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 /* 23546a3997SThomas Haynes * Copyright 2009 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> 682f172c55SRobert Thurlow #include <sys/priv.h> 69b9238976Sth 70b9238976Sth #include <rpc/types.h> 71b9238976Sth #include <rpc/auth.h> 72b9238976Sth #include <rpc/clnt.h> 73b9238976Sth 74b9238976Sth #include <nfs/nfs.h> 75b9238976Sth #include <nfs/nfs_clnt.h> 76b9238976Sth #include <nfs/nfs_acl.h> 77b9238976Sth #include <nfs/lm.h> 78b9238976Sth #include <nfs/nfs4.h> 79b9238976Sth #include <nfs/nfs4_kprot.h> 80b9238976Sth #include <nfs/rnode4.h> 81b9238976Sth #include <nfs/nfs4_clnt.h> 822f172c55SRobert Thurlow #include <nfs/nfsid_map.h> 832f172c55SRobert Thurlow #include <nfs/nfs4_idmap_impl.h> 84b9238976Sth 85b9238976Sth #include <vm/hat.h> 86b9238976Sth #include <vm/as.h> 87b9238976Sth #include <vm/page.h> 88b9238976Sth #include <vm/pvn.h> 89b9238976Sth #include <vm/seg.h> 90b9238976Sth #include <vm/seg_map.h> 91b9238976Sth #include <vm/seg_kpm.h> 92b9238976Sth #include <vm/seg_vn.h> 93b9238976Sth 94b9238976Sth #include <fs/fs_subr.h> 95b9238976Sth 96b9238976Sth #include <sys/ddi.h> 97b9238976Sth #include <sys/int_fmtio.h> 98b9238976Sth 99f39b8789Sth #include <sys/sunddi.h> 100b9238976Sth 101546a3997SThomas Haynes #include <sys/priv_names.h> 102546a3997SThomas Haynes 1032f172c55SRobert Thurlow extern zone_key_t nfs4clnt_zone_key; 1042f172c55SRobert Thurlow extern zone_key_t nfsidmap_zone_key; 1052f172c55SRobert Thurlow 106b9238976Sth /* 107b9238976Sth * The automatic unmounter thread stuff! 108b9238976Sth */ 109b9238976Sth static int nfs4_trigger_thread_timer = 20; /* in seconds */ 110b9238976Sth 111b9238976Sth /* 112b9238976Sth * Just a default.... 113b9238976Sth */ 114b9238976Sth static uint_t nfs4_trigger_mount_to = 240; 115b9238976Sth 116b9238976Sth typedef struct nfs4_trigger_globals { 117b9238976Sth kmutex_t ntg_forest_lock; 118b9238976Sth uint_t ntg_mount_to; 119b9238976Sth int ntg_thread_started; 120b9238976Sth nfs4_ephemeral_tree_t *ntg_forest; 121b9238976Sth } nfs4_trigger_globals_t; 122b9238976Sth 123b9238976Sth kmutex_t nfs4_ephemeral_thread_lock; 124b9238976Sth 125b9238976Sth zone_key_t nfs4_ephemeral_key = ZONE_KEY_UNINITIALIZED; 126b9238976Sth 127b9238976Sth static void nfs4_ephemeral_start_harvester(nfs4_trigger_globals_t *); 128b9238976Sth 129b9238976Sth /* 130b9238976Sth * Used for ephemeral mounts; contains data either duplicated from 131b9238976Sth * servinfo4_t, or hand-crafted, depending on type of ephemeral mount. 132b9238976Sth * 133b9238976Sth * It's intended that this structure is used solely for ephemeral 134b9238976Sth * mount-type specific data, for passing this data to 135b9238976Sth * nfs4_trigger_nargs_create(). 136b9238976Sth */ 137b9238976Sth typedef struct ephemeral_servinfo { 138b9238976Sth char *esi_hostname; 139b9238976Sth char *esi_netname; 140b9238976Sth char *esi_path; 141b9238976Sth int esi_path_len; 142b9238976Sth int esi_mount_flags; 143b9238976Sth struct netbuf *esi_addr; 144b9238976Sth struct netbuf *esi_syncaddr; 145b9238976Sth struct knetconfig *esi_knconf; 146b9238976Sth } ephemeral_servinfo_t; 147b9238976Sth 148b9238976Sth /* 149b9238976Sth * Collect together the mount-type specific and generic data args. 150b9238976Sth */ 151b9238976Sth typedef struct domount_args { 152b9238976Sth ephemeral_servinfo_t *dma_esi; 153b9238976Sth char *dma_hostlist; /* comma-sep. for RO failover */ 154b9238976Sth struct nfs_args *dma_nargs; 155b9238976Sth } domount_args_t; 156b9238976Sth 157b9238976Sth 158b9238976Sth /* 159b9238976Sth * The vnode ops functions for a trigger stub vnode 160b9238976Sth */ 161da6c28aaSamw static int nfs4_trigger_open(vnode_t **, int, cred_t *, caller_context_t *); 162da6c28aaSamw static int nfs4_trigger_getattr(vnode_t *, struct vattr *, int, cred_t *, 163da6c28aaSamw caller_context_t *); 164da6c28aaSamw static int nfs4_trigger_setattr(vnode_t *, struct vattr *, int, cred_t *, 165da6c28aaSamw caller_context_t *); 166da6c28aaSamw static int nfs4_trigger_access(vnode_t *, int, int, cred_t *, 167da6c28aaSamw caller_context_t *); 168da6c28aaSamw static int nfs4_trigger_readlink(vnode_t *, struct uio *, cred_t *, 169da6c28aaSamw caller_context_t *); 170da6c28aaSamw static int nfs4_trigger_lookup(vnode_t *, char *, vnode_t **, 171da6c28aaSamw struct pathname *, int, vnode_t *, cred_t *, caller_context_t *, 172da6c28aaSamw int *, pathname_t *); 173da6c28aaSamw static int nfs4_trigger_create(vnode_t *, char *, struct vattr *, 174da6c28aaSamw enum vcexcl, int, vnode_t **, cred_t *, int, caller_context_t *, 175da6c28aaSamw vsecattr_t *); 176da6c28aaSamw static int nfs4_trigger_remove(vnode_t *, char *, cred_t *, caller_context_t *, 177da6c28aaSamw int); 178da6c28aaSamw static int nfs4_trigger_link(vnode_t *, vnode_t *, char *, cred_t *, 179da6c28aaSamw caller_context_t *, int); 180da6c28aaSamw static int nfs4_trigger_rename(vnode_t *, char *, vnode_t *, char *, 181da6c28aaSamw cred_t *, caller_context_t *, int); 182da6c28aaSamw static int nfs4_trigger_mkdir(vnode_t *, char *, struct vattr *, 183da6c28aaSamw vnode_t **, cred_t *, caller_context_t *, int, vsecattr_t *vsecp); 184da6c28aaSamw static int nfs4_trigger_rmdir(vnode_t *, char *, vnode_t *, cred_t *, 185da6c28aaSamw caller_context_t *, int); 186da6c28aaSamw static int nfs4_trigger_symlink(vnode_t *, char *, struct vattr *, char *, 187da6c28aaSamw cred_t *, caller_context_t *, int); 188da6c28aaSamw static int nfs4_trigger_cmp(vnode_t *, vnode_t *, caller_context_t *); 189b9238976Sth 190b9238976Sth /* 191b9238976Sth * Regular NFSv4 vnodeops that we need to reference directly 192b9238976Sth */ 193da6c28aaSamw extern int nfs4_getattr(vnode_t *, struct vattr *, int, cred_t *, 194da6c28aaSamw caller_context_t *); 195da6c28aaSamw extern void nfs4_inactive(vnode_t *, cred_t *, caller_context_t *); 196b9238976Sth extern int nfs4_rwlock(vnode_t *, int, caller_context_t *); 197b9238976Sth extern void nfs4_rwunlock(vnode_t *, int, caller_context_t *); 198b9238976Sth extern int nfs4_lookup(vnode_t *, char *, vnode_t **, 199da6c28aaSamw struct pathname *, int, vnode_t *, cred_t *, 200da6c28aaSamw caller_context_t *, int *, pathname_t *); 201da6c28aaSamw extern int nfs4_pathconf(vnode_t *, int, ulong_t *, cred_t *, 202da6c28aaSamw caller_context_t *); 203da6c28aaSamw extern int nfs4_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *, 204da6c28aaSamw caller_context_t *); 205da6c28aaSamw extern int nfs4_fid(vnode_t *, fid_t *, caller_context_t *); 206da6c28aaSamw extern int nfs4_realvp(vnode_t *, vnode_t **, caller_context_t *); 207b9238976Sth 208546a3997SThomas Haynes static int nfs4_trigger_mount(vnode_t *, cred_t *, vnode_t **); 209b9238976Sth static int nfs4_trigger_domount(vnode_t *, domount_args_t *, vfs_t **, 2106962f5b8SThomas Haynes cred_t *, vnode_t **); 211d16da320SSimon Klinkert static int nfs4_trigger_domount_args_create(vnode_t *, cred_t *, 212d16da320SSimon Klinkert domount_args_t **dmap); 213b9238976Sth static void nfs4_trigger_domount_args_destroy(domount_args_t *dma, 214b9238976Sth vnode_t *vp); 2152f172c55SRobert Thurlow static ephemeral_servinfo_t *nfs4_trigger_esi_create(vnode_t *, servinfo4_t *, 2162f172c55SRobert Thurlow cred_t *); 217b9238976Sth static void nfs4_trigger_esi_destroy(ephemeral_servinfo_t *, vnode_t *); 218b9238976Sth static ephemeral_servinfo_t *nfs4_trigger_esi_create_mirrormount(vnode_t *, 219b9238976Sth servinfo4_t *); 2202f172c55SRobert Thurlow static ephemeral_servinfo_t *nfs4_trigger_esi_create_referral(vnode_t *, 2212f172c55SRobert Thurlow cred_t *); 222b9238976Sth static struct nfs_args *nfs4_trigger_nargs_create(mntinfo4_t *, servinfo4_t *, 223b9238976Sth ephemeral_servinfo_t *); 224b9238976Sth static void nfs4_trigger_nargs_destroy(struct nfs_args *); 225b9238976Sth static char *nfs4_trigger_create_mntopts(vfs_t *); 226b9238976Sth static void nfs4_trigger_destroy_mntopts(char *); 227b9238976Sth static int nfs4_trigger_add_mntopt(char *, char *, vfs_t *); 228b9238976Sth static enum clnt_stat nfs4_trigger_ping_server(servinfo4_t *, int); 2292f172c55SRobert Thurlow static enum clnt_stat nfs4_ping_server_common(struct knetconfig *, 2302f172c55SRobert Thurlow struct netbuf *, int); 231b9238976Sth 232b9238976Sth extern int umount2_engine(vfs_t *, int, cred_t *, int); 233b9238976Sth 234b9238976Sth vnodeops_t *nfs4_trigger_vnodeops; 235b9238976Sth 236b9238976Sth /* 237b9238976Sth * These are the vnodeops that we must define for stub vnodes. 238b9238976Sth * 239b9238976Sth * 240b9238976Sth * Many of the VOPs defined for NFSv4 do not need to be defined here, 241b9238976Sth * for various reasons. This will result in the VFS default function being 242b9238976Sth * used: 243b9238976Sth * 244b9238976Sth * - These VOPs require a previous VOP_OPEN to have occurred. That will have 245b9238976Sth * lost the reference to the stub vnode, meaning these should not be called: 246b9238976Sth * close, read, write, ioctl, readdir, seek. 247b9238976Sth * 248b9238976Sth * - These VOPs are meaningless for vnodes without data pages. Since the 249b9238976Sth * stub vnode is of type VDIR, these should not be called: 250b9238976Sth * space, getpage, putpage, map, addmap, delmap, pageio, fsync. 251b9238976Sth * 252b9238976Sth * - These VOPs are otherwise not applicable, and should not be called: 253b9238976Sth * dump, setsecattr. 254b9238976Sth * 255b9238976Sth * 256b9238976Sth * These VOPs we do not want to define, but nor do we want the VFS default 257b9238976Sth * action. Instead, we specify the VFS error function, with fs_error(), but 258b9238976Sth * note that fs_error() is not actually called. Instead it results in the 259b9238976Sth * use of the error function defined for the particular VOP, in vn_ops_table[]: 260b9238976Sth * 261b9238976Sth * - frlock, dispose, shrlock. 262b9238976Sth * 263b9238976Sth * 264b9238976Sth * These VOPs we define to use the corresponding regular NFSv4 vnodeop. 265b9238976Sth * NOTE: if any of these ops involve an OTW call with the stub FH, then 266b9238976Sth * that call must be wrapped with save_mnt_secinfo()/check_mnt_secinfo() 267b9238976Sth * to protect the security data in the servinfo4_t for the "parent" 268b9238976Sth * filesystem that contains the stub. 269b9238976Sth * 270b9238976Sth * - These VOPs should not trigger a mount, so that "ls -l" does not: 271b9238976Sth * pathconf, getsecattr. 272b9238976Sth * 273b9238976Sth * - These VOPs would not make sense to trigger: 274b9238976Sth * inactive, rwlock, rwunlock, fid, realvp. 275b9238976Sth */ 276b9238976Sth const fs_operation_def_t nfs4_trigger_vnodeops_template[] = { 277b9238976Sth VOPNAME_OPEN, { .vop_open = nfs4_trigger_open }, 278b9238976Sth VOPNAME_GETATTR, { .vop_getattr = nfs4_trigger_getattr }, 279b9238976Sth VOPNAME_SETATTR, { .vop_setattr = nfs4_trigger_setattr }, 280b9238976Sth VOPNAME_ACCESS, { .vop_access = nfs4_trigger_access }, 281b9238976Sth VOPNAME_LOOKUP, { .vop_lookup = nfs4_trigger_lookup }, 282b9238976Sth VOPNAME_CREATE, { .vop_create = nfs4_trigger_create }, 283b9238976Sth VOPNAME_REMOVE, { .vop_remove = nfs4_trigger_remove }, 284b9238976Sth VOPNAME_LINK, { .vop_link = nfs4_trigger_link }, 285b9238976Sth VOPNAME_RENAME, { .vop_rename = nfs4_trigger_rename }, 286b9238976Sth VOPNAME_MKDIR, { .vop_mkdir = nfs4_trigger_mkdir }, 287b9238976Sth VOPNAME_RMDIR, { .vop_rmdir = nfs4_trigger_rmdir }, 288b9238976Sth VOPNAME_SYMLINK, { .vop_symlink = nfs4_trigger_symlink }, 289b9238976Sth VOPNAME_READLINK, { .vop_readlink = nfs4_trigger_readlink }, 290b9238976Sth VOPNAME_INACTIVE, { .vop_inactive = nfs4_inactive }, 291b9238976Sth VOPNAME_FID, { .vop_fid = nfs4_fid }, 292b9238976Sth VOPNAME_RWLOCK, { .vop_rwlock = nfs4_rwlock }, 293b9238976Sth VOPNAME_RWUNLOCK, { .vop_rwunlock = nfs4_rwunlock }, 294b9238976Sth VOPNAME_REALVP, { .vop_realvp = nfs4_realvp }, 295b9238976Sth VOPNAME_GETSECATTR, { .vop_getsecattr = nfs4_getsecattr }, 296b9238976Sth VOPNAME_PATHCONF, { .vop_pathconf = nfs4_pathconf }, 297b9238976Sth VOPNAME_FRLOCK, { .error = fs_error }, 298b9238976Sth VOPNAME_DISPOSE, { .error = fs_error }, 299b9238976Sth VOPNAME_SHRLOCK, { .error = fs_error }, 300b9238976Sth VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 301b9238976Sth NULL, NULL 302b9238976Sth }; 303b9238976Sth 304d3a14591SThomas Haynes static void 305d708af74SThomas Haynes nfs4_ephemeral_tree_incr(nfs4_ephemeral_tree_t *net) 306d3a14591SThomas Haynes { 307d708af74SThomas Haynes ASSERT(mutex_owned(&net->net_cnt_lock)); 308d3a14591SThomas Haynes net->net_refcnt++; 309d3a14591SThomas Haynes ASSERT(net->net_refcnt != 0); 310d708af74SThomas Haynes } 311d708af74SThomas Haynes 312d708af74SThomas Haynes static void 313d708af74SThomas Haynes nfs4_ephemeral_tree_hold(nfs4_ephemeral_tree_t *net) 314d708af74SThomas Haynes { 315d708af74SThomas Haynes mutex_enter(&net->net_cnt_lock); 316d708af74SThomas Haynes nfs4_ephemeral_tree_incr(net); 317d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 318d3a14591SThomas Haynes } 319d3a14591SThomas Haynes 320d3a14591SThomas Haynes /* 321d3a14591SThomas Haynes * We need a safe way to decrement the refcnt whilst the 322d3a14591SThomas Haynes * lock is being held. 323d3a14591SThomas Haynes */ 324d3a14591SThomas Haynes static void 325d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(nfs4_ephemeral_tree_t *net) 326d3a14591SThomas Haynes { 327d3a14591SThomas Haynes ASSERT(mutex_owned(&net->net_cnt_lock)); 328d3a14591SThomas Haynes ASSERT(net->net_refcnt != 0); 329d3a14591SThomas Haynes net->net_refcnt--; 330d3a14591SThomas Haynes } 331d3a14591SThomas Haynes 332d3a14591SThomas Haynes static void 333d3a14591SThomas Haynes nfs4_ephemeral_tree_rele(nfs4_ephemeral_tree_t *net) 334d3a14591SThomas Haynes { 335d3a14591SThomas Haynes mutex_enter(&net->net_cnt_lock); 336d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 337d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 338d3a14591SThomas Haynes } 339d3a14591SThomas Haynes 340b9238976Sth /* 341b9238976Sth * Trigger ops for stub vnodes; for mirror mounts, etc. 342b9238976Sth * 343b9238976Sth * The general idea is that a "triggering" op will first call 344b9238976Sth * nfs4_trigger_mount(), which will find out whether a mount has already 345b9238976Sth * been triggered. 346b9238976Sth * 347b9238976Sth * If it has, then nfs4_trigger_mount() sets newvp to the root vnode 348b9238976Sth * of the covering vfs. 349b9238976Sth * 350b9238976Sth * If a mount has not yet been triggered, nfs4_trigger_mount() will do so, 351b9238976Sth * and again set newvp, as above. 352b9238976Sth * 353b9238976Sth * The triggering op may then re-issue the VOP by calling it on newvp. 354b9238976Sth * 355b9238976Sth * Note that some ops may perform custom action, and may or may not need 356b9238976Sth * to trigger a mount. 357b9238976Sth * 358b9238976Sth * Some ops need to call the regular NFSv4 vnodeop for a stub vnode. We 359b9238976Sth * obviously can't do this with VOP_<whatever>, since it's a stub vnode 360b9238976Sth * and that would just recurse. Instead, we call the v4 op directly, 361b9238976Sth * by name. This is OK, since we know that the vnode is for NFSv4, 362b9238976Sth * otherwise it couldn't be a stub. 363b9238976Sth * 364b9238976Sth */ 365b9238976Sth 366b9238976Sth static int 367da6c28aaSamw nfs4_trigger_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 368b9238976Sth { 369b9238976Sth int error; 370b9238976Sth vnode_t *newvp; 371b9238976Sth 372546a3997SThomas Haynes error = nfs4_trigger_mount(*vpp, cr, &newvp); 373b9238976Sth if (error) 374b9238976Sth return (error); 375b9238976Sth 376b9238976Sth /* Release the stub vnode, as we're losing the reference to it */ 377b9238976Sth VN_RELE(*vpp); 378b9238976Sth 379b9238976Sth /* Give the caller the root vnode of the newly-mounted fs */ 380b9238976Sth *vpp = newvp; 381b9238976Sth 382b9238976Sth /* return with VN_HELD(newvp) */ 383da6c28aaSamw return (VOP_OPEN(vpp, flag, cr, ct)); 384b9238976Sth } 385b9238976Sth 3862f172c55SRobert Thurlow void 3872f172c55SRobert Thurlow nfs4_fake_attrs(vnode_t *vp, struct vattr *vap) 3882f172c55SRobert Thurlow { 3892f172c55SRobert Thurlow uint_t mask; 3902f172c55SRobert Thurlow timespec_t now; 3912f172c55SRobert Thurlow 3922f172c55SRobert Thurlow /* 3932f172c55SRobert Thurlow * Set some attributes here for referrals. 3942f172c55SRobert Thurlow */ 3952f172c55SRobert Thurlow mask = vap->va_mask; 3962f172c55SRobert Thurlow bzero(vap, sizeof (struct vattr)); 3972f172c55SRobert Thurlow vap->va_mask = mask; 3982f172c55SRobert Thurlow vap->va_uid = 0; 3992f172c55SRobert Thurlow vap->va_gid = 0; 4002f172c55SRobert Thurlow vap->va_nlink = 1; 4012f172c55SRobert Thurlow vap->va_size = 1; 4022f172c55SRobert Thurlow gethrestime(&now); 4032f172c55SRobert Thurlow vap->va_atime = now; 4042f172c55SRobert Thurlow vap->va_mtime = now; 4052f172c55SRobert Thurlow vap->va_ctime = now; 4062f172c55SRobert Thurlow vap->va_type = VDIR; 4072f172c55SRobert Thurlow vap->va_mode = 0555; 4082f172c55SRobert Thurlow vap->va_fsid = vp->v_vfsp->vfs_dev; 4092f172c55SRobert Thurlow vap->va_rdev = 0; 4102f172c55SRobert Thurlow vap->va_blksize = MAXBSIZE; 4112f172c55SRobert Thurlow vap->va_nblocks = 1; 4122f172c55SRobert Thurlow vap->va_seq = 0; 4132f172c55SRobert Thurlow } 4142f172c55SRobert Thurlow 415b9238976Sth /* 416b9238976Sth * For the majority of cases, nfs4_trigger_getattr() will not trigger 417b9238976Sth * a mount. However, if ATTR_TRIGGER is set, we are being informed 418b9238976Sth * that we need to force the mount before we attempt to determine 419b9238976Sth * the attributes. The intent is an atomic operation for security 420b9238976Sth * testing. 4212f172c55SRobert Thurlow * 4222f172c55SRobert Thurlow * If we're not triggering a mount, we can still inquire about the 4232f172c55SRobert Thurlow * actual attributes from the server in the mirror mount case, 4242f172c55SRobert Thurlow * and will return manufactured attributes for a referral (see 4252f172c55SRobert Thurlow * the 'create' branch of find_referral_stubvp()). 