1b9238976Sth /* 2b9238976Sth * CDDL HEADER START 3b9238976Sth * 4b9238976Sth * The contents of this file are subject to the terms of the 5b9238976Sth * Common Development and Distribution License (the "License"). 6b9238976Sth * You may not use this file except in compliance with the License. 7b9238976Sth * 8b9238976Sth * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9b9238976Sth * or http://www.opensolaris.org/os/licensing. 10b9238976Sth * See the License for the specific language governing permissions 11b9238976Sth * and limitations under the License. 12b9238976Sth * 13b9238976Sth * When distributing Covered Code, include this CDDL HEADER in each 14b9238976Sth * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15b9238976Sth * If applicable, add the following below this CDDL HEADER, with the 16b9238976Sth * fields enclosed by brackets "[]" replaced with your own identifying 17b9238976Sth * information: Portions Copyright [yyyy] [name of copyright owner] 18b9238976Sth * 19b9238976Sth * CDDL HEADER END 20b9238976Sth */ 21b9238976Sth 22b9238976Sth /* 23eabd0450Sth * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 24b9238976Sth * Use is subject to license terms. 25b9238976Sth */ 26b9238976Sth 27b9238976Sth /* 28b9238976Sth * Support for ephemeral mounts, e.g. mirror-mounts. These mounts are 29b9238976Sth * triggered from a "stub" rnode via a special set of vnodeops. 30b9238976Sth */ 31b9238976Sth 32b9238976Sth #include <sys/param.h> 33b9238976Sth #include <sys/types.h> 34b9238976Sth #include <sys/systm.h> 35b9238976Sth #include <sys/cred.h> 36b9238976Sth #include <sys/time.h> 37b9238976Sth #include <sys/vnode.h> 38b9238976Sth #include <sys/vfs.h> 39b9238976Sth #include <sys/vfs_opreg.h> 40b9238976Sth #include <sys/file.h> 41b9238976Sth #include <sys/filio.h> 42b9238976Sth #include <sys/uio.h> 43b9238976Sth #include <sys/buf.h> 44b9238976Sth #include <sys/mman.h> 45b9238976Sth #include <sys/pathname.h> 46b9238976Sth #include <sys/dirent.h> 47b9238976Sth #include <sys/debug.h> 48b9238976Sth #include <sys/vmsystm.h> 49b9238976Sth #include <sys/fcntl.h> 50b9238976Sth #include <sys/flock.h> 51b9238976Sth #include <sys/swap.h> 52b9238976Sth #include <sys/errno.h> 53b9238976Sth #include <sys/strsubr.h> 54b9238976Sth #include <sys/sysmacros.h> 55b9238976Sth #include <sys/kmem.h> 56b9238976Sth #include <sys/mount.h> 57b9238976Sth #include <sys/cmn_err.h> 58b9238976Sth #include <sys/pathconf.h> 59b9238976Sth #include <sys/utsname.h> 60b9238976Sth #include <sys/dnlc.h> 61b9238976Sth #include <sys/acl.h> 62b9238976Sth #include <sys/systeminfo.h> 63b9238976Sth #include <sys/policy.h> 64b9238976Sth #include <sys/sdt.h> 65b9238976Sth #include <sys/list.h> 66b9238976Sth #include <sys/stat.h> 67b9238976Sth #include <sys/mntent.h> 68b9238976Sth 69b9238976Sth #include <rpc/types.h> 70b9238976Sth #include <rpc/auth.h> 71b9238976Sth #include <rpc/clnt.h> 72b9238976Sth 73b9238976Sth #include <nfs/nfs.h> 74b9238976Sth #include <nfs/nfs_clnt.h> 75b9238976Sth #include <nfs/nfs_acl.h> 76b9238976Sth #include <nfs/lm.h> 77b9238976Sth #include <nfs/nfs4.h> 78b9238976Sth #include <nfs/nfs4_kprot.h> 79b9238976Sth #include <nfs/rnode4.h> 80b9238976Sth #include <nfs/nfs4_clnt.h> 81b9238976Sth 82b9238976Sth #include <vm/hat.h> 83b9238976Sth #include <vm/as.h> 84b9238976Sth #include <vm/page.h> 85b9238976Sth #include <vm/pvn.h> 86b9238976Sth #include <vm/seg.h> 87b9238976Sth #include <vm/seg_map.h> 88b9238976Sth #include <vm/seg_kpm.h> 89b9238976Sth #include <vm/seg_vn.h> 90b9238976Sth 91b9238976Sth #include <fs/fs_subr.h> 92b9238976Sth 93b9238976Sth #include <sys/ddi.h> 94b9238976Sth #include <sys/int_fmtio.h> 95b9238976Sth 96f39b8789Sth #include <sys/sunddi.h> 97b9238976Sth 98b9238976Sth /* 99b9238976Sth * The automatic unmounter thread stuff! 100b9238976Sth */ 101b9238976Sth static int nfs4_trigger_thread_timer = 20; /* in seconds */ 102b9238976Sth 103b9238976Sth /* 104b9238976Sth * Just a default.... 105b9238976Sth */ 106b9238976Sth static uint_t nfs4_trigger_mount_to = 240; 107b9238976Sth 108b9238976Sth typedef struct nfs4_trigger_globals { 109b9238976Sth kmutex_t ntg_forest_lock; 110b9238976Sth uint_t ntg_mount_to; 111b9238976Sth int ntg_thread_started; 112b9238976Sth nfs4_ephemeral_tree_t *ntg_forest; 113b9238976Sth } nfs4_trigger_globals_t; 114b9238976Sth 115b9238976Sth kmutex_t nfs4_ephemeral_thread_lock; 116b9238976Sth 117b9238976Sth zone_key_t nfs4_ephemeral_key = ZONE_KEY_UNINITIALIZED; 118b9238976Sth 119b9238976Sth static void nfs4_ephemeral_start_harvester(nfs4_trigger_globals_t *); 120b9238976Sth 121b9238976Sth /* 122b9238976Sth * Used for ephemeral mounts; contains data either duplicated from 123b9238976Sth * servinfo4_t, or hand-crafted, depending on type of ephemeral mount. 124b9238976Sth * 125b9238976Sth * It's intended that this structure is used solely for ephemeral 126b9238976Sth * mount-type specific data, for passing this data to 127b9238976Sth * nfs4_trigger_nargs_create(). 128b9238976Sth */ 129b9238976Sth typedef struct ephemeral_servinfo { 130b9238976Sth char *esi_hostname; 131b9238976Sth char *esi_netname; 132b9238976Sth char *esi_path; 133b9238976Sth int esi_path_len; 134b9238976Sth int esi_mount_flags; 135b9238976Sth struct netbuf *esi_addr; 136b9238976Sth struct netbuf *esi_syncaddr; 137b9238976Sth struct knetconfig *esi_knconf; 138b9238976Sth } ephemeral_servinfo_t; 139b9238976Sth 140b9238976Sth /* 141b9238976Sth * Collect together the mount-type specific and generic data args. 142b9238976Sth */ 143b9238976Sth typedef struct domount_args { 144b9238976Sth ephemeral_servinfo_t *dma_esi; 145b9238976Sth char *dma_hostlist; /* comma-sep. for RO failover */ 146b9238976Sth struct nfs_args *dma_nargs; 147b9238976Sth } domount_args_t; 148b9238976Sth 149b9238976Sth 150b9238976Sth /* 151b9238976Sth * The vnode ops functions for a trigger stub vnode 152b9238976Sth */ 153da6c28aaSamw static int nfs4_trigger_open(vnode_t **, int, cred_t *, caller_context_t *); 154da6c28aaSamw static int nfs4_trigger_getattr(vnode_t *, struct vattr *, int, cred_t *, 155da6c28aaSamw caller_context_t *); 156da6c28aaSamw static int nfs4_trigger_setattr(vnode_t *, struct vattr *, int, cred_t *, 157da6c28aaSamw caller_context_t *); 158da6c28aaSamw static int nfs4_trigger_access(vnode_t *, int, int, cred_t *, 159da6c28aaSamw caller_context_t *); 160da6c28aaSamw static int nfs4_trigger_readlink(vnode_t *, struct uio *, cred_t *, 161da6c28aaSamw caller_context_t *); 162da6c28aaSamw static int nfs4_trigger_lookup(vnode_t *, char *, vnode_t **, 163da6c28aaSamw struct pathname *, int, vnode_t *, cred_t *, caller_context_t *, 164da6c28aaSamw int *, pathname_t *); 165da6c28aaSamw static int nfs4_trigger_create(vnode_t *, char *, struct vattr *, 166da6c28aaSamw enum vcexcl, int, vnode_t **, cred_t *, int, caller_context_t *, 167da6c28aaSamw vsecattr_t *); 168da6c28aaSamw static int nfs4_trigger_remove(vnode_t *, char *, cred_t *, caller_context_t *, 169da6c28aaSamw int); 170da6c28aaSamw static int nfs4_trigger_link(vnode_t *, vnode_t *, char *, cred_t *, 171da6c28aaSamw caller_context_t *, int); 172da6c28aaSamw static int nfs4_trigger_rename(vnode_t *, char *, vnode_t *, char *, 173da6c28aaSamw cred_t *, caller_context_t *, int); 174da6c28aaSamw static int nfs4_trigger_mkdir(vnode_t *, char *, struct vattr *, 175da6c28aaSamw vnode_t **, cred_t *, caller_context_t *, int, vsecattr_t *vsecp); 176da6c28aaSamw static int nfs4_trigger_rmdir(vnode_t *, char *, vnode_t *, cred_t *, 177da6c28aaSamw caller_context_t *, int); 178da6c28aaSamw static int nfs4_trigger_symlink(vnode_t *, char *, struct vattr *, char *, 179da6c28aaSamw cred_t *, caller_context_t *, int); 180da6c28aaSamw static int nfs4_trigger_cmp(vnode_t *, vnode_t *, caller_context_t *); 181b9238976Sth 182b9238976Sth /* 183b9238976Sth * Regular NFSv4 vnodeops that we need to reference directly 184b9238976Sth */ 185da6c28aaSamw extern int nfs4_getattr(vnode_t *, struct vattr *, int, cred_t *, 186da6c28aaSamw caller_context_t *); 187da6c28aaSamw extern void nfs4_inactive(vnode_t *, cred_t *, caller_context_t *); 188b9238976Sth extern int nfs4_rwlock(vnode_t *, int, caller_context_t *); 189b9238976Sth extern void nfs4_rwunlock(vnode_t *, int, caller_context_t *); 190b9238976Sth extern int nfs4_lookup(vnode_t *, char *, vnode_t **, 191da6c28aaSamw struct pathname *, int, vnode_t *, cred_t *, 192da6c28aaSamw caller_context_t *, int *, pathname_t *); 193da6c28aaSamw extern int nfs4_pathconf(vnode_t *, int, ulong_t *, cred_t *, 194da6c28aaSamw caller_context_t *); 195da6c28aaSamw extern int nfs4_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *, 196da6c28aaSamw caller_context_t *); 197da6c28aaSamw extern int nfs4_fid(vnode_t *, fid_t *, caller_context_t *); 198da6c28aaSamw extern int nfs4_realvp(vnode_t *, vnode_t **, caller_context_t *); 199b9238976Sth 200b9238976Sth static int nfs4_trigger_mount(vnode_t *, vnode_t **); 201b9238976Sth static int nfs4_trigger_domount(vnode_t *, domount_args_t *, vfs_t **, 202b9238976Sth cred_t *); 203b9238976Sth static domount_args_t *nfs4_trigger_domount_args_create(vnode_t *); 204b9238976Sth static void nfs4_trigger_domount_args_destroy(domount_args_t *dma, 205b9238976Sth vnode_t *vp); 206b9238976Sth static ephemeral_servinfo_t *nfs4_trigger_esi_create(vnode_t *, servinfo4_t *); 207b9238976Sth static void nfs4_trigger_esi_destroy(ephemeral_servinfo_t *, vnode_t *); 208b9238976Sth static ephemeral_servinfo_t *nfs4_trigger_esi_create_mirrormount(vnode_t *, 209b9238976Sth servinfo4_t *); 210b9238976Sth static struct nfs_args *nfs4_trigger_nargs_create(mntinfo4_t *, servinfo4_t *, 211b9238976Sth ephemeral_servinfo_t *); 212b9238976Sth static void nfs4_trigger_nargs_destroy(struct nfs_args *); 213b9238976Sth static char *nfs4_trigger_create_mntopts(vfs_t *); 214b9238976Sth static void nfs4_trigger_destroy_mntopts(char *); 215b9238976Sth static int nfs4_trigger_add_mntopt(char *, char *, vfs_t *); 216b9238976Sth static enum clnt_stat nfs4_trigger_ping_server(servinfo4_t *, int); 217b9238976Sth 218b9238976Sth extern int umount2_engine(vfs_t *, int, cred_t *, int); 219b9238976Sth 220b9238976Sth 221b9238976Sth vnodeops_t *nfs4_trigger_vnodeops; 222b9238976Sth 223b9238976Sth /* 224b9238976Sth * These are the vnodeops that we must define for stub vnodes. 225b9238976Sth * 226b9238976Sth * 227b9238976Sth * Many of the VOPs defined for NFSv4 do not need to be defined here, 228b9238976Sth * for various reasons. This will result in the VFS default function being 229b9238976Sth * used: 230b9238976Sth * 231b9238976Sth * - These VOPs require a previous VOP_OPEN to have occurred. That will have 232b9238976Sth * lost the reference to the stub vnode, meaning these should not be called: 233b9238976Sth * close, read, write, ioctl, readdir, seek. 234b9238976Sth * 235b9238976Sth * - These VOPs are meaningless for vnodes without data pages. Since the 236b9238976Sth * stub vnode is of type VDIR, these should not be called: 237b9238976Sth * space, getpage, putpage, map, addmap, delmap, pageio, fsync. 238b9238976Sth * 239b9238976Sth * - These VOPs are otherwise not applicable, and should not be called: 240b9238976Sth * dump, setsecattr. 241b9238976Sth * 242b9238976Sth * 243b9238976Sth * These VOPs we do not want to define, but nor do we want the VFS default 244b9238976Sth * action. Instead, we specify the VFS error function, with fs_error(), but 245b9238976Sth * note that fs_error() is not actually called. Instead it results in the 246b9238976Sth * use of the error function defined for the particular VOP, in vn_ops_table[]: 247b9238976Sth * 248b9238976Sth * - frlock, dispose, shrlock. 249b9238976Sth * 250b9238976Sth * 251b9238976Sth * These VOPs we define to use the corresponding regular NFSv4 vnodeop. 252b9238976Sth * NOTE: if any of these ops involve an OTW call with the stub FH, then 253b9238976Sth * that call must be wrapped with save_mnt_secinfo()/check_mnt_secinfo() 254b9238976Sth * to protect the security data in the servinfo4_t for the "parent" 255b9238976Sth * filesystem that contains the stub. 256b9238976Sth * 257b9238976Sth * - These VOPs should not trigger a mount, so that "ls -l" does not: 258b9238976Sth * pathconf, getsecattr. 259b9238976Sth * 260b9238976Sth * - These VOPs would not make sense to trigger: 261b9238976Sth * inactive, rwlock, rwunlock, fid, realvp. 