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 <sys/hash.h> 50 #include <sys/sysctl.h> 51 #include <nlm/nlm_prot.h> 52 #include <nlm/nlm.h> 53 54 FEATURE(nfsd, "NFSv4 server"); 55 56 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 57 extern int nfsrv_useacl; 58 extern int newnfs_numnfsd; 59 extern struct mount nfsv4root_mnt; 60 extern struct nfsrv_stablefirst nfsrv_stablefirst; 61 extern void (*nfsd_call_servertimer)(void); 62 extern SVCPOOL *nfsrvd_pool; 63 extern struct nfsv4lock nfsd_suspend_lock; 64 extern struct nfsclienthashhead *nfsclienthash; 65 extern struct nfslockhashhead *nfslockhash; 66 extern struct nfssessionhash *nfssessionhash; 67 extern int nfsrv_sessionhashsize; 68 extern struct nfsstatsv1 nfsstatsv1; 69 extern struct nfslayouthash *nfslayouthash; 70 extern int nfsrv_layouthashsize; 71 extern struct mtx nfsrv_dslock_mtx; 72 extern int nfs_pnfsiothreads; 73 extern struct nfsdontlisthead nfsrv_dontlisthead; 74 extern volatile int nfsrv_dontlistlen; 75 extern volatile int nfsrv_devidcnt; 76 extern int nfsrv_maxpnfsmirror; 77 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt; 78 NFSDLOCKMUTEX; 79 NFSSTATESPINLOCK; 80 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE]; 81 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE]; 82 struct mtx nfsrc_udpmtx; 83 struct mtx nfs_v4root_mutex; 84 struct mtx nfsrv_dontlistlock_mtx; 85 struct mtx nfsrv_recalllock_mtx; 86 struct nfsrvfh nfs_rootfh, nfs_pubfh; 87 int nfs_pubfhset = 0, nfs_rootfhset = 0; 88 struct proc *nfsd_master_proc = NULL; 89 int nfsd_debuglevel = 0; 90 static pid_t nfsd_master_pid = (pid_t)-1; 91 static char nfsd_master_comm[MAXCOMLEN + 1]; 92 static struct timeval nfsd_master_start; 93 static uint32_t nfsv4_sysid = 0; 94 static fhandle_t zerofh; 95 96 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, 97 struct ucred *); 98 99 int nfsrv_enable_crossmntpt = 1; 100 static int nfs_commit_blks; 101 static int nfs_commit_miss; 102 extern int nfsrv_issuedelegs; 103 extern int nfsrv_dolocallocks; 104 extern int nfsd_enable_stringtouid; 105 extern struct nfsdevicehead nfsrv_devidhead; 106 107 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, 108 NFSPROC_T *); 109 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, 110 int *, char *, fhandle_t *); 111 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, 112 NFSPROC_T *); 113 static int nfsrv_proxyds(struct nfsrv_descript *, struct vnode *, off_t, int, 114 struct ucred *, struct thread *, int, struct mbuf **, char *, 115 struct mbuf **, struct nfsvattr *, struct acl *); 116 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); 117 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, 118 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); 119 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, 120 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, 121 char *, int *); 122 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 123 struct vnode *, struct nfsmount **, int, struct acl *, int *); 124 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 125 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); 126 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 127 struct vnode *, struct nfsmount *, struct nfsvattr *); 128 static int nfsrv_putfhname(fhandle_t *, char *); 129 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, 130 struct pnfsdsfile *, struct vnode **, NFSPROC_T *); 131 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, 132 struct vnode *, NFSPROC_T *); 133 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); 134 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, 135 NFSPROC_T *); 136 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); 137 138 int nfs_pnfsio(task_fn_t *, void *); 139 140 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "NFS server"); 141 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, 142 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); 143 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 144 0, ""); 145 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 146 0, ""); 147 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, 148 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); 149 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, 150 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); 151 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 152 0, "Debug level for NFS server"); 153 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW, 154 &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names"); 155 static int nfsrv_pnfsgetdsattr = 1; 156 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, 157 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); 158 159 /* 160 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are 161 * not running. 162 * The dsN subdirectories for the increased values must have been created 163 * on all DS servers before this increase is done. 164 */ 165 u_int nfsrv_dsdirsize = 20; 166 static int 167 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) 168 { 169 int error, newdsdirsize; 170 171 newdsdirsize = nfsrv_dsdirsize; 172 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); 173 if (error != 0 || req->newptr == NULL) 174 return (error); 175 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || 176 newnfs_numnfsd != 0) 177 return (EINVAL); 178 nfsrv_dsdirsize = newdsdirsize; 179 return (0); 180 } 181 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, CTLTYPE_UINT | CTLFLAG_RW, 0, 182 sizeof(nfsrv_dsdirsize), sysctl_dsdirsize, "IU", 183 "Number of dsN subdirs on the DS servers"); 184 185 #define MAX_REORDERED_RPC 16 186 #define NUM_HEURISTIC 1031 187 #define NHUSE_INIT 64 188 #define NHUSE_INC 16 189 #define NHUSE_MAX 2048 190 191 static struct nfsheur { 192 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 193 off_t nh_nextoff; /* next offset for sequential detection */ 194 int nh_use; /* use count for selection */ 195 int nh_seqcount; /* heuristic */ 196 } nfsheur[NUM_HEURISTIC]; 197 198 199 /* 200 * Heuristic to detect sequential operation. 201 */ 202 static struct nfsheur * 203 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 204 { 205 struct nfsheur *nh; 206 int hi, try; 207 208 /* Locate best candidate. */ 209 try = 32; 210 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 211 nh = &nfsheur[hi]; 212 while (try--) { 213 if (nfsheur[hi].nh_vp == vp) { 214 nh = &nfsheur[hi]; 215 break; 216 } 217 if (nfsheur[hi].nh_use > 0) 218 --nfsheur[hi].nh_use; 219 hi = (hi + 1) % NUM_HEURISTIC; 220 if (nfsheur[hi].nh_use < nh->nh_use) 221 nh = &nfsheur[hi]; 222 } 223 224 /* Initialize hint if this is a new file. */ 225 if (nh->nh_vp != vp) { 226 nh->nh_vp = vp; 227 nh->nh_nextoff = uio->uio_offset; 228 nh->nh_use = NHUSE_INIT; 229 if (uio->uio_offset == 0) 230 nh->nh_seqcount = 4; 231 else 232 nh->nh_seqcount = 1; 233 } 234 235 /* Calculate heuristic. */ 236 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 237 uio->uio_offset == nh->nh_nextoff) { 238 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 239 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 240 if (nh->nh_seqcount > IO_SEQMAX) 241 nh->nh_seqcount = IO_SEQMAX; 242 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 243 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 244 /* Probably a reordered RPC, leave seqcount alone. */ 245 } else if (nh->nh_seqcount > 1) { 246 nh->nh_seqcount /= 2; 247 } else { 248 nh->nh_seqcount = 0; 249 } 250 nh->nh_use += NHUSE_INC; 251 if (nh->nh_use > NHUSE_MAX) 252 nh->nh_use = NHUSE_MAX; 253 return (nh); 254 } 255 256 /* 257 * Get attributes into nfsvattr structure. 258 */ 259 int 260 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, 261 struct nfsrv_descript *nd, struct thread *p, int vpislocked, 262 nfsattrbit_t *attrbitp) 263 { 264 int error, gotattr, lockedit = 0; 265 struct nfsvattr na; 266 267 if (vpislocked == 0) { 268 /* 269 * When vpislocked == 0, the vnode is either exclusively 270 * locked by this thread or not locked by this thread. 271 * As such, shared lock it, if not exclusively locked. 272 */ 273 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 274 lockedit = 1; 275 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); 276 } 277 } 278 279 /* 280 * Acquire the Change, Size and TimeModify attributes, as required. 281 * This needs to be done for regular files if: 282 * - non-NFSv4 RPCs or 283 * - when attrbitp == NULL or 284 * - an NFSv4 RPC with any of the above attributes in attrbitp. 285 * A return of 0 for nfsrv_proxyds() indicates that it has acquired 286 * these attributes. nfsrv_proxyds() will return an error if the 287 * server is not a pNFS one. 288 */ 289 gotattr = 0; 290 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || 291 (nd->nd_flag & ND_NFSV4) == 0 || 292 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || 293 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || 294 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || 295 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY))) { 296 error = nfsrv_proxyds(nd, vp, 0, 0, nd->nd_cred, p, 297 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL); 298 if (error == 0) 299 gotattr = 1; 300 } 301 302 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); 303 if (lockedit != 0) 304 NFSVOPUNLOCK(vp, 0); 305 306 /* 307 * If we got the Change, Size and Modify Time from the DS, 308 * replace them. 309 */ 310 if (gotattr != 0) { 311 nvap->na_atime = na.na_atime; 312 nvap->na_mtime = na.na_mtime; 313 nvap->na_filerev = na.na_filerev; 314 nvap->na_size = na.na_size; 315 } 316 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, 317 error, (uintmax_t)na.na_filerev); 318 319 NFSEXITCODE(error); 320 return (error); 321 } 322 323 /* 324 * Get a file handle for a vnode. 325 */ 326 int 327 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) 328 { 329 int error; 330 331 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); 332 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 333 error = VOP_VPTOFH(vp, &fhp->fh_fid); 334 335 NFSEXITCODE(error); 336 return (error); 337 } 338 339 /* 340 * Perform access checking for vnodes obtained from file handles that would 341 * refer to files already opened by a Unix client. You cannot just use 342 * vn_writechk() and VOP_ACCESSX() for two reasons. 343 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 344 * case. 345 * 2 - The owner is to be given access irrespective of mode bits for some 346 * operations, so that processes that chmod after opening a file don't 347 * break. 348 */ 349 int 350 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, 351 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, 352 u_int32_t *supportedtypep) 353 { 354 struct vattr vattr; 355 int error = 0, getret = 0; 356 357 if (vpislocked == 0) { 358 if (NFSVOPLOCK(vp, LK_SHARED) != 0) { 359 error = EPERM; 360 goto out; 361 } 362 } 363 if (accmode & VWRITE) { 364 /* Just vn_writechk() changed to check rdonly */ 365 /* 366 * Disallow write attempts on read-only file systems; 367 * unless the file is a socket or a block or character 368 * device resident on the file system. 369 */ 370 if (NFSVNO_EXRDONLY(exp) || 371 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 372 switch (vp->v_type) { 373 case VREG: 374 case VDIR: 375 case VLNK: 376 error = EROFS; 377 default: 378 break; 379 } 380 } 381 /* 382 * If there's shared text associated with 383 * the inode, try to free it up once. If 384 * we fail, we can't allow writing. 385 */ 386 if (VOP_IS_TEXT(vp) && error == 0) 387 error = ETXTBSY; 388 } 389 if (error != 0) { 390 if (vpislocked == 0) 391 NFSVOPUNLOCK(vp, 0); 392 goto out; 393 } 394 395 /* 396 * Should the override still be applied when ACLs are enabled? 397 */ 398 error = VOP_ACCESSX(vp, accmode, cred, p); 399 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { 400 /* 401 * Try again with VEXPLICIT_DENY, to see if the test for 402 * deletion is supported. 403 */ 404 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); 405 if (error == 0) { 406 if (vp->v_type == VDIR) { 407 accmode &= ~(VDELETE | VDELETE_CHILD); 408 accmode |= VWRITE; 409 error = VOP_ACCESSX(vp, accmode, cred, p); 410 } else if (supportedtypep != NULL) { 411 *supportedtypep &= ~NFSACCESS_DELETE; 412 } 413 } 414 } 415 416 /* 417 * Allow certain operations for the owner (reads and writes 418 * on files that are already open). 419 */ 420 if (override != NFSACCCHK_NOOVERRIDE && 421 (error == EPERM || error == EACCES)) { 422 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) 423 error = 0; 424 else if (override & NFSACCCHK_ALLOWOWNER) { 425 getret = VOP_GETATTR(vp, &vattr, cred); 426 if (getret == 0 && cred->cr_uid == vattr.va_uid) 427 error = 0; 428 } 429 } 430 if (vpislocked == 0) 431 NFSVOPUNLOCK(vp, 0); 432 433 out: 434 NFSEXITCODE(error); 435 return (error); 436 } 437 438 /* 439 * Set attribute(s) vnop. 440 */ 441 int 442 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 443 struct thread *p, struct nfsexstuff *exp) 444 { 445 u_quad_t savsize = 0; 446 int error, savedit; 447 448 /* 449 * If this is an exported file system and a pNFS service is running, 450 * don't VOP_SETATTR() of size for the MDS file system. 451 */ 452 savedit = 0; 453 error = 0; 454 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && 455 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && 456 nvap->na_vattr.va_size > 0) { 457 savsize = nvap->na_vattr.va_size; 458 nvap->na_vattr.va_size = VNOVAL; 459 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 460 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 461 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 462 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 463 nvap->na_vattr.va_mtime.tv_sec != VNOVAL) 464 savedit = 1; 465 else 466 savedit = 2; 467 } 468 if (savedit != 2) 469 error = VOP_SETATTR(vp, &nvap->na_vattr, cred); 470 if (savedit != 0) 471 nvap->na_vattr.va_size = savsize; 472 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 473 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 474 nvap->na_vattr.va_size != VNOVAL || 475 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 476 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 477 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { 478 /* For a pNFS server, set the attributes on the DS file. */ 479 error = nfsrv_proxyds(NULL, vp, 0, 0, cred, p, NFSPROC_SETATTR, 480 NULL, NULL, NULL, nvap, NULL); 481 if (error == ENOENT) 482 error = 0; 483 } 484 NFSEXITCODE(error); 485 return (error); 486 } 487 488 /* 489 * Set up nameidata for a lookup() call and do it. 490 */ 491 int 492 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, 493 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p, 494 struct vnode **retdirp) 495 { 496 struct componentname *cnp = &ndp->ni_cnd; 497 int i; 498 struct iovec aiov; 499 struct uio auio; 500 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; 501 int error = 0; 502 char *cp; 503 504 *retdirp = NULL; 505 cnp->cn_nameptr = cnp->cn_pnbuf; 506 ndp->ni_lcf = 0; 507 /* 508 * Extract and set starting directory. 509 */ 510 if (dp->v_type != VDIR) { 511 if (islocked) 512 vput(dp); 513 else 514 vrele(dp); 515 nfsvno_relpathbuf(ndp); 516 error = ENOTDIR; 517 goto out1; 518 } 519 if (islocked) 520 NFSVOPUNLOCK(dp, 0); 521 VREF(dp); 522 *retdirp = dp; 523 if (NFSVNO_EXRDONLY(exp)) 524 cnp->cn_flags |= RDONLY; 525 ndp->ni_segflg = UIO_SYSSPACE; 526 527 if (nd->nd_flag & ND_PUBLOOKUP) { 528 ndp->ni_loopcnt = 0; 529 if (cnp->cn_pnbuf[0] == '/') { 530 vrele(dp); 531 /* 532 * Check for degenerate pathnames here, since lookup() 533 * panics on them. 534 */ 535 for (i = 1; i < ndp->ni_pathlen; i++) 536 if (cnp->cn_pnbuf[i] != '/') 537 break; 538 if (i == ndp->ni_pathlen) { 539 error = NFSERR_ACCES; 540 goto out; 541 } 542 dp = rootvnode; 543 VREF(dp); 544 } 545 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || 546 (nd->nd_flag & ND_NFSV4) == 0) { 547 /* 548 * Only cross mount points for NFSv4 when doing a 549 * mount while traversing the file system above 550 * the mount point, unless nfsrv_enable_crossmntpt is set. 551 */ 552 cnp->cn_flags |= NOCROSSMOUNT; 553 } 554 555 /* 556 * Initialize for scan, set ni_startdir and bump ref on dp again 557 * because lookup() will dereference ni_startdir. 558 */ 559 560 cnp->cn_thread = p; 561 ndp->ni_startdir = dp; 562 ndp->ni_rootdir = rootvnode; 563 ndp->ni_topdir = NULL; 564 565 if (!lockleaf) 566 cnp->cn_flags |= LOCKLEAF; 567 for (;;) { 568 cnp->cn_nameptr = cnp->cn_pnbuf; 569 /* 570 * Call lookup() to do the real work. If an error occurs, 571 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and 572 * we do not have to dereference anything before returning. 573 * In either case ni_startdir will be dereferenced and NULLed 574 * out. 575 */ 576 error = lookup(ndp); 577 if (error) 578 break; 579 580 /* 581 * Check for encountering a symbolic link. Trivial 582 * termination occurs if no symlink encountered. 583 */ 584 if ((cnp->cn_flags & ISSYMLINK) == 0) { 585 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) 586 nfsvno_relpathbuf(ndp); 587 if (ndp->ni_vp && !lockleaf) 588 NFSVOPUNLOCK(ndp->ni_vp, 0); 589 break; 590 } 591 592 /* 593 * Validate symlink 594 */ 595 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) 596 NFSVOPUNLOCK(ndp->ni_dvp, 0); 597 if (!(nd->nd_flag & ND_PUBLOOKUP)) { 598 error = EINVAL; 599 goto badlink2; 600 } 601 602 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 603 error = ELOOP; 604 goto badlink2; 605 } 606 if (ndp->ni_pathlen > 1) 607 cp = uma_zalloc(namei_zone, M_WAITOK); 608 else 609 cp = cnp->cn_pnbuf; 610 aiov.iov_base = cp; 611 aiov.iov_len = MAXPATHLEN; 612 auio.uio_iov = &aiov; 613 auio.uio_iovcnt = 1; 614 auio.uio_offset = 0; 615 auio.uio_rw = UIO_READ; 616 auio.uio_segflg = UIO_SYSSPACE; 617 auio.uio_td = NULL; 618 auio.uio_resid = MAXPATHLEN; 619 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 620 if (error) { 621 badlink1: 622 if (ndp->ni_pathlen > 1) 623 uma_zfree(namei_zone, cp); 624 badlink2: 625 vrele(ndp->ni_dvp); 626 vput(ndp->ni_vp); 627 break; 628 } 629 linklen = MAXPATHLEN - auio.uio_resid; 630 if (linklen == 0) { 631 error = ENOENT; 632 goto badlink1; 633 } 634 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 635 error = ENAMETOOLONG; 636 goto badlink1; 637 } 638 639 /* 640 * Adjust or replace path 641 */ 642 if (ndp->ni_pathlen > 1) { 643 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 644 uma_zfree(namei_zone, cnp->cn_pnbuf); 645 cnp->cn_pnbuf = cp; 646 } else 647 cnp->cn_pnbuf[linklen] = '\0'; 648 ndp->ni_pathlen += linklen; 649 650 /* 651 * Cleanup refs for next loop and check if root directory 652 * should replace current directory. Normally ni_dvp 653 * becomes the new base directory and is cleaned up when 654 * we loop. Explicitly null pointers after invalidation 655 * to clarify operation. 656 */ 657 vput(ndp->ni_vp); 658 ndp->ni_vp = NULL; 659 660 if (cnp->cn_pnbuf[0] == '/') { 661 vrele(ndp->ni_dvp); 662 ndp->ni_dvp = ndp->ni_rootdir; 663 VREF(ndp->ni_dvp); 664 } 665 ndp->ni_startdir = ndp->ni_dvp; 666 ndp->ni_dvp = NULL; 667 } 668 if (!lockleaf) 669 cnp->cn_flags &= ~LOCKLEAF; 670 671 out: 672 if (error) { 673 nfsvno_relpathbuf(ndp); 674 ndp->ni_vp = NULL; 675 ndp->ni_dvp = NULL; 676 ndp->ni_startdir = NULL; 677 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { 678 ndp->ni_dvp = NULL; 679 } 680 681 out1: 682 NFSEXITCODE2(error, nd); 683 return (error); 684 } 685 686 /* 687 * Set up a pathname buffer and return a pointer to it and, optionally 688 * set a hash pointer. 689 */ 690 void 691 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) 692 { 693 struct componentname *cnp = &ndp->ni_cnd; 694 695 cnp->cn_flags |= (NOMACCHECK | HASBUF); 696 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 697 if (hashpp != NULL) 698 *hashpp = NULL; 699 *bufpp = cnp->cn_pnbuf; 700 } 701 702 /* 703 * Release the above path buffer, if not released by nfsvno_namei(). 