1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 */ 35 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include <sys/capsicum.h> 40 #include <sys/extattr.h> 41 42 /* 43 * Functions that perform the vfs operations required by the routines in 44 * nfsd_serv.c. It is hoped that this change will make the server more 45 * portable. 46 */ 47 48 #include <fs/nfs/nfsport.h> 49 #include <security/mac/mac_framework.h> 50 #include <sys/filio.h> 51 #include <sys/hash.h> 52 #include <sys/sysctl.h> 53 #include <nlm/nlm_prot.h> 54 #include <nlm/nlm.h> 55 56 FEATURE(nfsd, "NFSv4 server"); 57 58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 59 extern int nfsrv_useacl; 60 extern int newnfs_numnfsd; 61 extern struct mount nfsv4root_mnt; 62 extern struct nfsrv_stablefirst nfsrv_stablefirst; 63 extern void (*nfsd_call_servertimer)(void); 64 extern SVCPOOL *nfsrvd_pool; 65 extern struct nfsv4lock nfsd_suspend_lock; 66 extern struct nfsclienthashhead *nfsclienthash; 67 extern struct nfslockhashhead *nfslockhash; 68 extern struct nfssessionhash *nfssessionhash; 69 extern int nfsrv_sessionhashsize; 70 extern struct nfsstatsv1 nfsstatsv1; 71 extern struct nfslayouthash *nfslayouthash; 72 extern int nfsrv_layouthashsize; 73 extern struct mtx nfsrv_dslock_mtx; 74 extern int nfs_pnfsiothreads; 75 extern struct nfsdontlisthead nfsrv_dontlisthead; 76 extern volatile int nfsrv_dontlistlen; 77 extern volatile int nfsrv_devidcnt; 78 extern int nfsrv_maxpnfsmirror; 79 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt; 80 NFSDLOCKMUTEX; 81 NFSSTATESPINLOCK; 82 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE]; 83 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE]; 84 struct mtx nfsrc_udpmtx; 85 struct mtx nfs_v4root_mutex; 86 struct mtx nfsrv_dontlistlock_mtx; 87 struct mtx nfsrv_recalllock_mtx; 88 struct nfsrvfh nfs_rootfh, nfs_pubfh; 89 int nfs_pubfhset = 0, nfs_rootfhset = 0; 90 struct proc *nfsd_master_proc = NULL; 91 int nfsd_debuglevel = 0; 92 static pid_t nfsd_master_pid = (pid_t)-1; 93 static char nfsd_master_comm[MAXCOMLEN + 1]; 94 static struct timeval nfsd_master_start; 95 static uint32_t nfsv4_sysid = 0; 96 static fhandle_t zerofh; 97 98 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, 99 struct ucred *); 100 101 int nfsrv_enable_crossmntpt = 1; 102 static int nfs_commit_blks; 103 static int nfs_commit_miss; 104 extern int nfsrv_issuedelegs; 105 extern int nfsrv_dolocallocks; 106 extern int nfsd_enable_stringtouid; 107 extern struct nfsdevicehead nfsrv_devidhead; 108 109 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, 110 struct iovec **); 111 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, 112 int *); 113 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, 114 NFSPROC_T *); 115 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, 116 int *, char *, fhandle_t *); 117 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, 118 NFSPROC_T *); 119 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, 120 struct thread *, int, struct mbuf **, char *, struct mbuf **, 121 struct nfsvattr *, struct acl *, off_t *, int, bool *); 122 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); 123 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, 124 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); 125 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, 126 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, 127 char *, int *); 128 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 129 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 130 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 131 struct vnode *, struct nfsmount **, int, struct acl *, int *); 132 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 133 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); 134 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 135 struct vnode *, struct nfsmount *, struct nfsvattr *); 136 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, 137 NFSPROC_T *, struct nfsmount *); 138 static int nfsrv_putfhname(fhandle_t *, char *); 139 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, 140 struct pnfsdsfile *, struct vnode **, NFSPROC_T *); 141 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, 142 struct vnode *, NFSPROC_T *); 143 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); 144 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, 145 NFSPROC_T *); 146 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); 147 148 int nfs_pnfsio(task_fn_t *, void *); 149 150 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 151 "NFS server"); 152 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, 153 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); 154 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 155 0, ""); 156 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 157 0, ""); 158 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, 159 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); 160 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, 161 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); 162 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 163 0, "Debug level for NFS server"); 164 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW, 165 &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names"); 166 static int nfsrv_pnfsgetdsattr = 1; 167 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, 168 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); 169 170 /* 171 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are 172 * not running. 173 * The dsN subdirectories for the increased values must have been created 174 * on all DS servers before this increase is done. 175 */ 176 u_int nfsrv_dsdirsize = 20; 177 static int 178 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) 179 { 180 int error, newdsdirsize; 181 182 newdsdirsize = nfsrv_dsdirsize; 183 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); 184 if (error != 0 || req->newptr == NULL) 185 return (error); 186 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || 187 newnfs_numnfsd != 0) 188 return (EINVAL); 189 nfsrv_dsdirsize = newdsdirsize; 190 return (0); 191 } 192 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, 193 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), 194 sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); 195 196 #define MAX_REORDERED_RPC 16 197 #define NUM_HEURISTIC 1031 198 #define NHUSE_INIT 64 199 #define NHUSE_INC 16 200 #define NHUSE_MAX 2048 201 202 static struct nfsheur { 203 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 204 off_t nh_nextoff; /* next offset for sequential detection */ 205 int nh_use; /* use count for selection */ 206 int nh_seqcount; /* heuristic */ 207 } nfsheur[NUM_HEURISTIC]; 208 209 210 /* 211 * Heuristic to detect sequential operation. 212 */ 213 static struct nfsheur * 214 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 215 { 216 struct nfsheur *nh; 217 int hi, try; 218 219 /* Locate best candidate. */ 220 try = 32; 221 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 222 nh = &nfsheur[hi]; 223 while (try--) { 224 if (nfsheur[hi].nh_vp == vp) { 225 nh = &nfsheur[hi]; 226 break; 227 } 228 if (nfsheur[hi].nh_use > 0) 229 --nfsheur[hi].nh_use; 230 hi = (hi + 1) % NUM_HEURISTIC; 231 if (nfsheur[hi].nh_use < nh->nh_use) 232 nh = &nfsheur[hi]; 233 } 234 235 /* Initialize hint if this is a new file. */ 236 if (nh->nh_vp != vp) { 237 nh->nh_vp = vp; 238 nh->nh_nextoff = uio->uio_offset; 239 nh->nh_use = NHUSE_INIT; 240 if (uio->uio_offset == 0) 241 nh->nh_seqcount = 4; 242 else 243 nh->nh_seqcount = 1; 244 } 245 246 /* Calculate heuristic. */ 247 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 248 uio->uio_offset == nh->nh_nextoff) { 249 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 250 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 251 if (nh->nh_seqcount > IO_SEQMAX) 252 nh->nh_seqcount = IO_SEQMAX; 253 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 254 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 255 /* Probably a reordered RPC, leave seqcount alone. */ 256 } else if (nh->nh_seqcount > 1) { 257 nh->nh_seqcount /= 2; 258 } else { 259 nh->nh_seqcount = 0; 260 } 261 nh->nh_use += NHUSE_INC; 262 if (nh->nh_use > NHUSE_MAX) 263 nh->nh_use = NHUSE_MAX; 264 return (nh); 265 } 266 267 /* 268 * Get attributes into nfsvattr structure. 269 */ 270 int 271 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, 272 struct nfsrv_descript *nd, struct thread *p, int vpislocked, 273 nfsattrbit_t *attrbitp) 274 { 275 int error, gotattr, lockedit = 0; 276 struct nfsvattr na; 277 278 if (vpislocked == 0) { 279 /* 280 * When vpislocked == 0, the vnode is either exclusively 281 * locked by this thread or not locked by this thread. 282 * As such, shared lock it, if not exclusively locked. 283 */ 284 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 285 lockedit = 1; 286 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); 287 } 288 } 289 290 /* 291 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed 292 * attributes, as required. 293 * This needs to be done for regular files if: 294 * - non-NFSv4 RPCs or 295 * - when attrbitp == NULL or 296 * - an NFSv4 RPC with any of the above attributes in attrbitp. 297 * A return of 0 for nfsrv_proxyds() indicates that it has acquired 298 * these attributes. nfsrv_proxyds() will return an error if the 299 * server is not a pNFS one. 300 */ 301 gotattr = 0; 302 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || 303 (nd->nd_flag & ND_NFSV4) == 0 || 304 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || 305 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || 306 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || 307 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || 308 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { 309 error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, 310 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, 311 NULL); 312 if (error == 0) 313 gotattr = 1; 314 } 315 316 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); 317 if (lockedit != 0) 318 NFSVOPUNLOCK(vp); 319 320 /* 321 * If we got the Change, Size and Modify Time from the DS, 322 * replace them. 323 */ 324 if (gotattr != 0) { 325 nvap->na_atime = na.na_atime; 326 nvap->na_mtime = na.na_mtime; 327 nvap->na_filerev = na.na_filerev; 328 nvap->na_size = na.na_size; 329 nvap->na_bytes = na.na_bytes; 330 } 331 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, 332 error, (uintmax_t)na.na_filerev); 333 334 NFSEXITCODE(error); 335 return (error); 336 } 337 338 /* 339 * Get a file handle for a vnode. 