = zone);
+ mi->mi_zone = zone;
+ zone_init_ref(&mi->mi_zone_ref);
+ zone_hold_ref(zone, &mi->mi_zone_ref, ZONE_REF_NFS);
nfs_mi_zonelist_add(mi);
/*
--- a/usr/src/uts/common/fs/smbclnt/smbfs/smbfs.h Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/smbclnt/smbfs/smbfs.h Thu Aug 12 17:34:26 2010 -0700
@@ -33,8 +33,7 @@
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
*/
#ifndef _SMBFS_SMBFS_H
@@ -56,6 +55,7 @@
#include <sys/vfs.h>
#include <sys/vfs_opreg.h>
#include <sys/fs/smbfs_mount.h>
+#include <sys/zone.h>
/*
* Path component length
@@ -140,7 +140,7 @@
/*
* Zones support.
*/
- struct zone *smi_zone; /* Zone mounted in */
+ zone_ref_t smi_zone_ref; /* Zone FS is mounted in */
list_node_t smi_zone_node; /* Link to per-zone smi list */
/* Lock for the list is: smi_globals_t -> smg_lock */
--- a/usr/src/uts/common/fs/smbclnt/smbfs/smbfs_client.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/smbclnt/smbfs/smbfs_client.c Thu Aug 12 17:34:26 2010 -0700
@@ -20,8 +20,7 @@
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
*
* Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T.
* All rights reserved.
@@ -418,7 +417,7 @@
smi = VTOSMI(vp);
- ASSERT(curproc->p_zone == smi->smi_zone);
+ ASSERT(curproc->p_zone == smi->smi_zone_ref.zref_zone);
/*
* If asked for UID or GID, update n_uid, n_gid.
@@ -610,7 +609,7 @@
{
smi_globals_t *smg;
- smg = zone_getspecific(smi_list_key, smi->smi_zone);
+ smg = zone_getspecific(smi_list_key, smi->smi_zone_ref.zref_zone);
mutex_enter(&smg->smg_lock);
list_insert_head(&smg->smg_list, smi);
mutex_exit(&smg->smg_lock);
@@ -624,7 +623,7 @@
{
smi_globals_t *smg;
- smg = zone_getspecific(smi_list_key, smi->smi_zone);
+ smg = zone_getspecific(smi_list_key, smi->smi_zone_ref.zref_zone);
mutex_enter(&smg->smg_lock);
list_remove(&smg->smg_list, smi);
/*
--- a/usr/src/uts/common/fs/smbclnt/smbfs/smbfs_vfsops.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/smbclnt/smbfs/smbfs_vfsops.c Thu Aug 12 17:34:26 2010 -0700
@@ -314,8 +314,8 @@
if (smi == NULL)
return;
- if (smi->smi_zone != NULL)
- zone_rele(smi->smi_zone);
+ if (smi->smi_zone_ref.zref_zone != NULL)
+ zone_rele_ref(&smi->smi_zone_ref, ZONE_REF_SMBFS);
if (smi->smi_share != NULL)
smb_share_rele(smi->smi_share);
@@ -481,7 +481,14 @@
smi->smi_share = ssp;
ssp = NULL;
- smi->smi_zone = mntzone;
+
+ /*
+ * Convert the anonymous zone hold acquired via zone_hold() above
+ * into a zone reference.
+ */
+ zone_init_ref(&smi->smi_zone_ref);
+ zone_hold_ref(mntzone, &smi->smi_zone_ref, ZONE_REF_SMBFS);
+ zone_rele(mntzone);
mntzone = NULL;
/*
@@ -736,7 +743,7 @@
smi = VFTOSMI(vfsp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EPERM);
if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -770,7 +777,7 @@
hrtime_t now;
smb_cred_t scred;
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EPERM);
if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
--- a/usr/src/uts/common/fs/smbclnt/smbfs/smbfs_vnops.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/smbclnt/smbfs/smbfs_vnops.c Thu Aug 12 17:34:26 2010 -0700
@@ -257,7 +257,7 @@
smi = VTOSMI(vp);
ssp = smi->smi_share;
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -434,7 +434,7 @@
* open; if we happen to get here from the wrong zone we can't do
* anything over the wire.
*/
- if (smi->smi_zone != curproc->p_zone) {
+ if (smi->smi_zone_ref.zref_zone != curproc->p_zone) {
/*
* We could attempt to clean up locks, except we're sure
* that the current process didn't acquire any locks on
@@ -591,7 +591,7 @@
smi = VTOSMI(vp);
ssp = smi->smi_share;
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -676,7 +676,7 @@
smi = VTOSMI(vp);
ssp = smi->smi_share;
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -787,7 +787,7 @@
smi = VTOSMI(vp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -853,7 +853,7 @@
smi = VTOSMI(vp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -908,7 +908,7 @@
vfsp = vp->v_vfsp;
smi = VFTOSMI(vfsp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -980,7 +980,7 @@
int have_fid = 0;
uint32_t rights = 0;
- ASSERT(curproc->p_zone == VTOSMI(vp)->smi_zone);
+ ASSERT(curproc->p_zone == VTOSMI(vp)->smi_zone_ref.zref_zone);
/*
* There are no settable attributes on the XATTR dir,
@@ -1252,7 +1252,7 @@
vfsp = vp->v_vfsp;