262b9238976Sth */ 263b9238976Sth const fs_operation_def_t nfs4_trigger_vnodeops_template[] = { 264b9238976Sth VOPNAME_OPEN, { .vop_open = nfs4_trigger_open }, 265b9238976Sth VOPNAME_GETATTR, { .vop_getattr = nfs4_trigger_getattr }, 266b9238976Sth VOPNAME_SETATTR, { .vop_setattr = nfs4_trigger_setattr }, 267b9238976Sth VOPNAME_ACCESS, { .vop_access = nfs4_trigger_access }, 268b9238976Sth VOPNAME_LOOKUP, { .vop_lookup = nfs4_trigger_lookup }, 269b9238976Sth VOPNAME_CREATE, { .vop_create = nfs4_trigger_create }, 270b9238976Sth VOPNAME_REMOVE, { .vop_remove = nfs4_trigger_remove }, 271b9238976Sth VOPNAME_LINK, { .vop_link = nfs4_trigger_link }, 272b9238976Sth VOPNAME_RENAME, { .vop_rename = nfs4_trigger_rename }, 273b9238976Sth VOPNAME_MKDIR, { .vop_mkdir = nfs4_trigger_mkdir }, 274b9238976Sth VOPNAME_RMDIR, { .vop_rmdir = nfs4_trigger_rmdir }, 275b9238976Sth VOPNAME_SYMLINK, { .vop_symlink = nfs4_trigger_symlink }, 276b9238976Sth VOPNAME_READLINK, { .vop_readlink = nfs4_trigger_readlink }, 277b9238976Sth VOPNAME_INACTIVE, { .vop_inactive = nfs4_inactive }, 278b9238976Sth VOPNAME_FID, { .vop_fid = nfs4_fid }, 279b9238976Sth VOPNAME_RWLOCK, { .vop_rwlock = nfs4_rwlock }, 280b9238976Sth VOPNAME_RWUNLOCK, { .vop_rwunlock = nfs4_rwunlock }, 281b9238976Sth VOPNAME_REALVP, { .vop_realvp = nfs4_realvp }, 282b9238976Sth VOPNAME_GETSECATTR, { .vop_getsecattr = nfs4_getsecattr }, 283b9238976Sth VOPNAME_PATHCONF, { .vop_pathconf = nfs4_pathconf }, 284b9238976Sth VOPNAME_FRLOCK, { .error = fs_error }, 285b9238976Sth VOPNAME_DISPOSE, { .error = fs_error }, 286b9238976Sth VOPNAME_SHRLOCK, { .error = fs_error }, 287b9238976Sth VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 288b9238976Sth NULL, NULL 289b9238976Sth }; 290b9238976Sth 291*d3a14591SThomas Haynes static void 292*d3a14591SThomas Haynes nfs4_ephemeral_tree_hold(nfs4_ephemeral_tree_t *net) 293*d3a14591SThomas Haynes { 294*d3a14591SThomas Haynes mutex_enter(&net->net_cnt_lock); 295*d3a14591SThomas Haynes net->net_refcnt++; 296*d3a14591SThomas Haynes ASSERT(net->net_refcnt != 0); 297*d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 298*d3a14591SThomas Haynes } 299*d3a14591SThomas Haynes 300*d3a14591SThomas Haynes /* 301*d3a14591SThomas Haynes * We need a safe way to decrement the refcnt whilst the 302*d3a14591SThomas Haynes * lock is being held. 303*d3a14591SThomas Haynes */ 304*d3a14591SThomas Haynes static void 305*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(nfs4_ephemeral_tree_t *net) 306*d3a14591SThomas Haynes { 307*d3a14591SThomas Haynes ASSERT(mutex_owned(&net->net_cnt_lock)); 308*d3a14591SThomas Haynes ASSERT(net->net_refcnt != 0); 309*d3a14591SThomas Haynes net->net_refcnt--; 310*d3a14591SThomas Haynes } 311*d3a14591SThomas Haynes 312*d3a14591SThomas Haynes static void 313*d3a14591SThomas Haynes nfs4_ephemeral_tree_rele(nfs4_ephemeral_tree_t *net) 314*d3a14591SThomas Haynes { 315*d3a14591SThomas Haynes mutex_enter(&net->net_cnt_lock); 316*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 317*d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 318*d3a14591SThomas Haynes } 319*d3a14591SThomas Haynes 320b9238976Sth /* 321b9238976Sth * Trigger ops for stub vnodes; for mirror mounts, etc. 322b9238976Sth * 323b9238976Sth * The general idea is that a "triggering" op will first call 324b9238976Sth * nfs4_trigger_mount(), which will find out whether a mount has already 325b9238976Sth * been triggered. 326b9238976Sth * 327b9238976Sth * If it has, then nfs4_trigger_mount() sets newvp to the root vnode 328b9238976Sth * of the covering vfs. 329b9238976Sth * 330b9238976Sth * If a mount has not yet been triggered, nfs4_trigger_mount() will do so, 331b9238976Sth * and again set newvp, as above. 332b9238976Sth * 333b9238976Sth * The triggering op may then re-issue the VOP by calling it on newvp. 334b9238976Sth * 335b9238976Sth * Note that some ops may perform custom action, and may or may not need 336b9238976Sth * to trigger a mount. 337b9238976Sth * 338b9238976Sth * Some ops need to call the regular NFSv4 vnodeop for a stub vnode. We 339b9238976Sth * obviously can't do this with VOP_<whatever>, since it's a stub vnode 340b9238976Sth * and that would just recurse. Instead, we call the v4 op directly, 341b9238976Sth * by name. This is OK, since we know that the vnode is for NFSv4, 342b9238976Sth * otherwise it couldn't be a stub. 343b9238976Sth * 344b9238976Sth */ 345b9238976Sth 346b9238976Sth static int 347da6c28aaSamw nfs4_trigger_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 348b9238976Sth { 349b9238976Sth int error; 350b9238976Sth vnode_t *newvp; 351b9238976Sth 352b9238976Sth error = nfs4_trigger_mount(*vpp, &newvp); 353b9238976Sth if (error) 354b9238976Sth return (error); 355b9238976Sth 356b9238976Sth /* Release the stub vnode, as we're losing the reference to it */ 357b9238976Sth VN_RELE(*vpp); 358b9238976Sth 359b9238976Sth /* Give the caller the root vnode of the newly-mounted fs */ 360b9238976Sth *vpp = newvp; 361b9238976Sth 362b9238976Sth /* return with VN_HELD(newvp) */ 363da6c28aaSamw return (VOP_OPEN(vpp, flag, cr, ct)); 364b9238976Sth } 365b9238976Sth 366b9238976Sth /* 367b9238976Sth * For the majority of cases, nfs4_trigger_getattr() will not trigger 368b9238976Sth * a mount. However, if ATTR_TRIGGER is set, we are being informed 369b9238976Sth * that we need to force the mount before we attempt to determine 370b9238976Sth * the attributes. The intent is an atomic operation for security 371b9238976Sth * testing. 372b9238976Sth */ 373b9238976Sth static int 374da6c28aaSamw nfs4_trigger_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 375da6c28aaSamw caller_context_t *ct) 376b9238976Sth { 377b9238976Sth int error; 378b9238976Sth 379b9238976Sth if (flags & ATTR_TRIGGER) { 380b9238976Sth vnode_t *newvp; 381b9238976Sth 382b9238976Sth error = nfs4_trigger_mount(vp, &newvp); 383b9238976Sth if (error) 384b9238976Sth return (error); 385b9238976Sth 386da6c28aaSamw error = VOP_GETATTR(newvp, vap, flags, cr, ct); 387b9238976Sth VN_RELE(newvp); 388b9238976Sth } else { 389da6c28aaSamw error = nfs4_getattr(vp, vap, flags, cr, ct); 390b9238976Sth } 391b9238976Sth 392b9238976Sth return (error); 393b9238976Sth } 394b9238976Sth 395b9238976Sth static int 396b9238976Sth nfs4_trigger_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 397b9238976Sth caller_context_t *ct) 398b9238976Sth { 399b9238976Sth int error; 400b9238976Sth vnode_t *newvp; 401b9238976Sth 402b9238976Sth error = nfs4_trigger_mount(vp, &newvp); 403b9238976Sth if (error) 404b9238976Sth return (error); 405b9238976Sth 406b9238976Sth error = VOP_SETATTR(newvp, vap, flags, cr, ct); 407b9238976Sth VN_RELE(newvp); 408b9238976Sth 409b9238976Sth return (error); 410b9238976Sth } 411b9238976Sth 412b9238976Sth static int 413da6c28aaSamw nfs4_trigger_access(vnode_t *vp, int mode, int flags, cred_t *cr, 414da6c28aaSamw caller_context_t *ct) 415b9238976Sth { 416b9238976Sth int error; 417b9238976Sth vnode_t *newvp; 418b9238976Sth 419b9238976Sth error = nfs4_trigger_mount(vp, &newvp); 420b9238976Sth if (error) 421b9238976Sth return (error); 422b9238976Sth 423da6c28aaSamw error = VOP_ACCESS(newvp, mode, flags, cr, ct); 424b9238976Sth VN_RELE(newvp); 425b9238976Sth 426b9238976Sth return (error); 427b9238976Sth } 428b9238976Sth 429b9238976Sth static int 430da6c28aaSamw nfs4_trigger_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, 431da6c28aaSamw struct pathname *pnp, int flags, vnode_t *rdir, cred_t *cr, 432da6c28aaSamw caller_context_t *ct, int *deflags, pathname_t *rpnp) 433b9238976Sth { 434b9238976Sth int error; 435b9238976Sth vnode_t *newdvp; 436b9238976Sth rnode4_t *drp = VTOR4(dvp); 437b9238976Sth 438b9238976Sth ASSERT(RP_ISSTUB(drp)); 439b9238976Sth 440b9238976Sth /* for now, we only support mirror-mounts */ 441b9238976Sth ASSERT(RP_ISSTUB_MIRRORMOUNT(drp)); 442b9238976Sth 443b9238976Sth /* 444b9238976Sth * It's not legal to lookup ".." for an fs root, so we mustn't pass 445b9238976Sth * that up. Instead, pass onto the regular op, regardless of whether 446b9238976Sth * we've triggered a mount. 447b9238976Sth */ 448b9238976Sth if (strcmp(nm, "..") == 0) 449da6c28aaSamw return (nfs4_lookup(dvp, nm, vpp, pnp, flags, rdir, cr, 450da6c28aaSamw ct, deflags, rpnp)); 451b9238976Sth 452b9238976Sth error = nfs4_trigger_mount(dvp, &newdvp); 453b9238976Sth if (error) 454b9238976Sth return (error); 455b9238976Sth 456da6c28aaSamw error = VOP_LOOKUP(newdvp, nm, vpp, pnp, flags, rdir, cr, ct, 457da6c28aaSamw deflags, rpnp); 458b9238976Sth VN_RELE(newdvp); 459b9238976Sth 460b9238976Sth return (error); 461b9238976Sth } 462b9238976Sth 463b9238976Sth static int 464b9238976Sth nfs4_trigger_create(vnode_t *dvp, char *nm, struct vattr *va, 465da6c28aaSamw enum vcexcl exclusive, int mode, vnode_t **vpp, cred_t *cr, 466da6c28aaSamw int flags, caller_context_t *ct, vsecattr_t *vsecp) 467b9238976Sth { 468b9238976Sth int error; 469b9238976Sth vnode_t *newdvp; 470b9238976Sth 471b9238976Sth error = nfs4_trigger_mount(dvp, &newdvp); 472b9238976Sth if (error) 473b9238976Sth return (error); 474b9238976Sth 475da6c28aaSamw error = VOP_CREATE(newdvp, nm, va, exclusive, mode, vpp, cr, 476da6c28aaSamw flags, ct, vsecp); 477b9238976Sth VN_RELE(newdvp); 478b9238976Sth 479b9238976Sth return (error); 480b9238976Sth } 481b9238976Sth 482b9238976Sth static int 483da6c28aaSamw nfs4_trigger_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, 484da6c28aaSamw int flags) 485b9238976Sth { 486b9238976Sth int error; 487b9238976Sth vnode_t *newdvp; 488b9238976Sth 489b9238976Sth error = nfs4_trigger_mount(dvp, &newdvp); 490b9238976Sth if (error) 491b9238976Sth return (error); 492b9238976Sth 493da6c28aaSamw error = VOP_REMOVE(newdvp, nm, cr, ct, flags); 494b9238976Sth VN_RELE(newdvp); 495b9238976Sth 496b9238976Sth return (error); 497b9238976Sth } 498b9238976Sth 499b9238976Sth static int 500da6c28aaSamw nfs4_trigger_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr, 501da6c28aaSamw caller_context_t *ct, int flags) 502b9238976Sth { 503b9238976Sth int error; 504b9238976Sth vnode_t *newtdvp; 505b9238976Sth 506b9238976Sth error = nfs4_trigger_mount(tdvp, &newtdvp); 507b9238976Sth if (error) 508b9238976Sth return (error); 509b9238976Sth 510b9238976Sth /* 511b9238976Sth * We don't check whether svp is a stub. Let the NFSv4 code 512b9238976Sth * detect that error, and return accordingly. 513b9238976Sth */ 514da6c28aaSamw error = VOP_LINK(newtdvp, svp, tnm, cr, ct, flags); 515b9238976Sth VN_RELE(newtdvp); 516b9238976Sth 517b9238976Sth return (error); 518b9238976Sth } 519b9238976Sth 520b9238976Sth static int 521b9238976Sth nfs4_trigger_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 522da6c28aaSamw cred_t *cr, caller_context_t *ct, int flags) 523b9238976Sth { 524b9238976Sth int error; 525b9238976Sth vnode_t *newsdvp; 526b9238976Sth rnode4_t *tdrp = VTOR4(tdvp); 527b9238976Sth 528b9238976Sth /* 529b9238976Sth * We know that sdvp is a stub, otherwise we would not be here. 530b9238976Sth * 531b9238976Sth * If tdvp is also be a stub, there are two possibilities: it 532b9238976Sth * is either the same stub as sdvp [i.e. VN_CMP(sdvp, tdvp)] 533b9238976Sth * or it is a different stub [!VN_CMP(sdvp, tdvp)]. 534b9238976Sth * 535b9238976Sth * In the former case, just trigger sdvp, and treat tdvp as 536b9238976Sth * though it were not a stub. 537b9238976Sth * 538b9238976Sth * In the latter case, it might be a different stub for the 539b9238976Sth * same server fs as sdvp, or for a different server fs. 540b9238976Sth * Regardless, from the client perspective this would still 541b9238976Sth * be a cross-filesystem rename, and should not be allowed, 542b9238976Sth * so return EXDEV, without triggering either mount. 543b9238976Sth */ 544b9238976Sth if (RP_ISSTUB(tdrp) && !VN_CMP(sdvp, tdvp)) 545b9238976Sth return (EXDEV); 546b9238976Sth 547b9238976Sth error = nfs4_trigger_mount(sdvp, &newsdvp); 548b9238976Sth if (error) 549b9238976Sth return (error); 550b9238976Sth 551da6c28aaSamw error = VOP_RENAME(newsdvp, snm, tdvp, tnm, cr, ct, flags); 552b9238976Sth 553b9238976Sth VN_RELE(newsdvp); 554b9238976Sth 555b9238976Sth return (error); 556b9238976Sth } 557b9238976Sth 558da6c28aaSamw /* ARGSUSED */ 559b9238976Sth static int 560b9238976Sth nfs4_trigger_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, 561da6c28aaSamw cred_t *cr, caller_context_t *ct, int flags, vsecattr_t *vsecp) 562b9238976Sth { 563b9238976Sth int error; 564b9238976Sth vnode_t *newdvp; 565b9238976Sth 566b9238976Sth error = nfs4_trigger_mount(dvp, &newdvp); 567b9238976Sth if (error) 568b9238976Sth return (error); 569b9238976Sth 570da6c28aaSamw error = VOP_MKDIR(newdvp, nm, va, vpp, cr, ct, flags, vsecp); 571b9238976Sth VN_RELE(newdvp); 572b9238976Sth 573b9238976Sth return (error); 574b9238976Sth } 575b9238976Sth 576b9238976Sth static int 577da6c28aaSamw nfs4_trigger_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr, 578da6c28aaSamw caller_context_t *ct, int flags) 579b9238976Sth { 580b9238976Sth int error; 581b9238976Sth vnode_t *newdvp; 582b9238976Sth 583b9238976Sth error = nfs4_trigger_mount(dvp, &newdvp); 584b9238976Sth if (error) 585b9238976Sth return (error); 586b9238976Sth 587da6c28aaSamw error = VOP_RMDIR(newdvp, nm, cdir, cr, ct, flags); 588b9238976Sth VN_RELE(newdvp); 589b9238976Sth 590b9238976Sth return (error); 591b9238976Sth } 592b9238976Sth 593b9238976Sth static int 594b9238976Sth nfs4_trigger_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, 595da6c28aaSamw cred_t *cr, caller_context_t *ct, int flags) 596b9238976Sth { 597b9238976Sth int error; 598b9238976Sth vnode_t *newdvp; 599b9238976Sth 600b9238976Sth error = nfs4_trigger_mount(dvp, &newdvp); 601b9238976Sth if (error) 602b9238976Sth return (error); 603b9238976Sth 604da6c28aaSamw error = VOP_SYMLINK(newdvp, lnm, tva, tnm, cr, ct, flags); 605b9238976Sth VN_RELE(newdvp); 606b9238976Sth 607b9238976Sth return (error); 608b9238976Sth } 609b9238976Sth 610b9238976Sth static int 611da6c28aaSamw nfs4_trigger_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, 612da6c28aaSamw caller_context_t *ct) 613b9238976Sth { 614b9238976Sth int error; 615b9238976Sth vnode_t *newvp; 616b9238976Sth 617b9238976Sth error = nfs4_trigger_mount(vp, &newvp); 618b9238976Sth if (error) 619b9238976Sth return (error); 620b9238976Sth 621da6c28aaSamw error = VOP_READLINK(newvp, uiop, cr, ct); 622b9238976Sth VN_RELE(newvp); 623b9238976Sth 624b9238976Sth return (error); 625b9238976Sth } 626b9238976Sth 627b9238976Sth /* end of trigger vnode ops */ 628b9238976Sth 629b9238976Sth 630b9238976Sth /* 631b9238976Sth * Mount upon a trigger vnode; for mirror-mounts, etc. 632b9238976Sth * 633b9238976Sth * The mount may have already occurred, via another thread. If not, 634b9238976Sth * assemble the location information - which may require fetching - and 635b9238976Sth * perform the mount. 