340 */ 341 int 342 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) 343 { 344 int error; 345 346 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); 347 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 348 error = VOP_VPTOFH(vp, &fhp->fh_fid); 349 350 NFSEXITCODE(error); 351 return (error); 352 } 353 354 /* 355 * Perform access checking for vnodes obtained from file handles that would 356 * refer to files already opened by a Unix client. You cannot just use 357 * vn_writechk() and VOP_ACCESSX() for two reasons. 358 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 359 * case. 360 * 2 - The owner is to be given access irrespective of mode bits for some 361 * operations, so that processes that chmod after opening a file don't 362 * break. 363 */ 364 int 365 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, 366 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, 367 u_int32_t *supportedtypep) 368 { 369 struct vattr vattr; 370 int error = 0, getret = 0; 371 372 if (vpislocked == 0) { 373 if (NFSVOPLOCK(vp, LK_SHARED) != 0) { 374 error = EPERM; 375 goto out; 376 } 377 } 378 if (accmode & VWRITE) { 379 /* Just vn_writechk() changed to check rdonly */ 380 /* 381 * Disallow write attempts on read-only file systems; 382 * unless the file is a socket or a block or character 383 * device resident on the file system. 384 */ 385 if (NFSVNO_EXRDONLY(exp) || 386 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 387 switch (vp->v_type) { 388 case VREG: 389 case VDIR: 390 case VLNK: 391 error = EROFS; 392 default: 393 break; 394 } 395 } 396 /* 397 * If there's shared text associated with 398 * the inode, try to free it up once. If 399 * we fail, we can't allow writing. 400 */ 401 if (VOP_IS_TEXT(vp) && error == 0) 402 error = ETXTBSY; 403 } 404 if (error != 0) { 405 if (vpislocked == 0) 406 NFSVOPUNLOCK(vp); 407 goto out; 408 } 409 410 /* 411 * Should the override still be applied when ACLs are enabled? 412 */ 413 error = VOP_ACCESSX(vp, accmode, cred, p); 414 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { 415 /* 416 * Try again with VEXPLICIT_DENY, to see if the test for 417 * deletion is supported. 418 */ 419 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); 420 if (error == 0) { 421 if (vp->v_type == VDIR) { 422 accmode &= ~(VDELETE | VDELETE_CHILD); 423 accmode |= VWRITE; 424 error = VOP_ACCESSX(vp, accmode, cred, p); 425 } else if (supportedtypep != NULL) { 426 *supportedtypep &= ~NFSACCESS_DELETE; 427 } 428 } 429 } 430 431 /* 432 * Allow certain operations for the owner (reads and writes 433 * on files that are already open). 434 */ 435 if (override != NFSACCCHK_NOOVERRIDE && 436 (error == EPERM || error == EACCES)) { 437 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) 438 error = 0; 439 else if (override & NFSACCCHK_ALLOWOWNER) { 440 getret = VOP_GETATTR(vp, &vattr, cred); 441 if (getret == 0 && cred->cr_uid == vattr.va_uid) 442 error = 0; 443 } 444 } 445 if (vpislocked == 0) 446 NFSVOPUNLOCK(vp); 447 448 out: 449 NFSEXITCODE(error); 450 return (error); 451 } 452 453 /* 454 * Set attribute(s) vnop. 455 */ 456 int 457 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 458 struct thread *p, struct nfsexstuff *exp) 459 { 460 u_quad_t savsize = 0; 461 int error, savedit; 462 463 /* 464 * If this is an exported file system and a pNFS service is running, 465 * don't VOP_SETATTR() of size for the MDS file system. 466 */ 467 savedit = 0; 468 error = 0; 469 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && 470 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && 471 nvap->na_vattr.va_size > 0) { 472 savsize = nvap->na_vattr.va_size; 473 nvap->na_vattr.va_size = VNOVAL; 474 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 475 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 476 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 477 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 478 nvap->na_vattr.va_mtime.tv_sec != VNOVAL) 479 savedit = 1; 480 else 481 savedit = 2; 482 } 483 if (savedit != 2) 484 error = VOP_SETATTR(vp, &nvap->na_vattr, cred); 485 if (savedit != 0) 486 nvap->na_vattr.va_size = savsize; 487 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 488 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 489 nvap->na_vattr.va_size != VNOVAL || 490 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 491 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 492 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { 493 /* For a pNFS server, set the attributes on the DS file. */ 494 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, 495 NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); 496 if (error == ENOENT) 497 error = 0; 498 } 499 NFSEXITCODE(error); 500 return (error); 501 } 502 503 /* 504 * Set up nameidata for a lookup() call and do it. 505 */ 506 int 507 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, 508 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p, 509 struct vnode **retdirp) 510 { 511 struct componentname *cnp = &ndp->ni_cnd; 512 int i; 513 struct iovec aiov; 514 struct uio auio; 515 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; 516 int error = 0; 517 char *cp; 518 519 *retdirp = NULL; 520 cnp->cn_nameptr = cnp->cn_pnbuf; 521 ndp->ni_lcf = 0; 522 /* 523 * Extract and set starting directory. 524 */ 525 if (dp->v_type != VDIR) { 526 if (islocked) 527 vput(dp); 528 else 529 vrele(dp); 530 nfsvno_relpathbuf(ndp); 531 error = ENOTDIR; 532 goto out1; 533 } 534 if (islocked) 535 NFSVOPUNLOCK(dp); 536 VREF(dp); 537 *retdirp = dp; 538 if (NFSVNO_EXRDONLY(exp)) 539 cnp->cn_flags |= RDONLY; 540 ndp->ni_segflg = UIO_SYSSPACE; 541 542 if (nd->nd_flag & ND_PUBLOOKUP) { 543 ndp->ni_loopcnt = 0; 544 if (cnp->cn_pnbuf[0] == '/') { 545 vrele(dp); 546 /* 547 * Check for degenerate pathnames here, since lookup() 548 * panics on them. 549 */ 550 for (i = 1; i < ndp->ni_pathlen; i++) 551 if (cnp->cn_pnbuf[i] != '/') 552 break; 553 if (i == ndp->ni_pathlen) { 554 error = NFSERR_ACCES; 555 goto out; 556 } 557 dp = rootvnode; 558 VREF(dp); 559 } 560 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || 561 (nd->nd_flag & ND_NFSV4) == 0) { 562 /* 563 * Only cross mount points for NFSv4 when doing a 564 * mount while traversing the file system above 565 * the mount point, unless nfsrv_enable_crossmntpt is set. 566 */ 567 cnp->cn_flags |= NOCROSSMOUNT; 568 } 569 570 /* 571 * Initialize for scan, set ni_startdir and bump ref on dp again 572 * because lookup() will dereference ni_startdir. 573 */ 574 575 cnp->cn_thread = p; 576 ndp->ni_startdir = dp; 577 ndp->ni_rootdir = rootvnode; 578 ndp->ni_topdir = NULL; 579 580 if (!lockleaf) 581 cnp->cn_flags |= LOCKLEAF; 582 for (;;) { 583 cnp->cn_nameptr = cnp->cn_pnbuf; 584 /* 585 * Call lookup() to do the real work. If an error occurs, 586 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and 587 * we do not have to dereference anything before returning. 588 * In either case ni_startdir will be dereferenced and NULLed 589 * out. 590 */ 591 error = lookup(ndp); 592 if (error) 593 break; 594 595 /* 596 * Check for encountering a symbolic link. Trivial 597 * termination occurs if no symlink encountered. 598 */ 599 if ((cnp->cn_flags & ISSYMLINK) == 0) { 600 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) 601 nfsvno_relpathbuf(ndp); 602 if (ndp->ni_vp && !lockleaf) 603 NFSVOPUNLOCK(ndp->ni_vp); 604 break; 605 } 606 607 /* 608 * Validate symlink 609 */ 610 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) 611 NFSVOPUNLOCK(ndp->ni_dvp); 612 if (!(nd->nd_flag & ND_PUBLOOKUP)) { 613 error = EINVAL; 614 goto badlink2; 615 } 616 617 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 618 error = ELOOP; 619 goto badlink2; 620 } 621 if (ndp->ni_pathlen > 1) 622 cp = uma_zalloc(namei_zone, M_WAITOK); 623 else 624 cp = cnp->cn_pnbuf; 625 aiov.iov_base = cp; 626 aiov.iov_len = MAXPATHLEN; 627 auio.uio_iov = &aiov; 628 auio.uio_iovcnt = 1; 629 auio.uio_offset = 0; 630 auio.uio_rw = UIO_READ; 631 auio.uio_segflg = UIO_SYSSPACE; 632 auio.uio_td = NULL; 633 auio.uio_resid = MAXPATHLEN; 634 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 635 if (error) { 636 badlink1: 637 if (ndp->ni_pathlen > 1) 638 uma_zfree(namei_zone, cp); 639 badlink2: 640 vrele(ndp->ni_dvp); 641 vput(ndp->ni_vp); 642 break; 643 } 644 linklen = MAXPATHLEN - auio.uio_resid; 645 if (linklen == 0) { 646 error = ENOENT; 647 goto badlink1; 648 } 649 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 650 error = ENAMETOOLONG; 651 goto badlink1; 652 } 653 654 /* 655 * Adjust or replace path 656 */ 657 if (ndp->ni_pathlen > 1) { 658 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 659 uma_zfree(namei_zone, cnp->cn_pnbuf); 660 cnp->cn_pnbuf = cp; 661 } else 662 cnp->cn_pnbuf[linklen] = '\0'; 663 ndp->ni_pathlen += linklen; 664 665 /* 666 * Cleanup refs for next loop and check if root directory 667 * should replace current directory. Normally ni_dvp 668 * becomes the new base directory and is cleaned up when 669 * we loop. Explicitly null pointers after invalidation 670 * to clarify operation. 671 */ 672 vput(ndp->ni_vp); 673 ndp->ni_vp = NULL; 674 675 if (cnp->cn_pnbuf[0] == '/') { 676 vrele(ndp->ni_dvp); 677 ndp->ni_dvp = ndp->ni_rootdir; 678 VREF(ndp->ni_dvp); 679 } 680 ndp->ni_startdir = ndp->ni_dvp; 681 ndp->ni_dvp = NULL; 682 } 683 if (!lockleaf) 684 cnp->cn_flags &= ~LOCKLEAF; 685 686 out: 687 if (error) { 688 nfsvno_relpathbuf(ndp); 689 ndp->ni_vp = NULL; 690 ndp->ni_dvp = NULL; 691 ndp->ni_startdir = NULL; 692 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { 693 ndp->ni_dvp = NULL; 694 } 695 696 out1: 697 NFSEXITCODE2(error, nd); 698 return (error); 699 } 700 701 /* 702 * Set up a pathname buffer and return a pointer to it and, optionally 703 * set a hash pointer. 