426b9238976Sth */ 427b9238976Sth static int 428da6c28aaSamw nfs4_trigger_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 429da6c28aaSamw caller_context_t *ct) 430b9238976Sth { 431b9238976Sth int error; 432b9238976Sth 433b9238976Sth if (flags & ATTR_TRIGGER) { 434b9238976Sth vnode_t *newvp; 435b9238976Sth 436546a3997SThomas Haynes error = nfs4_trigger_mount(vp, cr, &newvp); 437b9238976Sth if (error) 438b9238976Sth return (error); 439b9238976Sth 440da6c28aaSamw error = VOP_GETATTR(newvp, vap, flags, cr, ct); 441b9238976Sth VN_RELE(newvp); 4422f172c55SRobert Thurlow 4432f172c55SRobert Thurlow } else if (RP_ISSTUB_MIRRORMOUNT(VTOR4(vp))) { 4442f172c55SRobert Thurlow 445da6c28aaSamw error = nfs4_getattr(vp, vap, flags, cr, ct); 4462f172c55SRobert Thurlow 4472f172c55SRobert Thurlow } else if (RP_ISSTUB_REFERRAL(VTOR4(vp))) { 4482f172c55SRobert Thurlow 4492f172c55SRobert Thurlow nfs4_fake_attrs(vp, vap); 4502f172c55SRobert Thurlow error = 0; 451b9238976Sth } 452b9238976Sth 453b9238976Sth return (error); 454b9238976Sth } 455b9238976Sth 456b9238976Sth static int 457b9238976Sth nfs4_trigger_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr, 458a17ce845SMarcel Telka caller_context_t *ct) 459b9238976Sth { 460b9238976Sth int error; 461b9238976Sth vnode_t *newvp; 462b9238976Sth 463546a3997SThomas Haynes error = nfs4_trigger_mount(vp, cr, &newvp); 464b9238976Sth if (error) 465b9238976Sth return (error); 466b9238976Sth 467b9238976Sth error = VOP_SETATTR(newvp, vap, flags, cr, ct); 468b9238976Sth VN_RELE(newvp); 469b9238976Sth 470b9238976Sth return (error); 471b9238976Sth } 472b9238976Sth 473b9238976Sth static int 474da6c28aaSamw nfs4_trigger_access(vnode_t *vp, int mode, int flags, cred_t *cr, 475da6c28aaSamw caller_context_t *ct) 476b9238976Sth { 477b9238976Sth int error; 478b9238976Sth vnode_t *newvp; 479b9238976Sth 480546a3997SThomas Haynes error = nfs4_trigger_mount(vp, cr, &newvp); 481b9238976Sth if (error) 482b9238976Sth return (error); 483b9238976Sth 484da6c28aaSamw error = VOP_ACCESS(newvp, mode, flags, cr, ct); 485b9238976Sth VN_RELE(newvp); 486b9238976Sth 487b9238976Sth return (error); 488b9238976Sth } 489b9238976Sth 490b9238976Sth static int 491da6c28aaSamw nfs4_trigger_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, 492da6c28aaSamw struct pathname *pnp, int flags, vnode_t *rdir, cred_t *cr, 493da6c28aaSamw caller_context_t *ct, int *deflags, pathname_t *rpnp) 494b9238976Sth { 495b9238976Sth int error; 496b9238976Sth vnode_t *newdvp; 497b9238976Sth rnode4_t *drp = VTOR4(dvp); 498b9238976Sth 499b9238976Sth ASSERT(RP_ISSTUB(drp)); 500b9238976Sth 501b9238976Sth /* 502b9238976Sth * It's not legal to lookup ".." for an fs root, so we mustn't pass 503b9238976Sth * that up. Instead, pass onto the regular op, regardless of whether 504b9238976Sth * we've triggered a mount. 505b9238976Sth */ 506b9238976Sth if (strcmp(nm, "..") == 0) 5072f172c55SRobert Thurlow if (RP_ISSTUB_MIRRORMOUNT(drp)) { 5082f172c55SRobert Thurlow return (nfs4_lookup(dvp, nm, vpp, pnp, flags, rdir, cr, 5092f172c55SRobert Thurlow ct, deflags, rpnp)); 5102f172c55SRobert Thurlow } else if (RP_ISSTUB_REFERRAL(drp)) { 5112f172c55SRobert Thurlow /* Return the parent vnode */ 5122f172c55SRobert Thurlow return (vtodv(dvp, vpp, cr, TRUE)); 5132f172c55SRobert Thurlow } 514b9238976Sth 515546a3997SThomas Haynes error = nfs4_trigger_mount(dvp, cr, &newdvp); 516b9238976Sth if (error) 517b9238976Sth return (error); 518b9238976Sth 519da6c28aaSamw error = VOP_LOOKUP(newdvp, nm, vpp, pnp, flags, rdir, cr, ct, 520da6c28aaSamw deflags, rpnp); 521b9238976Sth VN_RELE(newdvp); 522b9238976Sth 523b9238976Sth return (error); 524b9238976Sth } 525b9238976Sth 526b9238976Sth static int 527b9238976Sth nfs4_trigger_create(vnode_t *dvp, char *nm, struct vattr *va, 528da6c28aaSamw enum vcexcl exclusive, int mode, vnode_t **vpp, cred_t *cr, 529da6c28aaSamw int flags, caller_context_t *ct, vsecattr_t *vsecp) 530b9238976Sth { 531b9238976Sth int error; 532b9238976Sth vnode_t *newdvp; 533b9238976Sth 534546a3997SThomas Haynes error = nfs4_trigger_mount(dvp, cr, &newdvp); 535b9238976Sth if (error) 536b9238976Sth return (error); 537b9238976Sth 538da6c28aaSamw error = VOP_CREATE(newdvp, nm, va, exclusive, mode, vpp, cr, 539da6c28aaSamw flags, ct, vsecp); 540b9238976Sth VN_RELE(newdvp); 541b9238976Sth 542b9238976Sth return (error); 543b9238976Sth } 544b9238976Sth 545b9238976Sth static int 546da6c28aaSamw nfs4_trigger_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, 547da6c28aaSamw int flags) 548b9238976Sth { 549b9238976Sth int error; 550b9238976Sth vnode_t *newdvp; 551b9238976Sth 552546a3997SThomas Haynes error = nfs4_trigger_mount(dvp, cr, &newdvp); 553b9238976Sth if (error) 554b9238976Sth return (error); 555b9238976Sth 556da6c28aaSamw error = VOP_REMOVE(newdvp, nm, cr, ct, flags); 557b9238976Sth VN_RELE(newdvp); 558b9238976Sth 559b9238976Sth return (error); 560b9238976Sth } 561b9238976Sth 562b9238976Sth static int 563da6c28aaSamw nfs4_trigger_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr, 564da6c28aaSamw caller_context_t *ct, int flags) 565b9238976Sth { 566b9238976Sth int error; 567b9238976Sth vnode_t *newtdvp; 568b9238976Sth 569546a3997SThomas Haynes error = nfs4_trigger_mount(tdvp, cr, &newtdvp); 570b9238976Sth if (error) 571b9238976Sth return (error); 572b9238976Sth 573b9238976Sth /* 574b9238976Sth * We don't check whether svp is a stub. Let the NFSv4 code 575b9238976Sth * detect that error, and return accordingly. 576b9238976Sth */ 577da6c28aaSamw error = VOP_LINK(newtdvp, svp, tnm, cr, ct, flags); 578b9238976Sth VN_RELE(newtdvp); 579b9238976Sth 580b9238976Sth return (error); 581b9238976Sth } 582b9238976Sth 583b9238976Sth static int 584b9238976Sth nfs4_trigger_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 585da6c28aaSamw cred_t *cr, caller_context_t *ct, int flags) 586b9238976Sth { 587b9238976Sth int error; 588b9238976Sth vnode_t *newsdvp; 589b9238976Sth rnode4_t *tdrp = VTOR4(tdvp); 590b9238976Sth 591b9238976Sth /* 592b9238976Sth * We know that sdvp is a stub, otherwise we would not be here. 593b9238976Sth * 594b9238976Sth * If tdvp is also be a stub, there are two possibilities: it 595b9238976Sth * is either the same stub as sdvp [i.e. VN_CMP(sdvp, tdvp)] 596b9238976Sth * or it is a different stub [!VN_CMP(sdvp, tdvp)]. 597b9238976Sth * 598b9238976Sth * In the former case, just trigger sdvp, and treat tdvp as 599b9238976Sth * though it were not a stub. 600b9238976Sth * 601b9238976Sth * In the latter case, it might be a different stub for the 602b9238976Sth * same server fs as sdvp, or for a different server fs. 603b9238976Sth * Regardless, from the client perspective this would still 604b9238976Sth * be a cross-filesystem rename, and should not be allowed, 605b9238976Sth * so return EXDEV, without triggering either mount. 606b9238976Sth */ 607b9238976Sth if (RP_ISSTUB(tdrp) && !VN_CMP(sdvp, tdvp)) 608b9238976Sth return (EXDEV); 609b9238976Sth 610546a3997SThomas Haynes error = nfs4_trigger_mount(sdvp, cr, &newsdvp); 611b9238976Sth if (error) 612b9238976Sth return (error); 613b9238976Sth 614da6c28aaSamw error = VOP_RENAME(newsdvp, snm, tdvp, tnm, cr, ct, flags); 615b9238976Sth 616b9238976Sth VN_RELE(newsdvp); 617b9238976Sth 618b9238976Sth return (error); 619b9238976Sth } 620b9238976Sth 621da6c28aaSamw /* ARGSUSED */ 622b9238976Sth static int 623b9238976Sth nfs4_trigger_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, 624da6c28aaSamw cred_t *cr, caller_context_t *ct, int flags, vsecattr_t *vsecp) 625b9238976Sth { 626b9238976Sth int error; 627b9238976Sth vnode_t *newdvp; 628b9238976Sth 629546a3997SThomas Haynes error = nfs4_trigger_mount(dvp, cr, &newdvp); 630b9238976Sth if (error) 631b9238976Sth return (error); 632b9238976Sth 633da6c28aaSamw error = VOP_MKDIR(newdvp, nm, va, vpp, cr, ct, flags, vsecp); 634b9238976Sth VN_RELE(newdvp); 635b9238976Sth 636b9238976Sth return (error); 637b9238976Sth } 638b9238976Sth 639b9238976Sth static int 640da6c28aaSamw nfs4_trigger_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr, 641da6c28aaSamw caller_context_t *ct, int flags) 642b9238976Sth { 643b9238976Sth int error; 644b9238976Sth vnode_t *newdvp; 645b9238976Sth 646546a3997SThomas Haynes error = nfs4_trigger_mount(dvp, cr, &newdvp); 647b9238976Sth if (error) 648b9238976Sth return (error); 649b9238976Sth 650da6c28aaSamw error = VOP_RMDIR(newdvp, nm, cdir, cr, ct, flags); 651b9238976Sth VN_RELE(newdvp); 652b9238976Sth 653b9238976Sth return (error); 654b9238976Sth } 655b9238976Sth 656b9238976Sth static int 657b9238976Sth nfs4_trigger_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, 658da6c28aaSamw cred_t *cr, caller_context_t *ct, int flags) 659b9238976Sth { 660b9238976Sth int error; 661b9238976Sth vnode_t *newdvp; 662b9238976Sth 663546a3997SThomas Haynes error = nfs4_trigger_mount(dvp, cr, &newdvp); 664b9238976Sth if (error) 665b9238976Sth return (error); 666b9238976Sth 667da6c28aaSamw error = VOP_SYMLINK(newdvp, lnm, tva, tnm, cr, ct, flags); 668b9238976Sth VN_RELE(newdvp); 669b9238976Sth 670b9238976Sth return (error); 671b9238976Sth } 672b9238976Sth 673b9238976Sth static int 674da6c28aaSamw nfs4_trigger_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, 675da6c28aaSamw caller_context_t *ct) 676b9238976Sth { 677b9238976Sth int error; 678b9238976Sth vnode_t *newvp; 679b9238976Sth 680546a3997SThomas Haynes error = nfs4_trigger_mount(vp, cr, &newvp); 681b9238976Sth if (error) 682b9238976Sth return (error); 683b9238976Sth 684da6c28aaSamw error = VOP_READLINK(newvp, uiop, cr, ct); 685b9238976Sth VN_RELE(newvp); 686b9238976Sth 687b9238976Sth return (error); 688b9238976Sth } 689b9238976Sth 690b9238976Sth /* end of trigger vnode ops */ 691b9238976Sth 6926962f5b8SThomas Haynes /* 6936962f5b8SThomas Haynes * See if the mount has already been done by another caller. 6946962f5b8SThomas Haynes */ 6956962f5b8SThomas Haynes static int 6966962f5b8SThomas Haynes nfs4_trigger_mounted_already(vnode_t *vp, vnode_t **newvpp, 6976962f5b8SThomas Haynes bool_t *was_mounted, vfs_t **vfsp) 6986962f5b8SThomas Haynes { 6996962f5b8SThomas Haynes int error; 7006962f5b8SThomas Haynes mntinfo4_t *mi = VTOMI4(vp); 7016962f5b8SThomas Haynes 7026962f5b8SThomas Haynes *was_mounted = FALSE; 7036962f5b8SThomas Haynes 7046962f5b8SThomas Haynes error = vn_vfsrlock_wait(vp); 7056962f5b8SThomas Haynes if (error) 7066962f5b8SThomas Haynes return (error); 7076962f5b8SThomas Haynes 7086962f5b8SThomas Haynes *vfsp = vn_mountedvfs(vp); 7096962f5b8SThomas Haynes if (*vfsp != NULL) { 7106962f5b8SThomas Haynes /* the mount has already occurred */ 7116962f5b8SThomas Haynes error = VFS_ROOT(*vfsp, newvpp); 7126962f5b8SThomas Haynes if (!error) { 7136962f5b8SThomas Haynes /* need to update the reference time */ 7146962f5b8SThomas Haynes mutex_enter(&mi->mi_lock); 7156962f5b8SThomas Haynes if (mi->mi_ephemeral) 7166962f5b8SThomas Haynes mi->mi_ephemeral->ne_ref_time = 7176962f5b8SThomas Haynes gethrestime_sec(); 7186962f5b8SThomas Haynes mutex_exit(&mi->mi_lock); 7196962f5b8SThomas Haynes 7206962f5b8SThomas Haynes *was_mounted = TRUE; 7216962f5b8SThomas Haynes } 7226962f5b8SThomas Haynes } 7236962f5b8SThomas Haynes 7246962f5b8SThomas Haynes vn_vfsunlock(vp); 7256962f5b8SThomas Haynes return (0); 7266962f5b8SThomas Haynes } 7276962f5b8SThomas Haynes 728b9238976Sth /* 7292f172c55SRobert Thurlow * Mount upon a trigger vnode; for mirror-mounts, referrals, etc. 730b9238976Sth * 731b9238976Sth * The mount may have already occurred, via another thread. If not, 732b9238976Sth * assemble the location information - which may require fetching - and 733b9238976Sth * perform the mount. 734b9238976Sth * 735b9238976Sth * Sets newvp to be the root of the fs that is now covering vp. Note 736b9238976Sth * that we return with VN_HELD(*newvp). 737b9238976Sth * 738b9238976Sth * The caller is responsible for passing the VOP onto the covering fs. 739b9238976Sth */ 740b9238976Sth static int 741546a3997SThomas Haynes nfs4_trigger_mount(vnode_t *vp, cred_t *cr, vnode_t **newvpp) 742b9238976Sth { 743b9238976Sth int error; 744b9238976Sth vfs_t *vfsp; 745b9238976Sth rnode4_t *rp = VTOR4(vp); 746b9238976Sth mntinfo4_t *mi = VTOMI4(vp); 747b9238976Sth domount_args_t *dma; 748b9238976Sth 749b9238976Sth nfs4_ephemeral_tree_t *net; 750b9238976Sth 751b9238976Sth bool_t must_unlock = FALSE; 752b9238976Sth bool_t is_building = FALSE; 7536962f5b8SThomas Haynes bool_t was_mounted = FALSE; 754b9238976Sth 755546a3997SThomas Haynes cred_t *mcred = NULL; 756b9238976Sth 757b9238976Sth nfs4_trigger_globals_t *ntg; 758b9238976Sth 759b9238976Sth zone_t *zone = curproc->p_zone; 760b9238976Sth 761b9238976Sth ASSERT(RP_ISSTUB(rp)); 762b9238976Sth 763b9238976Sth *newvpp = NULL; 764b9238976Sth 765b9238976Sth /* 766b9238976Sth * Has the mount already occurred? 767b9238976Sth */ 7686962f5b8SThomas Haynes error = nfs4_trigger_mounted_already(vp, newvpp, 7696962f5b8SThomas Haynes &was_mounted, &vfsp); 7706962f5b8SThomas Haynes if (error || was_mounted) 771b9238976Sth goto done; 772b9238976Sth 773b9238976Sth ntg = zone_getspecific(nfs4_ephemeral_key, zone); 774b9238976Sth ASSERT(ntg != NULL); 775b9238976Sth 776b9238976Sth mutex_enter(&mi->mi_lock); 777b9238976Sth 778b9238976Sth /* 779b9238976Sth * We need to lock down the ephemeral tree. 780b9238976Sth */ 781b9238976Sth if (mi->mi_ephemeral_tree == NULL) { 782b9238976Sth net = kmem_zalloc(sizeof (*net), KM_SLEEP); 783b9238976Sth mutex_init(&net->net_tree_lock, NULL, MUTEX_DEFAULT, NULL); 784b9238976Sth mutex_init(&net->net_cnt_lock, NULL, MUTEX_DEFAULT, NULL); 785b9238976Sth net->net_refcnt = 1; 786b9238976Sth net->net_status = NFS4_EPHEMERAL_TREE_BUILDING; 787b9238976Sth is_building = TRUE; 788b9238976Sth 789b9238976Sth /* 790b9238976Sth * We need to add it to the zone specific list for 791b9238976Sth * automatic unmounting and harvesting of deadwood. 792b9238976Sth */ 793b9238976Sth mutex_enter(&ntg->ntg_forest_lock); 794b9238976Sth if (ntg->ntg_forest != NULL) 795b9238976Sth net->net_next = ntg->ntg_forest; 796b9238976Sth ntg->ntg_forest = net; 797b9238976Sth mutex_exit(&ntg->ntg_forest_lock); 798b9238976Sth 799b9238976Sth /* 800b9238976Sth * No lock order confusion with mi_lock because no 801b9238976Sth * other node could have grabbed net_tree_lock. 802b9238976Sth */ 803b9238976Sth mutex_enter(&net->net_tree_lock); 804b9238976Sth mi->mi_ephemeral_tree = net; 805b9238976Sth net->net_mount = mi; 806b9238976Sth mutex_exit(&mi->mi_lock); 807b87f76edSThomas Haynes 808b87f76edSThomas Haynes MI4_HOLD(mi); 809b87f76edSThomas Haynes VFS_HOLD(mi->mi_vfsp); 810b9238976Sth } else { 811b9238976Sth net = mi->mi_ephemeral_tree; 812d3a14591SThomas Haynes nfs4_ephemeral_tree_hold(net); 813d3a14591SThomas Haynes 814d708af74SThomas Haynes mutex_exit(&mi->mi_lock); 815d708af74SThomas Haynes 816d3a14591SThomas Haynes mutex_enter(&net->net_tree_lock); 817b9238976Sth 818b9238976Sth /* 819d3a14591SThomas Haynes * We can only procede if the tree is neither locked 820d3a14591SThomas Haynes * nor being torn down. 821b9238976Sth */ 822d3a14591SThomas Haynes mutex_enter(&net->net_cnt_lock); 823d3a14591SThomas Haynes if (net->net_status & NFS4_EPHEMERAL_TREE_PROCESSING) { 824d3a14591SThomas Haynes nfs4_ephemeral_tree_decr(net); 825d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 826d3a14591SThomas Haynes mutex_exit(&net->net_tree_lock); 827d3a14591SThomas Haynes 828d3a14591SThomas Haynes return (EIO); 829d3a14591SThomas Haynes } 830d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 831b9238976Sth } 832b9238976Sth 833b9238976Sth mutex_enter(&net->net_cnt_lock); 834b9238976Sth net->net_status |= NFS4_EPHEMERAL_TREE_MOUNTING; 835b9238976Sth mutex_exit(&net->net_cnt_lock); 836b9238976Sth 837b9238976Sth must_unlock = TRUE; 838b9238976Sth 839d16da320SSimon Klinkert error = nfs4_trigger_domount_args_create(vp, cr, &dma); 840d16da320SSimon Klinkert if (error) 841b9238976Sth goto done; 842b9238976Sth 843b9238976Sth /* 844b9238976Sth * Note that since we define mirror mounts to work 845546a3997SThomas Haynes * for any user, we simply extend the privileges of 846546a3997SThomas Haynes * the user's credentials to allow the mount to 847546a3997SThomas Haynes * proceed. 848b9238976Sth */ 849546a3997SThomas Haynes mcred = crdup(cr); 850546a3997SThomas Haynes if (mcred == NULL) { 851546a3997SThomas Haynes error = EINVAL; 852d16da320SSimon Klinkert nfs4_trigger_domount_args_destroy(dma, vp); 853546a3997SThomas Haynes goto done; 854546a3997SThomas Haynes } 855546a3997SThomas Haynes 856546a3997SThomas Haynes crset_zone_privall(mcred); 8572f172c55SRobert Thurlow if (is_system_labeled()) 8582f172c55SRobert Thurlow (void) setpflags(NET_MAC_AWARE, 1, mcred); 859b9238976Sth 8606962f5b8SThomas Haynes error = nfs4_trigger_domount(vp, dma, &vfsp, mcred, newvpp); 861b9238976Sth nfs4_trigger_domount_args_destroy(dma, vp); 862b9238976Sth 8632f172c55SRobert Thurlow DTRACE_PROBE2(nfs4clnt__func__referral__mount, 8642f172c55SRobert Thurlow vnode_t *, vp, int, error); 8652f172c55SRobert Thurlow 866546a3997SThomas Haynes crfree(mcred); 867b9238976Sth 868b9238976Sth done: 8696962f5b8SThomas Haynes 870b9238976Sth if (must_unlock) { 871b9238976Sth mutex_enter(&net->net_cnt_lock); 872b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_MOUNTING; 8732f172c55SRobert Thurlow 8742f172c55SRobert Thurlow /* 8752f172c55SRobert Thurlow * REFCNT: If we are the root of the tree, then we need 8762f172c55SRobert Thurlow * to keep a reference because we malloced the tree and 8772f172c55SRobert Thurlow * this is where we tied it to our mntinfo. 8782f172c55SRobert Thurlow * 8792f172c55SRobert Thurlow * If we are not the root of the tree, then our tie to 8802f172c55SRobert Thurlow * the mntinfo occured elsewhere and we need to 8812f172c55SRobert Thurlow * decrement the reference to the tree. 8822f172c55SRobert Thurlow */ 883b9238976Sth if (is_building) 884b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_BUILDING; 8852f172c55SRobert Thurlow else 8862f172c55SRobert Thurlow nfs4_ephemeral_tree_decr(net); 887b9238976Sth mutex_exit(&net->net_cnt_lock); 888b9238976Sth 889b9238976Sth mutex_exit(&net->net_tree_lock); 890b9238976Sth } 891b9238976Sth 892b9238976Sth if (!error && (newvpp == NULL || *newvpp == NULL)) 893b9238976Sth error = ENOSYS; 894b9238976Sth 895b9238976Sth return (error); 896b9238976Sth } 897b9238976Sth 898b9238976Sth /* 899b9238976Sth * Collect together both the generic & mount-type specific args. 900b9238976Sth */ 901d16da320SSimon Klinkert static int 902d16da320SSimon Klinkert nfs4_trigger_domount_args_create(vnode_t *vp, cred_t *cr, domount_args_t **dmap) 903b9238976Sth { 904b9238976Sth int nointr; 905b9238976Sth char *hostlist; 906b9238976Sth servinfo4_t *svp; 907b9238976Sth struct nfs_args *nargs, *nargs_head; 908b9238976Sth enum clnt_stat status; 909b9238976Sth ephemeral_servinfo_t *esi, *esi_first; 910b9238976Sth domount_args_t *dma; 911b9238976Sth mntinfo4_t *mi = VTOMI4(vp); 912b9238976Sth 913b9238976Sth nointr = !