Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/fs/nfsserver/nfs_nfsdport.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 */ 35 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include <sys/capsicum.h> 40 #include <sys/extattr.h> 41 42 /* 43 * Functions that perform the vfs operations required by the routines in 44 * nfsd_serv.c. It is hoped that this change will make the server more 45 * portable. 46 */ 47 48 #include <fs/nfs/nfsport.h> 49 #include <security/mac/mac_framework.h> 50 #include <sys/filio.h> 51 #include <sys/hash.h> 52 #include <sys/sysctl.h> 53 #include <nlm/nlm_prot.h> 54 #include <nlm/nlm.h> 55 56 FEATURE(nfsd, "NFSv4 server"); 57 58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 59 extern int nfsrv_useacl; 60 extern int newnfs_numnfsd; 61 extern struct mount nfsv4root_mnt; 62 extern struct nfsrv_stablefirst nfsrv_stablefirst; 63 extern void (*nfsd_call_servertimer)(void); 64 extern SVCPOOL *nfsrvd_pool; 65 extern struct nfsv4lock nfsd_suspend_lock; 66 extern struct nfsclienthashhead *nfsclienthash; 67 extern struct nfslockhashhead *nfslockhash; 68 extern struct nfssessionhash *nfssessionhash; 69 extern int nfsrv_sessionhashsize; 70 extern struct nfsstatsv1 nfsstatsv1; 71 extern struct nfslayouthash *nfslayouthash; 72 extern int nfsrv_layouthashsize; 73 extern struct mtx nfsrv_dslock_mtx; 74 extern int nfs_pnfsiothreads; 75 extern struct nfsdontlisthead nfsrv_dontlisthead; 76 extern volatile int nfsrv_dontlistlen; 77 extern volatile int nfsrv_devidcnt; 78 extern int nfsrv_maxpnfsmirror; 79 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt; 80 NFSDLOCKMUTEX; 81 NFSSTATESPINLOCK; 82 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE]; 83 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE]; 84 struct mtx nfsrc_udpmtx; 85 struct mtx nfs_v4root_mutex; 86 struct mtx nfsrv_dontlistlock_mtx; 87 struct mtx nfsrv_recalllock_mtx; 88 struct nfsrvfh nfs_rootfh, nfs_pubfh; 89 int nfs_pubfhset = 0, nfs_rootfhset = 0; 90 struct proc *nfsd_master_proc = NULL; 91 int nfsd_debuglevel = 0; 92 static pid_t nfsd_master_pid = (pid_t)-1; 93 static char nfsd_master_comm[MAXCOMLEN + 1]; 94 static struct timeval nfsd_master_start; 95 static uint32_t nfsv4_sysid = 0; 96 static fhandle_t zerofh; 97 98 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, 99 struct ucred *); 100 101 int nfsrv_enable_crossmntpt = 1; 102 static int nfs_commit_blks; 103 static int nfs_commit_miss; 104 extern int nfsrv_issuedelegs; 105 extern int nfsrv_dolocallocks; 106 extern int nfsd_enable_stringtouid; 107 extern struct nfsdevicehead nfsrv_devidhead; 108 109 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, 110 struct iovec **); 111 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **, 112 struct mbuf **, struct iovec **); 113 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, 114 int *); 115 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, 116 NFSPROC_T *); 117 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, 118 int *, char *, fhandle_t *); 119 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, 120 NFSPROC_T *); 121 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, 122 struct thread *, int, struct mbuf **, char *, struct mbuf **, 123 struct nfsvattr *, struct acl *, off_t *, int, bool *); 124 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); 125 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, 126 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); 127 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, 128 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, 129 char *, int *); 130 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 131 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 132 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 133 struct vnode *, struct nfsmount **, int, struct acl *, int *); 134 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 135 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); 136 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 137 struct vnode *, struct nfsmount *, struct nfsvattr *); 138 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, 139 NFSPROC_T *, struct nfsmount *); 140 static int nfsrv_putfhname(fhandle_t *, char *); 141 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, 142 struct pnfsdsfile *, struct vnode **, NFSPROC_T *); 143 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, 144 struct vnode *, NFSPROC_T *); 145 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); 146 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, 147 NFSPROC_T *); 148 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); 149 150 int nfs_pnfsio(task_fn_t *, void *); 151 152 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 153 "NFS server"); 154 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, 155 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); 156 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 157 0, ""); 158 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 159 0, ""); 160 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, 161 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); 162 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, 163 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); 164 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 165 0, "Debug level for NFS server"); 166 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW, 167 &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names"); 168 static int nfsrv_pnfsgetdsattr = 1; 169 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, 170 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); 171 172 /* 173 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are 174 * not running. 175 * The dsN subdirectories for the increased values must have been created 176 * on all DS servers before this increase is done. 177 */ 178 u_int nfsrv_dsdirsize = 20; 179 static int 180 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) 181 { 182 int error, newdsdirsize; 183 184 newdsdirsize = nfsrv_dsdirsize; 185 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); 186 if (error != 0 || req->newptr == NULL) 187 return (error); 188 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || 189 newnfs_numnfsd != 0) 190 return (EINVAL); 191 nfsrv_dsdirsize = newdsdirsize; 192 return (0); 193 } 194 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, 195 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), 196 sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); 197 198 #define MAX_REORDERED_RPC 16 199 #define NUM_HEURISTIC 1031 200 #define NHUSE_INIT 64 201 #define NHUSE_INC 16 202 #define NHUSE_MAX 2048 203 204 static struct nfsheur { 205 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 206 off_t nh_nextoff; /* next offset for sequential detection */ 207 int nh_use; /* use count for selection */ 208 int nh_seqcount; /* heuristic */ 209 } nfsheur[NUM_HEURISTIC]; 210 211 /* 212 * Heuristic to detect sequential operation. 213 */ 214 static struct nfsheur * 215 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 216 { 217 struct nfsheur *nh; 218 int hi, try; 219 220 /* Locate best candidate. */ 221 try = 32; 222 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 223 nh = &nfsheur[hi]; 224 while (try--) { 225 if (nfsheur[hi].nh_vp == vp) { 226 nh = &nfsheur[hi]; 227 break; 228 } 229 if (nfsheur[hi].nh_use > 0) 230 --nfsheur[hi].nh_use; 231 hi = (hi + 1) % NUM_HEURISTIC; 232 if (nfsheur[hi].nh_use < nh->nh_use) 233 nh = &nfsheur[hi]; 234 } 235 236 /* Initialize hint if this is a new file. */ 237 if (nh->nh_vp != vp) { 238 nh->nh_vp = vp; 239 nh->nh_nextoff = uio->uio_offset; 240 nh->nh_use = NHUSE_INIT; 241 if (uio->uio_offset == 0) 242 nh->nh_seqcount = 4; 243 else 244 nh->nh_seqcount = 1; 245 } 246 247 /* Calculate heuristic. */ 248 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 249 uio->uio_offset == nh->nh_nextoff) { 250 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 251 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 252 if (nh->nh_seqcount > IO_SEQMAX) 253 nh->nh_seqcount = IO_SEQMAX; 254 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 255 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 256 /* Probably a reordered RPC, leave seqcount alone. */ 257 } else if (nh->nh_seqcount > 1) { 258 nh->nh_seqcount /= 2; 259 } else { 260 nh->nh_seqcount = 0; 261 } 262 nh->nh_use += NHUSE_INC; 263 if (nh->nh_use > NHUSE_MAX) 264 nh->nh_use = NHUSE_MAX; 265 return (nh); 266 } 267 268 /* 269 * Get attributes into nfsvattr structure. 270 */ 271 int 272 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, 273 struct nfsrv_descript *nd, struct thread *p, int vpislocked, 274 nfsattrbit_t *attrbitp) 275 { 276 int error, gotattr, lockedit = 0; 277 struct nfsvattr na; 278 279 if (vpislocked == 0) { 280 /* 281 * When vpislocked == 0, the vnode is either exclusively 282 * locked by this thread or not locked by this thread. 283 * As such, shared lock it, if not exclusively locked. 284 */ 285 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 286 lockedit = 1; 287 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); 288 } 289 } 290 291 /* 292 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed 293 * attributes, as required. 294 * This needs to be done for regular files if: 295 * - non-NFSv4 RPCs or 296 * - when attrbitp == NULL or 297 * - an NFSv4 RPC with any of the above attributes in attrbitp. 298 * A return of 0 for nfsrv_proxyds() indicates that it has acquired 299 * these attributes. nfsrv_proxyds() will return an error if the 300 * server is not a pNFS one. 301 */ 302 gotattr = 0; 303 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || 304 (nd->nd_flag & ND_NFSV4) == 0 || 305 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || 306 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || 307 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || 308 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || 309 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { 310 error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, 311 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, 312 NULL); 313 if (error == 0) 314 gotattr = 1; 315 } 316 317 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); 318 if (lockedit != 0) 319 NFSVOPUNLOCK(vp); 320 321 /* 322 * If we got the Change, Size and Modify Time from the DS, 323 * replace them. 324 */ 325 if (gotattr != 0) { 326 nvap->na_atime = na.na_atime; 327 nvap->na_mtime = na.na_mtime; 328 nvap->na_filerev = na.na_filerev; 329 nvap->na_size = na.na_size; 330 nvap->na_bytes = na.na_bytes; 331 } 332 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, 333 error, (uintmax_t)na.na_filerev); 334 335 NFSEXITCODE(error); 336 return (error); 337 } 338 339 /* 340 * Get a file handle for a vnode. 341 */ 342 int 343 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) 344 { 345 int error; 346 347 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); 348 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 349 error = VOP_VPTOFH(vp, &fhp->fh_fid); 350 351 NFSEXITCODE(error); 352 return (error); 353 } 354 355 /* 356 * Perform access checking for vnodes obtained from file handles that would 357 * refer to files already opened by a Unix client. You cannot just use 358 * vn_writechk() and VOP_ACCESSX() for two reasons. 359 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 360 * case. 361 * 2 - The owner is to be given access irrespective of mode bits for some 362 * operations, so that processes that chmod after opening a file don't 363 * break. 364 */ 365 int 366 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, 367 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, 368 u_int32_t *supportedtypep) 369 { 370 struct vattr vattr; 371 int error = 0, getret = 0; 372 373 if (vpislocked == 0) { 374 if (NFSVOPLOCK(vp, LK_SHARED) != 0) { 375 error = EPERM; 376 goto out; 377 } 378 } 379 if (accmode & VWRITE) { 380 /* Just vn_writechk() changed to check rdonly */ 381 /* 382 * Disallow write attempts on read-only file systems; 383 * unless the file is a socket or a block or character 384 * device resident on the file system. 385 */ 386 if (NFSVNO_EXRDONLY(exp) || 387 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 388 switch (vp->v_type) { 389 case VREG: 390 case VDIR: 391 case VLNK: 392 error = EROFS; 393 default: 394 break; 395 } 396 } 397 /* 398 * If there's shared text associated with 399 * the inode, try to free it up once. If 400 * we fail, we can't allow writing. 401 */ 402 if (VOP_IS_TEXT(vp) && error == 0) 403 error = ETXTBSY; 404 } 405 if (error != 0) { 406 if (vpislocked == 0) 407 NFSVOPUNLOCK(vp); 408 goto out; 409 } 410 411 /* 412 * Should the override still be applied when ACLs are enabled? 413 */ 414 error = VOP_ACCESSX(vp, accmode, cred, p); 415 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { 416 /* 417 * Try again with VEXPLICIT_DENY, to see if the test for 418 * deletion is supported. 419 */ 420 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); 421 if (error == 0) { 422 if (vp->v_type == VDIR) { 423 accmode &= ~(VDELETE | VDELETE_CHILD); 424 accmode |= VWRITE; 425 error = VOP_ACCESSX(vp, accmode, cred, p); 426 } else if (supportedtypep != NULL) { 427 *supportedtypep &= ~NFSACCESS_DELETE; 428 } 429 } 430 } 431 432 /* 433 * Allow certain operations for the owner (reads and writes 434 * on files that are already open). 435 */ 436 if (override != NFSACCCHK_NOOVERRIDE && 437 (error == EPERM || error == EACCES)) { 438 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) 439 error = 0; 440 else if (override & NFSACCCHK_ALLOWOWNER) { 441 getret = VOP_GETATTR(vp, &vattr, cred); 442 if (getret == 0 && cred->cr_uid == vattr.va_uid) 443 error = 0; 444 } 445 } 446 if (vpislocked == 0) 447 NFSVOPUNLOCK(vp); 448 449 out: 450 NFSEXITCODE(error); 451 return (error); 452 } 453 454 /* 455 * Set attribute(s) vnop. 456 */ 457 int 458 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 459 struct thread *p, struct nfsexstuff *exp) 460 { 461 u_quad_t savsize = 0; 462 int error, savedit; 463 time_t savbtime; 464 465 /* 466 * If this is an exported file system and a pNFS service is running, 467 * don't VOP_SETATTR() of size for the MDS file system. 468 */ 469 savedit = 0; 470 error = 0; 471 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && 472 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && 473 nvap->na_vattr.va_size > 0) { 474 savsize = nvap->na_vattr.va_size; 475 nvap->na_vattr.va_size = VNOVAL; 476 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 477 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 478 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 479 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 480 nvap->na_vattr.va_mtime.tv_sec != VNOVAL) 481 savedit = 1; 482 else 483 savedit = 2; 484 } 485 if (savedit != 2) 486 error = VOP_SETATTR(vp, &nvap->na_vattr, cred); 487 if (savedit != 0) 488 nvap->na_vattr.va_size = savsize; 489 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 490 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 491 nvap->na_vattr.va_size != VNOVAL || 492 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 493 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 494 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { 495 /* Never modify birthtime on a DS file. */ 496 savbtime = nvap->na_vattr.va_birthtime.tv_sec; 497 nvap->na_vattr.va_birthtime.tv_sec = VNOVAL; 498 /* For a pNFS server, set the attributes on the DS file. */ 499 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, 500 NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); 501 nvap->na_vattr.va_birthtime.tv_sec = savbtime; 502 if (error == ENOENT) 503 error = 0; 504 } 505 NFSEXITCODE(error); 506 return (error); 507 } 508 509 /* 510 * Set up nameidata for a lookup() call and do it. 511 */ 512 int 513 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, 514 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p, 515 struct vnode **retdirp) 516 { 517 struct componentname *cnp = &ndp->ni_cnd; 518 int i; 519 struct iovec aiov; 520 struct uio auio; 521 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; 522 int error = 0; 523 char *cp; 524 525 *retdirp = NULL; 526 cnp->cn_nameptr = cnp->cn_pnbuf; 527 ndp->ni_lcf = 0; 528 /* 529 * Extract and set starting directory. 530 */ 531 if (dp->v_type != VDIR) { 532 if (islocked) 533 vput(dp); 534 else 535 vrele(dp); 536 nfsvno_relpathbuf(ndp); 537 error = ENOTDIR; 538 goto out1; 539 } 540 if (islocked) 541 NFSVOPUNLOCK(dp); 542 VREF(dp); 543 *retdirp = dp; 544 if (NFSVNO_EXRDONLY(exp)) 545 cnp->cn_flags |= RDONLY; 546 ndp->ni_segflg = UIO_SYSSPACE; 547 548 if (nd->nd_flag & ND_PUBLOOKUP) { 549 ndp->ni_loopcnt = 0; 550 if (cnp->cn_pnbuf[0] == '/') { 551 vrele(dp); 552 /* 553 * Check for degenerate pathnames here, since lookup() 554 * panics on them. 555 */ 556 for (i = 1; i < ndp->ni_pathlen; i++) 557 if (cnp->cn_pnbuf[i] != '/') 558 break; 559 if (i == ndp->ni_pathlen) { 560 error = NFSERR_ACCES; 561 goto out; 562 } 563 dp = rootvnode; 564 VREF(dp); 565 } 566 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || 567 (nd->nd_flag & ND_NFSV4) == 0) { 568 /* 569 * Only cross mount points for NFSv4 when doing a 570 * mount while traversing the file system above 571 * the mount point, unless nfsrv_enable_crossmntpt is set. 572 */ 573 cnp->cn_flags |= NOCROSSMOUNT; 574 } 575 576 /* 577 * Initialize for scan, set ni_startdir and bump ref on dp again 578 * because lookup() will dereference ni_startdir. 579 */ 580 581 cnp->cn_thread = p; 582 ndp->ni_startdir = dp; 583 ndp->ni_rootdir = rootvnode; 584 ndp->ni_topdir = NULL; 585 586 if (!lockleaf) 587 cnp->cn_flags |= LOCKLEAF; 588 for (;;) { 589 cnp->cn_nameptr = cnp->cn_pnbuf; 590 /* 591 * Call lookup() to do the real work. If an error occurs, 592 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and 593 * we do not have to dereference anything before returning. 594 * In either case ni_startdir will be dereferenced and NULLed 595 * out. 596 */ 597 error = lookup(ndp); 598 if (error) 599 break; 600 601 /* 602 * Check for encountering a symbolic link. Trivial 603 * termination occurs if no symlink encountered. 604 */ 605 if ((cnp->cn_flags & ISSYMLINK) == 0) { 606 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) 607 nfsvno_relpathbuf(ndp); 608 if (ndp->ni_vp && !lockleaf) 609 NFSVOPUNLOCK(ndp->ni_vp); 610 break; 611 } 612 613 /* 614 * Validate symlink 615 */ 616 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) 617 NFSVOPUNLOCK(ndp->ni_dvp); 618 if (!(nd->nd_flag & ND_PUBLOOKUP)) { 619 error = EINVAL; 620 goto badlink2; 621 } 622 623 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 624 error = ELOOP; 625 goto badlink2; 626 } 627 if (ndp->ni_pathlen > 1) 628 cp = uma_zalloc(namei_zone, M_WAITOK); 629 else 630 cp = cnp->cn_pnbuf; 631 aiov.iov_base = cp; 632 aiov.iov_len = MAXPATHLEN; 633 auio.uio_iov = &aiov; 634 auio.uio_iovcnt = 1; 635 auio.uio_offset = 0; 636 auio.uio_rw = UIO_READ; 637 auio.uio_segflg = UIO_SYSSPACE; 638 auio.uio_td = NULL; 639 auio.uio_resid = MAXPATHLEN; 640 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 641 if (error) { 642 badlink1: 643 if (ndp->ni_pathlen > 1) 644 uma_zfree(namei_zone, cp); 645 badlink2: 646 vrele(ndp->ni_dvp); 647 vput(ndp->ni_vp); 648 break; 649 } 650 linklen = MAXPATHLEN - auio.uio_resid; 651 if (linklen == 0) { 652 error = ENOENT; 653 goto badlink1; 654 } 655 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 656 error = ENAMETOOLONG; 657 goto badlink1; 658 } 659 660 /* 661 * Adjust or replace path 662 */ 663 if (ndp->ni_pathlen > 1) { 664 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 665 uma_zfree(namei_zone, cnp->cn_pnbuf); 666 cnp->cn_pnbuf = cp; 667 } else 668 cnp->cn_pnbuf[linklen] = '\0'; 669 ndp->ni_pathlen += linklen; 670 671 /* 672 * Cleanup refs for next loop and check if root directory 673 * should replace current directory. Normally ni_dvp 674 * becomes the new base directory and is cleaned up when 675 * we loop. Explicitly null pointers after invalidation 676 * to clarify operation. 677 */ 678 vput(ndp->ni_vp); 679 ndp->ni_vp = NULL; 680 681 if (cnp->cn_pnbuf[0] == '/') { 682 vrele(ndp->ni_dvp); 683 ndp->ni_dvp = ndp->ni_rootdir; 684 VREF(ndp->ni_dvp); 685 } 686 ndp->ni_startdir = ndp->ni_dvp; 687 ndp->ni_dvp = NULL; 688 } 689 if (!lockleaf) 690 cnp->cn_flags &= ~LOCKLEAF; 691 692 out: 693 if (error) { 694 nfsvno_relpathbuf(ndp); 695 ndp->ni_vp = NULL; 696 ndp->ni_dvp = NULL; 697 ndp->ni_startdir = NULL; 698 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { 699 ndp->ni_dvp = NULL; 700 } 701 702 out1: 703 NFSEXITCODE2(error, nd); 704 return (error); 705 } 706 707 /* 708 * Set up a pathname buffer and return a pointer to it and, optionally 709 * set a hash pointer. 