704 */ 705 void 706 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) 707 { 708 struct componentname *cnp = &ndp->ni_cnd; 709 710 cnp->cn_flags |= (NOMACCHECK | HASBUF); 711 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 712 if (hashpp != NULL) 713 *hashpp = NULL; 714 *bufpp = cnp->cn_pnbuf; 715 } 716 717 /* 718 * Release the above path buffer, if not released by nfsvno_namei(). 719 */ 720 void 721 nfsvno_relpathbuf(struct nameidata *ndp) 722 { 723 724 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0) 725 panic("nfsrelpath"); 726 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 727 ndp->ni_cnd.cn_flags &= ~HASBUF; 728 } 729 730 /* 731 * Readlink vnode op into an mbuf list. 732 */ 733 int 734 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p, 735 struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 736 { 737 struct iovec *iv; 738 struct uio io, *uiop = &io; 739 struct mbuf *mp, *mp3; 740 int len, tlen, error = 0; 741 742 len = NFS_MAXPATHLEN; 743 uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); 744 uiop->uio_iov = iv; 745 uiop->uio_offset = 0; 746 uiop->uio_resid = len; 747 uiop->uio_rw = UIO_READ; 748 uiop->uio_segflg = UIO_SYSSPACE; 749 uiop->uio_td = NULL; 750 error = VOP_READLINK(vp, uiop, cred); 751 free(iv, M_TEMP); 752 if (error) { 753 m_freem(mp3); 754 *lenp = 0; 755 goto out; 756 } 757 if (uiop->uio_resid > 0) { 758 len -= uiop->uio_resid; 759 tlen = NFSM_RNDUP(len); 760 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len); 761 } 762 *lenp = len; 763 *mpp = mp3; 764 *mpendp = mp; 765 766 out: 767 NFSEXITCODE(error); 768 return (error); 769 } 770 771 /* 772 * Create an mbuf chain and an associated iovec that can be used to Read 773 * or Getextattr of data. 774 * Upon success, return pointers to the first and last mbufs in the chain 775 * plus the malloc'd iovec and its iovlen. 776 */ 777 static int 778 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, 779 struct iovec **ivp) 780 { 781 struct mbuf *m, *m2 = NULL, *m3; 782 struct iovec *iv; 783 int i, left, siz; 784 785 left = len; 786 m3 = NULL; 787 /* 788 * Generate the mbuf list with the uio_iov ref. to it. 789 */ 790 i = 0; 791 while (left > 0) { 792 NFSMGET(m); 793 MCLGET(m, M_WAITOK); 794 m->m_len = 0; 795 siz = min(M_TRAILINGSPACE(m), left); 796 left -= siz; 797 i++; 798 if (m3) 799 m2->m_next = m; 800 else 801 m3 = m; 802 m2 = m; 803 } 804 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 805 m = m3; 806 left = len; 807 i = 0; 808 while (left > 0) { 809 if (m == NULL) 810 panic("nfsvno_read iov"); 811 siz = min(M_TRAILINGSPACE(m), left); 812 if (siz > 0) { 813 iv->iov_base = mtod(m, caddr_t) + m->m_len; 814 iv->iov_len = siz; 815 m->m_len += siz; 816 left -= siz; 817 iv++; 818 i++; 819 } 820 m = m->m_next; 821 } 822 *mpp = m3; 823 *mpendp = m2; 824 return (i); 825 } 826 827 /* 828 * Read vnode op call into mbuf list. 829 */ 830 int 831 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 832 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp) 833 { 834 struct mbuf *m; 835 struct iovec *iv; 836 int error = 0, len, tlen, ioflag = 0; 837 struct mbuf *m3; 838 struct uio io, *uiop = &io; 839 struct nfsheur *nh; 840 841 /* 842 * Attempt to read from a DS file. A return of ENOENT implies 843 * there is no DS file to read. 844 */ 845 error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, 846 NULL, mpendp, NULL, NULL, NULL, 0, NULL); 847 if (error != ENOENT) 848 return (error); 849 850 len = NFSM_RNDUP(cnt); 851 uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); 852 uiop->uio_iov = iv; 853 uiop->uio_offset = off; 854 uiop->uio_resid = len; 855 uiop->uio_rw = UIO_READ; 856 uiop->uio_segflg = UIO_SYSSPACE; 857 uiop->uio_td = NULL; 858 nh = nfsrv_sequential_heuristic(uiop, vp); 859 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 860 /* XXX KDM make this more systematic? */ 861 nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid; 862 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 863 free(iv, M_TEMP); 864 if (error) { 865 m_freem(m3); 866 *mpp = NULL; 867 goto out; 868 } 869 nh->nh_nextoff = uiop->uio_offset; 870 tlen = len - uiop->uio_resid; 871 cnt = cnt < tlen ? cnt : tlen; 872 tlen = NFSM_RNDUP(cnt); 873 if (tlen == 0) { 874 m_freem(m3); 875 m3 = NULL; 876 } else if (len != tlen || tlen != cnt) 877 nfsrv_adj(m3, len - tlen, tlen - cnt); 878 *mpp = m3; 879 *mpendp = m; 880 881 out: 882 NFSEXITCODE(error); 883 return (error); 884 } 885 886 /* 887 * Create the iovec for the mbuf chain passed in as an argument. 888 * The "cp" argument is where the data starts within the first mbuf in 889 * the chain. It returns the iovec and the iovcnt. 890 */ 891 static int 892 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, 893 int *iovcntp) 894 { 895 struct mbuf *mp; 896 struct iovec *ivp; 897 int cnt, i, len; 898 899 /* 900 * Loop through the mbuf chain, counting how many mbufs are a 901 * part of this write operation, so the iovec size is known. 902 */ 903 cnt = 0; 904 len = retlen; 905 mp = m; 906 i = mtod(mp, caddr_t) + mp->m_len - cp; 907 while (len > 0) { 908 if (i > 0) { 909 len -= i; 910 cnt++; 911 } 912 mp = mp->m_next; 913 if (!mp) { 914 if (len > 0) 915 return (EBADRPC); 916 } else 917 i = mp->m_len; 918 } 919 920 /* Now, create the iovec. */ 921 mp = m; 922 *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, 923 M_WAITOK); 924 *iovcntp = cnt; 925 i = mtod(mp, caddr_t) + mp->m_len - cp; 926 len = retlen; 927 while (len > 0) { 928 if (mp == NULL) 929 panic("nfsvno_write"); 930 if (i > 0) { 931 i = min(i, len); 932 ivp->iov_base = cp; 933 ivp->iov_len = i; 934 ivp++; 935 len -= i; 936 } 937 mp = mp->m_next; 938 if (mp) { 939 i = mp->m_len; 940 cp = mtod(mp, caddr_t); 941 } 942 } 943 return (0); 944 } 945 946 /* 947 * Write vnode op from an mbuf list. 948 */ 949 int 950 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, 951 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) 952 { 953 struct iovec *iv; 954 int cnt, ioflags, error; 955 struct uio io, *uiop = &io; 956 struct nfsheur *nh; 957 958 /* 959 * Attempt to write to a DS file. A return of ENOENT implies 960 * there is no DS file to write. 961 */ 962 error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, 963 &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); 964 if (error != ENOENT) { 965 *stable = NFSWRITE_FILESYNC; 966 return (error); 967 } 968 969 970 if (*stable == NFSWRITE_UNSTABLE) 971 ioflags = IO_NODELOCKED; 972 else 973 ioflags = (IO_SYNC | IO_NODELOCKED); 974 error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); 975 if (error != 0) 976 return (error); 977 uiop->uio_iov = iv; 978 uiop->uio_iovcnt = cnt; 979 uiop->uio_resid = retlen; 980 uiop->uio_rw = UIO_WRITE; 981 uiop->uio_segflg = UIO_SYSSPACE; 982 NFSUIOPROC(uiop, p); 983 uiop->uio_offset = off; 984 nh = nfsrv_sequential_heuristic(uiop, vp); 985 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 986 /* XXX KDM make this more systematic? */ 987 nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; 988 error = VOP_WRITE(vp, uiop, ioflags, cred); 989 if (error == 0) 990 nh->nh_nextoff = uiop->uio_offset; 991 free(iv, M_TEMP); 992 993 NFSEXITCODE(error); 994 return (error); 995 } 996 997 /* 998 * Common code for creating a regular file (plus special files for V2). 999 */ 1000 int 1001 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, 1002 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, 1003 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) 1004 { 1005 u_quad_t tempsize; 1006 int error; 1007 struct thread *p = curthread; 1008 1009 error = nd->nd_repstat; 1010 if (!error && ndp->ni_vp == NULL) { 1011 if (nvap->na_type == VREG || nvap->na_type == VSOCK) { 1012 vrele(ndp->ni_startdir); 1013 error = VOP_CREATE(ndp->ni_dvp, 1014 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1015 /* For a pNFS server, create the data file on a DS. */ 1016 if (error == 0 && nvap->na_type == VREG) { 1017 /* 1018 * Create a data file on a DS for a pNFS server. 1019 * This function just returns if not 1020 * running a pNFS DS or the creation fails. 1021 */ 1022 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1023 nd->nd_cred, p); 1024 } 1025 vput(ndp->ni_dvp); 1026 nfsvno_relpathbuf(ndp); 1027 if (!error) { 1028 if (*exclusive_flagp) { 1029 *exclusive_flagp = 0; 1030 NFSVNO_ATTRINIT(nvap); 1031 nvap->na_atime.tv_sec = cverf[0]; 1032 nvap->na_atime.tv_nsec = cverf[1]; 1033 error = VOP_SETATTR(ndp->ni_vp, 1034 &nvap->na_vattr, nd->nd_cred); 1035 if (error != 0) { 1036 vput(ndp->ni_vp); 1037 ndp->ni_vp = NULL; 1038 error = NFSERR_NOTSUPP; 1039 } 1040 } 1041 } 1042 /* 1043 * NFS V2 Only. nfsrvd_mknod() does this for V3. 1044 * (This implies, just get out on an error.) 1045 */ 1046 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || 1047 nvap->na_type == VFIFO) { 1048 if (nvap->na_type == VCHR && rdev == 0xffffffff) 1049 nvap->na_type = VFIFO; 1050 if (nvap->na_type != VFIFO && 1051 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { 1052 vrele(ndp->ni_startdir); 1053 nfsvno_relpathbuf(ndp); 1054 vput(ndp->ni_dvp); 1055 goto out; 1056 } 1057 nvap->na_rdev = rdev; 1058 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1059 &ndp->ni_cnd, &nvap->na_vattr); 1060 vput(ndp->ni_dvp); 1061 nfsvno_relpathbuf(ndp); 1062 vrele(ndp->ni_startdir); 1063 if (error) 1064 goto out; 1065 } else { 1066 vrele(ndp->ni_startdir); 1067 nfsvno_relpathbuf(ndp); 1068 vput(ndp->ni_dvp); 1069 error = ENXIO; 1070 goto out; 1071 } 1072 *vpp = ndp->ni_vp; 1073 } else { 1074 /* 1075 * Handle cases where error is already set and/or 1076 * the file exists. 