(mi->mi_flags & MI4_INT); 914b9238976Sth hostlist = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 915b9238976Sth 916b9238976Sth svp = mi->mi_curr_serv; 917b9238976Sth /* check if the current server is responding */ 918b9238976Sth status = nfs4_trigger_ping_server(svp, nointr); 919b9238976Sth if (status == RPC_SUCCESS) { 9202f172c55SRobert Thurlow esi_first = nfs4_trigger_esi_create(vp, svp, cr); 921b9238976Sth if (esi_first == NULL) { 922b9238976Sth kmem_free(hostlist, MAXPATHLEN); 923d16da320SSimon Klinkert return (EINVAL); 924b9238976Sth } 925b9238976Sth 926b9238976Sth (void) strlcpy(hostlist, esi_first->esi_hostname, MAXPATHLEN); 927b9238976Sth 928b9238976Sth nargs_head = nfs4_trigger_nargs_create(mi, svp, esi_first); 929b9238976Sth } else { 930b9238976Sth /* current server did not respond */ 931b9238976Sth esi_first = NULL; 932b9238976Sth nargs_head = NULL; 933b9238976Sth } 934b9238976Sth nargs = nargs_head; 935b9238976Sth 936b9238976Sth /* 937b9238976Sth * NFS RO failover. 938b9238976Sth * 939b9238976Sth * If we have multiple servinfo4 structures, linked via sv_next, 940b9238976Sth * we must create one nfs_args for each, linking the nfs_args via 941b9238976Sth * nfs_ext_u.nfs_extB.next. 942b9238976Sth * 943b9238976Sth * We need to build a corresponding esi for each, too, but that is 944b9238976Sth * used solely for building nfs_args, and may be immediately 945b9238976Sth * discarded, as domount() requires the info from just one esi, 946b9238976Sth * but all the nfs_args. 947b9238976Sth * 948b9238976Sth * Currently, the NFS mount code will hang if not all servers 949b9238976Sth * requested are available. To avoid that, we need to ping each 950b9238976Sth * server, here, and remove it from the list if it is not 951b9238976Sth * responding. This has the side-effect of that server then 952b9238976Sth * being permanently unavailable for this failover mount, even if 953b9238976Sth * it recovers. That's unfortunate, but the best we can do until 954b9238976Sth * the mount code path is fixed. 955b9238976Sth */ 956b9238976Sth 957b9238976Sth /* 958b9238976Sth * If the current server was down, loop indefinitely until we find 959b9238976Sth * at least one responsive server. 960b9238976Sth */ 961b9238976Sth do { 962b9238976Sth /* no locking needed for sv_next; it is only set at fs mount */ 963b9238976Sth for (svp = mi->mi_servers; svp != NULL; svp = svp->sv_next) { 964b9238976Sth struct nfs_args *next; 965b9238976Sth 966b9238976Sth /* 967b9238976Sth * nargs_head: the head of the nfs_args list 968b9238976Sth * nargs: the current tail of the list 969b9238976Sth * next: the newly-created element to be added 970b9238976Sth */ 971b9238976Sth 972b9238976Sth /* 973b9238976Sth * We've already tried the current server, above; 974b9238976Sth * if it was responding, we have already included it 975b9238976Sth * and it may now be ignored. 976b9238976Sth * 977b9238976Sth * Otherwise, try it again, since it may now have 978b9238976Sth * recovered. 979b9238976Sth */ 980b9238976Sth if (svp == mi->mi_curr_serv && esi_first != NULL) 981b9238976Sth continue; 982b9238976Sth 983b9238976Sth (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 984b9238976Sth if (svp->sv_flags & SV4_NOTINUSE) { 985b9238976Sth nfs_rw_exit(&svp->sv_lock); 986b9238976Sth continue; 987b9238976Sth } 988b9238976Sth nfs_rw_exit(&svp->sv_lock); 989b9238976Sth 990b9238976Sth /* check if the server is responding */ 991b9238976Sth status = nfs4_trigger_ping_server(svp, nointr); 992d16da320SSimon Klinkert if (status == RPC_INTR) { 993d16da320SSimon Klinkert kmem_free(hostlist, MAXPATHLEN); 994d16da320SSimon Klinkert nfs4_trigger_esi_destroy(esi_first, vp); 995d16da320SSimon Klinkert nargs = nargs_head; 996d16da320SSimon Klinkert while (nargs != NULL) { 997d16da320SSimon Klinkert next = nargs->nfs_ext_u.nfs_extB.next; 998d16da320SSimon Klinkert nfs4_trigger_nargs_destroy(nargs); 999d16da320SSimon Klinkert nargs = next; 1000d16da320SSimon Klinkert } 1001d16da320SSimon Klinkert return (EINTR); 1002d16da320SSimon Klinkert } else if (status != RPC_SUCCESS) { 1003d16da320SSimon Klinkert /* if the server did not respond, ignore it */ 1004b9238976Sth continue; 1005d16da320SSimon Klinkert } 1006b9238976Sth 10072f172c55SRobert Thurlow esi = nfs4_trigger_esi_create(vp, svp, cr); 1008b9238976Sth if (esi == NULL) 1009b9238976Sth continue; 1010b9238976Sth 1011b9238976Sth /* 1012b9238976Sth * If the original current server (mi_curr_serv) 1013b9238976Sth * was down when when we first tried it, 1014b9238976Sth * (i.e. esi_first == NULL), 1015b9238976Sth * we select this new server (svp) to be the server 1016b9238976Sth * that we will actually contact (esi_first). 1017b9238976Sth * 1018b9238976Sth * Note that it's possible that mi_curr_serv == svp, 1019b9238976Sth * if that mi_curr_serv was down but has now recovered. 1020b9238976Sth */ 1021b9238976Sth next = nfs4_trigger_nargs_create(mi, svp, esi); 1022b9238976Sth if (esi_first == NULL) { 1023b9238976Sth ASSERT(nargs == NULL); 1024b9238976Sth ASSERT(nargs_head == NULL); 1025b9238976Sth nargs_head = next; 1026b9238976Sth esi_first = esi; 1027b9238976Sth (void) strlcpy(hostlist, 1028b9238976Sth esi_first->esi_hostname, MAXPATHLEN); 1029b9238976Sth } else { 1030b9238976Sth ASSERT(nargs_head != NULL); 1031b9238976Sth nargs->nfs_ext_u.nfs_extB.next = next; 1032b9238976Sth (void) strlcat(hostlist, ",", MAXPATHLEN); 1033b9238976Sth (void) strlcat(hostlist, esi->esi_hostname, 1034b9238976Sth MAXPATHLEN); 1035b9238976Sth /* esi was only needed for hostname & nargs */ 1036b9238976Sth nfs4_trigger_esi_destroy(esi, vp); 1037b9238976Sth } 1038b9238976Sth 1039b9238976Sth nargs = next; 1040b9238976Sth } 1041b9238976Sth 1042b9238976Sth /* if we've had no response at all, wait a second */ 1043b9238976Sth if (esi_first == NULL) 1044b9238976Sth delay(drv_usectohz(1000000)); 1045b9238976Sth 1046b9238976Sth } while (esi_first == NULL); 1047b9238976Sth ASSERT(nargs_head != NULL); 1048b9238976Sth 1049b9238976Sth dma = kmem_zalloc(sizeof (domount_args_t), KM_SLEEP); 1050b9238976Sth dma->dma_esi = esi_first; 1051b9238976Sth dma->dma_hostlist = hostlist; 1052b9238976Sth dma->dma_nargs = nargs_head; 1053d16da320SSimon Klinkert *dmap = dma; 1054b9238976Sth 1055d16da320SSimon Klinkert return (0); 1056b9238976Sth } 1057b9238976Sth 1058b9238976Sth static void 1059b9238976Sth nfs4_trigger_domount_args_destroy(domount_args_t *dma, vnode_t *vp) 1060b9238976Sth { 1061b9238976Sth if (dma != NULL) { 1062b9238976Sth if (dma->dma_esi != NULL && vp != NULL) 1063b9238976Sth nfs4_trigger_esi_destroy(dma->dma_esi, vp); 1064b9238976Sth 1065b9238976Sth if (dma->dma_hostlist != NULL) 1066b9238976Sth kmem_free(dma->dma_hostlist, MAXPATHLEN); 1067b9238976Sth 1068b9238976Sth if (dma->dma_nargs != NULL) { 1069b9238976Sth struct nfs_args *nargs = dma->dma_nargs; 1070b9238976Sth 1071b9238976Sth do { 1072b9238976Sth struct nfs_args *next = 1073b9238976Sth nargs->nfs_ext_u.nfs_extB.next; 1074b9238976Sth 1075b9238976Sth nfs4_trigger_nargs_destroy(nargs); 1076b9238976Sth nargs = next; 1077b9238976Sth } while (nargs != NULL); 1078b9238976Sth } 1079b9238976Sth 1080b9238976Sth kmem_free(dma, sizeof (domount_args_t)); 1081b9238976Sth } 1082b9238976Sth } 1083b9238976Sth 1084b9238976Sth /* 1085b9238976Sth * The ephemeral_servinfo_t struct contains basic information we will need to 1086b9238976Sth * perform the mount. Whilst the structure is generic across different 1087b9238976Sth * types of ephemeral mount, the way we gather its contents differs. 1088b9238976Sth */ 1089b9238976Sth static ephemeral_servinfo_t * 10902f172c55SRobert Thurlow nfs4_trigger_esi_create(vnode_t *vp, servinfo4_t *svp, cred_t *cr) 1091b9238976Sth { 1092b9238976Sth ephemeral_servinfo_t *esi; 1093b9238976Sth rnode4_t *rp = VTOR4(vp); 1094b9238976Sth 1095b9238976Sth ASSERT(RP_ISSTUB(rp)); 1096b9238976Sth 1097b9238976Sth /* Call the ephemeral type-specific routine */ 1098b9238976Sth if (RP_ISSTUB_MIRRORMOUNT(rp)) 1099b9238976Sth esi = nfs4_trigger_esi_create_mirrormount(vp, svp); 11002f172c55SRobert Thurlow else if (RP_ISSTUB_REFERRAL(rp)) 11012f172c55SRobert Thurlow esi = nfs4_trigger_esi_create_referral(vp, cr); 1102b9238976Sth else 1103b9238976Sth esi = NULL; 1104b9238976Sth return (esi); 1105b9238976Sth } 1106b9238976Sth 1107b9238976Sth static void 1108b9238976Sth nfs4_trigger_esi_destroy(ephemeral_servinfo_t *esi, vnode_t *vp) 1109b9238976Sth { 1110b9238976Sth rnode4_t *rp = VTOR4(vp); 1111b9238976Sth 1112b9238976Sth ASSERT(RP_ISSTUB(rp)); 1113b9238976Sth 1114b9238976Sth /* Currently, no need for an ephemeral type-specific routine */ 1115b9238976Sth 1116b9238976Sth /* 1117b9238976Sth * The contents of ephemeral_servinfo_t goes into nfs_args, 1118b9238976Sth * and will be handled by nfs4_trigger_nargs_destroy(). 1119b9238976Sth * We need only free the structure itself. 1120b9238976Sth */ 1121b9238976Sth if (esi != NULL) 1122b9238976Sth kmem_free(esi, sizeof (ephemeral_servinfo_t)); 1123b9238976Sth } 1124b9238976Sth 1125b9238976Sth /* 1126b9238976Sth * Some of this may turn out to be common with other ephemeral types, 1127b9238976Sth * in which case it should be moved to nfs4_trigger_esi_create(), or a 1128b9238976Sth * common function called. 1129b9238976Sth */ 11302f172c55SRobert Thurlow 11312f172c55SRobert Thurlow /* 11322f172c55SRobert Thurlow * Mirror mounts case - should have all data available 11332f172c55SRobert Thurlow */ 1134b9238976Sth static ephemeral_servinfo_t * 1135b9238976Sth nfs4_trigger_esi_create_mirrormount(vnode_t *vp, servinfo4_t *svp) 1136b9238976Sth { 1137b9238976Sth char *stubpath; 1138b9238976Sth struct knetconfig *sikncp, *svkncp; 1139b9238976Sth struct netbuf *bufp; 1140b9238976Sth ephemeral_servinfo_t *esi; 1141b9238976Sth 1142b9238976Sth esi = kmem_zalloc(sizeof (ephemeral_servinfo_t), KM_SLEEP); 1143b9238976Sth 1144b9238976Sth /* initially set to be our type of ephemeral mount; may be added to */ 1145b9238976Sth esi->esi_mount_flags = NFSMNT_MIRRORMOUNT; 1146b9238976Sth 1147b9238976Sth /* 1148b9238976Sth * We're copying info from the stub rnode's servinfo4, but 1149b9238976Sth * we must create new copies, not pointers, since this information 1150b9238976Sth * is to be associated with the new mount, which will be 1151b9238976Sth * unmounted (and its structures freed) separately 1152b9238976Sth */ 1153b9238976Sth 1154b9238976Sth /* 1155b9238976Sth * Sizes passed to kmem_[z]alloc here must match those freed 1156b9238976Sth * in nfs4_free_args() 1157b9238976Sth */ 1158b9238976Sth 1159b9238976Sth /* 1160b9238976Sth * We hold sv_lock across kmem_zalloc() calls that may sleep, but this 1161b9238976Sth * is difficult to avoid: as we need to read svp to calculate the 1162b9238976Sth * sizes to be allocated. 1163b9238976Sth */ 1164b9238976Sth (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1165b9238976Sth 1166b9238976Sth esi->esi_hostname = kmem_zalloc(strlen(svp->sv_hostname) + 1, KM_SLEEP); 1167b9238976Sth (void) strcat(esi->esi_hostname, svp->sv_hostname); 1168b9238976Sth 1169b9238976Sth esi->esi_addr = kmem_zalloc(sizeof (struct netbuf), KM_SLEEP); 1170b9238976Sth bufp = esi->esi_addr; 1171b9238976Sth bufp->len = svp->sv_addr.len; 1172b9238976Sth bufp->maxlen = svp->sv_addr.maxlen; 1173b9238976Sth bufp->buf = kmem_zalloc(bufp->len, KM_SLEEP); 1174b9238976Sth bcopy(svp->sv_addr.buf, bufp->buf, bufp->len); 1175b9238976Sth 1176b9238976Sth esi->esi_knconf = kmem_zalloc(sizeof (*esi->esi_knconf), KM_SLEEP); 1177b9238976Sth sikncp = esi->esi_knconf; 1178b9238976Sth svkncp = svp->sv_knconf; 1179b9238976Sth sikncp->knc_semantics = svkncp->knc_semantics; 1180b9238976Sth sikncp->knc_protofmly = (caddr_t)kmem_zalloc(KNC_STRSIZE, KM_SLEEP); 1181b9238976Sth (void) strcat((char *)sikncp->knc_protofmly, 1182b9238976Sth (char *)svkncp->knc_protofmly); 1183b9238976Sth sikncp->knc_proto = (caddr_t)kmem_zalloc(KNC_STRSIZE, KM_SLEEP); 1184b9238976Sth (void) strcat((char *)sikncp->knc_proto, (char *)svkncp->knc_proto); 1185b9238976Sth sikncp->knc_rdev = svkncp->knc_rdev; 1186b9238976Sth 1187b9238976Sth /* 1188b9238976Sth * Used when AUTH_DH is negotiated. 1189b9238976Sth * 1190b9238976Sth * This is ephemeral mount-type specific, since it contains the 1191b9238976Sth * server's time-sync syncaddr. 1192b9238976Sth */ 1193b9238976Sth if (svp->sv_dhsec) { 1194b9238976Sth struct netbuf *bufp; 1195b9238976Sth sec_data_t *sdata; 1196b9238976Sth dh_k4_clntdata_t *data; 1197b9238976Sth 1198b9238976Sth sdata = svp->sv_dhsec; 1199b9238976Sth data = (dh_k4_clntdata_t *)sdata->data; 1200b9238976Sth ASSERT(sdata->rpcflavor == AUTH_DH); 1201b9238976Sth 1202b9238976Sth bufp = kmem_zalloc(sizeof (struct netbuf), KM_SLEEP); 1203b9238976Sth bufp->len = data->syncaddr.len; 1204b9238976Sth bufp->maxlen = data->syncaddr.maxlen; 1205b9238976Sth bufp->buf = kmem_zalloc(bufp->len, KM_SLEEP); 1206b9238976Sth bcopy(data->syncaddr.buf, bufp->buf, bufp->len); 1207b9238976Sth esi->esi_syncaddr = bufp; 1208b9238976Sth 1209b9238976Sth if (data->netname != NULL) { 1210b9238976Sth int nmlen = data->netnamelen; 1211b9238976Sth 1212b9238976Sth /* 1213b9238976Sth * We need to copy from a dh_k4_clntdata_t 1214b9238976Sth * netname/netnamelen pair to a NUL-terminated 1215b9238976Sth * netname string suitable for putting in nfs_args, 1216b9238976Sth * where the latter has no netnamelen field. 1217b9238976Sth */ 1218b9238976Sth esi->esi_netname = kmem_zalloc(nmlen + 1, KM_SLEEP); 1219b9238976Sth bcopy(data->netname, esi->esi_netname, nmlen); 1220b9238976Sth } 1221b9238976Sth } else { 1222b9238976Sth esi->esi_syncaddr = NULL; 1223b9238976Sth esi->esi_netname = NULL; 1224b9238976Sth } 1225b9238976Sth 1226b9238976Sth stubpath = fn_path(VTOSV(vp)->sv_name); 1227b9238976Sth /* step over initial '.', to avoid e.g. sv_path: "/tank./ws" */ 1228b9238976Sth ASSERT(*stubpath == '.'); 1229b9238976Sth stubpath += 1; 1230b9238976Sth 1231b9238976Sth /* for nfs_args->fh */ 12322f172c55SRobert Thurlow esi->esi_path_len = strlen(stubpath) + 1; 12332f172c55SRobert Thurlow if (strcmp(svp->sv_path, "/") != 0) 12342f172c55SRobert Thurlow esi->esi_path_len += strlen(svp->sv_path); 1235b9238976Sth esi->esi_path = kmem_zalloc(esi->esi_path_len, KM_SLEEP); 12362f172c55SRobert Thurlow if (strcmp(svp->sv_path, "/") != 0) 12372f172c55SRobert Thurlow (void) strcat(esi->esi_path, svp->sv_path); 1238b9238976Sth (void) strcat(esi->esi_path, stubpath); 1239b9238976Sth 1240b9238976Sth stubpath -= 1; 1241b9238976Sth /* stubpath allocated by fn_path() */ 1242b9238976Sth kmem_free(stubpath, strlen(stubpath) + 1); 1243b9238976Sth 1244b9238976Sth nfs_rw_exit(&svp->sv_lock); 1245b9238976Sth 1246b9238976Sth return (esi); 1247b9238976Sth } 1248b9238976Sth 12492f172c55SRobert Thurlow /* 12502f172c55SRobert Thurlow * Makes an upcall to NFSMAPID daemon to resolve hostname of NFS server to 12512f172c55SRobert Thurlow * get network information required to do the mount call. 12522f172c55SRobert Thurlow */ 12532f172c55SRobert Thurlow int 12542f172c55SRobert Thurlow nfs4_callmapid(utf8string *server, struct nfs_fsl_info *resp) 12552f172c55SRobert Thurlow { 12562f172c55SRobert Thurlow door_arg_t door_args; 12572f172c55SRobert Thurlow door_handle_t dh; 12582f172c55SRobert Thurlow XDR xdr; 12592f172c55SRobert Thurlow refd_door_args_t *xdr_argsp; 12602f172c55SRobert Thurlow refd_door_res_t *orig_resp; 12612f172c55SRobert Thurlow k_sigset_t smask; 12622f172c55SRobert Thurlow int xdr_len = 0; 12632f172c55SRobert Thurlow int res_len = 16; /* length of an ip adress */ 12642f172c55SRobert Thurlow int orig_reslen = res_len; 12652f172c55SRobert Thurlow int error = 0; 12662f172c55SRobert Thurlow struct nfsidmap_globals *nig; 12672f172c55SRobert Thurlow 12682f172c55SRobert Thurlow if (zone_status_get(curproc->p_zone) >= ZONE_IS_SHUTTING_DOWN) 12692f172c55SRobert Thurlow return (ECONNREFUSED); 12702f172c55SRobert Thurlow 12712f172c55SRobert Thurlow nig = zone_getspecific(nfsidmap_zone_key, nfs_zone()); 12722f172c55SRobert Thurlow ASSERT(nig != NULL); 12732f172c55SRobert Thurlow 12742f172c55SRobert Thurlow mutex_enter(&nig->nfsidmap_daemon_lock); 12752f172c55SRobert Thurlow dh = nig->nfsidmap_daemon_dh; 12762f172c55SRobert Thurlow if (dh == NULL) { 12772f172c55SRobert Thurlow mutex_exit(&nig->nfsidmap_daemon_lock); 12782f172c55SRobert Thurlow cmn_err(CE_NOTE, 12792f172c55SRobert Thurlow "nfs4_callmapid: nfsmapid daemon not " \ 12802f172c55SRobert Thurlow "running unable to resolve host name\n"); 12812f172c55SRobert Thurlow return (EINVAL); 12822f172c55SRobert Thurlow } 12832f172c55SRobert Thurlow door_ki_hold(dh); 12842f172c55SRobert Thurlow mutex_exit(&nig->nfsidmap_daemon_lock); 12852f172c55SRobert Thurlow 12862f172c55SRobert Thurlow xdr_len = xdr_sizeof(&(xdr_utf8string), server); 12872f172c55SRobert Thurlow 12882f172c55SRobert Thurlow xdr_argsp = kmem_zalloc(xdr_len + sizeof (*xdr_argsp), KM_SLEEP); 12892f172c55SRobert Thurlow xdr_argsp->xdr_len = xdr_len; 12902f172c55SRobert Thurlow xdr_argsp->cmd = NFSMAPID_SRV_NETINFO; 12912f172c55SRobert Thurlow 12922f172c55SRobert Thurlow xdrmem_create(&xdr, (char *)&xdr_argsp->xdr_arg, 12932f172c55SRobert Thurlow xdr_len, XDR_ENCODE); 12942f172c55SRobert Thurlow 12952f172c55SRobert Thurlow if (!xdr_utf8string(&xdr, server)) { 12962f172c55SRobert Thurlow kmem_free(xdr_argsp, xdr_len + sizeof (*xdr_argsp)); 12972f172c55SRobert Thurlow door_ki_rele(dh); 12982f172c55SRobert Thurlow return (1); 12992f172c55SRobert Thurlow } 13002f172c55SRobert Thurlow 13012f172c55SRobert Thurlow if (orig_reslen) 13022f172c55SRobert Thurlow orig_resp = kmem_alloc(orig_reslen, KM_SLEEP); 13032f172c55SRobert Thurlow 13042f172c55SRobert Thurlow door_args.data_ptr = (char *)xdr_argsp; 13052f172c55SRobert Thurlow door_args.data_size = sizeof (*xdr_argsp) + xdr_argsp->xdr_len; 13062f172c55SRobert Thurlow door_args.desc_ptr = NULL; 13072f172c55SRobert Thurlow door_args.desc_num = 0; 13082f172c55SRobert Thurlow door_args.rbuf = orig_resp ? (char *)orig_resp : NULL; 13092f172c55SRobert Thurlow door_args.rsize = res_len; 13102f172c55SRobert Thurlow 13112f172c55SRobert Thurlow sigintr(&smask, 1); 13122f172c55SRobert Thurlow error = door_ki_upcall(dh, &door_args); 13132f172c55SRobert Thurlow sigunintr(&smask); 13142f172c55SRobert Thurlow 13152f172c55SRobert Thurlow door_ki_rele(dh); 13162f172c55SRobert Thurlow 13172f172c55SRobert Thurlow kmem_free(xdr_argsp, xdr_len + sizeof (*xdr_argsp)); 13182f172c55SRobert Thurlow if (error) { 13192f172c55SRobert Thurlow kmem_free(orig_resp, orig_reslen); 13202f172c55SRobert Thurlow /* 13212f172c55SRobert Thurlow * There is no door to connect to. The referral daemon 13222f172c55SRobert Thurlow * must not be running yet. 13232f172c55SRobert Thurlow */ 13242f172c55SRobert Thurlow cmn_err(CE_WARN, 13252f172c55SRobert Thurlow "nfsmapid not running cannot resolve host name"); 13262f172c55SRobert Thurlow goto out; 13272f172c55SRobert Thurlow } 13282f172c55SRobert Thurlow 13292f172c55SRobert Thurlow /* 13302f172c55SRobert Thurlow * If the results buffer passed back are not the same as 13312f172c55SRobert Thurlow * what was sent free the old buffer and use the new one. 13322f172c55SRobert Thurlow */ 13332f172c55SRobert Thurlow if (orig_resp && orig_reslen) { 13342f172c55SRobert Thurlow refd_door_res_t *door_resp; 13352f172c55SRobert Thurlow 13362f172c55SRobert Thurlow door_resp = (refd_door_res_t *)door_args.rbuf; 13372f172c55SRobert Thurlow if ((void *)door_args.rbuf != orig_resp) 13382f172c55SRobert Thurlow kmem_free(orig_resp, orig_reslen); 13392f172c55SRobert Thurlow if (door_resp->res_status == 0) { 13402f172c55SRobert Thurlow xdrmem_create(&xdr, (char *)&door_resp->xdr_res, 13412f172c55SRobert Thurlow door_resp->xdr_len, XDR_DECODE); 13422f172c55SRobert Thurlow bzero(resp, sizeof (struct nfs_fsl_info)); 13432f172c55SRobert Thurlow if (!xdr_nfs_fsl_info(&xdr, resp)) { 13442f172c55SRobert Thurlow DTRACE_PROBE2( 13452f172c55SRobert Thurlow nfs4clnt__debug__referral__upcall__xdrfail, 13462f172c55SRobert Thurlow struct nfs_fsl_info *, resp, 13472f172c55SRobert Thurlow char *, "nfs4_callmapid"); 13482f172c55SRobert Thurlow error = EINVAL; 13492f172c55SRobert Thurlow } 13502f172c55SRobert Thurlow } else { 13512f172c55SRobert Thurlow DTRACE_PROBE2( 13522f172c55SRobert Thurlow nfs4clnt__debug__referral__upcall__badstatus, 13532f172c55SRobert Thurlow int, door_resp->res_status, 13542f172c55SRobert Thurlow char *, "nfs4_callmapid"); 13552f172c55SRobert Thurlow error = door_resp->res_status; 13562f172c55SRobert Thurlow } 13572f172c55SRobert Thurlow kmem_free(door_args.rbuf, door_args.rsize); 13582f172c55SRobert Thurlow } 13592f172c55SRobert Thurlow out: 13602f172c55SRobert Thurlow DTRACE_PROBE2(nfs4clnt__func__referral__upcall, 13612f172c55SRobert Thurlow char *, server, int, error); 13622f172c55SRobert Thurlow return (error); 13632f172c55SRobert Thurlow } 13642f172c55SRobert Thurlow 13652f172c55SRobert Thurlow /* 13662f172c55SRobert Thurlow * Fetches the fs_locations attribute. Typically called 13672f172c55SRobert Thurlow * from a Replication/Migration/Referrals/Mirror-mount context 13682f172c55SRobert Thurlow * 13692f172c55SRobert Thurlow * Fills in the attributes in garp. The caller is assumed 13702f172c55SRobert Thurlow * to have allocated memory for garp. 13712f172c55SRobert Thurlow * 13722f172c55SRobert Thurlow * lock: if set do not lock s_recovlock and mi_recovlock mutex, 13732f172c55SRobert Thurlow * it's already done by caller. Otherwise lock these mutexes 13742f172c55SRobert Thurlow * before doing the rfs4call(). 13752f172c55SRobert Thurlow * 13762f172c55SRobert Thurlow * Returns 13772f172c55SRobert Thurlow * 1 for success 13782f172c55SRobert Thurlow * 0 for failure 13792f172c55SRobert Thurlow */ 13802f172c55SRobert Thurlow int 13812f172c55SRobert Thurlow nfs4_fetch_locations(mntinfo4_t *mi, nfs4_sharedfh_t *sfh, char *nm, 13822f172c55SRobert Thurlow cred_t *cr, nfs4_ga_res_t *garp, COMPOUND4res_clnt *callres, bool_t lock) 13832f172c55SRobert Thurlow { 13842f172c55SRobert Thurlow COMPOUND4args_clnt args; 13852f172c55SRobert Thurlow COMPOUND4res_clnt res; 13862f172c55SRobert Thurlow nfs_argop4 *argop; 13872f172c55SRobert Thurlow int argoplist_size = 3 * sizeof (nfs_argop4); 13882f172c55SRobert Thurlow nfs4_server_t *sp = NULL; 13892f172c55SRobert Thurlow int doqueue = 1; 13902f172c55SRobert Thurlow nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS }; 13912f172c55SRobert Thurlow int retval = 1; 13922f172c55SRobert Thurlow struct nfs4_clnt *nfscl; 13932f172c55SRobert Thurlow 13942f172c55SRobert Thurlow if (lock == TRUE) 13952f172c55SRobert Thurlow (void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, 0); 13962f172c55SRobert Thurlow else 13972f172c55SRobert Thurlow ASSERT(nfs_rw_lock_held(&mi->mi_recovlock, RW_READER) || 13982f172c55SRobert Thurlow nfs_rw_lock_held(&mi->mi_recovlock, RW_WRITER)); 13992f172c55SRobert Thurlow 14002f172c55SRobert Thurlow sp = find_nfs4_server(mi); 14012f172c55SRobert Thurlow if (lock == TRUE) 14022f172c55SRobert Thurlow nfs_rw_exit(&mi->mi_recovlock); 14032f172c55SRobert Thurlow 14042f172c55SRobert Thurlow if (sp != NULL) 14052f172c55SRobert Thurlow mutex_exit(&sp->s_lock); 14062f172c55SRobert Thurlow 14072f172c55SRobert Thurlow if (lock == TRUE) { 14082f172c55SRobert Thurlow if (sp != NULL) 14092f172c55SRobert Thurlow (void) nfs_rw_enter_sig(&sp->s_recovlock, 14102f172c55SRobert Thurlow RW_WRITER, 0); 14112f172c55SRobert Thurlow (void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_WRITER, 0); 14122f172c55SRobert Thurlow } else { 14132f172c55SRobert Thurlow if (sp != NULL) { 14142f172c55SRobert Thurlow ASSERT(nfs_rw_lock_held(&sp->s_recovlock, RW_READER) || 14152f172c55SRobert Thurlow nfs_rw_lock_held(&sp->s_recovlock, RW_WRITER)); 14162f172c55SRobert Thurlow } 14172f172c55SRobert Thurlow } 14182f172c55SRobert Thurlow 14192f172c55SRobert Thurlow /* 14202f172c55SRobert Thurlow * Do we want to do the setup for recovery here? 14212f172c55SRobert Thurlow * 14222f172c55SRobert Thurlow * We know that the server responded to a null ping a very 14232f172c55SRobert Thurlow * short time ago, and we know that we intend to do a 14242f172c55SRobert Thurlow * single stateless operation - we want to fetch attributes, 14252f172c55SRobert Thurlow * so we know we can't encounter errors about state. If 14262f172c55SRobert Thurlow * something goes wrong with the GETATTR, like not being 14272f172c55SRobert Thurlow * able to get a response from the server or getting any 14282f172c55SRobert Thurlow * kind of FH error, we should fail the mount. 14292f172c55SRobert Thurlow * 14302f172c55SRobert Thurlow * We may want to re-visited this at a later time. 14312f172c55SRobert Thurlow */ 14322f172c55SRobert Thurlow argop = kmem_alloc(argoplist_size, KM_SLEEP); 14332f172c55SRobert Thurlow 14342f172c55SRobert Thurlow args.ctag = TAG_GETATTR_FSLOCATION; 14352f172c55SRobert Thurlow /* PUTFH LOOKUP GETATTR */ 14362f172c55SRobert Thurlow args.array_len = 3; 14372f172c55SRobert Thurlow args.array = argop; 14382f172c55SRobert Thurlow 14392f172c55SRobert Thurlow /* 0. putfh file */ 14402f172c55SRobert Thurlow argop[0].argop = OP_CPUTFH; 14412f172c55SRobert Thurlow argop[0].nfs_argop4_u.opcputfh.sfh = sfh; 14422f172c55SRobert Thurlow 14432f172c55SRobert Thurlow /* 1. lookup name, can't be dotdot */ 14442f172c55SRobert Thurlow argop[1].argop = OP_CLOOKUP; 14452f172c55SRobert Thurlow argop[1].nfs_argop4_u.opclookup.cname = nm; 14462f172c55SRobert Thurlow 14472f172c55SRobert Thurlow /* 2. file attrs */ 14482f172c55SRobert Thurlow argop[2].argop = OP_GETATTR; 14492f172c55SRobert Thurlow argop[2].nfs_argop4_u.opgetattr.attr_request = 14502f172c55SRobert Thurlow FATTR4_FSID_MASK | FATTR4_FS_LOCATIONS_MASK | 14512f172c55SRobert Thurlow FATTR4_MOUNTED_ON_FILEID_MASK; 14522f172c55SRobert Thurlow argop[2].nfs_argop4_u.opgetattr.mi = mi; 14532f172c55SRobert Thurlow 14542f172c55SRobert Thurlow rfs4call(mi, &args, &res, cr, &doqueue, 0, &e); 14552f172c55SRobert Thurlow 14562f172c55SRobert Thurlow if (lock == TRUE) { 14572f172c55SRobert Thurlow nfs_rw_exit(&mi->mi_recovlock); 14582f172c55SRobert Thurlow if (sp != NULL) 14592f172c55SRobert Thurlow nfs_rw_exit(&sp->s_recovlock); 14602f172c55SRobert Thurlow } 14612f172c55SRobert Thurlow 14622f172c55SRobert Thurlow nfscl = zone_getspecific(nfs4clnt_zone_key, nfs_zone()); 14632f172c55SRobert Thurlow nfscl->nfscl_stat.referrals.value.ui64++; 14642f172c55SRobert Thurlow DTRACE_PROBE3(nfs4clnt__func__referral__fsloc, 14652f172c55SRobert Thurlow nfs4_sharedfh_t *, sfh, char *, nm, nfs4_error_t *, &e); 14662f172c55SRobert Thurlow 14672f172c55SRobert Thurlow if (e.error != 0) { 14682f172c55SRobert Thurlow if (sp != NULL) 14692f172c55SRobert Thurlow nfs4_server_rele(sp); 14702f172c55SRobert Thurlow kmem_free(argop, argoplist_size); 14712f172c55SRobert Thurlow return (0); 14722f172c55SRobert Thurlow } 14732f172c55SRobert Thurlow 14742f172c55SRobert Thurlow /* 14752f172c55SRobert Thurlow * Check for all possible error conditions. 14762f172c55SRobert Thurlow * For valid replies without an ops array or for illegal 14772f172c55SRobert Thurlow * replies, return a failure. 14782f172c55SRobert Thurlow */ 14792f172c55SRobert Thurlow if (res.status != NFS4_OK || res.array_len < 3 || 14802f172c55SRobert Thurlow res.array[2].nfs_resop4_u.opgetattr.status != NFS4_OK) { 14812f172c55SRobert Thurlow retval = 0; 14822f172c55SRobert Thurlow goto exit; 14832f172c55SRobert Thurlow } 14842f172c55SRobert Thurlow 14852f172c55SRobert Thurlow /* 14862f172c55SRobert Thurlow * There isn't much value in putting the attributes 14872f172c55SRobert Thurlow * in the attr cache since fs_locations4 aren't 14882f172c55SRobert Thurlow * encountered very frequently, so just make them 14892f172c55SRobert Thurlow * available to the caller. 14902f172c55SRobert Thurlow */ 14912f172c55SRobert Thurlow *garp = res.array[2].nfs_resop4_u.opgetattr.ga_res; 14922f172c55SRobert Thurlow 14932f172c55SRobert Thurlow DTRACE_PROBE2(nfs4clnt__debug__referral__fsloc, 14942f172c55SRobert Thurlow nfs4_ga_res_t *, garp, char *, "nfs4_fetch_locations"); 14952f172c55SRobert Thurlow 14962f172c55SRobert Thurlow /* No fs_locations? -- return a failure */ 14972f172c55SRobert Thurlow if (garp->n4g_ext_res == NULL || 14982f172c55SRobert Thurlow garp->n4g_ext_res->n4g_fslocations.locations_val == NULL) { 14992f172c55SRobert Thurlow retval = 0; 15002f172c55SRobert Thurlow goto exit; 15012f172c55SRobert Thurlow } 15022f172c55SRobert Thurlow 15032f172c55SRobert Thurlow if (!garp->n4g_fsid_valid) 15042f172c55SRobert Thurlow retval = 0; 15052f172c55SRobert Thurlow 15062f172c55SRobert Thurlow exit: 15072f172c55SRobert Thurlow if (retval == 0) { 15082f172c55SRobert Thurlow /* the call was ok but failed validating the call results */ 1509a17ce845SMarcel Telka xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 15102f172c55SRobert Thurlow } else { 15112f172c55SRobert Thurlow ASSERT(callres != NULL); 15122f172c55SRobert Thurlow *callres = res; 15132f172c55SRobert Thurlow } 15142f172c55SRobert Thurlow 15152f172c55SRobert Thurlow if (sp != NULL) 15162f172c55SRobert Thurlow nfs4_server_rele(sp); 15172f172c55SRobert Thurlow kmem_free(argop, argoplist_size); 15182f172c55SRobert Thurlow return (retval); 15192f172c55SRobert Thurlow } 15202f172c55SRobert Thurlow 15212f172c55SRobert Thurlow /* tunable to disable referral mounts */ 15222f172c55SRobert Thurlow int nfs4_no_referrals = 0; 15232f172c55SRobert Thurlow 15242f172c55SRobert Thurlow /* 15252f172c55SRobert Thurlow * Returns NULL if the vnode cannot be created or found. 15262f172c55SRobert Thurlow */ 15272f172c55SRobert Thurlow vnode_t * 15282f172c55SRobert Thurlow find_referral_stubvp(vnode_t *dvp, char *nm, cred_t *cr) 15292f172c55SRobert Thurlow { 15302f172c55SRobert Thurlow nfs_fh4 *stub_fh, *dfh; 15312f172c55SRobert Thurlow nfs4_sharedfh_t *sfhp; 15322f172c55SRobert Thurlow char *newfhval; 15332f172c55SRobert Thurlow vnode_t *vp = NULL; 15342f172c55SRobert Thurlow fattr4_mounted_on_fileid mnt_on_fileid; 15352f172c55SRobert Thurlow nfs4_ga_res_t garp; 15362f172c55SRobert Thurlow mntinfo4_t *mi; 15372f172c55SRobert Thurlow COMPOUND4res_clnt callres; 15382f172c55SRobert Thurlow hrtime_t t; 15392f172c55SRobert Thurlow 15402f172c55SRobert Thurlow if (nfs4_no_referrals) 15412f172c55SRobert Thurlow return (NULL); 15422f172c55SRobert Thurlow 15432f172c55SRobert Thurlow /* 15442f172c55SRobert Thurlow * Get the mounted_on_fileid, unique on that server::fsid 15452f172c55SRobert Thurlow */ 15462f172c55SRobert Thurlow mi = VTOMI4(dvp); 15472f172c55SRobert Thurlow if (nfs4_fetch_locations(mi, VTOR4(dvp)->r_fh, nm, cr, 15482f172c55SRobert Thurlow &garp, &callres, FALSE) == 0) 15492f172c55SRobert Thurlow return (NULL); 15502f172c55SRobert Thurlow mnt_on_fileid = garp.n4g_mon_fid; 1551a17ce845SMarcel Telka xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&callres); 15522f172c55SRobert Thurlow 15532f172c55SRobert Thurlow /* 15542f172c55SRobert Thurlow * Build a fake filehandle from the dir FH and the mounted_on_fileid 15552f172c55SRobert Thurlow */ 15562f172c55SRobert Thurlow dfh = &VTOR4(dvp)->r_fh->sfh_fh; 15572f172c55SRobert Thurlow stub_fh = kmem_alloc(sizeof (nfs_fh4), KM_SLEEP); 15582f172c55SRobert Thurlow stub_fh->nfs_fh4_val = kmem_alloc(dfh->nfs_fh4_len + 15592f172c55SRobert Thurlow sizeof (fattr4_mounted_on_fileid), KM_SLEEP); 15602f172c55SRobert Thurlow newfhval = stub_fh->nfs_fh4_val; 15612f172c55SRobert Thurlow 15622f172c55SRobert Thurlow /* copy directory's file handle */ 15632f172c55SRobert Thurlow bcopy(dfh->nfs_fh4_val, newfhval, dfh->nfs_fh4_len); 15642f172c55SRobert Thurlow stub_fh->nfs_fh4_len = dfh->nfs_fh4_len; 15652f172c55SRobert Thurlow newfhval = newfhval + dfh->nfs_fh4_len; 15662f172c55SRobert Thurlow 15672f172c55SRobert Thurlow /* Add mounted_on_fileid. Use bcopy to avoid alignment problem */ 15682f172c55SRobert Thurlow bcopy((char *)&mnt_on_fileid, newfhval, 15692f172c55SRobert Thurlow sizeof (fattr4_mounted_on_fileid)); 15702f172c55SRobert Thurlow stub_fh->nfs_fh4_len += sizeof (fattr4_mounted_on_fileid); 15712f172c55SRobert Thurlow 15722f172c55SRobert Thurlow sfhp = sfh4_put(stub_fh, VTOMI4(dvp), NULL); 15732f172c55SRobert Thurlow kmem_free(stub_fh->nfs_fh4_val, dfh->nfs_fh4_len + 15742f172c55SRobert Thurlow sizeof (fattr4_mounted_on_fileid)); 15752f172c55SRobert Thurlow kmem_free(stub_fh, sizeof (nfs_fh4)); 15762f172c55SRobert Thurlow if (sfhp == NULL) 15772f172c55SRobert Thurlow return (NULL); 15782f172c55SRobert Thurlow 15792f172c55SRobert Thurlow t = gethrtime(); 15802f172c55SRobert Thurlow garp.n4g_va.va_type = VDIR; 15812f172c55SRobert Thurlow vp = makenfs4node(sfhp, NULL, dvp->v_vfsp, t, 15822f172c55SRobert Thurlow cr, dvp, fn_get(VTOSV(dvp)->sv_name, nm, sfhp)); 15832f172c55SRobert Thurlow 15842f172c55SRobert Thurlow if (vp != NULL) 15852f172c55SRobert Thurlow vp->v_type = VDIR; 15862f172c55SRobert Thurlow 15872f172c55SRobert Thurlow sfh4_rele(&sfhp); 15882f172c55SRobert Thurlow return (vp); 15892f172c55SRobert Thurlow } 15902f172c55SRobert Thurlow 15912f172c55SRobert Thurlow int 15922f172c55SRobert Thurlow nfs4_setup_referral(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr) 15932f172c55SRobert Thurlow { 15942f172c55SRobert Thurlow vnode_t *nvp; 15952f172c55SRobert Thurlow rnode4_t *rp; 15962f172c55SRobert Thurlow 15972f172c55SRobert Thurlow if ((nvp = find_referral_stubvp(dvp, nm, cr)) == NULL) 15982f172c55SRobert Thurlow return (EINVAL); 15992f172c55SRobert Thurlow 16002f172c55SRobert Thurlow rp = VTOR4(nvp); 16012f172c55SRobert Thurlow mutex_enter(&rp->r_statelock); 16022f172c55SRobert Thurlow r4_stub_referral(rp); 16032f172c55SRobert Thurlow mutex_exit(&rp->r_statelock); 16042f172c55SRobert Thurlow dnlc_enter(dvp, nm, nvp); 16052f172c55SRobert Thurlow 16062f172c55SRobert Thurlow if (*vpp != NULL) 16072f172c55SRobert Thurlow VN_RELE(*vpp); /* no longer need this vnode */ 16082f172c55SRobert Thurlow 16092f172c55SRobert Thurlow *vpp = nvp; 16102f172c55SRobert Thurlow 16112f172c55SRobert Thurlow return (0); 16122f172c55SRobert Thurlow } 16132f172c55SRobert Thurlow 16142f172c55SRobert Thurlow /* 16152f172c55SRobert Thurlow * Fetch the location information and resolve the new server. 16162f172c55SRobert Thurlow * Caller needs to free up the XDR data which is returned. 16172f172c55SRobert Thurlow * Input: mount info, shared filehandle, nodename 16182f172c55SRobert Thurlow * Return: Index to the result or Error(-1) 16192f172c55SRobert Thurlow * Output: FsLocations Info, Resolved Server Info. 16202f172c55SRobert Thurlow */ 16212f172c55SRobert Thurlow int 16222f172c55SRobert Thurlow nfs4_process_referral(mntinfo4_t *mi, nfs4_sharedfh_t *sfh, 16232f172c55SRobert Thurlow char *nm, cred_t *cr, nfs4_ga_res_t *grp, COMPOUND4res_clnt *res, 16242f172c55SRobert Thurlow struct nfs_fsl_info *fsloc) 16252f172c55SRobert Thurlow { 16262f172c55SRobert Thurlow fs_location4 *fsp; 16272f172c55SRobert Thurlow struct nfs_fsl_info nfsfsloc; 16282f172c55SRobert Thurlow int ret, i, error; 16292f172c55SRobert Thurlow nfs4_ga_res_t garp; 16302f172c55SRobert Thurlow COMPOUND4res_clnt callres; 16312f172c55SRobert Thurlow struct knetconfig *knc; 16322f172c55SRobert Thurlow 16332f172c55SRobert Thurlow ret = nfs4_fetch_locations(mi, sfh, nm, cr, &garp, &callres, TRUE); 16342f172c55SRobert Thurlow if (ret == 0) 16352f172c55SRobert Thurlow return (-1); 16362f172c55SRobert Thurlow 16372f172c55SRobert Thurlow /* 16382f172c55SRobert Thurlow * As a lame attempt to figuring out if we're 16392f172c55SRobert Thurlow * handling a migration event or a referral, 16402f172c55SRobert Thurlow * look for rnodes with this fsid in the rnode 16412f172c55SRobert Thurlow * cache. 