636b9238976Sth * 637b9238976Sth * Sets newvp to be the root of the fs that is now covering vp. Note 638b9238976Sth * that we return with VN_HELD(*newvp). 639b9238976Sth * 640b9238976Sth * The caller is responsible for passing the VOP onto the covering fs. 641b9238976Sth */ 642b9238976Sth static int 643b9238976Sth nfs4_trigger_mount(vnode_t *vp, vnode_t **newvpp) 644b9238976Sth { 645b9238976Sth int error; 646b9238976Sth vfs_t *vfsp; 647b9238976Sth rnode4_t *rp = VTOR4(vp); 648b9238976Sth mntinfo4_t *mi = VTOMI4(vp); 649b9238976Sth domount_args_t *dma; 650b9238976Sth 651b9238976Sth nfs4_ephemeral_tree_t *net; 652b9238976Sth 653b9238976Sth bool_t must_unlock = FALSE; 654b9238976Sth bool_t is_building = FALSE; 655b9238976Sth 656b9238976Sth cred_t *zcred; 657b9238976Sth 658b9238976Sth nfs4_trigger_globals_t *ntg; 659b9238976Sth 660b9238976Sth zone_t *zone = curproc->p_zone; 661b9238976Sth 662b9238976Sth ASSERT(RP_ISSTUB(rp)); 663b9238976Sth 664b9238976Sth /* for now, we only support mirror-mounts */ 665b9238976Sth ASSERT(RP_ISSTUB_MIRRORMOUNT(rp)); 666b9238976Sth 667b9238976Sth *newvpp = NULL; 668b9238976Sth 669b9238976Sth /* 670b9238976Sth * Has the mount already occurred? 671b9238976Sth */ 672b9238976Sth error = vn_vfsrlock_wait(vp); 673b9238976Sth if (error) 674b9238976Sth goto done; 675b9238976Sth vfsp = vn_mountedvfs(vp); 676b9238976Sth if (vfsp != NULL) { 677b9238976Sth /* the mount has already occurred */ 678b9238976Sth error = VFS_ROOT(vfsp, newvpp); 679b9238976Sth if (!error) { 680b9238976Sth /* need to update the reference time */ 681b9238976Sth mutex_enter(&mi->mi_lock); 682b9238976Sth if (mi->mi_ephemeral) 683b9238976Sth mi->mi_ephemeral->ne_ref_time = 684b9238976Sth gethrestime_sec(); 685b9238976Sth mutex_exit(&mi->mi_lock); 686b9238976Sth } 687b9238976Sth 688b9238976Sth vn_vfsunlock(vp); 689b9238976Sth goto done; 690b9238976Sth } 691b9238976Sth vn_vfsunlock(vp); 692b9238976Sth 693b9238976Sth ntg = zone_getspecific(nfs4_ephemeral_key, zone); 694b9238976Sth ASSERT(ntg != NULL); 695b9238976Sth 696b9238976Sth mutex_enter(&mi->mi_lock); 697b9238976Sth 698b9238976Sth /* 699b9238976Sth * We need to lock down the ephemeral tree. 700b9238976Sth */ 701b9238976Sth if (mi->mi_ephemeral_tree == NULL) { 702b9238976Sth net = kmem_zalloc(sizeof (*net), KM_SLEEP); 703b9238976Sth mutex_init(&net->net_tree_lock, NULL, MUTEX_DEFAULT, NULL); 704b9238976Sth mutex_init(&net->net_cnt_lock, NULL, MUTEX_DEFAULT, NULL); 705b9238976Sth net->net_refcnt = 1; 706b9238976Sth net->net_status = NFS4_EPHEMERAL_TREE_BUILDING; 707b9238976Sth is_building = TRUE; 708b9238976Sth 709b9238976Sth /* 710b9238976Sth * We need to add it to the zone specific list for 711b9238976Sth * automatic unmounting and harvesting of deadwood. 712b9238976Sth */ 713b9238976Sth mutex_enter(&ntg->ntg_forest_lock); 714b9238976Sth if (ntg->ntg_forest != NULL) 715b9238976Sth net->net_next = ntg->ntg_forest; 716b9238976Sth ntg->ntg_forest = net; 717b9238976Sth mutex_exit(&ntg->ntg_forest_lock); 718b9238976Sth 719b9238976Sth /* 720b9238976Sth * No lock order confusion with mi_lock because no 721b9238976Sth * other node could have grabbed net_tree_lock. 722b9238976Sth */ 723b9238976Sth mutex_enter(&net->net_tree_lock); 724b9238976Sth mi->mi_ephemeral_tree = net; 725b9238976Sth net->net_mount = mi; 726b9238976Sth mutex_exit(&mi->mi_lock); 727b9238976Sth } else { 728b9238976Sth net = mi->mi_ephemeral_tree; 729b9238976Sth mutex_exit(&mi->mi_lock); 730b9238976Sth 731*d3a14591SThomas Haynes nfs4_ephemeral_tree_hold(net); 732*d3a14591SThomas Haynes 733*d3a14591SThomas Haynes mutex_enter(&net->net_tree_lock); 734b9238976Sth 735b9238976Sth /* 736*d3a14591SThomas Haynes * We can only procede if the tree is neither locked 737*d3a14591SThomas Haynes * nor being torn down. 738b9238976Sth */ 739*d3a14591SThomas Haynes mutex_enter(&net->net_cnt_lock); 740*d3a14591SThomas Haynes if (net->net_status & NFS4_EPHEMERAL_TREE_PROCESSING) { 741*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 742*d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 743*d3a14591SThomas Haynes mutex_exit(&net->net_tree_lock); 744*d3a14591SThomas Haynes 745*d3a14591SThomas Haynes return (EIO); 746*d3a14591SThomas Haynes } 747*d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 748b9238976Sth } 749b9238976Sth 750b9238976Sth mutex_enter(&net->net_cnt_lock); 751b9238976Sth net->net_status |= NFS4_EPHEMERAL_TREE_MOUNTING; 752b9238976Sth mutex_exit(&net->net_cnt_lock); 753b9238976Sth 754b9238976Sth must_unlock = TRUE; 755b9238976Sth 756b9238976Sth dma = nfs4_trigger_domount_args_create(vp); 757b9238976Sth if (dma == NULL) { 758b9238976Sth error = EINVAL; 759b9238976Sth goto done; 760b9238976Sth } 761b9238976Sth 762b9238976Sth /* 763b9238976Sth * Need to be root for this call to make mount work. 764b9238976Sth * Note that since we define mirror mounts to work 765b9238976Sth * for any user, we allow the mount to proceed. And 766b9238976Sth * we realize that the server will perform security 767b9238976Sth * checks to make sure that the client is allowed 768b9238976Sth * access. Finally, once the mount takes place, 769b9238976Sth * directory permissions will ensure that the 770b9238976Sth * content is secure. 771b9238976Sth */ 772b9238976Sth zcred = zone_get_kcred(getzoneid()); 773b9238976Sth ASSERT(zcred != NULL); 774b9238976Sth 775b9238976Sth error = nfs4_trigger_domount(vp, dma, &vfsp, zcred); 776b9238976Sth nfs4_trigger_domount_args_destroy(dma, vp); 777b9238976Sth 778b9238976Sth crfree(zcred); 779b9238976Sth 780b9238976Sth if (!error) 781b9238976Sth error = VFS_ROOT(vfsp, newvpp); 782b9238976Sth done: 783b9238976Sth if (must_unlock) { 784b9238976Sth mutex_enter(&net->net_cnt_lock); 785b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_MOUNTING; 786b9238976Sth if (is_building) 787b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_BUILDING; 788*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 789b9238976Sth mutex_exit(&net->net_cnt_lock); 790b9238976Sth 791b9238976Sth mutex_exit(&net->net_tree_lock); 792b9238976Sth } 793b9238976Sth 794b9238976Sth if (!error && (newvpp == NULL || *newvpp == NULL)) 795b9238976Sth error = ENOSYS; 796b9238976Sth 797b9238976Sth return (error); 798b9238976Sth } 799b9238976Sth 800b9238976Sth /* 801b9238976Sth * Collect together both the generic & mount-type specific args. 802b9238976Sth */ 803b9238976Sth static domount_args_t * 804b9238976Sth nfs4_trigger_domount_args_create(vnode_t *vp) 805b9238976Sth { 806b9238976Sth int nointr; 807b9238976Sth char *hostlist; 808b9238976Sth servinfo4_t *svp; 809b9238976Sth struct nfs_args *nargs, *nargs_head; 810b9238976Sth enum clnt_stat status; 811b9238976Sth ephemeral_servinfo_t *esi, *esi_first; 812b9238976Sth domount_args_t *dma; 813b9238976Sth mntinfo4_t *mi = VTOMI4(vp); 814b9238976Sth 815b9238976Sth nointr = !(mi->mi_flags & MI4_INT); 816b9238976Sth hostlist = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 817b9238976Sth 818b9238976Sth svp = mi->mi_curr_serv; 819b9238976Sth /* check if the current server is responding */ 820b9238976Sth status = nfs4_trigger_ping_server(svp, nointr); 821b9238976Sth if (status == RPC_SUCCESS) { 822b9238976Sth esi_first = nfs4_trigger_esi_create(vp, svp); 823b9238976Sth if (esi_first == NULL) { 824b9238976Sth kmem_free(hostlist, MAXPATHLEN); 825b9238976Sth return (NULL); 826b9238976Sth } 827b9238976Sth 828b9238976Sth (void) strlcpy(hostlist, esi_first->esi_hostname, MAXPATHLEN); 829b9238976Sth 830b9238976Sth nargs_head = nfs4_trigger_nargs_create(mi, svp, esi_first); 831b9238976Sth } else { 832b9238976Sth /* current server did not respond */ 833b9238976Sth esi_first = NULL; 834b9238976Sth nargs_head = NULL; 835b9238976Sth } 836b9238976Sth nargs = nargs_head; 837b9238976Sth 838b9238976Sth /* 839b9238976Sth * NFS RO failover. 840b9238976Sth * 841b9238976Sth * If we have multiple servinfo4 structures, linked via sv_next, 842b9238976Sth * we must create one nfs_args for each, linking the nfs_args via 843b9238976Sth * nfs_ext_u.nfs_extB.next. 844b9238976Sth * 845b9238976Sth * We need to build a corresponding esi for each, too, but that is 846b9238976Sth * used solely for building nfs_args, and may be immediately 847b9238976Sth * discarded, as domount() requires the info from just one esi, 848b9238976Sth * but all the nfs_args. 849b9238976Sth * 850b9238976Sth * Currently, the NFS mount code will hang if not all servers 851b9238976Sth * requested are available. To avoid that, we need to ping each 852b9238976Sth * server, here, and remove it from the list if it is not 853b9238976Sth * responding. This has the side-effect of that server then 854b9238976Sth * being permanently unavailable for this failover mount, even if 855b9238976Sth * it recovers. That's unfortunate, but the best we can do until 856b9238976Sth * the mount code path is fixed. 857b9238976Sth */ 858b9238976Sth 859b9238976Sth /* 860b9238976Sth * If the current server was down, loop indefinitely until we find 861b9238976Sth * at least one responsive server. 862b9238976Sth */ 863b9238976Sth do { 864b9238976Sth /* no locking needed for sv_next; it is only set at fs mount */ 865b9238976Sth for (svp = mi->mi_servers; svp != NULL; svp = svp->sv_next) { 866b9238976Sth struct nfs_args *next; 867b9238976Sth 868b9238976Sth /* 869b9238976Sth * nargs_head: the head of the nfs_args list 870b9238976Sth * nargs: the current tail of the list 871b9238976Sth * next: the newly-created element to be added 872b9238976Sth */ 873b9238976Sth 874b9238976Sth /* 875b9238976Sth * We've already tried the current server, above; 876b9238976Sth * if it was responding, we have already included it 877b9238976Sth * and it may now be ignored. 878b9238976Sth * 879b9238976Sth * Otherwise, try it again, since it may now have 880b9238976Sth * recovered. 881b9238976Sth */ 882b9238976Sth if (svp == mi->mi_curr_serv && esi_first != NULL) 883b9238976Sth continue; 884b9238976Sth 885b9238976Sth (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 886b9238976Sth if (svp->sv_flags & SV4_NOTINUSE) { 887b9238976Sth nfs_rw_exit(&svp->sv_lock); 888b9238976Sth continue; 889b9238976Sth } 890b9238976Sth nfs_rw_exit(&svp->sv_lock); 891b9238976Sth 892b9238976Sth /* check if the server is responding */ 893b9238976Sth status = nfs4_trigger_ping_server(svp, nointr); 894b9238976Sth /* if the server did not respond, ignore it */ 895b9238976Sth if (status != RPC_SUCCESS) 896b9238976Sth continue; 897b9238976Sth 898b9238976Sth esi = nfs4_trigger_esi_create(vp, svp); 899b9238976Sth if (esi == NULL) 900b9238976Sth continue; 901b9238976Sth 902b9238976Sth /* 903b9238976Sth * If the original current server (mi_curr_serv) 904b9238976Sth * was down when when we first tried it, 905b9238976Sth * (i.e. esi_first == NULL), 906b9238976Sth * we select this new server (svp) to be the server 907b9238976Sth * that we will actually contact (esi_first). 908b9238976Sth * 909b9238976Sth * Note that it's possible that mi_curr_serv == svp, 910b9238976Sth * if that mi_curr_serv was down but has now recovered. 