1077 * 1 - clean up the lookup 1078 * 2 - iff !error and na_size set, truncate it 1079 */ 1080 vrele(ndp->ni_startdir); 1081 nfsvno_relpathbuf(ndp); 1082 *vpp = ndp->ni_vp; 1083 if (ndp->ni_dvp == *vpp) 1084 vrele(ndp->ni_dvp); 1085 else 1086 vput(ndp->ni_dvp); 1087 if (!error && nvap->na_size != VNOVAL) { 1088 error = nfsvno_accchk(*vpp, VWRITE, 1089 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1090 NFSACCCHK_VPISLOCKED, NULL); 1091 if (!error) { 1092 tempsize = nvap->na_size; 1093 NFSVNO_ATTRINIT(nvap); 1094 nvap->na_size = tempsize; 1095 error = VOP_SETATTR(*vpp, 1096 &nvap->na_vattr, nd->nd_cred); 1097 } 1098 } 1099 if (error) 1100 vput(*vpp); 1101 } 1102 1103 out: 1104 NFSEXITCODE(error); 1105 return (error); 1106 } 1107 1108 /* 1109 * Do a mknod vnode op. 1110 */ 1111 int 1112 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, 1113 struct thread *p) 1114 { 1115 int error = 0; 1116 enum vtype vtyp; 1117 1118 vtyp = nvap->na_type; 1119 /* 1120 * Iff doesn't exist, create it. 1121 */ 1122 if (ndp->ni_vp) { 1123 vrele(ndp->ni_startdir); 1124 nfsvno_relpathbuf(ndp); 1125 vput(ndp->ni_dvp); 1126 vrele(ndp->ni_vp); 1127 error = EEXIST; 1128 goto out; 1129 } 1130 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1131 vrele(ndp->ni_startdir); 1132 nfsvno_relpathbuf(ndp); 1133 vput(ndp->ni_dvp); 1134 error = NFSERR_BADTYPE; 1135 goto out; 1136 } 1137 if (vtyp == VSOCK) { 1138 vrele(ndp->ni_startdir); 1139 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 1140 &ndp->ni_cnd, &nvap->na_vattr); 1141 vput(ndp->ni_dvp); 1142 nfsvno_relpathbuf(ndp); 1143 } else { 1144 if (nvap->na_type != VFIFO && 1145 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { 1146 vrele(ndp->ni_startdir); 1147 nfsvno_relpathbuf(ndp); 1148 vput(ndp->ni_dvp); 1149 goto out; 1150 } 1151 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1152 &ndp->ni_cnd, &nvap->na_vattr); 1153 vput(ndp->ni_dvp); 1154 nfsvno_relpathbuf(ndp); 1155 vrele(ndp->ni_startdir); 1156 /* 1157 * Since VOP_MKNOD returns the ni_vp, I can't 1158 * see any reason to do the lookup. 1159 */ 1160 } 1161 1162 out: 1163 NFSEXITCODE(error); 1164 return (error); 1165 } 1166 1167 /* 1168 * Mkdir vnode op. 1169 */ 1170 int 1171 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, 1172 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 1173 { 1174 int error = 0; 1175 1176 if (ndp->ni_vp != NULL) { 1177 if (ndp->ni_dvp == ndp->ni_vp) 1178 vrele(ndp->ni_dvp); 1179 else 1180 vput(ndp->ni_dvp); 1181 vrele(ndp->ni_vp); 1182 nfsvno_relpathbuf(ndp); 1183 error = EEXIST; 1184 goto out; 1185 } 1186 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1187 &nvap->na_vattr); 1188 vput(ndp->ni_dvp); 1189 nfsvno_relpathbuf(ndp); 1190 1191 out: 1192 NFSEXITCODE(error); 1193 return (error); 1194 } 1195 1196 /* 1197 * symlink vnode op. 1198 */ 1199 int 1200 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, 1201 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, 1202 struct nfsexstuff *exp) 1203 { 1204 int error = 0; 1205 1206 if (ndp->ni_vp) { 1207 vrele(ndp->ni_startdir); 1208 nfsvno_relpathbuf(ndp); 1209 if (ndp->ni_dvp == ndp->ni_vp) 1210 vrele(ndp->ni_dvp); 1211 else 1212 vput(ndp->ni_dvp); 1213 vrele(ndp->ni_vp); 1214 error = EEXIST; 1215 goto out; 1216 } 1217 1218 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1219 &nvap->na_vattr, pathcp); 1220 vput(ndp->ni_dvp); 1221 vrele(ndp->ni_startdir); 1222 nfsvno_relpathbuf(ndp); 1223 /* 1224 * Although FreeBSD still had the lookup code in 1225 * it for 7/current, there doesn't seem to be any 1226 * point, since VOP_SYMLINK() returns the ni_vp. 1227 * Just vput it for v2. 1228 */ 1229 if (!not_v2 && !error) 1230 vput(ndp->ni_vp); 1231 1232 out: 1233 NFSEXITCODE(error); 1234 return (error); 1235 } 1236 1237 /* 1238 * Parse symbolic link arguments. 1239 * This function has an ugly side effect. It will malloc() an area for 1240 * the symlink and set iov_base to point to it, only if it succeeds. 1241 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must 1242 * be FREE'd later. 1243 */ 1244 int 1245 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, 1246 struct thread *p, char **pathcpp, int *lenp) 1247 { 1248 u_int32_t *tl; 1249 char *pathcp = NULL; 1250 int error = 0, len; 1251 struct nfsv2_sattr *sp; 1252 1253 *pathcpp = NULL; 1254 *lenp = 0; 1255 if ((nd->nd_flag & ND_NFSV3) && 1256 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) 1257 goto nfsmout; 1258 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 1259 len = fxdr_unsigned(int, *tl); 1260 if (len > NFS_MAXPATHLEN || len <= 0) { 1261 error = EBADRPC; 1262 goto nfsmout; 1263 } 1264 pathcp = malloc(len + 1, M_TEMP, M_WAITOK); 1265 error = nfsrv_mtostr(nd, pathcp, len); 1266 if (error) 1267 goto nfsmout; 1268 if (nd->nd_flag & ND_NFSV2) { 1269 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1270 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 1271 } 1272 *pathcpp = pathcp; 1273 *lenp = len; 1274 NFSEXITCODE2(0, nd); 1275 return (0); 1276 nfsmout: 1277 if (pathcp) 1278 free(pathcp, M_TEMP); 1279 NFSEXITCODE2(error, nd); 1280 return (error); 1281 } 1282 1283 /* 1284 * Remove a non-directory object. 1285 */ 1286 int 1287 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1288 struct thread *p, struct nfsexstuff *exp) 1289 { 1290 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS]; 1291 int error = 0, mirrorcnt; 1292 char fname[PNFS_FILENAME_LEN + 1]; 1293 fhandle_t fh; 1294 1295 vp = ndp->ni_vp; 1296 dsdvp[0] = NULL; 1297 if (vp->v_type == VDIR) 1298 error = NFSERR_ISDIR; 1299 else if (is_v4) 1300 error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0), 1301 p); 1302 if (error == 0) 1303 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); 1304 if (!error) 1305 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); 1306 if (error == 0 && dsdvp[0] != NULL) 1307 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1308 if (ndp->ni_dvp == vp) 1309 vrele(ndp->ni_dvp); 1310 else 1311 vput(ndp->ni_dvp); 1312 vput(vp); 1313 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) 1314 nfsvno_relpathbuf(ndp); 1315 NFSEXITCODE(error); 1316 return (error); 1317 } 1318 1319 /* 1320 * Remove a directory. 1321 */ 1322 int 1323 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1324 struct thread *p, struct nfsexstuff *exp) 1325 { 1326 struct vnode *vp; 1327 int error = 0; 1328 1329 vp = ndp->ni_vp; 1330 if (vp->v_type != VDIR) { 1331 error = ENOTDIR; 1332 goto out; 1333 } 1334 /* 1335 * No rmdir "." please. 1336 */ 1337 if (ndp->ni_dvp == vp) { 1338 error = EINVAL; 1339 goto out; 1340 } 1341 /* 1342 * The root of a mounted filesystem cannot be deleted. 1343 */ 1344 if (vp->v_vflag & VV_ROOT) 1345 error = EBUSY; 1346 out: 1347 if (!error) 1348 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); 1349 if (ndp->ni_dvp == vp) 1350 vrele(ndp->ni_dvp); 1351 else 1352 vput(ndp->ni_dvp); 1353 vput(vp); 1354 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) 1355 nfsvno_relpathbuf(ndp); 1356 NFSEXITCODE(error); 1357 return (error); 1358 } 1359 1360 /* 1361 * Rename vnode op. 1362 */ 1363 int 1364 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, 1365 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) 1366 { 1367 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS]; 1368 int error = 0, mirrorcnt; 1369 char fname[PNFS_FILENAME_LEN + 1]; 1370 fhandle_t fh; 1371 1372 dsdvp[0] = NULL; 1373 fvp = fromndp->ni_vp; 1374 if (ndstat) { 1375 vrele(fromndp->ni_dvp); 1376 vrele(fvp); 1377 error = ndstat; 1378 goto out1; 1379 } 1380 tdvp = tondp->ni_dvp; 1381 tvp = tondp->ni_vp; 1382 if (tvp != NULL) { 1383 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1384 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; 1385 goto out; 1386 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1387 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; 1388 goto out; 1389 } 1390 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1391 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1392 goto out; 1393 } 1394 1395 /* 1396 * A rename to '.' or '..' results in a prematurely 1397 * unlocked vnode on FreeBSD5, so I'm just going to fail that 1398 * here. 1399 */ 1400 if ((tondp->ni_cnd.cn_namelen == 1 && 1401 tondp->ni_cnd.cn_nameptr[0] == '.') || 1402 (tondp->ni_cnd.cn_namelen == 2 && 1403 tondp->ni_cnd.cn_nameptr[0] == '.' && 1404 tondp->ni_cnd.cn_nameptr[1] == '.')) { 1405 error = EINVAL; 1406 goto out; 1407 } 1408 } 1409 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1410 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1411 goto out; 1412 } 1413 if (fvp->v_mount != tdvp->v_mount) { 1414 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1415 goto out; 1416 } 1417 if (fvp == tdvp) { 1418 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; 1419 goto out; 1420 } 1421 if (fvp == tvp) { 1422 /* 1423 * If source and destination are the same, there is nothing to 1424 * do. Set error to -1 to indicate this. 1425 */ 1426 error = -1; 1427 goto out; 1428 } 1429 if (ndflag & ND_NFSV4) { 1430 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { 1431 error = nfsrv_checkremove(fvp, 0, NULL, 1432 (nfsquad_t)((u_quad_t)0), p); 1433 NFSVOPUNLOCK(fvp); 1434 } else 1435 error = EPERM; 1436 if (tvp && !error) 1437 error = nfsrv_checkremove(tvp, 1, NULL, 1438 (nfsquad_t)((u_quad_t)0), p); 1439 } else { 1440 /* 1441 * For NFSv2 and NFSv3, try to get rid of the delegation, so 1442 * that the NFSv4 client won't be confused by the rename. 1443 * Since nfsd_recalldelegation() can only be called on an 1444 * unlocked vnode at this point and fvp is the file that will 1445 * still exist after the rename, just do fvp. 1446 */ 1447 nfsd_recalldelegation(fvp, p); 1448 } 1449 if (error == 0 && tvp != NULL) { 1450 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh); 1451 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" 1452 " dsdvp=%p\n", dsdvp[0]); 1453 } 1454 out: 1455 if (!error) { 1456 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, 1457 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, 1458 &tondp->ni_cnd); 1459 } else { 1460 if (tdvp == tvp) 1461 vrele(tdvp); 1462 else 1463 vput(tdvp); 1464 if (tvp) 1465 vput(tvp); 1466 vrele(fromndp->ni_dvp); 1467 vrele(fvp); 1468 if (error == -1) 1469 error = 0; 1470 } 1471 1472 /* 1473 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and 1474 * if the rename succeeded, the DS file for the tvp needs to be 1475 * removed. 