16422f172c55SRobert Thurlow * 16432f172c55SRobert Thurlow * If we can find one or more such rnodes, it 16442f172c55SRobert Thurlow * means we're handling a migration event and 16452f172c55SRobert Thurlow * we want to bail out in that case. 16462f172c55SRobert Thurlow */ 16472f172c55SRobert Thurlow if (r4find_by_fsid(mi, &garp.n4g_fsid)) { 16482f172c55SRobert Thurlow DTRACE_PROBE3(nfs4clnt__debug__referral__migration, 16492f172c55SRobert Thurlow mntinfo4_t *, mi, nfs4_ga_res_t *, &garp, 16502f172c55SRobert Thurlow char *, "nfs4_process_referral"); 1651a17ce845SMarcel Telka xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&callres); 16522f172c55SRobert Thurlow return (-1); 16532f172c55SRobert Thurlow } 16542f172c55SRobert Thurlow 16552f172c55SRobert Thurlow /* 16562f172c55SRobert Thurlow * Find the first responsive server to mount. When we find 16572f172c55SRobert Thurlow * one, fsp will point to it. 16582f172c55SRobert Thurlow */ 16592f172c55SRobert Thurlow for (i = 0; i < garp.n4g_ext_res->n4g_fslocations.locations_len; i++) { 16602f172c55SRobert Thurlow 16612f172c55SRobert Thurlow fsp = &garp.n4g_ext_res->n4g_fslocations.locations_val[i]; 16622f172c55SRobert Thurlow if (fsp->server_len == 0 || fsp->server_val == NULL) 16632f172c55SRobert Thurlow continue; 16642f172c55SRobert Thurlow 16652f172c55SRobert Thurlow error = nfs4_callmapid(fsp->server_val, &nfsfsloc); 16662f172c55SRobert Thurlow if (error != 0) 16672f172c55SRobert Thurlow continue; 16682f172c55SRobert Thurlow 16692f172c55SRobert Thurlow error = nfs4_ping_server_common(nfsfsloc.knconf, 16702f172c55SRobert Thurlow nfsfsloc.addr, !(mi->mi_flags & MI4_INT)); 16712f172c55SRobert Thurlow if (error == RPC_SUCCESS) 16722f172c55SRobert Thurlow break; 16732f172c55SRobert Thurlow 16742f172c55SRobert Thurlow DTRACE_PROBE2(nfs4clnt__debug__referral__srvaddr, 16752f172c55SRobert Thurlow sockaddr_in *, (struct sockaddr_in *)nfsfsloc.addr->buf, 16762f172c55SRobert Thurlow char *, "nfs4_process_referral"); 16772f172c55SRobert Thurlow 1678a17ce845SMarcel Telka xdr_free(xdr_nfs_fsl_info, (char *)&nfsfsloc); 16792f172c55SRobert Thurlow } 16802f172c55SRobert Thurlow knc = nfsfsloc.knconf; 16812f172c55SRobert Thurlow if ((i >= garp.n4g_ext_res->n4g_fslocations.locations_len) || 16822f172c55SRobert Thurlow (knc->knc_protofmly == NULL) || (knc->knc_proto == NULL)) { 16832f172c55SRobert Thurlow DTRACE_PROBE2(nfs4clnt__debug__referral__nofsloc, 16842f172c55SRobert Thurlow nfs4_ga_res_t *, &garp, char *, "nfs4_process_referral"); 1685a17ce845SMarcel Telka xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&callres); 16862f172c55SRobert Thurlow return (-1); 16872f172c55SRobert Thurlow } 16882f172c55SRobert Thurlow 16892f172c55SRobert Thurlow /* Send the results back */ 16902f172c55SRobert Thurlow *fsloc = nfsfsloc; 16912f172c55SRobert Thurlow *grp = garp; 16922f172c55SRobert Thurlow *res = callres; 16932f172c55SRobert Thurlow return (i); 16942f172c55SRobert Thurlow } 16952f172c55SRobert Thurlow 16962f172c55SRobert Thurlow /* 16972f172c55SRobert Thurlow * Referrals case - need to fetch referral data and then upcall to 16982f172c55SRobert Thurlow * user-level to get complete mount data. 16992f172c55SRobert Thurlow */ 17002f172c55SRobert Thurlow static ephemeral_servinfo_t * 17012f172c55SRobert Thurlow nfs4_trigger_esi_create_referral(vnode_t *vp, cred_t *cr) 17022f172c55SRobert Thurlow { 17032f172c55SRobert Thurlow struct knetconfig *sikncp, *svkncp; 17042f172c55SRobert Thurlow struct netbuf *bufp; 17052f172c55SRobert Thurlow ephemeral_servinfo_t *esi; 17062f172c55SRobert Thurlow vnode_t *dvp; 17072f172c55SRobert Thurlow rnode4_t *drp; 17082f172c55SRobert Thurlow fs_location4 *fsp; 17092f172c55SRobert Thurlow struct nfs_fsl_info nfsfsloc; 17102f172c55SRobert Thurlow nfs4_ga_res_t garp; 17112f172c55SRobert Thurlow char *p; 17122f172c55SRobert Thurlow char fn[MAXNAMELEN]; 17132f172c55SRobert Thurlow int i, index = -1; 17142f172c55SRobert Thurlow mntinfo4_t *mi; 17152f172c55SRobert Thurlow COMPOUND4res_clnt callres; 17162f172c55SRobert Thurlow 17172f172c55SRobert Thurlow /* 17182f172c55SRobert Thurlow * If we're passed in a stub vnode that 17192f172c55SRobert Thurlow * isn't a "referral" stub, bail out 17202f172c55SRobert Thurlow * and return a failure 17212f172c55SRobert Thurlow */ 17222f172c55SRobert Thurlow if (!RP_ISSTUB_REFERRAL(VTOR4(vp))) 17232f172c55SRobert Thurlow return (NULL); 17242f172c55SRobert Thurlow 17252f172c55SRobert Thurlow if (vtodv(vp, &dvp, CRED(), TRUE) != 0) 17262f172c55SRobert Thurlow return (NULL); 17272f172c55SRobert Thurlow 17282f172c55SRobert Thurlow drp = VTOR4(dvp); 17292f172c55SRobert Thurlow if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR4(dvp))) { 17302f172c55SRobert Thurlow VN_RELE(dvp); 17312f172c55SRobert Thurlow return (NULL); 17322f172c55SRobert Thurlow } 17332f172c55SRobert Thurlow 17342f172c55SRobert Thurlow if (vtoname(vp, fn, MAXNAMELEN) != 0) { 17352f172c55SRobert Thurlow nfs_rw_exit(&drp->r_rwlock); 17362f172c55SRobert Thurlow VN_RELE(dvp); 17372f172c55SRobert Thurlow return (NULL); 17382f172c55SRobert Thurlow } 17392f172c55SRobert Thurlow 17402f172c55SRobert Thurlow mi = VTOMI4(dvp); 17412f172c55SRobert Thurlow index = nfs4_process_referral(mi, drp->r_fh, fn, cr, 17422f172c55SRobert Thurlow &garp, &callres, &nfsfsloc); 17432f172c55SRobert Thurlow nfs_rw_exit(&drp->r_rwlock); 17442f172c55SRobert Thurlow VN_RELE(dvp); 17452f172c55SRobert Thurlow if (index < 0) 17462f172c55SRobert Thurlow return (NULL); 17472f172c55SRobert Thurlow 17482f172c55SRobert Thurlow fsp = &garp.n4g_ext_res->n4g_fslocations.locations_val[index]; 17492f172c55SRobert Thurlow esi = kmem_zalloc(sizeof (ephemeral_servinfo_t), KM_SLEEP); 17502f172c55SRobert Thurlow 17512f172c55SRobert Thurlow /* initially set to be our type of ephemeral mount; may be added to */ 17522f172c55SRobert Thurlow esi->esi_mount_flags = NFSMNT_REFERRAL; 17532f172c55SRobert Thurlow 17542f172c55SRobert Thurlow esi->esi_hostname = 17552f172c55SRobert Thurlow kmem_zalloc(fsp->server_val->utf8string_len + 1, KM_SLEEP); 17562f172c55SRobert Thurlow bcopy(fsp->server_val->utf8string_val, esi->esi_hostname, 17572f172c55SRobert Thurlow fsp->server_val->utf8string_len); 17582f172c55SRobert Thurlow esi->esi_hostname[fsp->server_val->utf8string_len] = '\0'; 17592f172c55SRobert Thurlow 17602f172c55SRobert Thurlow bufp = kmem_alloc(sizeof (struct netbuf), KM_SLEEP); 17612f172c55SRobert Thurlow bufp->len = nfsfsloc.addr->len; 17622f172c55SRobert Thurlow bufp->maxlen = nfsfsloc.addr->maxlen; 17632f172c55SRobert Thurlow bufp->buf = kmem_zalloc(bufp->len, KM_SLEEP); 17642f172c55SRobert Thurlow bcopy(nfsfsloc.addr->buf, bufp->buf, bufp->len); 17652f172c55SRobert Thurlow esi->esi_addr = bufp; 17662f172c55SRobert Thurlow 17672f172c55SRobert Thurlow esi->esi_knconf = kmem_zalloc(sizeof (*esi->esi_knconf), KM_SLEEP); 17682f172c55SRobert Thurlow sikncp = esi->esi_knconf; 17692f172c55SRobert Thurlow 17702f172c55SRobert Thurlow DTRACE_PROBE2(nfs4clnt__debug__referral__nfsfsloc, 17712f172c55SRobert Thurlow struct nfs_fsl_info *, &nfsfsloc, 17722f172c55SRobert Thurlow char *, "nfs4_trigger_esi_create_referral"); 17732f172c55SRobert Thurlow 17742f172c55SRobert Thurlow svkncp = nfsfsloc.knconf; 17752f172c55SRobert Thurlow sikncp->knc_semantics = svkncp->knc_semantics; 17762f172c55SRobert Thurlow sikncp->knc_protofmly = (caddr_t)kmem_zalloc(KNC_STRSIZE, KM_SLEEP); 17772f172c55SRobert Thurlow (void) strlcat((char *)sikncp->knc_protofmly, 17782f172c55SRobert Thurlow (char *)svkncp->knc_protofmly, KNC_STRSIZE); 17792f172c55SRobert Thurlow sikncp->knc_proto = (caddr_t)kmem_zalloc(KNC_STRSIZE, KM_SLEEP); 17802f172c55SRobert Thurlow (void) strlcat((char *)sikncp->knc_proto, (char *)svkncp->knc_proto, 17812f172c55SRobert Thurlow KNC_STRSIZE); 17822f172c55SRobert Thurlow sikncp->knc_rdev = svkncp->knc_rdev; 17832f172c55SRobert Thurlow 17842f172c55SRobert Thurlow DTRACE_PROBE2(nfs4clnt__debug__referral__knetconf, 17852f172c55SRobert Thurlow struct knetconfig *, sikncp, 17862f172c55SRobert Thurlow char *, "nfs4_trigger_esi_create_referral"); 17872f172c55SRobert Thurlow 17882f172c55SRobert Thurlow esi->esi_netname = kmem_zalloc(nfsfsloc.netnm_len, KM_SLEEP); 17892f172c55SRobert Thurlow bcopy(nfsfsloc.netname, esi->esi_netname, nfsfsloc.netnm_len); 17902f172c55SRobert Thurlow esi->esi_syncaddr = NULL; 17912f172c55SRobert Thurlow 17922f172c55SRobert Thurlow esi->esi_path = p = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 17932f172c55SRobert Thurlow esi->esi_path_len = MAXPATHLEN; 17942f172c55SRobert Thurlow *p++ = '/'; 17952f172c55SRobert Thurlow for (i = 0; i < fsp->rootpath.pathname4_len; i++) { 17962f172c55SRobert Thurlow component4 *comp; 17972f172c55SRobert Thurlow 17982f172c55SRobert Thurlow comp = &fsp->rootpath.pathname4_val[i]; 17992f172c55SRobert Thurlow /* If no space, null the string and bail */ 18002f172c55SRobert Thurlow if ((p - esi->esi_path) + comp->utf8string_len + 1 > MAXPATHLEN) 18012f172c55SRobert Thurlow goto err; 18022f172c55SRobert Thurlow bcopy(comp->utf8string_val, p, comp->utf8string_len); 18032f172c55SRobert Thurlow p += comp->utf8string_len; 18042f172c55SRobert Thurlow *p++ = '/'; 18052f172c55SRobert Thurlow } 18062f172c55SRobert Thurlow if (fsp->rootpath.pathname4_len != 0) 18072f172c55SRobert Thurlow *(p - 1) = '\0'; 18082f172c55SRobert Thurlow else 18092f172c55SRobert Thurlow *p = '\0'; 18102f172c55SRobert Thurlow p = esi->esi_path; 18112f172c55SRobert Thurlow esi->esi_path = strdup(p); 18122f172c55SRobert Thurlow esi->esi_path_len = strlen(p) + 1; 18132f172c55SRobert Thurlow kmem_free(p, MAXPATHLEN); 18142f172c55SRobert Thurlow 18152f172c55SRobert Thurlow /* Allocated in nfs4_process_referral() */ 1816a17ce845SMarcel Telka xdr_free(xdr_nfs_fsl_info, (char *)&nfsfsloc); 1817a17ce845SMarcel Telka xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&callres); 18182f172c55SRobert Thurlow 18192f172c55SRobert Thurlow return (esi); 18202f172c55SRobert Thurlow err: 18212f172c55SRobert Thurlow kmem_free(esi->esi_path, esi->esi_path_len); 18222f172c55SRobert Thurlow kmem_free(esi->esi_hostname, fsp->server_val->utf8string_len + 1); 18232f172c55SRobert Thurlow kmem_free(esi->esi_addr->buf, esi->esi_addr->len); 18242f172c55SRobert Thurlow kmem_free(esi->esi_addr, sizeof (struct netbuf)); 18252f172c55SRobert Thurlow kmem_free(esi->esi_knconf->knc_protofmly, KNC_STRSIZE); 18262f172c55SRobert Thurlow kmem_free(esi->esi_knconf->knc_proto, KNC_STRSIZE); 18272f172c55SRobert Thurlow kmem_free(esi->esi_knconf, sizeof (*esi->esi_knconf)); 18282f172c55SRobert Thurlow kmem_free(esi->esi_netname, nfsfsloc.netnm_len); 18292f172c55SRobert Thurlow kmem_free(esi, sizeof (ephemeral_servinfo_t)); 1830a17ce845SMarcel Telka xdr_free(xdr_nfs_fsl_info, (char *)&nfsfsloc); 1831a17ce845SMarcel Telka xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&callres); 18322f172c55SRobert Thurlow return (NULL); 18332f172c55SRobert Thurlow } 18342f172c55SRobert Thurlow 1835b9238976Sth /* 1836b9238976Sth * Assemble the args, and call the generic VFS mount function to 1837b9238976Sth * finally perform the ephemeral mount. 1838b9238976Sth */ 1839b9238976Sth static int 1840b9238976Sth nfs4_trigger_domount(vnode_t *stubvp, domount_args_t *dma, vfs_t **vfsp, 18416962f5b8SThomas Haynes cred_t *cr, vnode_t **newvpp) 1842b9238976Sth { 1843b9238976Sth struct mounta *uap; 1844b9238976Sth char *mntpt, *orig_path, *path; 1845b9238976Sth const char *orig_mntpt; 1846b9238976Sth int retval; 1847b9238976Sth int mntpt_len; 1848b9238976Sth int spec_len; 1849b9238976Sth zone_t *zone = curproc->p_zone; 1850b9238976Sth bool_t has_leading_slash; 18516962f5b8SThomas Haynes int i; 1852b9238976Sth 1853b9238976Sth vfs_t *stubvfsp = stubvp->v_vfsp; 1854b9238976Sth ephemeral_servinfo_t *esi = dma->dma_esi; 1855b9238976Sth struct nfs_args *nargs = dma->dma_nargs; 1856b9238976Sth 1857b9238976Sth /* first, construct the mount point for the ephemeral mount */ 1858b9238976Sth orig_path = path = fn_path(VTOSV(stubvp)->sv_name); 1859b9238976Sth orig_mntpt = (char *)refstr_value(stubvfsp->vfs_mntpt); 1860b9238976Sth 1861b9238976Sth if (*orig_path == '.') 1862b9238976Sth orig_path++; 1863b9238976Sth 1864b9238976Sth /* 1865b9238976Sth * Get rid of zone's root path 1866b9238976Sth */ 1867b9238976Sth if (zone != global_zone) { 1868b9238976Sth /* 1869b9238976Sth * -1 for trailing '/' and -1 for EOS. 1870b9238976Sth */ 1871b9238976Sth if (strncmp(zone->zone_rootpath, orig_mntpt, 1872b9238976Sth zone->zone_rootpathlen - 1) == 0) { 1873b9238976Sth orig_mntpt += (zone->zone_rootpathlen - 2); 1874b9238976Sth } 1875b9238976Sth } 1876b9238976Sth 1877b9238976Sth mntpt_len = strlen(orig_mntpt) + strlen(orig_path); 1878b9238976Sth mntpt = kmem_zalloc(mntpt_len + 1, KM_SLEEP); 1879b9238976Sth (void) strcat(mntpt, orig_mntpt); 1880b9238976Sth (void) strcat(mntpt, orig_path); 1881b9238976Sth 1882b9238976Sth kmem_free(path, strlen(path) + 1); 1883b9238976Sth path = esi->esi_path; 1884b9238976Sth if (*path == '.') 1885b9238976Sth path++; 1886b9238976Sth if (path[0] == '/' && path[1] == '/') 1887b9238976Sth path++; 1888b9238976Sth has_leading_slash = (*path == '/'); 1889b9238976Sth 1890b9238976Sth spec_len = strlen(dma->dma_hostlist); 1891b9238976Sth spec_len += strlen(path); 1892b9238976Sth 1893b9238976Sth /* We are going to have to add this in */ 1894b9238976Sth if (!has_leading_slash) 1895b9238976Sth spec_len++; 1896b9238976Sth 1897b9238976Sth /* We need to get the ':' for dma_hostlist:esi_path */ 1898b9238976Sth spec_len++; 1899b9238976Sth 1900b9238976Sth uap = kmem_zalloc(sizeof (struct mounta), KM_SLEEP); 1901b9238976Sth uap->spec = kmem_zalloc(spec_len + 1, KM_SLEEP); 1902b9238976Sth (void) snprintf(uap->spec, spec_len + 1, "%s:%s%s", dma->dma_hostlist, 1903b9238976Sth has_leading_slash ? "" : "/", path); 1904b9238976Sth 1905b9238976Sth uap->dir = mntpt; 1906b9238976Sth 1907b9238976Sth uap->flags = MS_SYSSPACE | MS_DATA; 1908b9238976Sth /* fstype-independent mount options not covered elsewhere */ 1909b9238976Sth /* copy parent's mount(1M) "-m" flag */ 1910b9238976Sth if (stubvfsp->vfs_flag & VFS_NOMNTTAB) 1911b9238976Sth uap->flags |= MS_NOMNTTAB; 1912b9238976Sth 1913b9238976Sth uap->fstype = MNTTYPE_NFS4; 1914b9238976Sth uap->dataptr = (char *)nargs; 1915b9238976Sth /* not needed for MS_SYSSPACE */ 1916b9238976Sth uap->datalen = 0; 1917b9238976Sth 1918b9238976Sth /* use optptr to pass in extra mount options */ 1919b9238976Sth uap->flags |= MS_OPTIONSTR; 1920b9238976Sth uap->optptr = nfs4_trigger_create_mntopts(stubvfsp); 1921b9238976Sth if (uap->optptr == NULL) { 1922b9238976Sth retval = EINVAL; 1923b9238976Sth goto done; 1924b9238976Sth } 1925546a3997SThomas Haynes 1926b9238976Sth /* domount() expects us to count the trailing NUL */ 1927b9238976Sth uap->optlen = strlen(uap->optptr) + 1; 1928b9238976Sth 19296962f5b8SThomas Haynes /* 19306962f5b8SThomas Haynes * If we get EBUSY, we try again once to see if we can perform 19316962f5b8SThomas Haynes * the mount. We do this because of a spurious race condition. 19326962f5b8SThomas Haynes */ 19336962f5b8SThomas Haynes for (i = 0; i < 2; i++) { 19346962f5b8SThomas Haynes int error; 19356962f5b8SThomas Haynes bool_t was_mounted; 19366962f5b8SThomas Haynes 19376962f5b8SThomas Haynes retval = domount(NULL, uap, stubvp, cr, vfsp); 19386962f5b8SThomas Haynes if (retval == 0) { 19396962f5b8SThomas Haynes retval = VFS_ROOT(*vfsp, newvpp); 19406962f5b8SThomas Haynes VFS_RELE(*vfsp); 19416962f5b8SThomas Haynes break; 19426962f5b8SThomas Haynes } else if (retval != EBUSY) { 19436962f5b8SThomas Haynes break; 19446962f5b8SThomas Haynes } 19456962f5b8SThomas Haynes 19466962f5b8SThomas Haynes /* 19476962f5b8SThomas Haynes * We might find it mounted by the other racer... 19486962f5b8SThomas Haynes */ 19496962f5b8SThomas Haynes error = nfs4_trigger_mounted_already(stubvp, 19506962f5b8SThomas Haynes newvpp, &was_mounted, vfsp); 19516962f5b8SThomas Haynes if (error) { 19526962f5b8SThomas Haynes goto done; 19536962f5b8SThomas Haynes } else if (was_mounted) { 19546962f5b8SThomas Haynes retval = 0; 19556962f5b8SThomas Haynes break; 19566962f5b8SThomas Haynes } 19576962f5b8SThomas Haynes } 1958546a3997SThomas Haynes 1959b9238976Sth done: 1960b9238976Sth if (uap->optptr) 1961b9238976Sth nfs4_trigger_destroy_mntopts(uap->optptr); 1962b9238976Sth 1963b9238976Sth kmem_free(uap->spec, spec_len + 1); 1964b9238976Sth kmem_free(uap, sizeof (struct mounta)); 1965b9238976Sth kmem_free(mntpt, mntpt_len + 1); 1966b9238976Sth 1967b9238976Sth return (retval); 1968b9238976Sth } 1969b9238976Sth 1970b9238976Sth /* 1971b9238976Sth * Build an nfs_args structure for passing to domount(). 1972b9238976Sth * 1973b9238976Sth * Ephemeral mount-type specific data comes from the ephemeral_servinfo_t; 1974b9238976Sth * generic data - common to all ephemeral mount types - is read directly 1975b9238976Sth * from the parent mount's servinfo4_t and mntinfo4_t, via the stub vnode. 1976b9238976Sth */ 1977b9238976Sth static struct nfs_args * 1978b9238976Sth nfs4_trigger_nargs_create(mntinfo4_t *mi, servinfo4_t *svp, 1979b9238976Sth ephemeral_servinfo_t *esi) 1980b9238976Sth { 1981b9238976Sth sec_data_t *secdata; 1982b9238976Sth struct nfs_args *nargs; 1983b9238976Sth 1984b9238976Sth /* setup the nfs args */ 1985b9238976Sth nargs = kmem_zalloc(sizeof (struct nfs_args), KM_SLEEP); 1986b9238976Sth 1987b9238976Sth (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1988b9238976Sth 1989b9238976Sth nargs->addr = esi->esi_addr; 1990b9238976Sth 1991b9238976Sth /* for AUTH_DH by negotiation */ 1992b9238976Sth if (esi->esi_syncaddr || esi->esi_netname) { 1993b9238976Sth nargs->flags |= NFSMNT_SECURE; 1994b9238976Sth nargs->syncaddr = esi->esi_syncaddr; 1995b9238976Sth nargs->netname = esi->esi_netname; 1996b9238976Sth } 1997b9238976Sth 1998b9238976Sth nargs->flags |= NFSMNT_KNCONF; 1999b9238976Sth nargs->knconf = esi->esi_knconf; 2000b9238976Sth nargs->flags |= NFSMNT_HOSTNAME; 2001b9238976Sth nargs->hostname = esi->esi_hostname; 2002b9238976Sth nargs->fh = esi->esi_path; 2003b9238976Sth 2004b9238976Sth /* general mount settings, all copied from parent mount */ 2005b9238976Sth mutex_enter(&mi->mi_lock); 2006b9238976Sth 2007b9238976Sth if (!