710 */ 711 void 712 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) 713 { 714 struct componentname *cnp = &ndp->ni_cnd; 715 716 cnp->cn_flags |= (NOMACCHECK | HASBUF); 717 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 718 if (hashpp != NULL) 719 *hashpp = NULL; 720 *bufpp = cnp->cn_pnbuf; 721 } 722 723 /* 724 * Release the above path buffer, if not released by nfsvno_namei(). 725 */ 726 void 727 nfsvno_relpathbuf(struct nameidata *ndp) 728 { 729 730 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0) 731 panic("nfsrelpath"); 732 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 733 ndp->ni_cnd.cn_flags &= ~HASBUF; 734 } 735 736 /* 737 * Readlink vnode op into an mbuf list. 738 */ 739 int 740 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz, 741 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 742 { 743 struct iovec *iv; 744 struct uio io, *uiop = &io; 745 struct mbuf *mp, *mp3; 746 int len, tlen, error = 0; 747 748 len = NFS_MAXPATHLEN; 749 if (maxextsiz > 0) 750 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 751 &mp3, &mp, &iv); 752 else 753 uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); 754 uiop->uio_iov = iv; 755 uiop->uio_offset = 0; 756 uiop->uio_resid = len; 757 uiop->uio_rw = UIO_READ; 758 uiop->uio_segflg = UIO_SYSSPACE; 759 uiop->uio_td = NULL; 760 error = VOP_READLINK(vp, uiop, cred); 761 free(iv, M_TEMP); 762 if (error) { 763 m_freem(mp3); 764 *lenp = 0; 765 goto out; 766 } 767 if (uiop->uio_resid > 0) { 768 len -= uiop->uio_resid; 769 tlen = NFSM_RNDUP(len); 770 if (tlen == 0) { 771 m_freem(mp3); 772 mp3 = mp = NULL; 773 } else if (tlen != NFS_MAXPATHLEN || tlen != len) 774 mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, 775 tlen - len); 776 } 777 *lenp = len; 778 *mpp = mp3; 779 *mpendp = mp; 780 781 out: 782 NFSEXITCODE(error); 783 return (error); 784 } 785 786 /* 787 * Create an mbuf chain and an associated iovec that can be used to Read 788 * or Getextattr of data. 789 * Upon success, return pointers to the first and last mbufs in the chain 790 * plus the malloc'd iovec and its iovlen. 791 */ 792 static int 793 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, 794 struct iovec **ivp) 795 { 796 struct mbuf *m, *m2 = NULL, *m3; 797 struct iovec *iv; 798 int i, left, siz; 799 800 left = len; 801 m3 = NULL; 802 /* 803 * Generate the mbuf list with the uio_iov ref. to it. 804 */ 805 i = 0; 806 while (left > 0) { 807 NFSMGET(m); 808 MCLGET(m, M_WAITOK); 809 m->m_len = 0; 810 siz = min(M_TRAILINGSPACE(m), left); 811 left -= siz; 812 i++; 813 if (m3) 814 m2->m_next = m; 815 else 816 m3 = m; 817 m2 = m; 818 } 819 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 820 m = m3; 821 left = len; 822 i = 0; 823 while (left > 0) { 824 if (m == NULL) 825 panic("nfsrv_createiovec iov"); 826 siz = min(M_TRAILINGSPACE(m), left); 827 if (siz > 0) { 828 iv->iov_base = mtod(m, caddr_t) + m->m_len; 829 iv->iov_len = siz; 830 m->m_len += siz; 831 left -= siz; 832 iv++; 833 i++; 834 } 835 m = m->m_next; 836 } 837 *mpp = m3; 838 *mpendp = m2; 839 return (i); 840 } 841 842 /* 843 * Create an mbuf chain and an associated iovec that can be used to Read 844 * or Getextattr of data. 845 * Upon success, return pointers to the first and last mbufs in the chain 846 * plus the malloc'd iovec and its iovlen. 847 * Same as above, but creates ext_pgs mbuf(s). 848 */ 849 static int 850 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp, 851 struct mbuf **mpendp, struct iovec **ivp) 852 { 853 struct mbuf *m, *m2 = NULL, *m3; 854 struct iovec *iv; 855 int i, left, pgno, siz; 856 857 left = len; 858 m3 = NULL; 859 /* 860 * Generate the mbuf list with the uio_iov ref. to it. 861 */ 862 i = 0; 863 while (left > 0) { 864 siz = min(left, maxextsiz); 865 m = mb_alloc_ext_plus_pages(siz, M_WAITOK); 866 left -= siz; 867 i += m->m_epg_npgs; 868 if (m3 != NULL) 869 m2->m_next = m; 870 else 871 m3 = m; 872 m2 = m; 873 } 874 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 875 m = m3; 876 left = len; 877 i = 0; 878 pgno = 0; 879 while (left > 0) { 880 if (m == NULL) 881 panic("nfsvno_createiovec_extpgs iov"); 882 siz = min(PAGE_SIZE, left); 883 if (siz > 0) { 884 iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); 885 iv->iov_len = siz; 886 m->m_len += siz; 887 if (pgno == m->m_epg_npgs - 1) 888 m->m_epg_last_len = siz; 889 left -= siz; 890 iv++; 891 i++; 892 pgno++; 893 } 894 if (pgno == m->m_epg_npgs && left > 0) { 895 m = m->m_next; 896 if (m == NULL) 897 panic("nfsvno_createiovec_extpgs iov"); 898 pgno = 0; 899 } 900 } 901 *mpp = m3; 902 *mpendp = m2; 903 return (i); 904 } 905 906 /* 907 * Read vnode op call into mbuf list. 908 */ 909 int 910 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 911 int maxextsiz, struct thread *p, struct mbuf **mpp, 912 struct mbuf **mpendp) 913 { 914 struct mbuf *m; 915 struct iovec *iv; 916 int error = 0, len, tlen, ioflag = 0; 917 struct mbuf *m3; 918 struct uio io, *uiop = &io; 919 struct nfsheur *nh; 920 921 /* 922 * Attempt to read from a DS file. A return of ENOENT implies 923 * there is no DS file to read. 924 */ 925 error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, 926 NULL, mpendp, NULL, NULL, NULL, 0, NULL); 927 if (error != ENOENT) 928 return (error); 929 930 len = NFSM_RNDUP(cnt); 931 if (maxextsiz > 0) 932 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 933 &m3, &m, &iv); 934 else 935 uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); 936 uiop->uio_iov = iv; 937 uiop->uio_offset = off; 938 uiop->uio_resid = len; 939 uiop->uio_rw = UIO_READ; 940 uiop->uio_segflg = UIO_SYSSPACE; 941 uiop->uio_td = NULL; 942 nh = nfsrv_sequential_heuristic(uiop, vp); 943 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 944 /* XXX KDM make this more systematic? */ 945 nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid; 946 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 947 free(iv, M_TEMP); 948 if (error) { 949 m_freem(m3); 950 *mpp = NULL; 951 goto out; 952 } 953 nh->nh_nextoff = uiop->uio_offset; 954 tlen = len - uiop->uio_resid; 955 cnt = cnt < tlen ? cnt : tlen; 956 tlen = NFSM_RNDUP(cnt); 957 if (tlen == 0) { 958 m_freem(m3); 959 m3 = m = NULL; 960 } else if (len != tlen || tlen != cnt) 961 m = nfsrv_adj(m3, len - tlen, tlen - cnt); 962 *mpp = m3; 963 *mpendp = m; 964 965 out: 966 NFSEXITCODE(error); 967 return (error); 968 } 969 970 /* 971 * Create the iovec for the mbuf chain passed in as an argument. 972 * The "cp" argument is where the data starts within the first mbuf in 973 * the chain. It returns the iovec and the iovcnt. 974 */ 975 static int 976 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, 977 int *iovcntp) 978 { 979 struct mbuf *mp; 980 struct iovec *ivp; 981 int cnt, i, len; 982 983 /* 984 * Loop through the mbuf chain, counting how many mbufs are a 985 * part of this write operation, so the iovec size is known. 986 */ 987 cnt = 0; 988 len = retlen; 989 mp = m; 990 i = mtod(mp, caddr_t) + mp->m_len - cp; 991 while (len > 0) { 992 if (i > 0) { 993 len -= i; 994 cnt++; 995 } 996 mp = mp->m_next; 997 if (!mp) { 998 if (len > 0) 999 return (EBADRPC); 1000 } else 1001 i = mp->m_len; 1002 } 1003 1004 /* Now, create the iovec. */ 1005 mp = m; 1006 *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, 1007 M_WAITOK); 1008 *iovcntp = cnt; 1009 i = mtod(mp, caddr_t) + mp->m_len - cp; 1010 len = retlen; 1011 while (len > 0) { 1012 if (mp == NULL) 1013 panic("nfsrv_createiovecw"); 1014 if (i > 0) { 1015 i = min(i, len); 1016 ivp->iov_base = cp; 1017 ivp->iov_len = i; 1018 ivp++; 1019 len -= i; 1020 } 1021 mp = mp->m_next; 1022 if (mp) { 1023 i = mp->m_len; 1024 cp = mtod(mp, caddr_t); 1025 } 1026 } 1027 return (0); 1028 } 1029 1030 /* 1031 * Write vnode op from an mbuf list. 1032 */ 1033 int 1034 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, 1035 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) 1036 { 1037 struct iovec *iv; 1038 int cnt, ioflags, error; 1039 struct uio io, *uiop = &io; 1040 struct nfsheur *nh; 1041 1042 /* 1043 * Attempt to write to a DS file. A return of ENOENT implies 1044 * there is no DS file to write. 1045 */ 1046 error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, 1047 &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); 1048 if (error != ENOENT) { 1049 *stable = NFSWRITE_FILESYNC; 1050 return (error); 1051 } 1052 1053 if (*stable == NFSWRITE_UNSTABLE) 1054 ioflags = IO_NODELOCKED; 1055 else 1056 ioflags = (IO_SYNC | IO_NODELOCKED); 1057 error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); 1058 if (error != 0) 1059 return (error); 1060 uiop->uio_iov = iv; 1061 uiop->uio_iovcnt = cnt; 1062 uiop->uio_resid = retlen; 1063 uiop->uio_rw = UIO_WRITE; 1064 uiop->uio_segflg = UIO_SYSSPACE; 1065 NFSUIOPROC(uiop, p); 1066 uiop->uio_offset = off; 1067 nh = nfsrv_sequential_heuristic(uiop, vp); 1068 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 1069 /* XXX KDM make this more systematic? */ 1070 nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; 1071 error = VOP_WRITE(vp, uiop, ioflags, cred); 1072 if (error == 0) 1073 nh->nh_nextoff = uiop->uio_offset; 1074 free(iv, M_TEMP); 1075 1076 NFSEXITCODE(error); 1077 return (error); 1078 } 1079 1080 /* 1081 * Common code for creating a regular file (plus special files for V2). 1082 */ 1083 int 1084 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, 1085 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, 1086 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) 1087 { 1088 u_quad_t tempsize; 1089 int error; 1090 struct thread *p = curthread; 1091 1092 error = nd->nd_repstat; 1093 if (!error && ndp->ni_vp == NULL) { 1094 if (nvap->na_type == VREG || nvap->na_type == VSOCK) { 1095 vrele(ndp->ni_startdir); 1096 error = VOP_CREATE(ndp->ni_dvp, 1097 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1098 /* For a pNFS server, create the data file on a DS. */ 1099 if (error == 0 && nvap->na_type == VREG) { 1100 /* 1101 * Create a data file on a DS for a pNFS server. 1102 * This function just returns if not 1103 * running a pNFS DS or the creation fails. 1104 */ 1105 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1106 nd->nd_cred, p); 1107 } 1108 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1109 NULL, false); 1110 nfsvno_relpathbuf(ndp); 1111 if (!error) { 1112 if (*exclusive_flagp) { 1113 *exclusive_flagp = 0; 1114 NFSVNO_ATTRINIT(nvap); 1115 nvap->na_atime.tv_sec = cverf[0]; 1116 nvap->na_atime.tv_nsec = cverf[1]; 1117 error = VOP_SETATTR(ndp->ni_vp, 1118 &nvap->na_vattr, nd->nd_cred); 1119 if (error != 0) { 1120 vput(ndp->ni_vp); 1121 ndp->ni_vp = NULL; 1122 error = NFSERR_NOTSUPP; 1123 } 1124 } 1125 } 1126 /* 1127 * NFS V2 Only. nfsrvd_mknod() does this for V3. 1128 * (This implies, just get out on an error.) 1129 */ 1130 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || 1131 nvap->na_type == VFIFO) { 1132 if (nvap->na_type == VCHR && rdev == 0xffffffff) 1133 nvap->na_type = VFIFO; 1134 if (nvap->na_type != VFIFO && 1135 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { 1136 vrele(ndp->ni_startdir); 1137 nfsvno_relpathbuf(ndp); 1138 vput(ndp->ni_dvp); 1139 goto out; 1140 } 1141 nvap->na_rdev = rdev; 1142 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1143 &ndp->ni_cnd, &nvap->na_vattr); 1144 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1145 NULL, false); 1146 nfsvno_relpathbuf(ndp); 1147 vrele(ndp->ni_startdir); 1148 if (error) 1149 goto out; 1150 } else { 1151 vrele(ndp->ni_startdir); 1152 nfsvno_relpathbuf(ndp); 1153 vput(ndp->ni_dvp); 1154 error = ENXIO; 1155 goto out; 1156 } 1157 *vpp = ndp->ni_vp; 1158 } else { 1159 /* 1160 * Handle cases where error is already set and/or 1161 * the file exists. 1162 * 1 - clean up the lookup 1163 * 2 - iff !error and na_size set, truncate it 1164 */ 1165 vrele(ndp->ni_startdir); 1166 nfsvno_relpathbuf(ndp); 1167 *vpp = ndp->ni_vp; 1168 if (ndp->ni_dvp == *vpp) 1169 vrele(ndp->ni_dvp); 1170 else 1171 vput(ndp->ni_dvp); 1172 if (!error && nvap->na_size != VNOVAL) { 1173 error = nfsvno_accchk(*vpp, VWRITE, 1174 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1175 NFSACCCHK_VPISLOCKED, NULL); 1176 if (!error) { 1177 tempsize = nvap->na_size; 1178 NFSVNO_ATTRINIT(nvap); 1179 nvap->na_size = tempsize; 1180 error = VOP_SETATTR(*vpp, 1181 &nvap->na_vattr, nd->nd_cred); 1182 } 1183 } 1184 if (error) 1185 vput(*vpp); 1186 } 1187 1188 out: 1189 NFSEXITCODE(error); 1190 return (error); 1191 } 1192 1193 /* 1194 * Do a mknod vnode op. 1195 */ 1196 int 1197 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, 1198 struct thread *p) 1199 { 1200 int error = 0; 1201 enum vtype vtyp; 1202 1203 vtyp = nvap->na_type; 1204 /* 1205 * Iff doesn't exist, create it. 1206 */ 1207 if (ndp->ni_vp) { 1208 vrele(ndp->ni_startdir); 1209 nfsvno_relpathbuf(ndp); 1210 vput(ndp->ni_dvp); 1211 vrele(ndp->ni_vp); 1212 error = EEXIST; 1213 goto out; 1214 } 1215 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1216 vrele(ndp->ni_startdir); 1217 nfsvno_relpathbuf(ndp); 1218 vput(ndp->ni_dvp); 1219 error = NFSERR_BADTYPE; 1220 goto out; 1221 } 1222 if (vtyp == VSOCK) { 1223 vrele(ndp->ni_startdir); 1224 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 1225 &ndp->ni_cnd, &nvap->na_vattr); 1226 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1227 false); 1228 nfsvno_relpathbuf(ndp); 1229 } else { 1230 if (nvap->na_type != VFIFO && 1231 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { 1232 vrele(ndp->ni_startdir); 1233 nfsvno_relpathbuf(ndp); 1234 vput(ndp->ni_dvp); 1235 goto out; 1236 } 1237 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1238 &ndp->ni_cnd, &nvap->na_vattr); 1239 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1240 false); 1241 nfsvno_relpathbuf(ndp); 1242 vrele(ndp->ni_startdir); 1243 /* 1244 * Since VOP_MKNOD returns the ni_vp, I can't 1245 * see any reason to do the lookup. 1246 */ 1247 } 1248 1249 out: 1250 NFSEXITCODE(error); 1251 return (error); 1252 } 1253 1254 /* 1255 * Mkdir vnode op. 1256 */ 1257 int 1258 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, 1259 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 1260 { 1261 int error = 0; 1262 1263 if (ndp->ni_vp != NULL) { 1264 if (ndp->ni_dvp == ndp->ni_vp) 1265 vrele(ndp->ni_dvp); 1266 else 1267 vput(ndp->ni_dvp); 1268 vrele(ndp->ni_vp); 1269 nfsvno_relpathbuf(ndp); 1270 error = EEXIST; 1271 goto out; 1272 } 1273 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1274 &nvap->na_vattr); 1275 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); 1276 nfsvno_relpathbuf(ndp); 1277 1278 out: 1279 NFSEXITCODE(error); 1280 return (error); 1281 } 1282 1283 /* 1284 * symlink vnode op. 1285 */ 1286 int 1287 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, 1288 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, 1289 struct nfsexstuff *exp) 1290 { 1291 int error = 0; 1292 1293 if (ndp->ni_vp) { 1294 vrele(ndp->ni_startdir); 1295 nfsvno_relpathbuf(ndp); 1296 if (ndp->ni_dvp == ndp->ni_vp) 1297 vrele(ndp->ni_dvp); 1298 else 1299 vput(ndp->ni_dvp); 1300 vrele(ndp->ni_vp); 1301 error = EEXIST; 1302 goto out; 1303 } 1304 1305 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1306 &nvap->na_vattr, pathcp); 1307 /* 1308 * Although FreeBSD still had the lookup code in 1309 * it for 7/current, there doesn't seem to be any 1310 * point, since VOP_SYMLINK() returns the ni_vp. 1311 * Just vput it for v2. 1312 */ 1313 VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0); 1314 vrele(ndp->ni_startdir); 1315 nfsvno_relpathbuf(ndp); 1316 1317 out: 1318 NFSEXITCODE(error); 1319 return (error); 1320 } 1321 1322 /* 1323 * Parse symbolic link arguments. 1324 * This function has an ugly side effect. It will malloc() an area for 1325 * the symlink and set iov_base to point to it, only if it succeeds. 1326 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must 1327 * be FREE'd later. 1328 */ 1329 int 1330 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, 1331 struct thread *p, char **pathcpp, int *lenp) 1332 { 1333 u_int32_t *tl; 1334 char *pathcp = NULL; 1335 int error = 0, len; 1336 struct nfsv2_sattr *sp; 1337 1338 *pathcpp = NULL; 1339 *lenp = 0; 1340 if ((nd->nd_flag & ND_NFSV3) && 1341 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) 1342 goto nfsmout; 1343 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 1344 len = fxdr_unsigned(int, *tl); 1345 if (len > NFS_MAXPATHLEN || len <= 0) { 1346 error = EBADRPC; 1347 goto nfsmout; 1348 } 1349 pathcp = malloc(len + 1, M_TEMP, M_WAITOK); 1350 error = nfsrv_mtostr(nd, pathcp, len); 1351 if (error) 1352 goto nfsmout; 1353 if (nd->nd_flag & ND_NFSV2) { 1354 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1355 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 1356 } 1357 *pathcpp = pathcp; 1358 *lenp = len; 1359 NFSEXITCODE2(0, nd); 1360 return (0); 1361 nfsmout: 1362 if (pathcp) 1363 free(pathcp, M_TEMP); 1364 NFSEXITCODE2(error, nd); 1365 return (error); 1366 } 1367 1368 /* 1369 * Remove a non-directory object. 1370 */ 1371 int 1372 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1373 struct thread *p, struct nfsexstuff *exp) 1374 { 1375 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS]; 1376 int error = 0, mirrorcnt; 1377 char fname[PNFS_FILENAME_LEN + 1]; 1378 fhandle_t fh; 1379 1380 vp = ndp->ni_vp; 1381 dsdvp[0] = NULL; 1382 if (vp->v_type == VDIR) 1383 error = NFSERR_ISDIR; 1384 else if (is_v4) 1385 error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0), 1386 p); 1387 if (error == 0) 1388 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); 1389 if (!error) 1390 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); 1391 if (error == 0 && dsdvp[0] != NULL) 1392 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1393 if (ndp->ni_dvp == vp) 1394 vrele(ndp->ni_dvp); 1395 else 1396 vput(ndp->ni_dvp); 1397 vput(vp); 1398 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) 1399 nfsvno_relpathbuf(ndp); 1400 NFSEXITCODE(error); 1401 return (error); 1402 } 1403 1404 /* 1405 * Remove a directory. 1406 */ 1407 int 1408 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1409 struct thread *p, struct nfsexstuff *exp) 1410 { 1411 struct vnode *vp; 1412 int error = 0; 1413 1414 vp = ndp->ni_vp; 1415 if (vp->v_type != VDIR) { 1416 error = ENOTDIR; 1417 goto out; 1418 } 1419 /* 1420 * No rmdir "." please. 1421 */ 1422 if (ndp->ni_dvp == vp) { 1423 error = EINVAL; 1424 goto out; 1425 } 1426 /* 1427 * The root of a mounted filesystem cannot be deleted. 1428 */ 1429 if (vp->v_vflag & VV_ROOT) 1430 error = EBUSY; 1431 out: 1432 if (!error) 1433 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); 1434 if (ndp->ni_dvp == vp) 1435 vrele(ndp->ni_dvp); 1436 else 1437 vput(ndp->ni_dvp); 1438 vput(vp); 1439 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) 1440 nfsvno_relpathbuf(ndp); 1441 NFSEXITCODE(error); 1442 return (error); 1443 } 1444 1445 /* 1446 * Rename vnode op. 1447 */ 1448 int 1449 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, 1450 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) 1451 { 1452 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS]; 1453 int error = 0, mirrorcnt; 1454 char fname[PNFS_FILENAME_LEN + 1]; 1455 fhandle_t fh; 1456 1457 dsdvp[0] = NULL; 1458 fvp = fromndp->ni_vp; 1459 if (ndstat) { 1460 vrele(fromndp->ni_dvp); 1461 vrele(fvp); 1462 error = ndstat; 1463 goto out1; 1464 } 1465 tdvp = tondp->ni_dvp; 1466 tvp = tondp->ni_vp; 1467 if (tvp != NULL) { 1468 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1469 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; 1470 goto out; 1471 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1472 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; 1473 goto out; 1474 } 1475 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1476 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1477 goto out; 1478 } 1479 1480 /* 1481 * A rename to '.' or '..' results in a prematurely 1482 * unlocked vnode on FreeBSD5, so I'm just going to fail that 1483 * here. 1484 */ 1485 if ((tondp->ni_cnd.cn_namelen == 1 && 1486 tondp->ni_cnd.cn_nameptr[0] == '.') || 1487 (tondp->ni_cnd.cn_namelen == 2 && 1488 tondp->ni_cnd.cn_nameptr[0] == '.' && 1489 tondp->ni_cnd.cn_nameptr[1] == '.')) { 1490 error = EINVAL; 1491 goto out; 1492 } 1493 } 1494 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1495 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1496 goto out; 1497 } 1498 if (fvp->v_mount != tdvp->v_mount) { 1499 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1500 goto out; 1501 } 1502 if (fvp == tdvp) { 1503 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; 1504 goto out; 1505 } 1506 if (fvp == tvp) { 1507 /* 1508 * If source and destination are the same, there is nothing to 1509 * do. Set error to -1 to indicate this. 1510 */ 1511 error = -1; 1512 goto out; 1513 } 1514 if (ndflag & ND_NFSV4) { 1515 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { 1516 error = nfsrv_checkremove(fvp, 0, NULL, 1517 (nfsquad_t)((u_quad_t)0), p); 1518 NFSVOPUNLOCK(fvp); 1519 } else 1520 error = EPERM; 1521 if (tvp && !error) 1522 error = nfsrv_checkremove(tvp, 1, NULL, 1523 (nfsquad_t)((u_quad_t)0), p); 1524 } else { 1525 /* 1526 * For NFSv2 and NFSv3, try to get rid of the delegation, so 1527 * that the NFSv4 client won't be confused by the rename. 