1476 */ 1477 if (error == 0 && dsdvp[0] != NULL) { 1478 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1479 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); 1480 } 1481 1482 vrele(tondp->ni_startdir); 1483 nfsvno_relpathbuf(tondp); 1484 out1: 1485 vrele(fromndp->ni_startdir); 1486 nfsvno_relpathbuf(fromndp); 1487 NFSEXITCODE(error); 1488 return (error); 1489 } 1490 1491 /* 1492 * Link vnode op. 1493 */ 1494 int 1495 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, 1496 struct thread *p, struct nfsexstuff *exp) 1497 { 1498 struct vnode *xp; 1499 int error = 0; 1500 1501 xp = ndp->ni_vp; 1502 if (xp != NULL) { 1503 error = EEXIST; 1504 } else { 1505 xp = ndp->ni_dvp; 1506 if (vp->v_mount != xp->v_mount) 1507 error = EXDEV; 1508 } 1509 if (!error) { 1510 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1511 if (!VN_IS_DOOMED(vp)) 1512 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); 1513 else 1514 error = EPERM; 1515 if (ndp->ni_dvp == vp) 1516 vrele(ndp->ni_dvp); 1517 else 1518 vput(ndp->ni_dvp); 1519 NFSVOPUNLOCK(vp); 1520 } else { 1521 if (ndp->ni_dvp == ndp->ni_vp) 1522 vrele(ndp->ni_dvp); 1523 else 1524 vput(ndp->ni_dvp); 1525 if (ndp->ni_vp) 1526 vrele(ndp->ni_vp); 1527 } 1528 nfsvno_relpathbuf(ndp); 1529 NFSEXITCODE(error); 1530 return (error); 1531 } 1532 1533 /* 1534 * Do the fsync() appropriate for the commit. 1535 */ 1536 int 1537 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, 1538 struct thread *td) 1539 { 1540 int error = 0; 1541 1542 /* 1543 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of 1544 * file is done. At this time VOP_FSYNC does not accept offset and 1545 * byte count parameters so call VOP_FSYNC the whole file for now. 1546 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. 1547 * File systems that do not use the buffer cache (as indicated 1548 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). 1549 */ 1550 if (cnt == 0 || cnt > MAX_COMMIT_COUNT || 1551 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { 1552 /* 1553 * Give up and do the whole thing 1554 */ 1555 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1556 VM_OBJECT_WLOCK(vp->v_object); 1557 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); 1558 VM_OBJECT_WUNLOCK(vp->v_object); 1559 } 1560 error = VOP_FSYNC(vp, MNT_WAIT, td); 1561 } else { 1562 /* 1563 * Locate and synchronously write any buffers that fall 1564 * into the requested range. Note: we are assuming that 1565 * f_iosize is a power of 2. 1566 */ 1567 int iosize = vp->v_mount->mnt_stat.f_iosize; 1568 int iomask = iosize - 1; 1569 struct bufobj *bo; 1570 daddr_t lblkno; 1571 1572 /* 1573 * Align to iosize boundary, super-align to page boundary. 1574 */ 1575 if (off & iomask) { 1576 cnt += off & iomask; 1577 off &= ~(u_quad_t)iomask; 1578 } 1579 if (off & PAGE_MASK) { 1580 cnt += off & PAGE_MASK; 1581 off &= ~(u_quad_t)PAGE_MASK; 1582 } 1583 lblkno = off / iosize; 1584 1585 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1586 VM_OBJECT_WLOCK(vp->v_object); 1587 vm_object_page_clean(vp->v_object, off, off + cnt, 1588 OBJPC_SYNC); 1589 VM_OBJECT_WUNLOCK(vp->v_object); 1590 } 1591 1592 bo = &vp->v_bufobj; 1593 BO_LOCK(bo); 1594 while (cnt > 0) { 1595 struct buf *bp; 1596 1597 /* 1598 * If we have a buffer and it is marked B_DELWRI we 1599 * have to lock and write it. Otherwise the prior 1600 * write is assumed to have already been committed. 1601 * 1602 * gbincore() can return invalid buffers now so we 1603 * have to check that bit as well (though B_DELWRI 1604 * should not be set if B_INVAL is set there could be 1605 * a race here since we haven't locked the buffer). 1606 */ 1607 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 1608 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 1609 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { 1610 BO_LOCK(bo); 1611 continue; /* retry */ 1612 } 1613 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 1614 B_DELWRI) { 1615 bremfree(bp); 1616 bp->b_flags &= ~B_ASYNC; 1617 bwrite(bp); 1618 ++nfs_commit_miss; 1619 } else 1620 BUF_UNLOCK(bp); 1621 BO_LOCK(bo); 1622 } 1623 ++nfs_commit_blks; 1624 if (cnt < iosize) 1625 break; 1626 cnt -= iosize; 1627 ++lblkno; 1628 } 1629 BO_UNLOCK(bo); 1630 } 1631 NFSEXITCODE(error); 1632 return (error); 1633 } 1634 1635 /* 1636 * Statfs vnode op. 1637 */ 1638 int 1639 nfsvno_statfs(struct vnode *vp, struct statfs *sf) 1640 { 1641 struct statfs *tsf; 1642 int error; 1643 1644 tsf = NULL; 1645 if (nfsrv_devidcnt > 0) { 1646 /* For a pNFS service, get the DS numbers. */ 1647 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); 1648 error = nfsrv_pnfsstatfs(tsf, vp->v_mount); 1649 if (error != 0) { 1650 free(tsf, M_TEMP); 1651 tsf = NULL; 1652 } 1653 } 1654 error = VFS_STATFS(vp->v_mount, sf); 1655 if (error == 0) { 1656 if (tsf != NULL) { 1657 sf->f_blocks = tsf->f_blocks; 1658 sf->f_bavail = tsf->f_bavail; 1659 sf->f_bfree = tsf->f_bfree; 1660 sf->f_bsize = tsf->f_bsize; 1661 } 1662 /* 1663 * Since NFS handles these values as unsigned on the 1664 * wire, there is no way to represent negative values, 1665 * so set them to 0. Without this, they will appear 1666 * to be very large positive values for clients like 1667 * Solaris10. 1668 */ 1669 if (sf->f_bavail < 0) 1670 sf->f_bavail = 0; 1671 if (sf->f_ffree < 0) 1672 sf->f_ffree = 0; 1673 } 1674 free(tsf, M_TEMP); 1675 NFSEXITCODE(error); 1676 return (error); 1677 } 1678 1679 /* 1680 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but 1681 * must handle nfsrv_opencheck() calls after any other access checks. 1682 */ 1683 void 1684 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, 1685 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, 1686 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, 1687 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, 1688 struct nfsexstuff *exp, struct vnode **vpp) 1689 { 1690 struct vnode *vp = NULL; 1691 u_quad_t tempsize; 1692 struct nfsexstuff nes; 1693 struct thread *p = curthread; 1694 1695 if (ndp->ni_vp == NULL) 1696 nd->nd_repstat = nfsrv_opencheck(clientid, 1697 stateidp, stp, NULL, nd, p, nd->nd_repstat); 1698 if (!nd->nd_repstat) { 1699 if (ndp->ni_vp == NULL) { 1700 vrele(ndp->ni_startdir); 1701 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, 1702 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1703 /* For a pNFS server, create the data file on a DS. */ 1704 if (nd->nd_repstat == 0) { 1705 /* 1706 * Create a data file on a DS for a pNFS server. 1707 * This function just returns if not 1708 * running a pNFS DS or the creation fails. 1709 */ 1710 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1711 cred, p); 1712 } 1713 vput(ndp->ni_dvp); 1714 nfsvno_relpathbuf(ndp); 1715 if (!nd->nd_repstat) { 1716 if (*exclusive_flagp) { 1717 *exclusive_flagp = 0; 1718 NFSVNO_ATTRINIT(nvap); 1719 nvap->na_atime.tv_sec = cverf[0]; 1720 nvap->na_atime.tv_nsec = cverf[1]; 1721 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, 1722 &nvap->na_vattr, cred); 1723 if (nd->nd_repstat != 0) { 1724 vput(ndp->ni_vp); 1725 ndp->ni_vp = NULL; 1726 nd->nd_repstat = NFSERR_NOTSUPP; 1727 } else 1728 NFSSETBIT_ATTRBIT(attrbitp, 1729 NFSATTRBIT_TIMEACCESS); 1730 } else { 1731 nfsrv_fixattr(nd, ndp->ni_vp, nvap, 1732 aclp, p, attrbitp, exp); 1733 } 1734 } 1735 vp = ndp->ni_vp; 1736 } else { 1737 if (ndp->ni_startdir) 1738 vrele(ndp->ni_startdir); 1739 nfsvno_relpathbuf(ndp); 1740 vp = ndp->ni_vp; 1741 if (create == NFSV4OPEN_CREATE) { 1742 if (ndp->ni_dvp == vp) 1743 vrele(ndp->ni_dvp); 1744 else 1745 vput(ndp->ni_dvp); 1746 } 1747 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { 1748 if (ndp->ni_cnd.cn_flags & RDONLY) 1749 NFSVNO_SETEXRDONLY(&nes); 1750 else 1751 NFSVNO_EXINIT(&nes); 1752 nd->nd_repstat = nfsvno_accchk(vp, 1753 VWRITE, cred, &nes, p, 1754 NFSACCCHK_NOOVERRIDE, 1755 NFSACCCHK_VPISLOCKED, NULL); 1756 nd->nd_repstat = nfsrv_opencheck(clientid, 1757 stateidp, stp, vp, nd, p, nd->nd_repstat); 1758 if (!nd->nd_repstat) { 1759 tempsize = nvap->na_size; 1760 NFSVNO_ATTRINIT(nvap); 1761 nvap->na_size = tempsize; 1762 nd->nd_repstat = VOP_SETATTR(vp, 1763 &nvap->na_vattr, cred); 1764 } 1765 } else if (vp->v_type == VREG) { 1766 nd->nd_repstat = nfsrv_opencheck(clientid, 1767 stateidp, stp, vp, nd, p, nd->nd_repstat); 1768 } 1769 } 1770 } else { 1771 if (ndp->ni_cnd.cn_flags & HASBUF) 1772 nfsvno_relpathbuf(ndp); 1773 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) { 1774 vrele(ndp->ni_startdir); 1775 if (ndp->ni_dvp == ndp->ni_vp) 1776 vrele(ndp->ni_dvp); 1777 else 1778 vput(ndp->ni_dvp); 1779 if (ndp->ni_vp) 1780 vput(ndp->ni_vp); 1781 } 1782 } 1783 *vpp = vp; 1784 1785 NFSEXITCODE2(0, nd); 1786 } 1787 1788 /* 1789 * Updates the file rev and sets the mtime and ctime 1790 * to the current clock time, returning the va_filerev and va_Xtime 1791 * values. 1792 * Return ESTALE to indicate the vnode is VIRF_DOOMED. 1793 */ 1794 int 1795 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, 1796 struct nfsrv_descript *nd, struct thread *p) 1797 { 1798 struct vattr va; 1799 1800 VATTR_NULL(&va); 1801 vfs_timestamp(&va.va_mtime); 1802 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 1803 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 1804 if (VN_IS_DOOMED(vp)) 1805 return (ESTALE); 1806 } 1807 (void) VOP_SETATTR(vp, &va, nd->nd_cred); 1808 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); 1809 return (0); 1810 } 1811 1812 /* 1813 * Glue routine to nfsv4_fillattr(). 1814 */ 1815 int 1816 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, 1817 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, 1818 struct ucred *cred, struct thread *p, int isdgram, int reterr, 1819 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) 1820 { 1821 struct statfs *sf; 1822 int error; 1823 1824 sf = NULL; 1825 if (nfsrv_devidcnt > 0 && 1826 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || 1827 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || 1828 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { 1829 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); 1830 error = nfsrv_pnfsstatfs(sf, mp); 1831 if (error != 0) { 1832 free(sf, M_TEMP); 1833 sf = NULL; 1834 } 1835 } 1836 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, 1837 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, 1838 mounted_on_fileno, sf); 1839 free(sf, M_TEMP); 1840 NFSEXITCODE2(0, nd); 1841 return (error); 1842 } 1843 1844 /* Since the Readdir vnode ops vary, put the entire functions in here. */ 1845 /* 1846 * nfs readdir service 1847 * - mallocs what it thinks is enough to read 1848 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR 1849 * - calls VOP_READDIR() 1850 * - loops around building the reply 1851 * if the output generated exceeds count break out of loop 1852 * The NFSM_CLGET macro is used here so that the reply will be packed 1853 * tightly in mbuf clusters. 