704 */ 705 void 706 nfsvno_relpathbuf(struct nameidata *ndp) 707 { 708 709 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0) 710 panic("nfsrelpath"); 711 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 712 ndp->ni_cnd.cn_flags &= ~HASBUF; 713 } 714 715 /* 716 * Readlink vnode op into an mbuf list. 717 */ 718 int 719 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p, 720 struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 721 { 722 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 723 struct iovec *ivp = iv; 724 struct uio io, *uiop = &io; 725 struct mbuf *mp, *mp2 = NULL, *mp3 = NULL; 726 int i, len, tlen, error = 0; 727 728 len = 0; 729 i = 0; 730 while (len < NFS_MAXPATHLEN) { 731 NFSMGET(mp); 732 MCLGET(mp, M_WAITOK); 733 mp->m_len = M_SIZE(mp); 734 if (len == 0) { 735 mp3 = mp2 = mp; 736 } else { 737 mp2->m_next = mp; 738 mp2 = mp; 739 } 740 if ((len + mp->m_len) > NFS_MAXPATHLEN) { 741 mp->m_len = NFS_MAXPATHLEN - len; 742 len = NFS_MAXPATHLEN; 743 } else { 744 len += mp->m_len; 745 } 746 ivp->iov_base = mtod(mp, caddr_t); 747 ivp->iov_len = mp->m_len; 748 i++; 749 ivp++; 750 } 751 uiop->uio_iov = iv; 752 uiop->uio_iovcnt = i; 753 uiop->uio_offset = 0; 754 uiop->uio_resid = len; 755 uiop->uio_rw = UIO_READ; 756 uiop->uio_segflg = UIO_SYSSPACE; 757 uiop->uio_td = NULL; 758 error = VOP_READLINK(vp, uiop, cred); 759 if (error) { 760 m_freem(mp3); 761 *lenp = 0; 762 goto out; 763 } 764 if (uiop->uio_resid > 0) { 765 len -= uiop->uio_resid; 766 tlen = NFSM_RNDUP(len); 767 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len); 768 } 769 *lenp = len; 770 *mpp = mp3; 771 *mpendp = mp; 772 773 out: 774 NFSEXITCODE(error); 775 return (error); 776 } 777 778 /* 779 * Read vnode op call into mbuf list. 780 */ 781 int 782 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 783 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp) 784 { 785 struct mbuf *m; 786 int i; 787 struct iovec *iv; 788 struct iovec *iv2; 789 int error = 0, len, left, siz, tlen, ioflag = 0; 790 struct mbuf *m2 = NULL, *m3; 791 struct uio io, *uiop = &io; 792 struct nfsheur *nh; 793 794 /* 795 * Attempt to read from a DS file. A return of ENOENT implies 796 * there is no DS file to read. 797 */ 798 error = nfsrv_proxyds(NULL, vp, off, cnt, cred, p, NFSPROC_READDS, mpp, 799 NULL, mpendp, NULL, NULL); 800 if (error != ENOENT) 801 return (error); 802 803 len = left = NFSM_RNDUP(cnt); 804 m3 = NULL; 805 /* 806 * Generate the mbuf list with the uio_iov ref. to it. 807 */ 808 i = 0; 809 while (left > 0) { 810 NFSMGET(m); 811 MCLGET(m, M_WAITOK); 812 m->m_len = 0; 813 siz = min(M_TRAILINGSPACE(m), left); 814 left -= siz; 815 i++; 816 if (m3) 817 m2->m_next = m; 818 else 819 m3 = m; 820 m2 = m; 821 } 822 iv = malloc(i * sizeof (struct iovec), 823 M_TEMP, M_WAITOK); 824 uiop->uio_iov = iv2 = iv; 825 m = m3; 826 left = len; 827 i = 0; 828 while (left > 0) { 829 if (m == NULL) 830 panic("nfsvno_read iov"); 831 siz = min(M_TRAILINGSPACE(m), left); 832 if (siz > 0) { 833 iv->iov_base = mtod(m, caddr_t) + m->m_len; 834 iv->iov_len = siz; 835 m->m_len += siz; 836 left -= siz; 837 iv++; 838 i++; 839 } 840 m = m->m_next; 841 } 842 uiop->uio_iovcnt = i; 843 uiop->uio_offset = off; 844 uiop->uio_resid = len; 845 uiop->uio_rw = UIO_READ; 846 uiop->uio_segflg = UIO_SYSSPACE; 847 uiop->uio_td = NULL; 848 nh = nfsrv_sequential_heuristic(uiop, vp); 849 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 850 /* XXX KDM make this more systematic? */ 851 nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid; 852 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 853 free(iv2, M_TEMP); 854 if (error) { 855 m_freem(m3); 856 *mpp = NULL; 857 goto out; 858 } 859 nh->nh_nextoff = uiop->uio_offset; 860 tlen = len - uiop->uio_resid; 861 cnt = cnt < tlen ? cnt : tlen; 862 tlen = NFSM_RNDUP(cnt); 863 if (tlen == 0) { 864 m_freem(m3); 865 m3 = NULL; 866 } else if (len != tlen || tlen != cnt) 867 nfsrv_adj(m3, len - tlen, tlen - cnt); 868 *mpp = m3; 869 *mpendp = m2; 870 871 out: 872 NFSEXITCODE(error); 873 return (error); 874 } 875 876 /* 877 * Write vnode op from an mbuf list. 878 */ 879 int 880 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int *stable, 881 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) 882 { 883 struct iovec *ivp; 884 int i, len; 885 struct iovec *iv; 886 int ioflags, error; 887 struct uio io, *uiop = &io; 888 struct nfsheur *nh; 889 890 /* 891 * Attempt to write to a DS file. A return of ENOENT implies 892 * there is no DS file to write. 893 */ 894 error = nfsrv_proxyds(NULL, vp, off, retlen, cred, p, NFSPROC_WRITEDS, 895 &mp, cp, NULL, NULL, NULL); 896 if (error != ENOENT) { 897 *stable = NFSWRITE_FILESYNC; 898 return (error); 899 } 900 901 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, 902 M_WAITOK); 903 uiop->uio_iov = iv = ivp; 904 uiop->uio_iovcnt = cnt; 905 i = mtod(mp, caddr_t) + mp->m_len - cp; 906 len = retlen; 907 while (len > 0) { 908 if (mp == NULL) 909 panic("nfsvno_write"); 910 if (i > 0) { 911 i = min(i, len); 912 ivp->iov_base = cp; 913 ivp->iov_len = i; 914 ivp++; 915 len -= i; 916 } 917 mp = mp->m_next; 918 if (mp) { 919 i = mp->m_len; 920 cp = mtod(mp, caddr_t); 921 } 922 } 923 924 if (*stable == NFSWRITE_UNSTABLE) 925 ioflags = IO_NODELOCKED; 926 else 927 ioflags = (IO_SYNC | IO_NODELOCKED); 928 uiop->uio_resid = retlen; 929 uiop->uio_rw = UIO_WRITE; 930 uiop->uio_segflg = UIO_SYSSPACE; 931 NFSUIOPROC(uiop, p); 932 uiop->uio_offset = off; 933 nh = nfsrv_sequential_heuristic(uiop, vp); 934 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 935 /* XXX KDM make this more systematic? */ 936 nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; 937 error = VOP_WRITE(vp, uiop, ioflags, cred); 938 if (error == 0) 939 nh->nh_nextoff = uiop->uio_offset; 940 free(iv, M_TEMP); 941 942 NFSEXITCODE(error); 943 return (error); 944 } 945 946 /* 947 * Common code for creating a regular file (plus special files for V2). 948 */ 949 int 950 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, 951 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, 952 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) 953 { 954 u_quad_t tempsize; 955 int error; 956 struct thread *p = curthread; 957 958 error = nd->nd_repstat; 959 if (!error && ndp->ni_vp == NULL) { 960 if (nvap->na_type == VREG || nvap->na_type == VSOCK) { 961 vrele(ndp->ni_startdir); 962 error = VOP_CREATE(ndp->ni_dvp, 963 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 964 /* For a pNFS server, create the data file on a DS. */ 965 if (error == 0 && nvap->na_type == VREG) { 966 /* 967 * Create a data file on a DS for a pNFS server. 968 * This function just returns if not 969 * running a pNFS DS or the creation fails. 970 */ 971 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 972 nd->nd_cred, p); 973 } 974 vput(ndp->ni_dvp); 975 nfsvno_relpathbuf(ndp); 976 if (!error) { 977 if (*exclusive_flagp) { 978 *exclusive_flagp = 0; 979 NFSVNO_ATTRINIT(nvap); 980 nvap->na_atime.tv_sec = cverf[0]; 981 nvap->na_atime.tv_nsec = cverf[1]; 982 error = VOP_SETATTR(ndp->ni_vp, 983 &nvap->na_vattr, nd->nd_cred); 984 if (error != 0) { 985 vput(ndp->ni_vp); 986 ndp->ni_vp = NULL; 987 error = NFSERR_NOTSUPP; 988 } 989 } 990 } 991 /* 992 * NFS V2 Only. nfsrvd_mknod() does this for V3. 993 * (This implies, just get out on an error.) 994 */ 995 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || 996 nvap->na_type == VFIFO) { 997 if (nvap->na_type == VCHR && rdev == 0xffffffff) 998 nvap->na_type = VFIFO; 999 if (nvap->na_type != VFIFO && 1000 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { 1001 vrele(ndp->ni_startdir); 1002 nfsvno_relpathbuf(ndp); 1003 vput(ndp->ni_dvp); 1004 goto out; 1005 } 1006 nvap->na_rdev = rdev; 1007 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1008 &ndp->ni_cnd, &nvap->na_vattr); 1009 vput(ndp->ni_dvp); 1010 nfsvno_relpathbuf(ndp); 1011 vrele(ndp->ni_startdir); 1012 if (error) 1013 goto out; 1014 } else { 1015 vrele(ndp->ni_startdir); 1016 nfsvno_relpathbuf(ndp); 1017 vput(ndp->ni_dvp); 1018 error = ENXIO; 1019 goto out; 1020 } 1021 *vpp = ndp->ni_vp; 1022 } else { 1023 /* 1024 * Handle cases where error is already set and/or 1025 * the file exists. 1026 * 1 - clean up the lookup 1027 * 2 - iff !error and na_size set, truncate it 1028 */ 1029 vrele(ndp->ni_startdir); 1030 nfsvno_relpathbuf(ndp); 1031 *vpp = ndp->ni_vp; 1032 if (ndp->ni_dvp == *vpp) 1033 vrele(ndp->ni_dvp); 1034 else 1035 vput(ndp->ni_dvp); 1036 if (!error && nvap->na_size != VNOVAL) { 1037 error = nfsvno_accchk(*vpp, VWRITE, 1038 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1039 NFSACCCHK_VPISLOCKED, NULL); 1040 if (!error) { 1041 tempsize = nvap->na_size; 1042 NFSVNO_ATTRINIT(nvap); 1043 nvap->na_size = tempsize; 1044 error = VOP_SETATTR(*vpp, 1045 &nvap->na_vattr, nd->nd_cred); 1046 } 1047 } 1048 if (error) 1049 vput(*vpp); 1050 } 1051 1052 out: 1053 NFSEXITCODE(error); 1054 return (error); 1055 } 1056 1057 /* 1058 * Do a mknod vnode op. 1059 */ 1060 int 1061 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, 1062 struct thread *p) 1063 { 1064 int error = 0; 1065 enum vtype vtyp; 1066 1067 vtyp = nvap->na_type; 1068 /* 1069 * Iff doesn't exist, create it. 1070 */ 1071 if (ndp->ni_vp) { 1072 vrele(ndp->ni_startdir); 1073 nfsvno_relpathbuf(ndp); 1074 vput(ndp->ni_dvp); 1075 vrele(ndp->ni_vp); 1076 error = EEXIST; 1077 goto out; 1078 } 1079 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1080 vrele(ndp->ni_startdir); 1081 nfsvno_relpathbuf(ndp); 1082 vput(ndp->ni_dvp); 1083 error = NFSERR_BADTYPE; 1084 goto out; 1085 } 1086 if (vtyp == VSOCK) { 1087 vrele(ndp->ni_startdir); 1088 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 1089 &ndp->ni_cnd, &nvap->na_vattr); 1090 vput(ndp->ni_dvp); 1091 nfsvno_relpathbuf(ndp); 1092 } else { 1093 if (nvap->na_type != VFIFO && 1094 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { 1095 vrele(ndp->ni_startdir); 1096 nfsvno_relpathbuf(ndp); 1097 vput(ndp->ni_dvp); 1098 goto out; 1099 } 1100 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1101 &ndp->ni_cnd, &nvap->na_vattr); 1102 vput(ndp->ni_dvp); 1103 nfsvno_relpathbuf(ndp); 1104 vrele(ndp->ni_startdir); 1105 /* 1106 * Since VOP_MKNOD returns the ni_vp, I can't 1107 * see any reason to do the lookup. 1108 */ 1109 } 1110 1111 out: 1112 NFSEXITCODE(error); 1113 return (error); 1114 } 1115 1116 /* 1117 * Mkdir vnode op. 1118 */ 1119 int 1120 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, 1121 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 1122 { 1123 int error = 0; 1124 1125 if (ndp->ni_vp != NULL) { 1126 if (ndp->ni_dvp == ndp->ni_vp) 1127 vrele(ndp->ni_dvp); 1128 else 1129 vput(ndp->ni_dvp); 1130 vrele(ndp->ni_vp); 1131 nfsvno_relpathbuf(ndp); 1132 error = EEXIST; 1133 goto out; 1134 } 1135 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1136 &nvap->na_vattr); 1137 vput(ndp->ni_dvp); 1138 nfsvno_relpathbuf(ndp); 1139 1140 out: 1141 NFSEXITCODE(error); 1142 return (error); 1143 } 1144 1145 /* 1146 * symlink vnode op. 1147 */ 1148 int 1149 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, 1150 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, 1151 struct nfsexstuff *exp) 1152 { 1153 int error = 0; 1154 1155 if (ndp->ni_vp) { 1156 vrele(ndp->ni_startdir); 1157 nfsvno_relpathbuf(ndp); 1158 if (ndp->ni_dvp == ndp->ni_vp) 1159 vrele(ndp->ni_dvp); 1160 else 1161 vput(ndp->ni_dvp); 1162 vrele(ndp->ni_vp); 1163 error = EEXIST; 1164 goto out; 1165 } 1166 1167 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1168 &nvap->na_vattr, pathcp); 1169 vput(ndp->ni_dvp); 1170 vrele(ndp->ni_startdir); 1171 nfsvno_relpathbuf(ndp); 1172 /* 1173 * Although FreeBSD still had the lookup code in 1174 * it for 7/current, there doesn't seem to be any 1175 * point, since VOP_SYMLINK() returns the ni_vp. 1176 * Just vput it for v2. 1177 */ 1178 if (!not_v2 && !error) 1179 vput(ndp->ni_vp); 1180 1181 out: 1182 NFSEXITCODE(error); 1183 return (error); 1184 } 1185 1186 /* 1187 * Parse symbolic link arguments. 1188 * This function has an ugly side effect. It will malloc() an area for 1189 * the symlink and set iov_base to point to it, only if it succeeds. 1190 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must 1191 * be FREE'd later. 1192 */ 1193 int 1194 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, 1195 struct thread *p, char **pathcpp, int *lenp) 1196 { 1197 u_int32_t *tl; 1198 char *pathcp = NULL; 1199 int error = 0, len; 1200 struct nfsv2_sattr *sp; 1201 1202 *pathcpp = NULL; 1203 *lenp = 0; 1204 if ((nd->nd_flag & ND_NFSV3) && 1205 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) 1206 goto nfsmout; 1207 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 1208 len = fxdr_unsigned(int, *tl); 1209 if (len > NFS_MAXPATHLEN || len <= 0) { 1210 error = EBADRPC; 1211 goto nfsmout; 1212 } 1213 pathcp = malloc(len + 1, M_TEMP, M_WAITOK); 1214 error = nfsrv_mtostr(nd, pathcp, len); 1215 if (error) 1216 goto nfsmout; 1217 if (nd->nd_flag & ND_NFSV2) { 1218 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1219 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 1220 } 1221 *pathcpp = pathcp; 1222 *lenp = len; 1223 NFSEXITCODE2(0, nd); 1224 return (0); 1225 nfsmout: 1226 if (pathcp) 1227 free(pathcp, M_TEMP); 1228 NFSEXITCODE2(error, nd); 1229 return (error); 1230 } 1231 1232 /* 1233 * Remove a non-directory object. 1234 */ 1235 int 1236 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1237 struct thread *p, struct nfsexstuff *exp) 1238 { 1239 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS]; 1240 int error = 0, mirrorcnt; 1241 char fname[PNFS_FILENAME_LEN + 1]; 1242 fhandle_t fh; 1243 1244 vp = ndp->ni_vp; 1245 dsdvp[0] = NULL; 1246 if (vp->v_type == VDIR) 1247 error = NFSERR_ISDIR; 1248 else if (is_v4) 1249 error = nfsrv_checkremove(vp, 1, p); 1250 if (error == 0) 1251 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); 1252 if (!error) 1253 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); 1254 if (error == 0 && dsdvp[0] != NULL) 1255 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1256 if (ndp->ni_dvp == vp) 1257 vrele(ndp->ni_dvp); 1258 else 1259 vput(ndp->ni_dvp); 1260 vput(vp); 1261 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) 1262 nfsvno_relpathbuf(ndp); 1263 NFSEXITCODE(error); 1264 return (error); 1265 } 1266 1267 /* 1268 * Remove a directory. 1269 */ 1270 int 1271 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1272 struct thread *p, struct nfsexstuff *exp) 1273 { 1274 struct vnode *vp; 1275 int error = 0; 1276 1277 vp = ndp->ni_vp; 1278 if (vp->v_type != VDIR) { 1279 error = ENOTDIR; 1280 goto out; 1281 } 1282 /* 1283 * No rmdir "." please. 1284 */ 1285 if (ndp->ni_dvp == vp) { 1286 error = EINVAL; 1287 goto out; 1288 } 1289 /* 1290 * The root of a mounted filesystem cannot be deleted. 1291 */ 1292 if (vp->v_vflag & VV_ROOT) 1293 error = EBUSY; 1294 out: 1295 if (!error) 1296 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); 1297 if (ndp->ni_dvp == vp) 1298 vrele(ndp->ni_dvp); 1299 else 1300 vput(ndp->ni_dvp); 1301 vput(vp); 1302 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) 1303 nfsvno_relpathbuf(ndp); 1304 NFSEXITCODE(error); 1305 return (error); 1306 } 1307 1308 /* 1309 * Rename vnode op. 1310 */ 1311 int 1312 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, 1313 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) 1314 { 1315 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS]; 1316 int error = 0, mirrorcnt; 1317 char fname[PNFS_FILENAME_LEN + 1]; 1318 fhandle_t fh; 1319 1320 dsdvp[0] = NULL; 1321 fvp = fromndp->ni_vp; 1322 if (ndstat) { 1323 vrele(fromndp->ni_dvp); 1324 vrele(fvp); 1325 error = ndstat; 1326 goto out1; 1327 } 1328 tdvp = tondp->ni_dvp; 1329 tvp = tondp->ni_vp; 1330 if (tvp != NULL) { 1331 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1332 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; 1333 goto out; 1334 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1335 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; 1336 goto out; 1337 } 1338 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1339 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1340 goto out; 1341 } 1342 1343 /* 1344 * A rename to '.' or '..' results in a prematurely 1345 * unlocked vnode on FreeBSD5, so I'm just going to fail that 1346 * here. 1347 */ 1348 if ((tondp->ni_cnd.cn_namelen == 1 && 1349 tondp->ni_cnd.cn_nameptr[0] == '.') || 1350 (tondp->ni_cnd.cn_namelen == 2 && 1351 tondp->ni_cnd.cn_nameptr[0] == '.' && 1352 tondp->ni_cnd.cn_nameptr[1] == '.')) { 1353 error = EINVAL; 1354 goto out; 1355 } 1356 } 1357 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1358 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1359 goto out; 1360 } 1361 if (fvp->v_mount != tdvp->v_mount) { 1362 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1363 goto out; 1364 } 1365 if (fvp == tdvp) { 1366 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; 1367 goto out; 1368 } 1369 if (fvp == tvp) { 1370 /* 1371 * If source and destination are the same, there is nothing to 1372 * do. Set error to -1 to indicate this. 1373 */ 1374 error = -1; 1375 goto out; 1376 } 1377 if (ndflag & ND_NFSV4) { 1378 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { 1379 error = nfsrv_checkremove(fvp, 0, p); 1380 NFSVOPUNLOCK(fvp, 0); 1381 } else 1382 error = EPERM; 1383 if (tvp && !error) 1384 error = nfsrv_checkremove(tvp, 1, p); 1385 } else { 1386 /* 1387 * For NFSv2 and NFSv3, try to get rid of the delegation, so 1388 * that the NFSv4 client won't be confused by the rename. 1389 * Since nfsd_recalldelegation() can only be called on an 1390 * unlocked vnode at this point and fvp is the file that will 1391 * still exist after the rename, just do fvp. 1392 */ 1393 nfsd_recalldelegation(fvp, p); 1394 } 1395 if (error == 0 && tvp != NULL) { 1396 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh); 1397 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" 1398 " dsdvp=%p\n", dsdvp[0]); 1399 } 1400 out: 1401 if (!error) { 1402 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, 1403 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, 1404 &tondp->ni_cnd); 1405 } else { 1406 if (tdvp == tvp) 1407 vrele(tdvp); 1408 else 1409 vput(tdvp); 1410 if (tvp) 1411 vput(tvp); 1412 vrele(fromndp->ni_dvp); 1413 vrele(fvp); 1414 if (error == -1) 1415 error = 0; 1416 } 1417 1418 /* 1419 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and 1420 * if the rename succeeded, the DS file for the tvp needs to be 1421 * removed. 1422 */ 1423 if (error == 0 && dsdvp[0] != NULL) { 1424 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1425 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); 1426 } 1427 1428 vrele(tondp->ni_startdir); 1429 nfsvno_relpathbuf(tondp); 1430 out1: 1431 vrele(fromndp->ni_startdir); 1432 nfsvno_relpathbuf(fromndp); 1433 NFSEXITCODE(error); 1434 return (error); 1435 } 1436 1437 /* 1438 * Link vnode op. 1439 */ 1440 int 1441 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, 1442 struct thread *p, struct nfsexstuff *exp) 1443 { 1444 struct vnode *xp; 1445 int error = 0; 1446 1447 xp = ndp->ni_vp; 1448 if (xp != NULL) { 1449 error = EEXIST; 1450 } else { 1451 xp = ndp->ni_dvp; 1452 if (vp->v_mount != xp->v_mount) 1453 error = EXDEV; 1454 } 1455 if (!error) { 1456 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1457 if ((vp->v_iflag & VI_DOOMED) == 0) 1458 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); 1459 else 1460 error = EPERM; 1461 if (ndp->ni_dvp == vp) 1462 vrele(ndp->ni_dvp); 1463 else 1464 vput(ndp->ni_dvp); 1465 NFSVOPUNLOCK(vp, 0); 1466 } else { 1467 if (ndp->ni_dvp == ndp->ni_vp) 1468 vrele(ndp->ni_dvp); 1469 else 1470 vput(ndp->ni_dvp); 1471 if (ndp->ni_vp) 1472 vrele(ndp->ni_vp); 1473 } 1474 nfsvno_relpathbuf(ndp); 1475 NFSEXITCODE(error); 1476 return (error); 1477 } 1478 1479 /* 1480 * Do the fsync() appropriate for the commit. 1481 */ 1482 int 1483 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, 1484 struct thread *td) 1485 { 1486 int error = 0; 1487 1488 /* 1489 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of 1490 * file is done. At this time VOP_FSYNC does not accept offset and 1491 * byte count parameters so call VOP_FSYNC the whole file for now. 1492 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. 1493 * File systems that do not use the buffer cache (as indicated 1494 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). 1495 */ 1496 if (cnt == 0 || cnt > MAX_COMMIT_COUNT || 1497 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { 1498 /* 1499 * Give up and do the whole thing 1500 */ 1501 if (vp->v_object && 1502 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 1503 VM_OBJECT_WLOCK(vp->v_object); 1504 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); 1505 VM_OBJECT_WUNLOCK(vp->v_object); 1506 } 1507 error = VOP_FSYNC(vp, MNT_WAIT, td); 1508 } else { 1509 /* 1510 * Locate and synchronously write any buffers that fall 1511 * into the requested range. Note: we are assuming that 1512 * f_iosize is a power of 2. 1513 */ 1514 int iosize = vp->v_mount->mnt_stat.f_iosize; 1515 int iomask = iosize - 1; 1516 struct bufobj *bo; 1517 daddr_t lblkno; 1518 1519 /* 1520 * Align to iosize boundary, super-align to page boundary. 1521 */ 1522 if (off & iomask) { 1523 cnt += off & iomask; 1524 off &= ~(u_quad_t)iomask; 1525 } 1526 if (off & PAGE_MASK) { 1527 cnt += off & PAGE_MASK; 1528 off &= ~(u_quad_t)PAGE_MASK; 1529 } 1530 lblkno = off / iosize; 1531 1532 if (vp->v_object && 1533 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 1534 VM_OBJECT_WLOCK(vp->v_object); 1535 vm_object_page_clean(vp->v_object, off, off + cnt, 1536 OBJPC_SYNC); 1537 VM_OBJECT_WUNLOCK(vp->v_object); 1538 } 1539 1540 bo = &vp->v_bufobj; 1541 BO_LOCK(bo); 1542 while (cnt > 0) { 1543 struct buf *bp; 1544 1545 /* 1546 * If we have a buffer and it is marked B_DELWRI we 1547 * have to lock and write it. Otherwise the prior 1548 * write is assumed to have already been committed. 1549 * 1550 * gbincore() can return invalid buffers now so we 1551 * have to check that bit as well (though B_DELWRI 1552 * should not be set if B_INVAL is set there could be 1553 * a race here since we haven't locked the buffer). 