(mi->mi_flags & MI4_HARD)) 2008b9238976Sth nargs->flags |= NFSMNT_SOFT; 2009b9238976Sth 2010b9238976Sth nargs->flags |= NFSMNT_WSIZE | NFSMNT_RSIZE | NFSMNT_TIMEO | 2011b9238976Sth NFSMNT_RETRANS; 2012b9238976Sth nargs->wsize = mi->mi_stsize; 2013b9238976Sth nargs->rsize = mi->mi_tsize; 2014b9238976Sth nargs->timeo = mi->mi_timeo; 2015b9238976Sth nargs->retrans = mi->mi_retrans; 2016b9238976Sth 2017b9238976Sth if (mi->mi_flags & MI4_INT) 2018b9238976Sth nargs->flags |= NFSMNT_INT; 2019b9238976Sth if (mi->mi_flags & MI4_NOAC) 2020b9238976Sth nargs->flags |= NFSMNT_NOAC; 2021b9238976Sth 2022b9238976Sth nargs->flags |= NFSMNT_ACREGMIN | NFSMNT_ACREGMAX | NFSMNT_ACDIRMIN | 2023b9238976Sth NFSMNT_ACDIRMAX; 2024b9238976Sth nargs->acregmin = HR2SEC(mi->mi_acregmin); 2025b9238976Sth nargs->acregmax = HR2SEC(mi->mi_acregmax); 2026b9238976Sth nargs->acdirmin = HR2SEC(mi->mi_acdirmin); 2027b9238976Sth nargs->acdirmax = HR2SEC(mi->mi_acdirmax); 2028b9238976Sth 20292f172c55SRobert Thurlow /* add any specific flags for this type of ephemeral mount */ 20302f172c55SRobert Thurlow nargs->flags |= esi->esi_mount_flags; 20312f172c55SRobert Thurlow 2032b9238976Sth if (mi->mi_flags & MI4_NOCTO) 2033b9238976Sth nargs->flags |= NFSMNT_NOCTO; 2034b9238976Sth if (mi->mi_flags & MI4_GRPID) 2035b9238976Sth nargs->flags |= NFSMNT_GRPID; 2036b9238976Sth if (mi->mi_flags & MI4_LLOCK) 2037b9238976Sth nargs->flags |= NFSMNT_LLOCK; 2038b9238976Sth if (mi->mi_flags & MI4_NOPRINT) 2039b9238976Sth nargs->flags |= NFSMNT_NOPRINT; 2040b9238976Sth if (mi->mi_flags & MI4_DIRECTIO) 2041b9238976Sth nargs->flags |= NFSMNT_DIRECTIO; 20422f172c55SRobert Thurlow if (mi->mi_flags & MI4_PUBLIC && nargs->flags & NFSMNT_MIRRORMOUNT) 2043b9238976Sth nargs->flags |= NFSMNT_PUBLIC; 2044b9238976Sth 20452f172c55SRobert Thurlow /* Do some referral-specific option tweaking */ 20462f172c55SRobert Thurlow if (nargs->flags & NFSMNT_REFERRAL) { 20472f172c55SRobert Thurlow nargs->flags &= ~NFSMNT_DORDMA; 20482f172c55SRobert Thurlow nargs->flags |= NFSMNT_TRYRDMA; 20492f172c55SRobert Thurlow } 2050b9238976Sth 20512f172c55SRobert Thurlow mutex_exit(&mi->mi_lock); 2052b9238976Sth 2053b9238976Sth /* 2054b9238976Sth * Security data & negotiation policy. 2055b9238976Sth * 20562f172c55SRobert Thurlow * For mirror mounts, we need to preserve the parent mount's 20572f172c55SRobert Thurlow * preference for security negotiation, translating SV4_TRYSECDEFAULT 20582f172c55SRobert Thurlow * to NFSMNT_SECDEFAULT if present. 20592f172c55SRobert Thurlow * 20602f172c55SRobert Thurlow * For referrals, we always want security negotiation and will 20612f172c55SRobert Thurlow * set NFSMNT_SECDEFAULT and we will not copy current secdata. 20622f172c55SRobert Thurlow * The reason is that we can't negotiate down from a parent's 20632f172c55SRobert Thurlow * Kerberos flavor to AUTH_SYS. 2064b9238976Sth * 2065b9238976Sth * If SV4_TRYSECDEFAULT is not set, that indicates that a specific 2066b9238976Sth * security flavour was requested, with data in sv_secdata, and that 2067b9238976Sth * no negotiation should occur. If this specified flavour fails, that's 2068b9238976Sth * it. We will copy sv_secdata, and not set NFSMNT_SECDEFAULT. 2069b9238976Sth * 2070b9238976Sth * If SV4_TRYSECDEFAULT is set, then we start with a passed-in 2071b9238976Sth * default flavour, in sv_secdata, but then negotiate a new flavour. 2072b9238976Sth * Possible flavours are recorded in an array in sv_secinfo, with 2073b9238976Sth * currently in-use flavour pointed to by sv_currsec. 2074b9238976Sth * 2075b9238976Sth * If sv_currsec is set, i.e. if negotiation has already occurred, 2076b9238976Sth * we will copy sv_currsec. Otherwise, copy sv_secdata. Regardless, 2077b9238976Sth * we will set NFSMNT_SECDEFAULT, to enable negotiation. 2078b9238976Sth */ 20792f172c55SRobert Thurlow if (nargs->flags & NFSMNT_REFERRAL) { 20802f172c55SRobert Thurlow /* enable negotiation for referral mount */ 20812f172c55SRobert Thurlow nargs->flags |= NFSMNT_SECDEFAULT; 20822f172c55SRobert Thurlow secdata = kmem_alloc(sizeof (sec_data_t), KM_SLEEP); 20832f172c55SRobert Thurlow secdata->secmod = secdata->rpcflavor = AUTH_SYS; 20842f172c55SRobert Thurlow secdata->data = NULL; 2085b87f76edSThomas Haynes } else if (svp->sv_flags & SV4_TRYSECDEFAULT) { 20862f172c55SRobert Thurlow /* enable negotiation for mirror mount */ 2087b9238976Sth nargs->flags |= NFSMNT_SECDEFAULT; 2088b9238976Sth 2089b9238976Sth /* 2090b9238976Sth * As a starting point for negotiation, copy parent 2091b9238976Sth * mount's negotiated flavour (sv_currsec) if available, 2092b9238976Sth * or its passed-in flavour (sv_secdata) if not. 2093b9238976Sth */ 2094b9238976Sth if (svp->sv_currsec != NULL) 2095b9238976Sth secdata = copy_sec_data(svp->sv_currsec); 2096b9238976Sth else if (svp->sv_secdata != NULL) 2097b9238976Sth secdata = copy_sec_data(svp->sv_secdata); 2098b9238976Sth else 2099b9238976Sth secdata = NULL; 2100b9238976Sth } else { 2101b9238976Sth /* do not enable negotiation; copy parent's passed-in flavour */ 2102b9238976Sth if (svp->sv_secdata != NULL) 2103b9238976Sth secdata = copy_sec_data(svp->sv_secdata); 2104b9238976Sth else 2105b9238976Sth secdata = NULL; 2106b9238976Sth } 2107b9238976Sth 2108b9238976Sth nfs_rw_exit(&svp->sv_lock); 2109b9238976Sth 2110b9238976Sth nargs->flags |= NFSMNT_NEWARGS; 2111b9238976Sth nargs->nfs_args_ext = NFS_ARGS_EXTB; 2112b9238976Sth nargs->nfs_ext_u.nfs_extB.secdata = secdata; 2113b9238976Sth 2114b9238976Sth /* for NFS RO failover; caller will set if necessary */ 2115b9238976Sth nargs->nfs_ext_u.nfs_extB.next = NULL; 2116b9238976Sth 2117b9238976Sth return (nargs); 2118b9238976Sth } 2119b9238976Sth 2120b9238976Sth static void 2121b9238976Sth nfs4_trigger_nargs_destroy(struct nfs_args *nargs) 2122b9238976Sth { 2123b9238976Sth /* 2124b9238976Sth * Either the mount failed, in which case the data is not needed, or 2125b9238976Sth * nfs4_mount() has either taken copies of what it needs or, 2126b9238976Sth * where it has merely copied the ptr, it has set *our* ptr to NULL, 2127b9238976Sth * whereby nfs4_free_args() will ignore it. 2128b9238976Sth */ 2129b9238976Sth nfs4_free_args(nargs); 2130b9238976Sth kmem_free(nargs, sizeof (struct nfs_args)); 2131b9238976Sth } 2132b9238976Sth 2133b9238976Sth /* 2134b9238976Sth * When we finally get into the mounting, we need to add this 2135b9238976Sth * node to the ephemeral tree. 2136b9238976Sth * 2137b9238976Sth * This is called from nfs4_mount(). 2138b9238976Sth */ 2139d3a14591SThomas Haynes int 2140b9238976Sth nfs4_record_ephemeral_mount(mntinfo4_t *mi, vnode_t *mvp) 2141b9238976Sth { 2142b9238976Sth mntinfo4_t *mi_parent; 2143b9238976Sth nfs4_ephemeral_t *eph; 2144b9238976Sth nfs4_ephemeral_tree_t *net; 2145b9238976Sth 2146b9238976Sth nfs4_ephemeral_t *prior; 2147b9238976Sth nfs4_ephemeral_t *child; 2148b9238976Sth 2149b9238976Sth nfs4_ephemeral_t *peer; 2150b9238976Sth 2151b9238976Sth nfs4_trigger_globals_t *ntg; 2152b9238976Sth zone_t *zone = curproc->p_zone; 2153b9238976Sth 2154d3a14591SThomas Haynes int rc = 0; 2155d3a14591SThomas Haynes 2156b9238976Sth mi_parent = VTOMI4(mvp); 2157b9238976Sth 2158b9238976Sth /* 2159b9238976Sth * Get this before grabbing anything else! 2160b9238976Sth */ 2161b9238976Sth ntg = zone_getspecific(nfs4_ephemeral_key, zone); 2162b9238976Sth if (!ntg->ntg_thread_started) { 2163b9238976Sth nfs4_ephemeral_start_harvester(ntg); 2164b9238976Sth } 2165b9238976Sth 2166b9238976Sth mutex_enter(&mi_parent->mi_lock); 2167b9238976Sth mutex_enter(&mi->mi_lock); 2168b9238976Sth 2169d3a14591SThomas Haynes net = mi->mi_ephemeral_tree = 2170d3a14591SThomas Haynes mi_parent->mi_ephemeral_tree; 2171d3a14591SThomas Haynes 2172d3a14591SThomas Haynes /* 2173d3a14591SThomas Haynes * If the mi_ephemeral_tree is NULL, then it 2174d3a14591SThomas Haynes * means that either the harvester or a manual 2175d3a14591SThomas Haynes * umount has cleared the tree out right before 2176d3a14591SThomas Haynes * we got here. 2177d3a14591SThomas Haynes * 2178d3a14591SThomas Haynes * There is nothing we can do here, so return 2179d3a14591SThomas Haynes * to the caller and let them decide whether they 2180d3a14591SThomas Haynes * try again. 2181d3a14591SThomas Haynes */ 2182d3a14591SThomas Haynes if (net == NULL) { 2183d3a14591SThomas Haynes mutex_exit(&mi->mi_lock); 2184d3a14591SThomas Haynes mutex_exit(&mi_parent->mi_lock); 2185d3a14591SThomas Haynes 2186d3a14591SThomas Haynes return (EBUSY); 2187d3a14591SThomas Haynes } 2188d3a14591SThomas Haynes 21892f172c55SRobert Thurlow /* 21902f172c55SRobert Thurlow * We've just tied the mntinfo to the tree, so 21912f172c55SRobert Thurlow * now we bump the refcnt and hold it there until 21922f172c55SRobert Thurlow * this mntinfo is removed from the tree. 21932f172c55SRobert Thurlow */ 2194d3a14591SThomas Haynes nfs4_ephemeral_tree_hold(net); 2195d3a14591SThomas Haynes 2196b9238976Sth /* 2197b9238976Sth * We need to tack together the ephemeral mount 2198b9238976Sth * with this new mntinfo. 2199b9238976Sth */ 2200b9238976Sth eph = kmem_zalloc(sizeof (*eph), KM_SLEEP); 2201b9238976Sth eph->ne_mount = mi; 2202b87f76edSThomas Haynes MI4_HOLD(mi); 2203b87f76edSThomas Haynes VFS_HOLD(mi->mi_vfsp); 2204b9238976Sth eph->ne_ref_time = gethrestime_sec(); 2205b9238976Sth 2206b9238976Sth /* 2207b9238976Sth * We need to tell the ephemeral mount when 2208b9238976Sth * to time out. 2209b9238976Sth */ 2210b9238976Sth eph->ne_mount_to = ntg->ntg_mount_to; 2211b9238976Sth 2212b9238976Sth mi->mi_ephemeral = eph; 2213b9238976Sth 2214b9238976Sth /* 2215b9238976Sth * If the enclosing mntinfo4 is also ephemeral, 2216b9238976Sth * then we need to point to its enclosing parent. 2217b9238976Sth * Else the enclosing mntinfo4 is the enclosing parent. 2218b9238976Sth * 2219b9238976Sth * We also need to weave this ephemeral node 2220b9238976Sth * into the tree. 2221b9238976Sth */ 2222b9238976Sth if (mi_parent->mi_flags & MI4_EPHEMERAL) { 2223b9238976Sth /* 2224b9238976Sth * We need to decide if we are 2225b9238976Sth * the root node of this branch 2226b9238976Sth * or if we are a sibling of this 2227b9238976Sth * branch. 2228b9238976Sth */ 2229b9238976Sth prior = mi_parent->mi_ephemeral; 2230d3a14591SThomas Haynes if (prior == NULL) { 2231d3a14591SThomas Haynes /* 2232d3a14591SThomas Haynes * Race condition, clean up, and 2233d3a14591SThomas Haynes * let caller handle mntinfo. 2234d3a14591SThomas Haynes */ 2235d3a14591SThomas Haynes mi->mi_flags &= ~MI4_EPHEMERAL; 2236d3a14591SThomas Haynes mi->mi_ephemeral = NULL; 2237d3a14591SThomas Haynes kmem_free(eph, sizeof (*eph)); 2238b87f76edSThomas Haynes VFS_RELE(mi->mi_vfsp); 2239b87f76edSThomas Haynes MI4_RELE(mi); 22402f172c55SRobert Thurlow nfs4_ephemeral_tree_rele(net); 2241d3a14591SThomas Haynes rc = EBUSY; 2242b9238976Sth } else { 2243d3a14591SThomas Haynes if (prior->ne_child == NULL) { 2244d3a14591SThomas Haynes prior->ne_child = eph; 2245d3a14591SThomas Haynes } else { 2246d3a14591SThomas Haynes child = prior->ne_child; 2247b9238976Sth 2248d3a14591SThomas Haynes prior->ne_child = eph; 2249d3a14591SThomas Haynes eph->ne_peer = child; 2250b9238976Sth 2251d3a14591SThomas Haynes child->ne_prior = eph; 2252d3a14591SThomas Haynes } 2253b9238976Sth 2254d3a14591SThomas Haynes eph->ne_prior = prior; 2255d3a14591SThomas Haynes } 2256b9238976Sth } else { 2257b9238976Sth /* 2258b9238976Sth * The parent mntinfo4 is the non-ephemeral 2259b9238976Sth * root of the ephemeral tree. We 2260b9238976Sth * need to decide if we are the root 2261b9238976Sth * node of that tree or if we are a 2262b9238976Sth * sibling of the root node. 2263b9238976Sth * 2264b9238976Sth * We are the root if there is no 2265b9238976Sth * other node. 2266b9238976Sth */ 2267b9238976Sth if (net->net_root == NULL) { 2268b9238976Sth net->net_root = eph; 2269b9238976Sth } else { 2270b9238976Sth eph->ne_peer = peer = net->net_root; 2271b9238976Sth ASSERT(peer != NULL); 2272b9238976Sth net->net_root = eph; 2273b9238976Sth 2274b9238976Sth peer->ne_prior = eph; 2275b9238976Sth } 2276b9238976Sth 2277b9238976Sth eph->ne_prior = NULL; 2278b9238976Sth } 2279b9238976Sth 2280b9238976Sth mutex_exit(&mi->mi_lock); 2281b9238976Sth mutex_exit(&mi_parent->mi_lock); 2282d3a14591SThomas Haynes 2283d3a14591SThomas Haynes return (rc); 2284b9238976Sth } 2285b9238976Sth 2286b9238976Sth /* 2287b9238976Sth * Commit the changes to the ephemeral tree for removing this node. 2288b9238976Sth */ 2289b9238976Sth static void 2290b9238976Sth nfs4_ephemeral_umount_cleanup(nfs4_ephemeral_t *eph) 2291b9238976Sth { 2292b9238976Sth nfs4_ephemeral_t *e = eph; 2293b9238976Sth nfs4_ephemeral_t *peer; 2294b9238976Sth nfs4_ephemeral_t *prior; 2295b9238976Sth 2296b9238976Sth peer = eph->ne_peer; 2297b9238976Sth prior = e->ne_prior; 2298b9238976Sth 2299b9238976Sth /* 2300b9238976Sth * If this branch root was not the 2301b9238976Sth * tree root, then we need to fix back pointers. 2302b9238976Sth */ 2303b9238976Sth if (prior) { 2304b9238976Sth if (prior->ne_child == e) { 2305b9238976Sth prior->ne_child = peer; 2306b9238976Sth } else { 2307b9238976Sth prior->ne_peer = peer; 2308b9238976Sth } 2309b9238976Sth 2310b9238976Sth if (peer) 2311b9238976Sth peer->ne_prior = prior; 2312b9238976Sth } else if (peer) { 2313b9238976Sth peer->ne_mount->mi_ephemeral_tree->net_root = peer; 2314b9238976Sth peer->ne_prior = NULL; 2315b9238976Sth } else { 2316b9238976Sth e->ne_mount->mi_ephemeral_tree->net_root = NULL; 2317b9238976Sth } 2318b9238976Sth } 2319b9238976Sth 2320b9238976Sth /* 2321b9238976Sth * We want to avoid recursion at all costs. So we need to 2322b9238976Sth * unroll the tree. We do this by a depth first traversal to 2323b9238976Sth * leaf nodes. We blast away the leaf and work our way back 2324b9238976Sth * up and down the tree. 2325b9238976Sth */ 2326b9238976Sth static int 2327b9238976Sth nfs4_ephemeral_unmount_engine(nfs4_ephemeral_t *eph, 2328b9238976Sth int isTreeRoot, int flag, cred_t *cr) 2329b9238976Sth { 2330b9238976Sth nfs4_ephemeral_t *e = eph; 2331b9238976Sth nfs4_ephemeral_t *prior; 2332b9238976Sth mntinfo4_t *mi; 2333b9238976Sth vfs_t *vfsp; 2334b9238976Sth int error; 2335b9238976Sth 2336b9238976Sth /* 2337b9238976Sth * We use the loop while unrolling the ephemeral tree. 2338b9238976Sth */ 2339b9238976Sth for (;;) { 2340b9238976Sth /* 2341b9238976Sth * First we walk down the child. 2342b9238976Sth */ 2343b9238976Sth if (e->ne_child) { 2344b9238976Sth prior = e; 2345b9238976Sth e = e->ne_child; 2346b9238976Sth continue; 2347b9238976Sth } 2348b9238976Sth 2349b9238976Sth /* 2350b9238976Sth * If we are the root of the branch we are removing, 2351b9238976Sth * we end it here. But if the branch is the root of 2352b9238976Sth * the tree, we have to forge on. We do not consider 2353b9238976Sth * the peer list for the root because while it may 2354b9238976Sth * be okay to remove, it is both extra work and a 2355b9238976Sth * potential for a false-positive error to stall the 2356b9238976Sth * unmount attempt. 2357b9238976Sth */ 2358b9238976Sth if (e == eph && isTreeRoot == FALSE) 2359b9238976Sth return (0); 2360b9238976Sth 2361b9238976Sth /* 2362b9238976Sth * Next we walk down the peer list. 2363b9238976Sth */ 2364b9238976Sth if (e->ne_peer) { 2365b9238976Sth prior = e; 2366b9238976Sth e = e->ne_peer; 2367b9238976Sth continue; 2368b9238976Sth } 2369b9238976Sth 2370b9238976Sth /* 2371b9238976Sth * We can only remove the node passed in by the 2372b9238976Sth * caller if it is the root of the ephemeral tree. 2373b9238976Sth * Otherwise, the caller will remove it. 2374b9238976Sth */ 2375b9238976Sth if (e == eph && isTreeRoot == FALSE) 2376b9238976Sth return (0); 2377b9238976Sth 2378b9238976Sth /* 2379b9238976Sth * Okay, we have a leaf node, time 2380b9238976Sth * to prune it! 2381b9238976Sth * 2382b9238976Sth * Note that prior can only be NULL if 2383b9238976Sth * and only if it is the root of the 2384b9238976Sth * ephemeral tree. 2385b9238976Sth */ 2386b9238976Sth prior = e->ne_prior; 2387b9238976Sth 2388b9238976Sth mi = e->ne_mount; 2389b9238976Sth mutex_enter(&mi->mi_lock); 2390b9238976Sth vfsp = mi->mi_vfsp; 2391b87f76edSThomas Haynes ASSERT(vfsp != NULL); 2392b9238976Sth 2393b9238976Sth /* 2394b9238976Sth * Cleared by umount2_engine. 2395b9238976Sth */ 2396b9238976Sth VFS_HOLD(vfsp); 2397b9238976Sth 2398b9238976Sth /* 2399b9238976Sth * Inform nfs4_unmount to not recursively 2400b9238976Sth * descend into this node's children when it 2401b9238976Sth * gets processed. 2402b9238976Sth */ 2403b9238976Sth mi->mi_flags |= MI4_EPHEMERAL_RECURSED; 2404b9238976Sth mutex_exit(&mi->mi_lock); 2405b9238976Sth 2406b9238976Sth error = umount2_engine(vfsp, flag, cr, FALSE); 2407b9238976Sth if (error) { 2408b9238976Sth /* 2409b9238976Sth * We need to reenable nfs4_unmount's ability 2410b9238976Sth * to recursively descend on this node. 2411b9238976Sth */ 2412b9238976Sth mutex_enter(&mi->mi_lock); 2413b9238976Sth mi->mi_flags &= ~MI4_EPHEMERAL_RECURSED; 2414b9238976Sth mutex_exit(&mi->mi_lock); 2415b9238976Sth 2416b9238976Sth return (error); 2417b9238976Sth } 2418b9238976Sth 2419b9238976Sth /* 2420b9238976Sth * If we are the current node, we do not want to 2421b9238976Sth * touch anything else. At this point, the only 2422b9238976Sth * way the current node can have survived to here 2423b9238976Sth * is if it is the root of the ephemeral tree and 2424b9238976Sth * we are unmounting the enclosing mntinfo4. 