1854 * - it trims out records with d_fileno == 0 1855 * this doesn't matter for Unix clients, but they might confuse clients 1856 * for other os'. 1857 * - it trims out records with d_type == DT_WHT 1858 * these cannot be seen through NFS (unless we extend the protocol) 1859 * The alternate call nfsrvd_readdirplus() does lookups as well. 1860 * PS: The NFS protocol spec. does not clarify what the "count" byte 1861 * argument is a count of.. just name strings and file id's or the 1862 * entire reply rpc or ... 1863 * I tried just file name and id sizes and it confused the Sun client, 1864 * so I am using the full rpc size now. The "paranoia.." comment refers 1865 * to including the status longwords that are not a part of the dir. 1866 * "entry" structures, but are in the rpc. 1867 */ 1868 int 1869 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, 1870 struct vnode *vp, struct nfsexstuff *exp) 1871 { 1872 struct dirent *dp; 1873 u_int32_t *tl; 1874 int dirlen; 1875 char *cpos, *cend, *rbuf; 1876 struct nfsvattr at; 1877 int nlen, error = 0, getret = 1; 1878 int siz, cnt, fullsiz, eofflag, ncookies; 1879 u_int64_t off, toff, verf __unused; 1880 u_long *cookies = NULL, *cookiep; 1881 struct uio io; 1882 struct iovec iv; 1883 int is_ufs; 1884 struct thread *p = curthread; 1885 1886 if (nd->nd_repstat) { 1887 nfsrv_postopattr(nd, getret, &at); 1888 goto out; 1889 } 1890 if (nd->nd_flag & ND_NFSV2) { 1891 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1892 off = fxdr_unsigned(u_quad_t, *tl++); 1893 } else { 1894 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1895 off = fxdr_hyper(tl); 1896 tl += 2; 1897 verf = fxdr_hyper(tl); 1898 tl += 2; 1899 } 1900 toff = off; 1901 cnt = fxdr_unsigned(int, *tl); 1902 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 1903 cnt = NFS_SRVMAXDATA(nd); 1904 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 1905 fullsiz = siz; 1906 if (nd->nd_flag & ND_NFSV3) { 1907 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, 1908 NULL); 1909 #if 0 1910 /* 1911 * va_filerev is not sufficient as a cookie verifier, 1912 * since it is not supposed to change when entries are 1913 * removed/added unless that offset cookies returned to 1914 * the client are no longer valid. 1915 */ 1916 if (!nd->nd_repstat && toff && verf != at.na_filerev) 1917 nd->nd_repstat = NFSERR_BAD_COOKIE; 1918 #endif 1919 } 1920 if (!nd->nd_repstat && vp->v_type != VDIR) 1921 nd->nd_repstat = NFSERR_NOTDIR; 1922 if (nd->nd_repstat == 0 && cnt == 0) { 1923 if (nd->nd_flag & ND_NFSV2) 1924 /* NFSv2 does not have NFSERR_TOOSMALL */ 1925 nd->nd_repstat = EPERM; 1926 else 1927 nd->nd_repstat = NFSERR_TOOSMALL; 1928 } 1929 if (!nd->nd_repstat) 1930 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 1931 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1932 NFSACCCHK_VPISLOCKED, NULL); 1933 if (nd->nd_repstat) { 1934 vput(vp); 1935 if (nd->nd_flag & ND_NFSV3) 1936 nfsrv_postopattr(nd, getret, &at); 1937 goto out; 1938 } 1939 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 1940 rbuf = malloc(siz, M_TEMP, M_WAITOK); 1941 again: 1942 eofflag = 0; 1943 if (cookies) { 1944 free(cookies, M_TEMP); 1945 cookies = NULL; 1946 } 1947 1948 iv.iov_base = rbuf; 1949 iv.iov_len = siz; 1950 io.uio_iov = &iv; 1951 io.uio_iovcnt = 1; 1952 io.uio_offset = (off_t)off; 1953 io.uio_resid = siz; 1954 io.uio_segflg = UIO_SYSSPACE; 1955 io.uio_rw = UIO_READ; 1956 io.uio_td = NULL; 1957 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 1958 &cookies); 1959 off = (u_int64_t)io.uio_offset; 1960 if (io.uio_resid) 1961 siz -= io.uio_resid; 1962 1963 if (!cookies && !nd->nd_repstat) 1964 nd->nd_repstat = NFSERR_PERM; 1965 if (nd->nd_flag & ND_NFSV3) { 1966 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 1967 if (!nd->nd_repstat) 1968 nd->nd_repstat = getret; 1969 } 1970 1971 /* 1972 * Handles the failed cases. nd->nd_repstat == 0 past here. 1973 */ 1974 if (nd->nd_repstat) { 1975 vput(vp); 1976 free(rbuf, M_TEMP); 1977 if (cookies) 1978 free(cookies, M_TEMP); 1979 if (nd->nd_flag & ND_NFSV3) 1980 nfsrv_postopattr(nd, getret, &at); 1981 goto out; 1982 } 1983 /* 1984 * If nothing read, return eof 1985 * rpc reply 1986 */ 1987 if (siz == 0) { 1988 vput(vp); 1989 if (nd->nd_flag & ND_NFSV2) { 1990 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1991 } else { 1992 nfsrv_postopattr(nd, getret, &at); 1993 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1994 txdr_hyper(at.na_filerev, tl); 1995 tl += 2; 1996 } 1997 *tl++ = newnfs_false; 1998 *tl = newnfs_true; 1999 free(rbuf, M_TEMP); 2000 free(cookies, M_TEMP); 2001 goto out; 2002 } 2003 2004 /* 2005 * Check for degenerate cases of nothing useful read. 2006 * If so go try again 2007 */ 2008 cpos = rbuf; 2009 cend = rbuf + siz; 2010 dp = (struct dirent *)cpos; 2011 cookiep = cookies; 2012 2013 /* 2014 * For some reason FreeBSD's ufs_readdir() chooses to back the 2015 * directory offset up to a block boundary, so it is necessary to 2016 * skip over the records that precede the requested offset. This 2017 * requires the assumption that file offset cookies monotonically 2018 * increase. 2019 */ 2020 while (cpos < cend && ncookies > 0 && 2021 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2022 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { 2023 cpos += dp->d_reclen; 2024 dp = (struct dirent *)cpos; 2025 cookiep++; 2026 ncookies--; 2027 } 2028 if (cpos >= cend || ncookies == 0) { 2029 siz = fullsiz; 2030 toff = off; 2031 goto again; 2032 } 2033 vput(vp); 2034 2035 /* 2036 * dirlen is the size of the reply, including all XDR and must 2037 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate 2038 * if the XDR should be included in "count", but to be safe, we do. 2039 * (Include the two booleans at the end of the reply in dirlen now.) 2040 */ 2041 if (nd->nd_flag & ND_NFSV3) { 2042 nfsrv_postopattr(nd, getret, &at); 2043 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2044 txdr_hyper(at.na_filerev, tl); 2045 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2046 } else { 2047 dirlen = 2 * NFSX_UNSIGNED; 2048 } 2049 2050 /* Loop through the records and build reply */ 2051 while (cpos < cend && ncookies > 0) { 2052 nlen = dp->d_namlen; 2053 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2054 nlen <= NFS_MAXNAMLEN) { 2055 if (nd->nd_flag & ND_NFSV3) 2056 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2057 else 2058 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2059 if (dirlen > cnt) { 2060 eofflag = 0; 2061 break; 2062 } 2063 2064 /* 2065 * Build the directory record xdr from 2066 * the dirent entry. 2067 */ 2068 if (nd->nd_flag & ND_NFSV3) { 2069 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2070 *tl++ = newnfs_true; 2071 *tl++ = 0; 2072 } else { 2073 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2074 *tl++ = newnfs_true; 2075 } 2076 *tl = txdr_unsigned(dp->d_fileno); 2077 (void) nfsm_strtom(nd, dp->d_name, nlen); 2078 if (nd->nd_flag & ND_NFSV3) { 2079 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2080 *tl++ = 0; 2081 } else 2082 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 2083 *tl = txdr_unsigned(*cookiep); 2084 } 2085 cpos += dp->d_reclen; 2086 dp = (struct dirent *)cpos; 2087 cookiep++; 2088 ncookies--; 2089 } 2090 if (cpos < cend) 2091 eofflag = 0; 2092 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2093 *tl++ = newnfs_false; 2094 if (eofflag) 2095 *tl = newnfs_true; 2096 else 2097 *tl = newnfs_false; 2098 free(rbuf, M_TEMP); 2099 free(cookies, M_TEMP); 2100 2101 out: 2102 NFSEXITCODE2(0, nd); 2103 return (0); 2104 nfsmout: 2105 vput(vp); 2106 NFSEXITCODE2(error, nd); 2107 return (error); 2108 } 2109 2110 /* 2111 * Readdirplus for V3 and Readdir for V4. 2112 */ 2113 int 2114 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, 2115 struct vnode *vp, struct nfsexstuff *exp) 2116 { 2117 struct dirent *dp; 2118 u_int32_t *tl; 2119 int dirlen; 2120 char *cpos, *cend, *rbuf; 2121 struct vnode *nvp; 2122 fhandle_t nfh; 2123 struct nfsvattr nva, at, *nvap = &nva; 2124 struct mbuf *mb0, *mb1; 2125 struct nfsreferral *refp; 2126 int nlen, r, error = 0, getret = 1, usevget = 1; 2127 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; 2128 caddr_t bpos0, bpos1; 2129 u_int64_t off, toff, verf; 2130 u_long *cookies = NULL, *cookiep; 2131 nfsattrbit_t attrbits, rderrbits, savbits; 2132 struct uio io; 2133 struct iovec iv; 2134 struct componentname cn; 2135 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; 2136 struct mount *mp, *new_mp; 2137 uint64_t mounted_on_fileno; 2138 struct thread *p = curthread; 2139 2140 if (nd->nd_repstat) { 2141 nfsrv_postopattr(nd, getret, &at); 2142 goto out; 2143 } 2144 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2145 off = fxdr_hyper(tl); 2146 toff = off; 2147 tl += 2; 2148 verf = fxdr_hyper(tl); 2149 tl += 2; 2150 siz = fxdr_unsigned(int, *tl++); 2151 cnt = fxdr_unsigned(int, *tl); 2152 2153 /* 2154 * Use the server's maximum data transfer size as the upper bound 2155 * on reply datalen. 2156 */ 2157 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2158 cnt = NFS_SRVMAXDATA(nd); 2159 2160 /* 2161 * siz is a "hint" of how much directory information (name, fileid, 2162 * cookie) should be in the reply. At least one client "hints" 0, 2163 * so I set it to cnt for that case. I also round it up to the 2164 * next multiple of DIRBLKSIZ. 2165 * Since the size of a Readdirplus directory entry reply will always 2166 * be greater than a directory entry returned by VOP_READDIR(), it 2167 * does not make sense to read more than NFS_SRVMAXDATA() via 2168 * VOP_READDIR(). 