1554 */ 1555 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 1556 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 1557 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { 1558 BO_LOCK(bo); 1559 continue; /* retry */ 1560 } 1561 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 1562 B_DELWRI) { 1563 bremfree(bp); 1564 bp->b_flags &= ~B_ASYNC; 1565 bwrite(bp); 1566 ++nfs_commit_miss; 1567 } else 1568 BUF_UNLOCK(bp); 1569 BO_LOCK(bo); 1570 } 1571 ++nfs_commit_blks; 1572 if (cnt < iosize) 1573 break; 1574 cnt -= iosize; 1575 ++lblkno; 1576 } 1577 BO_UNLOCK(bo); 1578 } 1579 NFSEXITCODE(error); 1580 return (error); 1581 } 1582 1583 /* 1584 * Statfs vnode op. 1585 */ 1586 int 1587 nfsvno_statfs(struct vnode *vp, struct statfs *sf) 1588 { 1589 struct statfs *tsf; 1590 int error; 1591 1592 tsf = NULL; 1593 if (nfsrv_devidcnt > 0) { 1594 /* For a pNFS service, get the DS numbers. */ 1595 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); 1596 error = nfsrv_pnfsstatfs(tsf, vp->v_mount); 1597 if (error != 0) { 1598 free(tsf, M_TEMP); 1599 tsf = NULL; 1600 } 1601 } 1602 error = VFS_STATFS(vp->v_mount, sf); 1603 if (error == 0) { 1604 if (tsf != NULL) { 1605 sf->f_blocks = tsf->f_blocks; 1606 sf->f_bavail = tsf->f_bavail; 1607 sf->f_bfree = tsf->f_bfree; 1608 sf->f_bsize = tsf->f_bsize; 1609 } 1610 /* 1611 * Since NFS handles these values as unsigned on the 1612 * wire, there is no way to represent negative values, 1613 * so set them to 0. Without this, they will appear 1614 * to be very large positive values for clients like 1615 * Solaris10. 1616 */ 1617 if (sf->f_bavail < 0) 1618 sf->f_bavail = 0; 1619 if (sf->f_ffree < 0) 1620 sf->f_ffree = 0; 1621 } 1622 free(tsf, M_TEMP); 1623 NFSEXITCODE(error); 1624 return (error); 1625 } 1626 1627 /* 1628 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but 1629 * must handle nfsrv_opencheck() calls after any other access checks. 1630 */ 1631 void 1632 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, 1633 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, 1634 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, 1635 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, 1636 struct nfsexstuff *exp, struct vnode **vpp) 1637 { 1638 struct vnode *vp = NULL; 1639 u_quad_t tempsize; 1640 struct nfsexstuff nes; 1641 struct thread *p = curthread; 1642 1643 if (ndp->ni_vp == NULL) 1644 nd->nd_repstat = nfsrv_opencheck(clientid, 1645 stateidp, stp, NULL, nd, p, nd->nd_repstat); 1646 if (!nd->nd_repstat) { 1647 if (ndp->ni_vp == NULL) { 1648 vrele(ndp->ni_startdir); 1649 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, 1650 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1651 /* For a pNFS server, create the data file on a DS. */ 1652 if (nd->nd_repstat == 0) { 1653 /* 1654 * Create a data file on a DS for a pNFS server. 1655 * This function just returns if not 1656 * running a pNFS DS or the creation fails. 1657 */ 1658 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1659 cred, p); 1660 } 1661 vput(ndp->ni_dvp); 1662 nfsvno_relpathbuf(ndp); 1663 if (!nd->nd_repstat) { 1664 if (*exclusive_flagp) { 1665 *exclusive_flagp = 0; 1666 NFSVNO_ATTRINIT(nvap); 1667 nvap->na_atime.tv_sec = cverf[0]; 1668 nvap->na_atime.tv_nsec = cverf[1]; 1669 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, 1670 &nvap->na_vattr, cred); 1671 if (nd->nd_repstat != 0) { 1672 vput(ndp->ni_vp); 1673 ndp->ni_vp = NULL; 1674 nd->nd_repstat = NFSERR_NOTSUPP; 1675 } else 1676 NFSSETBIT_ATTRBIT(attrbitp, 1677 NFSATTRBIT_TIMEACCESS); 1678 } else { 1679 nfsrv_fixattr(nd, ndp->ni_vp, nvap, 1680 aclp, p, attrbitp, exp); 1681 } 1682 } 1683 vp = ndp->ni_vp; 1684 } else { 1685 if (ndp->ni_startdir) 1686 vrele(ndp->ni_startdir); 1687 nfsvno_relpathbuf(ndp); 1688 vp = ndp->ni_vp; 1689 if (create == NFSV4OPEN_CREATE) { 1690 if (ndp->ni_dvp == vp) 1691 vrele(ndp->ni_dvp); 1692 else 1693 vput(ndp->ni_dvp); 1694 } 1695 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { 1696 if (ndp->ni_cnd.cn_flags & RDONLY) 1697 NFSVNO_SETEXRDONLY(&nes); 1698 else 1699 NFSVNO_EXINIT(&nes); 1700 nd->nd_repstat = nfsvno_accchk(vp, 1701 VWRITE, cred, &nes, p, 1702 NFSACCCHK_NOOVERRIDE, 1703 NFSACCCHK_VPISLOCKED, NULL); 1704 nd->nd_repstat = nfsrv_opencheck(clientid, 1705 stateidp, stp, vp, nd, p, nd->nd_repstat); 1706 if (!nd->nd_repstat) { 1707 tempsize = nvap->na_size; 1708 NFSVNO_ATTRINIT(nvap); 1709 nvap->na_size = tempsize; 1710 nd->nd_repstat = VOP_SETATTR(vp, 1711 &nvap->na_vattr, cred); 1712 } 1713 } else if (vp->v_type == VREG) { 1714 nd->nd_repstat = nfsrv_opencheck(clientid, 1715 stateidp, stp, vp, nd, p, nd->nd_repstat); 1716 } 1717 } 1718 } else { 1719 if (ndp->ni_cnd.cn_flags & HASBUF) 1720 nfsvno_relpathbuf(ndp); 1721 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) { 1722 vrele(ndp->ni_startdir); 1723 if (ndp->ni_dvp == ndp->ni_vp) 1724 vrele(ndp->ni_dvp); 1725 else 1726 vput(ndp->ni_dvp); 1727 if (ndp->ni_vp) 1728 vput(ndp->ni_vp); 1729 } 1730 } 1731 *vpp = vp; 1732 1733 NFSEXITCODE2(0, nd); 1734 } 1735 1736 /* 1737 * Updates the file rev and sets the mtime and ctime 1738 * to the current clock time, returning the va_filerev and va_Xtime 1739 * values. 1740 * Return ESTALE to indicate the vnode is VI_DOOMED. 1741 */ 1742 int 1743 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, 1744 struct nfsrv_descript *nd, struct thread *p) 1745 { 1746 struct vattr va; 1747 1748 VATTR_NULL(&va); 1749 vfs_timestamp(&va.va_mtime); 1750 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 1751 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 1752 if ((vp->v_iflag & VI_DOOMED) != 0) 1753 return (ESTALE); 1754 } 1755 (void) VOP_SETATTR(vp, &va, nd->nd_cred); 1756 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); 1757 return (0); 1758 } 1759 1760 /* 1761 * Glue routine to nfsv4_fillattr(). 1762 */ 1763 int 1764 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, 1765 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, 1766 struct ucred *cred, struct thread *p, int isdgram, int reterr, 1767 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) 1768 { 1769 struct statfs *sf; 1770 int error; 1771 1772 sf = NULL; 1773 if (nfsrv_devidcnt > 0 && 1774 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || 1775 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || 1776 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { 1777 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); 1778 error = nfsrv_pnfsstatfs(sf, mp); 1779 if (error != 0) { 1780 free(sf, M_TEMP); 1781 sf = NULL; 1782 } 1783 } 1784 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, 1785 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, 1786 mounted_on_fileno, sf); 1787 free(sf, M_TEMP); 1788 NFSEXITCODE2(0, nd); 1789 return (error); 1790 } 1791 1792 /* Since the Readdir vnode ops vary, put the entire functions in here. */ 1793 /* 1794 * nfs readdir service 1795 * - mallocs what it thinks is enough to read 1796 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR 1797 * - calls VOP_READDIR() 1798 * - loops around building the reply 1799 * if the output generated exceeds count break out of loop 1800 * The NFSM_CLGET macro is used here so that the reply will be packed 1801 * tightly in mbuf clusters. 1802 * - it trims out records with d_fileno == 0 1803 * this doesn't matter for Unix clients, but they might confuse clients 1804 * for other os'. 1805 * - it trims out records with d_type == DT_WHT 1806 * these cannot be seen through NFS (unless we extend the protocol) 1807 * The alternate call nfsrvd_readdirplus() does lookups as well. 1808 * PS: The NFS protocol spec. does not clarify what the "count" byte 1809 * argument is a count of.. just name strings and file id's or the 1810 * entire reply rpc or ... 1811 * I tried just file name and id sizes and it confused the Sun client, 1812 * so I am using the full rpc size now. The "paranoia.." comment refers 1813 * to including the status longwords that are not a part of the dir. 1814 * "entry" structures, but are in the rpc. 1815 */ 1816 int 1817 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, 1818 struct vnode *vp, struct nfsexstuff *exp) 1819 { 1820 struct dirent *dp; 1821 u_int32_t *tl; 1822 int dirlen; 1823 char *cpos, *cend, *rbuf; 1824 struct nfsvattr at; 1825 int nlen, error = 0, getret = 1; 1826 int siz, cnt, fullsiz, eofflag, ncookies; 1827 u_int64_t off, toff, verf __unused; 1828 u_long *cookies = NULL, *cookiep; 1829 struct uio io; 1830 struct iovec iv; 1831 int is_ufs; 1832 struct thread *p = curthread; 1833 1834 if (nd->nd_repstat) { 1835 nfsrv_postopattr(nd, getret, &at); 1836 goto out; 1837 } 1838 if (nd->nd_flag & ND_NFSV2) { 1839 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1840 off = fxdr_unsigned(u_quad_t, *tl++); 1841 } else { 1842 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1843 off = fxdr_hyper(tl); 1844 tl += 2; 1845 verf = fxdr_hyper(tl); 1846 tl += 2; 1847 } 1848 toff = off; 1849 cnt = fxdr_unsigned(int, *tl); 1850 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 1851 cnt = NFS_SRVMAXDATA(nd); 1852 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 1853 fullsiz = siz; 1854 if (nd->nd_flag & ND_NFSV3) { 1855 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, 1856 NULL); 1857 #if 0 1858 /* 1859 * va_filerev is not sufficient as a cookie verifier, 1860 * since it is not supposed to change when entries are 1861 * removed/added unless that offset cookies returned to 1862 * the client are no longer valid. 1863 */ 1864 if (!nd->nd_repstat && toff && verf != at.na_filerev) 1865 nd->nd_repstat = NFSERR_BAD_COOKIE; 1866 #endif 1867 } 1868 if (!nd->nd_repstat && vp->v_type != VDIR) 1869 nd->nd_repstat = NFSERR_NOTDIR; 1870 if (nd->nd_repstat == 0 && cnt == 0) { 1871 if (nd->nd_flag & ND_NFSV2) 1872 /* NFSv2 does not have NFSERR_TOOSMALL */ 1873 nd->nd_repstat = EPERM; 1874 else 1875 nd->nd_repstat = NFSERR_TOOSMALL; 1876 } 1877 if (!nd->nd_repstat) 1878 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 1879 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1880 NFSACCCHK_VPISLOCKED, NULL); 1881 if (nd->nd_repstat) { 1882 vput(vp); 1883 if (nd->nd_flag & ND_NFSV3) 1884 nfsrv_postopattr(nd, getret, &at); 1885 goto out; 1886 } 1887 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 1888 rbuf = malloc(siz, M_TEMP, M_WAITOK); 1889 again: 1890 eofflag = 0; 1891 if (cookies) { 1892 free(cookies, M_TEMP); 1893 cookies = NULL; 1894 } 1895 1896 iv.iov_base = rbuf; 1897 iv.iov_len = siz; 1898 io.uio_iov = &iv; 1899 io.uio_iovcnt = 1; 1900 io.uio_offset = (off_t)off; 1901 io.uio_resid = siz; 1902 io.uio_segflg = UIO_SYSSPACE; 1903 io.uio_rw = UIO_READ; 1904 io.uio_td = NULL; 1905 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 1906 &cookies); 1907 off = (u_int64_t)io.uio_offset; 1908 if (io.uio_resid) 1909 siz -= io.uio_resid; 1910 1911 if (!cookies && !nd->nd_repstat) 1912 nd->nd_repstat = NFSERR_PERM; 1913 if (nd->nd_flag & ND_NFSV3) { 1914 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 1915 if (!nd->nd_repstat) 1916 nd->nd_repstat = getret; 1917 } 1918 1919 /* 1920 * Handles the failed cases. nd->nd_repstat == 0 past here. 1921 */ 1922 if (nd->nd_repstat) { 1923 vput(vp); 1924 free(rbuf, M_TEMP); 1925 if (cookies) 1926 free(cookies, M_TEMP); 1927 if (nd->nd_flag & ND_NFSV3) 1928 nfsrv_postopattr(nd, getret, &at); 1929 goto out; 1930 } 1931 /* 1932 * If nothing read, return eof 1933 * rpc reply 1934 */ 1935 if (siz == 0) { 1936 vput(vp); 1937 if (nd->nd_flag & ND_NFSV2) { 1938 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1939 } else { 1940 nfsrv_postopattr(nd, getret, &at); 1941 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1942 txdr_hyper(at.na_filerev, tl); 1943 tl += 2; 1944 } 1945 *tl++ = newnfs_false; 1946 *tl = newnfs_true; 1947 free(rbuf, M_TEMP); 1948 free(cookies, M_TEMP); 1949 goto out; 1950 } 1951 1952 /* 1953 * Check for degenerate cases of nothing useful read. 1954 * If so go try again 1955 */ 1956 cpos = rbuf; 1957 cend = rbuf + siz; 1958 dp = (struct dirent *)cpos; 1959 cookiep = cookies; 1960 1961 /* 1962 * For some reason FreeBSD's ufs_readdir() chooses to back the 1963 * directory offset up to a block boundary, so it is necessary to 1964 * skip over the records that precede the requested offset. This 1965 * requires the assumption that file offset cookies monotonically 1966 * increase. 1967 */ 1968 while (cpos < cend && ncookies > 0 && 1969 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 1970 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { 1971 cpos += dp->d_reclen; 1972 dp = (struct dirent *)cpos; 1973 cookiep++; 1974 ncookies--; 1975 } 1976 if (cpos >= cend || ncookies == 0) { 1977 siz = fullsiz; 1978 toff = off; 1979 goto again; 1980 } 1981 vput(vp); 1982 1983 /* 1984 * dirlen is the size of the reply, including all XDR and must 1985 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate 1986 * if the XDR should be included in "count", but to be safe, we do. 1987 * (Include the two booleans at the end of the reply in dirlen now.) 1988 */ 1989 if (nd->nd_flag & ND_NFSV3) { 1990 nfsrv_postopattr(nd, getret, &at); 1991 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1992 txdr_hyper(at.na_filerev, tl); 1993 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 1994 } else { 1995 dirlen = 2 * NFSX_UNSIGNED; 1996 } 1997 1998 /* Loop through the records and build reply */ 1999 while (cpos < cend && ncookies > 0) { 2000 nlen = dp->d_namlen; 2001 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2002 nlen <= NFS_MAXNAMLEN) { 2003 if (nd->nd_flag & ND_NFSV3) 2004 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2005 else 2006 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2007 if (dirlen > cnt) { 2008 eofflag = 0; 2009 break; 2010 } 2011 2012 /* 2013 * Build the directory record xdr from 2014 * the dirent entry. 2015 */ 2016 if (nd->nd_flag & ND_NFSV3) { 2017 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2018 *tl++ = newnfs_true; 2019 *tl++ = 0; 2020 } else { 2021 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2022 *tl++ = newnfs_true; 2023 } 2024 *tl = txdr_unsigned(dp->d_fileno); 2025 (void) nfsm_strtom(nd, dp->d_name, nlen); 2026 if (nd->nd_flag & ND_NFSV3) { 2027 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2028 *tl++ = 0; 2029 } else 2030 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 2031 *tl = txdr_unsigned(*cookiep); 2032 } 2033 cpos += dp->d_reclen; 2034 dp = (struct dirent *)cpos; 2035 cookiep++; 2036 ncookies--; 2037 } 2038 if (cpos < cend) 2039 eofflag = 0; 2040 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2041 *tl++ = newnfs_false; 2042 if (eofflag) 2043 *tl = newnfs_true; 2044 else 2045 *tl = newnfs_false; 2046 free(rbuf, M_TEMP); 2047 free(cookies, M_TEMP); 2048 2049 out: 2050 NFSEXITCODE2(0, nd); 2051 return (0); 2052 nfsmout: 2053 vput(vp); 2054 NFSEXITCODE2(error, nd); 2055 return (error); 2056 } 2057 2058 /* 2059 * Readdirplus for V3 and Readdir for V4. 2060 */ 2061 int 2062 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, 2063 struct vnode *vp, struct nfsexstuff *exp) 2064 { 2065 struct dirent *dp; 2066 u_int32_t *tl; 2067 int dirlen; 2068 char *cpos, *cend, *rbuf; 2069 struct vnode *nvp; 2070 fhandle_t nfh; 2071 struct nfsvattr nva, at, *nvap = &nva; 2072 struct mbuf *mb0, *mb1; 2073 struct nfsreferral *refp; 2074 int nlen, r, error = 0, getret = 1, usevget = 1; 2075 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; 2076 caddr_t bpos0, bpos1; 2077 u_int64_t off, toff, verf; 2078 u_long *cookies = NULL, *cookiep; 2079 nfsattrbit_t attrbits, rderrbits, savbits; 2080 struct uio io; 2081 struct iovec iv; 2082 struct componentname cn; 2083 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; 2084 struct mount *mp, *new_mp; 2085 uint64_t mounted_on_fileno; 2086 struct thread *p = curthread; 2087 2088 if (nd->nd_repstat) { 2089 nfsrv_postopattr(nd, getret, &at); 2090 goto out; 2091 } 2092 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2093 off = fxdr_hyper(tl); 2094 toff = off; 2095 tl += 2; 2096 verf = fxdr_hyper(tl); 2097 tl += 2; 2098 siz = fxdr_unsigned(int, *tl++); 2099 cnt = fxdr_unsigned(int, *tl); 2100 2101 /* 2102 * Use the server's maximum data transfer size as the upper bound 2103 * on reply datalen. 2104 */ 2105 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2106 cnt = NFS_SRVMAXDATA(nd); 2107 2108 /* 2109 * siz is a "hint" of how much directory information (name, fileid, 2110 * cookie) should be in the reply. At least one client "hints" 0, 2111 * so I set it to cnt for that case. I also round it up to the 2112 * next multiple of DIRBLKSIZ. 2113 * Since the size of a Readdirplus directory entry reply will always 2114 * be greater than a directory entry returned by VOP_READDIR(), it 2115 * does not make sense to read more than NFS_SRVMAXDATA() via 2116 * VOP_READDIR(). 2117 */ 2118 if (siz <= 0) 2119 siz = cnt; 2120 else if (siz > NFS_SRVMAXDATA(nd)) 2121 siz = NFS_SRVMAXDATA(nd); 2122 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2123 2124 if (nd->nd_flag & ND_NFSV4) { 2125 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 2126 if (error) 2127 goto nfsmout; 2128 NFSSET_ATTRBIT(&savbits, &attrbits); 2129 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits); 2130 NFSZERO_ATTRBIT(&rderrbits); 2131 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); 2132 } else { 2133 NFSZERO_ATTRBIT(&attrbits); 2134 } 2135 fullsiz = siz; 2136 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2137 #if 0 2138 if (!nd->nd_repstat) { 2139 if (off && verf != at.na_filerev) { 2140 /* 2141 * va_filerev is not sufficient as a cookie verifier, 2142 * since it is not supposed to change when entries are 2143 * removed/added unless that offset cookies returned to 2144 * the client are no longer valid. 2145 */ 2146 if (nd->nd_flag & ND_NFSV4) { 2147 nd->nd_repstat = NFSERR_NOTSAME; 2148 } else { 2149 nd->nd_repstat = NFSERR_BAD_COOKIE; 2150 } 2151 } 2152 } 2153 #endif 2154 if (!nd->nd_repstat && vp->v_type != VDIR) 2155 nd->nd_repstat = NFSERR_NOTDIR; 2156 if (!nd->nd_repstat && cnt == 0) 2157 nd->nd_repstat = NFSERR_TOOSMALL; 2158 if (!nd->nd_repstat) 2159 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2160 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2161 NFSACCCHK_VPISLOCKED, NULL); 2162 if (nd->nd_repstat) { 2163 vput(vp); 2164 if (nd->nd_flag & ND_NFSV3) 2165 nfsrv_postopattr(nd, getret, &at); 2166 goto out; 2167 } 2168 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2169 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; 2170 2171 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2172 again: 2173 eofflag = 0; 2174 if (cookies) { 2175 free(cookies, M_TEMP); 2176 cookies = NULL; 2177 } 2178 2179 iv.iov_base = rbuf; 2180 iv.iov_len = siz; 2181 io.uio_iov = &iv; 2182 io.uio_iovcnt = 1; 2183 io.uio_offset = (off_t)off; 2184 io.uio_resid = siz; 2185 io.uio_segflg = UIO_SYSSPACE; 2186 io.uio_rw = UIO_READ; 2187 io.uio_td = NULL; 2188 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2189 &cookies); 2190 off = (u_int64_t)io.uio_offset; 2191 if (io.uio_resid) 2192 siz -= io.uio_resid; 2193 2194 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2195 2196 if (!cookies && !nd->nd_repstat) 2197 nd->nd_repstat = NFSERR_PERM; 2198 if (!nd->nd_repstat) 2199 nd->nd_repstat = getret; 2200 if (nd->nd_repstat) { 2201 vput(vp); 2202 if (cookies) 2203 free(cookies, M_TEMP); 2204 free(rbuf, M_TEMP); 2205 if (nd->nd_flag & ND_NFSV3) 2206 nfsrv_postopattr(nd, getret, &at); 2207 goto out; 2208 } 2209 /* 2210 * If nothing read, return eof 2211 * rpc reply 2212 */ 2213 if (siz == 0) { 2214 vput(vp); 2215 if (nd->nd_flag & ND_NFSV3) 2216 nfsrv_postopattr(nd, getret, &at); 2217 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2218 txdr_hyper(at.na_filerev, tl); 2219 tl += 2; 2220 *tl++ = newnfs_false; 2221 *tl = newnfs_true; 2222 free(cookies, M_TEMP); 2223 free(rbuf, M_TEMP); 2224 goto out; 2225 } 2226 2227 /* 2228 * Check for degenerate cases of nothing useful read. 2229 * If so go try again 2230 */ 2231 cpos = rbuf; 2232 cend = rbuf + siz; 2233 dp = (struct dirent *)cpos; 2234 cookiep = cookies; 2235 2236 /* 2237 * For some reason FreeBSD's ufs_readdir() chooses to back the 2238 * directory offset up to a block boundary, so it is necessary to 2239 * skip over the records that precede the requested offset. This 2240 * requires the assumption that file offset cookies monotonically 2241 * increase. 2242 */ 2243 while (cpos < cend && ncookies > 0 && 2244 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2245 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || 2246 ((nd->nd_flag & ND_NFSV4) && 2247 ((dp->d_namlen == 1 && dp->d_name[0] == '.') || 2248 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { 2249 cpos += dp->d_reclen; 2250 dp = (struct dirent *)cpos; 2251 cookiep++; 2252 ncookies--; 2253 } 2254 if (cpos >= cend || ncookies == 0) { 2255 siz = fullsiz; 2256 toff = off; 2257 goto again; 2258 } 2259 2260 /* 2261 * Busy the file system so that the mount point won't go away 2262 * and, as such, VFS_VGET() can be used safely. 2263 */ 2264 mp = vp->v_mount; 2265 vfs_ref(mp); 2266 NFSVOPUNLOCK(vp, 0); 2267 nd->nd_repstat = vfs_busy(mp, 0); 2268 vfs_rel(mp); 2269 if (nd->nd_repstat != 0) { 2270 vrele(vp); 2271 free(cookies, M_TEMP); 2272 free(rbuf, M_TEMP); 2273 if (nd->nd_flag & ND_NFSV3) 2274 nfsrv_postopattr(nd, getret, &at); 2275 goto out; 2276 } 2277 2278 /* 2279 * Check to see if entries in this directory can be safely acquired 2280 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. 2281 * ZFS snapshot directories need VOP_LOOKUP(), so that any 2282 * automount of the snapshot directory that is required will 2283 * be done. 2284 * This needs to be done here for NFSv4, since NFSv4 never does 2285 * a VFS_VGET() for "." or "..". 