911b9238976Sth */ 912b9238976Sth next = nfs4_trigger_nargs_create(mi, svp, esi); 913b9238976Sth if (esi_first == NULL) { 914b9238976Sth ASSERT(nargs == NULL); 915b9238976Sth ASSERT(nargs_head == NULL); 916b9238976Sth nargs_head = next; 917b9238976Sth esi_first = esi; 918b9238976Sth (void) strlcpy(hostlist, 919b9238976Sth esi_first->esi_hostname, MAXPATHLEN); 920b9238976Sth } else { 921b9238976Sth ASSERT(nargs_head != NULL); 922b9238976Sth nargs->nfs_ext_u.nfs_extB.next = next; 923b9238976Sth (void) strlcat(hostlist, ",", MAXPATHLEN); 924b9238976Sth (void) strlcat(hostlist, esi->esi_hostname, 925b9238976Sth MAXPATHLEN); 926b9238976Sth /* esi was only needed for hostname & nargs */ 927b9238976Sth nfs4_trigger_esi_destroy(esi, vp); 928b9238976Sth } 929b9238976Sth 930b9238976Sth nargs = next; 931b9238976Sth } 932b9238976Sth 933b9238976Sth /* if we've had no response at all, wait a second */ 934b9238976Sth if (esi_first == NULL) 935b9238976Sth delay(drv_usectohz(1000000)); 936b9238976Sth 937b9238976Sth } while (esi_first == NULL); 938b9238976Sth ASSERT(nargs_head != NULL); 939b9238976Sth 940b9238976Sth dma = kmem_zalloc(sizeof (domount_args_t), KM_SLEEP); 941b9238976Sth dma->dma_esi = esi_first; 942b9238976Sth dma->dma_hostlist = hostlist; 943b9238976Sth dma->dma_nargs = nargs_head; 944b9238976Sth 945b9238976Sth return (dma); 946b9238976Sth } 947b9238976Sth 948b9238976Sth static void 949b9238976Sth nfs4_trigger_domount_args_destroy(domount_args_t *dma, vnode_t *vp) 950b9238976Sth { 951b9238976Sth if (dma != NULL) { 952b9238976Sth if (dma->dma_esi != NULL && vp != NULL) 953b9238976Sth nfs4_trigger_esi_destroy(dma->dma_esi, vp); 954b9238976Sth 955b9238976Sth if (dma->dma_hostlist != NULL) 956b9238976Sth kmem_free(dma->dma_hostlist, MAXPATHLEN); 957b9238976Sth 958b9238976Sth if (dma->dma_nargs != NULL) { 959b9238976Sth struct nfs_args *nargs = dma->dma_nargs; 960b9238976Sth 961b9238976Sth do { 962b9238976Sth struct nfs_args *next = 963b9238976Sth nargs->nfs_ext_u.nfs_extB.next; 964b9238976Sth 965b9238976Sth nfs4_trigger_nargs_destroy(nargs); 966b9238976Sth nargs = next; 967b9238976Sth } while (nargs != NULL); 968b9238976Sth } 969b9238976Sth 970b9238976Sth kmem_free(dma, sizeof (domount_args_t)); 971b9238976Sth } 972b9238976Sth } 973b9238976Sth 974b9238976Sth /* 975b9238976Sth * The ephemeral_servinfo_t struct contains basic information we will need to 976b9238976Sth * perform the mount. Whilst the structure is generic across different 977b9238976Sth * types of ephemeral mount, the way we gather its contents differs. 978b9238976Sth */ 979b9238976Sth static ephemeral_servinfo_t * 980b9238976Sth nfs4_trigger_esi_create(vnode_t *vp, servinfo4_t *svp) 981b9238976Sth { 982b9238976Sth ephemeral_servinfo_t *esi; 983b9238976Sth rnode4_t *rp = VTOR4(vp); 984b9238976Sth 985b9238976Sth ASSERT(RP_ISSTUB(rp)); 986b9238976Sth 987b9238976Sth /* Call the ephemeral type-specific routine */ 988b9238976Sth if (RP_ISSTUB_MIRRORMOUNT(rp)) 989b9238976Sth esi = nfs4_trigger_esi_create_mirrormount(vp, svp); 990b9238976Sth else 991b9238976Sth esi = NULL; 992b9238976Sth 993b9238976Sth /* for now, we only support mirror-mounts */ 994b9238976Sth ASSERT(esi != NULL); 995b9238976Sth 996b9238976Sth return (esi); 997b9238976Sth } 998b9238976Sth 999b9238976Sth static void 1000b9238976Sth nfs4_trigger_esi_destroy(ephemeral_servinfo_t *esi, vnode_t *vp) 1001b9238976Sth { 1002b9238976Sth rnode4_t *rp = VTOR4(vp); 1003b9238976Sth 1004b9238976Sth ASSERT(RP_ISSTUB(rp)); 1005b9238976Sth 1006b9238976Sth /* for now, we only support mirror-mounts */ 1007b9238976Sth ASSERT(RP_ISSTUB_MIRRORMOUNT(rp)); 1008b9238976Sth 1009b9238976Sth /* Currently, no need for an ephemeral type-specific routine */ 1010b9238976Sth 1011b9238976Sth /* 1012b9238976Sth * The contents of ephemeral_servinfo_t goes into nfs_args, 1013b9238976Sth * and will be handled by nfs4_trigger_nargs_destroy(). 1014b9238976Sth * We need only free the structure itself. 1015b9238976Sth */ 1016b9238976Sth if (esi != NULL) 1017b9238976Sth kmem_free(esi, sizeof (ephemeral_servinfo_t)); 1018b9238976Sth } 1019b9238976Sth 1020b9238976Sth /* 1021b9238976Sth * Some of this may turn out to be common with other ephemeral types, 1022b9238976Sth * in which case it should be moved to nfs4_trigger_esi_create(), or a 1023b9238976Sth * common function called. 1024b9238976Sth */ 1025b9238976Sth static ephemeral_servinfo_t * 1026b9238976Sth nfs4_trigger_esi_create_mirrormount(vnode_t *vp, servinfo4_t *svp) 1027b9238976Sth { 1028b9238976Sth char *stubpath; 1029b9238976Sth struct knetconfig *sikncp, *svkncp; 1030b9238976Sth struct netbuf *bufp; 1031b9238976Sth ephemeral_servinfo_t *esi; 1032b9238976Sth 1033b9238976Sth esi = kmem_zalloc(sizeof (ephemeral_servinfo_t), KM_SLEEP); 1034b9238976Sth 1035b9238976Sth /* initially set to be our type of ephemeral mount; may be added to */ 1036b9238976Sth esi->esi_mount_flags = NFSMNT_MIRRORMOUNT; 1037b9238976Sth 1038b9238976Sth /* 1039b9238976Sth * We're copying info from the stub rnode's servinfo4, but 1040b9238976Sth * we must create new copies, not pointers, since this information 1041b9238976Sth * is to be associated with the new mount, which will be 1042b9238976Sth * unmounted (and its structures freed) separately 1043b9238976Sth */ 1044b9238976Sth 1045b9238976Sth /* 1046b9238976Sth * Sizes passed to kmem_[z]alloc here must match those freed 1047b9238976Sth * in nfs4_free_args() 1048b9238976Sth */ 1049b9238976Sth 1050b9238976Sth /* 1051b9238976Sth * We hold sv_lock across kmem_zalloc() calls that may sleep, but this 1052b9238976Sth * is difficult to avoid: as we need to read svp to calculate the 1053b9238976Sth * sizes to be allocated. 1054b9238976Sth */ 1055b9238976Sth (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1056b9238976Sth 1057b9238976Sth esi->esi_hostname = kmem_zalloc(strlen(svp->sv_hostname) + 1, KM_SLEEP); 1058b9238976Sth (void) strcat(esi->esi_hostname, svp->sv_hostname); 1059b9238976Sth 1060b9238976Sth esi->esi_addr = kmem_zalloc(sizeof (struct netbuf), KM_SLEEP); 1061b9238976Sth bufp = esi->esi_addr; 1062b9238976Sth bufp->len = svp->sv_addr.len; 1063b9238976Sth bufp->maxlen = svp->sv_addr.maxlen; 1064b9238976Sth bufp->buf = kmem_zalloc(bufp->len, KM_SLEEP); 1065b9238976Sth bcopy(svp->sv_addr.buf, bufp->buf, bufp->len); 1066b9238976Sth 1067b9238976Sth esi->esi_knconf = kmem_zalloc(sizeof (*esi->esi_knconf), KM_SLEEP); 1068b9238976Sth sikncp = esi->esi_knconf; 1069b9238976Sth svkncp = svp->sv_knconf; 1070b9238976Sth sikncp->knc_semantics = svkncp->knc_semantics; 1071b9238976Sth sikncp->knc_protofmly = (caddr_t)kmem_zalloc(KNC_STRSIZE, KM_SLEEP); 1072b9238976Sth (void) strcat((char *)sikncp->knc_protofmly, 1073b9238976Sth (char *)svkncp->knc_protofmly); 1074b9238976Sth sikncp->knc_proto = (caddr_t)kmem_zalloc(KNC_STRSIZE, KM_SLEEP); 1075b9238976Sth (void) strcat((char *)sikncp->knc_proto, (char *)svkncp->knc_proto); 1076b9238976Sth sikncp->knc_rdev = svkncp->knc_rdev; 1077b9238976Sth 1078b9238976Sth /* 1079b9238976Sth * Used when AUTH_DH is negotiated. 1080b9238976Sth * 1081b9238976Sth * This is ephemeral mount-type specific, since it contains the 1082b9238976Sth * server's time-sync syncaddr. 1083b9238976Sth */ 1084b9238976Sth if (svp->sv_dhsec) { 1085b9238976Sth struct netbuf *bufp; 1086b9238976Sth sec_data_t *sdata; 1087b9238976Sth dh_k4_clntdata_t *data; 1088b9238976Sth 1089b9238976Sth sdata = svp->sv_dhsec; 1090b9238976Sth data = (dh_k4_clntdata_t *)sdata->data; 1091b9238976Sth ASSERT(sdata->rpcflavor == AUTH_DH); 1092b9238976Sth 1093b9238976Sth bufp = kmem_zalloc(sizeof (struct netbuf), KM_SLEEP); 1094b9238976Sth bufp->len = data->syncaddr.len; 1095b9238976Sth bufp->maxlen = data->syncaddr.maxlen; 1096b9238976Sth bufp->buf = kmem_zalloc(bufp->len, KM_SLEEP); 1097b9238976Sth bcopy(data->syncaddr.buf, bufp->buf, bufp->len); 1098b9238976Sth esi->esi_syncaddr = bufp; 1099b9238976Sth 1100b9238976Sth if (data->netname != NULL) { 1101b9238976Sth int nmlen = data->netnamelen; 1102b9238976Sth 1103b9238976Sth /* 1104b9238976Sth * We need to copy from a dh_k4_clntdata_t 1105b9238976Sth * netname/netnamelen pair to a NUL-terminated 1106b9238976Sth * netname string suitable for putting in nfs_args, 1107b9238976Sth * where the latter has no netnamelen field. 1108b9238976Sth */ 1109b9238976Sth esi->esi_netname = kmem_zalloc(nmlen + 1, KM_SLEEP); 1110b9238976Sth bcopy(data->netname, esi->esi_netname, nmlen); 1111b9238976Sth } 1112b9238976Sth } else { 1113b9238976Sth esi->esi_syncaddr = NULL; 1114b9238976Sth esi->esi_netname = NULL; 1115b9238976Sth } 1116b9238976Sth 1117b9238976Sth stubpath = fn_path(VTOSV(vp)->sv_name); 1118b9238976Sth /* step over initial '.', to avoid e.g. sv_path: "/tank./ws" */ 1119b9238976Sth ASSERT(*stubpath == '.'); 1120b9238976Sth stubpath += 1; 1121b9238976Sth 1122b9238976Sth /* for nfs_args->fh */ 1123b9238976Sth esi->esi_path_len = strlen(svp->sv_path) + strlen(stubpath) + 1; 1124b9238976Sth esi->esi_path = kmem_zalloc(esi->esi_path_len, KM_SLEEP); 1125b9238976Sth (void) strcat(esi->esi_path, svp->sv_path); 1126b9238976Sth (void) strcat(esi->esi_path, stubpath); 1127b9238976Sth 1128b9238976Sth stubpath -= 1; 1129b9238976Sth /* stubpath allocated by fn_path() */ 1130b9238976Sth kmem_free(stubpath, strlen(stubpath) + 1); 1131b9238976Sth 1132b9238976Sth nfs_rw_exit(&svp->sv_lock); 1133b9238976Sth 1134b9238976Sth return (esi); 1135b9238976Sth } 1136b9238976Sth 1137b9238976Sth /* 1138b9238976Sth * Assemble the args, and call the generic VFS mount function to 1139b9238976Sth * finally perform the ephemeral mount. 1140b9238976Sth */ 1141b9238976Sth static int 1142b9238976Sth nfs4_trigger_domount(vnode_t *stubvp, domount_args_t *dma, vfs_t **vfsp, 1143b9238976Sth cred_t *cr) 1144b9238976Sth { 1145b9238976Sth struct mounta *uap; 1146b9238976Sth char *mntpt, *orig_path, *path; 1147b9238976Sth const char *orig_mntpt; 1148b9238976Sth int retval; 1149b9238976Sth int mntpt_len; 1150b9238976Sth int spec_len; 1151b9238976Sth zone_t *zone = curproc->p_zone; 1152b9238976Sth bool_t has_leading_slash; 1153b9238976Sth 1154b9238976Sth vfs_t *stubvfsp = stubvp->v_vfsp; 1155b9238976Sth ephemeral_servinfo_t *esi = dma->dma_esi; 1156b9238976Sth struct nfs_args *nargs = dma->dma_nargs; 1157b9238976Sth 1158b9238976Sth /* first, construct the mount point for the ephemeral mount */ 1159b9238976Sth orig_path = path = fn_path(VTOSV(stubvp)->sv_name); 1160b9238976Sth orig_mntpt = (char *)refstr_value(stubvfsp->vfs_mntpt); 1161b9238976Sth 1162b9238976Sth if (*orig_path == '.') 1163b9238976Sth orig_path++; 1164b9238976Sth 1165b9238976Sth /* 1166b9238976Sth * Get rid of zone's root path 1167b9238976Sth */ 1168b9238976Sth if (zone != global_zone) { 1169b9238976Sth /* 1170b9238976Sth * -1 for trailing '/' and -1 for EOS. 1171b9238976Sth */ 1172b9238976Sth if (strncmp(zone->zone_rootpath, orig_mntpt, 1173b9238976Sth zone->zone_rootpathlen - 1) == 0) { 1174b9238976Sth orig_mntpt += (zone->zone_rootpathlen - 2); 1175b9238976Sth } 1176b9238976Sth } 1177b9238976Sth 1178b9238976Sth mntpt_len = strlen(orig_mntpt) + strlen(orig_path); 1179b9238976Sth mntpt = kmem_zalloc(mntpt_len + 1, KM_SLEEP); 1180b9238976Sth (void) strcat(mntpt, orig_mntpt); 1181b9238976Sth (void) strcat(mntpt, orig_path); 1182b9238976Sth 1183b9238976Sth kmem_free(path, strlen(path) + 1); 1184b9238976Sth path = esi->esi_path; 1185b9238976Sth if (*path == '.') 1186b9238976Sth path++; 1187b9238976Sth if (path[0] == '/' && path[1] == '/') 1188b9238976Sth path++; 1189b9238976Sth has_leading_slash = (*path == '/'); 1190b9238976Sth 1191b9238976Sth spec_len = strlen(dma->dma_hostlist); 1192b9238976Sth spec_len += strlen(path); 1193b9238976Sth 1194b9238976Sth /* We are going to have to add this in */ 1195b9238976Sth if (!has_leading_slash) 1196b9238976Sth spec_len++; 1197b9238976Sth 1198b9238976Sth /* We need to get the ':' for dma_hostlist:esi_path */ 1199b9238976Sth spec_len++; 1200b9238976Sth 1201b9238976Sth uap = kmem_zalloc(sizeof (struct mounta), KM_SLEEP); 1202b9238976Sth uap->spec = kmem_zalloc(spec_len + 1, KM_SLEEP); 1203b9238976Sth (void) snprintf(uap->spec, spec_len + 1, "%s:%s%s", dma->dma_hostlist, 1204b9238976Sth has_leading_slash ? "" : "/", path); 1205b9238976Sth 1206b9238976Sth uap->dir = mntpt; 1207b9238976Sth 1208b9238976Sth uap->flags = MS_SYSSPACE | MS_DATA; 1209b9238976Sth /* fstype-independent mount options not covered elsewhere */ 1210b9238976Sth /* copy parent's mount(1M) "-m" flag */ 1211b9238976Sth if (stubvfsp->vfs_flag & VFS_NOMNTTAB) 1212b9238976Sth uap->flags |= MS_NOMNTTAB; 1213b9238976Sth 1214b9238976Sth uap->fstype = MNTTYPE_NFS4; 1215b9238976Sth uap->dataptr = (char *)nargs; 1216b9238976Sth /* not needed for MS_SYSSPACE */ 1217b9238976Sth uap->datalen = 0; 1218b9238976Sth 1219b9238976Sth /* use optptr to pass in extra mount options */ 1220b9238976Sth uap->flags |= MS_OPTIONSTR; 1221b9238976Sth uap->optptr = nfs4_trigger_create_mntopts(stubvfsp); 1222b9238976Sth if (uap->optptr == NULL) { 1223b9238976Sth retval = EINVAL; 1224b9238976Sth goto done; 1225b9238976Sth } 1226b9238976Sth /* domount() expects us to count the trailing NUL */ 1227b9238976Sth uap->optlen = strlen(uap->optptr) + 1; 1228b9238976Sth 1229b9238976Sth retval = domount(NULL, uap, stubvp, cr, vfsp); 1230b9238976Sth if (retval == 0) 1231b9238976Sth VFS_RELE(*vfsp); 1232b9238976Sth done: 1233b9238976Sth if (uap->optptr) 1234b9238976Sth nfs4_trigger_destroy_mntopts(uap->optptr); 1235b9238976Sth 1236b9238976Sth kmem_free(uap->spec, spec_len + 1); 1237b9238976Sth kmem_free(uap, sizeof (struct mounta)); 1238b9238976Sth kmem_free(mntpt, mntpt_len + 1); 1239b9238976Sth 1240b9238976Sth return (retval); 1241b9238976Sth } 1242b9238976Sth 1243b9238976Sth /* 1244b9238976Sth * Build an nfs_args structure for passing to domount(). 1245b9238976Sth * 1246b9238976Sth * Ephemeral mount-type specific data comes from the ephemeral_servinfo_t; 1247b9238976Sth * generic data - common to all ephemeral mount types - is read directly 1248b9238976Sth * from the parent mount's servinfo4_t and mntinfo4_t, via the stub vnode. 