1528 * Since nfsd_recalldelegation() can only be called on an 1529 * unlocked vnode at this point and fvp is the file that will 1530 * still exist after the rename, just do fvp. 1531 */ 1532 nfsd_recalldelegation(fvp, p); 1533 } 1534 if (error == 0 && tvp != NULL) { 1535 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh); 1536 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" 1537 " dsdvp=%p\n", dsdvp[0]); 1538 } 1539 out: 1540 if (!error) { 1541 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, 1542 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, 1543 &tondp->ni_cnd); 1544 } else { 1545 if (tdvp == tvp) 1546 vrele(tdvp); 1547 else 1548 vput(tdvp); 1549 if (tvp) 1550 vput(tvp); 1551 vrele(fromndp->ni_dvp); 1552 vrele(fvp); 1553 if (error == -1) 1554 error = 0; 1555 } 1556 1557 /* 1558 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and 1559 * if the rename succeeded, the DS file for the tvp needs to be 1560 * removed. 1561 */ 1562 if (error == 0 && dsdvp[0] != NULL) { 1563 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1564 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); 1565 } 1566 1567 vrele(tondp->ni_startdir); 1568 nfsvno_relpathbuf(tondp); 1569 out1: 1570 vrele(fromndp->ni_startdir); 1571 nfsvno_relpathbuf(fromndp); 1572 NFSEXITCODE(error); 1573 return (error); 1574 } 1575 1576 /* 1577 * Link vnode op. 1578 */ 1579 int 1580 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, 1581 struct thread *p, struct nfsexstuff *exp) 1582 { 1583 struct vnode *xp; 1584 int error = 0; 1585 1586 xp = ndp->ni_vp; 1587 if (xp != NULL) { 1588 error = EEXIST; 1589 } else { 1590 xp = ndp->ni_dvp; 1591 if (vp->v_mount != xp->v_mount) 1592 error = EXDEV; 1593 } 1594 if (!error) { 1595 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1596 if (!VN_IS_DOOMED(vp)) 1597 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); 1598 else 1599 error = EPERM; 1600 if (ndp->ni_dvp == vp) { 1601 vrele(ndp->ni_dvp); 1602 NFSVOPUNLOCK(vp); 1603 } else { 1604 vref(vp); 1605 VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true); 1606 } 1607 } else { 1608 if (ndp->ni_dvp == ndp->ni_vp) 1609 vrele(ndp->ni_dvp); 1610 else 1611 vput(ndp->ni_dvp); 1612 if (ndp->ni_vp) 1613 vrele(ndp->ni_vp); 1614 } 1615 nfsvno_relpathbuf(ndp); 1616 NFSEXITCODE(error); 1617 return (error); 1618 } 1619 1620 /* 1621 * Do the fsync() appropriate for the commit. 1622 */ 1623 int 1624 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, 1625 struct thread *td) 1626 { 1627 int error = 0; 1628 1629 /* 1630 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of 1631 * file is done. At this time VOP_FSYNC does not accept offset and 1632 * byte count parameters so call VOP_FSYNC the whole file for now. 1633 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. 1634 * File systems that do not use the buffer cache (as indicated 1635 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). 1636 */ 1637 if (cnt == 0 || cnt > MAX_COMMIT_COUNT || 1638 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { 1639 /* 1640 * Give up and do the whole thing 1641 */ 1642 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1643 VM_OBJECT_WLOCK(vp->v_object); 1644 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); 1645 VM_OBJECT_WUNLOCK(vp->v_object); 1646 } 1647 error = VOP_FSYNC(vp, MNT_WAIT, td); 1648 } else { 1649 /* 1650 * Locate and synchronously write any buffers that fall 1651 * into the requested range. Note: we are assuming that 1652 * f_iosize is a power of 2. 1653 */ 1654 int iosize = vp->v_mount->mnt_stat.f_iosize; 1655 int iomask = iosize - 1; 1656 struct bufobj *bo; 1657 daddr_t lblkno; 1658 1659 /* 1660 * Align to iosize boundary, super-align to page boundary. 1661 */ 1662 if (off & iomask) { 1663 cnt += off & iomask; 1664 off &= ~(u_quad_t)iomask; 1665 } 1666 if (off & PAGE_MASK) { 1667 cnt += off & PAGE_MASK; 1668 off &= ~(u_quad_t)PAGE_MASK; 1669 } 1670 lblkno = off / iosize; 1671 1672 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1673 VM_OBJECT_WLOCK(vp->v_object); 1674 vm_object_page_clean(vp->v_object, off, off + cnt, 1675 OBJPC_SYNC); 1676 VM_OBJECT_WUNLOCK(vp->v_object); 1677 } 1678 1679 bo = &vp->v_bufobj; 1680 BO_LOCK(bo); 1681 while (cnt > 0) { 1682 struct buf *bp; 1683 1684 /* 1685 * If we have a buffer and it is marked B_DELWRI we 1686 * have to lock and write it. Otherwise the prior 1687 * write is assumed to have already been committed. 1688 * 1689 * gbincore() can return invalid buffers now so we 1690 * have to check that bit as well (though B_DELWRI 1691 * should not be set if B_INVAL is set there could be 1692 * a race here since we haven't locked the buffer). 1693 */ 1694 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 1695 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 1696 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { 1697 BO_LOCK(bo); 1698 continue; /* retry */ 1699 } 1700 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 1701 B_DELWRI) { 1702 bremfree(bp); 1703 bp->b_flags &= ~B_ASYNC; 1704 bwrite(bp); 1705 ++nfs_commit_miss; 1706 } else 1707 BUF_UNLOCK(bp); 1708 BO_LOCK(bo); 1709 } 1710 ++nfs_commit_blks; 1711 if (cnt < iosize) 1712 break; 1713 cnt -= iosize; 1714 ++lblkno; 1715 } 1716 BO_UNLOCK(bo); 1717 } 1718 NFSEXITCODE(error); 1719 return (error); 1720 } 1721 1722 /* 1723 * Statfs vnode op. 1724 */ 1725 int 1726 nfsvno_statfs(struct vnode *vp, struct statfs *sf) 1727 { 1728 struct statfs *tsf; 1729 int error; 1730 1731 tsf = NULL; 1732 if (nfsrv_devidcnt > 0) { 1733 /* For a pNFS service, get the DS numbers. */ 1734 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); 1735 error = nfsrv_pnfsstatfs(tsf, vp->v_mount); 1736 if (error != 0) { 1737 free(tsf, M_TEMP); 1738 tsf = NULL; 1739 } 1740 } 1741 error = VFS_STATFS(vp->v_mount, sf); 1742 if (error == 0) { 1743 if (tsf != NULL) { 1744 sf->f_blocks = tsf->f_blocks; 1745 sf->f_bavail = tsf->f_bavail; 1746 sf->f_bfree = tsf->f_bfree; 1747 sf->f_bsize = tsf->f_bsize; 1748 } 1749 /* 1750 * Since NFS handles these values as unsigned on the 1751 * wire, there is no way to represent negative values, 1752 * so set them to 0. Without this, they will appear 1753 * to be very large positive values for clients like 1754 * Solaris10. 1755 */ 1756 if (sf->f_bavail < 0) 1757 sf->f_bavail = 0; 1758 if (sf->f_ffree < 0) 1759 sf->f_ffree = 0; 1760 } 1761 free(tsf, M_TEMP); 1762 NFSEXITCODE(error); 1763 return (error); 1764 } 1765 1766 /* 1767 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but 1768 * must handle nfsrv_opencheck() calls after any other access checks. 1769 */ 1770 void 1771 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, 1772 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, 1773 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, 1774 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, 1775 struct nfsexstuff *exp, struct vnode **vpp) 1776 { 1777 struct vnode *vp = NULL; 1778 u_quad_t tempsize; 1779 struct nfsexstuff nes; 1780 struct thread *p = curthread; 1781 1782 if (ndp->ni_vp == NULL) 1783 nd->nd_repstat = nfsrv_opencheck(clientid, 1784 stateidp, stp, NULL, nd, p, nd->nd_repstat); 1785 if (!nd->nd_repstat) { 1786 if (ndp->ni_vp == NULL) { 1787 vrele(ndp->ni_startdir); 1788 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, 1789 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1790 /* For a pNFS server, create the data file on a DS. */ 1791 if (nd->nd_repstat == 0) { 1792 /* 1793 * Create a data file on a DS for a pNFS server. 1794 * This function just returns if not 1795 * running a pNFS DS or the creation fails. 1796 */ 1797 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1798 cred, p); 1799 } 1800 VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ? 1801 &ndp->ni_vp : NULL, false); 1802 nfsvno_relpathbuf(ndp); 1803 if (!nd->nd_repstat) { 1804 if (*exclusive_flagp) { 1805 *exclusive_flagp = 0; 1806 NFSVNO_ATTRINIT(nvap); 1807 nvap->na_atime.tv_sec = cverf[0]; 1808 nvap->na_atime.tv_nsec = cverf[1]; 1809 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, 1810 &nvap->na_vattr, cred); 1811 if (nd->nd_repstat != 0) { 1812 vput(ndp->ni_vp); 1813 ndp->ni_vp = NULL; 1814 nd->nd_repstat = NFSERR_NOTSUPP; 1815 } else 1816 NFSSETBIT_ATTRBIT(attrbitp, 1817 NFSATTRBIT_TIMEACCESS); 1818 } else { 1819 nfsrv_fixattr(nd, ndp->ni_vp, nvap, 1820 aclp, p, attrbitp, exp); 1821 } 1822 } 1823 vp = ndp->ni_vp; 1824 } else { 1825 if (ndp->ni_startdir) 1826 vrele(ndp->ni_startdir); 1827 nfsvno_relpathbuf(ndp); 1828 vp = ndp->ni_vp; 1829 if (create == NFSV4OPEN_CREATE) { 1830 if (ndp->ni_dvp == vp) 1831 vrele(ndp->ni_dvp); 1832 else 1833 vput(ndp->ni_dvp); 1834 } 1835 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { 1836 if (ndp->ni_cnd.cn_flags & RDONLY) 1837 NFSVNO_SETEXRDONLY(&nes); 1838 else 1839 NFSVNO_EXINIT(&nes); 1840 nd->nd_repstat = nfsvno_accchk(vp, 1841 VWRITE, cred, &nes, p, 1842 NFSACCCHK_NOOVERRIDE, 1843 NFSACCCHK_VPISLOCKED, NULL); 1844 nd->nd_repstat = nfsrv_opencheck(clientid, 1845 stateidp, stp, vp, nd, p, nd->nd_repstat); 1846 if (!nd->nd_repstat) { 1847 tempsize = nvap->na_size; 1848 NFSVNO_ATTRINIT(nvap); 1849 nvap->na_size = tempsize; 1850 nd->nd_repstat = VOP_SETATTR(vp, 1851 &nvap->na_vattr, cred); 1852 } 1853 } else if (vp->v_type == VREG) { 1854 nd->nd_repstat = nfsrv_opencheck(clientid, 1855 stateidp, stp, vp, nd, p, nd->nd_repstat); 1856 } 1857 } 1858 } else { 1859 if (ndp->ni_cnd.cn_flags & HASBUF) 1860 nfsvno_relpathbuf(ndp); 1861 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) { 1862 vrele(ndp->ni_startdir); 1863 if (ndp->ni_dvp == ndp->ni_vp) 1864 vrele(ndp->ni_dvp); 1865 else 1866 vput(ndp->ni_dvp); 1867 if (ndp->ni_vp) 1868 vput(ndp->ni_vp); 1869 } 1870 } 1871 *vpp = vp; 1872 1873 NFSEXITCODE2(0, nd); 1874 } 1875 1876 /* 1877 * Updates the file rev and sets the mtime and ctime 1878 * to the current clock time, returning the va_filerev and va_Xtime 1879 * values. 1880 * Return ESTALE to indicate the vnode is VIRF_DOOMED. 1881 */ 1882 int 1883 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, 1884 struct nfsrv_descript *nd, struct thread *p) 1885 { 1886 struct vattr va; 1887 1888 VATTR_NULL(&va); 1889 vfs_timestamp(&va.va_mtime); 1890 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 1891 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 1892 if (VN_IS_DOOMED(vp)) 1893 return (ESTALE); 1894 } 1895 (void) VOP_SETATTR(vp, &va, nd->nd_cred); 1896 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); 1897 return (0); 1898 } 1899 1900 /* 1901 * Glue routine to nfsv4_fillattr(). 1902 */ 1903 int 1904 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, 1905 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, 1906 struct ucred *cred, struct thread *p, int isdgram, int reterr, 1907 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) 1908 { 1909 struct statfs *sf; 1910 int error; 1911 1912 sf = NULL; 1913 if (nfsrv_devidcnt > 0 && 1914 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || 1915 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || 1916 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { 1917 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); 1918 error = nfsrv_pnfsstatfs(sf, mp); 1919 if (error != 0) { 1920 free(sf, M_TEMP); 1921 sf = NULL; 1922 } 1923 } 1924 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, 1925 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, 1926 mounted_on_fileno, sf); 1927 free(sf, M_TEMP); 1928 NFSEXITCODE2(0, nd); 1929 return (error); 1930 } 1931 1932 /* Since the Readdir vnode ops vary, put the entire functions in here. */ 1933 /* 1934 * nfs readdir service 1935 * - mallocs what it thinks is enough to read 1936 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR 1937 * - calls VOP_READDIR() 1938 * - loops around building the reply 1939 * if the output generated exceeds count break out of loop 1940 * The NFSM_CLGET macro is used here so that the reply will be packed 1941 * tightly in mbuf clusters. 1942 * - it trims out records with d_fileno == 0 1943 * this doesn't matter for Unix clients, but they might confuse clients 1944 * for other os'. 1945 * - it trims out records with d_type == DT_WHT 1946 * these cannot be seen through NFS (unless we extend the protocol) 1947 * The alternate call nfsrvd_readdirplus() does lookups as well. 1948 * PS: The NFS protocol spec. does not clarify what the "count" byte 1949 * argument is a count of.. just name strings and file id's or the 1950 * entire reply rpc or ... 1951 * I tried just file name and id sizes and it confused the Sun client, 1952 * so I am using the full rpc size now. The "paranoia.." comment refers 1953 * to including the status longwords that are not a part of the dir. 1954 * "entry" structures, but are in the rpc. 1955 */ 1956 int 1957 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, 1958 struct vnode *vp, struct nfsexstuff *exp) 1959 { 1960 struct dirent *dp; 1961 u_int32_t *tl; 1962 int dirlen; 1963 char *cpos, *cend, *rbuf; 1964 struct nfsvattr at; 1965 int nlen, error = 0, getret = 1; 1966 int siz, cnt, fullsiz, eofflag, ncookies; 1967 u_int64_t off, toff, verf __unused; 1968 u_long *cookies = NULL, *cookiep; 1969 struct uio io; 1970 struct iovec iv; 1971 int is_ufs; 1972 struct thread *p = curthread; 1973 1974 if (nd->nd_repstat) { 1975 nfsrv_postopattr(nd, getret, &at); 1976 goto out; 1977 } 1978 if (nd->nd_flag & ND_NFSV2) { 1979 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1980 off = fxdr_unsigned(u_quad_t, *tl++); 1981 } else { 1982 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1983 off = fxdr_hyper(tl); 1984 tl += 2; 1985 verf = fxdr_hyper(tl); 1986 tl += 2; 1987 } 1988 toff = off; 1989 cnt = fxdr_unsigned(int, *tl); 1990 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 1991 cnt = NFS_SRVMAXDATA(nd); 1992 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 1993 fullsiz = siz; 1994 if (nd->nd_flag & ND_NFSV3) { 1995 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, 1996 NULL); 1997 #if 0 1998 /* 1999 * va_filerev is not sufficient as a cookie verifier, 2000 * since it is not supposed to change when entries are 2001 * removed/added unless that offset cookies returned to 2002 * the client are no longer valid. 2003 */ 2004 if (!nd->nd_repstat && toff && verf != at.na_filerev) 2005 nd->nd_repstat = NFSERR_BAD_COOKIE; 2006 #endif 2007 } 2008 if (!nd->nd_repstat && vp->v_type != VDIR) 2009 nd->nd_repstat = NFSERR_NOTDIR; 2010 if (nd->nd_repstat == 0 && cnt == 0) { 2011 if (nd->nd_flag & ND_NFSV2) 2012 /* NFSv2 does not have NFSERR_TOOSMALL */ 2013 nd->nd_repstat = EPERM; 2014 else 2015 nd->nd_repstat = NFSERR_TOOSMALL; 2016 } 2017 if (!nd->nd_repstat) 2018 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2019 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2020 NFSACCCHK_VPISLOCKED, NULL); 2021 if (nd->nd_repstat) { 2022 vput(vp); 2023 if (nd->nd_flag & ND_NFSV3) 2024 nfsrv_postopattr(nd, getret, &at); 2025 goto out; 2026 } 2027 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2028 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2029 again: 2030 eofflag = 0; 2031 if (cookies) { 2032 free(cookies, M_TEMP); 2033 cookies = NULL; 2034 } 2035 2036 iv.iov_base = rbuf; 2037 iv.iov_len = siz; 2038 io.uio_iov = &iv; 2039 io.uio_iovcnt = 1; 2040 io.uio_offset = (off_t)off; 2041 io.uio_resid = siz; 2042 io.uio_segflg = UIO_SYSSPACE; 2043 io.uio_rw = UIO_READ; 2044 io.uio_td = NULL; 2045 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2046 &cookies); 2047 off = (u_int64_t)io.uio_offset; 2048 if (io.uio_resid) 2049 siz -= io.uio_resid; 2050 2051 if (!cookies && !nd->nd_repstat) 2052 nd->nd_repstat = NFSERR_PERM; 2053 if (nd->nd_flag & ND_NFSV3) { 2054 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2055 if (!nd->nd_repstat) 2056 nd->nd_repstat = getret; 2057 } 2058 2059 /* 2060 * Handles the failed cases. nd->nd_repstat == 0 past here. 2061 */ 2062 if (nd->nd_repstat) { 2063 vput(vp); 2064 free(rbuf, M_TEMP); 2065 if (cookies) 2066 free(cookies, M_TEMP); 2067 if (nd->nd_flag & ND_NFSV3) 2068 nfsrv_postopattr(nd, getret, &at); 2069 goto out; 2070 } 2071 /* 2072 * If nothing read, return eof 2073 * rpc reply 2074 */ 2075 if (siz == 0) { 2076 vput(vp); 2077 if (nd->nd_flag & ND_NFSV2) { 2078 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2079 } else { 2080 nfsrv_postopattr(nd, getret, &at); 2081 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2082 txdr_hyper(at.na_filerev, tl); 2083 tl += 2; 2084 } 2085 *tl++ = newnfs_false; 2086 *tl = newnfs_true; 2087 free(rbuf, M_TEMP); 2088 free(cookies, M_TEMP); 2089 goto out; 2090 } 2091 2092 /* 2093 * Check for degenerate cases of nothing useful read. 2094 * If so go try again 2095 */ 2096 cpos = rbuf; 2097 cend = rbuf + siz; 2098 dp = (struct dirent *)cpos; 2099 cookiep = cookies; 2100 2101 /* 2102 * For some reason FreeBSD's ufs_readdir() chooses to back the 2103 * directory offset up to a block boundary, so it is necessary to 2104 * skip over the records that precede the requested offset. This 2105 * requires the assumption that file offset cookies monotonically 2106 * increase. 2107 */ 2108 while (cpos < cend && ncookies > 0 && 2109 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2110 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { 2111 cpos += dp->d_reclen; 2112 dp = (struct dirent *)cpos; 2113 cookiep++; 2114 ncookies--; 2115 } 2116 if (cpos >= cend || ncookies == 0) { 2117 siz = fullsiz; 2118 toff = off; 2119 goto again; 2120 } 2121 vput(vp); 2122 2123 /* 2124 * If cnt > MCLBYTES and the reply will not be saved, use 2125 * ext_pgs mbufs for TLS. 2126 * For NFSv4.0, we do not know for sure if the reply will 2127 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2128 */ 2129 if (cnt > MCLBYTES && siz > MCLBYTES && 2130 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2131 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2132 nd->nd_flag |= ND_EXTPG; 2133 2134 /* 2135 * dirlen is the size of the reply, including all XDR and must 2136 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate 2137 * if the XDR should be included in "count", but to be safe, we do. 2138 * (Include the two booleans at the end of the reply in dirlen now.) 2139 */ 2140 if (nd->nd_flag & ND_NFSV3) { 2141 nfsrv_postopattr(nd, getret, &at); 2142 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2143 txdr_hyper(at.na_filerev, tl); 2144 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2145 } else { 2146 dirlen = 2 * NFSX_UNSIGNED; 2147 } 2148 2149 /* Loop through the records and build reply */ 2150 while (cpos < cend && ncookies > 0) { 2151 nlen = dp->d_namlen; 2152 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2153 nlen <= NFS_MAXNAMLEN) { 2154 if (nd->nd_flag & ND_NFSV3) 2155 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2156 else 2157 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2158 if (dirlen > cnt) { 2159 eofflag = 0; 2160 break; 2161 } 2162 2163 /* 2164 * Build the directory record xdr from 2165 * the dirent entry. 2166 */ 2167 if (nd->nd_flag & ND_NFSV3) { 2168 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2169 *tl++ = newnfs_true; 2170 *tl++ = 0; 2171 } else { 2172 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2173 *tl++ = newnfs_true; 2174 } 2175 *tl = txdr_unsigned(dp->d_fileno); 2176 (void) nfsm_strtom(nd, dp->d_name, nlen); 2177 if (nd->nd_flag & ND_NFSV3) { 2178 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2179 *tl++ = 0; 2180 } else 2181 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 2182 *tl = txdr_unsigned(*cookiep); 2183 } 2184 cpos += dp->d_reclen; 2185 dp = (struct dirent *)cpos; 2186 cookiep++; 2187 ncookies--; 2188 } 2189 if (cpos < cend) 2190 eofflag = 0; 2191 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2192 *tl++ = newnfs_false; 2193 if (eofflag) 2194 *tl = newnfs_true; 2195 else 2196 *tl = newnfs_false; 2197 free(rbuf, M_TEMP); 2198 free(cookies, M_TEMP); 2199 2200 out: 2201 NFSEXITCODE2(0, nd); 2202 return (0); 2203 nfsmout: 2204 vput(vp); 2205 NFSEXITCODE2(error, nd); 2206 return (error); 2207 } 2208 2209 /* 2210 * Readdirplus for V3 and Readdir for V4. 