2286 */ 2287 if (is_zfs == 1) { 2288 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); 2289 if (r == EOPNOTSUPP) { 2290 usevget = 0; 2291 cn.cn_nameiop = LOOKUP; 2292 cn.cn_lkflags = LK_SHARED | LK_RETRY; 2293 cn.cn_cred = nd->nd_cred; 2294 cn.cn_thread = p; 2295 } else if (r == 0) 2296 vput(nvp); 2297 } 2298 2299 /* 2300 * Save this position, in case there is an error before one entry 2301 * is created. 2302 */ 2303 mb0 = nd->nd_mb; 2304 bpos0 = nd->nd_bpos; 2305 2306 /* 2307 * Fill in the first part of the reply. 2308 * dirlen is the reply length in bytes and cannot exceed cnt. 2309 * (Include the two booleans at the end of the reply in dirlen now, 2310 * so we recognize when we have exceeded cnt.) 2311 */ 2312 if (nd->nd_flag & ND_NFSV3) { 2313 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2314 nfsrv_postopattr(nd, getret, &at); 2315 } else { 2316 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; 2317 } 2318 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); 2319 txdr_hyper(at.na_filerev, tl); 2320 2321 /* 2322 * Save this position, in case there is an empty reply needed. 2323 */ 2324 mb1 = nd->nd_mb; 2325 bpos1 = nd->nd_bpos; 2326 2327 /* Loop through the records and build reply */ 2328 entrycnt = 0; 2329 while (cpos < cend && ncookies > 0 && dirlen < cnt) { 2330 nlen = dp->d_namlen; 2331 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2332 nlen <= NFS_MAXNAMLEN && 2333 ((nd->nd_flag & ND_NFSV3) || nlen > 2 || 2334 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) 2335 || (nlen == 1 && dp->d_name[0] != '.'))) { 2336 /* 2337 * Save the current position in the reply, in case 2338 * this entry exceeds cnt. 2339 */ 2340 mb1 = nd->nd_mb; 2341 bpos1 = nd->nd_bpos; 2342 2343 /* 2344 * For readdir_and_lookup get the vnode using 2345 * the file number. 2346 */ 2347 nvp = NULL; 2348 refp = NULL; 2349 r = 0; 2350 at_root = 0; 2351 needs_unbusy = 0; 2352 new_mp = mp; 2353 mounted_on_fileno = (uint64_t)dp->d_fileno; 2354 if ((nd->nd_flag & ND_NFSV3) || 2355 NFSNONZERO_ATTRBIT(&savbits)) { 2356 if (nd->nd_flag & ND_NFSV4) 2357 refp = nfsv4root_getreferral(NULL, 2358 vp, dp->d_fileno); 2359 if (refp == NULL) { 2360 if (usevget) 2361 r = VFS_VGET(mp, dp->d_fileno, 2362 LK_SHARED, &nvp); 2363 else 2364 r = EOPNOTSUPP; 2365 if (r == EOPNOTSUPP) { 2366 if (usevget) { 2367 usevget = 0; 2368 cn.cn_nameiop = LOOKUP; 2369 cn.cn_lkflags = 2370 LK_SHARED | 2371 LK_RETRY; 2372 cn.cn_cred = 2373 nd->nd_cred; 2374 cn.cn_thread = p; 2375 } 2376 cn.cn_nameptr = dp->d_name; 2377 cn.cn_namelen = nlen; 2378 cn.cn_flags = ISLASTCN | 2379 NOFOLLOW | LOCKLEAF; 2380 if (nlen == 2 && 2381 dp->d_name[0] == '.' && 2382 dp->d_name[1] == '.') 2383 cn.cn_flags |= 2384 ISDOTDOT; 2385 if (NFSVOPLOCK(vp, LK_SHARED) 2386 != 0) { 2387 nd->nd_repstat = EPERM; 2388 break; 2389 } 2390 if ((vp->v_vflag & VV_ROOT) != 0 2391 && (cn.cn_flags & ISDOTDOT) 2392 != 0) { 2393 vref(vp); 2394 nvp = vp; 2395 r = 0; 2396 } else { 2397 r = VOP_LOOKUP(vp, &nvp, 2398 &cn); 2399 if (vp != nvp) 2400 NFSVOPUNLOCK(vp, 2401 0); 2402 } 2403 } 2404 2405 /* 2406 * For NFSv4, check to see if nvp is 2407 * a mount point and get the mount 2408 * point vnode, as required. 2409 */ 2410 if (r == 0 && 2411 nfsrv_enable_crossmntpt != 0 && 2412 (nd->nd_flag & ND_NFSV4) != 0 && 2413 nvp->v_type == VDIR && 2414 nvp->v_mountedhere != NULL) { 2415 new_mp = nvp->v_mountedhere; 2416 r = vfs_busy(new_mp, 0); 2417 vput(nvp); 2418 nvp = NULL; 2419 if (r == 0) { 2420 r = VFS_ROOT(new_mp, 2421 LK_SHARED, &nvp); 2422 needs_unbusy = 1; 2423 if (r == 0) 2424 at_root = 1; 2425 } 2426 } 2427 } 2428 2429 /* 2430 * If we failed to look up the entry, then it 2431 * has become invalid, most likely removed. 2432 */ 2433 if (r != 0) { 2434 if (needs_unbusy) 2435 vfs_unbusy(new_mp); 2436 goto invalid; 2437 } 2438 KASSERT(refp != NULL || nvp != NULL, 2439 ("%s: undetected lookup error", __func__)); 2440 2441 if (refp == NULL && 2442 ((nd->nd_flag & ND_NFSV3) || 2443 NFSNONZERO_ATTRBIT(&attrbits))) { 2444 r = nfsvno_getfh(nvp, &nfh, p); 2445 if (!r) 2446 r = nfsvno_getattr(nvp, nvap, nd, p, 2447 1, &attrbits); 2448 if (r == 0 && is_zfs == 1 && 2449 nfsrv_enable_crossmntpt != 0 && 2450 (nd->nd_flag & ND_NFSV4) != 0 && 2451 nvp->v_type == VDIR && 2452 vp->v_mount != nvp->v_mount) { 2453 /* 2454 * For a ZFS snapshot, there is a 2455 * pseudo mount that does not set 2456 * v_mountedhere, so it needs to 2457 * be detected via a different 2458 * mount structure. 2459 */ 2460 at_root = 1; 2461 if (new_mp == mp) 2462 new_mp = nvp->v_mount; 2463 } 2464 } 2465 2466 /* 2467 * If we failed to get attributes of the entry, 2468 * then just skip it for NFSv3 (the traditional 2469 * behavior in the old NFS server). 2470 * For NFSv4 the behavior is controlled by 2471 * RDATTRERROR: we either ignore the error or 2472 * fail the request. 2473 * Note that RDATTRERROR is never set for NFSv3. 2474 */ 2475 if (r != 0) { 2476 if (!NFSISSET_ATTRBIT(&attrbits, 2477 NFSATTRBIT_RDATTRERROR)) { 2478 vput(nvp); 2479 if (needs_unbusy != 0) 2480 vfs_unbusy(new_mp); 2481 if ((nd->nd_flag & ND_NFSV3)) 2482 goto invalid; 2483 nd->nd_repstat = r; 2484 break; 2485 } 2486 } 2487 } 2488 2489 /* 2490 * Build the directory record xdr 2491 */ 2492 if (nd->nd_flag & ND_NFSV3) { 2493 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2494 *tl++ = newnfs_true; 2495 *tl++ = 0; 2496 *tl = txdr_unsigned(dp->d_fileno); 2497 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2498 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2499 *tl++ = 0; 2500 *tl = txdr_unsigned(*cookiep); 2501 nfsrv_postopattr(nd, 0, nvap); 2502 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1); 2503 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); 2504 if (nvp != NULL) 2505 vput(nvp); 2506 } else { 2507 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2508 *tl++ = newnfs_true; 2509 *tl++ = 0; 2510 *tl = txdr_unsigned(*cookiep); 2511 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2512 if (nvp != NULL) { 2513 supports_nfsv4acls = 2514 nfs_supportsnfsv4acls(nvp); 2515 NFSVOPUNLOCK(nvp, 0); 2516 } else 2517 supports_nfsv4acls = 0; 2518 if (refp != NULL) { 2519 dirlen += nfsrv_putreferralattr(nd, 2520 &savbits, refp, 0, 2521 &nd->nd_repstat); 2522 if (nd->nd_repstat) { 2523 if (nvp != NULL) 2524 vrele(nvp); 2525 if (needs_unbusy != 0) 2526 vfs_unbusy(new_mp); 2527 break; 2528 } 2529 } else if (r) { 2530 dirlen += nfsvno_fillattr(nd, new_mp, 2531 nvp, nvap, &nfh, r, &rderrbits, 2532 nd->nd_cred, p, isdgram, 0, 2533 supports_nfsv4acls, at_root, 2534 mounted_on_fileno); 2535 } else { 2536 dirlen += nfsvno_fillattr(nd, new_mp, 2537 nvp, nvap, &nfh, r, &attrbits, 2538 nd->nd_cred, p, isdgram, 0, 2539 supports_nfsv4acls, at_root, 2540 mounted_on_fileno); 2541 } 2542 if (nvp != NULL) 2543 vrele(nvp); 2544 dirlen += (3 * NFSX_UNSIGNED); 2545 } 2546 if (needs_unbusy != 0) 2547 vfs_unbusy(new_mp); 2548 if (dirlen <= cnt) 2549 entrycnt++; 2550 } 2551 invalid: 2552 cpos += dp->d_reclen; 2553 dp = (struct dirent *)cpos; 2554 cookiep++; 2555 ncookies--; 2556 } 2557 vrele(vp); 2558 vfs_unbusy(mp); 2559 2560 /* 2561 * If dirlen > cnt, we must strip off the last entry. If that 2562 * results in an empty reply, report NFSERR_TOOSMALL. 2563 */ 2564 if (dirlen > cnt || nd->nd_repstat) { 2565 if (!nd->nd_repstat && entrycnt == 0) 2566 nd->nd_repstat = NFSERR_TOOSMALL; 2567 if (nd->nd_repstat) { 2568 newnfs_trimtrailing(nd, mb0, bpos0); 2569 if (nd->nd_flag & ND_NFSV3) 2570 nfsrv_postopattr(nd, getret, &at); 2571 } else 2572 newnfs_trimtrailing(nd, mb1, bpos1); 2573 eofflag = 0; 2574 } else if (cpos < cend) 2575 eofflag = 0; 2576 if (!nd->nd_repstat) { 2577 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2578 *tl++ = newnfs_false; 2579 if (eofflag) 2580 *tl = newnfs_true; 2581 else 2582 *tl = newnfs_false; 2583 } 2584 free(cookies, M_TEMP); 2585 free(rbuf, M_TEMP); 2586 2587 out: 2588 NFSEXITCODE2(0, nd); 2589 return (0); 2590 nfsmout: 2591 vput(vp); 2592 NFSEXITCODE2(error, nd); 2593 return (error); 2594 } 2595 2596 /* 2597 * Get the settable attributes out of the mbuf list. 2598 * (Return 0 or EBADRPC) 2599 */ 2600 int 2601 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2602 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2603 { 2604 u_int32_t *tl; 2605 struct nfsv2_sattr *sp; 2606 int error = 0, toclient = 0; 2607 2608 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 2609 case ND_NFSV2: 2610 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2611 /* 2612 * Some old clients didn't fill in the high order 16bits. 2613 * --> check the low order 2 bytes for 0xffff 2614 */ 2615 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 2616 nvap->na_mode = nfstov_mode(sp->sa_mode); 2617 if (sp->sa_uid != newnfs_xdrneg1) 2618 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 2619 if (sp->sa_gid != newnfs_xdrneg1) 2620 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 2621 if (sp->sa_size != newnfs_xdrneg1) 2622 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 2623 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 2624 #ifdef notyet 2625 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 2626 #else 2627 nvap->na_atime.tv_sec = 2628 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 2629 nvap->na_atime.tv_nsec = 0; 2630 #endif 2631 } 2632 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 2633 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 2634 break; 2635 case ND_NFSV3: 2636 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2637 if (*tl == newnfs_true) { 2638 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2639 nvap->na_mode = nfstov_mode(*tl); 2640 } 2641 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2642 if (*tl == newnfs_true) { 2643 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2644 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 2645 } 2646 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2647 if (*tl == newnfs_true) { 2648 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2649 nvap->na_gid = fxdr_unsigned(gid_t, *tl); 2650 } 2651 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2652 if (*tl == newnfs_true) { 2653 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2654 nvap->na_size = fxdr_hyper(tl); 2655 } 2656 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2657 switch (fxdr_unsigned(int, *tl)) { 2658 case NFSV3SATTRTIME_TOCLIENT: 2659 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2660 fxdr_nfsv3time(tl, &nvap->na_atime); 2661 toclient = 1; 2662 break; 2663 case NFSV3SATTRTIME_TOSERVER: 2664 vfs_timestamp(&nvap->na_atime); 2665 nvap->na_vaflags |= VA_UTIMES_NULL; 2666 break; 2667 } 2668 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2669 switch (fxdr_unsigned(int, *tl)) { 2670 case NFSV3SATTRTIME_TOCLIENT: 2671 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2672 fxdr_nfsv3time(tl, &nvap->na_mtime); 2673 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2674 break; 2675 case NFSV3SATTRTIME_TOSERVER: 2676 vfs_timestamp(&nvap->na_mtime); 2677 if (!toclient) 2678 nvap->na_vaflags |= VA_UTIMES_NULL; 2679 break; 2680 } 2681 break; 2682 case ND_NFSV4: 2683 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); 2684 } 2685 nfsmout: 2686 NFSEXITCODE2(error, nd); 2687 return (error); 2688 } 2689 2690 /* 2691 * Handle the setable attributes for V4. 2692 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 2693 */ 2694 int 2695 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2696 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2697 { 2698 u_int32_t *tl; 2699 int attrsum = 0; 2700 int i, j; 2701 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 2702 int toclient = 0; 2703 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 2704 uid_t uid; 2705 gid_t gid; 2706 2707 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 2708 if (error) 2709 goto nfsmout; 2710 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2711 attrsize = fxdr_unsigned(int, *tl); 2712 2713 /* 2714 * Loop around getting the setable attributes. If an unsupported 2715 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 2716 */ 2717 if (retnotsup) { 2718 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2719 bitpos = NFSATTRBIT_MAX; 2720 } else { 2721 bitpos = 0; 2722 } 2723 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 2724 if (attrsum > attrsize) { 2725 error = NFSERR_BADXDR; 2726 goto nfsmout; 2727 } 2728 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 2729 switch (bitpos) { 2730 case NFSATTRBIT_SIZE: 2731 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 2732 if (vp != NULL && vp->v_type != VREG) { 2733 error = (vp->v_type == VDIR) ? NFSERR_ISDIR : 2734 NFSERR_INVAL; 2735 goto nfsmout; 2736 } 2737 nvap->na_size = fxdr_hyper(tl); 2738 attrsum += NFSX_HYPER; 2739 break; 2740 case NFSATTRBIT_ACL: 2741 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize, 2742 p); 2743 if (error) 2744 goto nfsmout; 2745 if (aceerr && !nd->nd_repstat) 2746 nd->nd_repstat = aceerr; 2747 attrsum += aclsize; 2748 break; 2749 case NFSATTRBIT_ARCHIVE: 2750 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2751 if (!nd->nd_repstat) 2752 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2753 attrsum += NFSX_UNSIGNED; 2754 break; 2755 case NFSATTRBIT_HIDDEN: 2756 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2757 if (!nd->nd_repstat) 2758 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2759 attrsum += NFSX_UNSIGNED; 2760 break; 2761 case NFSATTRBIT_MIMETYPE: 2762 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2763 i = fxdr_unsigned(int, *tl); 2764 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 2765 if (error) 2766 goto nfsmout; 2767 if (!nd->nd_repstat) 2768 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2769 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 2770 break; 2771 case NFSATTRBIT_MODE: 2772 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2773 nvap->na_mode = nfstov_mode(*tl); 2774 attrsum += NFSX_UNSIGNED; 2775 break; 2776 case NFSATTRBIT_OWNER: 2777 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2778 j = fxdr_unsigned(int, *tl); 2779 if (j < 0) { 2780 error = NFSERR_BADXDR; 2781 goto nfsmout; 2782 } 2783 if (j > NFSV4_SMALLSTR) 2784 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 2785 else 2786 cp = namestr; 2787 error = nfsrv_mtostr(nd, cp, j); 2788 if (error) { 2789 if (j > NFSV4_SMALLSTR) 2790 free(cp, M_NFSSTRING); 2791 goto nfsmout; 2792 } 2793 if (!nd->nd_repstat) { 2794 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, 2795 &uid); 2796 if (!nd->nd_repstat) 2797 nvap->na_uid = uid; 2798 } 2799 if (j > NFSV4_SMALLSTR) 2800 free(cp, M_NFSSTRING); 2801 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 2802 break; 2803 case NFSATTRBIT_OWNERGROUP: 2804 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2805 j = fxdr_unsigned(int, *tl); 2806 if (j < 0) { 2807 error = NFSERR_BADXDR; 2808 goto nfsmout; 2809 } 2810 if (j > NFSV4_SMALLSTR) 2811 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 2812 else 2813 cp = namestr; 2814 error = nfsrv_mtostr(nd, cp, j); 2815 if (error) { 2816 if (j > NFSV4_SMALLSTR) 2817 free(cp, M_NFSSTRING); 2818 goto nfsmout; 2819 } 2820 if (!nd->nd_repstat) { 2821 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, 2822 &gid); 2823 if (!nd->nd_repstat) 2824 nvap->na_gid = gid; 2825 } 2826 if (j > NFSV4_SMALLSTR) 2827 free(cp, M_NFSSTRING); 2828 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 2829 break; 2830 case NFSATTRBIT_SYSTEM: 2831 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2832 if (!nd->nd_repstat) 2833 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2834 attrsum += NFSX_UNSIGNED; 2835 break; 2836 case NFSATTRBIT_TIMEACCESSSET: 2837 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2838 attrsum += NFSX_UNSIGNED; 2839 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 2840 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2841 fxdr_nfsv4time(tl, &nvap->na_atime); 2842 toclient = 1; 2843 attrsum += NFSX_V4TIME; 2844 } else { 2845 vfs_timestamp(&nvap->na_atime); 2846 nvap->na_vaflags |= VA_UTIMES_NULL; 2847 } 2848 break; 2849 case NFSATTRBIT_TIMEBACKUP: 2850 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2851 if (!nd->nd_repstat) 2852 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2853 attrsum += NFSX_V4TIME; 2854 break; 2855 case NFSATTRBIT_TIMECREATE: 2856 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2857 if (!nd->nd_repstat) 2858 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2859 attrsum += NFSX_V4TIME; 2860 break; 2861 case NFSATTRBIT_TIMEMODIFYSET: 2862 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2863 attrsum += NFSX_UNSIGNED; 2864 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 2865 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2866 fxdr_nfsv4time(tl, &nvap->na_mtime); 2867 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2868 attrsum += NFSX_V4TIME; 2869 } else { 2870 vfs_timestamp(&nvap->na_mtime); 2871 if (!toclient) 2872 nvap->na_vaflags |= VA_UTIMES_NULL; 2873 } 2874 break; 2875 default: 2876 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2877 /* 2878 * set bitpos so we drop out of the loop. 2879 */ 2880 bitpos = NFSATTRBIT_MAX; 2881 break; 2882 } 2883 } 2884 2885 /* 2886 * some clients pad the attrlist, so we need to skip over the 2887 * padding. 2888 */ 2889 if (attrsum > attrsize) { 2890 error = NFSERR_BADXDR; 2891 } else { 2892 attrsize = NFSM_RNDUP(attrsize); 2893 if (attrsum < attrsize) 2894 error = nfsm_advance(nd, attrsize - attrsum, -1); 2895 } 2896 nfsmout: 2897 NFSEXITCODE2(error, nd); 2898 return (error); 2899 } 2900 2901 /* 2902 * Check/setup export credentials. 2903 */ 2904 int 2905 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 2906 struct ucred *credanon) 2907 { 2908 int error = 0; 2909 2910 /* 2911 * Check/setup credentials. 2912 */ 2913 if (nd->nd_flag & ND_GSS) 2914 exp->nes_exflag &= ~MNT_EXPORTANON; 2915 2916 /* 2917 * Check to see if the operation is allowed for this security flavor. 2918 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to 2919 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. 2920 * Also, allow Secinfo, so that it can acquire the correct flavor(s). 2921 */ 2922 if (nfsvno_testexp(nd, exp) && 2923 nd->nd_procnum != NFSV4OP_SECINFO && 2924 nd->nd_procnum != NFSPROC_FSINFO) { 2925 if (nd->nd_flag & ND_NFSV4) 2926 error = NFSERR_WRONGSEC; 2927 else 2928 error = (NFSERR_AUTHERR | AUTH_TOOWEAK); 2929 goto out; 2930 } 2931 2932 /* 2933 * Check to see if the file system is exported V4 only. 2934 */ 2935 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 2936 error = NFSERR_PROGNOTV4; 2937 goto out; 2938 } 2939 2940 /* 2941 * Now, map the user credentials. 2942 * (Note that ND_AUTHNONE will only be set for an NFSv3 2943 * Fsinfo RPC. If set for anything else, this code might need 2944 * to change.) 2945 */ 2946 if (NFSVNO_EXPORTED(exp)) { 2947 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || 2948 NFSVNO_EXPORTANON(exp) || 2949 (nd->nd_flag & ND_AUTHNONE) != 0) { 2950 nd->nd_cred->cr_uid = credanon->cr_uid; 2951 nd->nd_cred->cr_gid = credanon->cr_gid; 2952 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 2953 credanon->cr_groups); 2954 } else if ((nd->nd_flag & ND_GSS) == 0) { 2955 /* 2956 * If using AUTH_SYS, call nfsrv_getgrpscred() to see 2957 * if there is a replacement credential with a group 2958 * list set up by "nfsuserd -manage-gids". 2959 * If there is no replacement, nfsrv_getgrpscred() 2960 * simply returns its argument. 2961 */ 2962 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); 2963 } 2964 } 2965 2966 out: 2967 NFSEXITCODE2(error, nd); 2968 return (error); 2969 } 2970 2971 /* 2972 * Check exports. 2973 */ 2974 int 2975 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 2976 struct ucred **credp) 2977 { 2978 int i, error, *secflavors; 2979 2980 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 2981 &exp->nes_numsecflavor, &secflavors); 2982 if (error) { 2983 if (nfs_rootfhset) { 2984 exp->nes_exflag = 0; 2985 exp->nes_numsecflavor = 0; 2986 error = 0; 2987 } 2988 } else { 2989 /* Copy the security flavors. */ 2990 for (i = 0; i < exp->nes_numsecflavor; i++) 2991 exp->nes_secflavors[i] = secflavors[i]; 2992 } 2993 NFSEXITCODE(error); 2994 return (error); 2995 } 2996 2997 /* 2998 * Get a vnode for a file handle and export stuff. 2999 */ 3000 int 3001 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 3002 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 3003 struct ucred **credp) 3004 { 3005 int i, error, *secflavors; 3006 3007 *credp = NULL; 3008 exp->nes_numsecflavor = 0; 3009 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); 3010 if (error != 0) 3011 /* Make sure the server replies ESTALE to the client. */ 3012 error = ESTALE; 3013 if (nam && !error) { 3014 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3015 &exp->nes_numsecflavor, &secflavors); 3016 if (error) { 3017 if (nfs_rootfhset) { 3018 exp->nes_exflag = 0; 3019 exp->nes_numsecflavor = 0; 3020 error = 0; 3021 } else { 3022 vput(*vpp); 3023 } 3024 } else { 3025 /* Copy the security flavors. */ 3026 for (i = 0; i < exp->nes_numsecflavor; i++) 3027 exp->nes_secflavors[i] = secflavors[i]; 3028 } 3029 } 3030 NFSEXITCODE(error); 3031 return (error); 3032 } 3033 3034 /* 3035 * nfsd_fhtovp() - convert a fh to a vnode ptr 3036 * - look up fsid in mount list (if not found ret error) 3037 * - get vp and export rights by calling nfsvno_fhtovp() 3038 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 3039 * for AUTH_SYS 3040 * - if mpp != NULL, return the mount point so that it can 3041 * be used for vn_finished_write() by the caller 3042 */ 3043 void 3044 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 3045 struct vnode **vpp, struct nfsexstuff *exp, 3046 struct mount **mpp, int startwrite) 3047 { 3048 struct mount *mp; 3049 struct ucred *credanon; 3050 fhandle_t *fhp; 3051 3052 fhp = (fhandle_t *)nfp->nfsrvfh_data; 3053 /* 3054 * Check for the special case of the nfsv4root_fh. 3055 */ 3056 mp = vfs_busyfs(&fhp->fh_fsid); 3057 if (mpp != NULL) 3058 *mpp = mp; 3059 if (mp == NULL) { 3060 *vpp = NULL; 3061 nd->nd_repstat = ESTALE; 3062 goto out; 3063 } 3064 3065 if (startwrite) { 3066 vn_start_write(NULL, mpp, V_WAIT); 3067 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) 3068 lktype = LK_EXCLUSIVE; 3069 } 3070 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 3071 &credanon); 3072 vfs_unbusy(mp); 3073 3074 /* 3075 * For NFSv4 without a pseudo root fs, unexported file handles 3076 * can be returned, so that Lookup works everywhere. 