smi = VFTOSMI(vfsp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -1281,7 +1281,7 @@
np = VTOSMB(vp);
smi = VTOSMI(vp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -1399,7 +1399,7 @@
vfs = dvp->v_vfsp;
smi = VFTOSMI(vfs);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EPERM);
if (smi->smi_flags & SMI_DEAD || vfs->vfs_flag & VFS_UNMOUNTED)
@@ -1456,7 +1456,7 @@
smi = VTOSMI(dvp);
dnp = VTOSMB(dvp);
- ASSERT(curproc->p_zone == smi->smi_zone);
+ ASSERT(curproc->p_zone == smi->smi_zone_ref.zref_zone);
#ifdef NOT_YET
vcp = SSTOVC(smi->smi_share);
@@ -1775,7 +1775,7 @@
dnp = VTOSMB(dvp);
vp = NULL;
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EPERM);
if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -2007,7 +2007,7 @@
smi = VTOSMI(dvp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EPERM);
if (smi->smi_flags & SMI_DEAD || dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -2101,8 +2101,8 @@
{
/* vnode_t *realvp; */
- if (curproc->p_zone != VTOSMI(odvp)->smi_zone ||
- curproc->p_zone != VTOSMI(ndvp)->smi_zone)
+ if (curproc->p_zone != VTOSMI(odvp)->smi_zone_ref.zref_zone ||
+ curproc->p_zone != VTOSMI(ndvp)->smi_zone_ref.zref_zone)
return (EPERM);
if (VTOSMI(odvp)->smi_flags & SMI_DEAD ||
@@ -2133,7 +2133,7 @@
struct smb_cred scred;
/* enum smbfsstat status; */
- ASSERT(curproc->p_zone == VTOSMI(odvp)->smi_zone);
+ ASSERT(curproc->p_zone == VTOSMI(odvp)->smi_zone_ref.zref_zone);
if (strcmp(onm, ".") == 0 || strcmp(onm, "..") == 0 ||
strcmp(nnm, ".") == 0 || strcmp(nnm, "..") == 0)
@@ -2361,7 +2361,7 @@
int nmlen = strlen(name);
int error, hiderr;
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EPERM);
if (smi->smi_flags & SMI_DEAD || dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -2440,7 +2440,7 @@
struct smb_cred scred;
int error;
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EPERM);
if (smi->smi_flags & SMI_DEAD || dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -2536,7 +2536,7 @@
smi = VTOSMI(vp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -2599,7 +2599,7 @@
int nmlen, error;
ushort_t reclen;
- ASSERT(curproc->p_zone == VTOSMI(vp)->smi_zone);
+ ASSERT(curproc->p_zone == VTOSMI(vp)->smi_zone_ref.zref_zone);
/* Make sure we serialize for n_dirseq use. */
ASSERT(smbfs_rw_lock_held(&np->r_lkserlock, RW_WRITER));
@@ -2840,7 +2840,7 @@
smi = VTOSMI(vp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EPERM);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -2871,7 +2871,7 @@
offset_t offset, struct flk_callback *flk_cbp, cred_t *cr,
caller_context_t *ct)
{
- if (curproc->p_zone != VTOSMI(vp)->smi_zone)
+ if (curproc->p_zone != VTOSMI(vp)->smi_zone_ref.zref_zone)
return (EIO);
if (VTOSMI(vp)->smi_flags & SMI_LLOCK)
@@ -2897,7 +2897,7 @@
smi = VTOSMI(vp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -2951,7 +2951,7 @@
vfs = vp->v_vfsp;
smi = VFTOSMI(vfs);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -3019,7 +3019,7 @@
vfsp = vp->v_vfsp;
smi = VFTOSMI(vfsp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -3060,7 +3060,7 @@
vfsp = vp->v_vfsp;
smi = VFTOSMI(vfsp);
- if (curproc->p_zone != smi->smi_zone)
+ if (curproc->p_zone != smi->smi_zone_ref.zref_zone)
return (EIO);
if (smi->smi_flags & SMI_DEAD || vfsp->vfs_flag & VFS_UNMOUNTED)
@@ -3102,7 +3102,7 @@
smbfs_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr,
caller_context_t *ct)
{
- if (curproc->p_zone != VTOSMI(vp)->smi_zone)
+ if (curproc->p_zone != VTOSMI(vp)->smi_zone_ref.zref_zone)
return (EIO);
if (VTOSMI(vp)->smi_flags & SMI_LLOCK)
--- a/usr/src/uts/common/fs/vfs.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/fs/vfs.c Thu Aug 12 17:34:26 2010 -0700
@@ -3513,11 +3513,13 @@
/*
* The zone that owns the mount is the one that performed the mount.
* Note that this isn't necessarily the same as the zone mounted into.
- * The corresponding zone_rele() will be done when the vfs_t is
- * being free'd.
+ * The corresponding zone_rele_ref() will be done when the vfs_t
+ * is being free'd.
*/
vfsp->vfs_zone = curproc->p_zone;
- zone_hold(vfsp->vfs_zone);
+ zone_init_ref(&vfsp->vfs_implp->vi_zone_ref);
+ zone_hold_ref(vfsp->vfs_zone, &vfsp->vfs_implp->vi_zone_ref,
+ ZONE_REF_VFS);
/*
* Find the zone mounted into, and put this mount on its vfs list.