2211 */ 2212 int 2213 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, 2214 struct vnode *vp, struct nfsexstuff *exp) 2215 { 2216 struct dirent *dp; 2217 u_int32_t *tl; 2218 int dirlen; 2219 char *cpos, *cend, *rbuf; 2220 struct vnode *nvp; 2221 fhandle_t nfh; 2222 struct nfsvattr nva, at, *nvap = &nva; 2223 struct mbuf *mb0, *mb1; 2224 struct nfsreferral *refp; 2225 int nlen, r, error = 0, getret = 1, usevget = 1; 2226 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; 2227 caddr_t bpos0, bpos1; 2228 u_int64_t off, toff, verf; 2229 u_long *cookies = NULL, *cookiep; 2230 nfsattrbit_t attrbits, rderrbits, savbits; 2231 struct uio io; 2232 struct iovec iv; 2233 struct componentname cn; 2234 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; 2235 struct mount *mp, *new_mp; 2236 uint64_t mounted_on_fileno; 2237 struct thread *p = curthread; 2238 int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1; 2239 2240 if (nd->nd_repstat) { 2241 nfsrv_postopattr(nd, getret, &at); 2242 goto out; 2243 } 2244 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2245 off = fxdr_hyper(tl); 2246 toff = off; 2247 tl += 2; 2248 verf = fxdr_hyper(tl); 2249 tl += 2; 2250 siz = fxdr_unsigned(int, *tl++); 2251 cnt = fxdr_unsigned(int, *tl); 2252 2253 /* 2254 * Use the server's maximum data transfer size as the upper bound 2255 * on reply datalen. 2256 */ 2257 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2258 cnt = NFS_SRVMAXDATA(nd); 2259 2260 /* 2261 * siz is a "hint" of how much directory information (name, fileid, 2262 * cookie) should be in the reply. At least one client "hints" 0, 2263 * so I set it to cnt for that case. I also round it up to the 2264 * next multiple of DIRBLKSIZ. 2265 * Since the size of a Readdirplus directory entry reply will always 2266 * be greater than a directory entry returned by VOP_READDIR(), it 2267 * does not make sense to read more than NFS_SRVMAXDATA() via 2268 * VOP_READDIR(). 2269 */ 2270 if (siz <= 0) 2271 siz = cnt; 2272 else if (siz > NFS_SRVMAXDATA(nd)) 2273 siz = NFS_SRVMAXDATA(nd); 2274 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2275 2276 if (nd->nd_flag & ND_NFSV4) { 2277 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 2278 if (error) 2279 goto nfsmout; 2280 NFSSET_ATTRBIT(&savbits, &attrbits); 2281 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); 2282 NFSZERO_ATTRBIT(&rderrbits); 2283 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); 2284 } else { 2285 NFSZERO_ATTRBIT(&attrbits); 2286 } 2287 fullsiz = siz; 2288 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2289 #if 0 2290 if (!nd->nd_repstat) { 2291 if (off && verf != at.na_filerev) { 2292 /* 2293 * va_filerev is not sufficient as a cookie verifier, 2294 * since it is not supposed to change when entries are 2295 * removed/added unless that offset cookies returned to 2296 * the client are no longer valid. 2297 */ 2298 if (nd->nd_flag & ND_NFSV4) { 2299 nd->nd_repstat = NFSERR_NOTSAME; 2300 } else { 2301 nd->nd_repstat = NFSERR_BAD_COOKIE; 2302 } 2303 } 2304 } 2305 #endif 2306 if (!nd->nd_repstat && vp->v_type != VDIR) 2307 nd->nd_repstat = NFSERR_NOTDIR; 2308 if (!nd->nd_repstat && cnt == 0) 2309 nd->nd_repstat = NFSERR_TOOSMALL; 2310 if (!nd->nd_repstat) 2311 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2312 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2313 NFSACCCHK_VPISLOCKED, NULL); 2314 if (nd->nd_repstat) { 2315 vput(vp); 2316 if (nd->nd_flag & ND_NFSV3) 2317 nfsrv_postopattr(nd, getret, &at); 2318 goto out; 2319 } 2320 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2321 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; 2322 2323 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2324 again: 2325 eofflag = 0; 2326 if (cookies) { 2327 free(cookies, M_TEMP); 2328 cookies = NULL; 2329 } 2330 2331 iv.iov_base = rbuf; 2332 iv.iov_len = siz; 2333 io.uio_iov = &iv; 2334 io.uio_iovcnt = 1; 2335 io.uio_offset = (off_t)off; 2336 io.uio_resid = siz; 2337 io.uio_segflg = UIO_SYSSPACE; 2338 io.uio_rw = UIO_READ; 2339 io.uio_td = NULL; 2340 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2341 &cookies); 2342 off = (u_int64_t)io.uio_offset; 2343 if (io.uio_resid) 2344 siz -= io.uio_resid; 2345 2346 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2347 2348 if (!cookies && !nd->nd_repstat) 2349 nd->nd_repstat = NFSERR_PERM; 2350 if (!nd->nd_repstat) 2351 nd->nd_repstat = getret; 2352 if (nd->nd_repstat) { 2353 vput(vp); 2354 if (cookies) 2355 free(cookies, M_TEMP); 2356 free(rbuf, M_TEMP); 2357 if (nd->nd_flag & ND_NFSV3) 2358 nfsrv_postopattr(nd, getret, &at); 2359 goto out; 2360 } 2361 /* 2362 * If nothing read, return eof 2363 * rpc reply 2364 */ 2365 if (siz == 0) { 2366 vput(vp); 2367 if (nd->nd_flag & ND_NFSV3) 2368 nfsrv_postopattr(nd, getret, &at); 2369 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2370 txdr_hyper(at.na_filerev, tl); 2371 tl += 2; 2372 *tl++ = newnfs_false; 2373 *tl = newnfs_true; 2374 free(cookies, M_TEMP); 2375 free(rbuf, M_TEMP); 2376 goto out; 2377 } 2378 2379 /* 2380 * Check for degenerate cases of nothing useful read. 2381 * If so go try again 2382 */ 2383 cpos = rbuf; 2384 cend = rbuf + siz; 2385 dp = (struct dirent *)cpos; 2386 cookiep = cookies; 2387 2388 /* 2389 * For some reason FreeBSD's ufs_readdir() chooses to back the 2390 * directory offset up to a block boundary, so it is necessary to 2391 * skip over the records that precede the requested offset. This 2392 * requires the assumption that file offset cookies monotonically 2393 * increase. 2394 */ 2395 while (cpos < cend && ncookies > 0 && 2396 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2397 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || 2398 ((nd->nd_flag & ND_NFSV4) && 2399 ((dp->d_namlen == 1 && dp->d_name[0] == '.') || 2400 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { 2401 cpos += dp->d_reclen; 2402 dp = (struct dirent *)cpos; 2403 cookiep++; 2404 ncookies--; 2405 } 2406 if (cpos >= cend || ncookies == 0) { 2407 siz = fullsiz; 2408 toff = off; 2409 goto again; 2410 } 2411 2412 /* 2413 * Busy the file system so that the mount point won't go away 2414 * and, as such, VFS_VGET() can be used safely. 2415 */ 2416 mp = vp->v_mount; 2417 vfs_ref(mp); 2418 NFSVOPUNLOCK(vp); 2419 nd->nd_repstat = vfs_busy(mp, 0); 2420 vfs_rel(mp); 2421 if (nd->nd_repstat != 0) { 2422 vrele(vp); 2423 free(cookies, M_TEMP); 2424 free(rbuf, M_TEMP); 2425 if (nd->nd_flag & ND_NFSV3) 2426 nfsrv_postopattr(nd, getret, &at); 2427 goto out; 2428 } 2429 2430 /* 2431 * Check to see if entries in this directory can be safely acquired 2432 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. 2433 * ZFS snapshot directories need VOP_LOOKUP(), so that any 2434 * automount of the snapshot directory that is required will 2435 * be done. 2436 * This needs to be done here for NFSv4, since NFSv4 never does 2437 * a VFS_VGET() for "." or "..". 2438 */ 2439 if (is_zfs == 1) { 2440 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); 2441 if (r == EOPNOTSUPP) { 2442 usevget = 0; 2443 cn.cn_nameiop = LOOKUP; 2444 cn.cn_lkflags = LK_SHARED | LK_RETRY; 2445 cn.cn_cred = nd->nd_cred; 2446 cn.cn_thread = p; 2447 } else if (r == 0) 2448 vput(nvp); 2449 } 2450 2451 /* 2452 * If the reply is likely to exceed MCLBYTES and the reply will 2453 * not be saved, use ext_pgs mbufs for TLS. 2454 * It is difficult to predict how large each entry will be and 2455 * how many entries have been read, so just assume the directory 2456 * entries grow by a factor of 4 when attributes are included. 2457 * For NFSv4.0, we do not know for sure if the reply will 2458 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2459 */ 2460 if (cnt > MCLBYTES && siz > MCLBYTES / 4 && 2461 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2462 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2463 nd->nd_flag |= ND_EXTPG; 2464 2465 /* 2466 * Save this position, in case there is an error before one entry 2467 * is created. 2468 */ 2469 mb0 = nd->nd_mb; 2470 bpos0 = nd->nd_bpos; 2471 bextpg0 = nd->nd_bextpg; 2472 bextpgsiz0 = nd->nd_bextpgsiz; 2473 2474 /* 2475 * Fill in the first part of the reply. 2476 * dirlen is the reply length in bytes and cannot exceed cnt. 2477 * (Include the two booleans at the end of the reply in dirlen now, 2478 * so we recognize when we have exceeded cnt.) 2479 */ 2480 if (nd->nd_flag & ND_NFSV3) { 2481 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2482 nfsrv_postopattr(nd, getret, &at); 2483 } else { 2484 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; 2485 } 2486 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); 2487 txdr_hyper(at.na_filerev, tl); 2488 2489 /* 2490 * Save this position, in case there is an empty reply needed. 2491 */ 2492 mb1 = nd->nd_mb; 2493 bpos1 = nd->nd_bpos; 2494 bextpg1 = nd->nd_bextpg; 2495 bextpgsiz1 = nd->nd_bextpgsiz; 2496 2497 /* Loop through the records and build reply */ 2498 entrycnt = 0; 2499 while (cpos < cend && ncookies > 0 && dirlen < cnt) { 2500 nlen = dp->d_namlen; 2501 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2502 nlen <= NFS_MAXNAMLEN && 2503 ((nd->nd_flag & ND_NFSV3) || nlen > 2 || 2504 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) 2505 || (nlen == 1 && dp->d_name[0] != '.'))) { 2506 /* 2507 * Save the current position in the reply, in case 2508 * this entry exceeds cnt. 2509 */ 2510 mb1 = nd->nd_mb; 2511 bpos1 = nd->nd_bpos; 2512 bextpg1 = nd->nd_bextpg; 2513 bextpgsiz1 = nd->nd_bextpgsiz; 2514 2515 /* 2516 * For readdir_and_lookup get the vnode using 2517 * the file number. 2518 */ 2519 nvp = NULL; 2520 refp = NULL; 2521 r = 0; 2522 at_root = 0; 2523 needs_unbusy = 0; 2524 new_mp = mp; 2525 mounted_on_fileno = (uint64_t)dp->d_fileno; 2526 if ((nd->nd_flag & ND_NFSV3) || 2527 NFSNONZERO_ATTRBIT(&savbits)) { 2528 if (nd->nd_flag & ND_NFSV4) 2529 refp = nfsv4root_getreferral(NULL, 2530 vp, dp->d_fileno); 2531 if (refp == NULL) { 2532 if (usevget) 2533 r = VFS_VGET(mp, dp->d_fileno, 2534 LK_SHARED, &nvp); 2535 else 2536 r = EOPNOTSUPP; 2537 if (r == EOPNOTSUPP) { 2538 if (usevget) { 2539 usevget = 0; 2540 cn.cn_nameiop = LOOKUP; 2541 cn.cn_lkflags = 2542 LK_SHARED | 2543 LK_RETRY; 2544 cn.cn_cred = 2545 nd->nd_cred; 2546 cn.cn_thread = p; 2547 } 2548 cn.cn_nameptr = dp->d_name; 2549 cn.cn_namelen = nlen; 2550 cn.cn_flags = ISLASTCN | 2551 NOFOLLOW | LOCKLEAF; 2552 if (nlen == 2 && 2553 dp->d_name[0] == '.' && 2554 dp->d_name[1] == '.') 2555 cn.cn_flags |= 2556 ISDOTDOT; 2557 if (NFSVOPLOCK(vp, LK_SHARED) 2558 != 0) { 2559 nd->nd_repstat = EPERM; 2560 break; 2561 } 2562 if ((vp->v_vflag & VV_ROOT) != 0 2563 && (cn.cn_flags & ISDOTDOT) 2564 != 0) { 2565 vref(vp); 2566 nvp = vp; 2567 r = 0; 2568 } else { 2569 r = VOP_LOOKUP(vp, &nvp, 2570 &cn); 2571 if (vp != nvp) 2572 NFSVOPUNLOCK(vp); 2573 } 2574 } 2575 2576 /* 2577 * For NFSv4, check to see if nvp is 2578 * a mount point and get the mount 2579 * point vnode, as required. 2580 */ 2581 if (r == 0 && 2582 nfsrv_enable_crossmntpt != 0 && 2583 (nd->nd_flag & ND_NFSV4) != 0 && 2584 nvp->v_type == VDIR && 2585 nvp->v_mountedhere != NULL) { 2586 new_mp = nvp->v_mountedhere; 2587 r = vfs_busy(new_mp, 0); 2588 vput(nvp); 2589 nvp = NULL; 2590 if (r == 0) { 2591 r = VFS_ROOT(new_mp, 2592 LK_SHARED, &nvp); 2593 needs_unbusy = 1; 2594 if (r == 0) 2595 at_root = 1; 2596 } 2597 } 2598 } 2599 2600 /* 2601 * If we failed to look up the entry, then it 2602 * has become invalid, most likely removed. 2603 */ 2604 if (r != 0) { 2605 if (needs_unbusy) 2606 vfs_unbusy(new_mp); 2607 goto invalid; 2608 } 2609 KASSERT(refp != NULL || nvp != NULL, 2610 ("%s: undetected lookup error", __func__)); 2611 2612 if (refp == NULL && 2613 ((nd->nd_flag & ND_NFSV3) || 2614 NFSNONZERO_ATTRBIT(&attrbits))) { 2615 r = nfsvno_getfh(nvp, &nfh, p); 2616 if (!r) 2617 r = nfsvno_getattr(nvp, nvap, nd, p, 2618 1, &attrbits); 2619 if (r == 0 && is_zfs == 1 && 2620 nfsrv_enable_crossmntpt != 0 && 2621 (nd->nd_flag & ND_NFSV4) != 0 && 2622 nvp->v_type == VDIR && 2623 vp->v_mount != nvp->v_mount) { 2624 /* 2625 * For a ZFS snapshot, there is a 2626 * pseudo mount that does not set 2627 * v_mountedhere, so it needs to 2628 * be detected via a different 2629 * mount structure. 2630 */ 2631 at_root = 1; 2632 if (new_mp == mp) 2633 new_mp = nvp->v_mount; 2634 } 2635 } 2636 2637 /* 2638 * If we failed to get attributes of the entry, 2639 * then just skip it for NFSv3 (the traditional 2640 * behavior in the old NFS server). 2641 * For NFSv4 the behavior is controlled by 2642 * RDATTRERROR: we either ignore the error or 2643 * fail the request. 2644 * Note that RDATTRERROR is never set for NFSv3. 2645 */ 2646 if (r != 0) { 2647 if (!NFSISSET_ATTRBIT(&attrbits, 2648 NFSATTRBIT_RDATTRERROR)) { 2649 vput(nvp); 2650 if (needs_unbusy != 0) 2651 vfs_unbusy(new_mp); 2652 if ((nd->nd_flag & ND_NFSV3)) 2653 goto invalid; 2654 nd->nd_repstat = r; 2655 break; 2656 } 2657 } 2658 } 2659 2660 /* 2661 * Build the directory record xdr 2662 */ 2663 if (nd->nd_flag & ND_NFSV3) { 2664 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2665 *tl++ = newnfs_true; 2666 *tl++ = 0; 2667 *tl = txdr_unsigned(dp->d_fileno); 2668 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2669 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2670 *tl++ = 0; 2671 *tl = txdr_unsigned(*cookiep); 2672 nfsrv_postopattr(nd, 0, nvap); 2673 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1); 2674 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); 2675 if (nvp != NULL) 2676 vput(nvp); 2677 } else { 2678 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2679 *tl++ = newnfs_true; 2680 *tl++ = 0; 2681 *tl = txdr_unsigned(*cookiep); 2682 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2683 if (nvp != NULL) { 2684 supports_nfsv4acls = 2685 nfs_supportsnfsv4acls(nvp); 2686 NFSVOPUNLOCK(nvp); 2687 } else 2688 supports_nfsv4acls = 0; 2689 if (refp != NULL) { 2690 dirlen += nfsrv_putreferralattr(nd, 2691 &savbits, refp, 0, 2692 &nd->nd_repstat); 2693 if (nd->nd_repstat) { 2694 if (nvp != NULL) 2695 vrele(nvp); 2696 if (needs_unbusy != 0) 2697 vfs_unbusy(new_mp); 2698 break; 2699 } 2700 } else if (r) { 2701 dirlen += nfsvno_fillattr(nd, new_mp, 2702 nvp, nvap, &nfh, r, &rderrbits, 2703 nd->nd_cred, p, isdgram, 0, 2704 supports_nfsv4acls, at_root, 2705 mounted_on_fileno); 2706 } else { 2707 dirlen += nfsvno_fillattr(nd, new_mp, 2708 nvp, nvap, &nfh, r, &attrbits, 2709 nd->nd_cred, p, isdgram, 0, 2710 supports_nfsv4acls, at_root, 2711 mounted_on_fileno); 2712 } 2713 if (nvp != NULL) 2714 vrele(nvp); 2715 dirlen += (3 * NFSX_UNSIGNED); 2716 } 2717 if (needs_unbusy != 0) 2718 vfs_unbusy(new_mp); 2719 if (dirlen <= cnt) 2720 entrycnt++; 2721 } 2722 invalid: 2723 cpos += dp->d_reclen; 2724 dp = (struct dirent *)cpos; 2725 cookiep++; 2726 ncookies--; 2727 } 2728 vrele(vp); 2729 vfs_unbusy(mp); 2730 2731 /* 2732 * If dirlen > cnt, we must strip off the last entry. If that 2733 * results in an empty reply, report NFSERR_TOOSMALL. 2734 */ 2735 if (dirlen > cnt || nd->nd_repstat) { 2736 if (!nd->nd_repstat && entrycnt == 0) 2737 nd->nd_repstat = NFSERR_TOOSMALL; 2738 if (nd->nd_repstat) { 2739 nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); 2740 if (nd->nd_flag & ND_NFSV3) 2741 nfsrv_postopattr(nd, getret, &at); 2742 } else 2743 nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); 2744 eofflag = 0; 2745 } else if (cpos < cend) 2746 eofflag = 0; 2747 if (!nd->nd_repstat) { 2748 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2749 *tl++ = newnfs_false; 2750 if (eofflag) 2751 *tl = newnfs_true; 2752 else 2753 *tl = newnfs_false; 2754 } 2755 free(cookies, M_TEMP); 2756 free(rbuf, M_TEMP); 2757 2758 out: 2759 NFSEXITCODE2(0, nd); 2760 return (0); 2761 nfsmout: 2762 vput(vp); 2763 NFSEXITCODE2(error, nd); 2764 return (error); 2765 } 2766 2767 /* 2768 * Get the settable attributes out of the mbuf list. 2769 * (Return 0 or EBADRPC) 2770 */ 2771 int 2772 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2773 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2774 { 2775 u_int32_t *tl; 2776 struct nfsv2_sattr *sp; 2777 int error = 0, toclient = 0; 2778 2779 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 2780 case ND_NFSV2: 2781 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2782 /* 2783 * Some old clients didn't fill in the high order 16bits. 2784 * --> check the low order 2 bytes for 0xffff 2785 */ 2786 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 2787 nvap->na_mode = nfstov_mode(sp->sa_mode); 2788 if (sp->sa_uid != newnfs_xdrneg1) 2789 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 2790 if (sp->sa_gid != newnfs_xdrneg1) 2791 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 2792 if (sp->sa_size != newnfs_xdrneg1) 2793 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 2794 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 2795 #ifdef notyet 2796 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 2797 #else 2798 nvap->na_atime.tv_sec = 2799 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 2800 nvap->na_atime.tv_nsec = 0; 2801 #endif 2802 } 2803 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 2804 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 2805 break; 2806 case ND_NFSV3: 2807 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2808 if (*tl == newnfs_true) { 2809 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2810 nvap->na_mode = nfstov_mode(*tl); 2811 } 2812 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2813 if (*tl == newnfs_true) { 2814 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2815 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 2816 } 2817 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2818 if (*tl == newnfs_true) { 2819 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2820 nvap->na_gid = fxdr_unsigned(gid_t, *tl); 2821 } 2822 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2823 if (*tl == newnfs_true) { 2824 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2825 nvap->na_size = fxdr_hyper(tl); 2826 } 2827 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2828 switch (fxdr_unsigned(int, *tl)) { 2829 case NFSV3SATTRTIME_TOCLIENT: 2830 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2831 fxdr_nfsv3time(tl, &nvap->na_atime); 2832 toclient = 1; 2833 break; 2834 case NFSV3SATTRTIME_TOSERVER: 2835 vfs_timestamp(&nvap->na_atime); 2836 nvap->na_vaflags |= VA_UTIMES_NULL; 2837 break; 2838 } 2839 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2840 switch (fxdr_unsigned(int, *tl)) { 2841 case NFSV3SATTRTIME_TOCLIENT: 2842 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2843 fxdr_nfsv3time(tl, &nvap->na_mtime); 2844 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2845 break; 2846 case NFSV3SATTRTIME_TOSERVER: 2847 vfs_timestamp(&nvap->na_mtime); 2848 if (!toclient) 2849 nvap->na_vaflags |= VA_UTIMES_NULL; 2850 break; 2851 } 2852 break; 2853 case ND_NFSV4: 2854 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); 2855 } 2856 nfsmout: 2857 NFSEXITCODE2(error, nd); 2858 return (error); 2859 } 2860 2861 /* 2862 * Handle the setable attributes for V4. 