3077 */ 3078 if (!nd->nd_repstat && exp->nes_exflag == 0 && 3079 !(nd->nd_flag & ND_NFSV4)) { 3080 vput(*vpp); 3081 nd->nd_repstat = EACCES; 3082 } 3083 3084 /* 3085 * Personally, I've never seen any point in requiring a 3086 * reserved port#, since only in the rare case where the 3087 * clients are all boxes with secure system privileges, 3088 * does it provide any enhanced security, but... some people 3089 * believe it to be useful and keep putting this code back in. 3090 * (There is also some "security checker" out there that 3091 * complains if the nfs server doesn't enforce this.) 3092 * However, note the following: 3093 * RFC3530 (NFSv4) specifies that a reserved port# not be 3094 * required. 3095 * RFC2623 recommends that, if a reserved port# is checked for, 3096 * that there be a way to turn that off--> ifdef'd. 3097 */ 3098 #ifdef NFS_REQRSVPORT 3099 if (!nd->nd_repstat) { 3100 struct sockaddr_in *saddr; 3101 struct sockaddr_in6 *saddr6; 3102 3103 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 3104 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 3105 if (!(nd->nd_flag & ND_NFSV4) && 3106 ((saddr->sin_family == AF_INET && 3107 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 3108 (saddr6->sin6_family == AF_INET6 && 3109 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 3110 vput(*vpp); 3111 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3112 } 3113 } 3114 #endif /* NFS_REQRSVPORT */ 3115 3116 /* 3117 * Check/setup credentials. 3118 */ 3119 if (!nd->nd_repstat) { 3120 nd->nd_saveduid = nd->nd_cred->cr_uid; 3121 nd->nd_repstat = nfsd_excred(nd, exp, credanon); 3122 if (nd->nd_repstat) 3123 vput(*vpp); 3124 } 3125 if (credanon != NULL) 3126 crfree(credanon); 3127 if (nd->nd_repstat) { 3128 if (startwrite) 3129 vn_finished_write(mp); 3130 *vpp = NULL; 3131 if (mpp != NULL) 3132 *mpp = NULL; 3133 } 3134 3135 out: 3136 NFSEXITCODE2(0, nd); 3137 } 3138 3139 /* 3140 * glue for fp. 3141 */ 3142 static int 3143 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 3144 { 3145 struct filedesc *fdp; 3146 struct file *fp; 3147 int error = 0; 3148 3149 fdp = p->td_proc->p_fd; 3150 if (fd < 0 || fd >= fdp->fd_nfiles || 3151 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { 3152 error = EBADF; 3153 goto out; 3154 } 3155 *fpp = fp; 3156 3157 out: 3158 NFSEXITCODE(error); 3159 return (error); 3160 } 3161 3162 /* 3163 * Called from nfssvc() to update the exports list. Just call 3164 * vfs_export(). This has to be done, since the v4 root fake fs isn't 3165 * in the mount list. 3166 */ 3167 int 3168 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 3169 { 3170 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 3171 int error = 0; 3172 struct nameidata nd; 3173 fhandle_t fh; 3174 3175 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export); 3176 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 3177 nfs_rootfhset = 0; 3178 else if (error == 0) { 3179 if (nfsexargp->fspec == NULL) { 3180 error = EPERM; 3181 goto out; 3182 } 3183 /* 3184 * If fspec != NULL, this is the v4root path. 3185 */ 3186 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, 3187 nfsexargp->fspec, p); 3188 if ((error = namei(&nd)) != 0) 3189 goto out; 3190 error = nfsvno_getfh(nd.ni_vp, &fh, p); 3191 vrele(nd.ni_vp); 3192 if (!error) { 3193 nfs_rootfh.nfsrvfh_len = NFSX_MYFH; 3194 NFSBCOPY((caddr_t)&fh, 3195 nfs_rootfh.nfsrvfh_data, 3196 sizeof (fhandle_t)); 3197 nfs_rootfhset = 1; 3198 } 3199 } 3200 3201 out: 3202 NFSEXITCODE(error); 3203 return (error); 3204 } 3205 3206 /* 3207 * This function needs to test to see if the system is near its limit 3208 * for memory allocation via malloc() or mget() and return True iff 3209 * either of these resources are near their limit. 3210 * XXX (For now, this is just a stub.) 3211 */ 3212 int nfsrv_testmalloclimit = 0; 3213 int 3214 nfsrv_mallocmget_limit(void) 3215 { 3216 static int printmesg = 0; 3217 static int testval = 1; 3218 3219 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 3220 if ((printmesg++ % 100) == 0) 3221 printf("nfsd: malloc/mget near limit\n"); 3222 return (1); 3223 } 3224 return (0); 3225 } 3226 3227 /* 3228 * BSD specific initialization of a mount point. 3229 */ 3230 void 3231 nfsd_mntinit(void) 3232 { 3233 static int inited = 0; 3234 3235 if (inited) 3236 return; 3237 inited = 1; 3238 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 3239 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist); 3240 TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist); 3241 nfsv4root_mnt.mnt_export = NULL; 3242 TAILQ_INIT(&nfsv4root_opt); 3243 TAILQ_INIT(&nfsv4root_newopt); 3244 nfsv4root_mnt.mnt_opt = &nfsv4root_opt; 3245 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt; 3246 nfsv4root_mnt.mnt_nvnodelistsize = 0; 3247 nfsv4root_mnt.mnt_activevnodelistsize = 0; 3248 } 3249 3250 /* 3251 * Get a vnode for a file handle, without checking exports, etc. 3252 */ 3253 struct vnode * 3254 nfsvno_getvp(fhandle_t *fhp) 3255 { 3256 struct mount *mp; 3257 struct vnode *vp; 3258 int error; 3259 3260 mp = vfs_busyfs(&fhp->fh_fsid); 3261 if (mp == NULL) 3262 return (NULL); 3263 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 3264 vfs_unbusy(mp); 3265 if (error) 3266 return (NULL); 3267 return (vp); 3268 } 3269 3270 /* 3271 * Do a local VOP_ADVLOCK(). 3272 */ 3273 int 3274 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 3275 u_int64_t end, struct thread *td) 3276 { 3277 int error = 0; 3278 struct flock fl; 3279 u_int64_t tlen; 3280 3281 if (nfsrv_dolocallocks == 0) 3282 goto out; 3283 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); 3284 3285 fl.l_whence = SEEK_SET; 3286 fl.l_type = ftype; 3287 fl.l_start = (off_t)first; 3288 if (end == NFS64BITSSET) { 3289 fl.l_len = 0; 3290 } else { 3291 tlen = end - first; 3292 fl.l_len = (off_t)tlen; 3293 } 3294 /* 3295 * For FreeBSD8, the l_pid and l_sysid must be set to the same 3296 * values for all calls, so that all locks will be held by the 3297 * nfsd server. (The nfsd server handles conflicts between the 3298 * various clients.) 3299 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 3300 * bytes, so it can't be put in l_sysid. 3301 */ 3302 if (nfsv4_sysid == 0) 3303 nfsv4_sysid = nlm_acquire_next_sysid(); 3304 fl.l_pid = (pid_t)0; 3305 fl.l_sysid = (int)nfsv4_sysid; 3306 3307 if (ftype == F_UNLCK) 3308 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 3309 (F_POSIX | F_REMOTE)); 3310 else 3311 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 3312 (F_POSIX | F_REMOTE)); 3313 3314 out: 3315 NFSEXITCODE(error); 3316 return (error); 3317 } 3318 3319 /* 3320 * Check the nfsv4 root exports. 3321 */ 3322 int 3323 nfsvno_v4rootexport(struct nfsrv_descript *nd) 3324 { 3325 struct ucred *credanon; 3326 int exflags, error = 0, numsecflavor, *secflavors, i; 3327 3328 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags, 3329 &credanon, &numsecflavor, &secflavors); 3330 if (error) { 3331 error = NFSERR_PROGUNAVAIL; 3332 goto out; 3333 } 3334 if (credanon != NULL) 3335 crfree(credanon); 3336 for (i = 0; i < numsecflavor; i++) { 3337 if (secflavors[i] == AUTH_SYS) 3338 nd->nd_flag |= ND_EXAUTHSYS; 3339 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3340 nd->nd_flag |= ND_EXGSS; 3341 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3342 nd->nd_flag |= ND_EXGSSINTEGRITY; 3343 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3344 nd->nd_flag |= ND_EXGSSPRIVACY; 3345 } 3346 3347 out: 3348 NFSEXITCODE(error); 3349 return (error); 3350 } 3351 3352 /* 3353 * Nfs server pseudo system call for the nfsd's 3354 */ 3355 /* 3356 * MPSAFE 3357 */ 3358 static int 3359 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3360 { 3361 struct file *fp; 3362 struct nfsd_addsock_args sockarg; 3363 struct nfsd_nfsd_args nfsdarg; 3364 struct nfsd_nfsd_oargs onfsdarg; 3365 struct nfsd_pnfsd_args pnfsdarg; 3366 struct vnode *vp, *nvp, *curdvp; 3367 struct pnfsdsfile *pf; 3368 struct nfsdevice *ds, *fds; 3369 cap_rights_t rights; 3370 int buflen, error, ret; 3371 char *buf, *cp, *cp2, *cp3; 3372 char fname[PNFS_FILENAME_LEN + 1]; 3373 3374 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3375 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3376 if (error) 3377 goto out; 3378 /* 3379 * Since we don't know what rights might be required, 3380 * pretend that we need them all. It is better to be too 3381 * careful than too reckless. 3382 */ 3383 error = fget(td, sockarg.sock, 3384 cap_rights_init(&rights, CAP_SOCK_SERVER), &fp); 3385 if (error != 0) 3386 goto out; 3387 if (fp->f_type != DTYPE_SOCKET) { 3388 fdrop(fp, td); 3389 error = EPERM; 3390 goto out; 3391 } 3392 error = nfsrvd_addsock(fp); 3393 fdrop(fp, td); 3394 } else if (uap->flag & NFSSVC_NFSDNFSD) { 3395 if (uap->argp == NULL) { 3396 error = EINVAL; 3397 goto out; 3398 } 3399 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { 3400 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); 3401 if (error == 0) { 3402 nfsdarg.principal = onfsdarg.principal; 3403 nfsdarg.minthreads = onfsdarg.minthreads; 3404 nfsdarg.maxthreads = onfsdarg.maxthreads; 3405 nfsdarg.version = 1; 3406 nfsdarg.addr = NULL; 3407 nfsdarg.addrlen = 0; 3408 nfsdarg.dnshost = NULL; 3409 nfsdarg.dnshostlen = 0; 3410 nfsdarg.dspath = NULL; 3411 nfsdarg.dspathlen = 0; 3412 nfsdarg.mdspath = NULL; 3413 nfsdarg.mdspathlen = 0; 3414 nfsdarg.mirrorcnt = 1; 3415 } 3416 } else 3417 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); 3418 if (error) 3419 goto out; 3420 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && 3421 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && 3422 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && 3423 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && 3424 nfsdarg.mirrorcnt >= 1 && 3425 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && 3426 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && 3427 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { 3428 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" 3429 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, 3430 nfsdarg.dspathlen, nfsdarg.dnshostlen, 3431 nfsdarg.mdspathlen, nfsdarg.mirrorcnt); 3432 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); 3433 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); 3434 if (error != 0) { 3435 free(cp, M_TEMP); 3436 goto out; 3437 } 3438 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ 3439 nfsdarg.addr = cp; 3440 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); 3441 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); 3442 if (error != 0) { 3443 free(nfsdarg.addr, M_TEMP); 3444 free(cp, M_TEMP); 3445 goto out; 3446 } 3447 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ 3448 nfsdarg.dnshost = cp; 3449 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); 3450 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); 3451 if (error != 0) { 3452 free(nfsdarg.addr, M_TEMP); 3453 free(nfsdarg.dnshost, M_TEMP); 3454 free(cp, M_TEMP); 3455 goto out; 3456 } 3457 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ 3458 nfsdarg.dspath = cp; 3459 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); 3460 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); 3461 if (error != 0) { 3462 free(nfsdarg.addr, M_TEMP); 3463 free(nfsdarg.dnshost, M_TEMP); 3464 free(nfsdarg.dspath, M_TEMP); 3465 free(cp, M_TEMP); 3466 goto out; 3467 } 3468 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ 3469 nfsdarg.mdspath = cp; 3470 } else { 3471 nfsdarg.addr = NULL; 3472 nfsdarg.addrlen = 0; 3473 nfsdarg.dnshost = NULL; 3474 nfsdarg.dnshostlen = 0; 3475 nfsdarg.dspath = NULL; 3476 nfsdarg.dspathlen = 0; 3477 nfsdarg.mdspath = NULL; 3478 nfsdarg.mdspathlen = 0; 3479 nfsdarg.mirrorcnt = 1; 3480 } 3481 error = nfsrvd_nfsd(td, &nfsdarg); 3482 free(nfsdarg.addr, M_TEMP); 3483 free(nfsdarg.dnshost, M_TEMP); 3484 free(nfsdarg.dspath, M_TEMP); 3485 free(nfsdarg.mdspath, M_TEMP); 3486 } else if (uap->flag & NFSSVC_PNFSDS) { 3487 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); 3488 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || 3489 pnfsdarg.op == PNFSDOP_FORCEDELDS)) { 3490 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3491 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, 3492 NULL); 3493 if (error == 0) 3494 error = nfsrv_deldsserver(pnfsdarg.op, cp, td); 3495 free(cp, M_TEMP); 3496 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { 3497 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3498 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; 3499 buf = malloc(buflen, M_TEMP, M_WAITOK); 3500 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, 3501 NULL); 3502 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); 3503 if (error == 0 && pnfsdarg.dspath != NULL) { 3504 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3505 error = copyinstr(pnfsdarg.dspath, cp2, 3506 PATH_MAX + 1, NULL); 3507 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", 3508 error); 3509 } else 3510 cp2 = NULL; 3511 if (error == 0 && pnfsdarg.curdspath != NULL) { 3512 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3513 error = copyinstr(pnfsdarg.curdspath, cp3, 3514 PATH_MAX + 1, NULL); 3515 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", 3516 error); 3517 } else 3518 cp3 = NULL; 3519 curdvp = NULL; 3520 fds = NULL; 3521 if (error == 0) 3522 error = nfsrv_mdscopymr(cp, cp2, cp3, buf, 3523 &buflen, fname, td, &vp, &nvp, &pf, &ds, 3524 &fds); 3525 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); 3526 if (error == 0) { 3527 if (pf->dsf_dir >= nfsrv_dsdirsize) { 3528 printf("copymr: dsdir out of range\n"); 3529 pf->dsf_dir = 0; 3530 } 3531 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); 3532 error = nfsrv_copymr(vp, nvp, 3533 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, 3534 (struct pnfsdsfile *)buf, 3535 buflen / sizeof(*pf), td->td_ucred, td); 3536 vput(vp); 3537 vput(nvp); 3538 if (fds != NULL && error == 0) { 3539 curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; 3540 ret = vn_lock(curdvp, LK_EXCLUSIVE); 3541 if (ret == 0) { 3542 nfsrv_dsremove(curdvp, fname, 3543 td->td_ucred, td); 3544 NFSVOPUNLOCK(curdvp, 0); 3545 } 3546 } 3547 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); 3548 } 3549 free(cp, M_TEMP); 3550 free(cp2, M_TEMP); 3551 free(cp3, M_TEMP); 3552 free(buf, M_TEMP); 3553 } 3554 } else { 3555 error = nfssvc_srvcall(td, uap, td->td_ucred); 3556 } 3557 3558 out: 3559 NFSEXITCODE(error); 3560 return (error); 3561 } 3562 3563 static int 3564 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 3565 { 3566 struct nfsex_args export; 3567 struct file *fp = NULL; 3568 int stablefd, len; 3569 struct nfsd_clid adminrevoke; 3570 struct nfsd_dumplist dumplist; 3571 struct nfsd_dumpclients *dumpclients; 3572 struct nfsd_dumplocklist dumplocklist; 3573 struct nfsd_dumplocks *dumplocks; 3574 struct nameidata nd; 3575 vnode_t vp; 3576 int error = EINVAL, igotlock; 3577 struct proc *procp; 3578 static int suspend_nfsd = 0; 3579 3580 if (uap->flag & NFSSVC_PUBLICFH) { 3581 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 3582 sizeof (fhandle_t)); 3583 error = copyin(uap->argp, 3584 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 3585 if (!error) 3586 nfs_pubfhset = 1; 3587 } else if (uap->flag & NFSSVC_V4ROOTEXPORT) { 3588 error = copyin(uap->argp,(caddr_t)&export, 3589 sizeof (struct nfsex_args)); 3590 if (!error) 3591 error = nfsrv_v4rootexport(&export, cred, p); 3592 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 3593 nfs_pubfhset = 0; 3594 error = 0; 3595 } else if (uap->flag & NFSSVC_STABLERESTART) { 3596 error = copyin(uap->argp, (caddr_t)&stablefd, 3597 sizeof (int)); 3598 if (!error) 3599 error = fp_getfvp(p, stablefd, &fp, &vp); 3600 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 3601 error = EBADF; 3602 if (!error && newnfs_numnfsd != 0) 3603 error = EPERM; 3604 if (!error) { 3605 nfsrv_stablefirst.nsf_fp = fp; 3606 nfsrv_setupstable(p); 3607 } 3608 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 3609 error = copyin(uap->argp, (caddr_t)&adminrevoke, 3610 sizeof (struct nfsd_clid)); 3611 if (!error) 3612 error = nfsrv_adminrevoke(&adminrevoke, p); 3613 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 3614 error = copyin(uap->argp, (caddr_t)&dumplist, 3615 sizeof (struct nfsd_dumplist)); 3616 if (!error && (dumplist.ndl_size < 1 || 3617 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 3618 error = EPERM; 3619 if (!error) { 3620 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 3621 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 3622 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 3623 error = copyout(dumpclients, 3624 CAST_USER_ADDR_T(dumplist.ndl_list), len); 3625 free(dumpclients, M_TEMP); 3626 } 3627 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 3628 error = copyin(uap->argp, (caddr_t)&dumplocklist, 3629 sizeof (struct nfsd_dumplocklist)); 3630 if (!error && (dumplocklist.ndllck_size < 1 || 3631 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 3632 error = EPERM; 3633 if (!error) 3634 error = nfsrv_lookupfilename(&nd, 3635 dumplocklist.ndllck_fname, p); 3636 if (!error) { 3637 len = sizeof (struct nfsd_dumplocks) * 3638 dumplocklist.ndllck_size; 3639 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 3640 nfsrv_dumplocks(nd.ni_vp, dumplocks, 3641 dumplocklist.ndllck_size, p); 3642 vput(nd.ni_vp); 3643 error = copyout(dumplocks, 3644 CAST_USER_ADDR_T(dumplocklist.ndllck_list), len); 3645 free(dumplocks, M_TEMP); 3646 } 3647 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 3648 procp = p->td_proc; 3649 PROC_LOCK(procp); 3650 nfsd_master_pid = procp->p_pid; 3651 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 3652 nfsd_master_start = procp->p_stats->p_start; 3653 nfsd_master_proc = procp; 3654 PROC_UNLOCK(procp); 3655 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { 3656 NFSLOCKV4ROOTMUTEX(); 3657 if (suspend_nfsd == 0) { 3658 /* Lock out all nfsd threads */ 3659 do { 3660 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1, 3661 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 3662 } while (igotlock == 0 && suspend_nfsd == 0); 3663 suspend_nfsd = 1; 3664 } 3665 NFSUNLOCKV4ROOTMUTEX(); 3666 error = 0; 3667 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { 3668 NFSLOCKV4ROOTMUTEX(); 3669 if (suspend_nfsd != 0) { 3670 nfsv4_unlock(&nfsd_suspend_lock, 0); 3671 suspend_nfsd = 0; 3672 } 3673 NFSUNLOCKV4ROOTMUTEX(); 3674 error = 0; 3675 } 3676 3677 NFSEXITCODE(error); 3678 return (error); 3679 } 3680 3681 /* 3682 * Check exports. 3683 * Returns 0 if ok, 1 otherwise. 3684 */ 3685 int 3686 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 3687 { 3688 int i; 3689 3690 /* 3691 * This seems odd, but allow the case where the security flavor 3692 * list is empty. This happens when NFSv4 is traversing non-exported 3693 * file systems. Exported file systems should always have a non-empty 3694 * security flavor list. 3695 */ 3696 if (exp->nes_numsecflavor == 0) 3697 return (0); 3698 3699 for (i = 0; i < exp->nes_numsecflavor; i++) { 3700 /* 3701 * The tests for privacy and integrity must be first, 3702 * since ND_GSS is set for everything but AUTH_SYS. 3703 */ 3704 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 3705 (nd->nd_flag & ND_GSSPRIVACY)) 3706 return (0); 3707 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 3708 (nd->nd_flag & ND_GSSINTEGRITY)) 3709 return (0); 3710 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 3711 (nd->nd_flag & ND_GSS)) 3712 return (0); 3713 if (exp->nes_secflavors[i] == AUTH_SYS && 3714 (nd->nd_flag & ND_GSS) == 0) 3715 return (0); 3716 } 3717 return (1); 3718 } 3719 3720 /* 3721 * Calculate a hash value for the fid in a file handle. 3722 */ 3723 uint32_t 3724 nfsrv_hashfh(fhandle_t *fhp) 3725 { 3726 uint32_t hashval; 3727 3728 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 3729 return (hashval); 3730 } 3731 3732 /* 3733 * Calculate a hash value for the sessionid. 