@@ -4342,7 +4344,8 @@
VFS_FREEVFS(vfsp);
lofi_remove(vfsp);
if (vfsp->vfs_zone)
- zone_rele(vfsp->vfs_zone);
+ zone_rele_ref(&vfsp->vfs_implp->vi_zone_ref,
+ ZONE_REF_VFS);
vfs_freemnttab(vfsp);
vfs_free(vfsp);
}
--- a/usr/src/uts/common/io/lofi.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/io/lofi.c Thu Aug 12 17:34:26 2010 -0700
@@ -373,8 +373,8 @@
lsp->ls_uncomp_seg_sz = 0;
}
- rctl_decr_lofi(lsp->ls_zone, 1);
- zone_rele(lsp->ls_zone);
+ rctl_decr_lofi(lsp->ls_zone.zref_zone, 1);
+ zone_rele_ref(&lsp->ls_zone, ZONE_REF_LOFI);
mutex_destroy(&lsp->ls_comp_cache_lock);
mutex_destroy(&lsp->ls_comp_bufs_lock);
@@ -418,7 +418,7 @@
/* lofi_destroy() frees lsp */
next = list_next(&lofi_list, lsp);
- if (lsp->ls_zone->zone_id != zoneid)
+ if (lsp->ls_zone.zref_zone->zone_id != zoneid)
continue;
/*
@@ -1610,7 +1610,7 @@
lofi_access(struct lofi_state *lsp)
{
ASSERT(MUTEX_HELD(&lofi_lock));
- if (INGLOBALZONE(curproc) || lsp->ls_zone == curproc->p_zone)
+ if (INGLOBALZONE(curproc) || lsp->ls_zone.zref_zone == curzone)
return (0);
return (EPERM);
}
@@ -2184,8 +2184,8 @@
newdev = makedevice(getmajor(dev), minor);
lsp->ls_dev = newdev;
- lsp->ls_zone = zone_find_by_id(getzoneid());
- ASSERT(lsp->ls_zone != NULL);
+ zone_init_ref(&lsp->ls_zone);
+ zone_hold_ref(curzone, &lsp->ls_zone, ZONE_REF_LOFI);
lsp->ls_uncomp_seg_sz = 0;
lsp->ls_comp_algorithm[0] = '\0';
lsp->ls_crypto_offset = 0;
--- a/usr/src/uts/common/nfs/nfs4_clnt.h Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/nfs/nfs4_clnt.h Thu Aug 12 17:34:26 2010 -0700
@@ -19,8 +19,7 @@
* CDDL HEADER END
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -48,6 +47,7 @@
#include <rpc/auth.h>
#include <sys/door.h>
#include <sys/condvar_impl.h>
+#include <sys/zone.h>
#ifdef __cplusplus
extern "C" {
@@ -1083,7 +1083,8 @@
/*
* Zones support.
*/
- struct zone *mi_zone; /* Zone mounted in */
+ struct zone *mi_zone; /* Zone in which FS is mounted */
+ zone_ref_t mi_zone_ref; /* Reference to aforementioned zone */
list_node_t mi_zone_node; /* linkage into per-zone mi list */
/*
--- a/usr/src/uts/common/nfs/nfs_clnt.h Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/nfs/nfs_clnt.h Thu Aug 12 17:34:26 2010 -0700
@@ -19,8 +19,7 @@
* CDDL HEADER END
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
*/
/* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
@@ -37,6 +36,7 @@
#include <nfs/rnode.h>
#include <sys/list.h>
#include <sys/condvar_impl.h>
+#include <sys/zone.h>
#ifdef __cplusplus
extern "C" {
@@ -287,6 +287,7 @@
#define ORIG_KNCONF(mi) (mi->mi_curr_serv->sv_origknconf ? \
mi->mi_curr_serv->sv_origknconf : mi->mi_curr_serv->sv_knconf)
+#if defined(_KERNEL)
/*
* NFS private data per mounted file system
* The mi_lock mutex protects the following fields:
@@ -418,7 +419,8 @@
/*
* Zones support.
*/
- struct zone *mi_zone; /* Zone mounted in */
+ struct zone *mi_zone; /* Zone in which FS is mounted */
+ zone_ref_t mi_zone_ref; /* Reference to aforementioned zone */
list_node_t mi_zone_node; /* Linkage into per-zone mi list */
/*
* Serializes threads in failover_remap.
@@ -427,6 +429,7 @@
*/
kmutex_t mi_remap_lock;
} mntinfo_t;
+#endif /* _KERNEL */
/*
* vfs pointer to mount info
--- a/usr/src/uts/common/os/ipc.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/os/ipc.c Thu Aug 12 17:34:26 2010 -0700
@@ -19,8 +19,7 @@
* CDDL HEADER END
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -842,7 +841,7 @@
ASSERT(IPC_FREE(perm)); /* ipc_rmid clears IPC_ALLOC */
s->ipcs_dtor(perm);
project_rele(perm->ipc_proj);
- zone_rele(perm->ipc_zone);
+ zone_rele_ref(&perm->ipc_zone_ref, ZONE_REF_IPC);
kmem_free(perm, s->ipcs_ssize);
}
}
@@ -1064,12 +1063,13 @@
ASSERT(IPC_FREE(newperm));
/*
- * Set ipc_proj and ipc_zone so that future calls to ipc_cleanup()
+ * Set ipc_proj and ipc_zone_ref so that future calls to ipc_cleanup()
* clean up the necessary state. This must be done before the
* potential call to ipcs_dtor() below.
*/
newperm->ipc_proj = pp->p_task->tk_proj;
- newperm->ipc_zone = pp->p_zone;
+ zone_init_ref(&newperm->ipc_zone_ref);
+ zone_hold_ref(pp->p_zone, &newperm->ipc_zone_ref, ZONE_REF_IPC);
mutex_enter(&service->ipcs_lock);
/*
@@ -1097,6 +1097,7 @@
errout:
mutex_exit(&service->ipcs_lock);
service->ipcs_dtor(newperm);
+ zone_rele_ref(&newperm->ipc_zone_ref, ZONE_REF_IPC);
kmem_free(newperm, service->ipcs_ssize);
return (error);
}
@@ -1118,7 +1119,6 @@
ASSERT(MUTEX_HELD(&curproc->p_lock));
(void) project_hold(perm->ipc_proj);
- (void) zone_hold(perm->ipc_zone);
mutex_exit(&curproc->p_lock);
/*
@@ -1173,6 +1173,8 @@
mutex_exit(&service->ipcs_lock);
service->ipcs_dtor(perm);
}
+ if (perm->ipc_zone_ref.zref_zone != NULL)
+ zone_rele_ref(&perm->ipc_zone_ref, ZONE_REF_IPC);
kmem_free(perm, service->ipcs_ssize);
}
--- a/usr/src/uts/common/os/shm.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/os/shm.c Thu Aug 12 17:34:26 2010 -0700
@@ -20,8 +20,7 @@
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
*/
/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
@@ -605,7 +604,7 @@
rsize = ptob(btopr(sp->shm_segsz));
ipcs_lock(shm_svc);
sp->shm_perm.ipc_proj->kpj_data.kpd_shmmax -= rsize;
- sp->shm_perm.ipc_zone->zone_shmmax -= rsize;
+ sp->shm_perm.ipc_zone_ref.zref_zone->zone_shmmax -= rsize;
ipcs_unlock(shm_svc);
}
}
@@ -924,13 +923,13 @@
sp->shm_perm.ipc_proj->kpj_rctls, pp, rsize,
RCA_SAFE) & RCT_DENY) ||
(rctl_test(rc_zone_shmmax,
- sp->shm_perm.ipc_zone->zone_rctls, pp, rsize,
+ sp->shm_perm.ipc_zone_ref.zref_zone->zone_rctls, pp, rsize,
RCA_SAFE) & RCT_DENY)) {
ipc_cleanup(shm_svc, (kipc_perm_t *)sp);
return (EINVAL);
}
sp->shm_perm.ipc_proj->kpj_data.kpd_shmmax += rsize;
- sp->shm_perm.ipc_zone->zone_shmmax += rsize;
+ sp->shm_perm.ipc_zone_ref.zref_zone->zone_shmmax += rsize;
lock = ipc_commit_end(shm_svc, &sp->shm_perm);
}
@@ -1258,7 +1257,7 @@
struct anon_map *amp = sp->shm_amp;
zone_t *zone;
- zone = sp->shm_perm.ipc_zone;
+ zone = sp->shm_perm.ipc_zone_ref.zref_zone;
ASSERT(zone != NULL);
/*
* Free up the anon_map.