2863 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 2864 */ 2865 int 2866 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2867 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2868 { 2869 u_int32_t *tl; 2870 int attrsum = 0; 2871 int i, j; 2872 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 2873 int moderet, toclient = 0; 2874 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 2875 uid_t uid; 2876 gid_t gid; 2877 u_short mode, mask; /* Same type as va_mode. */ 2878 struct vattr va; 2879 2880 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 2881 if (error) 2882 goto nfsmout; 2883 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2884 attrsize = fxdr_unsigned(int, *tl); 2885 2886 /* 2887 * Loop around getting the setable attributes. If an unsupported 2888 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 2889 */ 2890 if (retnotsup) { 2891 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2892 bitpos = NFSATTRBIT_MAX; 2893 } else { 2894 bitpos = 0; 2895 } 2896 moderet = 0; 2897 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 2898 if (attrsum > attrsize) { 2899 error = NFSERR_BADXDR; 2900 goto nfsmout; 2901 } 2902 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 2903 switch (bitpos) { 2904 case NFSATTRBIT_SIZE: 2905 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 2906 if (vp != NULL && vp->v_type != VREG) { 2907 error = (vp->v_type == VDIR) ? NFSERR_ISDIR : 2908 NFSERR_INVAL; 2909 goto nfsmout; 2910 } 2911 nvap->na_size = fxdr_hyper(tl); 2912 attrsum += NFSX_HYPER; 2913 break; 2914 case NFSATTRBIT_ACL: 2915 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize, 2916 p); 2917 if (error) 2918 goto nfsmout; 2919 if (aceerr && !nd->nd_repstat) 2920 nd->nd_repstat = aceerr; 2921 attrsum += aclsize; 2922 break; 2923 case NFSATTRBIT_ARCHIVE: 2924 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2925 if (!nd->nd_repstat) 2926 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2927 attrsum += NFSX_UNSIGNED; 2928 break; 2929 case NFSATTRBIT_HIDDEN: 2930 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2931 if (!nd->nd_repstat) 2932 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2933 attrsum += NFSX_UNSIGNED; 2934 break; 2935 case NFSATTRBIT_MIMETYPE: 2936 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2937 i = fxdr_unsigned(int, *tl); 2938 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 2939 if (error) 2940 goto nfsmout; 2941 if (!nd->nd_repstat) 2942 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2943 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 2944 break; 2945 case NFSATTRBIT_MODE: 2946 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ 2947 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2948 nvap->na_mode = nfstov_mode(*tl); 2949 attrsum += NFSX_UNSIGNED; 2950 break; 2951 case NFSATTRBIT_OWNER: 2952 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2953 j = fxdr_unsigned(int, *tl); 2954 if (j < 0) { 2955 error = NFSERR_BADXDR; 2956 goto nfsmout; 2957 } 2958 if (j > NFSV4_SMALLSTR) 2959 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 2960 else 2961 cp = namestr; 2962 error = nfsrv_mtostr(nd, cp, j); 2963 if (error) { 2964 if (j > NFSV4_SMALLSTR) 2965 free(cp, M_NFSSTRING); 2966 goto nfsmout; 2967 } 2968 if (!nd->nd_repstat) { 2969 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, 2970 &uid); 2971 if (!nd->nd_repstat) 2972 nvap->na_uid = uid; 2973 } 2974 if (j > NFSV4_SMALLSTR) 2975 free(cp, M_NFSSTRING); 2976 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 2977 break; 2978 case NFSATTRBIT_OWNERGROUP: 2979 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2980 j = fxdr_unsigned(int, *tl); 2981 if (j < 0) { 2982 error = NFSERR_BADXDR; 2983 goto nfsmout; 2984 } 2985 if (j > NFSV4_SMALLSTR) 2986 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 2987 else 2988 cp = namestr; 2989 error = nfsrv_mtostr(nd, cp, j); 2990 if (error) { 2991 if (j > NFSV4_SMALLSTR) 2992 free(cp, M_NFSSTRING); 2993 goto nfsmout; 2994 } 2995 if (!nd->nd_repstat) { 2996 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, 2997 &gid); 2998 if (!nd->nd_repstat) 2999 nvap->na_gid = gid; 3000 } 3001 if (j > NFSV4_SMALLSTR) 3002 free(cp, M_NFSSTRING); 3003 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3004 break; 3005 case NFSATTRBIT_SYSTEM: 3006 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3007 if (!nd->nd_repstat) 3008 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3009 attrsum += NFSX_UNSIGNED; 3010 break; 3011 case NFSATTRBIT_TIMEACCESSSET: 3012 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3013 attrsum += NFSX_UNSIGNED; 3014 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3015 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3016 fxdr_nfsv4time(tl, &nvap->na_atime); 3017 toclient = 1; 3018 attrsum += NFSX_V4TIME; 3019 } else { 3020 vfs_timestamp(&nvap->na_atime); 3021 nvap->na_vaflags |= VA_UTIMES_NULL; 3022 } 3023 break; 3024 case NFSATTRBIT_TIMEBACKUP: 3025 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3026 if (!nd->nd_repstat) 3027 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3028 attrsum += NFSX_V4TIME; 3029 break; 3030 case NFSATTRBIT_TIMECREATE: 3031 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3032 fxdr_nfsv4time(tl, &nvap->na_btime); 3033 attrsum += NFSX_V4TIME; 3034 break; 3035 case NFSATTRBIT_TIMEMODIFYSET: 3036 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3037 attrsum += NFSX_UNSIGNED; 3038 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3039 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3040 fxdr_nfsv4time(tl, &nvap->na_mtime); 3041 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3042 attrsum += NFSX_V4TIME; 3043 } else { 3044 vfs_timestamp(&nvap->na_mtime); 3045 if (!toclient) 3046 nvap->na_vaflags |= VA_UTIMES_NULL; 3047 } 3048 break; 3049 case NFSATTRBIT_MODESETMASKED: 3050 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3051 mode = fxdr_unsigned(u_short, *tl++); 3052 mask = fxdr_unsigned(u_short, *tl); 3053 /* 3054 * vp == NULL implies an Open/Create operation. 3055 * This attribute can only be used for Setattr and 3056 * only for NFSv4.1 or higher. 3057 * If moderet != 0, a mode attribute has also been 3058 * specified and this attribute cannot be done in the 3059 * same Setattr operation. 3060 */ 3061 if ((nd->nd_flag & ND_NFSV41) == 0) 3062 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3063 else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 || 3064 vp == NULL) 3065 nd->nd_repstat = NFSERR_INVAL; 3066 else if (moderet == 0) 3067 moderet = VOP_GETATTR(vp, &va, nd->nd_cred); 3068 if (moderet == 0) 3069 nvap->na_mode = (mode & mask) | 3070 (va.va_mode & ~mask); 3071 else 3072 nd->nd_repstat = moderet; 3073 attrsum += 2 * NFSX_UNSIGNED; 3074 break; 3075 default: 3076 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3077 /* 3078 * set bitpos so we drop out of the loop. 3079 */ 3080 bitpos = NFSATTRBIT_MAX; 3081 break; 3082 } 3083 } 3084 3085 /* 3086 * some clients pad the attrlist, so we need to skip over the 3087 * padding. 3088 */ 3089 if (attrsum > attrsize) { 3090 error = NFSERR_BADXDR; 3091 } else { 3092 attrsize = NFSM_RNDUP(attrsize); 3093 if (attrsum < attrsize) 3094 error = nfsm_advance(nd, attrsize - attrsum, -1); 3095 } 3096 nfsmout: 3097 NFSEXITCODE2(error, nd); 3098 return (error); 3099 } 3100 3101 /* 3102 * Check/setup export credentials. 3103 */ 3104 int 3105 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 3106 struct ucred *credanon) 3107 { 3108 int error = 0; 3109 3110 /* 3111 * Check/setup credentials. 3112 */ 3113 if (nd->nd_flag & ND_GSS) 3114 exp->nes_exflag &= ~MNT_EXPORTANON; 3115 3116 /* 3117 * Check to see if the operation is allowed for this security flavor. 3118 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to 3119 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. 3120 * Also, allow Secinfo, so that it can acquire the correct flavor(s). 3121 */ 3122 if (nfsvno_testexp(nd, exp) && 3123 nd->nd_procnum != NFSV4OP_SECINFO && 3124 nd->nd_procnum != NFSPROC_FSINFO) { 3125 if (nd->nd_flag & ND_NFSV4) 3126 error = NFSERR_WRONGSEC; 3127 else 3128 error = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3129 goto out; 3130 } 3131 3132 /* 3133 * Check to see if the file system is exported V4 only. 3134 */ 3135 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 3136 error = NFSERR_PROGNOTV4; 3137 goto out; 3138 } 3139 3140 /* 3141 * Now, map the user credentials. 3142 * (Note that ND_AUTHNONE will only be set for an NFSv3 3143 * Fsinfo RPC. If set for anything else, this code might need 3144 * to change.) 3145 */ 3146 if (NFSVNO_EXPORTED(exp)) { 3147 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || 3148 NFSVNO_EXPORTANON(exp) || 3149 (nd->nd_flag & ND_AUTHNONE) != 0) { 3150 nd->nd_cred->cr_uid = credanon->cr_uid; 3151 nd->nd_cred->cr_gid = credanon->cr_gid; 3152 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 3153 credanon->cr_groups); 3154 } else if ((nd->nd_flag & ND_GSS) == 0) { 3155 /* 3156 * If using AUTH_SYS, call nfsrv_getgrpscred() to see 3157 * if there is a replacement credential with a group 3158 * list set up by "nfsuserd -manage-gids". 3159 * If there is no replacement, nfsrv_getgrpscred() 3160 * simply returns its argument. 3161 */ 3162 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); 3163 } 3164 } 3165 3166 out: 3167 NFSEXITCODE2(error, nd); 3168 return (error); 3169 } 3170 3171 /* 3172 * Check exports. 3173 */ 3174 int 3175 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 3176 struct ucred **credp) 3177 { 3178 int error; 3179 3180 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3181 &exp->nes_numsecflavor, exp->nes_secflavors); 3182 if (error) { 3183 if (nfs_rootfhset) { 3184 exp->nes_exflag = 0; 3185 exp->nes_numsecflavor = 0; 3186 error = 0; 3187 } 3188 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3189 MAXSECFLAVORS) { 3190 printf("nfsvno_checkexp: numsecflavors out of range\n"); 3191 exp->nes_numsecflavor = 0; 3192 error = EACCES; 3193 } 3194 NFSEXITCODE(error); 3195 return (error); 3196 } 3197 3198 /* 3199 * Get a vnode for a file handle and export stuff. 3200 */ 3201 int 3202 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 3203 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 3204 struct ucred **credp) 3205 { 3206 int error; 3207 3208 *credp = NULL; 3209 exp->nes_numsecflavor = 0; 3210 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); 3211 if (error != 0) 3212 /* Make sure the server replies ESTALE to the client. */ 3213 error = ESTALE; 3214 if (nam && !error) { 3215 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3216 &exp->nes_numsecflavor, exp->nes_secflavors); 3217 if (error) { 3218 if (nfs_rootfhset) { 3219 exp->nes_exflag = 0; 3220 exp->nes_numsecflavor = 0; 3221 error = 0; 3222 } else { 3223 vput(*vpp); 3224 } 3225 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3226 MAXSECFLAVORS) { 3227 printf("nfsvno_fhtovp: numsecflavors out of range\n"); 3228 exp->nes_numsecflavor = 0; 3229 error = EACCES; 3230 vput(*vpp); 3231 } 3232 } 3233 NFSEXITCODE(error); 3234 return (error); 3235 } 3236 3237 /* 3238 * nfsd_fhtovp() - convert a fh to a vnode ptr 3239 * - look up fsid in mount list (if not found ret error) 3240 * - get vp and export rights by calling nfsvno_fhtovp() 3241 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 3242 * for AUTH_SYS 3243 * - if mpp != NULL, return the mount point so that it can 3244 * be used for vn_finished_write() by the caller 3245 */ 3246 void 3247 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 3248 struct vnode **vpp, struct nfsexstuff *exp, 3249 struct mount **mpp, int startwrite) 3250 { 3251 struct mount *mp, *mpw; 3252 struct ucred *credanon; 3253 fhandle_t *fhp; 3254 int error; 3255 3256 if (mpp != NULL) 3257 *mpp = NULL; 3258 *vpp = NULL; 3259 fhp = (fhandle_t *)nfp->nfsrvfh_data; 3260 mp = vfs_busyfs(&fhp->fh_fsid); 3261 if (mp == NULL) { 3262 nd->nd_repstat = ESTALE; 3263 goto out; 3264 } 3265 3266 if (startwrite) { 3267 mpw = mp; 3268 error = vn_start_write(NULL, &mpw, V_WAIT); 3269 if (error != 0) { 3270 mpw = NULL; 3271 vfs_unbusy(mp); 3272 nd->nd_repstat = ESTALE; 3273 goto out; 3274 } 3275 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) 3276 lktype = LK_EXCLUSIVE; 3277 } else 3278 mpw = NULL; 3279 3280 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 3281 &credanon); 3282 vfs_unbusy(mp); 3283 3284 /* 3285 * For NFSv4 without a pseudo root fs, unexported file handles 3286 * can be returned, so that Lookup works everywhere. 3287 */ 3288 if (!nd->nd_repstat && exp->nes_exflag == 0 && 3289 !(nd->nd_flag & ND_NFSV4)) { 3290 vput(*vpp); 3291 *vpp = NULL; 3292 nd->nd_repstat = EACCES; 3293 } 3294 3295 /* 3296 * If TLS is required by the export, check the flags in nd_flag. 3297 */ 3298 if (nd->nd_repstat == 0 && ((NFSVNO_EXTLS(exp) && 3299 (nd->nd_flag & ND_TLS) == 0) || 3300 (NFSVNO_EXTLSCERT(exp) && 3301 (nd->nd_flag & ND_TLSCERT) == 0) || 3302 (NFSVNO_EXTLSCERTUSER(exp) && 3303 (nd->nd_flag & ND_TLSCERTUSER) == 0))) { 3304 vput(*vpp); 3305 nd->nd_repstat = NFSERR_ACCES; 3306 } 3307 3308 /* 3309 * Personally, I've never seen any point in requiring a 3310 * reserved port#, since only in the rare case where the 3311 * clients are all boxes with secure system privileges, 3312 * does it provide any enhanced security, but... some people 3313 * believe it to be useful and keep putting this code back in. 3314 * (There is also some "security checker" out there that 3315 * complains if the nfs server doesn't enforce this.) 3316 * However, note the following: 3317 * RFC3530 (NFSv4) specifies that a reserved port# not be 3318 * required. 3319 * RFC2623 recommends that, if a reserved port# is checked for, 3320 * that there be a way to turn that off--> ifdef'd. 3321 */ 3322 #ifdef NFS_REQRSVPORT 3323 if (!nd->nd_repstat) { 3324 struct sockaddr_in *saddr; 3325 struct sockaddr_in6 *saddr6; 3326 3327 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 3328 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 3329 if (!(nd->nd_flag & ND_NFSV4) && 3330 ((saddr->sin_family == AF_INET && 3331 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 3332 (saddr6->sin6_family == AF_INET6 && 3333 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 3334 vput(*vpp); 3335 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3336 } 3337 } 3338 #endif /* NFS_REQRSVPORT */ 3339 3340 /* 3341 * Check/setup credentials. 3342 */ 3343 if (!nd->nd_repstat) { 3344 nd->nd_saveduid = nd->nd_cred->cr_uid; 3345 nd->nd_repstat = nfsd_excred(nd, exp, credanon); 3346 if (nd->nd_repstat) 3347 vput(*vpp); 3348 } 3349 if (credanon != NULL) 3350 crfree(credanon); 3351 if (nd->nd_repstat) { 3352 vn_finished_write(mpw); 3353 *vpp = NULL; 3354 } else if (mpp != NULL) { 3355 *mpp = mpw; 3356 } 3357 3358 out: 3359 NFSEXITCODE2(0, nd); 3360 } 3361 3362 /* 3363 * glue for fp. 3364 */ 3365 static int 3366 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 3367 { 3368 struct filedesc *fdp; 3369 struct file *fp; 3370 int error = 0; 3371 3372 fdp = p->td_proc->p_fd; 3373 if (fd < 0 || fd >= fdp->fd_nfiles || 3374 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { 3375 error = EBADF; 3376 goto out; 3377 } 3378 *fpp = fp; 3379 3380 out: 3381 NFSEXITCODE(error); 3382 return (error); 3383 } 3384 3385 /* 3386 * Called from nfssvc() to update the exports list. Just call 3387 * vfs_export(). This has to be done, since the v4 root fake fs isn't 3388 * in the mount list. 3389 */ 3390 int 3391 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 3392 { 3393 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 3394 int error = 0; 3395 struct nameidata nd; 3396 fhandle_t fh; 3397 3398 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export); 3399 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 3400 nfs_rootfhset = 0; 3401 else if (error == 0) { 3402 if (nfsexargp->fspec == NULL) { 3403 error = EPERM; 3404 goto out; 3405 } 3406 /* 3407 * If fspec != NULL, this is the v4root path. 3408 */ 3409 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, 3410 nfsexargp->fspec, p); 3411 if ((error = namei(&nd)) != 0) 3412 goto out; 3413 error = nfsvno_getfh(nd.ni_vp, &fh, p); 3414 vrele(nd.ni_vp); 3415 if (!error) { 3416 nfs_rootfh.nfsrvfh_len = NFSX_MYFH; 3417 NFSBCOPY((caddr_t)&fh, 3418 nfs_rootfh.nfsrvfh_data, 3419 sizeof (fhandle_t)); 3420 nfs_rootfhset = 1; 3421 } 3422 } 3423 3424 out: 3425 NFSEXITCODE(error); 3426 return (error); 3427 } 3428 3429 /* 3430 * This function needs to test to see if the system is near its limit 3431 * for memory allocation via malloc() or mget() and return True iff 3432 * either of these resources are near their limit. 3433 * XXX (For now, this is just a stub.) 3434 */ 3435 int nfsrv_testmalloclimit = 0; 3436 int 3437 nfsrv_mallocmget_limit(void) 3438 { 3439 static int printmesg = 0; 3440 static int testval = 1; 3441 3442 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 3443 if ((printmesg++ % 100) == 0) 3444 printf("nfsd: malloc/mget near limit\n"); 3445 return (1); 3446 } 3447 return (0); 3448 } 3449 3450 /* 3451 * BSD specific initialization of a mount point. 3452 */ 3453 void 3454 nfsd_mntinit(void) 3455 { 3456 static int inited = 0; 3457 3458 if (inited) 3459 return; 3460 inited = 1; 3461 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 3462 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist); 3463 TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist); 3464 nfsv4root_mnt.mnt_export = NULL; 3465 TAILQ_INIT(&nfsv4root_opt); 3466 TAILQ_INIT(&nfsv4root_newopt); 3467 nfsv4root_mnt.mnt_opt = &nfsv4root_opt; 3468 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt; 3469 nfsv4root_mnt.mnt_nvnodelistsize = 0; 3470 nfsv4root_mnt.mnt_lazyvnodelistsize = 0; 3471 } 3472 3473 /* 3474 * Get a vnode for a file handle, without checking exports, etc. 3475 */ 3476 struct vnode * 3477 nfsvno_getvp(fhandle_t *fhp) 3478 { 3479 struct mount *mp; 3480 struct vnode *vp; 3481 int error; 3482 3483 mp = vfs_busyfs(&fhp->fh_fsid); 3484 if (mp == NULL) 3485 return (NULL); 3486 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 3487 vfs_unbusy(mp); 3488 if (error) 3489 return (NULL); 3490 return (vp); 3491 } 3492 3493 /* 3494 * Do a local VOP_ADVLOCK(). 3495 */ 3496 int 3497 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 3498 u_int64_t end, struct thread *td) 3499 { 3500 int error = 0; 3501 struct flock fl; 3502 u_int64_t tlen; 3503 3504 if (nfsrv_dolocallocks == 0) 3505 goto out; 3506 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); 3507 3508 fl.l_whence = SEEK_SET; 3509 fl.l_type = ftype; 3510 fl.l_start = (off_t)first; 3511 if (end == NFS64BITSSET) { 3512 fl.l_len = 0; 3513 } else { 3514 tlen = end - first; 3515 fl.l_len = (off_t)tlen; 3516 } 3517 /* 3518 * For FreeBSD8, the l_pid and l_sysid must be set to the same 3519 * values for all calls, so that all locks will be held by the 3520 * nfsd server. (The nfsd server handles conflicts between the 3521 * various clients.) 3522 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 3523 * bytes, so it can't be put in l_sysid. 3524 */ 3525 if (nfsv4_sysid == 0) 3526 nfsv4_sysid = nlm_acquire_next_sysid(); 3527 fl.l_pid = (pid_t)0; 3528 fl.l_sysid = (int)nfsv4_sysid; 3529 3530 if (ftype == F_UNLCK) 3531 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 3532 (F_POSIX | F_REMOTE)); 3533 else 3534 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 3535 (F_POSIX | F_REMOTE)); 3536 3537 out: 3538 NFSEXITCODE(error); 3539 return (error); 3540 } 3541 3542 /* 3543 * Check the nfsv4 root exports. 3544 */ 3545 int 3546 nfsvno_v4rootexport(struct nfsrv_descript *nd) 3547 { 3548 struct ucred *credanon; 3549 int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; 3550 uint64_t exflags; 3551 3552 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags, 3553 &credanon, &numsecflavor, secflavors); 3554 if (error) { 3555 error = NFSERR_PROGUNAVAIL; 3556 goto out; 3557 } 3558 if (credanon != NULL) 3559 crfree(credanon); 3560 for (i = 0; i < numsecflavor; i++) { 3561 if (secflavors[i] == AUTH_SYS) 3562 nd->nd_flag |= ND_EXAUTHSYS; 3563 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3564 nd->nd_flag |= ND_EXGSS; 3565 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3566 nd->nd_flag |= ND_EXGSSINTEGRITY; 3567 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3568 nd->nd_flag |= ND_EXGSSPRIVACY; 3569 } 3570 3571 /* And set ND_EXxx flags for TLS. */ 3572 if ((exflags & MNT_EXTLS) != 0) { 3573 nd->nd_flag |= ND_EXTLS; 3574 if ((exflags & MNT_EXTLSCERT) != 0) 3575 nd->nd_flag |= ND_EXTLSCERT; 3576 if ((exflags & MNT_EXTLSCERTUSER) != 0) 3577 nd->nd_flag |= ND_EXTLSCERTUSER; 3578 } 3579 3580 out: 3581 NFSEXITCODE(error); 3582 return (error); 3583 } 3584 3585 /* 3586 * Nfs server pseudo system call for the nfsd's 3587 */ 3588 /* 3589 * MPSAFE 3590 */ 3591 static int 3592 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3593 { 3594 struct file *fp; 3595 struct nfsd_addsock_args sockarg; 3596 struct nfsd_nfsd_args nfsdarg; 3597 struct nfsd_nfsd_oargs onfsdarg; 3598 struct nfsd_pnfsd_args pnfsdarg; 3599 struct vnode *vp, *nvp, *curdvp; 3600 struct pnfsdsfile *pf; 3601 struct nfsdevice *ds, *fds; 3602 cap_rights_t rights; 3603 int buflen, error, ret; 3604 char *buf, *cp, *cp2, *cp3; 3605 char fname[PNFS_FILENAME_LEN + 1]; 3606 3607 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3608 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3609 if (error) 3610 goto out; 3611 /* 3612 * Since we don't know what rights might be required, 3613 * pretend that we need them all. It is better to be too 3614 * careful than too reckless. 