3734 */ 3735 uint32_t 3736 nfsrv_hashsessionid(uint8_t *sessionid) 3737 { 3738 uint32_t hashval; 3739 3740 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); 3741 return (hashval); 3742 } 3743 3744 /* 3745 * Signal the userland master nfsd to backup the stable restart file. 3746 */ 3747 void 3748 nfsrv_backupstable(void) 3749 { 3750 struct proc *procp; 3751 3752 if (nfsd_master_proc != NULL) { 3753 procp = pfind(nfsd_master_pid); 3754 /* Try to make sure it is the correct process. */ 3755 if (procp == nfsd_master_proc && 3756 procp->p_stats->p_start.tv_sec == 3757 nfsd_master_start.tv_sec && 3758 procp->p_stats->p_start.tv_usec == 3759 nfsd_master_start.tv_usec && 3760 strcmp(procp->p_comm, nfsd_master_comm) == 0) 3761 kern_psignal(procp, SIGUSR2); 3762 else 3763 nfsd_master_proc = NULL; 3764 3765 if (procp != NULL) 3766 PROC_UNLOCK(procp); 3767 } 3768 } 3769 3770 /* 3771 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. 3772 * The arguments are in a structure, so that they can be passed through 3773 * taskqueue for a kernel process to execute this function. 3774 */ 3775 struct nfsrvdscreate { 3776 int done; 3777 int inprog; 3778 struct task tsk; 3779 struct ucred *tcred; 3780 struct vnode *dvp; 3781 NFSPROC_T *p; 3782 struct pnfsdsfile *pf; 3783 int err; 3784 fhandle_t fh; 3785 struct vattr va; 3786 struct vattr createva; 3787 }; 3788 3789 int 3790 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, 3791 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, 3792 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) 3793 { 3794 struct vnode *nvp; 3795 struct nameidata named; 3796 struct vattr va; 3797 char *bufp; 3798 u_long *hashp; 3799 struct nfsnode *np; 3800 struct nfsmount *nmp; 3801 int error; 3802 3803 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, 3804 LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE); 3805 nfsvno_setpathbuf(&named, &bufp, &hashp); 3806 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 3807 named.ni_cnd.cn_thread = p; 3808 named.ni_cnd.cn_nameptr = bufp; 3809 if (fnamep != NULL) { 3810 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); 3811 named.ni_cnd.cn_namelen = strlen(bufp); 3812 } else 3813 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); 3814 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); 3815 3816 /* Create the date file in the DS mount. */ 3817 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 3818 if (error == 0) { 3819 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); 3820 NFSVOPUNLOCK(dvp, 0); 3821 if (error == 0) { 3822 /* Set the ownership of the file. */ 3823 error = VOP_SETATTR(nvp, nvap, tcred); 3824 NFSD_DEBUG(4, "nfsrv_dscreate:" 3825 " setattr-uid=%d\n", error); 3826 if (error != 0) 3827 vput(nvp); 3828 } 3829 if (error != 0) 3830 printf("pNFS: pnfscreate failed=%d\n", error); 3831 } else 3832 printf("pNFS: pnfscreate vnlock=%d\n", error); 3833 if (error == 0) { 3834 np = VTONFS(nvp); 3835 nmp = VFSTONFS(nvp->v_mount); 3836 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") 3837 != 0 || nmp->nm_nam->sa_len > sizeof( 3838 struct sockaddr_in6) || 3839 np->n_fhp->nfh_len != NFSX_MYFH) { 3840 printf("Bad DS file: fstype=%s salen=%d" 3841 " fhlen=%d\n", 3842 nvp->v_mount->mnt_vfc->vfc_name, 3843 nmp->nm_nam->sa_len, np->n_fhp->nfh_len); 3844 error = ENOENT; 3845 } 3846 3847 /* Set extattrs for the DS on the MDS file. */ 3848 if (error == 0) { 3849 if (dsa != NULL) { 3850 error = VOP_GETATTR(nvp, &va, tcred); 3851 if (error == 0) { 3852 dsa->dsa_filerev = va.va_filerev; 3853 dsa->dsa_size = va.va_size; 3854 dsa->dsa_atime = va.va_atime; 3855 dsa->dsa_mtime = va.va_mtime; 3856 } 3857 } 3858 if (error == 0) { 3859 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, 3860 NFSX_MYFH); 3861 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, 3862 nmp->nm_nam->sa_len); 3863 NFSBCOPY(named.ni_cnd.cn_nameptr, 3864 pf->dsf_filename, 3865 sizeof(pf->dsf_filename)); 3866 } 3867 } else 3868 printf("pNFS: pnfscreate can't get DS" 3869 " attr=%d\n", error); 3870 if (nvpp != NULL && error == 0) 3871 *nvpp = nvp; 3872 else 3873 vput(nvp); 3874 } 3875 nfsvno_relpathbuf(&named); 3876 return (error); 3877 } 3878 3879 /* 3880 * Start up the thread that will execute nfsrv_dscreate(). 3881 */ 3882 static void 3883 start_dscreate(void *arg, int pending) 3884 { 3885 struct nfsrvdscreate *dsc; 3886 3887 dsc = (struct nfsrvdscreate *)arg; 3888 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, 3889 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); 3890 dsc->done = 1; 3891 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); 3892 } 3893 3894 /* 3895 * Create a pNFS data file on the Data Server(s). 3896 */ 3897 static void 3898 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, 3899 NFSPROC_T *p) 3900 { 3901 struct nfsrvdscreate *dsc, *tdsc; 3902 struct nfsdevice *ds, *tds, *fds; 3903 struct mount *mp; 3904 struct pnfsdsfile *pf, *tpf; 3905 struct pnfsdsattr dsattr; 3906 struct vattr va; 3907 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 3908 struct nfsmount *nmp; 3909 fhandle_t fh; 3910 uid_t vauid; 3911 gid_t vagid; 3912 u_short vamode; 3913 struct ucred *tcred; 3914 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; 3915 int failpos, timo; 3916 3917 /* Get a DS server directory in a round-robin order. */ 3918 mirrorcnt = 1; 3919 mp = vp->v_mount; 3920 ds = fds = NULL; 3921 NFSDDSLOCK(); 3922 /* 3923 * Search for the first entry that handles this MDS fs, but use the 3924 * first entry for all MDS fs's otherwise. 3925 */ 3926 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { 3927 if (tds->nfsdev_nmp != NULL) { 3928 if (tds->nfsdev_mdsisset == 0 && ds == NULL) 3929 ds = tds; 3930 else if (tds->nfsdev_mdsisset != 0 && 3931 mp->mnt_stat.f_fsid.val[0] == 3932 tds->nfsdev_mdsfsid.val[0] && 3933 mp->mnt_stat.f_fsid.val[1] == 3934 tds->nfsdev_mdsfsid.val[1]) { 3935 ds = fds = tds; 3936 break; 3937 } 3938 } 3939 } 3940 if (ds == NULL) { 3941 NFSDDSUNLOCK(); 3942 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); 3943 return; 3944 } 3945 i = dsdir[0] = ds->nfsdev_nextdir; 3946 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; 3947 dvp[0] = ds->nfsdev_dsdir[i]; 3948 tds = TAILQ_NEXT(ds, nfsdev_list); 3949 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { 3950 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { 3951 if (tds->nfsdev_nmp != NULL && 3952 ((tds->nfsdev_mdsisset == 0 && fds == NULL) || 3953 (tds->nfsdev_mdsisset != 0 && fds != NULL && 3954 mp->mnt_stat.f_fsid.val[0] == 3955 tds->nfsdev_mdsfsid.val[0] && 3956 mp->mnt_stat.f_fsid.val[1] == 3957 tds->nfsdev_mdsfsid.val[1]))) { 3958 dsdir[mirrorcnt] = i; 3959 dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; 3960 mirrorcnt++; 3961 if (mirrorcnt >= nfsrv_maxpnfsmirror) 3962 break; 3963 } 3964 } 3965 } 3966 /* Put at end of list to implement round-robin usage. */ 3967 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 3968 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 3969 NFSDDSUNLOCK(); 3970 dsc = NULL; 3971 if (mirrorcnt > 1) 3972 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, 3973 M_WAITOK | M_ZERO); 3974 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | 3975 M_ZERO); 3976 3977 error = nfsvno_getfh(vp, &fh, p); 3978 if (error == 0) 3979 error = VOP_GETATTR(vp, &va, cred); 3980 if (error == 0) { 3981 /* Set the attributes for "vp" to Setattr the DS vp. */ 3982 vauid = va.va_uid; 3983 vagid = va.va_gid; 3984 vamode = va.va_mode; 3985 VATTR_NULL(&va); 3986 va.va_uid = vauid; 3987 va.va_gid = vagid; 3988 va.va_mode = vamode; 3989 va.va_size = 0; 3990 } else 3991 printf("pNFS: pnfscreate getfh+attr=%d\n", error); 3992 3993 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, 3994 cred->cr_gid); 3995 /* Make data file name based on FH. */ 3996 tcred = newnfs_getcred(); 3997 3998 /* 3999 * Create the file on each DS mirror, using kernel process(es) for the 4000 * additional mirrors. 4001 */ 4002 failpos = -1; 4003 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { 4004 tpf->dsf_dir = dsdir[i]; 4005 tdsc->tcred = tcred; 4006 tdsc->p = p; 4007 tdsc->pf = tpf; 4008 tdsc->createva = *vap; 4009 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); 4010 tdsc->va = va; 4011 tdsc->dvp = dvp[i]; 4012 tdsc->done = 0; 4013 tdsc->inprog = 0; 4014 tdsc->err = 0; 4015 ret = EIO; 4016 if (nfs_pnfsiothreads != 0) { 4017 ret = nfs_pnfsio(start_dscreate, tdsc); 4018 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); 4019 } 4020 if (ret != 0) { 4021 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, 4022 NULL, tcred, p, NULL); 4023 if (ret != 0) { 4024 KASSERT(error == 0, ("nfsrv_dscreate err=%d", 4025 error)); 4026 if (failpos == -1 && nfsds_failerr(ret)) 4027 failpos = i; 4028 else 4029 error = ret; 4030 } 4031 } 4032 } 4033 if (error == 0) { 4034 tpf->dsf_dir = dsdir[mirrorcnt - 1]; 4035 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, 4036 &dsattr, NULL, tcred, p, NULL); 4037 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { 4038 failpos = mirrorcnt - 1; 4039 error = 0; 4040 } 4041 } 4042 timo = hz / 50; /* Wait for 20msec. */ 4043 if (timo < 1) 4044 timo = 1; 4045 /* Wait for kernel task(s) to complete. */ 4046 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { 4047 while (tdsc->inprog != 0 && tdsc->done == 0) 4048 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); 4049 if (tdsc->err != 0) { 4050 if (failpos == -1 && nfsds_failerr(tdsc->err)) 4051 failpos = i; 4052 else if (error == 0) 4053 error = tdsc->err; 4054 } 4055 } 4056 4057 /* 4058 * If failpos has been set, that mirror has failed, so it needs 4059 * to be disabled. 4060 */ 4061 if (failpos >= 0) { 4062 nmp = VFSTONFS(dvp[failpos]->v_mount); 4063 NFSLOCKMNT(nmp); 4064 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4065 NFSMNTP_CANCELRPCS)) == 0) { 4066 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4067 NFSUNLOCKMNT(nmp); 4068 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4069 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, 4070 ds); 4071 if (ds != NULL) 4072 nfsrv_killrpcs(nmp); 4073 NFSLOCKMNT(nmp); 4074 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4075 wakeup(nmp); 4076 } 4077 NFSUNLOCKMNT(nmp); 4078 } 4079 4080 NFSFREECRED(tcred); 4081 if (error == 0) { 4082 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); 4083 4084 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", 4085 mirrorcnt, nfsrv_maxpnfsmirror); 4086 /* 4087 * For all mirrors that couldn't be created, fill in the 4088 * *pf structure, but with an IP address == 0.0.0.0. 4089 */ 4090 tpf = pf + mirrorcnt; 4091 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { 4092 *tpf = *pf; 4093 tpf->dsf_sin.sin_family = AF_INET; 4094 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); 4095 tpf->dsf_sin.sin_addr.s_addr = 0; 4096 tpf->dsf_sin.sin_port = 0; 4097 } 4098 4099 error = vn_extattr_set(vp, IO_NODELOCKED, 4100 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 4101 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); 4102 if (error == 0) 4103 error = vn_extattr_set(vp, IO_NODELOCKED, 4104 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", 4105 sizeof(dsattr), (char *)&dsattr, p); 4106 if (error != 0) 4107 printf("pNFS: pnfscreate setextattr=%d\n", 4108 error); 4109 } else 4110 printf("pNFS: pnfscreate=%d\n", error); 4111 free(pf, M_TEMP); 4112 free(dsc, M_TEMP); 4113 } 4114 4115 /* 4116 * Get the information needed to remove the pNFS Data Server file from the 4117 * Metadata file. Upon success, ddvp is set non-NULL to the locked 4118 * DS directory vnode. The caller must unlock *ddvp when done with it. 4119 */ 4120 static void 4121 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, 4122 int *mirrorcntp, char *fname, fhandle_t *fhp) 4123 { 4124 struct vattr va; 4125 struct ucred *tcred; 4126 char *buf; 4127 int buflen, error; 4128 4129 dvpp[0] = NULL; 4130 /* If not an exported regular file or not a pNFS server, just return. */ 4131 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4132 nfsrv_devidcnt == 0) 4133 return; 4134 4135 /* Check to see if this is the last hard link. */ 4136 tcred = newnfs_getcred(); 4137 error = VOP_GETATTR(vp, &va, tcred); 4138 NFSFREECRED(tcred); 4139 if (error != 0) { 4140 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); 4141 return; 4142 } 4143 if (va.va_nlink > 1) 4144 return; 4145 4146 error = nfsvno_getfh(vp, fhp, p); 4147 if (error != 0) { 4148 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); 4149 return; 4150 } 4151 4152 buflen = 1024; 4153 buf = malloc(buflen, M_TEMP, M_WAITOK); 4154 /* Get the directory vnode for the DS mount and the file handle. */ 4155 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, 4156 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); 4157 free(buf, M_TEMP); 4158 if (error != 0) 4159 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); 4160 } 4161 4162 /* 4163 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. 4164 * The arguments are in a structure, so that they can be passed through 4165 * taskqueue for a kernel process to execute this function. 4166 */ 4167 struct nfsrvdsremove { 4168 int done; 4169 int inprog; 4170 struct task tsk; 4171 struct ucred *tcred; 4172 struct vnode *dvp; 4173 NFSPROC_T *p; 4174 int err; 4175 char fname[PNFS_FILENAME_LEN + 1]; 4176 }; 4177 4178 static int 4179 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, 4180 NFSPROC_T *p) 4181 { 4182 struct nameidata named; 4183 struct vnode *nvp; 4184 char *bufp; 4185 u_long *hashp; 4186 int error; 4187 4188 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4189 if (error != 0) 4190 return (error); 4191 named.ni_cnd.cn_nameiop = DELETE; 4192 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 4193 named.ni_cnd.cn_cred = tcred; 4194 named.ni_cnd.cn_thread = p; 4195 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; 4196 nfsvno_setpathbuf(&named, &bufp, &hashp); 4197 named.ni_cnd.cn_nameptr = bufp; 4198 named.ni_cnd.cn_namelen = strlen(fname); 4199 strlcpy(bufp, fname, NAME_MAX); 4200 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); 4201 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 4202 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); 4203 if (error == 0) { 4204 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); 4205 vput(nvp); 4206 } 4207 NFSVOPUNLOCK(dvp, 0); 4208 nfsvno_relpathbuf(&named); 4209 if (error != 0) 4210 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); 4211 return (error); 4212 } 4213 4214 /* 4215 * Start up the thread that will execute nfsrv_dsremove(). 4216 */ 4217 static void 4218 start_dsremove(void *arg, int pending) 4219 { 4220 struct nfsrvdsremove *dsrm; 4221 4222 dsrm = (struct nfsrvdsremove *)arg; 4223 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, 4224 dsrm->p); 4225 dsrm->done = 1; 4226 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); 4227 } 4228 4229 /* 4230 * Remove a pNFS data file from a Data Server. 4231 * nfsrv_pnfsremovesetup() must have been called before the MDS file was 4232 * removed to set up the dvp and fill in the FH. 4233 */ 4234 static void 4235 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, 4236 NFSPROC_T *p) 4237 { 4238 struct ucred *tcred; 4239 struct nfsrvdsremove *dsrm, *tdsrm; 4240 struct nfsdevice *ds; 4241 struct nfsmount *nmp; 4242 int failpos, i, ret, timo; 4243 4244 tcred = newnfs_getcred(); 4245 dsrm = NULL; 4246 if (mirrorcnt > 1) 4247 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); 4248 /* 4249 * Remove the file on each DS mirror, using kernel process(es) for the 4250 * additional mirrors. 4251 */ 4252 failpos = -1; 4253 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4254 tdsrm->tcred = tcred; 4255 tdsrm->p = p; 4256 tdsrm->dvp = dvp[i]; 4257 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); 4258 tdsrm->inprog = 0; 4259 tdsrm->done = 0; 4260 tdsrm->err = 0; 4261 ret = EIO; 4262 if (nfs_pnfsiothreads != 0) { 4263 ret = nfs_pnfsio(start_dsremove, tdsrm); 4264 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); 4265 } 4266 if (ret != 0) { 4267 ret = nfsrv_dsremove(dvp[i], fname, tcred, p); 4268 if (failpos == -1 && nfsds_failerr(ret)) 4269 failpos = i; 4270 } 4271 } 4272 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); 4273 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) 4274 failpos = mirrorcnt - 1; 4275 timo = hz / 50; /* Wait for 20msec. */ 4276 if (timo < 1) 4277 timo = 1; 4278 /* Wait for kernel task(s) to complete. */ 4279 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4280 while (tdsrm->inprog != 0 && tdsrm->done == 0) 4281 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); 4282 if (failpos == -1 && nfsds_failerr(tdsrm->err)) 4283 failpos = i; 4284 } 4285 4286 /* 4287 * If failpos has been set, that mirror has failed, so it needs 4288 * to be disabled. 4289 */ 4290 if (failpos >= 0) { 4291 nmp = VFSTONFS(dvp[failpos]->v_mount); 4292 NFSLOCKMNT(nmp); 4293 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4294 NFSMNTP_CANCELRPCS)) == 0) { 4295 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4296 NFSUNLOCKMNT(nmp); 4297 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4298 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, 4299 ds); 4300 if (ds != NULL) 4301 nfsrv_killrpcs(nmp); 4302 NFSLOCKMNT(nmp); 4303 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4304 wakeup(nmp); 4305 } 4306 NFSUNLOCKMNT(nmp); 4307 } 4308 4309 /* Get rid all layouts for the file. */ 4310 nfsrv_freefilelayouts(fhp); 4311 4312 NFSFREECRED(tcred); 4313 free(dsrm, M_TEMP); 4314 } 4315 4316 /* 4317 * Generate a file name based on the file handle and put it in *bufp. 4318 * Return the number of bytes generated. 4319 */ 4320 static int 4321 nfsrv_putfhname(fhandle_t *fhp, char *bufp) 4322 { 4323 int i; 4324 uint8_t *cp; 4325 const uint8_t *hexdigits = "0123456789abcdef"; 4326 4327 cp = (uint8_t *)fhp; 4328 for (i = 0; i < sizeof(*fhp); i++) { 4329 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; 4330 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; 4331 } 4332 bufp[2 * i] = '\0'; 4333 return (2 * i); 4334 } 4335 4336 /* 4337 * Update the Metadata file's attributes from the DS file when a Read/Write 4338 * layout is returned. 4339 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN 4340 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. 4341 */ 4342 int 4343 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 4344 { 4345 struct ucred *tcred; 4346 int error; 4347 4348 /* Do this as root so that it won't fail with EACCES. */ 4349 tcred = newnfs_getcred(); 4350 error = nfsrv_proxyds(NULL, vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, 4351 NULL, NULL, NULL, nap, NULL); 4352 NFSFREECRED(tcred); 4353 return (error); 4354 } 4355 4356 /* 4357 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. 4358 */ 4359 static int 4360 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, 4361 NFSPROC_T *p) 4362 { 4363 int error; 4364 4365 error = nfsrv_proxyds(NULL, vp, 0, 0, cred, p, NFSPROC_SETACL, 4366 NULL, NULL, NULL, NULL, aclp); 4367 return (error); 4368 } 4369 4370 static int 4371 nfsrv_proxyds(struct nfsrv_descript *nd, struct vnode *vp, off_t off, int cnt, 4372 struct ucred *cred, struct thread *p, int ioproc, struct mbuf **mpp, 4373 char *cp, struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp) 4374 { 4375 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; 4376 fhandle_t fh[NFSDEV_MAXMIRRORS]; 4377 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4378 struct nfsdevice *ds; 4379 struct pnfsdsattr dsattr; 4380 char *buf; 4381 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; 4382 4383 NFSD_DEBUG(4, "in nfsrv_proxyds\n"); 4384 /* 4385 * If not a regular file, not exported or not a pNFS server, 4386 * just return ENOENT. 4387 */ 4388 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4389 nfsrv_devidcnt == 0) 4390 return (ENOENT); 4391 4392 buflen = 1024; 4393 buf = malloc(buflen, M_TEMP, M_WAITOK); 4394 error = 0; 4395 4396 /* 4397 * For Getattr, get the Change attribute (va_filerev) and size (va_size) 4398 * from the MetaData file's extended attribute. 4399 */ 4400 if (ioproc == NFSPROC_GETATTR) { 4401 error = vn_extattr_get(vp, IO_NODELOCKED, 4402 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, 4403 p); 4404 if (error == 0 && buflen != sizeof(dsattr)) 4405 error = ENXIO; 4406 if (error == 0) { 4407 NFSBCOPY(buf, &dsattr, buflen); 4408 nap->na_filerev = dsattr.dsa_filerev; 4409 nap->na_size = dsattr.dsa_size; 4410 nap->na_atime = dsattr.dsa_atime; 4411 nap->na_mtime = dsattr.dsa_mtime; 4412 4413 /* 4414 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() 4415 * returns 0, just return now. nfsrv_checkdsattr() 4416 * returns 0 if there is no Read/Write layout 4417 * plus either an Open/Write_access or Write 4418 * delegation issued to a client for the file. 4419 */ 4420 if (nfsrv_pnfsgetdsattr == 0 || 4421 nfsrv_checkdsattr(nd, vp, p) == 0) { 4422 free(buf, M_TEMP); 4423 return (error); 4424 } 4425 } 4426 4427 /* 4428 * Clear ENOATTR so the code below will attempt to do a 4429 * nfsrv_getattrdsrpc() to get the attributes and (re)create 4430 * the extended attribute. 4431 */ 4432 if (error == ENOATTR) 4433 error = 0; 4434 } 4435 4436 origmircnt = -1; 4437 trycnt = 0; 4438 tryagain: 4439 if (error == 0) { 4440 buflen = 1024; 4441 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == 4442 LK_EXCLUSIVE) 4443 printf("nfsrv_proxyds: Readds vp exclusively locked\n"); 4444 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, 4445 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, 4446 NULL, NULL); 4447 if (error == 0) { 4448 for (i = 0; i < mirrorcnt; i++) 4449 nmp[i] = VFSTONFS(dvp[i]->v_mount); 4450 } else 4451 printf("pNFS: proxy getextattr sockaddr=%d\n", error); 4452 } else 4453 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); 4454 if (error == 0) { 4455 failpos = -1; 4456 if (origmircnt == -1) 4457 origmircnt = mirrorcnt; 4458 /* 4459 * If failpos is set to a mirror#, then that mirror has 4460 * failed and will be disabled. For Read and Getattr, the 4461 * function only tries one mirror, so if that mirror has 4462 * failed, it will need to be retried. As such, increment 4463 * tryitagain for these cases. 