--- a/usr/src/uts/common/os/zone.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/os/zone.c Thu Aug 12 17:34:26 2010 -0700
@@ -223,6 +223,7 @@
#include <sys/pool.h>
#include <sys/pool_pset.h>
#include <sys/pset.h>
+#include <sys/strlog.h>
#include <sys/sysmacros.h>
#include <sys/callb.h>
#include <sys/vmparam.h>
@@ -249,6 +250,17 @@
#include <vm/seg.h>
#include <sys/mac.h>
+/*
+ * This constant specifies the number of seconds that threads waiting for
+ * subsystems to release a zone's general-purpose references will wait before
+ * they log the zone's reference counts. The constant's value shouldn't
+ * be so small that reference counts are unnecessarily reported for zones
+ * whose references are slowly released. On the other hand, it shouldn't be so
+ * large that users reboot their systems out of frustration over hung zones
+ * before the system logs the zones' reference counts.
+ */
+#define ZONE_DESTROY_TIMEOUT_SECS 60
+
/* List of data link IDs which are accessible from the zone */
typedef struct zone_dl {
datalink_id_t zdl_id;
@@ -338,6 +350,20 @@
};
/*
+ * This array contains the names of the subsystems listed in zone_ref_subsys_t
+ * (see sys/zone.h).
+ */
+static char *zone_ref_subsys_names[] = {
+ "NFS", /* ZONE_REF_NFS */
+ "NFSv4", /* ZONE_REF_NFSV4 */
+ "SMBFS", /* ZONE_REF_SMBFS */
+ "MNTFS", /* ZONE_REF_MNTFS */
+ "LOFI", /* ZONE_REF_LOFI */
+ "VFS", /* ZONE_REF_VFS */
+ "IPC" /* ZONE_REF_IPC */
+};
+
+/*
* This isn't static so lint doesn't complain.
*/
rctl_hndl_t rc_zone_cpu_shares;
@@ -1881,6 +1907,8 @@
zone0.zone_lockedmem_kstat = NULL;
zone0.zone_swapresv_kstat = NULL;
zone0.zone_nprocs_kstat = NULL;
+ list_create(&zone0.zone_ref_list, sizeof (zone_ref_t),
+ offsetof(zone_ref_t, zref_linkage));
list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
offsetof(struct zsd_entry, zsd_linkage));
list_insert_head(&zone_active, &zone0);
@@ -2144,6 +2172,7 @@
ASSERT(zone->zone_kcred == NULL);
ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
+ ASSERT(list_is_empty(&zone->zone_ref_list));
/*
* Remove any zone caps.
@@ -2160,6 +2189,7 @@
mutex_exit(&zone_deathrow_lock);
}
+ list_destroy(&zone->zone_ref_list);
zone_free_zsd(zone);
zone_free_datasets(zone);
list_destroy(&zone->zone_dl_list);
@@ -2494,6 +2524,11 @@
* to force halt/reboot to block waiting for the zone_cred_ref to drop
* to 0. This can be useful to flush out other sources of cached creds
* that may be less innocuous than the driver case.
+ *
+ * Zones also provide a tracked reference counting mechanism in which zone
+ * references are represented by "crumbs" (zone_ref structures). Crumbs help
+ * debuggers determine the sources of leaked zone references. See
+ * zone_hold_ref() and zone_rele_ref() below for more information.
*/
int zone_wait_for_cred = 0;
@@ -2506,6 +2541,14 @@
ASSERT(z->zone_ref != 0);
}
+/*
+ * Increment the specified zone's reference count. The zone's zone_t structure
+ * will not be freed as long as the zone's reference count is nonzero.
+ * Decrement the zone's reference count via zone_rele().
+ *
+ * NOTE: This function should only be used to hold zones for short periods of
+ * time. Use zone_hold_ref() if the zone must be held for a long time.
+ */
void
zone_hold(zone_t *z)
{
@@ -2522,14 +2565,27 @@
#define ZONE_IS_UNREF(zone) ((zone)->zone_ref == 1 && \
(!zone_wait_for_cred || (zone)->zone_cred_ref == 0))
-void
-zone_rele(zone_t *z)
+/*
+ * Common zone reference release function invoked by zone_rele() and
+ * zone_rele_ref(). If subsys is ZONE_REF_NUM_SUBSYS, then the specified
+ * zone's subsystem-specific reference counters are not affected by the
+ * release. If ref is not NULL, then the zone_ref_t to which it refers is
+ * removed from the specified zone's reference list. ref must be non-NULL iff
+ * subsys is not ZONE_REF_NUM_SUBSYS.