1249b9238976Sth */ 1250b9238976Sth static struct nfs_args * 1251b9238976Sth nfs4_trigger_nargs_create(mntinfo4_t *mi, servinfo4_t *svp, 1252b9238976Sth ephemeral_servinfo_t *esi) 1253b9238976Sth { 1254b9238976Sth sec_data_t *secdata; 1255b9238976Sth struct nfs_args *nargs; 1256b9238976Sth 1257b9238976Sth /* setup the nfs args */ 1258b9238976Sth nargs = kmem_zalloc(sizeof (struct nfs_args), KM_SLEEP); 1259b9238976Sth 1260b9238976Sth (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1261b9238976Sth 1262b9238976Sth nargs->addr = esi->esi_addr; 1263b9238976Sth 1264b9238976Sth /* for AUTH_DH by negotiation */ 1265b9238976Sth if (esi->esi_syncaddr || esi->esi_netname) { 1266b9238976Sth nargs->flags |= NFSMNT_SECURE; 1267b9238976Sth nargs->syncaddr = esi->esi_syncaddr; 1268b9238976Sth nargs->netname = esi->esi_netname; 1269b9238976Sth } 1270b9238976Sth 1271b9238976Sth nargs->flags |= NFSMNT_KNCONF; 1272b9238976Sth nargs->knconf = esi->esi_knconf; 1273b9238976Sth nargs->flags |= NFSMNT_HOSTNAME; 1274b9238976Sth nargs->hostname = esi->esi_hostname; 1275b9238976Sth nargs->fh = esi->esi_path; 1276b9238976Sth 1277b9238976Sth /* general mount settings, all copied from parent mount */ 1278b9238976Sth mutex_enter(&mi->mi_lock); 1279b9238976Sth 1280b9238976Sth if (!(mi->mi_flags & MI4_HARD)) 1281b9238976Sth nargs->flags |= NFSMNT_SOFT; 1282b9238976Sth 1283b9238976Sth nargs->flags |= NFSMNT_WSIZE | NFSMNT_RSIZE | NFSMNT_TIMEO | 1284b9238976Sth NFSMNT_RETRANS; 1285b9238976Sth nargs->wsize = mi->mi_stsize; 1286b9238976Sth nargs->rsize = mi->mi_tsize; 1287b9238976Sth nargs->timeo = mi->mi_timeo; 1288b9238976Sth nargs->retrans = mi->mi_retrans; 1289b9238976Sth 1290b9238976Sth if (mi->mi_flags & MI4_INT) 1291b9238976Sth nargs->flags |= NFSMNT_INT; 1292b9238976Sth if (mi->mi_flags & MI4_NOAC) 1293b9238976Sth nargs->flags |= NFSMNT_NOAC; 1294b9238976Sth 1295b9238976Sth nargs->flags |= NFSMNT_ACREGMIN | NFSMNT_ACREGMAX | NFSMNT_ACDIRMIN | 1296b9238976Sth NFSMNT_ACDIRMAX; 1297b9238976Sth nargs->acregmin = HR2SEC(mi->mi_acregmin); 1298b9238976Sth nargs->acregmax = HR2SEC(mi->mi_acregmax); 1299b9238976Sth nargs->acdirmin = HR2SEC(mi->mi_acdirmin); 1300b9238976Sth nargs->acdirmax = HR2SEC(mi->mi_acdirmax); 1301b9238976Sth 1302b9238976Sth if (mi->mi_flags & MI4_NOCTO) 1303b9238976Sth nargs->flags |= NFSMNT_NOCTO; 1304b9238976Sth if (mi->mi_flags & MI4_GRPID) 1305b9238976Sth nargs->flags |= NFSMNT_GRPID; 1306b9238976Sth if (mi->mi_flags & MI4_LLOCK) 1307b9238976Sth nargs->flags |= NFSMNT_LLOCK; 1308b9238976Sth if (mi->mi_flags & MI4_NOPRINT) 1309b9238976Sth nargs->flags |= NFSMNT_NOPRINT; 1310b9238976Sth if (mi->mi_flags & MI4_DIRECTIO) 1311b9238976Sth nargs->flags |= NFSMNT_DIRECTIO; 1312b9238976Sth if (mi->mi_flags & MI4_PUBLIC) 1313b9238976Sth nargs->flags |= NFSMNT_PUBLIC; 1314b9238976Sth 1315b9238976Sth mutex_exit(&mi->mi_lock); 1316b9238976Sth 1317b9238976Sth /* add any specific flags for this type of ephemeral mount */ 1318b9238976Sth nargs->flags |= esi->esi_mount_flags; 1319b9238976Sth 1320b9238976Sth /* 1321b9238976Sth * Security data & negotiation policy. 1322b9238976Sth * 1323b9238976Sth * We need to preserve the parent mount's preference for security 1324b9238976Sth * negotiation, translating SV4_TRYSECDEFAULT -> NFSMNT_SECDEFAULT. 1325b9238976Sth * 1326b9238976Sth * If SV4_TRYSECDEFAULT is not set, that indicates that a specific 1327b9238976Sth * security flavour was requested, with data in sv_secdata, and that 1328b9238976Sth * no negotiation should occur. If this specified flavour fails, that's 1329b9238976Sth * it. We will copy sv_secdata, and not set NFSMNT_SECDEFAULT. 1330b9238976Sth * 1331b9238976Sth * If SV4_TRYSECDEFAULT is set, then we start with a passed-in 1332b9238976Sth * default flavour, in sv_secdata, but then negotiate a new flavour. 1333b9238976Sth * Possible flavours are recorded in an array in sv_secinfo, with 1334b9238976Sth * currently in-use flavour pointed to by sv_currsec. 1335b9238976Sth * 1336b9238976Sth * If sv_currsec is set, i.e. if negotiation has already occurred, 1337b9238976Sth * we will copy sv_currsec. Otherwise, copy sv_secdata. Regardless, 1338b9238976Sth * we will set NFSMNT_SECDEFAULT, to enable negotiation. 1339b9238976Sth */ 1340b9238976Sth if (svp->sv_flags & SV4_TRYSECDEFAULT) { 1341b9238976Sth /* enable negotiation for ephemeral mount */ 1342b9238976Sth nargs->flags |= NFSMNT_SECDEFAULT; 1343b9238976Sth 1344b9238976Sth /* 1345b9238976Sth * As a starting point for negotiation, copy parent 1346b9238976Sth * mount's negotiated flavour (sv_currsec) if available, 1347b9238976Sth * or its passed-in flavour (sv_secdata) if not. 1348b9238976Sth */ 1349b9238976Sth if (svp->sv_currsec != NULL) 1350b9238976Sth secdata = copy_sec_data(svp->sv_currsec); 1351b9238976Sth else if (svp->sv_secdata != NULL) 1352b9238976Sth secdata = copy_sec_data(svp->sv_secdata); 1353b9238976Sth else 1354b9238976Sth secdata = NULL; 1355b9238976Sth } else { 1356b9238976Sth /* do not enable negotiation; copy parent's passed-in flavour */ 1357b9238976Sth if (svp->sv_secdata != NULL) 1358b9238976Sth secdata = copy_sec_data(svp->sv_secdata); 1359b9238976Sth else 1360b9238976Sth secdata = NULL; 1361b9238976Sth } 1362b9238976Sth 1363b9238976Sth nfs_rw_exit(&svp->sv_lock); 1364b9238976Sth 1365b9238976Sth nargs->flags |= NFSMNT_NEWARGS; 1366b9238976Sth nargs->nfs_args_ext = NFS_ARGS_EXTB; 1367b9238976Sth nargs->nfs_ext_u.nfs_extB.secdata = secdata; 1368b9238976Sth 1369b9238976Sth /* for NFS RO failover; caller will set if necessary */ 1370b9238976Sth nargs->nfs_ext_u.nfs_extB.next = NULL; 1371b9238976Sth 1372b9238976Sth return (nargs); 1373b9238976Sth } 1374b9238976Sth 1375b9238976Sth static void 1376b9238976Sth nfs4_trigger_nargs_destroy(struct nfs_args *nargs) 1377b9238976Sth { 1378b9238976Sth /* 1379b9238976Sth * Either the mount failed, in which case the data is not needed, or 1380b9238976Sth * nfs4_mount() has either taken copies of what it needs or, 1381b9238976Sth * where it has merely copied the ptr, it has set *our* ptr to NULL, 1382b9238976Sth * whereby nfs4_free_args() will ignore it. 1383b9238976Sth */ 1384b9238976Sth nfs4_free_args(nargs); 1385b9238976Sth kmem_free(nargs, sizeof (struct nfs_args)); 1386b9238976Sth } 1387b9238976Sth 1388b9238976Sth /* 1389b9238976Sth * When we finally get into the mounting, we need to add this 1390b9238976Sth * node to the ephemeral tree. 1391b9238976Sth * 1392b9238976Sth * This is called from nfs4_mount(). 1393b9238976Sth */ 1394*d3a14591SThomas Haynes int 1395b9238976Sth nfs4_record_ephemeral_mount(mntinfo4_t *mi, vnode_t *mvp) 1396b9238976Sth { 1397b9238976Sth mntinfo4_t *mi_parent; 1398b9238976Sth nfs4_ephemeral_t *eph; 1399b9238976Sth nfs4_ephemeral_tree_t *net; 1400b9238976Sth 1401b9238976Sth nfs4_ephemeral_t *prior; 1402b9238976Sth nfs4_ephemeral_t *child; 1403b9238976Sth 1404b9238976Sth nfs4_ephemeral_t *peer; 1405b9238976Sth 1406b9238976Sth nfs4_trigger_globals_t *ntg; 1407b9238976Sth zone_t *zone = curproc->p_zone; 1408b9238976Sth 1409*d3a14591SThomas Haynes int rc = 0; 1410*d3a14591SThomas Haynes 1411b9238976Sth mi_parent = VTOMI4(mvp); 1412b9238976Sth 1413b9238976Sth /* 1414b9238976Sth * Get this before grabbing anything else! 1415b9238976Sth */ 1416b9238976Sth ntg = zone_getspecific(nfs4_ephemeral_key, zone); 1417b9238976Sth if (!ntg->ntg_thread_started) { 1418b9238976Sth nfs4_ephemeral_start_harvester(ntg); 1419b9238976Sth } 1420b9238976Sth 1421b9238976Sth mutex_enter(&mi_parent->mi_lock); 1422b9238976Sth mutex_enter(&mi->mi_lock); 1423b9238976Sth 1424*d3a14591SThomas Haynes net = mi->mi_ephemeral_tree = 1425*d3a14591SThomas Haynes mi_parent->mi_ephemeral_tree; 1426*d3a14591SThomas Haynes 1427*d3a14591SThomas Haynes /* 1428*d3a14591SThomas Haynes * If the mi_ephemeral_tree is NULL, then it 1429*d3a14591SThomas Haynes * means that either the harvester or a manual 1430*d3a14591SThomas Haynes * umount has cleared the tree out right before 1431*d3a14591SThomas Haynes * we got here. 1432*d3a14591SThomas Haynes * 1433*d3a14591SThomas Haynes * There is nothing we can do here, so return 1434*d3a14591SThomas Haynes * to the caller and let them decide whether they 1435*d3a14591SThomas Haynes * try again. 1436*d3a14591SThomas Haynes */ 1437*d3a14591SThomas Haynes if (net == NULL) { 1438*d3a14591SThomas Haynes mutex_exit(&mi->mi_lock); 1439*d3a14591SThomas Haynes mutex_exit(&mi_parent->mi_lock); 1440*d3a14591SThomas Haynes 1441*d3a14591SThomas Haynes return (EBUSY); 1442*d3a14591SThomas Haynes } 1443*d3a14591SThomas Haynes 1444*d3a14591SThomas Haynes nfs4_ephemeral_tree_hold(net); 1445*d3a14591SThomas Haynes 1446b9238976Sth /* 1447b9238976Sth * We need to tack together the ephemeral mount 1448b9238976Sth * with this new mntinfo. 1449b9238976Sth */ 1450b9238976Sth eph = kmem_zalloc(sizeof (*eph), KM_SLEEP); 1451b9238976Sth eph->ne_mount = mi; 1452b9238976Sth eph->ne_ref_time = gethrestime_sec(); 1453b9238976Sth 1454b9238976Sth /* 1455b9238976Sth * We need to tell the ephemeral mount when 1456b9238976Sth * to time out. 1457b9238976Sth */ 1458b9238976Sth eph->ne_mount_to = ntg->ntg_mount_to; 1459b9238976Sth 1460b9238976Sth mi->mi_flags |= MI4_EPHEMERAL; 1461b9238976Sth mi->mi_ephemeral = eph; 1462b9238976Sth 1463b9238976Sth /* 1464b9238976Sth * If the enclosing mntinfo4 is also ephemeral, 1465b9238976Sth * then we need to point to its enclosing parent. 1466b9238976Sth * Else the enclosing mntinfo4 is the enclosing parent. 1467b9238976Sth * 1468b9238976Sth * We also need to weave this ephemeral node 1469b9238976Sth * into the tree. 1470b9238976Sth */ 1471b9238976Sth if (mi_parent->mi_flags & MI4_EPHEMERAL) { 1472b9238976Sth /* 1473b9238976Sth * We need to decide if we are 1474b9238976Sth * the root node of this branch 1475b9238976Sth * or if we are a sibling of this 1476b9238976Sth * branch. 1477b9238976Sth */ 1478b9238976Sth prior = mi_parent->mi_ephemeral; 1479*d3a14591SThomas Haynes if (prior == NULL) { 1480*d3a14591SThomas Haynes /* 1481*d3a14591SThomas Haynes * Race condition, clean up, and 1482*d3a14591SThomas Haynes * let caller handle mntinfo. 1483*d3a14591SThomas Haynes */ 1484*d3a14591SThomas Haynes mi->mi_flags &= ~MI4_EPHEMERAL; 1485*d3a14591SThomas Haynes mi->mi_ephemeral = NULL; 1486*d3a14591SThomas Haynes kmem_free(eph, sizeof (*eph)); 1487*d3a14591SThomas Haynes rc = EBUSY; 1488b9238976Sth } else { 1489*d3a14591SThomas Haynes if (prior->ne_child == NULL) { 1490*d3a14591SThomas Haynes prior->ne_child = eph; 1491*d3a14591SThomas Haynes } else { 1492*d3a14591SThomas Haynes child = prior->ne_child; 1493b9238976Sth 1494*d3a14591SThomas Haynes prior->ne_child = eph; 1495*d3a14591SThomas Haynes eph->ne_peer = child; 1496b9238976Sth 1497*d3a14591SThomas Haynes child->ne_prior = eph; 1498*d3a14591SThomas Haynes } 1499b9238976Sth 1500*d3a14591SThomas Haynes eph->ne_prior = prior; 1501*d3a14591SThomas Haynes } 1502b9238976Sth } else { 1503b9238976Sth /* 1504b9238976Sth * The parent mntinfo4 is the non-ephemeral 1505b9238976Sth * root of the ephemeral tree. We 1506b9238976Sth * need to decide if we are the root 1507b9238976Sth * node of that tree or if we are a 1508b9238976Sth * sibling of the root node. 1509b9238976Sth * 1510b9238976Sth * We are the root if there is no 1511b9238976Sth * other node. 1512b9238976Sth */ 1513b9238976Sth if (net->net_root == NULL) { 1514b9238976Sth net->net_root = eph; 1515b9238976Sth } else { 1516b9238976Sth eph->ne_peer = peer = net->net_root; 1517b9238976Sth ASSERT(peer != NULL); 1518b9238976Sth net->net_root = eph; 1519b9238976Sth 1520b9238976Sth peer->ne_prior = eph; 1521b9238976Sth } 1522b9238976Sth 1523b9238976Sth eph->ne_prior = NULL; 1524b9238976Sth } 1525b9238976Sth 1526*d3a14591SThomas Haynes nfs4_ephemeral_tree_rele(net); 1527*d3a14591SThomas Haynes 1528b9238976Sth mutex_exit(&mi->mi_lock); 1529b9238976Sth mutex_exit(&mi_parent->mi_lock); 1530*d3a14591SThomas Haynes 1531*d3a14591SThomas Haynes return (rc); 1532b9238976Sth } 1533b9238976Sth 1534b9238976Sth /* 1535b9238976Sth * Commit the changes to the ephemeral tree for removing this node. 1536b9238976Sth */ 1537b9238976Sth static void 1538b9238976Sth nfs4_ephemeral_umount_cleanup(nfs4_ephemeral_t *eph) 1539b9238976Sth { 1540b9238976Sth nfs4_ephemeral_t *e = eph; 1541b9238976Sth nfs4_ephemeral_t *peer; 1542b9238976Sth nfs4_ephemeral_t *prior; 1543b9238976Sth 1544b9238976Sth peer = eph->ne_peer; 1545b9238976Sth prior = e->ne_prior; 1546b9238976Sth 1547b9238976Sth /* 1548b9238976Sth * If this branch root was not the 1549b9238976Sth * tree root, then we need to fix back pointers. 1550b9238976Sth */ 1551b9238976Sth if (prior) { 1552b9238976Sth if (prior->ne_child == e) { 1553b9238976Sth prior->ne_child = peer; 1554b9238976Sth } else { 1555b9238976Sth prior->ne_peer = peer; 1556b9238976Sth } 1557b9238976Sth 1558b9238976Sth if (peer) 1559b9238976Sth peer->ne_prior = prior; 1560b9238976Sth } else if (peer) { 1561b9238976Sth peer->ne_mount->mi_ephemeral_tree->net_root = peer; 1562b9238976Sth peer->ne_prior = NULL; 1563b9238976Sth } else { 1564b9238976Sth e->ne_mount->mi_ephemeral_tree->net_root = NULL; 1565b9238976Sth } 1566b9238976Sth } 1567b9238976Sth 1568b9238976Sth /* 1569b9238976Sth * We want to avoid recursion at all costs. So we need to 1570b9238976Sth * unroll the tree. We do this by a depth first traversal to 1571b9238976Sth * leaf nodes. We blast away the leaf and work our way back 1572b9238976Sth * up and down the tree. 1573b9238976Sth */ 1574b9238976Sth static int 1575b9238976Sth nfs4_ephemeral_unmount_engine(nfs4_ephemeral_t *eph, 1576b9238976Sth int isTreeRoot, int flag, cred_t *cr) 1577b9238976Sth { 1578b9238976Sth nfs4_ephemeral_t *e = eph; 1579b9238976Sth nfs4_ephemeral_t *prior; 1580b9238976Sth mntinfo4_t *mi; 1581b9238976Sth vfs_t *vfsp; 1582b9238976Sth int error; 1583b9238976Sth 1584b9238976Sth /* 1585b9238976Sth * We use the loop while unrolling the ephemeral tree. 1586b9238976Sth */ 1587b9238976Sth for (;;) { 1588b9238976Sth /* 1589b9238976Sth * First we walk down the child. 