2169 */ 2170 if (siz <= 0) 2171 siz = cnt; 2172 else if (siz > NFS_SRVMAXDATA(nd)) 2173 siz = NFS_SRVMAXDATA(nd); 2174 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2175 2176 if (nd->nd_flag & ND_NFSV4) { 2177 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 2178 if (error) 2179 goto nfsmout; 2180 NFSSET_ATTRBIT(&savbits, &attrbits); 2181 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); 2182 NFSZERO_ATTRBIT(&rderrbits); 2183 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); 2184 } else { 2185 NFSZERO_ATTRBIT(&attrbits); 2186 } 2187 fullsiz = siz; 2188 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2189 #if 0 2190 if (!nd->nd_repstat) { 2191 if (off && verf != at.na_filerev) { 2192 /* 2193 * va_filerev is not sufficient as a cookie verifier, 2194 * since it is not supposed to change when entries are 2195 * removed/added unless that offset cookies returned to 2196 * the client are no longer valid. 2197 */ 2198 if (nd->nd_flag & ND_NFSV4) { 2199 nd->nd_repstat = NFSERR_NOTSAME; 2200 } else { 2201 nd->nd_repstat = NFSERR_BAD_COOKIE; 2202 } 2203 } 2204 } 2205 #endif 2206 if (!nd->nd_repstat && vp->v_type != VDIR) 2207 nd->nd_repstat = NFSERR_NOTDIR; 2208 if (!nd->nd_repstat && cnt == 0) 2209 nd->nd_repstat = NFSERR_TOOSMALL; 2210 if (!nd->nd_repstat) 2211 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2212 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2213 NFSACCCHK_VPISLOCKED, NULL); 2214 if (nd->nd_repstat) { 2215 vput(vp); 2216 if (nd->nd_flag & ND_NFSV3) 2217 nfsrv_postopattr(nd, getret, &at); 2218 goto out; 2219 } 2220 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2221 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; 2222 2223 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2224 again: 2225 eofflag = 0; 2226 if (cookies) { 2227 free(cookies, M_TEMP); 2228 cookies = NULL; 2229 } 2230 2231 iv.iov_base = rbuf; 2232 iv.iov_len = siz; 2233 io.uio_iov = &iv; 2234 io.uio_iovcnt = 1; 2235 io.uio_offset = (off_t)off; 2236 io.uio_resid = siz; 2237 io.uio_segflg = UIO_SYSSPACE; 2238 io.uio_rw = UIO_READ; 2239 io.uio_td = NULL; 2240 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2241 &cookies); 2242 off = (u_int64_t)io.uio_offset; 2243 if (io.uio_resid) 2244 siz -= io.uio_resid; 2245 2246 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2247 2248 if (!cookies && !nd->nd_repstat) 2249 nd->nd_repstat = NFSERR_PERM; 2250 if (!nd->nd_repstat) 2251 nd->nd_repstat = getret; 2252 if (nd->nd_repstat) { 2253 vput(vp); 2254 if (cookies) 2255 free(cookies, M_TEMP); 2256 free(rbuf, M_TEMP); 2257 if (nd->nd_flag & ND_NFSV3) 2258 nfsrv_postopattr(nd, getret, &at); 2259 goto out; 2260 } 2261 /* 2262 * If nothing read, return eof 2263 * rpc reply 2264 */ 2265 if (siz == 0) { 2266 vput(vp); 2267 if (nd->nd_flag & ND_NFSV3) 2268 nfsrv_postopattr(nd, getret, &at); 2269 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2270 txdr_hyper(at.na_filerev, tl); 2271 tl += 2; 2272 *tl++ = newnfs_false; 2273 *tl = newnfs_true; 2274 free(cookies, M_TEMP); 2275 free(rbuf, M_TEMP); 2276 goto out; 2277 } 2278 2279 /* 2280 * Check for degenerate cases of nothing useful read. 2281 * If so go try again 2282 */ 2283 cpos = rbuf; 2284 cend = rbuf + siz; 2285 dp = (struct dirent *)cpos; 2286 cookiep = cookies; 2287 2288 /* 2289 * For some reason FreeBSD's ufs_readdir() chooses to back the 2290 * directory offset up to a block boundary, so it is necessary to 2291 * skip over the records that precede the requested offset. This 2292 * requires the assumption that file offset cookies monotonically 2293 * increase. 2294 */ 2295 while (cpos < cend && ncookies > 0 && 2296 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2297 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || 2298 ((nd->nd_flag & ND_NFSV4) && 2299 ((dp->d_namlen == 1 && dp->d_name[0] == '.') || 2300 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { 2301 cpos += dp->d_reclen; 2302 dp = (struct dirent *)cpos; 2303 cookiep++; 2304 ncookies--; 2305 } 2306 if (cpos >= cend || ncookies == 0) { 2307 siz = fullsiz; 2308 toff = off; 2309 goto again; 2310 } 2311 2312 /* 2313 * Busy the file system so that the mount point won't go away 2314 * and, as such, VFS_VGET() can be used safely. 2315 */ 2316 mp = vp->v_mount; 2317 vfs_ref(mp); 2318 NFSVOPUNLOCK(vp); 2319 nd->nd_repstat = vfs_busy(mp, 0); 2320 vfs_rel(mp); 2321 if (nd->nd_repstat != 0) { 2322 vrele(vp); 2323 free(cookies, M_TEMP); 2324 free(rbuf, M_TEMP); 2325 if (nd->nd_flag & ND_NFSV3) 2326 nfsrv_postopattr(nd, getret, &at); 2327 goto out; 2328 } 2329 2330 /* 2331 * Check to see if entries in this directory can be safely acquired 2332 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. 2333 * ZFS snapshot directories need VOP_LOOKUP(), so that any 2334 * automount of the snapshot directory that is required will 2335 * be done. 2336 * This needs to be done here for NFSv4, since NFSv4 never does 2337 * a VFS_VGET() for "." or "..". 2338 */ 2339 if (is_zfs == 1) { 2340 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); 2341 if (r == EOPNOTSUPP) { 2342 usevget = 0; 2343 cn.cn_nameiop = LOOKUP; 2344 cn.cn_lkflags = LK_SHARED | LK_RETRY; 2345 cn.cn_cred = nd->nd_cred; 2346 cn.cn_thread = p; 2347 } else if (r == 0) 2348 vput(nvp); 2349 } 2350 2351 /* 2352 * Save this position, in case there is an error before one entry 2353 * is created. 2354 */ 2355 mb0 = nd->nd_mb; 2356 bpos0 = nd->nd_bpos; 2357 2358 /* 2359 * Fill in the first part of the reply. 2360 * dirlen is the reply length in bytes and cannot exceed cnt. 2361 * (Include the two booleans at the end of the reply in dirlen now, 2362 * so we recognize when we have exceeded cnt.) 2363 */ 2364 if (nd->nd_flag & ND_NFSV3) { 2365 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2366 nfsrv_postopattr(nd, getret, &at); 2367 } else { 2368 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; 2369 } 2370 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); 2371 txdr_hyper(at.na_filerev, tl); 2372 2373 /* 2374 * Save this position, in case there is an empty reply needed. 2375 */ 2376 mb1 = nd->nd_mb; 2377 bpos1 = nd->nd_bpos; 2378 2379 /* Loop through the records and build reply */ 2380 entrycnt = 0; 2381 while (cpos < cend && ncookies > 0 && dirlen < cnt) { 2382 nlen = dp->d_namlen; 2383 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2384 nlen <= NFS_MAXNAMLEN && 2385 ((nd->nd_flag & ND_NFSV3) || nlen > 2 || 2386 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) 2387 || (nlen == 1 && dp->d_name[0] != '.'))) { 2388 /* 2389 * Save the current position in the reply, in case 2390 * this entry exceeds cnt. 2391 */ 2392 mb1 = nd->nd_mb; 2393 bpos1 = nd->nd_bpos; 2394 2395 /* 2396 * For readdir_and_lookup get the vnode using 2397 * the file number. 2398 */ 2399 nvp = NULL; 2400 refp = NULL; 2401 r = 0; 2402 at_root = 0; 2403 needs_unbusy = 0; 2404 new_mp = mp; 2405 mounted_on_fileno = (uint64_t)dp->d_fileno; 2406 if ((nd->nd_flag & ND_NFSV3) || 2407 NFSNONZERO_ATTRBIT(&savbits)) { 2408 if (nd->nd_flag & ND_NFSV4) 2409 refp = nfsv4root_getreferral(NULL, 2410 vp, dp->d_fileno); 2411 if (refp == NULL) { 2412 if (usevget) 2413 r = VFS_VGET(mp, dp->d_fileno, 2414 LK_SHARED, &nvp); 2415 else 2416 r = EOPNOTSUPP; 2417 if (r == EOPNOTSUPP) { 2418 if (usevget) { 2419 usevget = 0; 2420 cn.cn_nameiop = LOOKUP; 2421 cn.cn_lkflags = 2422 LK_SHARED | 2423 LK_RETRY; 2424 cn.cn_cred = 2425 nd->nd_cred; 2426 cn.cn_thread = p; 2427 } 2428 cn.cn_nameptr = dp->d_name; 2429 cn.cn_namelen = nlen; 2430 cn.cn_flags = ISLASTCN | 2431 NOFOLLOW | LOCKLEAF; 2432 if (nlen == 2 && 2433 dp->d_name[0] == '.' && 2434 dp->d_name[1] == '.') 2435 cn.cn_flags |= 2436 ISDOTDOT; 2437 if (NFSVOPLOCK(vp, LK_SHARED) 2438 != 0) { 2439 nd->nd_repstat = EPERM; 2440 break; 2441 } 2442 if ((vp->v_vflag & VV_ROOT) != 0 2443 && (cn.cn_flags & ISDOTDOT) 2444 != 0) { 2445 vref(vp); 2446 nvp = vp; 2447 r = 0; 2448 } else { 2449 r = VOP_LOOKUP(vp, &nvp, 2450 &cn); 2451 if (vp != nvp) 2452 NFSVOPUNLOCK(vp); 2453 } 2454 } 2455 2456 /* 2457 * For NFSv4, check to see if nvp is 2458 * a mount point and get the mount 2459 * point vnode, as required. 2460 */ 2461 if (r == 0 && 2462 nfsrv_enable_crossmntpt != 0 && 2463 (nd->nd_flag & ND_NFSV4) != 0 && 2464 nvp->v_type == VDIR && 2465 nvp->v_mountedhere != NULL) { 2466 new_mp = nvp->v_mountedhere; 2467 r = vfs_busy(new_mp, 0); 2468 vput(nvp); 2469 nvp = NULL; 2470 if (r == 0) { 2471 r = VFS_ROOT(new_mp, 2472 LK_SHARED, &nvp); 2473 needs_unbusy = 1; 2474 if (r == 0) 2475 at_root = 1; 2476 } 2477 } 2478 } 2479 2480 /* 2481 * If we failed to look up the entry, then it 2482 * has become invalid, most likely removed. 2483 */ 2484 if (r != 0) { 2485 if (needs_unbusy) 2486 vfs_unbusy(new_mp); 2487 goto invalid; 2488 } 2489 KASSERT(refp != NULL || nvp != NULL, 2490 ("%s: undetected lookup error", __func__)); 2491 2492 if (refp == NULL && 2493 ((nd->nd_flag & ND_NFSV3) || 2494 NFSNONZERO_ATTRBIT(&attrbits))) { 2495 r = nfsvno_getfh(nvp, &nfh, p); 2496 if (!r) 2497 r = nfsvno_getattr(nvp, nvap, nd, p, 2498 1, &attrbits); 2499 if (r == 0 && is_zfs == 1 && 2500 nfsrv_enable_crossmntpt != 0 && 2501 (nd->nd_flag & ND_NFSV4) != 0 && 2502 nvp->v_type == VDIR && 2503 vp->v_mount != nvp->v_mount) { 2504 /* 2505 * For a ZFS snapshot, there is a 2506 * pseudo mount that does not set 2507 * v_mountedhere, so it needs to 2508 * be detected via a different 2509 * mount structure. 2510 */ 2511 at_root = 1; 2512 if (new_mp == mp) 2513 new_mp = nvp->v_mount; 2514 } 2515 } 2516 2517 /* 2518 * If we failed to get attributes of the entry, 2519 * then just skip it for NFSv3 (the traditional 2520 * behavior in the old NFS server). 2521 * For NFSv4 the behavior is controlled by 2522 * RDATTRERROR: we either ignore the error or 2523 * fail the request. 2524 * Note that RDATTRERROR is never set for NFSv3. 