+ */
+static void
+zone_rele_common(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
{
boolean_t wakeup;
mutex_enter(&z->zone_lock);
ASSERT(z->zone_ref != 0);
z->zone_ref--;
+ if (subsys != ZONE_REF_NUM_SUBSYS) {
+ ASSERT(z->zone_subsys_ref[subsys] != 0);
+ z->zone_subsys_ref[subsys]--;
+ list_remove(&z->zone_ref_list, ref);
+ }
if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
/* no more refs, free the structure */
mutex_exit(&z->zone_lock);
@@ -2551,6 +2607,83 @@
}
}
+/*
+ * Decrement the specified zone's reference count. The specified zone will
+ * cease to exist after this function returns if the reference count drops to
+ * zero. This function should be paired with zone_hold().
+ */
+void
+zone_rele(zone_t *z)
+{
+ zone_rele_common(z, NULL, ZONE_REF_NUM_SUBSYS);
+}
+
+/*
+ * Initialize a zone reference structure. This function must be invoked for
+ * a reference structure before the structure is passed to zone_hold_ref().
+ */
+void
+zone_init_ref(zone_ref_t *ref)
+{
+ ref->zref_zone = NULL;
+ list_link_init(&ref->zref_linkage);
+}
+
+/*
+ * Acquire a reference to zone z. The caller must specify the
+ * zone_ref_subsys_t constant associated with its subsystem. The specified
+ * zone_ref_t structure will represent a reference to the specified zone. Use
+ * zone_rele_ref() to release the reference.
+ *
+ * The referenced zone_t structure will not be freed as long as the zone_t's
+ * zone_status field is not ZONE_IS_DEAD and the zone has outstanding
+ * references.
+ *
+ * NOTE: The zone_ref_t structure must be initialized before it is used.
+ * See zone_init_ref() above.
+ */
+void
+zone_hold_ref(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
+{
+ ASSERT(subsys >= 0 && subsys < ZONE_REF_NUM_SUBSYS);
+
+ /*
+ * Prevent consumers from reusing a reference structure before
+ * releasing it.
+ */
+ VERIFY(ref->zref_zone == NULL);
+
+ ref->zref_zone = z;
+ mutex_enter(&z->zone_lock);
+ zone_hold_locked(z);
+ z->zone_subsys_ref[subsys]++;
+ ASSERT(z->zone_subsys_ref[subsys] != 0);
+ list_insert_head(&z->zone_ref_list, ref);
+ mutex_exit(&z->zone_lock);
+}
+
+/*
+ * Release the zone reference represented by the specified zone_ref_t.
+ * The reference is invalid after it's released; however, the zone_ref_t
+ * structure can be reused without having to invoke zone_init_ref().
+ * subsys should be the same value that was passed to zone_hold_ref()
+ * when the reference was acquired.
+ */
+void
+zone_rele_ref(zone_ref_t *ref, zone_ref_subsys_t subsys)
+{
+ zone_rele_common(ref->zref_zone, ref, subsys);
+
+ /*
+ * Set the zone_ref_t's zref_zone field to NULL to generate panics
+ * when consumers dereference the reference. This helps us catch
+ * consumers who use released references. Furthermore, this lets
+ * consumers reuse the zone_ref_t structure without having to
+ * invoke zone_init_ref().
+ */
+ ref->zref_zone = NULL;
+}
+
void
zone_cred_hold(zone_t *z)
{
@@ -3984,6 +4117,8 @@
mutex_init(&zone->zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&zone->zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&zone->zone_cv, NULL, CV_DEFAULT, NULL);
+ list_create(&zone->zone_ref_list, sizeof (zone_ref_t),
+ offsetof(zone_ref_t, zref_linkage));
list_create(&zone->zone_zsd, sizeof (struct zsd_entry),
offsetof(struct zsd_entry, zsd_linkage));
list_create(&zone->zone_datasets, sizeof (zone_dataset_t),
@@ -4269,7 +4404,7 @@
* zone_kcred. To free the zone, we call crfree, which will call
* zone_cred_rele, which will call zone_free.
*/
- ASSERT(zone->zone_cred_ref == 1); /* for zone_kcred */
+ ASSERT(zone->zone_cred_ref == 1);
ASSERT(zone->zone_kcred->cr_ref == 1);
ASSERT(zone->zone_ref == 0);
zkcr = zone->zone_kcred;
@@ -4555,6 +4690,101 @@
}
/*
+ * Log the specified zone's reference counts. The caller should not be
+ * holding the zone's zone_lock.
+ */
+static void
+zone_log_refcounts(zone_t *zone)
+{
+ char *buffer;
+ char *buffer_position;
+ uint32_t buffer_size;
+ uint32_t index;
+ uint_t ref;
+ uint_t cred_ref;
+
+ /*
+ * Construct a string representing the subsystem-specific reference
+ * counts. The counts are printed in ascending order by index into the
+ * zone_t::zone_subsys_ref array. The list will be surrounded by
+ * square brackets [] and will only contain nonzero reference counts.
+ *
+ * The buffer will hold two square bracket characters plus ten digits,
+ * one colon, one space, one comma, and some characters for a
+ * subsystem name per subsystem-specific reference count. (Unsigned 32-
+ * bit integers have at most ten decimal digits.) The last
+ * reference count's comma is replaced by the closing square
+ * bracket and a NULL character to terminate the string.
+ *
+ * NOTE: We have to grab the zone's zone_lock to create a consistent
+ * snapshot of the zone's reference counters.
+ *
+ * First, figure out how much space the string buffer will need.
+ * The buffer's size is stored in buffer_size.
+ */
+ buffer_size = 2; /* for the square brackets */
+ mutex_enter(&zone->zone_lock);
+ zone->zone_flags |= ZF_REFCOUNTS_LOGGED;
+ ref = zone->zone_ref;
+ cred_ref = zone->zone_cred_ref;
+ for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index)
+ if (zone->zone_subsys_ref[index] != 0)
+ buffer_size += strlen(zone_ref_subsys_names[index]) +
+ 13;
+ if (buffer_size == 2) {
+ /*
+ * No subsystems had nonzero reference counts. Don't bother
+ * with allocating a buffer; just log the general-purpose and
+ * credential reference counts.