1590b9238976Sth */ 1591b9238976Sth if (e->ne_child) { 1592b9238976Sth prior = e; 1593b9238976Sth e = e->ne_child; 1594b9238976Sth continue; 1595b9238976Sth } 1596b9238976Sth 1597b9238976Sth /* 1598b9238976Sth * If we are the root of the branch we are removing, 1599b9238976Sth * we end it here. But if the branch is the root of 1600b9238976Sth * the tree, we have to forge on. We do not consider 1601b9238976Sth * the peer list for the root because while it may 1602b9238976Sth * be okay to remove, it is both extra work and a 1603b9238976Sth * potential for a false-positive error to stall the 1604b9238976Sth * unmount attempt. 1605b9238976Sth */ 1606b9238976Sth if (e == eph && isTreeRoot == FALSE) 1607b9238976Sth return (0); 1608b9238976Sth 1609b9238976Sth /* 1610b9238976Sth * Next we walk down the peer list. 1611b9238976Sth */ 1612b9238976Sth if (e->ne_peer) { 1613b9238976Sth prior = e; 1614b9238976Sth e = e->ne_peer; 1615b9238976Sth continue; 1616b9238976Sth } 1617b9238976Sth 1618b9238976Sth /* 1619b9238976Sth * We can only remove the node passed in by the 1620b9238976Sth * caller if it is the root of the ephemeral tree. 1621b9238976Sth * Otherwise, the caller will remove it. 1622b9238976Sth */ 1623b9238976Sth if (e == eph && isTreeRoot == FALSE) 1624b9238976Sth return (0); 1625b9238976Sth 1626b9238976Sth /* 1627b9238976Sth * Okay, we have a leaf node, time 1628b9238976Sth * to prune it! 1629b9238976Sth * 1630b9238976Sth * Note that prior can only be NULL if 1631b9238976Sth * and only if it is the root of the 1632b9238976Sth * ephemeral tree. 1633b9238976Sth */ 1634b9238976Sth prior = e->ne_prior; 1635b9238976Sth 1636b9238976Sth mi = e->ne_mount; 1637b9238976Sth mutex_enter(&mi->mi_lock); 1638b9238976Sth vfsp = mi->mi_vfsp; 1639b9238976Sth 1640b9238976Sth /* 1641b9238976Sth * Cleared by umount2_engine. 1642b9238976Sth */ 1643b9238976Sth VFS_HOLD(vfsp); 1644b9238976Sth 1645b9238976Sth /* 1646b9238976Sth * Inform nfs4_unmount to not recursively 1647b9238976Sth * descend into this node's children when it 1648b9238976Sth * gets processed. 1649b9238976Sth */ 1650b9238976Sth mi->mi_flags |= MI4_EPHEMERAL_RECURSED; 1651b9238976Sth mutex_exit(&mi->mi_lock); 1652b9238976Sth 1653b9238976Sth error = umount2_engine(vfsp, flag, cr, FALSE); 1654b9238976Sth if (error) { 1655b9238976Sth /* 1656b9238976Sth * We need to reenable nfs4_unmount's ability 1657b9238976Sth * to recursively descend on this node. 1658b9238976Sth */ 1659b9238976Sth mutex_enter(&mi->mi_lock); 1660b9238976Sth mi->mi_flags &= ~MI4_EPHEMERAL_RECURSED; 1661b9238976Sth mutex_exit(&mi->mi_lock); 1662b9238976Sth 1663b9238976Sth return (error); 1664b9238976Sth } 1665b9238976Sth 1666b9238976Sth /* 1667b9238976Sth * If we are the current node, we do not want to 1668b9238976Sth * touch anything else. At this point, the only 1669b9238976Sth * way the current node can have survived to here 1670b9238976Sth * is if it is the root of the ephemeral tree and 1671b9238976Sth * we are unmounting the enclosing mntinfo4. 1672b9238976Sth */ 1673b9238976Sth if (e == eph) { 1674b9238976Sth ASSERT(prior == NULL); 1675b9238976Sth return (0); 1676b9238976Sth } 1677b9238976Sth 1678b9238976Sth /* 1679b9238976Sth * Stitch up the prior node. Note that since 1680b9238976Sth * we have handled the root of the tree, prior 1681b9238976Sth * must be non-NULL. 1682b9238976Sth */ 1683b9238976Sth ASSERT(prior != NULL); 1684b9238976Sth if (prior->ne_child == e) { 1685b9238976Sth prior->ne_child = NULL; 1686b9238976Sth } else { 1687b9238976Sth ASSERT(prior->ne_peer == e); 1688b9238976Sth 1689b9238976Sth prior->ne_peer = NULL; 1690b9238976Sth } 1691b9238976Sth 1692b9238976Sth e = prior; 1693b9238976Sth } 1694b9238976Sth 1695b9238976Sth /* NOTREACHED */ 1696b9238976Sth } 1697b9238976Sth 1698b9238976Sth /* 1699b9238976Sth * Common code to safely release net_cnt_lock and net_tree_lock 1700b9238976Sth */ 1701b9238976Sth void 1702b9238976Sth nfs4_ephemeral_umount_unlock(bool_t *pmust_unlock, 1703b9238976Sth nfs4_ephemeral_tree_t **pnet) 1704b9238976Sth { 1705b9238976Sth nfs4_ephemeral_tree_t *net = *pnet; 1706b9238976Sth 1707b9238976Sth if (*pmust_unlock) { 1708b9238976Sth mutex_enter(&net->net_cnt_lock); 1709b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_UMOUNTING; 1710*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 1711b9238976Sth mutex_exit(&net->net_cnt_lock); 1712b9238976Sth 1713b9238976Sth mutex_exit(&net->net_tree_lock); 1714b9238976Sth 1715b9238976Sth *pmust_unlock = FALSE; 1716b9238976Sth } 1717b9238976Sth } 1718b9238976Sth 1719b9238976Sth /* 1720b9238976Sth * While we may have removed any child or sibling nodes of this 1721b9238976Sth * ephemeral node, we can not nuke it until we know that there 1722b9238976Sth * were no actived vnodes on it. This will do that final 1723b9238976Sth * work once we know it is not busy. 1724b9238976Sth */ 1725b9238976Sth void 1726b9238976Sth nfs4_ephemeral_umount_activate(mntinfo4_t *mi, bool_t *pmust_unlock, 1727b9238976Sth nfs4_ephemeral_tree_t **pnet) 1728b9238976Sth { 1729b9238976Sth /* 1730b9238976Sth * Now we need to get rid of the ephemeral data if it exists. 1731b9238976Sth */ 1732b9238976Sth mutex_enter(&mi->mi_lock); 1733b9238976Sth if (mi->mi_ephemeral) { 1734b9238976Sth /* 1735b9238976Sth * If we are the root node of an ephemeral branch 1736b9238976Sth * which is being removed, then we need to fixup 1737b9238976Sth * pointers into and out of the node. 1738b9238976Sth */ 1739b9238976Sth if (!(mi->mi_flags & MI4_EPHEMERAL_RECURSED)) 1740b9238976Sth nfs4_ephemeral_umount_cleanup(mi->mi_ephemeral); 1741b9238976Sth 1742b9238976Sth ASSERT(mi->mi_ephemeral != NULL); 1743b9238976Sth 1744b9238976Sth kmem_free(mi->mi_ephemeral, sizeof (*mi->mi_ephemeral)); 1745b9238976Sth mi->mi_ephemeral = NULL; 1746b9238976Sth } 1747b9238976Sth mutex_exit(&mi->mi_lock); 1748b9238976Sth 1749b9238976Sth nfs4_ephemeral_umount_unlock(pmust_unlock, pnet); 1750b9238976Sth } 1751b9238976Sth 1752b9238976Sth /* 1753b9238976Sth * Unmount an ephemeral node. 1754b9238976Sth */ 1755b9238976Sth int 1756b9238976Sth nfs4_ephemeral_umount(mntinfo4_t *mi, int flag, cred_t *cr, 1757b9238976Sth bool_t *pmust_unlock, nfs4_ephemeral_tree_t **pnet) 1758b9238976Sth { 1759b9238976Sth int error = 0; 1760b9238976Sth nfs4_ephemeral_t *eph; 1761b9238976Sth nfs4_ephemeral_tree_t *net; 1762b9238976Sth int is_derooting = FALSE; 1763b9238976Sth int is_recursed = FALSE; 1764*d3a14591SThomas Haynes int was_locked = FALSE; 1765*d3a14591SThomas Haynes 1766*d3a14591SThomas Haynes /* 1767*d3a14591SThomas Haynes * Make sure to set the default state for cleaning 1768*d3a14591SThomas Haynes * up the tree in the caller (and on the way out). 1769*d3a14591SThomas Haynes */ 1770*d3a14591SThomas Haynes *pmust_unlock = FALSE; 1771b9238976Sth 1772b9238976Sth /* 1773b9238976Sth * The active vnodes on this file system may be ephemeral 1774b9238976Sth * children. We need to check for and try to unmount them 1775b9238976Sth * here. If any can not be unmounted, we are going 1776b9238976Sth * to return EBUSY. 1777b9238976Sth */ 1778b9238976Sth mutex_enter(&mi->mi_lock); 1779b9238976Sth 1780b9238976Sth /* 1781b9238976Sth * If an ephemeral tree, we need to check to see if 1782b9238976Sth * the lock is already held. If it is, then we need 1783b9238976Sth * to see if we are being called as a result of 1784b9238976Sth * the recursive removal of some node of the tree or 1785b9238976Sth * if we are another attempt to remove the tree. 1786b9238976Sth * 1787b9238976Sth * mi_flags & MI4_EPHEMERAL indicates an ephemeral 1788b9238976Sth * node. mi_ephemeral being non-NULL also does this. 1789b9238976Sth * 1790b9238976Sth * mi_ephemeral_tree being non-NULL is sufficient 1791b9238976Sth * to also indicate either it is an ephemeral node 1792b9238976Sth * or the enclosing mntinfo4. 1793b9238976Sth * 1794b9238976Sth * Do we need MI4_EPHEMERAL? Yes, it is useful for 1795b9238976Sth * when we delete the ephemeral node and need to 1796b9238976Sth * differentiate from an ephemeral node and the 1797b9238976Sth * enclosing root node. 1798b9238976Sth */ 1799b9238976Sth *pnet = net = mi->mi_ephemeral_tree; 1800eabd0450Sth if (net == NULL) { 1801b9238976Sth mutex_exit(&mi->mi_lock); 1802eabd0450Sth return (0); 1803eabd0450Sth } 1804b9238976Sth 1805eabd0450Sth eph = mi->mi_ephemeral; 1806eabd0450Sth is_recursed = mi->mi_flags & MI4_EPHEMERAL_RECURSED; 1807eabd0450Sth is_derooting = (eph == NULL); 1808b9238976Sth 1809eabd0450Sth /* 1810eabd0450Sth * If this is not recursion, then we need to 1811eabd0450Sth * grab a ref count. 1812eabd0450Sth * 1813eabd0450Sth * But wait, we also do not want to do that 1814eabd0450Sth * if a harvester thread has already grabbed 1815eabd0450Sth * the lock. 1816eabd0450Sth */ 1817eabd0450Sth if (!is_recursed) { 1818eabd0450Sth mutex_enter(&net->net_cnt_lock); 1819eabd0450Sth if (net->net_status & 1820eabd0450Sth NFS4_EPHEMERAL_TREE_LOCKED) { 1821b9238976Sth /* 1822*d3a14591SThomas Haynes * If the tree is locked, we need 1823*d3a14591SThomas Haynes * to decide whether we are the 1824*d3a14591SThomas Haynes * harvester or some explicit call 1825*d3a14591SThomas Haynes * for a umount. The only way that 1826*d3a14591SThomas Haynes * we are the harvester is if 1827*d3a14591SThomas Haynes * MS_SYSSPACE is set. 1828*d3a14591SThomas Haynes * 1829*d3a14591SThomas Haynes * We only let the harvester through 1830*d3a14591SThomas Haynes * at this point. 1831eabd0450Sth * 1832eabd0450Sth * We return EBUSY so that the 1833eabd0450Sth * caller knows something is 1834eabd0450Sth * going on. Note that by that 1835eabd0450Sth * time, the umount in the other 1836eabd0450Sth * thread may have already occured. 1837b9238976Sth */ 1838*d3a14591SThomas Haynes if (!(flag & MS_SYSSPACE)) { 1839*d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 1840*d3a14591SThomas Haynes mutex_exit(&mi->mi_lock); 1841*d3a14591SThomas Haynes 1842*d3a14591SThomas Haynes return (EBUSY); 1843*d3a14591SThomas Haynes } 1844*d3a14591SThomas Haynes 1845*d3a14591SThomas Haynes was_locked = TRUE; 1846*d3a14591SThomas Haynes } else { 1847eabd0450Sth net->net_refcnt++; 1848*d3a14591SThomas Haynes ASSERT(net->net_refcnt != 0); 1849*d3a14591SThomas Haynes } 1850*d3a14591SThomas Haynes 1851eabd0450Sth mutex_exit(&net->net_cnt_lock); 1852eabd0450Sth } 1853eabd0450Sth mutex_exit(&mi->mi_lock); 1854b9238976Sth 1855eabd0450Sth /* 1856*d3a14591SThomas Haynes * If we are not the harvester, we need to check 1857*d3a14591SThomas Haynes * to see if we need to grab the tree lock. 1858eabd0450Sth */ 1859*d3a14591SThomas Haynes if (was_locked == FALSE) { 1860*d3a14591SThomas Haynes /* 1861*d3a14591SThomas Haynes * If we grab the lock, it means that no other 1862*d3a14591SThomas Haynes * operation is working on the tree. If we don't 1863*d3a14591SThomas Haynes * grab it, we need to decide if this is because 1864*d3a14591SThomas Haynes * we are a recursive call or a new operation. 1865*d3a14591SThomas Haynes */ 1866*d3a14591SThomas Haynes if (mutex_tryenter(&net->net_tree_lock)) { 1867*d3a14591SThomas Haynes *pmust_unlock = TRUE; 1868*d3a14591SThomas Haynes } else { 1869b9238976Sth /* 1870*d3a14591SThomas Haynes * If we are a recursive call, we can 1871*d3a14591SThomas Haynes * proceed without the lock. 1872*d3a14591SThomas Haynes * Otherwise we have to wait until 1873*d3a14591SThomas Haynes * the lock becomes free. 1874b9238976Sth */ 1875*d3a14591SThomas Haynes if (!is_recursed) { 1876*d3a14591SThomas Haynes mutex_enter(&net->net_cnt_lock); 1877*d3a14591SThomas Haynes if (net->net_status & 1878*d3a14591SThomas Haynes (NFS4_EPHEMERAL_TREE_DEROOTING 1879*d3a14591SThomas Haynes | NFS4_EPHEMERAL_TREE_INVALID)) { 1880*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 1881*d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 1882*d3a14591SThomas Haynes goto is_busy; 1883*d3a14591SThomas Haynes } 1884*d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 1885b9238976Sth 1886*d3a14591SThomas Haynes /* 1887*d3a14591SThomas Haynes * We can't hold any other locks whilst 1888*d3a14591SThomas Haynes * we wait on this to free up. 1889*d3a14591SThomas Haynes */ 1890*d3a14591SThomas Haynes mutex_enter(&net->net_tree_lock); 1891b9238976Sth 1892*d3a14591SThomas Haynes /* 1893*d3a14591SThomas Haynes * Note that while mi->mi_ephemeral 1894*d3a14591SThomas Haynes * may change and thus we have to 1895*d3a14591SThomas Haynes * update eph, it is the case that 1896*d3a14591SThomas Haynes * we have tied down net and 1897*d3a14591SThomas Haynes * do not care if mi->mi_ephemeral_tree 1898*d3a14591SThomas Haynes * has changed. 