2425b9238976Sth */ 2426b9238976Sth if (e == eph) { 2427b9238976Sth ASSERT(prior == NULL); 2428b9238976Sth return (0); 2429b9238976Sth } 2430b9238976Sth 2431b9238976Sth /* 2432b9238976Sth * Stitch up the prior node. Note that since 2433b9238976Sth * we have handled the root of the tree, prior 2434b9238976Sth * must be non-NULL. 2435b9238976Sth */ 2436b9238976Sth ASSERT(prior != NULL); 2437b9238976Sth if (prior->ne_child == e) { 2438b9238976Sth prior->ne_child = NULL; 2439b9238976Sth } else { 2440b9238976Sth ASSERT(prior->ne_peer == e); 2441b9238976Sth 2442b9238976Sth prior->ne_peer = NULL; 2443b9238976Sth } 2444b9238976Sth 2445b9238976Sth e = prior; 2446b9238976Sth } 2447b9238976Sth 2448b9238976Sth /* NOTREACHED */ 2449b9238976Sth } 2450b9238976Sth 2451b9238976Sth /* 2452b9238976Sth * Common code to safely release net_cnt_lock and net_tree_lock 2453b9238976Sth */ 2454b9238976Sth void 2455b9238976Sth nfs4_ephemeral_umount_unlock(bool_t *pmust_unlock, 24562f172c55SRobert Thurlow nfs4_ephemeral_tree_t **pnet) 2457b9238976Sth { 2458b9238976Sth nfs4_ephemeral_tree_t *net = *pnet; 2459b9238976Sth 2460b9238976Sth if (*pmust_unlock) { 2461b9238976Sth mutex_enter(&net->net_cnt_lock); 2462b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_UMOUNTING; 2463b9238976Sth mutex_exit(&net->net_cnt_lock); 2464b9238976Sth 2465b9238976Sth mutex_exit(&net->net_tree_lock); 2466b9238976Sth 2467b9238976Sth *pmust_unlock = FALSE; 2468b9238976Sth } 2469b9238976Sth } 2470b9238976Sth 2471b9238976Sth /* 2472b9238976Sth * While we may have removed any child or sibling nodes of this 2473b9238976Sth * ephemeral node, we can not nuke it until we know that there 2474b9238976Sth * were no actived vnodes on it. This will do that final 2475b9238976Sth * work once we know it is not busy. 2476b9238976Sth */ 2477b9238976Sth void 2478b9238976Sth nfs4_ephemeral_umount_activate(mntinfo4_t *mi, bool_t *pmust_unlock, 24792f172c55SRobert Thurlow nfs4_ephemeral_tree_t **pnet) 2480b9238976Sth { 2481b9238976Sth /* 2482b9238976Sth * Now we need to get rid of the ephemeral data if it exists. 2483b9238976Sth */ 2484b9238976Sth mutex_enter(&mi->mi_lock); 2485b9238976Sth if (mi->mi_ephemeral) { 2486b9238976Sth /* 2487b9238976Sth * If we are the root node of an ephemeral branch 2488b9238976Sth * which is being removed, then we need to fixup 2489b9238976Sth * pointers into and out of the node. 2490b9238976Sth */ 2491b9238976Sth if (!(mi->mi_flags & MI4_EPHEMERAL_RECURSED)) 2492b9238976Sth nfs4_ephemeral_umount_cleanup(mi->mi_ephemeral); 2493b9238976Sth 24942f172c55SRobert Thurlow nfs4_ephemeral_tree_rele(*pnet); 2495b9238976Sth ASSERT(mi->mi_ephemeral != NULL); 2496b9238976Sth 2497b9238976Sth kmem_free(mi->mi_ephemeral, sizeof (*mi->mi_ephemeral)); 2498b9238976Sth mi->mi_ephemeral = NULL; 2499b87f76edSThomas Haynes VFS_RELE(mi->mi_vfsp); 2500b87f76edSThomas Haynes MI4_RELE(mi); 2501b9238976Sth } 2502b9238976Sth mutex_exit(&mi->mi_lock); 2503b9238976Sth 25042f172c55SRobert Thurlow nfs4_ephemeral_umount_unlock(pmust_unlock, pnet); 2505b9238976Sth } 2506b9238976Sth 2507b9238976Sth /* 2508b9238976Sth * Unmount an ephemeral node. 25092f172c55SRobert Thurlow * 25102f172c55SRobert Thurlow * Note that if this code fails, then it must unlock. 25112f172c55SRobert Thurlow * 25122f172c55SRobert Thurlow * If it succeeds, then the caller must be prepared to do so. 2513b9238976Sth */ 2514b9238976Sth int 2515b9238976Sth nfs4_ephemeral_umount(mntinfo4_t *mi, int flag, cred_t *cr, 25162f172c55SRobert Thurlow bool_t *pmust_unlock, nfs4_ephemeral_tree_t **pnet) 2517b9238976Sth { 2518b9238976Sth int error = 0; 2519b9238976Sth nfs4_ephemeral_t *eph; 2520b9238976Sth nfs4_ephemeral_tree_t *net; 2521b9238976Sth int is_derooting = FALSE; 2522b9238976Sth int is_recursed = FALSE; 2523d3a14591SThomas Haynes int was_locked = FALSE; 2524d3a14591SThomas Haynes 2525d3a14591SThomas Haynes /* 2526d3a14591SThomas Haynes * Make sure to set the default state for cleaning 2527d3a14591SThomas Haynes * up the tree in the caller (and on the way out). 2528d3a14591SThomas Haynes */ 25292f172c55SRobert Thurlow *pmust_unlock = FALSE; 2530b9238976Sth 2531b9238976Sth /* 2532b9238976Sth * The active vnodes on this file system may be ephemeral 2533b9238976Sth * children. We need to check for and try to unmount them 2534b9238976Sth * here. If any can not be unmounted, we are going 2535b9238976Sth * to return EBUSY. 2536b9238976Sth */ 2537b9238976Sth mutex_enter(&mi->mi_lock); 2538b9238976Sth 2539b9238976Sth /* 2540b9238976Sth * If an ephemeral tree, we need to check to see if 2541b9238976Sth * the lock is already held. If it is, then we need 2542b9238976Sth * to see if we are being called as a result of 2543b9238976Sth * the recursive removal of some node of the tree or 2544b9238976Sth * if we are another attempt to remove the tree. 2545b9238976Sth * 2546b9238976Sth * mi_flags & MI4_EPHEMERAL indicates an ephemeral 2547b9238976Sth * node. mi_ephemeral being non-NULL also does this. 2548b9238976Sth * 2549b9238976Sth * mi_ephemeral_tree being non-NULL is sufficient 2550b9238976Sth * to also indicate either it is an ephemeral node 2551b9238976Sth * or the enclosing mntinfo4. 2552b9238976Sth * 2553b9238976Sth * Do we need MI4_EPHEMERAL? Yes, it is useful for 2554b9238976Sth * when we delete the ephemeral node and need to 2555b9238976Sth * differentiate from an ephemeral node and the 2556b9238976Sth * enclosing root node. 2557b9238976Sth */ 2558b9238976Sth *pnet = net = mi->mi_ephemeral_tree; 2559eabd0450Sth if (net == NULL) { 2560b9238976Sth mutex_exit(&mi->mi_lock); 2561eabd0450Sth return (0); 2562eabd0450Sth } 2563b9238976Sth 2564eabd0450Sth eph = mi->mi_ephemeral; 2565eabd0450Sth is_recursed = mi->mi_flags & MI4_EPHEMERAL_RECURSED; 2566eabd0450Sth is_derooting = (eph == NULL); 2567b9238976Sth 25682f172c55SRobert Thurlow mutex_enter(&net->net_cnt_lock); 25692f172c55SRobert Thurlow 2570eabd0450Sth /* 2571eabd0450Sth * If this is not recursion, then we need to 25722f172c55SRobert Thurlow * check to see if a harvester thread has 25732f172c55SRobert Thurlow * already grabbed the lock. 2574eabd0450Sth * 25752f172c55SRobert Thurlow * After we exit this branch, we may not 25762f172c55SRobert Thurlow * blindly return, we need to jump to 25772f172c55SRobert Thurlow * is_busy! 2578eabd0450Sth */ 2579eabd0450Sth if (!is_recursed) { 2580eabd0450Sth if (net->net_status & 2581eabd0450Sth NFS4_EPHEMERAL_TREE_LOCKED) { 2582b9238976Sth /* 2583d3a14591SThomas Haynes * If the tree is locked, we need 2584d3a14591SThomas Haynes * to decide whether we are the 2585d3a14591SThomas Haynes * harvester or some explicit call 2586d3a14591SThomas Haynes * for a umount. The only way that 2587d3a14591SThomas Haynes * we are the harvester is if 2588d3a14591SThomas Haynes * MS_SYSSPACE is set. 2589d3a14591SThomas Haynes * 2590d3a14591SThomas Haynes * We only let the harvester through 2591d3a14591SThomas Haynes * at this point. 2592eabd0450Sth * 2593eabd0450Sth * We return EBUSY so that the 2594eabd0450Sth * caller knows something is 2595eabd0450Sth * going on. Note that by that 2596eabd0450Sth * time, the umount in the other 2597eabd0450Sth * thread may have already occured. 2598b9238976Sth */ 2599d3a14591SThomas Haynes if (!(flag & MS_SYSSPACE)) { 2600d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 2601d3a14591SThomas Haynes mutex_exit(&mi->mi_lock); 2602d3a14591SThomas Haynes 2603d3a14591SThomas Haynes return (EBUSY); 2604d3a14591SThomas Haynes } 2605d3a14591SThomas Haynes 2606d3a14591SThomas Haynes was_locked = TRUE; 2607d3a14591SThomas Haynes } 2608eabd0450Sth } 26092f172c55SRobert Thurlow 26102f172c55SRobert Thurlow mutex_exit(&net->net_cnt_lock); 2611eabd0450Sth mutex_exit(&mi->mi_lock); 2612b9238976Sth 2613eabd0450Sth /* 2614d3a14591SThomas Haynes * If we are not the harvester, we need to check 2615d3a14591SThomas Haynes * to see if we need to grab the tree lock. 2616eabd0450Sth */ 2617d3a14591SThomas Haynes if (was_locked == FALSE) { 2618d3a14591SThomas Haynes /* 2619d3a14591SThomas Haynes * If we grab the lock, it means that no other 2620d3a14591SThomas Haynes * operation is working on the tree. If we don't 2621d3a14591SThomas Haynes * grab it, we need to decide if this is because 2622d3a14591SThomas Haynes * we are a recursive call or a new operation. 2623d3a14591SThomas Haynes */ 2624d3a14591SThomas Haynes if (mutex_tryenter(&net->net_tree_lock)) { 2625d3a14591SThomas Haynes *pmust_unlock = TRUE; 2626d3a14591SThomas Haynes } else { 2627b9238976Sth /* 2628d3a14591SThomas Haynes * If we are a recursive call, we can 2629d3a14591SThomas Haynes * proceed without the lock. 2630d3a14591SThomas Haynes * Otherwise we have to wait until 2631d3a14591SThomas Haynes * the lock becomes free. 2632b9238976Sth */ 2633d3a14591SThomas Haynes if (!is_recursed) { 2634d3a14591SThomas Haynes mutex_enter(&net->net_cnt_lock); 2635d3a14591SThomas Haynes if (net->net_status & 2636d3a14591SThomas Haynes (NFS4_EPHEMERAL_TREE_DEROOTING 2637d3a14591SThomas Haynes | NFS4_EPHEMERAL_TREE_INVALID)) { 2638d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 2639d3a14591SThomas Haynes goto is_busy; 2640d3a14591SThomas Haynes } 2641d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 2642b9238976Sth 2643d3a14591SThomas Haynes /* 2644d3a14591SThomas Haynes * We can't hold any other locks whilst 2645d3a14591SThomas Haynes * we wait on this to free up. 2646d3a14591SThomas Haynes */ 2647d3a14591SThomas Haynes mutex_enter(&net->net_tree_lock); 2648b9238976Sth 2649d3a14591SThomas Haynes /* 2650d3a14591SThomas Haynes * Note that while mi->mi_ephemeral 2651d3a14591SThomas Haynes * may change and thus we have to 2652d3a14591SThomas Haynes * update eph, it is the case that 2653d3a14591SThomas Haynes * we have tied down net and 2654d3a14591SThomas Haynes * do not care if mi->mi_ephemeral_tree 2655d3a14591SThomas Haynes * has changed. 2656d3a14591SThomas Haynes */ 2657d3a14591SThomas Haynes mutex_enter(&mi->mi_lock); 2658d3a14591SThomas Haynes eph = mi->mi_ephemeral; 2659d3a14591SThomas Haynes mutex_exit(&mi->mi_lock); 2660d3a14591SThomas Haynes 2661d3a14591SThomas Haynes /* 2662d3a14591SThomas Haynes * Okay, we need to see if either the 2663d3a14591SThomas Haynes * tree got nuked or the current node 2664d3a14591SThomas Haynes * got nuked. Both of which will cause 2665d3a14591SThomas Haynes * an error. 2666d3a14591SThomas Haynes * 2667d3a14591SThomas Haynes * Note that a subsequent retry of the 2668d3a14591SThomas Haynes * umount shall work. 2669d3a14591SThomas Haynes */ 2670d3a14591SThomas Haynes mutex_enter(&net->net_cnt_lock); 2671d3a14591SThomas Haynes if (net->net_status & 2672d3a14591SThomas Haynes NFS4_EPHEMERAL_TREE_INVALID || 2673d3a14591SThomas Haynes (!is_derooting && eph == NULL)) { 2674d3a14591SThomas Haynes mutex_exit(&net->net_cnt_lock); 2675d3a14591SThomas Haynes mutex_exit(&net->net_tree_lock); 2676d3a14591SThomas Haynes goto is_busy; 2677d3a14591SThomas Haynes } 2678eabd0450Sth mutex_exit(&net->net_cnt_lock); 2679d3a14591SThomas Haynes *pmust_unlock = TRUE; 2680eabd0450Sth } 2681eabd0450Sth } 2682eabd0450Sth } 2683eabd0450Sth 2684eabd0450Sth /* 2685eabd0450Sth * Only once we have grabbed the lock can we mark what we 2686eabd0450Sth * are planning on doing to the ephemeral tree. 2687eabd0450Sth */ 2688eabd0450Sth if (*pmust_unlock) { 2689eabd0450Sth mutex_enter(&net->net_cnt_lock); 2690eabd0450Sth net->net_status |= NFS4_EPHEMERAL_TREE_UMOUNTING; 2691eabd0450Sth 2692eabd0450Sth /* 2693eabd0450Sth * Check to see if we are nuking the root. 2694eabd0450Sth */ 2695eabd0450Sth if (is_derooting) 2696eabd0450Sth net->net_status |= 2697eabd0450Sth NFS4_EPHEMERAL_TREE_DEROOTING; 2698eabd0450Sth mutex_exit(&net->net_cnt_lock); 2699eabd0450Sth } 2700eabd0450Sth 2701eabd0450Sth if (!is_derooting) { 2702eabd0450Sth /* 2703eabd0450Sth * Only work on children if the caller has not already 2704eabd0450Sth * done so. 2705eabd0450Sth */ 2706eabd0450Sth if (!is_recursed) { 2707eabd0450Sth ASSERT(eph != NULL); 2708eabd0450Sth 2709eabd0450Sth error = nfs4_ephemeral_unmount_engine(eph, 2710eabd0450Sth FALSE, flag, cr); 2711eabd0450Sth if (error) 2712eabd0450Sth goto is_busy; 2713eabd0450Sth } 2714eabd0450Sth } else { 2715eabd0450Sth eph = net->net_root; 2716eabd0450Sth 2717eabd0450Sth /* 2718eabd0450Sth * Only work if there is something there. 2719eabd0450Sth */ 2720eabd0450Sth if (eph) { 2721eabd0450Sth error = nfs4_ephemeral_unmount_engine(eph, TRUE, 2722eabd0450Sth flag, cr); 2723eabd0450Sth if (error) { 2724eabd0450Sth mutex_enter(&net->net_cnt_lock); 2725eabd0450Sth net->net_status &= 2726eabd0450Sth ~NFS4_EPHEMERAL_TREE_DEROOTING; 2727eabd0450Sth mutex_exit(&net->net_cnt_lock); 2728eabd0450Sth goto is_busy; 2729eabd0450Sth } 2730b9238976Sth 2731b9238976Sth /* 2732eabd0450Sth * Nothing else which goes wrong will 2733eabd0450Sth * invalidate the blowing away of the 2734eabd0450Sth * ephmeral tree. 2735b9238976Sth */ 2736eabd0450Sth net->net_root = NULL; 2737b9238976Sth } 2738eabd0450Sth 2739eabd0450Sth /* 2740eabd0450Sth * We have derooted and we have caused the tree to be 2741d3a14591SThomas Haynes * invalidated. 2742eabd0450Sth */ 2743eabd0450Sth mutex_enter(&net->net_cnt_lock); 2744eabd0450Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_DEROOTING; 2745eabd0450Sth net->net_status |= NFS4_EPHEMERAL_TREE_INVALID; 27462f172c55SRobert Thurlow DTRACE_NFSV4_1(nfs4clnt__dbg__ephemeral__tree__derooting, 27472f172c55SRobert Thurlow uint_t, net->net_refcnt); 27482f172c55SRobert Thurlow 27492f172c55SRobert Thurlow /* 27502f172c55SRobert Thurlow * We will not finalize this node, so safe to 27512f172c55SRobert Thurlow * release it. 27522f172c55SRobert Thurlow */ 27532f172c55SRobert Thurlow nfs4_ephemeral_tree_decr(net); 2754eabd0450Sth mutex_exit(&net->net_cnt_lock); 2755eabd0450Sth 2756d3a14591SThomas Haynes if (was_locked == FALSE) 2757d3a14591SThomas Haynes mutex_exit(&net->net_tree_lock); 2758d3a14591SThomas Haynes 2759d3a14591SThomas Haynes /* 2760d3a14591SThomas Haynes * We have just blown away any notation of this 27612f172c55SRobert Thurlow * tree being locked or having a refcnt. 27622f172c55SRobert Thurlow * We can't let the caller try to clean things up. 2763d3a14591SThomas Haynes */ 2764d3a14591SThomas Haynes *pmust_unlock = FALSE; 2765d3a14591SThomas Haynes 2766eabd0450Sth /* 2767d708af74SThomas Haynes * At this point, the tree should no longer be 2768d708af74SThomas Haynes * associated with the mntinfo4. We need to pull 2769d708af74SThomas Haynes * it off there and let the harvester take 2770eabd0450Sth * care of it once the refcnt drops. 2771eabd0450Sth */ 2772eabd0450Sth mutex_enter(&mi->mi_lock); 2773eabd0450Sth mi->mi_ephemeral_tree = NULL; 2774b9238976Sth mutex_exit(&mi->mi_lock); 2775b9238976Sth } 2776b9238976Sth 2777b9238976Sth return (0); 2778b9238976Sth 2779b9238976Sth is_busy: 2780b9238976Sth 27812f172c55SRobert Thurlow nfs4_ephemeral_umount_unlock(pmust_unlock, pnet); 2782b9238976Sth 2783b9238976Sth return (error); 2784b9238976Sth } 2785b9238976Sth 2786b9238976Sth /* 2787b9238976Sth * Do the umount and record any error in the parent. 2788b9238976Sth */ 2789b9238976Sth static void 2790b9238976Sth nfs4_ephemeral_record_umount(vfs_t *vfsp, int flag, 2791b9238976Sth nfs4_ephemeral_t *e, nfs4_ephemeral_t *prior) 2792b9238976Sth { 2793b9238976Sth int error; 2794b9238976Sth 2795b87f76edSThomas Haynes /* 2796b87f76edSThomas Haynes * Only act on if the fs is still mounted. 2797b87f76edSThomas Haynes */ 2798b87f76edSThomas Haynes if (vfsp == NULL) 2799b87f76edSThomas Haynes return; 2800b87f76edSThomas Haynes 2801b9238976Sth error = umount2_engine(vfsp, flag, kcred, FALSE); 2802b9238976Sth if (error) { 2803b9238976Sth if (prior) { 2804b9238976Sth if (prior->ne_child == e) 2805b9238976Sth prior->ne_state |= 2806b9238976Sth NFS4_EPHEMERAL_CHILD_ERROR; 2807b9238976Sth else 2808b9238976Sth prior->ne_state |= 2809b9238976Sth NFS4_EPHEMERAL_PEER_ERROR; 2810b9238976Sth } 2811b9238976Sth } 2812b9238976Sth } 2813b9238976Sth 2814b9238976Sth /* 2815b9238976Sth * For each tree in the forest (where the forest is in 2816b9238976Sth * effect all of the ephemeral trees for this zone), 2817b9238976Sth * scan to see if a node can be unmounted. Note that 2818b9238976Sth * unlike nfs4_ephemeral_unmount_engine(), we do 2819b9238976Sth * not process the current node before children or 2820b9238976Sth * siblings. I.e., if a node can be unmounted, we 2821b9238976Sth * do not recursively check to see if the nodes 2822b9238976Sth * hanging off of it can also be unmounted. 2823b9238976Sth * 2824b9238976Sth * Instead, we delve down deep to try and remove the 2825b9238976Sth * children first. Then, because we share code with 2826b9238976Sth * nfs4_ephemeral_unmount_engine(), we will try 2827b9238976Sth * them again. This could be a performance issue in 2828b9238976Sth * the future. 2829b9238976Sth * 2830b9238976Sth * Also note that unlike nfs4_ephemeral_unmount_engine(), 2831b9238976Sth * we do not halt on an error. We will not remove the 2832b9238976Sth * current node, but we will keep on trying to remove 2833b9238976Sth * the others. 2834b9238976Sth * 2835b9238976Sth * force indicates that we want the unmount to occur 2836b9238976Sth * even if there is something blocking it. 2837b9238976Sth * 2838b9238976Sth * time_check indicates that we want to see if the 2839b9238976Sth * mount has expired past mount_to or not. Typically 2840b9238976Sth * we want to do this and only on a shutdown of the 2841b9238976Sth * zone would we want to ignore the check. 2842b9238976Sth */ 2843b9238976Sth static void 2844b9238976Sth nfs4_ephemeral_harvest_forest(nfs4_trigger_globals_t *ntg, 2845b9238976Sth bool_t force, bool_t time_check) 2846b9238976Sth { 2847b9238976Sth nfs4_ephemeral_tree_t *net; 2848b9238976Sth nfs4_ephemeral_tree_t *prev = NULL; 2849b9238976Sth nfs4_ephemeral_tree_t *next; 2850b9238976Sth nfs4_ephemeral_t *e; 2851b9238976Sth nfs4_ephemeral_t *prior; 2852b9238976Sth time_t now = gethrestime_sec(); 2853b9238976Sth 2854b9238976Sth nfs4_ephemeral_tree_t *harvest = NULL; 2855b9238976Sth 2856b9238976Sth int flag; 2857b9238976Sth 2858b9238976Sth mntinfo4_t *mi; 2859b9238976Sth vfs_t *vfsp; 2860b9238976Sth 2861b9238976Sth if (force) 2862d3a14591SThomas Haynes flag = MS_FORCE | MS_SYSSPACE; 2863b9238976Sth else 2864d3a14591SThomas Haynes flag = MS_SYSSPACE; 2865b9238976Sth 2866b9238976Sth mutex_enter(&ntg->ntg_forest_lock); 2867b9238976Sth for (net = ntg->ntg_forest; net != NULL; net = next) { 2868b9238976Sth next = net->net_next; 2869b9238976Sth 2870d3a14591SThomas Haynes nfs4_ephemeral_tree_hold(net); 2871b9238976Sth 2872b9238976Sth mutex_enter(&net->net_tree_lock); 2873b9238976Sth 2874b9238976Sth /* 2875b9238976Sth * Let the unmount code know that the 2876b9238976Sth * tree is already locked! 