+ */
+ mutex_exit(&zone->zone_lock);
+ (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
+ "Zone '%s' (ID: %d) is shutting down, but %u zone "
+ "references and %u credential references are still extant",
+ zone->zone_name, zone->zone_id, ref, cred_ref);
+ return;
+ }
+
+ /*
+ * buffer_size contains the exact number of characters that the
+ * buffer will need. Allocate the buffer and fill it with nonzero
+ * subsystem-specific reference counts. Surround the results with
+ * square brackets afterwards.
+ */
+ buffer = kmem_alloc(buffer_size, KM_SLEEP);
+ buffer_position = &buffer[1];
+ for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index) {
+ /*
+ * NOTE: The DDI's version of sprintf() returns a pointer to
+ * the modified buffer rather than the number of bytes written
+ * (as in snprintf(3C)). This is unfortunate and annoying.
+ * Therefore, we'll use snprintf() with INT_MAX to get the
+ * number of bytes written. Using INT_MAX is safe because
+ * the buffer is perfectly sized for the data: we'll never
+ * overrun the buffer.
+ */
+ if (zone->zone_subsys_ref[index] != 0)
+ buffer_position += snprintf(buffer_position, INT_MAX,
+ "%s: %u,", zone_ref_subsys_names[index],
+ zone->zone_subsys_ref[index]);
+ }
+ mutex_exit(&zone->zone_lock);
+ buffer[0] = '[';
+ ASSERT((uintptr_t)(buffer_position - buffer) < buffer_size);
+ ASSERT(buffer_position[0] == '\0' && buffer_position[-1] == ',');
+ buffer_position[-1] = ']';
+
+ /*
+ * Log the reference counts and free the message buffer.
+ */
+ (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
+ "Zone '%s' (ID: %d) is shutting down, but %u zone references and "
+ "%u credential references are still extant %s", zone->zone_name,
+ zone->zone_id, ref, cred_ref, buffer);
+ kmem_free(buffer, buffer_size);
+}
+
+/*
* Systemcall entry point to finalize the zone halt process. The caller
* must have already successfully called zone_shutdown().
*
@@ -4568,6 +4798,8 @@
uint64_t uniqid;
zone_t *zone;
zone_status_t status;
+ clock_t wait_time;
+ boolean_t log_refcounts;
if (secpolicy_zone_config(CRED()) != 0)
return (set_errno(EPERM));
@@ -4611,9 +4843,12 @@
zone_rele(zone);
zone = NULL; /* potentially free'd */
+ log_refcounts = B_FALSE;
+ wait_time = SEC_TO_TICK(ZONE_DESTROY_TIMEOUT_SECS);
mutex_enter(&zonehash_lock);
for (; /* ever */; ) {
boolean_t unref;
+ boolean_t refs_have_been_logged;
if ((zone = zone_find_all_by_id(zoneid)) == NULL ||
zone->zone_uniqid != uniqid) {
@@ -4626,6 +4861,8 @@
}
mutex_enter(&zone->zone_lock);
unref = ZONE_IS_UNREF(zone);
+ refs_have_been_logged = (zone->zone_flags &
+ ZF_REFCOUNTS_LOGGED);
mutex_exit(&zone->zone_lock);
if (unref) {
/*
@@ -4638,12 +4875,69 @@
break;
}
+ /*
+ * Wait for zone_rele_common() or zone_cred_rele() to signal
+ * zone_destroy_cv. zone_destroy_cv is signaled only when
+ * some zone's general-purpose reference count reaches one.
+ * If ZONE_DESTROY_TIMEOUT_SECS seconds elapse while waiting
+ * on zone_destroy_cv, then log the zone's reference counts and
+ * continue to wait for zone_rele() and zone_cred_rele().
+ */
+ if (!refs_have_been_logged) {
+ if (!log_refcounts) {
+ /*
+ * This thread hasn't timed out waiting on
+ * zone_destroy_cv yet. Wait wait_time clock
+ * ticks (initially ZONE_DESTROY_TIMEOUT_SECS
+ * seconds) for the zone's references to clear.
+ */
+ ASSERT(wait_time > 0);
+ wait_time = cv_reltimedwait_sig(
+ &zone_destroy_cv, &zonehash_lock, wait_time,
+ TR_SEC);
+ if (wait_time > 0) {
+ /*
+ * A thread in zone_rele() or
+ * zone_cred_rele() signaled
+ * zone_destroy_cv before this thread's
+ * wait timed out. The zone might have
+ * only one reference left; find out!
+ */
+ continue;
+ } else if (wait_time == 0) {
+ /* The thread's process was signaled. */
+ mutex_exit(&zonehash_lock);
+ return (set_errno(EINTR));
+ }
+
+ /*
+ * The thread timed out while waiting on
+ * zone_destroy_cv. Even though the thread
+ * timed out, it has to check whether another
+ * thread woke up from zone_destroy_cv and
+ * destroyed the zone.
+ *
+ * If the zone still exists and has more than
+ * one unreleased general-purpose reference,
+ * then log the zone's reference counts.
+ */
+ log_refcounts = B_TRUE;
+ continue;
+ }
+
+ /*
+ * The thread already timed out on zone_destroy_cv while
+ * waiting for subsystems to release the zone's last
+ * general-purpose references. Log the zone's reference
+ * counts and wait indefinitely on zone_destroy_cv.