1899*d3a14591SThomas Haynes */ 1900*d3a14591SThomas Haynes mutex_enter(&mi->mi_lock); 1901*d3a14591SThomas Haynes eph = mi->mi_ephemeral; 1902*d3a14591SThomas Haynes mutex_exit(&mi->mi_lock); 1903*d3a14591SThomas Haynes 1904*d3a14591SThomas Haynes /* 1905*d3a14591SThomas Haynes * Okay, we need to see if either the 1906*d3a14591SThomas Haynes * tree got nuked or the current node 1907*d3a14591SThomas Haynes * got nuked. Both of which will cause 1908*d3a14591SThomas Haynes * an error. 1909*d3a14591SThomas Haynes * 1910*d3a14591SThomas Haynes * Note that a subsequent retry of the 1911*d3a14591SThomas Haynes * umount shall work. 1912*d3a14591SThomas Haynes */ 1913*d3a14591SThomas Haynes mutex_enter(&net->net_cnt_lock); 1914*d3a14591SThomas Haynes if (net->net_status & 1915*d3a14591SThomas Haynes NFS4_EPHEMERAL_TREE_INVALID || 1916*d3a14591SThomas Haynes (!is_derooting && eph == NULL)) { 1917*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 1918*d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 1919*d3a14591SThomas Haynes mutex_exit(&net->net_tree_lock); 1920*d3a14591SThomas Haynes goto is_busy; 1921*d3a14591SThomas Haynes } 1922eabd0450Sth mutex_exit(&net->net_cnt_lock); 1923*d3a14591SThomas Haynes *pmust_unlock = TRUE; 1924eabd0450Sth } 1925eabd0450Sth } 1926eabd0450Sth } 1927eabd0450Sth 1928eabd0450Sth /* 1929eabd0450Sth * Only once we have grabbed the lock can we mark what we 1930eabd0450Sth * are planning on doing to the ephemeral tree. 1931eabd0450Sth */ 1932eabd0450Sth if (*pmust_unlock) { 1933eabd0450Sth mutex_enter(&net->net_cnt_lock); 1934eabd0450Sth net->net_status |= NFS4_EPHEMERAL_TREE_UMOUNTING; 1935eabd0450Sth 1936eabd0450Sth /* 1937eabd0450Sth * Check to see if we are nuking the root. 1938eabd0450Sth */ 1939eabd0450Sth if (is_derooting) 1940eabd0450Sth net->net_status |= 1941eabd0450Sth NFS4_EPHEMERAL_TREE_DEROOTING; 1942eabd0450Sth mutex_exit(&net->net_cnt_lock); 1943eabd0450Sth } 1944eabd0450Sth 1945eabd0450Sth if (!is_derooting) { 1946eabd0450Sth /* 1947eabd0450Sth * Only work on children if the caller has not already 1948eabd0450Sth * done so. 1949eabd0450Sth */ 1950eabd0450Sth if (!is_recursed) { 1951eabd0450Sth ASSERT(eph != NULL); 1952eabd0450Sth 1953eabd0450Sth error = nfs4_ephemeral_unmount_engine(eph, 1954eabd0450Sth FALSE, flag, cr); 1955eabd0450Sth if (error) 1956eabd0450Sth goto is_busy; 1957eabd0450Sth } 1958eabd0450Sth } else { 1959eabd0450Sth eph = net->net_root; 1960eabd0450Sth 1961eabd0450Sth /* 1962eabd0450Sth * Only work if there is something there. 1963eabd0450Sth */ 1964eabd0450Sth if (eph) { 1965eabd0450Sth error = nfs4_ephemeral_unmount_engine(eph, TRUE, 1966eabd0450Sth flag, cr); 1967eabd0450Sth if (error) { 1968eabd0450Sth mutex_enter(&net->net_cnt_lock); 1969eabd0450Sth net->net_status &= 1970eabd0450Sth ~NFS4_EPHEMERAL_TREE_DEROOTING; 1971eabd0450Sth mutex_exit(&net->net_cnt_lock); 1972eabd0450Sth goto is_busy; 1973eabd0450Sth } 1974b9238976Sth 1975b9238976Sth /* 1976eabd0450Sth * Nothing else which goes wrong will 1977eabd0450Sth * invalidate the blowing away of the 1978eabd0450Sth * ephmeral tree. 1979b9238976Sth */ 1980eabd0450Sth net->net_root = NULL; 1981b9238976Sth } 1982eabd0450Sth 1983eabd0450Sth /* 1984eabd0450Sth * We have derooted and we have caused the tree to be 1985*d3a14591SThomas Haynes * invalidated. 1986eabd0450Sth */ 1987eabd0450Sth mutex_enter(&net->net_cnt_lock); 1988eabd0450Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_DEROOTING; 1989eabd0450Sth net->net_status |= NFS4_EPHEMERAL_TREE_INVALID; 1990*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 1991eabd0450Sth mutex_exit(&net->net_cnt_lock); 1992eabd0450Sth 1993*d3a14591SThomas Haynes if (was_locked == FALSE) 1994*d3a14591SThomas Haynes mutex_exit(&net->net_tree_lock); 1995*d3a14591SThomas Haynes 1996*d3a14591SThomas Haynes /* 1997*d3a14591SThomas Haynes * We have just blown away any notation of this 1998*d3a14591SThomas Haynes * tree being locked. We can't let the caller 1999*d3a14591SThomas Haynes * try to clean things up. 2000*d3a14591SThomas Haynes */ 2001*d3a14591SThomas Haynes *pmust_unlock = FALSE; 2002*d3a14591SThomas Haynes 2003eabd0450Sth /* 2004eabd0450Sth * At this point, the tree should no 2005eabd0450Sth * longer be associated with the 2006eabd0450Sth * mntinfo4. We need to pull it off 2007eabd0450Sth * there and let the harvester take 2008eabd0450Sth * care of it once the refcnt drops. 2009eabd0450Sth */ 2010eabd0450Sth mutex_enter(&mi->mi_lock); 2011eabd0450Sth mi->mi_ephemeral_tree = NULL; 2012b9238976Sth mutex_exit(&mi->mi_lock); 2013b9238976Sth } 2014b9238976Sth 2015b9238976Sth return (0); 2016b9238976Sth 2017b9238976Sth is_busy: 2018b9238976Sth 2019b9238976Sth nfs4_ephemeral_umount_unlock(pmust_unlock, pnet); 2020b9238976Sth 2021b9238976Sth return (error); 2022b9238976Sth } 2023b9238976Sth 2024b9238976Sth /* 2025b9238976Sth * Do the umount and record any error in the parent. 2026b9238976Sth */ 2027b9238976Sth static void 2028b9238976Sth nfs4_ephemeral_record_umount(vfs_t *vfsp, int flag, 2029b9238976Sth nfs4_ephemeral_t *e, nfs4_ephemeral_t *prior) 2030b9238976Sth { 2031b9238976Sth int error; 2032b9238976Sth 2033b9238976Sth error = umount2_engine(vfsp, flag, kcred, FALSE); 2034b9238976Sth if (error) { 2035b9238976Sth if (prior) { 2036b9238976Sth if (prior->ne_child == e) 2037b9238976Sth prior->ne_state |= 2038b9238976Sth NFS4_EPHEMERAL_CHILD_ERROR; 2039b9238976Sth else 2040b9238976Sth prior->ne_state |= 2041b9238976Sth NFS4_EPHEMERAL_PEER_ERROR; 2042b9238976Sth } 2043b9238976Sth } 2044b9238976Sth } 2045b9238976Sth 2046b9238976Sth /* 2047b9238976Sth * For each tree in the forest (where the forest is in 2048b9238976Sth * effect all of the ephemeral trees for this zone), 2049b9238976Sth * scan to see if a node can be unmounted. Note that 2050b9238976Sth * unlike nfs4_ephemeral_unmount_engine(), we do 2051b9238976Sth * not process the current node before children or 2052b9238976Sth * siblings. I.e., if a node can be unmounted, we 2053b9238976Sth * do not recursively check to see if the nodes 2054b9238976Sth * hanging off of it can also be unmounted. 2055b9238976Sth * 2056b9238976Sth * Instead, we delve down deep to try and remove the 2057b9238976Sth * children first. Then, because we share code with 2058b9238976Sth * nfs4_ephemeral_unmount_engine(), we will try 2059b9238976Sth * them again. This could be a performance issue in 2060b9238976Sth * the future. 2061b9238976Sth * 2062b9238976Sth * Also note that unlike nfs4_ephemeral_unmount_engine(), 2063b9238976Sth * we do not halt on an error. We will not remove the 2064b9238976Sth * current node, but we will keep on trying to remove 2065b9238976Sth * the others. 2066b9238976Sth * 2067b9238976Sth * force indicates that we want the unmount to occur 2068b9238976Sth * even if there is something blocking it. 2069b9238976Sth * 2070b9238976Sth * time_check indicates that we want to see if the 2071b9238976Sth * mount has expired past mount_to or not. Typically 2072b9238976Sth * we want to do this and only on a shutdown of the 2073b9238976Sth * zone would we want to ignore the check. 2074b9238976Sth */ 2075b9238976Sth static void 2076b9238976Sth nfs4_ephemeral_harvest_forest(nfs4_trigger_globals_t *ntg, 2077b9238976Sth bool_t force, bool_t time_check) 2078b9238976Sth { 2079b9238976Sth nfs4_ephemeral_tree_t *net; 2080b9238976Sth nfs4_ephemeral_tree_t *prev = NULL; 2081b9238976Sth nfs4_ephemeral_tree_t *next; 2082b9238976Sth nfs4_ephemeral_t *e; 2083b9238976Sth nfs4_ephemeral_t *prior; 2084b9238976Sth time_t now = gethrestime_sec(); 2085b9238976Sth 2086b9238976Sth nfs4_ephemeral_tree_t *harvest = NULL; 2087b9238976Sth 2088b9238976Sth int flag; 2089b9238976Sth 2090b9238976Sth mntinfo4_t *mi; 2091b9238976Sth vfs_t *vfsp; 2092b9238976Sth 2093b9238976Sth if (force) 2094*d3a14591SThomas Haynes flag = MS_FORCE | MS_SYSSPACE; 2095b9238976Sth else 2096*d3a14591SThomas Haynes flag = MS_SYSSPACE; 2097b9238976Sth 2098b9238976Sth mutex_enter(&ntg->ntg_forest_lock); 2099b9238976Sth for (net = ntg->ntg_forest; net != NULL; net = next) { 2100b9238976Sth next = net->net_next; 2101b9238976Sth 2102*d3a14591SThomas Haynes nfs4_ephemeral_tree_hold(net); 2103b9238976Sth 2104b9238976Sth mutex_enter(&net->net_tree_lock); 2105b9238976Sth 2106b9238976Sth /* 2107b9238976Sth * Let the unmount code know that the 2108b9238976Sth * tree is already locked! 2109b9238976Sth */ 2110b9238976Sth mutex_enter(&net->net_cnt_lock); 2111b9238976Sth net->net_status |= NFS4_EPHEMERAL_TREE_LOCKED; 2112b9238976Sth mutex_exit(&net->net_cnt_lock); 2113b9238976Sth 2114b9238976Sth /* 2115b9238976Sth * If the intent is force all ephemeral nodes to 2116b9238976Sth * be unmounted in this zone, we can short circuit a 2117b9238976Sth * lot of tree traversal and simply zap the root node. 2118b9238976Sth */ 2119b9238976Sth if (force) { 2120b9238976Sth if (net->net_root) { 2121b9238976Sth mi = net->net_root->ne_mount; 2122b9238976Sth vfsp = mi->mi_vfsp; 2123b9238976Sth 2124b9238976Sth /* 2125b9238976Sth * Cleared by umount2_engine. 2126b9238976Sth */ 2127b9238976Sth VFS_HOLD(vfsp); 2128b9238976Sth 2129b9238976Sth (void) umount2_engine(vfsp, flag, 2130b9238976Sth kcred, FALSE); 2131b9238976Sth 2132b9238976Sth goto check_done; 2133b9238976Sth } 2134b9238976Sth } 2135b9238976Sth 2136b9238976Sth e = net->net_root; 2137b9238976Sth if (e) 2138b9238976Sth e->ne_state = NFS4_EPHEMERAL_VISIT_CHILD; 2139b9238976Sth 2140b9238976Sth while (e) { 2141b9238976Sth if (e->ne_state == NFS4_EPHEMERAL_VISIT_CHILD) { 2142b9238976Sth e->ne_state = NFS4_EPHEMERAL_VISIT_SIBLING; 2143b9238976Sth if (e->ne_child) { 2144b9238976Sth e = e->ne_child; 2145b9238976Sth e->ne_state = 2146b9238976Sth NFS4_EPHEMERAL_VISIT_CHILD; 2147b9238976Sth } 2148b9238976Sth 2149b9238976Sth continue; 2150b9238976Sth } else if (e->ne_state == 2151b9238976Sth NFS4_EPHEMERAL_VISIT_SIBLING) { 2152b9238976Sth e->ne_state = NFS4_EPHEMERAL_PROCESS_ME; 2153b9238976Sth if (e->ne_peer) { 2154b9238976Sth e = e->ne_peer; 2155b9238976Sth e->ne_state = 2156b9238976Sth NFS4_EPHEMERAL_VISIT_CHILD; 2157b9238976Sth } 2158b9238976Sth 2159b9238976Sth continue; 2160b9238976Sth } else if (e->ne_state == 2161b9238976Sth NFS4_EPHEMERAL_CHILD_ERROR) { 2162b9238976Sth prior = e->ne_prior; 2163b9238976Sth 2164b9238976Sth /* 2165b9238976Sth * If a child reported an error, do 2166b9238976Sth * not bother trying to unmount. 2167b9238976Sth * 2168b9238976Sth * If your prior node is a parent, 2169b9238976Sth * pass the error up such that they 2170b9238976Sth * also do not try to unmount. 2171b9238976Sth * 2172b9238976Sth * However, if your prior is a sibling, 2173b9238976Sth * let them try to unmount if they can. 2174b9238976Sth */ 2175b9238976Sth if (prior) { 2176b9238976Sth if (prior->ne_child == e) 2177b9238976Sth prior->ne_state |= 2178b9238976Sth NFS4_EPHEMERAL_CHILD_ERROR; 2179b9238976Sth else 2180b9238976Sth prior->ne_state |= 2181b9238976Sth NFS4_EPHEMERAL_PEER_ERROR; 2182b9238976Sth } 2183b9238976Sth 2184b9238976Sth /* 2185b9238976Sth * Clear the error and if needed, process peers. 2186b9238976Sth * 2187b9238976Sth * Once we mask out the error, we know whether 2188b9238976Sth * or we have to process another node. 2189b9238976Sth */ 2190b9238976Sth e->ne_state &= ~NFS4_EPHEMERAL_CHILD_ERROR; 2191b9238976Sth if (e->ne_state == NFS4_EPHEMERAL_PROCESS_ME) 2192b9238976Sth e = prior; 2193b9238976Sth 2194b9238976Sth continue; 2195b9238976Sth } else if (e->ne_state == 2196b9238976Sth NFS4_EPHEMERAL_PEER_ERROR) { 2197b9238976Sth prior = e->ne_prior; 2198b9238976Sth 2199b9238976Sth if (prior) { 2200b9238976Sth if (prior->ne_child == e) 2201b9238976Sth prior->ne_state = 2202b9238976Sth NFS4_EPHEMERAL_CHILD_ERROR; 2203b9238976Sth else 2204b9238976Sth prior->ne_state = 2205b9238976Sth NFS4_EPHEMERAL_PEER_ERROR; 2206b9238976Sth } 2207b9238976Sth 2208b9238976Sth /* 2209b9238976Sth * Clear the error from this node and do the 2210b9238976Sth * correct processing. 2211b9238976Sth */ 2212b9238976Sth e->ne_state &= ~NFS4_EPHEMERAL_PEER_ERROR; 2213b9238976Sth continue; 2214b9238976Sth } 2215b9238976Sth 2216b9238976Sth prior = e->ne_prior; 2217b9238976Sth e->ne_state = NFS4_EPHEMERAL_OK; 2218b9238976Sth 2219b9238976Sth /* 2220b9238976Sth * It must be the case that we need to process 2221b9238976Sth * this node. 2222b9238976Sth */ 2223b9238976Sth if (!time_check || 2224b9238976Sth now - e->ne_ref_time > e->ne_mount_to) { 2225b9238976Sth mi = e->ne_mount; 2226b9238976Sth vfsp = mi->mi_vfsp; 2227b9238976Sth 2228b9238976Sth /* 2229b9238976Sth * Cleared by umount2_engine. 2230b9238976Sth */ 2231b9238976Sth VFS_HOLD(vfsp); 2232b9238976Sth 2233b9238976Sth /* 2234b9238976Sth * Note that we effectively work down to the 2235b9238976Sth * leaf nodes first, try to unmount them, 2236b9238976Sth * then work our way back up into the leaf 2237b9238976Sth * nodes. 