4464 * For Write, Setattr and Setacl, the function tries all 4465 * mirrors and will not return an error for the case where 4466 * one mirror has failed. For these cases, the functioning 4467 * mirror(s) will have been modified, so a retry isn't 4468 * necessary. These functions will set failpos for the 4469 * failed mirror#. 4470 */ 4471 if (ioproc == NFSPROC_READDS) { 4472 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], 4473 mpp, mpp2); 4474 if (nfsds_failerr(error) && mirrorcnt > 1) { 4475 /* 4476 * Setting failpos will cause the mirror 4477 * to be disabled and then a retry of this 4478 * read is required. 4479 */ 4480 failpos = 0; 4481 error = 0; 4482 trycnt++; 4483 } 4484 } else if (ioproc == NFSPROC_WRITEDS) 4485 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, 4486 &nmp[0], mirrorcnt, mpp, cp, &failpos); 4487 else if (ioproc == NFSPROC_SETATTR) 4488 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], 4489 mirrorcnt, nap, &failpos); 4490 else if (ioproc == NFSPROC_SETACL) 4491 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], 4492 mirrorcnt, aclp, &failpos); 4493 else { 4494 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, 4495 vp, nmp[mirrorcnt - 1], nap); 4496 if (nfsds_failerr(error) && mirrorcnt > 1) { 4497 /* 4498 * Setting failpos will cause the mirror 4499 * to be disabled and then a retry of this 4500 * getattr is required. 4501 */ 4502 failpos = mirrorcnt - 1; 4503 error = 0; 4504 trycnt++; 4505 } 4506 } 4507 ds = NULL; 4508 if (failpos >= 0) { 4509 failnmp = nmp[failpos]; 4510 NFSLOCKMNT(failnmp); 4511 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | 4512 NFSMNTP_CANCELRPCS)) == 0) { 4513 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4514 NFSUNLOCKMNT(failnmp); 4515 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, 4516 failnmp, p); 4517 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", 4518 failpos, ds); 4519 if (ds != NULL) 4520 nfsrv_killrpcs(failnmp); 4521 NFSLOCKMNT(failnmp); 4522 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4523 wakeup(failnmp); 4524 } 4525 NFSUNLOCKMNT(failnmp); 4526 } 4527 for (i = 0; i < mirrorcnt; i++) 4528 NFSVOPUNLOCK(dvp[i], 0); 4529 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, 4530 trycnt); 4531 /* Try the Read/Getattr again if a mirror was deleted. */ 4532 if (ds != NULL && trycnt > 0 && trycnt < origmircnt) 4533 goto tryagain; 4534 } else { 4535 /* Return ENOENT for any Extended Attribute error. */ 4536 error = ENOENT; 4537 } 4538 free(buf, M_TEMP); 4539 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); 4540 return (error); 4541 } 4542 4543 /* 4544 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended 4545 * attribute. 4546 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs 4547 * to be checked. If it points to a NULL nmp, then it returns 4548 * a suitable destination. 4549 * curnmp - If non-NULL, it is the source mount for the copy. 4550 */ 4551 int 4552 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, 4553 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, 4554 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, 4555 struct nfsmount *curnmp, int *ippos, int *dsdirp) 4556 { 4557 struct vnode *dvp, *nvp, **tdvpp; 4558 struct mount *mp; 4559 struct nfsmount *nmp, *newnmp; 4560 struct sockaddr *sad; 4561 struct sockaddr_in *sin; 4562 struct nfsdevice *ds, *tds, *fndds; 4563 struct pnfsdsfile *pf; 4564 uint32_t dsdir; 4565 int error, fhiszero, fnd, gotone, i, mirrorcnt; 4566 4567 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); 4568 *mirrorcntp = 1; 4569 tdvpp = dvpp; 4570 if (nvpp != NULL) 4571 *nvpp = NULL; 4572 if (dvpp != NULL) 4573 *dvpp = NULL; 4574 if (ippos != NULL) 4575 *ippos = -1; 4576 if (newnmpp != NULL) 4577 newnmp = *newnmpp; 4578 else 4579 newnmp = NULL; 4580 mp = vp->v_mount; 4581 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 4582 "pnfsd.dsfile", buflenp, buf, p); 4583 mirrorcnt = *buflenp / sizeof(*pf); 4584 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || 4585 *buflenp != sizeof(*pf) * mirrorcnt)) 4586 error = ENOATTR; 4587 4588 pf = (struct pnfsdsfile *)buf; 4589 /* If curnmp != NULL, check for a match in the mirror list. */ 4590 if (curnmp != NULL && error == 0) { 4591 fnd = 0; 4592 for (i = 0; i < mirrorcnt; i++, pf++) { 4593 sad = (struct sockaddr *)&pf->dsf_sin; 4594 if (nfsaddr2_match(sad, curnmp->nm_nam)) { 4595 if (ippos != NULL) 4596 *ippos = i; 4597 fnd = 1; 4598 break; 4599 } 4600 } 4601 if (fnd == 0) 4602 error = ENXIO; 4603 } 4604 4605 gotone = 0; 4606 pf = (struct pnfsdsfile *)buf; 4607 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, 4608 error); 4609 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { 4610 fhiszero = 0; 4611 sad = (struct sockaddr *)&pf->dsf_sin; 4612 sin = &pf->dsf_sin; 4613 dsdir = pf->dsf_dir; 4614 if (dsdir >= nfsrv_dsdirsize) { 4615 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); 4616 error = ENOATTR; 4617 } else if (nvpp != NULL && newnmp != NULL && 4618 nfsaddr2_match(sad, newnmp->nm_nam)) 4619 error = EEXIST; 4620 if (error == 0) { 4621 if (ippos != NULL && curnmp == NULL && 4622 sad->sa_family == AF_INET && 4623 sin->sin_addr.s_addr == 0) 4624 *ippos = i; 4625 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) 4626 fhiszero = 1; 4627 /* Use the socket address to find the mount point. */ 4628 fndds = NULL; 4629 NFSDDSLOCK(); 4630 /* Find a match for the IP address. */ 4631 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 4632 if (ds->nfsdev_nmp != NULL) { 4633 dvp = ds->nfsdev_dvp; 4634 nmp = VFSTONFS(dvp->v_mount); 4635 if (nmp != ds->nfsdev_nmp) 4636 printf("different2 nmp %p %p\n", 4637 nmp, ds->nfsdev_nmp); 4638 if (nfsaddr2_match(sad, nmp->nm_nam)) { 4639 fndds = ds; 4640 break; 4641 } 4642 } 4643 } 4644 if (fndds != NULL && newnmpp != NULL && 4645 newnmp == NULL) { 4646 /* Search for a place to make a mirror copy. */ 4647 TAILQ_FOREACH(tds, &nfsrv_devidhead, 4648 nfsdev_list) { 4649 if (tds->nfsdev_nmp != NULL && 4650 fndds != tds && 4651 ((tds->nfsdev_mdsisset == 0 && 4652 fndds->nfsdev_mdsisset == 0) || 4653 (tds->nfsdev_mdsisset != 0 && 4654 fndds->nfsdev_mdsisset != 0 && 4655 tds->nfsdev_mdsfsid.val[0] == 4656 mp->mnt_stat.f_fsid.val[0] && 4657 tds->nfsdev_mdsfsid.val[1] == 4658 mp->mnt_stat.f_fsid.val[1]))) { 4659 *newnmpp = tds->nfsdev_nmp; 4660 break; 4661 } 4662 } 4663 if (tds != NULL) { 4664 /* 4665 * Move this entry to the end of the 4666 * list, so it won't be selected as 4667 * easily the next time. 4668 */ 4669 TAILQ_REMOVE(&nfsrv_devidhead, tds, 4670 nfsdev_list); 4671 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, 4672 nfsdev_list); 4673 } 4674 } 4675 NFSDDSUNLOCK(); 4676 if (fndds != NULL) { 4677 dvp = fndds->nfsdev_dsdir[dsdir]; 4678 if (lktype != 0 || fhiszero != 0 || 4679 (nvpp != NULL && *nvpp == NULL)) { 4680 if (fhiszero != 0) 4681 error = vn_lock(dvp, 4682 LK_EXCLUSIVE); 4683 else if (lktype != 0) 4684 error = vn_lock(dvp, lktype); 4685 else 4686 error = vn_lock(dvp, LK_SHARED); 4687 /* 4688 * If the file handle is all 0's, try to 4689 * do a Lookup against the DS to acquire 4690 * it. 4691 * If dvpp == NULL or the Lookup fails, 4692 * unlock dvp after the call. 4693 */ 4694 if (error == 0 && (fhiszero != 0 || 4695 (nvpp != NULL && *nvpp == NULL))) { 4696 error = nfsrv_pnfslookupds(vp, 4697 dvp, pf, &nvp, p); 4698 if (error == 0) { 4699 if (fhiszero != 0) 4700 nfsrv_pnfssetfh( 4701 vp, pf, 4702 devid, 4703 fnamep, 4704 nvp, p); 4705 if (nvpp != NULL && 4706 *nvpp == NULL) { 4707 *nvpp = nvp; 4708 *dsdirp = dsdir; 4709 } else 4710 vput(nvp); 4711 } 4712 if (error != 0 || lktype == 0) 4713 NFSVOPUNLOCK(dvp, 0); 4714 } 4715 } 4716 if (error == 0) { 4717 gotone++; 4718 NFSD_DEBUG(4, "gotone=%d\n", gotone); 4719 if (devid != NULL) { 4720 NFSBCOPY(fndds->nfsdev_deviceid, 4721 devid, NFSX_V4DEVICEID); 4722 devid += NFSX_V4DEVICEID; 4723 } 4724 if (dvpp != NULL) 4725 *tdvpp++ = dvp; 4726 if (fhp != NULL) 4727 NFSBCOPY(&pf->dsf_fh, fhp++, 4728 NFSX_MYFH); 4729 if (fnamep != NULL && gotone == 1) 4730 strlcpy(fnamep, 4731 pf->dsf_filename, 4732 sizeof(pf->dsf_filename)); 4733 } else 4734 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " 4735 "err=%d\n", error); 4736 } 4737 } 4738 } 4739 if (error == 0 && gotone == 0) 4740 error = ENOENT; 4741 4742 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, 4743 error); 4744 if (error == 0) 4745 *mirrorcntp = gotone; 4746 else { 4747 if (gotone > 0 && dvpp != NULL) { 4748 /* 4749 * If the error didn't occur on the first one and 4750 * dvpp != NULL, the one(s) prior to the failure will 4751 * have locked dvp's that need to be unlocked. 4752 */ 4753 for (i = 0; i < gotone; i++) { 4754 NFSVOPUNLOCK(*dvpp, 0); 4755 *dvpp++ = NULL; 4756 } 4757 } 4758 /* 4759 * If it found the vnode to be copied from before a failure, 4760 * it needs to be vput()'d. 4761 */ 4762 if (nvpp != NULL && *nvpp != NULL) { 4763 vput(*nvpp); 4764 *nvpp = NULL; 4765 } 4766 } 4767 return (error); 4768 } 4769 4770 /* 4771 * Set the extended attribute for the Change attribute. 4772 */ 4773 static int 4774 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 4775 { 4776 struct pnfsdsattr dsattr; 4777 int error; 4778 4779 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); 4780 dsattr.dsa_filerev = nap->na_filerev; 4781 dsattr.dsa_size = nap->na_size; 4782 dsattr.dsa_atime = nap->na_atime; 4783 dsattr.dsa_mtime = nap->na_mtime; 4784 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 4785 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); 4786 if (error != 0) 4787 printf("pNFS: setextattr=%d\n", error); 4788 return (error); 4789 } 4790 4791 static int 4792 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 4793 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) 4794 { 4795 uint32_t *tl; 4796 struct nfsrv_descript *nd; 4797 nfsv4stateid_t st; 4798 struct mbuf *m, *m2; 4799 int error = 0, retlen, tlen, trimlen; 4800 4801 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); 4802 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 4803 *mpp = NULL; 4804 /* 4805 * Use a stateid where other is an alternating 01010 pattern and 4806 * seqid is 0xffffffff. This value is not defined as special by 4807 * the RFC and is used by the FreeBSD NFS server to indicate an 4808 * MDS->DS proxy operation. 4809 */ 4810 st.other[0] = 0x55555555; 4811 st.other[1] = 0x55555555; 4812 st.other[2] = 0x55555555; 4813 st.seqid = 0xffffffff; 4814 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), 4815 NULL, NULL, 0, 0); 4816 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 4817 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); 4818 txdr_hyper(off, tl); 4819 *(tl + 2) = txdr_unsigned(len); 4820 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 4821 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 4822 if (error != 0) { 4823 free(nd, M_TEMP); 4824 return (error); 4825 } 4826 if (nd->nd_repstat == 0) { 4827 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 4828 NFSM_STRSIZ(retlen, len); 4829 if (retlen > 0) { 4830 /* Trim off the pre-data XDR from the mbuf chain. */ 4831 m = nd->nd_mrep; 4832 while (m != NULL && m != nd->nd_md) { 4833 if (m->m_next == nd->nd_md) { 4834 m->m_next = NULL; 4835 m_freem(nd->nd_mrep); 4836 nd->nd_mrep = m = nd->nd_md; 4837 } else 4838 m = m->m_next; 4839 } 4840 if (m == NULL) { 4841 printf("nfsrv_readdsrpc: busted mbuf list\n"); 4842 error = ENOENT; 4843 goto nfsmout; 4844 } 4845 4846 /* 4847 * Now, adjust first mbuf so that any XDR before the 4848 * read data is skipped over. 4849 */ 4850 trimlen = nd->nd_dpos - mtod(m, char *); 4851 if (trimlen > 0) { 4852 m->m_len -= trimlen; 4853 NFSM_DATAP(m, trimlen); 4854 } 4855 4856 /* 4857 * Truncate the mbuf chain at retlen bytes of data, 4858 * plus XDR padding that brings the length up to a 4859 * multiple of 4. 4860 */ 4861 tlen = NFSM_RNDUP(retlen); 4862 do { 4863 if (m->m_len >= tlen) { 4864 m->m_len = tlen; 4865 tlen = 0; 4866 m2 = m->m_next; 4867 m->m_next = NULL; 4868 m_freem(m2); 4869 break; 4870 } 4871 tlen -= m->m_len; 4872 m = m->m_next; 4873 } while (m != NULL); 4874 if (tlen > 0) { 4875 printf("nfsrv_readdsrpc: busted mbuf list\n"); 4876 error = ENOENT; 4877 goto nfsmout; 4878 } 4879 *mpp = nd->nd_mrep; 4880 *mpendp = m; 4881 nd->nd_mrep = NULL; 4882 } 4883 } else 4884 error = nd->nd_repstat; 4885 nfsmout: 4886 /* If nd->nd_mrep is already NULL, this is a no-op. */ 4887 m_freem(nd->nd_mrep); 4888 free(nd, M_TEMP); 4889 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); 4890 return (error); 4891 } 4892 4893 /* 4894 * Do a write RPC on a DS data file, using this structure for the arguments, 4895 * so that this function can be executed by a separate kernel process. 4896 */ 4897 struct nfsrvwritedsdorpc { 4898 int done; 4899 int inprog; 4900 struct task tsk; 4901 fhandle_t fh; 4902 off_t off; 4903 int len; 4904 struct nfsmount *nmp; 4905 struct ucred *cred; 4906 NFSPROC_T *p; 4907 struct mbuf *m; 4908 int err; 4909 }; 4910 4911 static int 4912 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, 4913 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) 4914 { 4915 uint32_t *tl; 4916 struct nfsrv_descript *nd; 4917 nfsattrbit_t attrbits; 4918 nfsv4stateid_t st; 4919 int commit, error, retlen; 4920 4921 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 4922 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, 4923 sizeof(fhandle_t), NULL, NULL, 0, 0); 4924 4925 /* 4926 * Use a stateid where other is an alternating 01010 pattern and 4927 * seqid is 0xffffffff. This value is not defined as special by 4928 * the RFC and is used by the FreeBSD NFS server to indicate an 4929 * MDS->DS proxy operation. 4930 */ 4931 st.other[0] = 0x55555555; 4932 st.other[1] = 0x55555555; 4933 st.other[2] = 0x55555555; 4934 st.seqid = 0xffffffff; 4935 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 4936 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); 4937 txdr_hyper(off, tl); 4938 tl += 2; 4939 /* 4940 * Do all writes FileSync, since the server doesn't hold onto dirty 4941 * buffers. Since clients should be accessing the DS servers directly 4942 * using the pNFS layouts, this just needs to work correctly as a 4943 * fallback. 4944 */ 4945 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); 4946 *tl = txdr_unsigned(len); 4947 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); 4948 4949 /* Put data in mbuf chain. */ 4950 nd->nd_mb->m_next = m; 4951 4952 /* Set nd_mb and nd_bpos to end of data. */ 4953 while (m->m_next != NULL) 4954 m = m->m_next; 4955 nd->nd_mb = m; 4956 nd->nd_bpos = mtod(m, char *) + m->m_len; 4957 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); 4958 4959 /* Do a Getattr for Size, Change and Modify Time. */ 4960 NFSZERO_ATTRBIT(&attrbits); 4961 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 4962 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 4963 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 4964 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 4965 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 4966 *tl = txdr_unsigned(NFSV4OP_GETATTR); 4967 (void) nfsrv_putattrbit(nd, &attrbits); 4968 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 4969 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 4970 if (error != 0) { 4971 free(nd, M_TEMP); 4972 return (error); 4973 } 4974 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); 4975 /* Get rid of weak cache consistency data for now. */ 4976 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 4977 (ND_NFSV4 | ND_V4WCCATTR)) { 4978 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 4979 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 4980 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); 4981 if (error != 0) 4982 goto nfsmout; 4983 /* 4984 * Get rid of Op# and status for next op. 4985 */ 4986 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 4987 if (*++tl != 0) 4988 nd->nd_flag |= ND_NOMOREDATA; 4989 } 4990 if (nd->nd_repstat == 0) { 4991 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); 4992 retlen = fxdr_unsigned(int, *tl++); 4993 commit = fxdr_unsigned(int, *tl); 4994 if (commit != NFSWRITE_FILESYNC) 4995 error = NFSERR_IO; 4996 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", 4997 retlen, commit, error); 4998 } else 4999 error = nd->nd_repstat; 5000 /* We have no use for the Write Verifier since we use FileSync. */ 5001 5002 /* 5003 * Get the Change, Size, Access Time and Modify Time attributes and set 5004 * on the Metadata file, so its attributes will be what the file's 5005 * would be if it had been written. 5006 */ 5007 if (error == 0) { 5008 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5009 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5010 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5011 } 5012 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); 5013 nfsmout: 5014 m_freem(nd->nd_mrep); 5015 free(nd, M_TEMP); 5016 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); 5017 return (error); 5018 } 5019 5020 /* 5021 * Start up the thread that will execute nfsrv_writedsdorpc(). 5022 */ 5023 static void 5024 start_writedsdorpc(void *arg, int pending) 5025 { 5026 struct nfsrvwritedsdorpc *drpc; 5027 5028 drpc = (struct nfsrvwritedsdorpc *)arg; 5029 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5030 drpc->len, NULL, drpc->m, drpc->cred, drpc->p); 5031 drpc->done = 1; 5032 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); 5033 } 5034 5035 static int 5036 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5037 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5038 struct mbuf **mpp, char *cp, int *failposp) 5039 { 5040 struct nfsrvwritedsdorpc *drpc, *tdrpc; 5041 struct nfsvattr na; 5042 struct mbuf *m; 5043 int error, i, offs, ret, timo; 5044 5045 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); 5046 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); 5047 drpc = NULL; 5048 if (mirrorcnt > 1) 5049 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5050 M_WAITOK); 5051 5052 /* Calculate offset in mbuf chain that data starts. */ 5053 offs = cp - mtod(*mpp, char *); 5054 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); 5055 5056 /* 5057 * Do the write RPC for every DS, using a separate kernel process 5058 * for every DS except the last one. 5059 */ 5060 error = 0; 5061 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5062 tdrpc->done = 0; 5063 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5064 tdrpc->off = off; 5065 tdrpc->len = len; 5066 tdrpc->nmp = *nmpp; 5067 tdrpc->cred = cred; 5068 tdrpc->p = p; 5069 tdrpc->inprog = 0; 5070 tdrpc->err = 0; 5071 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5072 ret = EIO; 5073 if (nfs_pnfsiothreads != 0) { 5074 ret = nfs_pnfsio(start_writedsdorpc, tdrpc); 5075 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", 5076 ret); 5077 } 5078 if (ret != 0) { 5079 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, 5080 tdrpc->m, cred, p); 5081 if (nfsds_failerr(ret) && *failposp == -1) 5082 *failposp = i; 5083 else if (error == 0 && ret != 0) 5084 error = ret; 5085 } 5086 nmpp++; 5087 fhp++; 5088 } 5089 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5090 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); 5091 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5092 *failposp = mirrorcnt - 1; 5093 else if (error == 0 && ret != 0) 5094 error = ret; 5095 if (error == 0) 5096 error = nfsrv_setextattr(vp, &na, p); 5097 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); 5098 tdrpc = drpc; 5099 timo = hz / 50; /* Wait for 20msec. */ 5100 if (timo < 1) 5101 timo = 1; 5102 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5103 /* Wait for RPCs on separate threads to complete. */ 5104 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5105 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); 5106 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5107 *failposp = i; 5108 else if (error == 0 && tdrpc->err != 0) 5109 error = tdrpc->err; 5110 } 5111 free(drpc, M_TEMP); 5112 return (error); 5113 } 5114 5115 static int 5116 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5117 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, 5118 struct nfsvattr *dsnap) 5119 { 5120 uint32_t *tl; 5121 struct nfsrv_descript *nd; 5122 nfsv4stateid_t st; 5123 nfsattrbit_t attrbits; 5124 int error; 5125 5126 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); 5127 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 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_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5139 NULL, NULL, 0, 0); 5140 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5141 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); 5142 5143 /* Do a Getattr for Size, Change, Access Time and Modify Time. */ 5144 NFSZERO_ATTRBIT(&attrbits); 5145 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5146 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5147 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5148 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5149 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5150 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5151 (void) nfsrv_putattrbit(nd, &attrbits); 5152 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5153 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5154 if (error != 0) { 5155 free(nd, M_TEMP); 5156 return (error); 5157 } 5158 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", 5159 nd->nd_repstat); 5160 /* Get rid of weak cache consistency data for now. */ 5161 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5162 (ND_NFSV4 | ND_V4WCCATTR)) { 5163 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5164 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5165 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); 5166 if (error != 0) 5167 goto nfsmout; 5168 /* 5169 * Get rid of Op# and status for next op. 5170 */ 5171 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5172 if (*++tl != 0) 5173 nd->nd_flag |= ND_NOMOREDATA; 5174 } 5175 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 5176 if (error != 0) 5177 goto nfsmout; 5178 if (nd->nd_repstat != 0) 5179 error = nd->nd_repstat; 5180 /* 5181 * Get the Change, Size, Access Time and Modify Time attributes and set 5182 * on the Metadata file, so its attributes will be what the file's 5183 * would be if it had been written. 