+ */
+ zone_log_refcounts(zone);
+ }
if (cv_wait_sig(&zone_destroy_cv, &zonehash_lock) == 0) {
- /* Signaled */
+ /* The thread's process was signaled. */
mutex_exit(&zonehash_lock);
return (set_errno(EINTR));
}
-
}
/*
--- a/usr/src/uts/common/sys/fs/mntdata.h Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/sys/fs/mntdata.h Thu Aug 12 17:34:26 2010 -0700
@@ -19,8 +19,7 @@
* CDDL HEADER END
*/
/*
- * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
*/
#ifndef _SYS_MNTFS_MNTDATA_H
@@ -29,6 +28,7 @@
#include <sys/vnode.h>
#include <sys/poll.h>
#include <sys/mnttab.h>
+#include <sys/zone.h>
#ifdef __cplusplus
extern "C" {
@@ -70,18 +70,6 @@
mntsnap_t mnt_ioctl; /* data for ioctl() */
} mntnode_t;
-struct zone;
-
-typedef struct mntdata {
- struct zone *mnt_zone; /* zone for mount point */
- uint_t mnt_nopen; /* count of vnodes open */
- size_t mnt_size; /* size of last normal snapshot */
- size_t mnt_hidden_size; /* size of last hidden snapshot */
- timespec_t mnt_mtime; /* mtime at last normal snapshot */
- timespec_t mnt_hidden_mtime; /* mtime at last hidden snapshot */
- struct mntnode mnt_node; /* embedded mntnode */
-} mntdata_t;
-
/*
* Conversion macros.
*/
@@ -95,6 +83,16 @@
#if defined(_KERNEL)
+typedef struct mntdata {
+ zone_ref_t mnt_zone_ref; /* zone for mount point */
+ uint_t mnt_nopen; /* count of vnodes open */
+ size_t mnt_size; /* size of last normal snapshot */
+ size_t mnt_hidden_size; /* size of last hidden snapshot */
+ timespec_t mnt_mtime; /* mtime at last normal snapshot */
+ timespec_t mnt_hidden_mtime; /* mtime at last hidden snapshot */
+ struct mntnode mnt_node; /* embedded mntnode */
+} mntdata_t;
+
/*
* Value for a mntsnap_t's mnts_flags.
*/
--- a/usr/src/uts/common/sys/ipc_impl.h Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/sys/ipc_impl.h Thu Aug 12 17:34:26 2010 -0700
@@ -19,15 +19,12 @@
* CDDL HEADER END
*/
/*
- * Copyright 2006 Sun Microsystems, Inc. All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
*/
#ifndef _IPC_IMPL_H
#define _IPC_IMPL_H
-#pragma ident "%Z%%M% %I% %E% SMI"
-
#include <sys/types.h>
#include <sys/ipc.h>
#include <sys/mutex.h>
@@ -151,7 +148,7 @@
(*(rctl_qty_t *)(((char *)&p->ipc_proj->kpj_data.kpd_ipc) + \
s->ipcs_rctlofs))
#define IPC_ZONE_USAGE(p, s) \
- (*(rctl_qty_t *)(((char *)&p->ipc_zone->zone_ipc) + \
+ (*(rctl_qty_t *)(((char *)&p->ipc_zone_ref.zref_zone->zone_ipc) + \
s->ipcs_rctlofs))
#define IPC_LOCKED(s, o) \
MUTEX_HELD(&s->ipcs_table[IPC_INDEX(o->ipc_id)].ipct_lock)
@@ -172,7 +169,7 @@
kproject_t *ipc_proj; /* creator's project */
uint_t ipc_id; /* id */
zoneid_t ipc_zoneid; /* creator's zone id */
- zone_t *ipc_zone; /* creator's zone */
+ zone_ref_t ipc_zone_ref; /* reference to creator's zone */
} kipc_perm_t;
typedef struct ipc_slot {
--- a/usr/src/uts/common/sys/lofi.h Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/sys/lofi.h Thu Aug 12 17:34:26 2010 -0700
@@ -229,7 +229,7 @@
struct dk_geom ls_dkg;
struct vtoc ls_vtoc;
struct dk_cinfo ls_ci;
- zone_t *ls_zone;
+ zone_ref_t ls_zone;
list_node_t ls_list; /* all lofis */
dev_t ls_dev; /* this node's dev_t */
--- a/usr/src/uts/common/sys/vfs.h Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/sys/vfs.h Thu Aug 12 17:34:26 2010 -0700
@@ -33,6 +33,7 @@
#ifndef _SYS_VFS_H
#define _SYS_VFS_H
+#include <sys/zone.h>
#include <sys/types.h>
#include <sys/t_lock.h>
#include <sys/cred.h>
@@ -173,26 +174,6 @@
extern kmutex_t vskstat_tree_lock;
/*
- * Private vfs data, NOT to be used by a file system implementation.
- */
-
-#define VFS_FEATURE_MAXSZ 4
-
-typedef struct vfs_impl {
- /* Counted array - Bitmap of vfs features */
- uint32_t vi_featureset[VFS_FEATURE_MAXSZ];
- /*
- * Support for statistics on the vnode operations
- */
- vsk_anchor_t *vi_vskap; /* anchor for vopstats' kstat */
- vopstats_t *vi_fstypevsp; /* ptr to per-fstype vopstats */
- vopstats_t vi_vopstats; /* per-mount vnode op stats */
-
- timespec_t vi_hrctime; /* High-res creation time */
-} vfs_impl_t;
-
-
-/*
* Structure per mounted file system. Each mounted file system has
* an array of operations and an instance record.
*
@@ -241,7 +222,7 @@
refstr_t *vfs_resource; /* mounted resource name */
refstr_t *vfs_mntpt; /* mount point name */
time_t vfs_mtime; /* time we were mounted */
- vfs_impl_t *vfs_implp; /* impl specific data */
+ struct vfs_impl *vfs_implp; /* impl specific data */
/*
* Zones support. Note that the zone that "owns" the mount isn't
* necessarily the same as the zone in which the zone is visible.
@@ -448,6 +429,28 @@
#define VFSSP_VERBATIM 0x1 /* do not prefix the supplied path */
#if defined(_KERNEL)
+
+/*
+ * Private vfs data, NOT to be used by a file system implementation.