2238b9238976Sth * 2239b9238976Sth * Also note that we deal with a lot of 2240b9238976Sth * complexity by sharing the work with 2241b9238976Sth * the manual unmount code. 2242b9238976Sth */ 2243b9238976Sth nfs4_ephemeral_record_umount(vfsp, flag, 2244b9238976Sth e, prior); 2245b9238976Sth } 2246b9238976Sth 2247b9238976Sth e = prior; 2248b9238976Sth } 2249b9238976Sth 2250b9238976Sth check_done: 2251b9238976Sth 2252b9238976Sth /* 2253*d3a14591SThomas Haynes * At this point we are done processing this tree. 2254*d3a14591SThomas Haynes * 2255*d3a14591SThomas Haynes * If the tree is invalid and we are the only reference 2256*d3a14591SThomas Haynes * to it, then we push it on the local linked list 2257*d3a14591SThomas Haynes * to remove it at the end. We avoid that action now 2258*d3a14591SThomas Haynes * to keep the tree processing going along at a fair clip. 2259*d3a14591SThomas Haynes * 2260*d3a14591SThomas Haynes * Else, even if we are the only reference, we drop 2261*d3a14591SThomas Haynes * our hold on the current tree and allow it to be 2262*d3a14591SThomas Haynes * reused as needed. 2263b9238976Sth */ 2264b9238976Sth mutex_enter(&net->net_cnt_lock); 2265b9238976Sth if (net->net_refcnt == 1 && 2266b9238976Sth net->net_status & NFS4_EPHEMERAL_TREE_INVALID) { 2267*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 2268b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_LOCKED; 2269b9238976Sth mutex_exit(&net->net_cnt_lock); 2270b9238976Sth mutex_exit(&net->net_tree_lock); 2271b9238976Sth 2272b9238976Sth if (prev) 2273b9238976Sth prev->net_next = net->net_next; 2274b9238976Sth else 2275b9238976Sth ntg->ntg_forest = net->net_next; 2276b9238976Sth 2277b9238976Sth net->net_next = harvest; 2278b9238976Sth harvest = net; 2279b9238976Sth continue; 2280b9238976Sth } 2281b9238976Sth 2282*d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 2283b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_LOCKED; 2284b9238976Sth mutex_exit(&net->net_cnt_lock); 2285b9238976Sth mutex_exit(&net->net_tree_lock); 2286b9238976Sth 2287b9238976Sth prev = net; 2288b9238976Sth } 2289b9238976Sth mutex_exit(&ntg->ntg_forest_lock); 2290b9238976Sth 2291b9238976Sth for (net = harvest; net != NULL; net = next) { 2292b9238976Sth next = net->net_next; 2293b9238976Sth 2294b9238976Sth mutex_destroy(&net->net_tree_lock); 2295b9238976Sth mutex_destroy(&net->net_cnt_lock); 2296b9238976Sth kmem_free(net, sizeof (*net)); 2297b9238976Sth } 2298b9238976Sth } 2299b9238976Sth 2300b9238976Sth /* 2301b9238976Sth * This is the thread which decides when the harvesting 2302b9238976Sth * can proceed and when to kill it off for this zone. 2303b9238976Sth */ 2304b9238976Sth static void 2305b9238976Sth nfs4_ephemeral_harvester(nfs4_trigger_globals_t *ntg) 2306b9238976Sth { 2307b9238976Sth clock_t timeleft; 2308b9238976Sth zone_t *zone = curproc->p_zone; 2309b9238976Sth 2310b9238976Sth for (;;) { 2311b9238976Sth timeleft = zone_status_timedwait(zone, lbolt + 2312b9238976Sth nfs4_trigger_thread_timer * hz, ZONE_IS_SHUTTING_DOWN); 2313b9238976Sth 2314b9238976Sth /* 2315b9238976Sth * zone is exiting... 2316b9238976Sth */ 2317b9238976Sth if (timeleft != -1) { 2318b9238976Sth ASSERT(zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN); 2319b9238976Sth zthread_exit(); 2320b9238976Sth /* NOTREACHED */ 2321b9238976Sth } 2322b9238976Sth 2323b9238976Sth /* 2324b9238976Sth * Only bother scanning if there is potential 2325b9238976Sth * work to be done. 2326b9238976Sth */ 2327b9238976Sth if (ntg->ntg_forest == NULL) 2328b9238976Sth continue; 2329b9238976Sth 2330b9238976Sth /* 2331b9238976Sth * Now scan the list and get rid of everything which 2332b9238976Sth * is old. 2333b9238976Sth */ 2334b9238976Sth nfs4_ephemeral_harvest_forest(ntg, FALSE, TRUE); 2335b9238976Sth } 2336b9238976Sth 2337b9238976Sth /* NOTREACHED */ 2338b9238976Sth } 2339b9238976Sth 2340b9238976Sth /* 2341b9238976Sth * The zone specific glue needed to start the unmount harvester. 2342b9238976Sth * 2343b9238976Sth * Note that we want to avoid holding the mutex as long as possible, 2344b9238976Sth * hence the multiple checks. 2345b9238976Sth * 2346b9238976Sth * The caller should avoid us getting down here in the first 2347b9238976Sth * place. 2348b9238976Sth */ 2349b9238976Sth static void 2350b9238976Sth nfs4_ephemeral_start_harvester(nfs4_trigger_globals_t *ntg) 2351b9238976Sth { 2352b9238976Sth /* 2353b9238976Sth * It got started before we got here... 2354b9238976Sth */ 2355b9238976Sth if (ntg->ntg_thread_started) 2356b9238976Sth return; 2357b9238976Sth 2358b9238976Sth mutex_enter(&nfs4_ephemeral_thread_lock); 2359b9238976Sth 2360b9238976Sth if (ntg->ntg_thread_started) { 2361b9238976Sth mutex_exit(&nfs4_ephemeral_thread_lock); 2362b9238976Sth return; 2363b9238976Sth } 2364b9238976Sth 2365b9238976Sth /* 2366b9238976Sth * Start the unmounter harvester thread for this zone. 2367b9238976Sth */ 2368b9238976Sth (void) zthread_create(NULL, 0, nfs4_ephemeral_harvester, 2369b9238976Sth ntg, 0, minclsyspri); 2370b9238976Sth 2371b9238976Sth ntg->ntg_thread_started = TRUE; 2372b9238976Sth mutex_exit(&nfs4_ephemeral_thread_lock); 2373b9238976Sth } 2374b9238976Sth 2375b9238976Sth /*ARGSUSED*/ 2376b9238976Sth static void * 2377b9238976Sth nfs4_ephemeral_zsd_create(zoneid_t zoneid) 2378b9238976Sth { 2379b9238976Sth nfs4_trigger_globals_t *ntg; 2380b9238976Sth 2381b9238976Sth ntg = kmem_zalloc(sizeof (*ntg), KM_SLEEP); 2382b9238976Sth ntg->ntg_thread_started = FALSE; 2383b9238976Sth 2384b9238976Sth /* 2385b9238976Sth * This is the default.... 2386b9238976Sth */ 2387b9238976Sth ntg->ntg_mount_to = nfs4_trigger_thread_timer; 2388b9238976Sth 2389b9238976Sth mutex_init(&ntg->ntg_forest_lock, NULL, 2390b9238976Sth MUTEX_DEFAULT, NULL); 2391b9238976Sth 2392b9238976Sth return (ntg); 2393b9238976Sth } 2394b9238976Sth 2395b9238976Sth /* 2396b9238976Sth * Try a nice gentle walk down the forest and convince 2397b9238976Sth * all of the trees to gracefully give it up. 2398b9238976Sth */ 2399b9238976Sth /*ARGSUSED*/ 2400b9238976Sth static void 2401b9238976Sth nfs4_ephemeral_zsd_shutdown(zoneid_t zoneid, void *arg) 2402b9238976Sth { 2403b9238976Sth nfs4_trigger_globals_t *ntg = arg; 2404b9238976Sth 2405b9238976Sth if (!ntg) 2406b9238976Sth return; 2407b9238976Sth 2408b9238976Sth nfs4_ephemeral_harvest_forest(ntg, FALSE, FALSE); 2409b9238976Sth } 2410b9238976Sth 2411b9238976Sth /* 2412b9238976Sth * Race along the forest and rip all of the trees out by 2413b9238976Sth * their rootballs! 2414b9238976Sth */ 2415b9238976Sth /*ARGSUSED*/ 2416b9238976Sth static void 2417b9238976Sth nfs4_ephemeral_zsd_destroy(zoneid_t zoneid, void *arg) 2418b9238976Sth { 2419b9238976Sth nfs4_trigger_globals_t *ntg = arg; 2420b9238976Sth 2421b9238976Sth if (!ntg) 2422b9238976Sth return; 2423b9238976Sth 2424b9238976Sth nfs4_ephemeral_harvest_forest(ntg, TRUE, FALSE); 2425b9238976Sth 2426b9238976Sth mutex_destroy(&ntg->ntg_forest_lock); 2427b9238976Sth kmem_free(ntg, sizeof (*ntg)); 2428b9238976Sth } 2429b9238976Sth 2430b9238976Sth /* 2431b9238976Sth * This is the zone independent cleanup needed for 2432b9238976Sth * emphemeral mount processing. 2433b9238976Sth */ 2434b9238976Sth void 2435b9238976Sth nfs4_ephemeral_fini(void) 2436b9238976Sth { 2437b9238976Sth (void) zone_key_delete(nfs4_ephemeral_key); 2438b9238976Sth mutex_destroy(&nfs4_ephemeral_thread_lock); 2439b9238976Sth } 2440b9238976Sth 2441b9238976Sth /* 2442b9238976Sth * This is the zone independent initialization needed for 2443b9238976Sth * emphemeral mount processing. 2444b9238976Sth */ 2445b9238976Sth void 2446b9238976Sth nfs4_ephemeral_init(void) 2447b9238976Sth { 2448b9238976Sth mutex_init(&nfs4_ephemeral_thread_lock, NULL, MUTEX_DEFAULT, 2449b9238976Sth NULL); 2450b9238976Sth 2451b9238976Sth zone_key_create(&nfs4_ephemeral_key, nfs4_ephemeral_zsd_create, 2452b9238976Sth nfs4_ephemeral_zsd_shutdown, nfs4_ephemeral_zsd_destroy); 2453b9238976Sth } 2454b9238976Sth 2455b9238976Sth /* 2456b9238976Sth * nfssys() calls this function to set the per-zone 2457b9238976Sth * value of mount_to to drive when an ephemeral mount is 2458b9238976Sth * timed out. Each mount will grab a copy of this value 2459b9238976Sth * when mounted. 2460b9238976Sth */ 2461b9238976Sth void 2462b9238976Sth nfs4_ephemeral_set_mount_to(uint_t mount_to) 2463b9238976Sth { 2464b9238976Sth nfs4_trigger_globals_t *ntg; 2465b9238976Sth zone_t *zone = curproc->p_zone; 2466b9238976Sth 2467b9238976Sth ntg = zone_getspecific(nfs4_ephemeral_key, zone); 2468b9238976Sth 2469b9238976Sth ntg->ntg_mount_to = mount_to; 2470b9238976Sth } 2471b9238976Sth 2472b9238976Sth /* 2473b9238976Sth * Walk the list of v4 mount options; if they are currently set in vfsp, 2474b9238976Sth * append them to a new comma-separated mount option string, and return it. 2475b9238976Sth * 2476b9238976Sth * Caller should free by calling nfs4_trigger_destroy_mntopts(). 2477b9238976Sth */ 2478b9238976Sth static char * 2479b9238976Sth nfs4_trigger_create_mntopts(vfs_t *vfsp) 2480b9238976Sth { 2481b9238976Sth uint_t i; 2482b9238976Sth char *mntopts; 2483b9238976Sth struct vfssw *vswp; 2484b9238976Sth mntopts_t *optproto; 2485b9238976Sth 2486b9238976Sth mntopts = kmem_zalloc(MAX_MNTOPT_STR, KM_SLEEP); 2487b9238976Sth 2488b9238976Sth /* get the list of applicable mount options for v4; locks *vswp */ 2489b9238976Sth vswp = vfs_getvfssw(MNTTYPE_NFS4); 2490b9238976Sth optproto = &vswp->vsw_optproto; 2491b9238976Sth 2492b9238976Sth for (i = 0; i < optproto->mo_count; i++) { 2493b9238976Sth struct mntopt *mop = &optproto->mo_list[i]; 2494b9238976Sth 2495b9238976Sth if (mop->mo_flags & MO_EMPTY) 2496b9238976Sth continue; 2497b9238976Sth 2498b9238976Sth if (nfs4_trigger_add_mntopt(mntopts, mop->mo_name, vfsp)) { 2499b9238976Sth kmem_free(mntopts, MAX_MNTOPT_STR); 2500b9238976Sth vfs_unrefvfssw(vswp); 2501b9238976Sth return (NULL); 2502b9238976Sth } 2503b9238976Sth } 2504b9238976Sth 2505b9238976Sth vfs_unrefvfssw(vswp); 2506b9238976Sth 2507b9238976Sth /* 2508b9238976Sth * MNTOPT_XATTR is not in the v4 mount opt proto list, 2509b9238976Sth * and it may only be passed via MS_OPTIONSTR, so we 2510b9238976Sth * must handle it here. 2511b9238976Sth * 2512b9238976Sth * Ideally, it would be in the list, but NFS does not specify its 2513b9238976Sth * own opt proto list, it uses instead the default one. Since 2514b9238976Sth * not all filesystems support extended attrs, it would not be 2515b9238976Sth * appropriate to add it there. 2516b9238976Sth */ 2517b9238976Sth if (nfs4_trigger_add_mntopt(mntopts, MNTOPT_XATTR, vfsp) || 2518b9238976Sth nfs4_trigger_add_mntopt(mntopts, MNTOPT_NOXATTR, vfsp)) { 2519b9238976Sth kmem_free(mntopts, MAX_MNTOPT_STR); 2520b9238976Sth return (NULL); 2521b9238976Sth } 2522b9238976Sth 2523b9238976Sth return (mntopts); 2524b9238976Sth } 2525b9238976Sth 2526b9238976Sth static void 2527b9238976Sth nfs4_trigger_destroy_mntopts(char *mntopts) 2528b9238976Sth { 2529b9238976Sth if (mntopts) 2530b9238976Sth kmem_free(mntopts, MAX_MNTOPT_STR); 2531b9238976Sth } 2532b9238976Sth 2533b9238976Sth /* 2534b9238976Sth * Check a single mount option (optname). Add to mntopts if it is set in VFS. 2535b9238976Sth */ 2536b9238976Sth static int 2537b9238976Sth nfs4_trigger_add_mntopt(char *mntopts, char *optname, vfs_t *vfsp) 2538b9238976Sth { 2539b9238976Sth if (mntopts == NULL || optname == NULL || vfsp == NULL) 2540b9238976Sth return (EINVAL); 2541b9238976Sth 2542b9238976Sth if (vfs_optionisset(vfsp, optname, NULL)) { 2543b9238976Sth size_t mntoptslen = strlen(mntopts); 2544b9238976Sth size_t optnamelen = strlen(optname); 2545b9238976Sth 2546b9238976Sth /* +1 for ',', +1 for NUL */ 2547b9238976Sth if (mntoptslen + optnamelen + 2 > MAX_MNTOPT_STR) 2548b9238976Sth return (EOVERFLOW); 2549b9238976Sth 2550b9238976Sth /* first or subsequent mount option? */ 2551b9238976Sth if (*mntopts != '\0') 2552b9238976Sth (void) strcat(mntopts, ","); 2553b9238976Sth 2554b9238976Sth (void) strcat(mntopts, optname); 2555b9238976Sth } 2556b9238976Sth 2557b9238976Sth return (0); 2558b9238976Sth } 2559b9238976Sth 2560b9238976Sth static enum clnt_stat 2561b9238976Sth nfs4_trigger_ping_server(servinfo4_t *svp, int nointr) 2562b9238976Sth { 2563b9238976Sth int retries, error; 2564b9238976Sth uint_t max_msgsize; 2565b9238976Sth enum clnt_stat status; 2566b9238976Sth CLIENT *cl; 2567b9238976Sth struct timeval timeout; 2568b9238976Sth 2569b9238976Sth /* as per recov_newserver() */ 2570b9238976Sth max_msgsize = 0; 2571b9238976Sth retries = 1; 2572b9238976Sth timeout.tv_sec = 2; 2573b9238976Sth timeout.tv_usec = 0; 2574b9238976Sth 2575b9238976Sth error = clnt_tli_kcreate(svp->sv_knconf, &svp->sv_addr, NFS_PROGRAM, 2576b9238976Sth NFS_V4, max_msgsize, retries, CRED(), &cl); 2577b9238976Sth if (error) 2578b9238976Sth return (RPC_FAILED); 2579b9238976Sth 2580b9238976Sth if (nointr) 2581b9238976Sth cl->cl_nosignal = TRUE; 2582b9238976Sth status = CLNT_CALL(cl, RFS_NULL, xdr_void, NULL, xdr_void, NULL, 2583b9238976Sth timeout); 2584b9238976Sth if (nointr) 2585b9238976Sth cl->cl_nosignal = FALSE; 2586b9238976Sth 2587b9238976Sth AUTH_DESTROY(cl->cl_auth); 2588b9238976Sth CLNT_DESTROY(cl); 2589b9238976Sth 2590b9238976Sth return (status); 2591b9238976Sth } 2592