5184 */ 5185 if (error == 0) { 5186 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5187 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5188 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5189 } 5190 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); 5191 nfsmout: 5192 m_freem(nd->nd_mrep); 5193 free(nd, M_TEMP); 5194 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); 5195 return (error); 5196 } 5197 5198 struct nfsrvsetattrdsdorpc { 5199 int done; 5200 int inprog; 5201 struct task tsk; 5202 fhandle_t fh; 5203 struct nfsmount *nmp; 5204 struct vnode *vp; 5205 struct ucred *cred; 5206 NFSPROC_T *p; 5207 struct nfsvattr na; 5208 struct nfsvattr dsna; 5209 int err; 5210 }; 5211 5212 /* 5213 * Start up the thread that will execute nfsrv_setattrdsdorpc(). 5214 */ 5215 static void 5216 start_setattrdsdorpc(void *arg, int pending) 5217 { 5218 struct nfsrvsetattrdsdorpc *drpc; 5219 5220 drpc = (struct nfsrvsetattrdsdorpc *)arg; 5221 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, 5222 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); 5223 drpc->done = 1; 5224 } 5225 5226 static int 5227 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5228 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5229 struct nfsvattr *nap, int *failposp) 5230 { 5231 struct nfsrvsetattrdsdorpc *drpc, *tdrpc; 5232 struct nfsvattr na; 5233 int error, i, ret, timo; 5234 5235 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); 5236 drpc = NULL; 5237 if (mirrorcnt > 1) 5238 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5239 M_WAITOK); 5240 5241 /* 5242 * Do the setattr RPC for every DS, using a separate kernel process 5243 * for every DS except the last one. 5244 */ 5245 error = 0; 5246 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5247 tdrpc->done = 0; 5248 tdrpc->inprog = 0; 5249 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5250 tdrpc->nmp = *nmpp; 5251 tdrpc->vp = vp; 5252 tdrpc->cred = cred; 5253 tdrpc->p = p; 5254 tdrpc->na = *nap; 5255 tdrpc->err = 0; 5256 ret = EIO; 5257 if (nfs_pnfsiothreads != 0) { 5258 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); 5259 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", 5260 ret); 5261 } 5262 if (ret != 0) { 5263 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, 5264 &na); 5265 if (nfsds_failerr(ret) && *failposp == -1) 5266 *failposp = i; 5267 else if (error == 0 && ret != 0) 5268 error = ret; 5269 } 5270 nmpp++; 5271 fhp++; 5272 } 5273 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); 5274 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5275 *failposp = mirrorcnt - 1; 5276 else if (error == 0 && ret != 0) 5277 error = ret; 5278 if (error == 0) 5279 error = nfsrv_setextattr(vp, &na, p); 5280 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); 5281 tdrpc = drpc; 5282 timo = hz / 50; /* Wait for 20msec. */ 5283 if (timo < 1) 5284 timo = 1; 5285 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5286 /* Wait for RPCs on separate threads to complete. */ 5287 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5288 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); 5289 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5290 *failposp = i; 5291 else if (error == 0 && tdrpc->err != 0) 5292 error = tdrpc->err; 5293 } 5294 free(drpc, M_TEMP); 5295 return (error); 5296 } 5297 5298 /* 5299 * Do a Setattr of an NFSv4 ACL on the DS file. 5300 */ 5301 static int 5302 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5303 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) 5304 { 5305 struct nfsrv_descript *nd; 5306 nfsv4stateid_t st; 5307 nfsattrbit_t attrbits; 5308 int error; 5309 5310 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); 5311 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5312 /* 5313 * Use a stateid where other is an alternating 01010 pattern and 5314 * seqid is 0xffffffff. This value is not defined as special by 5315 * the RFC and is used by the FreeBSD NFS server to indicate an 5316 * MDS->DS proxy operation. 5317 */ 5318 st.other[0] = 0x55555555; 5319 st.other[1] = 0x55555555; 5320 st.other[2] = 0x55555555; 5321 st.seqid = 0xffffffff; 5322 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5323 NULL, NULL, 0, 0); 5324 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5325 NFSZERO_ATTRBIT(&attrbits); 5326 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); 5327 /* 5328 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), 5329 * so passing in the metadata "vp" will be ok, since it is of 5330 * the same type (VREG). 5331 */ 5332 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, 5333 NULL, 0, 0, 0, 0, 0, NULL); 5334 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5335 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5336 if (error != 0) { 5337 free(nd, M_TEMP); 5338 return (error); 5339 } 5340 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", 5341 nd->nd_repstat); 5342 error = nd->nd_repstat; 5343 m_freem(nd->nd_mrep); 5344 free(nd, M_TEMP); 5345 return (error); 5346 } 5347 5348 struct nfsrvsetacldsdorpc { 5349 int done; 5350 int inprog; 5351 struct task tsk; 5352 fhandle_t fh; 5353 struct nfsmount *nmp; 5354 struct vnode *vp; 5355 struct ucred *cred; 5356 NFSPROC_T *p; 5357 struct acl *aclp; 5358 int err; 5359 }; 5360 5361 /* 5362 * Start up the thread that will execute nfsrv_setacldsdorpc(). 5363 */ 5364 static void 5365 start_setacldsdorpc(void *arg, int pending) 5366 { 5367 struct nfsrvsetacldsdorpc *drpc; 5368 5369 drpc = (struct nfsrvsetacldsdorpc *)arg; 5370 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, 5371 drpc->vp, drpc->nmp, drpc->aclp); 5372 drpc->done = 1; 5373 } 5374 5375 static int 5376 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5377 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, 5378 int *failposp) 5379 { 5380 struct nfsrvsetacldsdorpc *drpc, *tdrpc; 5381 int error, i, ret, timo; 5382 5383 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); 5384 drpc = NULL; 5385 if (mirrorcnt > 1) 5386 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5387 M_WAITOK); 5388 5389 /* 5390 * Do the setattr RPC for every DS, using a separate kernel process 5391 * for every DS except the last one. 5392 */ 5393 error = 0; 5394 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5395 tdrpc->done = 0; 5396 tdrpc->inprog = 0; 5397 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5398 tdrpc->nmp = *nmpp; 5399 tdrpc->vp = vp; 5400 tdrpc->cred = cred; 5401 tdrpc->p = p; 5402 tdrpc->aclp = aclp; 5403 tdrpc->err = 0; 5404 ret = EIO; 5405 if (nfs_pnfsiothreads != 0) { 5406 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); 5407 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", 5408 ret); 5409 } 5410 if (ret != 0) { 5411 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, 5412 aclp); 5413 if (nfsds_failerr(ret) && *failposp == -1) 5414 *failposp = i; 5415 else if (error == 0 && ret != 0) 5416 error = ret; 5417 } 5418 nmpp++; 5419 fhp++; 5420 } 5421 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); 5422 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5423 *failposp = mirrorcnt - 1; 5424 else if (error == 0 && ret != 0) 5425 error = ret; 5426 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); 5427 tdrpc = drpc; 5428 timo = hz / 50; /* Wait for 20msec. */ 5429 if (timo < 1) 5430 timo = 1; 5431 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5432 /* Wait for RPCs on separate threads to complete. */ 5433 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5434 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); 5435 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5436 *failposp = i; 5437 else if (error == 0 && tdrpc->err != 0) 5438 error = tdrpc->err; 5439 } 5440 free(drpc, M_TEMP); 5441 return (error); 5442 } 5443 5444 /* 5445 * Getattr call to the DS for the Modify, Size and Change attributes. 5446 */ 5447 static int 5448 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5449 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) 5450 { 5451 struct nfsrv_descript *nd; 5452 int error; 5453 nfsattrbit_t attrbits; 5454 5455 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); 5456 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5457 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, 5458 sizeof(fhandle_t), NULL, NULL, 0, 0); 5459 NFSZERO_ATTRBIT(&attrbits); 5460 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5461 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5462 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5463 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5464 (void) nfsrv_putattrbit(nd, &attrbits); 5465 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5466 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5467 if (error != 0) { 5468 free(nd, M_TEMP); 5469 return (error); 5470 } 5471 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", 5472 nd->nd_repstat); 5473 if (nd->nd_repstat == 0) { 5474 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, 5475 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, 5476 NULL, NULL); 5477 /* 5478 * We can only save the updated values in the extended 5479 * attribute if the vp is exclusively locked. 5480 * This should happen when any of the following operations 5481 * occur on the vnode: 5482 * Close, Delegreturn, LayoutCommit, LayoutReturn 5483 * As such, the updated extended attribute should get saved 5484 * before nfsrv_checkdsattr() returns 0 and allows the cached 5485 * attributes to be returned without calling this function. 5486 */ 5487 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { 5488 error = nfsrv_setextattr(vp, nap, p); 5489 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", 5490 error); 5491 } 5492 } else 5493 error = nd->nd_repstat; 5494 m_freem(nd->nd_mrep); 5495 free(nd, M_TEMP); 5496 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); 5497 return (error); 5498 } 5499 5500 /* 5501 * Get the device id and file handle for a DS file. 5502 */ 5503 int 5504 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, 5505 fhandle_t *fhp, char *devid) 5506 { 5507 int buflen, error; 5508 char *buf; 5509 5510 buflen = 1024; 5511 buf = malloc(buflen, M_TEMP, M_WAITOK); 5512 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, 5513 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); 5514 free(buf, M_TEMP); 5515 return (error); 5516 } 5517 5518 /* 5519 * Do a Lookup against the DS for the filename. 5520 */ 5521 static int 5522 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, 5523 struct vnode **nvpp, NFSPROC_T *p) 5524 { 5525 struct nameidata named; 5526 struct ucred *tcred; 5527 char *bufp; 5528 u_long *hashp; 5529 struct vnode *nvp; 5530 int error; 5531 5532 tcred = newnfs_getcred(); 5533 named.ni_cnd.cn_nameiop = LOOKUP; 5534 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; 5535 named.ni_cnd.cn_cred = tcred; 5536 named.ni_cnd.cn_thread = p; 5537 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; 5538 nfsvno_setpathbuf(&named, &bufp, &hashp); 5539 named.ni_cnd.cn_nameptr = bufp; 5540 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); 5541 strlcpy(bufp, pf->dsf_filename, NAME_MAX); 5542 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); 5543 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 5544 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); 5545 NFSFREECRED(tcred); 5546 nfsvno_relpathbuf(&named); 5547 if (error == 0) 5548 *nvpp = nvp; 5549 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); 5550 return (error); 5551 } 5552 5553 /* 5554 * Set the file handle to the correct one. 5555 */ 5556 static void 5557 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, 5558 char *fnamep, struct vnode *nvp, NFSPROC_T *p) 5559 { 5560 struct nfsnode *np; 5561 int ret; 5562 5563 np = VTONFS(nvp); 5564 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); 5565 /* 5566 * We can only do a vn_set_extattr() if the vnode is exclusively 5567 * locked and vn_start_write() has been done. If devid != NULL or 5568 * fnamep != NULL or the vnode is shared locked, vn_start_write() 5569 * may not have been done. 5570 * If not done now, it will be done on a future call. 5571 */ 5572 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == 5573 LK_EXCLUSIVE) 5574 ret = vn_extattr_set(vp, IO_NODELOCKED, 5575 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), 5576 (char *)pf, p); 5577 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); 5578 } 5579 5580 /* 5581 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point 5582 * when the DS has failed. 5583 */ 5584 void 5585 nfsrv_killrpcs(struct nfsmount *nmp) 5586 { 5587 5588 /* 5589 * Call newnfs_nmcancelreqs() to cause 5590 * any RPCs in progress on the mount point to 5591 * fail. 5592 * This will cause any process waiting for an 5593 * RPC to complete while holding a vnode lock 5594 * on the mounted-on vnode (such as "df" or 5595 * a non-forced "umount") to fail. 5596 * This will unlock the mounted-on vnode so 5597 * a forced dismount can succeed. 5598 * The NFSMNTP_CANCELRPCS flag should be set when this function is 5599 * called. 5600 */ 5601 newnfs_nmcancelreqs(nmp); 5602 } 5603 5604 /* 5605 * Sum up the statfs info for each of the DSs, so that the client will 5606 * receive the total for all DSs. 5607 */ 5608 static int 5609 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) 5610 { 5611 struct statfs *tsf; 5612 struct nfsdevice *ds; 5613 struct vnode **dvpp, **tdvpp, *dvp; 5614 uint64_t tot; 5615 int cnt, error = 0, i; 5616 5617 if (nfsrv_devidcnt <= 0) 5618 return (ENXIO); 5619 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 5620 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 5621 5622 /* Get an array of the dvps for the DSs. */ 5623 tdvpp = dvpp; 5624 i = 0; 5625 NFSDDSLOCK(); 5626 /* First, search for matches for same file system. */ 5627 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5628 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && 5629 ds->nfsdev_mdsfsid.val[0] == mp->mnt_stat.f_fsid.val[0] && 5630 ds->nfsdev_mdsfsid.val[1] == mp->mnt_stat.f_fsid.val[1]) { 5631 if (++i > nfsrv_devidcnt) 5632 break; 5633 *tdvpp++ = ds->nfsdev_dvp; 5634 } 5635 } 5636 /* 5637 * If no matches for same file system, total all servers not assigned 5638 * to a file system. 5639 */ 5640 if (i == 0) { 5641 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5642 if (ds->nfsdev_nmp != NULL && 5643 ds->nfsdev_mdsisset == 0) { 5644 if (++i > nfsrv_devidcnt) 5645 break; 5646 *tdvpp++ = ds->nfsdev_dvp; 5647 } 5648 } 5649 } 5650 NFSDDSUNLOCK(); 5651 cnt = i; 5652 5653 /* Do a VFS_STATFS() for each of the DSs and sum them up. */ 5654 tdvpp = dvpp; 5655 for (i = 0; i < cnt && error == 0; i++) { 5656 dvp = *tdvpp++; 5657 error = VFS_STATFS(dvp->v_mount, tsf); 5658 if (error == 0) { 5659 if (sf->f_bsize == 0) { 5660 if (tsf->f_bsize > 0) 5661 sf->f_bsize = tsf->f_bsize; 5662 else 5663 sf->f_bsize = 8192; 5664 } 5665 if (tsf->f_blocks > 0) { 5666 if (sf->f_bsize != tsf->f_bsize) { 5667 tot = tsf->f_blocks * tsf->f_bsize; 5668 sf->f_blocks += (tot / sf->f_bsize); 5669 } else 5670 sf->f_blocks += tsf->f_blocks; 5671 } 5672 if (tsf->f_bfree > 0) { 5673 if (sf->f_bsize != tsf->f_bsize) { 5674 tot = tsf->f_bfree * tsf->f_bsize; 5675 sf->f_bfree += (tot / sf->f_bsize); 5676 } else 5677 sf->f_bfree += tsf->f_bfree; 5678 } 5679 if (tsf->f_bavail > 0) { 5680 if (sf->f_bsize != tsf->f_bsize) { 5681 tot = tsf->f_bavail * tsf->f_bsize; 5682 sf->f_bavail += (tot / sf->f_bsize); 5683 } else 5684 sf->f_bavail += tsf->f_bavail; 5685 } 5686 } 5687 } 5688 free(tsf, M_TEMP); 5689 free(dvpp, M_TEMP); 5690 return (error); 5691 } 5692 5693 /* 5694 * Set an NFSv4 acl. 5695 */ 5696 int 5697 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) 5698 { 5699 int error; 5700 5701 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { 5702 error = NFSERR_ATTRNOTSUPP; 5703 goto out; 5704 } 5705 /* 5706 * With NFSv4 ACLs, chmod(2) may need to add additional entries. 5707 * Make sure it has enough room for that - splitting every entry 5708 * into two and appending "canonical six" entries at the end. 5709 * Cribbed out of kern/vfs_acl.c - Rick M. 5710 */ 5711 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { 5712 error = NFSERR_ATTRNOTSUPP; 5713 goto out; 5714 } 5715 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 5716 if (error == 0) { 5717 error = nfsrv_dssetacl(vp, aclp, cred, p); 5718 if (error == ENOENT) 5719 error = 0; 5720 } 5721 5722 out: 5723 NFSEXITCODE(error); 5724 return (error); 5725 } 5726 5727 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); 5728 5729 /* 5730 * Called once to initialize data structures... 5731 */ 5732 static int 5733 nfsd_modevent(module_t mod, int type, void *data) 5734 { 5735 int error = 0, i; 5736 static int loaded = 0; 5737 5738 switch (type) { 5739 case MOD_LOAD: 5740 if (loaded) 5741 goto out; 5742 newnfs_portinit(); 5743 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 5744 mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL, 5745 MTX_DEF); 5746 mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL, 5747 MTX_DEF); 5748 } 5749 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); 5750 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); 5751 mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF); 5752 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); 5753 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); 5754 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0); 5755 nfsrvd_initcache(); 5756 nfsd_init(); 5757 NFSD_LOCK(); 5758 nfsrvd_init(0); 5759 NFSD_UNLOCK(); 5760 nfsd_mntinit(); 5761 #ifdef VV_DISABLEDELEG 5762 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; 5763 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; 5764 #endif 5765 nfsd_call_servertimer = nfsrv_servertimer; 5766 nfsd_call_nfsd = nfssvc_nfsd; 5767 loaded = 1; 5768 break; 5769 5770 case MOD_UNLOAD: 5771 if (newnfs_numnfsd != 0) { 5772 error = EBUSY; 5773 break; 5774 } 5775 5776 #ifdef VV_DISABLEDELEG 5777 vn_deleg_ops.vndeleg_recall = NULL; 5778 vn_deleg_ops.vndeleg_disable = NULL; 5779 #endif 5780 nfsd_call_servertimer = NULL; 5781 nfsd_call_nfsd = NULL; 5782 5783 /* Clean out all NFSv4 state. */ 5784 nfsrv_throwawayallstate(curthread); 5785 5786 /* Clean the NFS server reply cache */ 5787 nfsrvd_cleancache(); 5788 5789 /* Free up the krpc server pool. */ 5790 if (nfsrvd_pool != NULL) 5791 svcpool_destroy(nfsrvd_pool); 5792 5793 /* and get rid of the locks */ 5794 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 5795 mtx_destroy(&nfsrchash_table[i].mtx); 5796 mtx_destroy(&nfsrcahash_table[i].mtx); 5797 } 5798 mtx_destroy(&nfsrc_udpmtx); 5799 mtx_destroy(&nfs_v4root_mutex); 5800 mtx_destroy(&nfsv4root_mnt.mnt_mtx); 5801 mtx_destroy(&nfsrv_dontlistlock_mtx); 5802 mtx_destroy(&nfsrv_recalllock_mtx); 5803 for (i = 0; i < nfsrv_sessionhashsize; i++) 5804 mtx_destroy(&nfssessionhash[i].mtx); 5805 if (nfslayouthash != NULL) { 5806 for (i = 0; i < nfsrv_layouthashsize; i++) 5807 mtx_destroy(&nfslayouthash[i].mtx); 5808 free(nfslayouthash, M_NFSDSESSION); 5809 } 5810 lockdestroy(&nfsv4root_mnt.mnt_explock); 5811 free(nfsclienthash, M_NFSDCLIENT); 5812 free(nfslockhash, M_NFSDLOCKFILE); 5813 free(nfssessionhash, M_NFSDSESSION); 5814 loaded = 0; 5815 break; 5816 default: 5817 error = EOPNOTSUPP; 5818 break; 5819 } 5820 5821 out: 5822 NFSEXITCODE(error); 5823 return (error); 5824 } 5825 static moduledata_t nfsd_mod = { 5826 "nfsd", 5827 nfsd_modevent, 5828 NULL, 5829 }; 5830 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); 5831 5832 /* So that loader and kldload(2) can find us, wherever we are.. */ 5833 MODULE_VERSION(nfsd, 1); 5834 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); 5835 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1); 5836 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); 5837 MODULE_DEPEND(nfsd, krpc, 1, 1, 1); 5838 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); 5839 5840