+ */
+
+#define VFS_FEATURE_MAXSZ 4
+
+typedef struct vfs_impl {
+ /* Counted array - Bitmap of vfs features */
+ uint32_t vi_featureset[VFS_FEATURE_MAXSZ];
+ /*
+ * Support for statistics on the vnode operations
+ */
+ vsk_anchor_t *vi_vskap; /* anchor for vopstats' kstat */
+ vopstats_t *vi_fstypevsp; /* ptr to per-fstype vopstats */
+ vopstats_t vi_vopstats; /* per-mount vnode op stats */
+
+ timespec_t vi_hrctime; /* High-res creation time */
+
+ zone_ref_t vi_zone_ref; /* reference to zone */
+} vfs_impl_t;
+
/*
* Public operations.
*/
--- a/usr/src/uts/common/sys/zone.h Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/sys/zone.h Thu Aug 12 17:34:26 2010 -0700
@@ -204,7 +204,12 @@
#define ZE_AREMOUNTS 2 /* there are mounts within the zone */
#define ZE_LABELINUSE 3 /* label is already in use by some other zone */
-/* zone_status */
+/*
+ * zone_status values
+ *
+ * You must modify zone_status_names in mdb(1M)'s genunix module
+ * (genunix/zone.c) when you modify this enum.
+ */
typedef enum {
ZONE_IS_UNINITIALIZED = 0,
ZONE_IS_INITIALIZED,
@@ -274,11 +279,22 @@
*/
#define ZONE_DOOR_PATH ZONES_TMPDIR "/%s.zoneadmd_door"
+
/* zone_flags */
+/*
+ * Threads that read or write the following flag must hold zone_lock.
+ */
+#define ZF_REFCOUNTS_LOGGED 0x1 /* a thread logged the zone's refs */
+
+/*
+ * The following threads are set when the zone is created and never changed.
+ * Threads that test for these flags don't have to hold zone_lock.
+ */
#define ZF_HASHED_LABEL 0x2 /* zone has a unique label */
#define ZF_IS_SCRATCH 0x4 /* scratch zone */
#define ZF_NET_EXCL 0x8 /* Zone has an exclusive IP stack */
+
/* zone_create flags */
#define ZCF_NET_EXCL 0x1 /* Create a zone with exclusive IP */
@@ -298,6 +314,7 @@
#ifdef _KERNEL
+
/*
* We need to protect the definition of 'list_t' from userland applications and
* libraries which may be defining ther own versions.
@@ -310,6 +327,38 @@
struct brand;
/*
+ * Each of these constants identifies a kernel subsystem that acquires and
+ * releases zone references. Each subsystem that invokes
+ * zone_hold_ref() and zone_rele_ref() should specify the
+ * zone_ref_subsys_t constant associated with the subsystem. Tracked holds
+ * help users and developers quickly identify subsystems that stall zone
+ * shutdowns indefinitely.
+ *
+ * NOTE: You must modify zone_ref_subsys_names in usr/src/uts/common/os/zone.c
+ * when you modify this enumeration.
+ */
+typedef enum zone_ref_subsys {
+ ZONE_REF_NFS, /* NFS */
+ ZONE_REF_NFSV4, /* NFSv4 */
+ ZONE_REF_SMBFS, /* SMBFS */
+ ZONE_REF_MNTFS, /* MNTFS */
+ ZONE_REF_LOFI, /* LOFI devices */
+ ZONE_REF_VFS, /* VFS infrastructure */
+ ZONE_REF_IPC, /* IPC infrastructure */
+ ZONE_REF_NUM_SUBSYS /* This must be the last entry. */
+} zone_ref_subsys_t;
+
+/*
+ * zone_ref represents a general-purpose references to a zone. Each zone's
+ * references are linked into the zone's zone_t::zone_ref_list. This allows
+ * debuggers to walk zones' references.
+ */
+typedef struct zone_ref {
+ struct zone *zref_zone; /* the zone to which the reference refers */
+ list_node_t zref_linkage; /* linkage for zone_t::zone_ref_list */
+} zone_ref_t;
+
+/*
* Structure to record list of ZFS datasets exported to a zone.
*/
typedef struct zone_dataset {
@@ -353,6 +402,8 @@
* zone_lock protects the following fields of a zone_t:
* zone_ref
* zone_cred_ref
+ * zone_subsys_ref
+ * zone_ref_list
* zone_ntasks
* zone_flags
* zone_zsd
@@ -368,6 +419,14 @@
uint_t zone_ref; /* count of zone_hold()s on zone */
uint_t zone_cred_ref; /* count of zone_hold_cred()s on zone */
/*
+ * Fixed-sized array of subsystem-specific reference counts
+ * The sum of all of the counts must be less than or equal to zone_ref.
+ * The array is indexed by the counts' subsystems' zone_ref_subsys_t
+ * constants.
+ */
+ uint_t zone_subsys_ref[ZONE_REF_NUM_SUBSYS];
+ list_t zone_ref_list; /* list of zone_ref_t structs */
+ /*
* zone_rootvp and zone_rootpath can never be modified once set.
*/
struct vnode *zone_rootvp; /* zone's root vnode */
@@ -494,6 +553,9 @@
extern void zone_init(void);
extern void zone_hold(zone_t *);
extern void zone_rele(zone_t *);
+extern void zone_init_ref(zone_ref_t *);
+extern void zone_hold_ref(zone_t *, zone_ref_t *, zone_ref_subsys_t);
+extern void zone_rele_ref(zone_ref_t *, zone_ref_subsys_t);
extern void zone_cred_hold(zone_t *);
extern void zone_cred_rele(zone_t *);
extern void zone_task_hold(zone_t *);
--- a/usr/src/uts/common/vm/seg_spt.c Thu Aug 12 16:05:23 2010 -0700
+++ b/usr/src/uts/common/vm/seg_spt.c Thu Aug 12 17:34:26 2010 -0700
@@ -466,7 +466,7 @@
* The zone will never be NULL, as a fully created
* shm always has an owning zone.
*/
- zone = sp->shm_perm.ipc_zone;
+ zone = sp->shm_perm.ipc_zone_ref.zref_zone;
ASSERT(zone != NULL);
if (anon_resv_zone(ptob(more_pgs), zone) == 0) {
err = ENOMEM;