3615 */ 3616 error = fget(td, sockarg.sock, 3617 cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); 3618 if (error != 0) 3619 goto out; 3620 if (fp->f_type != DTYPE_SOCKET) { 3621 fdrop(fp, td); 3622 error = EPERM; 3623 goto out; 3624 } 3625 error = nfsrvd_addsock(fp); 3626 fdrop(fp, td); 3627 } else if (uap->flag & NFSSVC_NFSDNFSD) { 3628 if (uap->argp == NULL) { 3629 error = EINVAL; 3630 goto out; 3631 } 3632 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { 3633 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); 3634 if (error == 0) { 3635 nfsdarg.principal = onfsdarg.principal; 3636 nfsdarg.minthreads = onfsdarg.minthreads; 3637 nfsdarg.maxthreads = onfsdarg.maxthreads; 3638 nfsdarg.version = 1; 3639 nfsdarg.addr = NULL; 3640 nfsdarg.addrlen = 0; 3641 nfsdarg.dnshost = NULL; 3642 nfsdarg.dnshostlen = 0; 3643 nfsdarg.dspath = NULL; 3644 nfsdarg.dspathlen = 0; 3645 nfsdarg.mdspath = NULL; 3646 nfsdarg.mdspathlen = 0; 3647 nfsdarg.mirrorcnt = 1; 3648 } 3649 } else 3650 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); 3651 if (error) 3652 goto out; 3653 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && 3654 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && 3655 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && 3656 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && 3657 nfsdarg.mirrorcnt >= 1 && 3658 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && 3659 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && 3660 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { 3661 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" 3662 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, 3663 nfsdarg.dspathlen, nfsdarg.dnshostlen, 3664 nfsdarg.mdspathlen, nfsdarg.mirrorcnt); 3665 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); 3666 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); 3667 if (error != 0) { 3668 free(cp, M_TEMP); 3669 goto out; 3670 } 3671 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ 3672 nfsdarg.addr = cp; 3673 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); 3674 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); 3675 if (error != 0) { 3676 free(nfsdarg.addr, M_TEMP); 3677 free(cp, M_TEMP); 3678 goto out; 3679 } 3680 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ 3681 nfsdarg.dnshost = cp; 3682 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); 3683 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); 3684 if (error != 0) { 3685 free(nfsdarg.addr, M_TEMP); 3686 free(nfsdarg.dnshost, M_TEMP); 3687 free(cp, M_TEMP); 3688 goto out; 3689 } 3690 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ 3691 nfsdarg.dspath = cp; 3692 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); 3693 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); 3694 if (error != 0) { 3695 free(nfsdarg.addr, M_TEMP); 3696 free(nfsdarg.dnshost, M_TEMP); 3697 free(nfsdarg.dspath, M_TEMP); 3698 free(cp, M_TEMP); 3699 goto out; 3700 } 3701 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ 3702 nfsdarg.mdspath = cp; 3703 } else { 3704 nfsdarg.addr = NULL; 3705 nfsdarg.addrlen = 0; 3706 nfsdarg.dnshost = NULL; 3707 nfsdarg.dnshostlen = 0; 3708 nfsdarg.dspath = NULL; 3709 nfsdarg.dspathlen = 0; 3710 nfsdarg.mdspath = NULL; 3711 nfsdarg.mdspathlen = 0; 3712 nfsdarg.mirrorcnt = 1; 3713 } 3714 error = nfsrvd_nfsd(td, &nfsdarg); 3715 free(nfsdarg.addr, M_TEMP); 3716 free(nfsdarg.dnshost, M_TEMP); 3717 free(nfsdarg.dspath, M_TEMP); 3718 free(nfsdarg.mdspath, M_TEMP); 3719 } else if (uap->flag & NFSSVC_PNFSDS) { 3720 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); 3721 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || 3722 pnfsdarg.op == PNFSDOP_FORCEDELDS)) { 3723 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3724 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, 3725 NULL); 3726 if (error == 0) 3727 error = nfsrv_deldsserver(pnfsdarg.op, cp, td); 3728 free(cp, M_TEMP); 3729 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { 3730 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3731 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; 3732 buf = malloc(buflen, M_TEMP, M_WAITOK); 3733 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, 3734 NULL); 3735 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); 3736 if (error == 0 && pnfsdarg.dspath != NULL) { 3737 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3738 error = copyinstr(pnfsdarg.dspath, cp2, 3739 PATH_MAX + 1, NULL); 3740 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", 3741 error); 3742 } else 3743 cp2 = NULL; 3744 if (error == 0 && pnfsdarg.curdspath != NULL) { 3745 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3746 error = copyinstr(pnfsdarg.curdspath, cp3, 3747 PATH_MAX + 1, NULL); 3748 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", 3749 error); 3750 } else 3751 cp3 = NULL; 3752 curdvp = NULL; 3753 fds = NULL; 3754 if (error == 0) 3755 error = nfsrv_mdscopymr(cp, cp2, cp3, buf, 3756 &buflen, fname, td, &vp, &nvp, &pf, &ds, 3757 &fds); 3758 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); 3759 if (error == 0) { 3760 if (pf->dsf_dir >= nfsrv_dsdirsize) { 3761 printf("copymr: dsdir out of range\n"); 3762 pf->dsf_dir = 0; 3763 } 3764 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); 3765 error = nfsrv_copymr(vp, nvp, 3766 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, 3767 (struct pnfsdsfile *)buf, 3768 buflen / sizeof(*pf), td->td_ucred, td); 3769 vput(vp); 3770 vput(nvp); 3771 if (fds != NULL && error == 0) { 3772 curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; 3773 ret = vn_lock(curdvp, LK_EXCLUSIVE); 3774 if (ret == 0) { 3775 nfsrv_dsremove(curdvp, fname, 3776 td->td_ucred, td); 3777 NFSVOPUNLOCK(curdvp); 3778 } 3779 } 3780 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); 3781 } 3782 free(cp, M_TEMP); 3783 free(cp2, M_TEMP); 3784 free(cp3, M_TEMP); 3785 free(buf, M_TEMP); 3786 } 3787 } else { 3788 error = nfssvc_srvcall(td, uap, td->td_ucred); 3789 } 3790 3791 out: 3792 NFSEXITCODE(error); 3793 return (error); 3794 } 3795 3796 static int 3797 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 3798 { 3799 struct nfsex_args export; 3800 struct nfsex_oldargs oexp; 3801 struct file *fp = NULL; 3802 int stablefd, i, len; 3803 struct nfsd_clid adminrevoke; 3804 struct nfsd_dumplist dumplist; 3805 struct nfsd_dumpclients *dumpclients; 3806 struct nfsd_dumplocklist dumplocklist; 3807 struct nfsd_dumplocks *dumplocks; 3808 struct nameidata nd; 3809 vnode_t vp; 3810 int error = EINVAL, igotlock; 3811 struct proc *procp; 3812 gid_t *grps; 3813 static int suspend_nfsd = 0; 3814 3815 if (uap->flag & NFSSVC_PUBLICFH) { 3816 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 3817 sizeof (fhandle_t)); 3818 error = copyin(uap->argp, 3819 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 3820 if (!error) 3821 nfs_pubfhset = 1; 3822 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 3823 (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { 3824 error = copyin(uap->argp,(caddr_t)&export, 3825 sizeof (struct nfsex_args)); 3826 if (!error) { 3827 grps = NULL; 3828 if (export.export.ex_ngroups > NGROUPS_MAX || 3829 export.export.ex_ngroups < 0) 3830 error = EINVAL; 3831 else if (export.export.ex_ngroups > 0) { 3832 grps = malloc(export.export.ex_ngroups * 3833 sizeof(gid_t), M_TEMP, M_WAITOK); 3834 error = copyin(export.export.ex_groups, grps, 3835 export.export.ex_ngroups * sizeof(gid_t)); 3836 export.export.ex_groups = grps; 3837 } else 3838 export.export.ex_groups = NULL; 3839 if (!error) 3840 error = nfsrv_v4rootexport(&export, cred, p); 3841 free(grps, M_TEMP); 3842 } 3843 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 3844 NFSSVC_V4ROOTEXPORT) { 3845 error = copyin(uap->argp,(caddr_t)&oexp, 3846 sizeof (struct nfsex_oldargs)); 3847 if (!error) { 3848 memset(&export.export, 0, sizeof(export.export)); 3849 export.export.ex_flags = (uint64_t)oexp.export.ex_flags; 3850 export.export.ex_root = oexp.export.ex_root; 3851 export.export.ex_uid = oexp.export.ex_anon.cr_uid; 3852 export.export.ex_ngroups = 3853 oexp.export.ex_anon.cr_ngroups; 3854 export.export.ex_groups = NULL; 3855 if (export.export.ex_ngroups > XU_NGROUPS || 3856 export.export.ex_ngroups < 0) 3857 error = EINVAL; 3858 else if (export.export.ex_ngroups > 0) { 3859 export.export.ex_groups = malloc( 3860 export.export.ex_ngroups * sizeof(gid_t), 3861 M_TEMP, M_WAITOK); 3862 for (i = 0; i < export.export.ex_ngroups; i++) 3863 export.export.ex_groups[i] = 3864 oexp.export.ex_anon.cr_groups[i]; 3865 } 3866 export.export.ex_addr = oexp.export.ex_addr; 3867 export.export.ex_addrlen = oexp.export.ex_addrlen; 3868 export.export.ex_mask = oexp.export.ex_mask; 3869 export.export.ex_masklen = oexp.export.ex_masklen; 3870 export.export.ex_indexfile = oexp.export.ex_indexfile; 3871 export.export.ex_numsecflavors = 3872 oexp.export.ex_numsecflavors; 3873 if (export.export.ex_numsecflavors >= MAXSECFLAVORS || 3874 export.export.ex_numsecflavors < 0) 3875 error = EINVAL; 3876 else { 3877 for (i = 0; i < export.export.ex_numsecflavors; 3878 i++) 3879 export.export.ex_secflavors[i] = 3880 oexp.export.ex_secflavors[i]; 3881 } 3882 export.fspec = oexp.fspec; 3883 if (error == 0) 3884 error = nfsrv_v4rootexport(&export, cred, p); 3885 free(export.export.ex_groups, M_TEMP); 3886 } 3887 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 3888 nfs_pubfhset = 0; 3889 error = 0; 3890 } else if (uap->flag & NFSSVC_STABLERESTART) { 3891 error = copyin(uap->argp, (caddr_t)&stablefd, 3892 sizeof (int)); 3893 if (!error) 3894 error = fp_getfvp(p, stablefd, &fp, &vp); 3895 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 3896 error = EBADF; 3897 if (!error && newnfs_numnfsd != 0) 3898 error = EPERM; 3899 if (!error) { 3900 nfsrv_stablefirst.nsf_fp = fp; 3901 nfsrv_setupstable(p); 3902 } 3903 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 3904 error = copyin(uap->argp, (caddr_t)&adminrevoke, 3905 sizeof (struct nfsd_clid)); 3906 if (!error) 3907 error = nfsrv_adminrevoke(&adminrevoke, p); 3908 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 3909 error = copyin(uap->argp, (caddr_t)&dumplist, 3910 sizeof (struct nfsd_dumplist)); 3911 if (!error && (dumplist.ndl_size < 1 || 3912 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 3913 error = EPERM; 3914 if (!error) { 3915 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 3916 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 3917 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 3918 error = copyout(dumpclients, dumplist.ndl_list, len); 3919 free(dumpclients, M_TEMP); 3920 } 3921 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 3922 error = copyin(uap->argp, (caddr_t)&dumplocklist, 3923 sizeof (struct nfsd_dumplocklist)); 3924 if (!error && (dumplocklist.ndllck_size < 1 || 3925 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 3926 error = EPERM; 3927 if (!error) 3928 error = nfsrv_lookupfilename(&nd, 3929 dumplocklist.ndllck_fname, p); 3930 if (!error) { 3931 len = sizeof (struct nfsd_dumplocks) * 3932 dumplocklist.ndllck_size; 3933 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 3934 nfsrv_dumplocks(nd.ni_vp, dumplocks, 3935 dumplocklist.ndllck_size, p); 3936 vput(nd.ni_vp); 3937 error = copyout(dumplocks, dumplocklist.ndllck_list, 3938 len); 3939 free(dumplocks, M_TEMP); 3940 } 3941 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 3942 procp = p->td_proc; 3943 PROC_LOCK(procp); 3944 nfsd_master_pid = procp->p_pid; 3945 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 3946 nfsd_master_start = procp->p_stats->p_start; 3947 nfsd_master_proc = procp; 3948 PROC_UNLOCK(procp); 3949 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { 3950 NFSLOCKV4ROOTMUTEX(); 3951 if (suspend_nfsd == 0) { 3952 /* Lock out all nfsd threads */ 3953 do { 3954 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1, 3955 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 3956 } while (igotlock == 0 && suspend_nfsd == 0); 3957 suspend_nfsd = 1; 3958 } 3959 NFSUNLOCKV4ROOTMUTEX(); 3960 error = 0; 3961 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { 3962 NFSLOCKV4ROOTMUTEX(); 3963 if (suspend_nfsd != 0) { 3964 nfsv4_unlock(&nfsd_suspend_lock, 0); 3965 suspend_nfsd = 0; 3966 } 3967 NFSUNLOCKV4ROOTMUTEX(); 3968 error = 0; 3969 } 3970 3971 NFSEXITCODE(error); 3972 return (error); 3973 } 3974 3975 /* 3976 * Check exports. 3977 * Returns 0 if ok, 1 otherwise. 3978 */ 3979 int 3980 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 3981 { 3982 int i; 3983 3984 /* 3985 * This seems odd, but allow the case where the security flavor 3986 * list is empty. This happens when NFSv4 is traversing non-exported 3987 * file systems. Exported file systems should always have a non-empty 3988 * security flavor list. 3989 */ 3990 if (exp->nes_numsecflavor == 0) 3991 return (0); 3992 3993 for (i = 0; i < exp->nes_numsecflavor; i++) { 3994 /* 3995 * The tests for privacy and integrity must be first, 3996 * since ND_GSS is set for everything but AUTH_SYS. 3997 */ 3998 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 3999 (nd->nd_flag & ND_GSSPRIVACY)) 4000 return (0); 4001 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 4002 (nd->nd_flag & ND_GSSINTEGRITY)) 4003 return (0); 4004 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 4005 (nd->nd_flag & ND_GSS)) 4006 return (0); 4007 if (exp->nes_secflavors[i] == AUTH_SYS && 4008 (nd->nd_flag & ND_GSS) == 0) 4009 return (0); 4010 } 4011 return (1); 4012 } 4013 4014 /* 4015 * Calculate a hash value for the fid in a file handle. 4016 */ 4017 uint32_t 4018 nfsrv_hashfh(fhandle_t *fhp) 4019 { 4020 uint32_t hashval; 4021 4022 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 4023 return (hashval); 4024 } 4025 4026 /* 4027 * Calculate a hash value for the sessionid. 4028 */ 4029 uint32_t 4030 nfsrv_hashsessionid(uint8_t *sessionid) 4031 { 4032 uint32_t hashval; 4033 4034 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); 4035 return (hashval); 4036 } 4037 4038 /* 4039 * Signal the userland master nfsd to backup the stable restart file. 4040 */ 4041 void 4042 nfsrv_backupstable(void) 4043 { 4044 struct proc *procp; 4045 4046 if (nfsd_master_proc != NULL) { 4047 procp = pfind(nfsd_master_pid); 4048 /* Try to make sure it is the correct process. */ 4049 if (procp == nfsd_master_proc && 4050 procp->p_stats->p_start.tv_sec == 4051 nfsd_master_start.tv_sec && 4052 procp->p_stats->p_start.tv_usec == 4053 nfsd_master_start.tv_usec && 4054 strcmp(procp->p_comm, nfsd_master_comm) == 0) 4055 kern_psignal(procp, SIGUSR2); 4056 else 4057 nfsd_master_proc = NULL; 4058 4059 if (procp != NULL) 4060 PROC_UNLOCK(procp); 4061 } 4062 } 4063 4064 /* 4065 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. 4066 * The arguments are in a structure, so that they can be passed through 4067 * taskqueue for a kernel process to execute this function. 4068 */ 4069 struct nfsrvdscreate { 4070 int done; 4071 int inprog; 4072 struct task tsk; 4073 struct ucred *tcred; 4074 struct vnode *dvp; 4075 NFSPROC_T *p; 4076 struct pnfsdsfile *pf; 4077 int err; 4078 fhandle_t fh; 4079 struct vattr va; 4080 struct vattr createva; 4081 }; 4082 4083 int 4084 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, 4085 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, 4086 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) 4087 { 4088 struct vnode *nvp; 4089 struct nameidata named; 4090 struct vattr va; 4091 char *bufp; 4092 u_long *hashp; 4093 struct nfsnode *np; 4094 struct nfsmount *nmp; 4095 int error; 4096 4097 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, 4098 LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE); 4099 nfsvno_setpathbuf(&named, &bufp, &hashp); 4100 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 4101 named.ni_cnd.cn_thread = p; 4102 named.ni_cnd.cn_nameptr = bufp; 4103 if (fnamep != NULL) { 4104 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); 4105 named.ni_cnd.cn_namelen = strlen(bufp); 4106 } else 4107 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); 4108 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); 4109 4110 /* Create the date file in the DS mount. */ 4111 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4112 if (error == 0) { 4113 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); 4114 vref(dvp); 4115 VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false); 4116 if (error == 0) { 4117 /* Set the ownership of the file. */ 4118 error = VOP_SETATTR(nvp, nvap, tcred); 4119 NFSD_DEBUG(4, "nfsrv_dscreate:" 4120 " setattr-uid=%d\n", error); 4121 if (error != 0) 4122 vput(nvp); 4123 } 4124 if (error != 0) 4125 printf("pNFS: pnfscreate failed=%d\n", error); 4126 } else 4127 printf("pNFS: pnfscreate vnlock=%d\n", error); 4128 if (error == 0) { 4129 np = VTONFS(nvp); 4130 nmp = VFSTONFS(nvp->v_mount); 4131 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") 4132 != 0 || nmp->nm_nam->sa_len > sizeof( 4133 struct sockaddr_in6) || 4134 np->n_fhp->nfh_len != NFSX_MYFH) { 4135 printf("Bad DS file: fstype=%s salen=%d" 4136 " fhlen=%d\n", 4137 nvp->v_mount->mnt_vfc->vfc_name, 4138 nmp->nm_nam->sa_len, np->n_fhp->nfh_len); 4139 error = ENOENT; 4140 } 4141 4142 /* Set extattrs for the DS on the MDS file. */ 4143 if (error == 0) { 4144 if (dsa != NULL) { 4145 error = VOP_GETATTR(nvp, &va, tcred); 4146 if (error == 0) { 4147 dsa->dsa_filerev = va.va_filerev; 4148 dsa->dsa_size = va.va_size; 4149 dsa->dsa_atime = va.va_atime; 4150 dsa->dsa_mtime = va.va_mtime; 4151 dsa->dsa_bytes = va.va_bytes; 4152 } 4153 } 4154 if (error == 0) { 4155 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, 4156 NFSX_MYFH); 4157 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, 4158 nmp->nm_nam->sa_len); 4159 NFSBCOPY(named.ni_cnd.cn_nameptr, 4160 pf->dsf_filename, 4161 sizeof(pf->dsf_filename)); 4162 } 4163 } else 4164 printf("pNFS: pnfscreate can't get DS" 4165 " attr=%d\n", error); 4166 if (nvpp != NULL && error == 0) 4167 *nvpp = nvp; 4168 else 4169 vput(nvp); 4170 } 4171 nfsvno_relpathbuf(&named); 4172 return (error); 4173 } 4174 4175 /* 4176 * Start up the thread that will execute nfsrv_dscreate(). 4177 */ 4178 static void 4179 start_dscreate(void *arg, int pending) 4180 { 4181 struct nfsrvdscreate *dsc; 4182 4183 dsc = (struct nfsrvdscreate *)arg; 4184 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, 4185 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); 4186 dsc->done = 1; 4187 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); 4188 } 4189 4190 /* 4191 * Create a pNFS data file on the Data Server(s). 4192 */ 4193 static void 4194 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, 4195 NFSPROC_T *p) 4196 { 4197 struct nfsrvdscreate *dsc, *tdsc = NULL; 4198 struct nfsdevice *ds, *tds, *fds; 4199 struct mount *mp; 4200 struct pnfsdsfile *pf, *tpf; 4201 struct pnfsdsattr dsattr; 4202 struct vattr va; 4203 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4204 struct nfsmount *nmp; 4205 fhandle_t fh; 4206 uid_t vauid; 4207 gid_t vagid; 4208 u_short vamode; 4209 struct ucred *tcred; 4210 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; 4211 int failpos, timo; 4212 4213 /* Get a DS server directory in a round-robin order. */ 4214 mirrorcnt = 1; 4215 mp = vp->v_mount; 4216 ds = fds = NULL; 4217 NFSDDSLOCK(); 4218 /* 4219 * Search for the first entry that handles this MDS fs, but use the 4220 * first entry for all MDS fs's otherwise. 4221 */ 4222 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { 4223 if (tds->nfsdev_nmp != NULL) { 4224 if (tds->nfsdev_mdsisset == 0 && ds == NULL) 4225 ds = tds; 4226 else if (tds->nfsdev_mdsisset != 0 && fsidcmp( 4227 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { 4228 ds = fds = tds; 4229 break; 4230 } 4231 } 4232 } 4233 if (ds == NULL) { 4234 NFSDDSUNLOCK(); 4235 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); 4236 return; 4237 } 4238 i = dsdir[0] = ds->nfsdev_nextdir; 4239 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; 4240 dvp[0] = ds->nfsdev_dsdir[i]; 4241 tds = TAILQ_NEXT(ds, nfsdev_list); 4242 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { 4243 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { 4244 if (tds->nfsdev_nmp != NULL && 4245 ((tds->nfsdev_mdsisset == 0 && fds == NULL) || 4246 (tds->nfsdev_mdsisset != 0 && fds != NULL && 4247 fsidcmp(&mp->mnt_stat.f_fsid, 4248 &tds->nfsdev_mdsfsid) == 0))) { 4249 dsdir[mirrorcnt] = i; 4250 dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; 4251 mirrorcnt++; 4252 if (mirrorcnt >= nfsrv_maxpnfsmirror) 4253 break; 4254 } 4255 } 4256 } 4257 /* Put at end of list to implement round-robin usage. */ 4258 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 4259 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 4260 NFSDDSUNLOCK(); 4261 dsc = NULL; 4262 if (mirrorcnt > 1) 4263 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, 4264 M_WAITOK | M_ZERO); 4265 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | 4266 M_ZERO); 4267 4268 error = nfsvno_getfh(vp, &fh, p); 4269 if (error == 0) 4270 error = VOP_GETATTR(vp, &va, cred); 4271 if (error == 0) { 4272 /* Set the attributes for "vp" to Setattr the DS vp. */ 4273 vauid = va.va_uid; 4274 vagid = va.va_gid; 4275 vamode = va.va_mode; 4276 VATTR_NULL(&va); 4277 va.va_uid = vauid; 4278 va.va_gid = vagid; 4279 va.va_mode = vamode; 4280 va.va_size = 0; 4281 } else 4282 printf("pNFS: pnfscreate getfh+attr=%d\n", error); 4283 4284 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, 4285 cred->cr_gid); 4286 /* Make data file name based on FH. */ 4287 tcred = newnfs_getcred(); 4288 4289 /* 4290 * Create the file on each DS mirror, using kernel process(es) for the 4291 * additional mirrors. 