2877b9238976Sth */ 2878b9238976Sth mutex_enter(&net->net_cnt_lock); 2879b9238976Sth net->net_status |= NFS4_EPHEMERAL_TREE_LOCKED; 2880b9238976Sth mutex_exit(&net->net_cnt_lock); 2881b9238976Sth 2882b9238976Sth /* 2883b9238976Sth * If the intent is force all ephemeral nodes to 2884b9238976Sth * be unmounted in this zone, we can short circuit a 2885b9238976Sth * lot of tree traversal and simply zap the root node. 2886b9238976Sth */ 2887b9238976Sth if (force) { 2888b9238976Sth if (net->net_root) { 2889b9238976Sth mi = net->net_root->ne_mount; 2890b87f76edSThomas Haynes 2891b9238976Sth vfsp = mi->mi_vfsp; 2892b87f76edSThomas Haynes ASSERT(vfsp != NULL); 2893b9238976Sth 2894b9238976Sth /* 2895b9238976Sth * Cleared by umount2_engine. 2896b9238976Sth */ 2897b9238976Sth VFS_HOLD(vfsp); 2898b9238976Sth 2899b9238976Sth (void) umount2_engine(vfsp, flag, 2900b9238976Sth kcred, FALSE); 2901b9238976Sth 2902b9238976Sth goto check_done; 2903b9238976Sth } 2904b9238976Sth } 2905b9238976Sth 2906b9238976Sth e = net->net_root; 2907b9238976Sth if (e) 2908b9238976Sth e->ne_state = NFS4_EPHEMERAL_VISIT_CHILD; 2909b9238976Sth 2910b9238976Sth while (e) { 2911b9238976Sth if (e->ne_state == NFS4_EPHEMERAL_VISIT_CHILD) { 2912b9238976Sth e->ne_state = NFS4_EPHEMERAL_VISIT_SIBLING; 2913b9238976Sth if (e->ne_child) { 2914b9238976Sth e = e->ne_child; 2915b9238976Sth e->ne_state = 2916b9238976Sth NFS4_EPHEMERAL_VISIT_CHILD; 2917b9238976Sth } 2918b9238976Sth 2919b9238976Sth continue; 2920b9238976Sth } else if (e->ne_state == 2921b9238976Sth NFS4_EPHEMERAL_VISIT_SIBLING) { 2922b9238976Sth e->ne_state = NFS4_EPHEMERAL_PROCESS_ME; 2923b9238976Sth if (e->ne_peer) { 2924b9238976Sth e = e->ne_peer; 2925b9238976Sth e->ne_state = 2926b9238976Sth NFS4_EPHEMERAL_VISIT_CHILD; 2927b9238976Sth } 2928b9238976Sth 2929b9238976Sth continue; 2930b9238976Sth } else if (e->ne_state == 2931b9238976Sth NFS4_EPHEMERAL_CHILD_ERROR) { 2932b9238976Sth prior = e->ne_prior; 2933b9238976Sth 2934b9238976Sth /* 2935b9238976Sth * If a child reported an error, do 2936b9238976Sth * not bother trying to unmount. 2937b9238976Sth * 2938b9238976Sth * If your prior node is a parent, 2939b9238976Sth * pass the error up such that they 2940b9238976Sth * also do not try to unmount. 2941b9238976Sth * 2942b9238976Sth * However, if your prior is a sibling, 2943b9238976Sth * let them try to unmount if they can. 2944b9238976Sth */ 2945b9238976Sth if (prior) { 2946b9238976Sth if (prior->ne_child == e) 2947b9238976Sth prior->ne_state |= 2948b9238976Sth NFS4_EPHEMERAL_CHILD_ERROR; 2949b9238976Sth else 2950b9238976Sth prior->ne_state |= 2951b9238976Sth NFS4_EPHEMERAL_PEER_ERROR; 2952b9238976Sth } 2953b9238976Sth 2954b9238976Sth /* 2955b9238976Sth * Clear the error and if needed, process peers. 2956b9238976Sth * 2957b9238976Sth * Once we mask out the error, we know whether 2958b9238976Sth * or we have to process another node. 2959b9238976Sth */ 2960b9238976Sth e->ne_state &= ~NFS4_EPHEMERAL_CHILD_ERROR; 2961b9238976Sth if (e->ne_state == NFS4_EPHEMERAL_PROCESS_ME) 2962b9238976Sth e = prior; 2963b9238976Sth 2964b9238976Sth continue; 2965b9238976Sth } else if (e->ne_state == 2966b9238976Sth NFS4_EPHEMERAL_PEER_ERROR) { 2967b9238976Sth prior = e->ne_prior; 2968b9238976Sth 2969b9238976Sth if (prior) { 2970b9238976Sth if (prior->ne_child == e) 2971b9238976Sth prior->ne_state = 2972b9238976Sth NFS4_EPHEMERAL_CHILD_ERROR; 2973b9238976Sth else 2974b9238976Sth prior->ne_state = 2975b9238976Sth NFS4_EPHEMERAL_PEER_ERROR; 2976b9238976Sth } 2977b9238976Sth 2978b9238976Sth /* 2979b9238976Sth * Clear the error from this node and do the 2980b9238976Sth * correct processing. 2981b9238976Sth */ 2982b9238976Sth e->ne_state &= ~NFS4_EPHEMERAL_PEER_ERROR; 2983b9238976Sth continue; 2984b9238976Sth } 2985b9238976Sth 2986b9238976Sth prior = e->ne_prior; 2987b9238976Sth e->ne_state = NFS4_EPHEMERAL_OK; 2988b9238976Sth 2989b9238976Sth /* 2990b9238976Sth * It must be the case that we need to process 2991b9238976Sth * this node. 2992b9238976Sth */ 2993b9238976Sth if (!time_check || 2994b9238976Sth now - e->ne_ref_time > e->ne_mount_to) { 2995b9238976Sth mi = e->ne_mount; 2996b9238976Sth vfsp = mi->mi_vfsp; 2997b9238976Sth 2998b9238976Sth /* 2999b9238976Sth * Cleared by umount2_engine. 3000b9238976Sth */ 3001b87f76edSThomas Haynes if (vfsp != NULL) 3002b87f76edSThomas Haynes VFS_HOLD(vfsp); 3003b9238976Sth 3004b9238976Sth /* 3005b9238976Sth * Note that we effectively work down to the 3006b9238976Sth * leaf nodes first, try to unmount them, 3007b9238976Sth * then work our way back up into the leaf 3008b9238976Sth * nodes. 3009b9238976Sth * 3010b9238976Sth * Also note that we deal with a lot of 3011b9238976Sth * complexity by sharing the work with 3012b9238976Sth * the manual unmount code. 3013b9238976Sth */ 3014b9238976Sth nfs4_ephemeral_record_umount(vfsp, flag, 3015b9238976Sth e, prior); 3016b9238976Sth } 3017b9238976Sth 3018b9238976Sth e = prior; 3019b9238976Sth } 3020b9238976Sth 3021b9238976Sth check_done: 3022b9238976Sth 3023b9238976Sth /* 3024d3a14591SThomas Haynes * At this point we are done processing this tree. 3025d3a14591SThomas Haynes * 30262f172c55SRobert Thurlow * If the tree is invalid and we were the only reference 3027d3a14591SThomas Haynes * to it, then we push it on the local linked list 3028d3a14591SThomas Haynes * to remove it at the end. We avoid that action now 3029d3a14591SThomas Haynes * to keep the tree processing going along at a fair clip. 3030d3a14591SThomas Haynes * 30312f172c55SRobert Thurlow * Else, even if we were the only reference, we 30322f172c55SRobert Thurlow * allow it to be reused as needed. 3033b9238976Sth */ 3034b9238976Sth mutex_enter(&net->net_cnt_lock); 30352f172c55SRobert Thurlow nfs4_ephemeral_tree_decr(net); 30362f172c55SRobert Thurlow if (net->net_refcnt == 0 && 3037b9238976Sth net->net_status & NFS4_EPHEMERAL_TREE_INVALID) { 3038b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_LOCKED; 3039b9238976Sth mutex_exit(&net->net_cnt_lock); 3040b9238976Sth mutex_exit(&net->net_tree_lock); 3041b9238976Sth 3042b9238976Sth if (prev) 3043b9238976Sth prev->net_next = net->net_next; 3044b9238976Sth else 3045b9238976Sth ntg->ntg_forest = net->net_next; 3046b9238976Sth 3047b9238976Sth net->net_next = harvest; 3048b9238976Sth harvest = net; 3049b87f76edSThomas Haynes 3050b87f76edSThomas Haynes VFS_RELE(net->net_mount->mi_vfsp); 3051b87f76edSThomas Haynes MI4_RELE(net->net_mount); 3052b87f76edSThomas Haynes 3053b9238976Sth continue; 3054b9238976Sth } 3055b9238976Sth 3056b9238976Sth net->net_status &= ~NFS4_EPHEMERAL_TREE_LOCKED; 3057b9238976Sth mutex_exit(&net->net_cnt_lock); 3058b9238976Sth mutex_exit(&net->net_tree_lock); 3059b9238976Sth 3060b9238976Sth prev = net; 3061b9238976Sth } 3062b9238976Sth mutex_exit(&ntg->ntg_forest_lock); 3063b9238976Sth 3064b9238976Sth for (net = harvest; net != NULL; net = next) { 3065b9238976Sth next = net->net_next; 3066b9238976Sth 3067b9238976Sth mutex_destroy(&net->net_tree_lock); 3068b9238976Sth mutex_destroy(&net->net_cnt_lock); 3069b9238976Sth kmem_free(net, sizeof (*net)); 3070b9238976Sth } 3071b9238976Sth } 3072b9238976Sth 3073b9238976Sth /* 3074b9238976Sth * This is the thread which decides when the harvesting 3075b9238976Sth * can proceed and when to kill it off for this zone. 3076b9238976Sth */ 3077b9238976Sth static void 3078b9238976Sth nfs4_ephemeral_harvester(nfs4_trigger_globals_t *ntg) 3079b9238976Sth { 3080b9238976Sth clock_t timeleft; 3081b9238976Sth zone_t *zone = curproc->p_zone; 3082b9238976Sth 3083b9238976Sth for (;;) { 3084d3d50737SRafael Vanoni timeleft = zone_status_timedwait(zone, ddi_get_lbolt() + 3085b9238976Sth nfs4_trigger_thread_timer * hz, ZONE_IS_SHUTTING_DOWN); 3086b9238976Sth 3087b9238976Sth /* 3088b9238976Sth * zone is exiting... 3089b9238976Sth */ 3090b9238976Sth if (timeleft != -1) { 3091b9238976Sth ASSERT(zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN); 3092b9238976Sth zthread_exit(); 3093b9238976Sth /* NOTREACHED */ 3094b9238976Sth } 3095b9238976Sth 3096b9238976Sth /* 3097b9238976Sth * Only bother scanning if there is potential 3098b9238976Sth * work to be done. 3099b9238976Sth */ 3100b9238976Sth if (ntg->ntg_forest == NULL) 3101b9238976Sth continue; 3102b9238976Sth 3103b9238976Sth /* 3104b9238976Sth * Now scan the list and get rid of everything which 3105b9238976Sth * is old. 3106b9238976Sth */ 3107b9238976Sth nfs4_ephemeral_harvest_forest(ntg, FALSE, TRUE); 3108b9238976Sth } 3109b9238976Sth 3110b9238976Sth /* NOTREACHED */ 3111b9238976Sth } 3112b9238976Sth 3113b9238976Sth /* 3114b9238976Sth * The zone specific glue needed to start the unmount harvester. 3115b9238976Sth * 3116b9238976Sth * Note that we want to avoid holding the mutex as long as possible, 3117b9238976Sth * hence the multiple checks. 3118b9238976Sth * 3119b9238976Sth * The caller should avoid us getting down here in the first 3120b9238976Sth * place. 3121b9238976Sth */ 3122b9238976Sth static void 3123b9238976Sth nfs4_ephemeral_start_harvester(nfs4_trigger_globals_t *ntg) 3124b9238976Sth { 3125b9238976Sth /* 3126b9238976Sth * It got started before we got here... 3127b9238976Sth */ 3128b9238976Sth if (ntg->ntg_thread_started) 3129b9238976Sth return; 3130b9238976Sth 3131b9238976Sth mutex_enter(&nfs4_ephemeral_thread_lock); 3132b9238976Sth 3133b9238976Sth if (ntg->ntg_thread_started) { 3134b9238976Sth mutex_exit(&nfs4_ephemeral_thread_lock); 3135b9238976Sth return; 3136b9238976Sth } 3137b9238976Sth 3138b9238976Sth /* 3139b9238976Sth * Start the unmounter harvester thread for this zone. 3140b9238976Sth */ 3141b9238976Sth (void) zthread_create(NULL, 0, nfs4_ephemeral_harvester, 3142b9238976Sth ntg, 0, minclsyspri); 3143b9238976Sth 3144b9238976Sth ntg->ntg_thread_started = TRUE; 3145b9238976Sth mutex_exit(&nfs4_ephemeral_thread_lock); 3146b9238976Sth } 3147b9238976Sth 3148b9238976Sth /*ARGSUSED*/ 3149b9238976Sth static void * 3150b9238976Sth nfs4_ephemeral_zsd_create(zoneid_t zoneid) 3151b9238976Sth { 3152b9238976Sth nfs4_trigger_globals_t *ntg; 3153b9238976Sth 3154b9238976Sth ntg = kmem_zalloc(sizeof (*ntg), KM_SLEEP); 3155b9238976Sth ntg->ntg_thread_started = FALSE; 3156b9238976Sth 3157b9238976Sth /* 3158b9238976Sth * This is the default.... 3159b9238976Sth */ 3160*cb5401f4SMarcel Telka ntg->ntg_mount_to = nfs4_trigger_mount_to; 3161b9238976Sth 3162b9238976Sth mutex_init(&ntg->ntg_forest_lock, NULL, 3163b9238976Sth MUTEX_DEFAULT, NULL); 3164b9238976Sth 3165b9238976Sth return (ntg); 3166b9238976Sth } 3167b9238976Sth 3168b9238976Sth /* 3169b9238976Sth * Try a nice gentle walk down the forest and convince 3170b9238976Sth * all of the trees to gracefully give it up. 3171b9238976Sth */ 3172b9238976Sth /*ARGSUSED*/ 3173b9238976Sth static void 3174b9238976Sth nfs4_ephemeral_zsd_shutdown(zoneid_t zoneid, void *arg) 3175b9238976Sth { 3176b9238976Sth nfs4_trigger_globals_t *ntg = arg; 3177b9238976Sth 3178b9238976Sth if (!ntg) 3179b9238976Sth return; 3180b9238976Sth 3181b9238976Sth nfs4_ephemeral_harvest_forest(ntg, FALSE, FALSE); 3182b9238976Sth } 3183b9238976Sth 3184b9238976Sth /* 3185b9238976Sth * Race along the forest and rip all of the trees out by 3186b9238976Sth * their rootballs! 3187b9238976Sth */ 3188b9238976Sth /*ARGSUSED*/ 3189b9238976Sth static void 3190b9238976Sth nfs4_ephemeral_zsd_destroy(zoneid_t zoneid, void *arg) 3191b9238976Sth { 3192b9238976Sth nfs4_trigger_globals_t *ntg = arg; 3193b9238976Sth 3194b9238976Sth if (!ntg) 3195b9238976Sth return; 3196b9238976Sth 3197b9238976Sth nfs4_ephemeral_harvest_forest(ntg, TRUE, FALSE); 3198b9238976Sth 3199b9238976Sth mutex_destroy(&ntg->ntg_forest_lock); 3200b9238976Sth kmem_free(ntg, sizeof (*ntg)); 3201b9238976Sth } 3202b9238976Sth 3203b9238976Sth /* 3204b9238976Sth * This is the zone independent cleanup needed for 3205b9238976Sth * emphemeral mount processing. 3206b9238976Sth */ 3207b9238976Sth void 3208b9238976Sth nfs4_ephemeral_fini(void) 3209b9238976Sth { 3210b9238976Sth (void) zone_key_delete(nfs4_ephemeral_key); 3211b9238976Sth mutex_destroy(&nfs4_ephemeral_thread_lock); 3212b9238976Sth } 3213b9238976Sth 3214b9238976Sth /* 3215b9238976Sth * This is the zone independent initialization needed for 3216b9238976Sth * emphemeral mount processing. 3217b9238976Sth */ 3218b9238976Sth void 3219b9238976Sth nfs4_ephemeral_init(void) 3220b9238976Sth { 3221b9238976Sth mutex_init(&nfs4_ephemeral_thread_lock, NULL, MUTEX_DEFAULT, 3222b9238976Sth NULL); 3223b9238976Sth 3224b9238976Sth zone_key_create(&nfs4_ephemeral_key, nfs4_ephemeral_zsd_create, 3225b9238976Sth nfs4_ephemeral_zsd_shutdown, nfs4_ephemeral_zsd_destroy); 3226b9238976Sth } 3227b9238976Sth 3228b9238976Sth /* 3229b9238976Sth * nfssys() calls this function to set the per-zone 3230b9238976Sth * value of mount_to to drive when an ephemeral mount is 3231b9238976Sth * timed out. Each mount will grab a copy of this value 3232b9238976Sth * when mounted. 3233b9238976Sth */ 3234b9238976Sth void 3235b9238976Sth nfs4_ephemeral_set_mount_to(uint_t mount_to) 3236b9238976Sth { 3237b9238976Sth nfs4_trigger_globals_t *ntg; 3238b9238976Sth zone_t *zone = curproc->p_zone; 3239b9238976Sth 3240b9238976Sth ntg = zone_getspecific(nfs4_ephemeral_key, zone); 3241b9238976Sth 3242b9238976Sth ntg->ntg_mount_to = mount_to; 3243b9238976Sth } 3244b9238976Sth 3245b9238976Sth /* 3246b9238976Sth * Walk the list of v4 mount options; if they are currently set in vfsp, 3247b9238976Sth * append them to a new comma-separated mount option string, and return it. 3248b9238976Sth * 3249b9238976Sth * Caller should free by calling nfs4_trigger_destroy_mntopts(). 3250b9238976Sth */ 3251b9238976Sth static char * 3252b9238976Sth nfs4_trigger_create_mntopts(vfs_t *vfsp) 3253b9238976Sth { 3254b9238976Sth uint_t i; 3255b9238976Sth char *mntopts; 3256b9238976Sth struct vfssw *vswp; 3257b9238976Sth mntopts_t *optproto; 3258b9238976Sth 3259b9238976Sth mntopts = kmem_zalloc(MAX_MNTOPT_STR, KM_SLEEP); 3260b9238976Sth 3261b9238976Sth /* get the list of applicable mount options for v4; locks *vswp */ 3262b9238976Sth vswp = vfs_getvfssw(MNTTYPE_NFS4); 3263b9238976Sth optproto = &vswp->vsw_optproto; 3264b9238976Sth 3265b9238976Sth for (i = 0; i < optproto->mo_count; i++) { 3266b9238976Sth struct mntopt *mop = &optproto->mo_list[i]; 3267b9238976Sth 3268b9238976Sth if (mop->mo_flags & MO_EMPTY) 3269b9238976Sth continue; 3270b9238976Sth 3271b9238976Sth if (nfs4_trigger_add_mntopt(mntopts, mop->mo_name, vfsp)) { 3272b9238976Sth kmem_free(mntopts, MAX_MNTOPT_STR); 3273b9238976Sth vfs_unrefvfssw(vswp); 3274b9238976Sth return (NULL); 3275b9238976Sth } 3276b9238976Sth } 3277b9238976Sth 3278b9238976Sth vfs_unrefvfssw(vswp); 3279b9238976Sth 3280b9238976Sth /* 3281b9238976Sth * MNTOPT_XATTR is not in the v4 mount opt proto list, 3282b9238976Sth * and it may only be passed via MS_OPTIONSTR, so we 3283b9238976Sth * must handle it here. 3284b9238976Sth * 3285b9238976Sth * Ideally, it would be in the list, but NFS does not specify its 3286b9238976Sth * own opt proto list, it uses instead the default one. Since 3287b9238976Sth * not all filesystems support extended attrs, it would not be 3288b9238976Sth * appropriate to add it there. 3289b9238976Sth */ 3290b9238976Sth if (nfs4_trigger_add_mntopt(mntopts, MNTOPT_XATTR, vfsp) || 3291b9238976Sth nfs4_trigger_add_mntopt(mntopts, MNTOPT_NOXATTR, vfsp)) { 3292b9238976Sth kmem_free(mntopts, MAX_MNTOPT_STR); 3293b9238976Sth return (NULL); 3294b9238976Sth } 3295b9238976Sth 3296b9238976Sth return (mntopts); 3297b9238976Sth } 3298b9238976Sth 3299b9238976Sth static void 3300b9238976Sth nfs4_trigger_destroy_mntopts(char *mntopts) 3301b9238976Sth { 3302b9238976Sth if (mntopts) 3303b9238976Sth kmem_free(mntopts, MAX_MNTOPT_STR); 3304b9238976Sth } 3305b9238976Sth 3306b9238976Sth /* 3307b9238976Sth * Check a single mount option (optname). Add to mntopts if it is set in VFS. 3308b9238976Sth */ 3309b9238976Sth static int 3310b9238976Sth nfs4_trigger_add_mntopt(char *mntopts, char *optname, vfs_t *vfsp) 3311b9238976Sth { 3312b9238976Sth if (mntopts == NULL || optname == NULL || vfsp == NULL) 3313b9238976Sth return (EINVAL); 3314b9238976Sth 3315b9238976Sth if (vfs_optionisset(vfsp, optname, NULL)) { 3316b9238976Sth size_t mntoptslen = strlen(mntopts); 3317b9238976Sth size_t optnamelen = strlen(optname); 3318b9238976Sth 3319b9238976Sth /* +1 for ',', +1 for NUL */ 3320b9238976Sth if (mntoptslen + optnamelen + 2 > MAX_MNTOPT_STR) 3321b9238976Sth return (EOVERFLOW); 3322b9238976Sth 3323b9238976Sth /* first or subsequent mount option? */ 3324b9238976Sth if (*mntopts != '\0') 3325b9238976Sth (void) strcat(mntopts, ","); 3326b9238976Sth 3327b9238976Sth (void) strcat(mntopts, optname); 3328b9238976Sth } 3329b9238976Sth 3330b9238976Sth return (0); 3331b9238976Sth } 3332b9238976Sth 3333b9238976Sth static enum clnt_stat 33342f172c55SRobert Thurlow nfs4_ping_server_common(struct knetconfig *knc, struct netbuf *addr, int nointr) 3335b9238976Sth { 33362f172c55SRobert Thurlow int retries; 3337b9238976Sth uint_t max_msgsize; 3338b9238976Sth enum clnt_stat status; 3339b9238976Sth CLIENT *cl; 3340b9238976Sth struct timeval timeout; 3341b9238976Sth 3342b9238976Sth /* as per recov_newserver() */ 3343b9238976Sth max_msgsize = 0; 3344b9238976Sth retries = 1; 3345b9238976Sth timeout.tv_sec = 2; 3346b9238976Sth timeout.tv_usec = 0; 3347b9238976Sth 33482f172c55SRobert Thurlow if (clnt_tli_kcreate(knc, addr, NFS_PROGRAM, NFS_V4, 33492f172c55SRobert Thurlow max_msgsize, retries, CRED(), &cl) != 0) 3350b9238976Sth return (RPC_FAILED); 3351b9238976Sth 3352b9238976Sth if (nointr) 3353b9238976Sth cl->cl_nosignal = TRUE; 3354b9238976Sth status = CLNT_CALL(cl, RFS_NULL, xdr_void, NULL, xdr_void, NULL, 3355b9238976Sth timeout); 3356b9238976Sth if (nointr) 3357b9238976Sth cl->cl_nosignal = FALSE; 3358b9238976Sth 3359b9238976Sth AUTH_DESTROY(cl->cl_auth); 3360b9238976Sth CLNT_DESTROY(cl); 3361b9238976Sth 3362b9238976Sth return (status); 3363b9238976Sth } 33642f172c55SRobert Thurlow 33652f172c55SRobert Thurlow static enum clnt_stat 33662f172c55SRobert Thurlow nfs4_trigger_ping_server(servinfo4_t *svp, int nointr) 33672f172c55SRobert Thurlow { 33682f172c55SRobert Thurlow return (nfs4_ping_server_common(svp->sv_knconf, &svp->sv_addr, nointr)); 33692f172c55SRobert Thurlow } 3370