4292 */ 4293 failpos = -1; 4294 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { 4295 tpf->dsf_dir = dsdir[i]; 4296 tdsc->tcred = tcred; 4297 tdsc->p = p; 4298 tdsc->pf = tpf; 4299 tdsc->createva = *vap; 4300 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); 4301 tdsc->va = va; 4302 tdsc->dvp = dvp[i]; 4303 tdsc->done = 0; 4304 tdsc->inprog = 0; 4305 tdsc->err = 0; 4306 ret = EIO; 4307 if (nfs_pnfsiothreads != 0) { 4308 ret = nfs_pnfsio(start_dscreate, tdsc); 4309 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); 4310 } 4311 if (ret != 0) { 4312 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, 4313 NULL, tcred, p, NULL); 4314 if (ret != 0) { 4315 KASSERT(error == 0, ("nfsrv_dscreate err=%d", 4316 error)); 4317 if (failpos == -1 && nfsds_failerr(ret)) 4318 failpos = i; 4319 else 4320 error = ret; 4321 } 4322 } 4323 } 4324 if (error == 0) { 4325 tpf->dsf_dir = dsdir[mirrorcnt - 1]; 4326 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, 4327 &dsattr, NULL, tcred, p, NULL); 4328 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { 4329 failpos = mirrorcnt - 1; 4330 error = 0; 4331 } 4332 } 4333 timo = hz / 50; /* Wait for 20msec. */ 4334 if (timo < 1) 4335 timo = 1; 4336 /* Wait for kernel task(s) to complete. */ 4337 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { 4338 while (tdsc->inprog != 0 && tdsc->done == 0) 4339 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); 4340 if (tdsc->err != 0) { 4341 if (failpos == -1 && nfsds_failerr(tdsc->err)) 4342 failpos = i; 4343 else if (error == 0) 4344 error = tdsc->err; 4345 } 4346 } 4347 4348 /* 4349 * If failpos has been set, that mirror has failed, so it needs 4350 * to be disabled. 4351 */ 4352 if (failpos >= 0) { 4353 nmp = VFSTONFS(dvp[failpos]->v_mount); 4354 NFSLOCKMNT(nmp); 4355 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4356 NFSMNTP_CANCELRPCS)) == 0) { 4357 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4358 NFSUNLOCKMNT(nmp); 4359 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4360 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, 4361 ds); 4362 if (ds != NULL) 4363 nfsrv_killrpcs(nmp); 4364 NFSLOCKMNT(nmp); 4365 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4366 wakeup(nmp); 4367 } 4368 NFSUNLOCKMNT(nmp); 4369 } 4370 4371 NFSFREECRED(tcred); 4372 if (error == 0) { 4373 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); 4374 4375 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", 4376 mirrorcnt, nfsrv_maxpnfsmirror); 4377 /* 4378 * For all mirrors that couldn't be created, fill in the 4379 * *pf structure, but with an IP address == 0.0.0.0. 4380 */ 4381 tpf = pf + mirrorcnt; 4382 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { 4383 *tpf = *pf; 4384 tpf->dsf_sin.sin_family = AF_INET; 4385 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); 4386 tpf->dsf_sin.sin_addr.s_addr = 0; 4387 tpf->dsf_sin.sin_port = 0; 4388 } 4389 4390 error = vn_extattr_set(vp, IO_NODELOCKED, 4391 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 4392 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); 4393 if (error == 0) 4394 error = vn_extattr_set(vp, IO_NODELOCKED, 4395 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", 4396 sizeof(dsattr), (char *)&dsattr, p); 4397 if (error != 0) 4398 printf("pNFS: pnfscreate setextattr=%d\n", 4399 error); 4400 } else 4401 printf("pNFS: pnfscreate=%d\n", error); 4402 free(pf, M_TEMP); 4403 free(dsc, M_TEMP); 4404 } 4405 4406 /* 4407 * Get the information needed to remove the pNFS Data Server file from the 4408 * Metadata file. Upon success, ddvp is set non-NULL to the locked 4409 * DS directory vnode. The caller must unlock *ddvp when done with it. 4410 */ 4411 static void 4412 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, 4413 int *mirrorcntp, char *fname, fhandle_t *fhp) 4414 { 4415 struct vattr va; 4416 struct ucred *tcred; 4417 char *buf; 4418 int buflen, error; 4419 4420 dvpp[0] = NULL; 4421 /* If not an exported regular file or not a pNFS server, just return. */ 4422 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4423 nfsrv_devidcnt == 0) 4424 return; 4425 4426 /* Check to see if this is the last hard link. */ 4427 tcred = newnfs_getcred(); 4428 error = VOP_GETATTR(vp, &va, tcred); 4429 NFSFREECRED(tcred); 4430 if (error != 0) { 4431 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); 4432 return; 4433 } 4434 if (va.va_nlink > 1) 4435 return; 4436 4437 error = nfsvno_getfh(vp, fhp, p); 4438 if (error != 0) { 4439 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); 4440 return; 4441 } 4442 4443 buflen = 1024; 4444 buf = malloc(buflen, M_TEMP, M_WAITOK); 4445 /* Get the directory vnode for the DS mount and the file handle. */ 4446 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, 4447 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); 4448 free(buf, M_TEMP); 4449 if (error != 0) 4450 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); 4451 } 4452 4453 /* 4454 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. 4455 * The arguments are in a structure, so that they can be passed through 4456 * taskqueue for a kernel process to execute this function. 4457 */ 4458 struct nfsrvdsremove { 4459 int done; 4460 int inprog; 4461 struct task tsk; 4462 struct ucred *tcred; 4463 struct vnode *dvp; 4464 NFSPROC_T *p; 4465 int err; 4466 char fname[PNFS_FILENAME_LEN + 1]; 4467 }; 4468 4469 static int 4470 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, 4471 NFSPROC_T *p) 4472 { 4473 struct nameidata named; 4474 struct vnode *nvp; 4475 char *bufp; 4476 u_long *hashp; 4477 int error; 4478 4479 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4480 if (error != 0) 4481 return (error); 4482 named.ni_cnd.cn_nameiop = DELETE; 4483 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 4484 named.ni_cnd.cn_cred = tcred; 4485 named.ni_cnd.cn_thread = p; 4486 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; 4487 nfsvno_setpathbuf(&named, &bufp, &hashp); 4488 named.ni_cnd.cn_nameptr = bufp; 4489 named.ni_cnd.cn_namelen = strlen(fname); 4490 strlcpy(bufp, fname, NAME_MAX); 4491 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); 4492 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 4493 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); 4494 if (error == 0) { 4495 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); 4496 vput(nvp); 4497 } 4498 NFSVOPUNLOCK(dvp); 4499 nfsvno_relpathbuf(&named); 4500 if (error != 0) 4501 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); 4502 return (error); 4503 } 4504 4505 /* 4506 * Start up the thread that will execute nfsrv_dsremove(). 4507 */ 4508 static void 4509 start_dsremove(void *arg, int pending) 4510 { 4511 struct nfsrvdsremove *dsrm; 4512 4513 dsrm = (struct nfsrvdsremove *)arg; 4514 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, 4515 dsrm->p); 4516 dsrm->done = 1; 4517 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); 4518 } 4519 4520 /* 4521 * Remove a pNFS data file from a Data Server. 4522 * nfsrv_pnfsremovesetup() must have been called before the MDS file was 4523 * removed to set up the dvp and fill in the FH. 4524 */ 4525 static void 4526 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, 4527 NFSPROC_T *p) 4528 { 4529 struct ucred *tcred; 4530 struct nfsrvdsremove *dsrm, *tdsrm; 4531 struct nfsdevice *ds; 4532 struct nfsmount *nmp; 4533 int failpos, i, ret, timo; 4534 4535 tcred = newnfs_getcred(); 4536 dsrm = NULL; 4537 if (mirrorcnt > 1) 4538 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); 4539 /* 4540 * Remove the file on each DS mirror, using kernel process(es) for the 4541 * additional mirrors. 4542 */ 4543 failpos = -1; 4544 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4545 tdsrm->tcred = tcred; 4546 tdsrm->p = p; 4547 tdsrm->dvp = dvp[i]; 4548 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); 4549 tdsrm->inprog = 0; 4550 tdsrm->done = 0; 4551 tdsrm->err = 0; 4552 ret = EIO; 4553 if (nfs_pnfsiothreads != 0) { 4554 ret = nfs_pnfsio(start_dsremove, tdsrm); 4555 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); 4556 } 4557 if (ret != 0) { 4558 ret = nfsrv_dsremove(dvp[i], fname, tcred, p); 4559 if (failpos == -1 && nfsds_failerr(ret)) 4560 failpos = i; 4561 } 4562 } 4563 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); 4564 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) 4565 failpos = mirrorcnt - 1; 4566 timo = hz / 50; /* Wait for 20msec. */ 4567 if (timo < 1) 4568 timo = 1; 4569 /* Wait for kernel task(s) to complete. */ 4570 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4571 while (tdsrm->inprog != 0 && tdsrm->done == 0) 4572 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); 4573 if (failpos == -1 && nfsds_failerr(tdsrm->err)) 4574 failpos = i; 4575 } 4576 4577 /* 4578 * If failpos has been set, that mirror has failed, so it needs 4579 * to be disabled. 4580 */ 4581 if (failpos >= 0) { 4582 nmp = VFSTONFS(dvp[failpos]->v_mount); 4583 NFSLOCKMNT(nmp); 4584 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4585 NFSMNTP_CANCELRPCS)) == 0) { 4586 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4587 NFSUNLOCKMNT(nmp); 4588 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4589 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, 4590 ds); 4591 if (ds != NULL) 4592 nfsrv_killrpcs(nmp); 4593 NFSLOCKMNT(nmp); 4594 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4595 wakeup(nmp); 4596 } 4597 NFSUNLOCKMNT(nmp); 4598 } 4599 4600 /* Get rid all layouts for the file. */ 4601 nfsrv_freefilelayouts(fhp); 4602 4603 NFSFREECRED(tcred); 4604 free(dsrm, M_TEMP); 4605 } 4606 4607 /* 4608 * Generate a file name based on the file handle and put it in *bufp. 4609 * Return the number of bytes generated. 4610 */ 4611 static int 4612 nfsrv_putfhname(fhandle_t *fhp, char *bufp) 4613 { 4614 int i; 4615 uint8_t *cp; 4616 const uint8_t *hexdigits = "0123456789abcdef"; 4617 4618 cp = (uint8_t *)fhp; 4619 for (i = 0; i < sizeof(*fhp); i++) { 4620 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; 4621 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; 4622 } 4623 bufp[2 * i] = '\0'; 4624 return (2 * i); 4625 } 4626 4627 /* 4628 * Update the Metadata file's attributes from the DS file when a Read/Write 4629 * layout is returned. 4630 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN 4631 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. 4632 */ 4633 int 4634 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 4635 { 4636 struct ucred *tcred; 4637 int error; 4638 4639 /* Do this as root so that it won't fail with EACCES. */ 4640 tcred = newnfs_getcred(); 4641 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, 4642 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); 4643 NFSFREECRED(tcred); 4644 return (error); 4645 } 4646 4647 /* 4648 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. 4649 */ 4650 static int 4651 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, 4652 NFSPROC_T *p) 4653 { 4654 int error; 4655 4656 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, 4657 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); 4658 return (error); 4659 } 4660 4661 static int 4662 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 4663 struct thread *p, int ioproc, struct mbuf **mpp, char *cp, 4664 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, 4665 off_t *offp, int content, bool *eofp) 4666 { 4667 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; 4668 fhandle_t fh[NFSDEV_MAXMIRRORS]; 4669 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4670 struct nfsdevice *ds; 4671 struct pnfsdsattr dsattr; 4672 struct opnfsdsattr odsattr; 4673 char *buf; 4674 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; 4675 4676 NFSD_DEBUG(4, "in nfsrv_proxyds\n"); 4677 /* 4678 * If not a regular file, not exported or not a pNFS server, 4679 * just return ENOENT. 4680 */ 4681 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4682 nfsrv_devidcnt == 0) 4683 return (ENOENT); 4684 4685 buflen = 1024; 4686 buf = malloc(buflen, M_TEMP, M_WAITOK); 4687 error = 0; 4688 4689 /* 4690 * For Getattr, get the Change attribute (va_filerev) and size (va_size) 4691 * from the MetaData file's extended attribute. 4692 */ 4693 if (ioproc == NFSPROC_GETATTR) { 4694 error = vn_extattr_get(vp, IO_NODELOCKED, 4695 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, 4696 p); 4697 if (error == 0) { 4698 if (buflen == sizeof(odsattr)) { 4699 NFSBCOPY(buf, &odsattr, buflen); 4700 nap->na_filerev = odsattr.dsa_filerev; 4701 nap->na_size = odsattr.dsa_size; 4702 nap->na_atime = odsattr.dsa_atime; 4703 nap->na_mtime = odsattr.dsa_mtime; 4704 /* 4705 * Fake na_bytes by rounding up na_size. 4706 * Since we don't know the block size, just 4707 * use BLKDEV_IOSIZE. 4708 */ 4709 nap->na_bytes = (odsattr.dsa_size + 4710 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); 4711 } else if (buflen == sizeof(dsattr)) { 4712 NFSBCOPY(buf, &dsattr, buflen); 4713 nap->na_filerev = dsattr.dsa_filerev; 4714 nap->na_size = dsattr.dsa_size; 4715 nap->na_atime = dsattr.dsa_atime; 4716 nap->na_mtime = dsattr.dsa_mtime; 4717 nap->na_bytes = dsattr.dsa_bytes; 4718 } else 4719 error = ENXIO; 4720 } 4721 if (error == 0) { 4722 /* 4723 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() 4724 * returns 0, just return now. nfsrv_checkdsattr() 4725 * returns 0 if there is no Read/Write layout 4726 * plus either an Open/Write_access or Write 4727 * delegation issued to a client for the file. 4728 */ 4729 if (nfsrv_pnfsgetdsattr == 0 || 4730 nfsrv_checkdsattr(vp, p) == 0) { 4731 free(buf, M_TEMP); 4732 return (error); 4733 } 4734 } 4735 4736 /* 4737 * Clear ENOATTR so the code below will attempt to do a 4738 * nfsrv_getattrdsrpc() to get the attributes and (re)create 4739 * the extended attribute. 4740 */ 4741 if (error == ENOATTR) 4742 error = 0; 4743 } 4744 4745 origmircnt = -1; 4746 trycnt = 0; 4747 tryagain: 4748 if (error == 0) { 4749 buflen = 1024; 4750 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == 4751 LK_EXCLUSIVE) 4752 printf("nfsrv_proxyds: Readds vp exclusively locked\n"); 4753 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, 4754 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, 4755 NULL, NULL); 4756 if (error == 0) { 4757 for (i = 0; i < mirrorcnt; i++) 4758 nmp[i] = VFSTONFS(dvp[i]->v_mount); 4759 } else 4760 printf("pNFS: proxy getextattr sockaddr=%d\n", error); 4761 } else 4762 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); 4763 if (error == 0) { 4764 failpos = -1; 4765 if (origmircnt == -1) 4766 origmircnt = mirrorcnt; 4767 /* 4768 * If failpos is set to a mirror#, then that mirror has 4769 * failed and will be disabled. For Read, Getattr and Seek, the 4770 * function only tries one mirror, so if that mirror has 4771 * failed, it will need to be retried. As such, increment 4772 * tryitagain for these cases. 4773 * For Write, Setattr and Setacl, the function tries all 4774 * mirrors and will not return an error for the case where 4775 * one mirror has failed. For these cases, the functioning 4776 * mirror(s) will have been modified, so a retry isn't 4777 * necessary. These functions will set failpos for the 4778 * failed mirror#. 4779 */ 4780 if (ioproc == NFSPROC_READDS) { 4781 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], 4782 mpp, mpp2); 4783 if (nfsds_failerr(error) && mirrorcnt > 1) { 4784 /* 4785 * Setting failpos will cause the mirror 4786 * to be disabled and then a retry of this 4787 * read is required. 4788 */ 4789 failpos = 0; 4790 error = 0; 4791 trycnt++; 4792 } 4793 } else if (ioproc == NFSPROC_WRITEDS) 4794 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, 4795 &nmp[0], mirrorcnt, mpp, cp, &failpos); 4796 else if (ioproc == NFSPROC_SETATTR) 4797 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], 4798 mirrorcnt, nap, &failpos); 4799 else if (ioproc == NFSPROC_SETACL) 4800 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], 4801 mirrorcnt, aclp, &failpos); 4802 else if (ioproc == NFSPROC_SEEKDS) { 4803 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, 4804 p, nmp[0]); 4805 if (nfsds_failerr(error) && mirrorcnt > 1) { 4806 /* 4807 * Setting failpos will cause the mirror 4808 * to be disabled and then a retry of this 4809 * read is required. 4810 */ 4811 failpos = 0; 4812 error = 0; 4813 trycnt++; 4814 } 4815 } else if (ioproc == NFSPROC_ALLOCATE) 4816 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, 4817 &nmp[0], mirrorcnt, &failpos); 4818 else { 4819 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, 4820 vp, nmp[mirrorcnt - 1], nap); 4821 if (nfsds_failerr(error) && mirrorcnt > 1) { 4822 /* 4823 * Setting failpos will cause the mirror 4824 * to be disabled and then a retry of this 4825 * getattr is required. 4826 */ 4827 failpos = mirrorcnt - 1; 4828 error = 0; 4829 trycnt++; 4830 } 4831 } 4832 ds = NULL; 4833 if (failpos >= 0) { 4834 failnmp = nmp[failpos]; 4835 NFSLOCKMNT(failnmp); 4836 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | 4837 NFSMNTP_CANCELRPCS)) == 0) { 4838 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4839 NFSUNLOCKMNT(failnmp); 4840 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, 4841 failnmp, p); 4842 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", 4843 failpos, ds); 4844 if (ds != NULL) 4845 nfsrv_killrpcs(failnmp); 4846 NFSLOCKMNT(failnmp); 4847 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4848 wakeup(failnmp); 4849 } 4850 NFSUNLOCKMNT(failnmp); 4851 } 4852 for (i = 0; i < mirrorcnt; i++) 4853 NFSVOPUNLOCK(dvp[i]); 4854 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, 4855 trycnt); 4856 /* Try the Read/Getattr again if a mirror was deleted. */ 4857 if (ds != NULL && trycnt > 0 && trycnt < origmircnt) 4858 goto tryagain; 4859 } else { 4860 /* Return ENOENT for any Extended Attribute error. */ 4861 error = ENOENT; 4862 } 4863 free(buf, M_TEMP); 4864 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); 4865 return (error); 4866 } 4867 4868 /* 4869 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended 4870 * attribute. 4871 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs 4872 * to be checked. If it points to a NULL nmp, then it returns 4873 * a suitable destination. 4874 * curnmp - If non-NULL, it is the source mount for the copy. 4875 */ 4876 int 4877 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, 4878 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, 4879 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, 4880 struct nfsmount *curnmp, int *ippos, int *dsdirp) 4881 { 4882 struct vnode *dvp, *nvp = NULL, **tdvpp; 4883 struct mount *mp; 4884 struct nfsmount *nmp, *newnmp; 4885 struct sockaddr *sad; 4886 struct sockaddr_in *sin; 4887 struct nfsdevice *ds, *tds, *fndds; 4888 struct pnfsdsfile *pf; 4889 uint32_t dsdir; 4890 int error, fhiszero, fnd, gotone, i, mirrorcnt; 4891 4892 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); 4893 *mirrorcntp = 1; 4894 tdvpp = dvpp; 4895 if (nvpp != NULL) 4896 *nvpp = NULL; 4897 if (dvpp != NULL) 4898 *dvpp = NULL; 4899 if (ippos != NULL) 4900 *ippos = -1; 4901 if (newnmpp != NULL) 4902 newnmp = *newnmpp; 4903 else 4904 newnmp = NULL; 4905 mp = vp->v_mount; 4906 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 4907 "pnfsd.dsfile", buflenp, buf, p); 4908 mirrorcnt = *buflenp / sizeof(*pf); 4909 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || 4910 *buflenp != sizeof(*pf) * mirrorcnt)) 4911 error = ENOATTR; 4912 4913 pf = (struct pnfsdsfile *)buf; 4914 /* If curnmp != NULL, check for a match in the mirror list. */ 4915 if (curnmp != NULL && error == 0) { 4916 fnd = 0; 4917 for (i = 0; i < mirrorcnt; i++, pf++) { 4918 sad = (struct sockaddr *)&pf->dsf_sin; 4919 if (nfsaddr2_match(sad, curnmp->nm_nam)) { 4920 if (ippos != NULL) 4921 *ippos = i; 4922 fnd = 1; 4923 break; 4924 } 4925 } 4926 if (fnd == 0) 4927 error = ENXIO; 4928 } 4929 4930 gotone = 0; 4931 pf = (struct pnfsdsfile *)buf; 4932 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, 4933 error); 4934 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { 4935 fhiszero = 0; 4936 sad = (struct sockaddr *)&pf->dsf_sin; 4937 sin = &pf->dsf_sin; 4938 dsdir = pf->dsf_dir; 4939 if (dsdir >= nfsrv_dsdirsize) { 4940 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); 4941 error = ENOATTR; 4942 } else if (nvpp != NULL && newnmp != NULL && 4943 nfsaddr2_match(sad, newnmp->nm_nam)) 4944 error = EEXIST; 4945 if (error == 0) { 4946 if (ippos != NULL && curnmp == NULL && 4947 sad->sa_family == AF_INET && 4948 sin->sin_addr.s_addr == 0) 4949 *ippos = i; 4950 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) 4951 fhiszero = 1; 4952 /* Use the socket address to find the mount point. */ 4953 fndds = NULL; 4954 NFSDDSLOCK(); 4955 /* Find a match for the IP address. */ 4956 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 4957 if (ds->nfsdev_nmp != NULL) { 4958 dvp = ds->nfsdev_dvp; 4959 nmp = VFSTONFS(dvp->v_mount); 4960 if (nmp != ds->nfsdev_nmp) 4961 printf("different2 nmp %p %p\n", 4962 nmp, ds->nfsdev_nmp); 4963 if (nfsaddr2_match(sad, nmp->nm_nam)) { 4964 fndds = ds; 4965 break; 4966 } 4967 } 4968 } 4969 if (fndds != NULL && newnmpp != NULL && 4970 newnmp == NULL) { 4971 /* Search for a place to make a mirror copy. */ 4972 TAILQ_FOREACH(tds, &nfsrv_devidhead, 4973 nfsdev_list) { 4974 if (tds->nfsdev_nmp != NULL && 4975 fndds != tds && 4976 ((tds->nfsdev_mdsisset == 0 && 4977 fndds->nfsdev_mdsisset == 0) || 4978 (tds->nfsdev_mdsisset != 0 && 4979 fndds->nfsdev_mdsisset != 0 && 4980 fsidcmp(&tds->nfsdev_mdsfsid, 4981 &mp->mnt_stat.f_fsid) == 0))) { 4982 *newnmpp = tds->nfsdev_nmp; 4983 break; 4984 } 4985 } 4986 if (tds != NULL) { 4987 /* 4988 * Move this entry to the end of the 4989 * list, so it won't be selected as 4990 * easily the next time. 4991 */ 4992 TAILQ_REMOVE(&nfsrv_devidhead, tds, 4993 nfsdev_list); 4994 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, 4995 nfsdev_list); 4996 } 4997 } 4998 NFSDDSUNLOCK(); 4999 if (fndds != NULL) { 5000 dvp = fndds->nfsdev_dsdir[dsdir]; 5001 if (lktype != 0 || fhiszero != 0 || 5002 (nvpp != NULL && *nvpp == NULL)) { 5003 if (fhiszero != 0) 5004 error = vn_lock(dvp, 5005 LK_EXCLUSIVE); 5006 else if (lktype != 0) 5007 error = vn_lock(dvp, lktype); 5008 else 5009 error = vn_lock(dvp, LK_SHARED); 5010 /* 5011 * If the file handle is all 0's, try to 5012 * do a Lookup against the DS to acquire 5013 * it. 5014 * If dvpp == NULL or the Lookup fails, 5015 * unlock dvp after the call. 5016 */ 5017 if (error == 0 && (fhiszero != 0 || 5018 (nvpp != NULL && *nvpp == NULL))) { 5019 error = nfsrv_pnfslookupds(vp, 5020 dvp, pf, &nvp, p); 5021 if (error == 0) { 5022 if (fhiszero != 0) 5023 nfsrv_pnfssetfh( 5024 vp, pf, 5025 devid, 5026 fnamep, 5027 nvp, p); 5028 if (nvpp != NULL && 5029 *nvpp == NULL) { 5030 *nvpp = nvp; 5031 *dsdirp = dsdir; 5032 } else 5033 vput(nvp); 5034 } 5035 if (error != 0 || lktype == 0) 5036 NFSVOPUNLOCK(dvp); 5037 } 5038 } 5039 if (error == 0) { 5040 gotone++; 5041 NFSD_DEBUG(4, "gotone=%d\n", gotone); 5042 if (devid != NULL) { 5043 NFSBCOPY(fndds->nfsdev_deviceid, 5044 devid, NFSX_V4DEVICEID); 5045 devid += NFSX_V4DEVICEID; 5046 } 5047 if (dvpp != NULL) 5048 *tdvpp++ = dvp; 5049 if (fhp != NULL) 5050 NFSBCOPY(&pf->dsf_fh, fhp++, 5051 NFSX_MYFH); 5052 if (fnamep != NULL && gotone == 1) 5053 strlcpy(fnamep, 5054 pf->dsf_filename, 5055 sizeof(pf->dsf_filename)); 5056 } else 5057 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " 5058 "err=%d\n", error); 5059 } 5060 } 5061 } 5062 if (error == 0 && gotone == 0) 5063 error = ENOENT; 5064 5065 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, 5066 error); 5067 if (error == 0) 5068 *mirrorcntp = gotone; 5069 else { 5070 if (gotone > 0 && dvpp != NULL) { 5071 /* 5072 * If the error didn't occur on the first one and 5073 * dvpp != NULL, the one(s) prior to the failure will 5074 * have locked dvp's that need to be unlocked. 5075 */ 5076 for (i = 0; i < gotone; i++) { 5077 NFSVOPUNLOCK(*dvpp); 5078 *dvpp++ = NULL; 5079 } 5080 } 5081 /* 5082 * If it found the vnode to be copied from before a failure, 5083 * it needs to be vput()'d. 5084 */ 5085 if (nvpp != NULL && *nvpp != NULL) { 5086 vput(*nvpp); 5087 *nvpp = NULL; 5088 } 5089 } 5090 return (error); 5091 } 5092 5093 /* 5094 * Set the extended attribute for the Change attribute. 5095 */ 5096 static int 5097 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 5098 { 5099 struct pnfsdsattr dsattr; 5100 int error; 5101 5102 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); 5103 dsattr.dsa_filerev = nap->na_filerev; 5104 dsattr.dsa_size = nap->na_size; 5105 dsattr.dsa_atime = nap->na_atime; 5106 dsattr.dsa_mtime = nap->na_mtime; 5107 dsattr.dsa_bytes = nap->na_bytes; 5108 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5109 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); 5110 if (error != 0) 5111 printf("pNFS: setextattr=%d\n", error); 5112 return (error); 5113 } 5114 5115 static int 5116 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5117 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) 5118 { 5119 uint32_t *tl; 5120 struct nfsrv_descript *nd; 5121 nfsv4stateid_t st; 5122 struct mbuf *m, *m2; 5123 int error = 0, retlen, tlen, trimlen; 5124 5125 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); 5126 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5127 *mpp = NULL; 5128 /* 5129 * Use a stateid where other is an alternating 01010 pattern and 5130 * seqid is 0xffffffff. This value is not defined as special by 5131 * the RFC and is used by the FreeBSD NFS server to indicate an 5132 * MDS->DS proxy operation. 5133 */ 5134 st.other[0] = 0x55555555; 5135 st.other[1] = 0x55555555; 5136 st.other[2] = 0x55555555; 5137 st.seqid = 0xffffffff; 5138 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5139 NULL, NULL, 0, 0); 5140 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5141 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); 5142 txdr_hyper(off, tl); 5143 *(tl + 2) = txdr_unsigned(len); 5144 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5145 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5146 if (error != 0) { 5147 free(nd, M_TEMP); 5148 return (error); 5149 } 5150 if (nd->nd_repstat == 0) { 5151 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 5152 NFSM_STRSIZ(retlen, len); 5153 if (retlen > 0) { 5154 /* Trim off the pre-data XDR from the mbuf chain. */ 5155 m = nd->nd_mrep; 5156 while (m != NULL && m != nd->nd_md) { 5157 if (m->m_next == nd->nd_md) { 5158 m->m_next = NULL; 5159 m_freem(nd->nd_mrep); 5160 nd->nd_mrep = m = nd->nd_md; 5161 } else 5162 m = m->m_next; 5163 } 5164 if (m == NULL) { 5165 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5166 error = ENOENT; 5167 goto nfsmout; 5168 } 5169 5170 /* 5171 * Now, adjust first mbuf so that any XDR before the 5172 * read data is skipped over. 5173 */ 5174 trimlen = nd->nd_dpos - mtod(m, char *); 5175 if (trimlen > 0) { 5176 m->m_len -= trimlen; 5177 NFSM_DATAP(m, trimlen); 5178 } 5179 5180 /* 5181 * Truncate the mbuf chain at retlen bytes of data, 5182 * plus XDR padding that brings the length up to a 5183 * multiple of 4. 5184 */ 5185 tlen = NFSM_RNDUP(retlen); 5186 do { 5187 if (m->m_len >= tlen) { 5188 m->m_len = tlen; 5189 tlen = 0; 5190 m2 = m->m_next; 5191 m->m_next = NULL; 5192 m_freem(m2); 5193 break; 5194 } 5195 tlen -= m->m_len; 5196 m = m->m_next; 5197 } while (m != NULL); 5198 if (tlen > 0) { 5199 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5200 error = ENOENT; 5201 goto nfsmout; 5202 } 5203 *mpp = nd->nd_mrep; 5204 *mpendp = m; 5205 nd->nd_mrep = NULL; 5206 } 5207 } else 5208 error = nd->nd_repstat; 5209 nfsmout: 5210 /* If nd->nd_mrep is already NULL, this is a no-op. */ 5211 m_freem(nd->nd_mrep); 5212 free(nd, M_TEMP); 5213 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); 5214 return (error); 5215 } 5216 5217 /* 5218 * Do a write RPC on a DS data file, using this structure for the arguments, 5219 * so that this function can be executed by a separate kernel process. 5220 */ 5221 struct nfsrvwritedsdorpc { 5222 int done; 5223 int inprog; 5224 struct task tsk; 5225 fhandle_t fh; 5226 off_t off; 5227 int len; 5228 struct nfsmount *nmp; 5229 struct ucred *cred; 5230 NFSPROC_T *p; 5231 struct mbuf *m; 5232 int err; 5233 }; 5234 5235 static int 5236 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, 5237 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) 5238 { 5239 uint32_t *tl; 5240 struct nfsrv_descript *nd; 5241 nfsattrbit_t attrbits; 5242 nfsv4stateid_t st; 5243 int commit, error, retlen; 5244 5245 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5246 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, 5247 sizeof(fhandle_t), NULL, NULL, 0, 0); 5248 5249 /* 5250 * Use a stateid where other is an alternating 01010 pattern and 5251 * seqid is 0xffffffff. This value is not defined as special by 5252 * the RFC and is used by the FreeBSD NFS server to indicate an 5253 * MDS->DS proxy operation. 5254 */ 5255 st.other[0] = 0x55555555; 5256 st.other[1] = 0x55555555; 5257 st.other[2] = 0x55555555; 5258 st.seqid = 0xffffffff; 5259 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5260 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); 5261 txdr_hyper(off, tl); 5262 tl += 2; 5263 /* 5264 * Do all writes FileSync, since the server doesn't hold onto dirty 5265 * buffers. Since clients should be accessing the DS servers directly 5266 * using the pNFS layouts, this just needs to work correctly as a 5267 * fallback. 5268 */ 5269 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); 5270 *tl = txdr_unsigned(len); 5271 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); 5272 5273 /* Put data in mbuf chain. */ 5274 nd->nd_mb->m_next = m; 5275 5276 /* Set nd_mb and nd_bpos to end of data. */ 5277 while (m->m_next != NULL) 5278 m = m->m_next; 5279 nd->nd_mb = m; 5280 nfsm_set(nd, m->m_len); 5281 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); 5282 5283 /* Do a Getattr for the attributes that change upon writing. */ 5284 NFSZERO_ATTRBIT(&attrbits); 5285 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5286 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5287 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5288 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5289 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5290 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5291 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5292 (void) nfsrv_putattrbit(nd, &attrbits); 5293 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5294 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5295 if (error != 0) { 5296 free(nd, M_TEMP); 5297 return (error); 5298 } 5299 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); 5300 /* Get rid of weak cache consistency data for now. */ 5301 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5302 (ND_NFSV4 | ND_V4WCCATTR)) { 5303 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5304 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5305 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); 5306 if (error != 0) 5307 goto nfsmout; 5308 /* 5309 * Get rid of Op# and status for next op. 5310 */ 5311 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5312 if (*++tl != 0) 5313 nd->nd_flag |= ND_NOMOREDATA; 5314 } 5315 if (nd->nd_repstat == 0) { 5316 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); 5317 retlen = fxdr_unsigned(int, *tl++); 5318 commit = fxdr_unsigned(int, *tl); 5319 if (commit != NFSWRITE_FILESYNC) 5320 error = NFSERR_IO; 5321 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", 5322 retlen, commit, error); 5323 } else 5324 error = nd->nd_repstat; 5325 /* We have no use for the Write Verifier since we use FileSync. */ 5326 5327 /* 5328 * Get the Change, Size, Access Time and Modify Time attributes and set 5329 * on the Metadata file, so its attributes will be what the file's 5330 * would be if it had been written. 5331 */ 5332 if (error == 0) { 5333 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5334 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5335 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5336 } 5337 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); 5338 nfsmout: 5339 m_freem(nd->nd_mrep); 5340 free(nd, M_TEMP); 5341 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); 5342 return (error); 5343 } 5344 5345 /* 5346 * Start up the thread that will execute nfsrv_writedsdorpc(). 5347 */ 5348 static void 5349 start_writedsdorpc(void *arg, int pending) 5350 { 5351 struct nfsrvwritedsdorpc *drpc; 5352 5353 drpc = (struct nfsrvwritedsdorpc *)arg; 5354 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5355 drpc->len, NULL, drpc->m, drpc->cred, drpc->p); 5356 drpc->done = 1; 5357 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); 5358 } 5359 5360 static int 5361 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5362 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5363 struct mbuf **mpp, char *cp, int *failposp) 5364 { 5365 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; 5366 struct nfsvattr na; 5367 struct mbuf *m; 5368 int error, i, offs, ret, timo; 5369 5370 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); 5371 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); 5372 drpc = NULL; 5373 if (mirrorcnt > 1) 5374 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5375 M_WAITOK); 5376 5377 /* Calculate offset in mbuf chain that data starts. */ 5378 offs = cp - mtod(*mpp, char *); 5379 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); 5380 5381 /* 5382 * Do the write RPC for every DS, using a separate kernel process 5383 * for every DS except the last one. 5384 */ 5385 error = 0; 5386 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5387 tdrpc->done = 0; 5388 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5389 tdrpc->off = off; 5390 tdrpc->len = len; 5391 tdrpc->nmp = *nmpp; 5392 tdrpc->cred = cred; 5393 tdrpc->p = p; 5394 tdrpc->inprog = 0; 5395 tdrpc->err = 0; 5396 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5397 ret = EIO; 5398 if (nfs_pnfsiothreads != 0) { 5399 ret = nfs_pnfsio(start_writedsdorpc, tdrpc); 5400 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", 5401 ret); 5402 } 5403 if (ret != 0) { 5404 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, 5405 tdrpc->m, cred, p); 5406 if (nfsds_failerr(ret) && *failposp == -1) 5407 *failposp = i; 5408 else if (error == 0 && ret != 0) 5409 error = ret; 5410 } 5411 nmpp++; 5412 fhp++; 5413 } 5414 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5415 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); 5416 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5417 *failposp = mirrorcnt - 1; 5418 else if (error == 0 && ret != 0) 5419 error = ret; 5420 if (error == 0) 5421 error = nfsrv_setextattr(vp, &na, p); 5422 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); 5423 tdrpc = drpc; 5424 timo = hz / 50; /* Wait for 20msec. */ 5425 if (timo < 1) 5426 timo = 1; 5427 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5428 /* Wait for RPCs on separate threads to complete. */ 5429 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5430 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); 5431 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5432 *failposp = i; 5433 else if (error == 0 && tdrpc->err != 0) 5434 error = tdrpc->err; 5435 } 5436 free(drpc, M_TEMP); 5437 return (error); 5438 } 5439 5440 /* 5441 * Do a allocate RPC on a DS data file, using this structure for the arguments, 5442 * so that this function can be executed by a separate kernel process. 5443 */ 5444 struct nfsrvallocatedsdorpc { 5445 int done; 5446 int inprog; 5447 struct task tsk; 5448 fhandle_t fh; 5449 off_t off; 5450 off_t len; 5451 struct nfsmount *nmp; 5452 struct ucred *cred; 5453 NFSPROC_T *p; 5454 int err; 5455 }; 5456 5457 static int 5458 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5459 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5460 { 5461 uint32_t *tl; 5462 struct nfsrv_descript *nd; 5463 nfsattrbit_t attrbits; 5464 nfsv4stateid_t st; 5465 int error; 5466 5467 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5468 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, 5469 sizeof(fhandle_t), NULL, NULL, 0, 0); 5470 5471 /* 5472 * Use a stateid where other is an alternating 01010 pattern and 5473 * seqid is 0xffffffff. This value is not defined as special by 5474 * the RFC and is used by the FreeBSD NFS server to indicate an 5475 * MDS->DS proxy operation. 5476 */ 5477 st.other[0] = 0x55555555; 5478 st.other[1] = 0x55555555; 5479 st.other[2] = 0x55555555; 5480 st.seqid = 0xffffffff; 5481 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5482 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5483 txdr_hyper(off, tl); tl += 2; 5484 txdr_hyper(len, tl); tl += 2; 5485 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); 5486 5487 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5488 NFSGETATTR_ATTRBIT(&attrbits); 5489 nfsrv_putattrbit(nd, &attrbits); 5490 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5491 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5492 if (error != 0) { 5493 free(nd, M_TEMP); 5494 return (error); 5495 } 5496 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", 5497 nd->nd_repstat); 5498 if (nd->nd_repstat == 0) { 5499 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5500 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5501 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5502 } else 5503 error = nd->nd_repstat; 5504 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); 5505 nfsmout: 5506 m_freem(nd->nd_mrep); 5507 free(nd, M_TEMP); 5508 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); 5509 return (error); 5510 } 5511 5512 /* 5513 * Start up the thread that will execute nfsrv_allocatedsdorpc(). 5514 */ 5515 static void 5516 start_allocatedsdorpc(void *arg, int pending) 5517 { 5518 struct nfsrvallocatedsdorpc *drpc; 5519 5520 drpc = (struct nfsrvallocatedsdorpc *)arg; 5521 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5522 drpc->len, NULL, drpc->cred, drpc->p); 5523 drpc->done = 1; 5524 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); 5525 } 5526 5527 static int 5528 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5529 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5530 int *failposp) 5531 { 5532 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; 5533 struct nfsvattr na; 5534 int error, i, ret, timo; 5535 5536 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); 5537 drpc = NULL; 5538 if (mirrorcnt > 1) 5539 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5540 M_WAITOK); 5541 5542 /* 5543 * Do the allocate RPC for every DS, using a separate kernel process 5544 * for every DS except the last one. 5545 */ 5546 error = 0; 5547 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5548 tdrpc->done = 0; 5549 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5550 tdrpc->off = off; 5551 tdrpc->len = len; 5552 tdrpc->nmp = *nmpp; 5553 tdrpc->cred = cred; 5554 tdrpc->p = p; 5555 tdrpc->inprog = 0; 5556 tdrpc->err = 0; 5557 ret = EIO; 5558 if (nfs_pnfsiothreads != 0) { 5559 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); 5560 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", 5561 ret); 5562 } 5563 if (ret != 0) { 5564 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, 5565 cred, p); 5566 if (nfsds_failerr(ret) && *failposp == -1) 5567 *failposp = i; 5568 else if (error == 0 && ret != 0) 5569 error = ret; 5570 } 5571 nmpp++; 5572 fhp++; 5573 } 5574 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5575 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5576 *failposp = mirrorcnt - 1; 5577 else if (error == 0 && ret != 0) 5578 error = ret; 5579 if (error == 0) 5580 error = nfsrv_setextattr(vp, &na, p); 5581 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); 5582 tdrpc = drpc; 5583 timo = hz / 50; /* Wait for 20msec. */ 5584 if (timo < 1) 5585 timo = 1; 5586 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5587 /* Wait for RPCs on separate threads to complete. */ 5588 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5589 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5590 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5591 *failposp = i; 5592 else if (error == 0 && tdrpc->err != 0) 5593 error = tdrpc->err; 5594 } 5595 free(drpc, M_TEMP); 5596 return (error); 5597 } 5598 5599 static int 5600 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5601 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, 5602 struct nfsvattr *dsnap) 5603 { 5604 uint32_t *tl; 5605 struct nfsrv_descript *nd; 5606 nfsv4stateid_t st; 5607 nfsattrbit_t attrbits; 5608 int error; 5609 5610 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); 5611 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5612 /* 5613 * Use a stateid where other is an alternating 01010 pattern and 5614 * seqid is 0xffffffff. This value is not defined as special by 5615 * the RFC and is used by the FreeBSD NFS server to indicate an 5616 * MDS->DS proxy operation. 5617 */ 5618 st.other[0] = 0x55555555; 5619 st.other[1] = 0x55555555; 5620 st.other[2] = 0x55555555; 5621 st.seqid = 0xffffffff; 5622 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5623 NULL, NULL, 0, 0); 5624 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5625 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); 5626 5627 /* Do a Getattr for the attributes that change due to writing. */ 5628 NFSZERO_ATTRBIT(&attrbits); 5629 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5630 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5631 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5632 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5633 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5634 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5635 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5636 (void) nfsrv_putattrbit(nd, &attrbits); 5637 error =