2525 */ 2526 if (r != 0) { 2527 if (!NFSISSET_ATTRBIT(&attrbits, 2528 NFSATTRBIT_RDATTRERROR)) { 2529 vput(nvp); 2530 if (needs_unbusy != 0) 2531 vfs_unbusy(new_mp); 2532 if ((nd->nd_flag & ND_NFSV3)) 2533 goto invalid; 2534 nd->nd_repstat = r; 2535 break; 2536 } 2537 } 2538 } 2539 2540 /* 2541 * Build the directory record xdr 2542 */ 2543 if (nd->nd_flag & ND_NFSV3) { 2544 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2545 *tl++ = newnfs_true; 2546 *tl++ = 0; 2547 *tl = txdr_unsigned(dp->d_fileno); 2548 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2549 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2550 *tl++ = 0; 2551 *tl = txdr_unsigned(*cookiep); 2552 nfsrv_postopattr(nd, 0, nvap); 2553 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1); 2554 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); 2555 if (nvp != NULL) 2556 vput(nvp); 2557 } else { 2558 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2559 *tl++ = newnfs_true; 2560 *tl++ = 0; 2561 *tl = txdr_unsigned(*cookiep); 2562 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2563 if (nvp != NULL) { 2564 supports_nfsv4acls = 2565 nfs_supportsnfsv4acls(nvp); 2566 NFSVOPUNLOCK(nvp); 2567 } else 2568 supports_nfsv4acls = 0; 2569 if (refp != NULL) { 2570 dirlen += nfsrv_putreferralattr(nd, 2571 &savbits, refp, 0, 2572 &nd->nd_repstat); 2573 if (nd->nd_repstat) { 2574 if (nvp != NULL) 2575 vrele(nvp); 2576 if (needs_unbusy != 0) 2577 vfs_unbusy(new_mp); 2578 break; 2579 } 2580 } else if (r) { 2581 dirlen += nfsvno_fillattr(nd, new_mp, 2582 nvp, nvap, &nfh, r, &rderrbits, 2583 nd->nd_cred, p, isdgram, 0, 2584 supports_nfsv4acls, at_root, 2585 mounted_on_fileno); 2586 } else { 2587 dirlen += nfsvno_fillattr(nd, new_mp, 2588 nvp, nvap, &nfh, r, &attrbits, 2589 nd->nd_cred, p, isdgram, 0, 2590 supports_nfsv4acls, at_root, 2591 mounted_on_fileno); 2592 } 2593 if (nvp != NULL) 2594 vrele(nvp); 2595 dirlen += (3 * NFSX_UNSIGNED); 2596 } 2597 if (needs_unbusy != 0) 2598 vfs_unbusy(new_mp); 2599 if (dirlen <= cnt) 2600 entrycnt++; 2601 } 2602 invalid: 2603 cpos += dp->d_reclen; 2604 dp = (struct dirent *)cpos; 2605 cookiep++; 2606 ncookies--; 2607 } 2608 vrele(vp); 2609 vfs_unbusy(mp); 2610 2611 /* 2612 * If dirlen > cnt, we must strip off the last entry. If that 2613 * results in an empty reply, report NFSERR_TOOSMALL. 2614 */ 2615 if (dirlen > cnt || nd->nd_repstat) { 2616 if (!nd->nd_repstat && entrycnt == 0) 2617 nd->nd_repstat = NFSERR_TOOSMALL; 2618 if (nd->nd_repstat) { 2619 newnfs_trimtrailing(nd, mb0, bpos0); 2620 if (nd->nd_flag & ND_NFSV3) 2621 nfsrv_postopattr(nd, getret, &at); 2622 } else 2623 newnfs_trimtrailing(nd, mb1, bpos1); 2624 eofflag = 0; 2625 } else if (cpos < cend) 2626 eofflag = 0; 2627 if (!nd->nd_repstat) { 2628 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2629 *tl++ = newnfs_false; 2630 if (eofflag) 2631 *tl = newnfs_true; 2632 else 2633 *tl = newnfs_false; 2634 } 2635 free(cookies, M_TEMP); 2636 free(rbuf, M_TEMP); 2637 2638 out: 2639 NFSEXITCODE2(0, nd); 2640 return (0); 2641 nfsmout: 2642 vput(vp); 2643 NFSEXITCODE2(error, nd); 2644 return (error); 2645 } 2646 2647 /* 2648 * Get the settable attributes out of the mbuf list. 2649 * (Return 0 or EBADRPC) 2650 */ 2651 int 2652 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2653 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2654 { 2655 u_int32_t *tl; 2656 struct nfsv2_sattr *sp; 2657 int error = 0, toclient = 0; 2658 2659 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 2660 case ND_NFSV2: 2661 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2662 /* 2663 * Some old clients didn't fill in the high order 16bits. 2664 * --> check the low order 2 bytes for 0xffff 2665 */ 2666 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 2667 nvap->na_mode = nfstov_mode(sp->sa_mode); 2668 if (sp->sa_uid != newnfs_xdrneg1) 2669 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 2670 if (sp->sa_gid != newnfs_xdrneg1) 2671 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 2672 if (sp->sa_size != newnfs_xdrneg1) 2673 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 2674 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 2675 #ifdef notyet 2676 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 2677 #else 2678 nvap->na_atime.tv_sec = 2679 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 2680 nvap->na_atime.tv_nsec = 0; 2681 #endif 2682 } 2683 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 2684 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 2685 break; 2686 case ND_NFSV3: 2687 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2688 if (*tl == newnfs_true) { 2689 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2690 nvap->na_mode = nfstov_mode(*tl); 2691 } 2692 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2693 if (*tl == newnfs_true) { 2694 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2695 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 2696 } 2697 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2698 if (*tl == newnfs_true) { 2699 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2700 nvap->na_gid = fxdr_unsigned(gid_t, *tl); 2701 } 2702 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2703 if (*tl == newnfs_true) { 2704 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2705 nvap->na_size = fxdr_hyper(tl); 2706 } 2707 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2708 switch (fxdr_unsigned(int, *tl)) { 2709 case NFSV3SATTRTIME_TOCLIENT: 2710 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2711 fxdr_nfsv3time(tl, &nvap->na_atime); 2712 toclient = 1; 2713 break; 2714 case NFSV3SATTRTIME_TOSERVER: 2715 vfs_timestamp(&nvap->na_atime); 2716 nvap->na_vaflags |= VA_UTIMES_NULL; 2717 break; 2718 } 2719 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2720 switch (fxdr_unsigned(int, *tl)) { 2721 case NFSV3SATTRTIME_TOCLIENT: 2722 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2723 fxdr_nfsv3time(tl, &nvap->na_mtime); 2724 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2725 break; 2726 case NFSV3SATTRTIME_TOSERVER: 2727 vfs_timestamp(&nvap->na_mtime); 2728 if (!toclient) 2729 nvap->na_vaflags |= VA_UTIMES_NULL; 2730 break; 2731 } 2732 break; 2733 case ND_NFSV4: 2734 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); 2735 } 2736 nfsmout: 2737 NFSEXITCODE2(error, nd); 2738 return (error); 2739 } 2740 2741 /* 2742 * Handle the setable attributes for V4. 2743 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 2744 */ 2745 int 2746 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2747 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2748 { 2749 u_int32_t *tl; 2750 int attrsum = 0; 2751 int i, j; 2752 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 2753 int moderet, toclient = 0; 2754 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 2755 uid_t uid; 2756 gid_t gid; 2757 u_short mode, mask; /* Same type as va_mode. */ 2758 struct vattr va; 2759 2760 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 2761 if (error) 2762 goto nfsmout; 2763 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2764 attrsize = fxdr_unsigned(int, *tl); 2765 2766 /* 2767 * Loop around getting the setable attributes. If an unsupported 2768 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 2769 */ 2770 if (retnotsup) { 2771 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2772 bitpos = NFSATTRBIT_MAX; 2773 } else { 2774 bitpos = 0; 2775 } 2776 moderet = 0; 2777 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 2778 if (attrsum > attrsize) { 2779 error = NFSERR_BADXDR; 2780 goto nfsmout; 2781 } 2782 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 2783 switch (bitpos) { 2784 case NFSATTRBIT_SIZE: 2785 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 2786 if (vp != NULL && vp->v_type != VREG) { 2787 error = (vp->v_type == VDIR) ? NFSERR_ISDIR : 2788 NFSERR_INVAL; 2789 goto nfsmout; 2790 } 2791 nvap->na_size = fxdr_hyper(tl); 2792 attrsum += NFSX_HYPER; 2793 break; 2794 case NFSATTRBIT_ACL: 2795 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize, 2796 p); 2797 if (error) 2798 goto nfsmout; 2799 if (aceerr && !nd->nd_repstat) 2800 nd->nd_repstat = aceerr; 2801 attrsum += aclsize; 2802 break; 2803 case NFSATTRBIT_ARCHIVE: 2804 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2805 if (!nd->nd_repstat) 2806 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2807 attrsum += NFSX_UNSIGNED; 2808 break; 2809 case NFSATTRBIT_HIDDEN: 2810 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2811 if (!nd->nd_repstat) 2812 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2813 attrsum += NFSX_UNSIGNED; 2814 break; 2815 case NFSATTRBIT_MIMETYPE: 2816 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2817 i = fxdr_unsigned(int, *tl); 2818 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 2819 if (error) 2820 goto nfsmout; 2821 if (!nd->nd_repstat) 2822 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2823 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 2824 break; 2825 case NFSATTRBIT_MODE: 2826 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ 2827 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2828 nvap->na_mode = nfstov_mode(*tl); 2829 attrsum += NFSX_UNSIGNED; 2830 break; 2831 case NFSATTRBIT_OWNER: 2832 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2833 j = fxdr_unsigned(int, *tl); 2834 if (j < 0) { 2835 error = NFSERR_BADXDR; 2836 goto nfsmout; 2837 } 2838 if (j > NFSV4_SMALLSTR) 2839 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 2840 else 2841 cp = namestr; 2842 error = nfsrv_mtostr(nd, cp, j); 2843 if (error) { 2844 if (j > NFSV4_SMALLSTR) 2845 free(cp, M_NFSSTRING); 2846 goto nfsmout; 2847 } 2848 if (!nd->nd_repstat) { 2849 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, 2850 &uid); 2851 if (!nd->nd_repstat) 2852 nvap->na_uid = uid; 2853 } 2854 if (j > NFSV4_SMALLSTR) 2855 free(cp, M_NFSSTRING); 2856 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 2857 break; 2858 case NFSATTRBIT_OWNERGROUP: 2859 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2860 j = fxdr_unsigned(int, *tl); 2861 if (j < 0) { 2862 error = NFSERR_BADXDR; 2863 goto nfsmout; 2864 } 2865 if (j > NFSV4_SMALLSTR) 2866 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 2867 else 2868 cp = namestr; 2869 error = nfsrv_mtostr(nd, cp, j); 2870 if (error) { 2871 if (j > NFSV4_SMALLSTR) 2872 free(cp, M_NFSSTRING); 2873 goto nfsmout; 2874 } 2875 if (!nd->nd_repstat) { 2876 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, 2877 &gid); 2878 if (!nd->nd_repstat) 2879 nvap->na_gid = gid; 2880 } 2881 if (j > NFSV4_SMALLSTR) 2882 free(cp, M_NFSSTRING); 2883 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 2884 break; 2885 case NFSATTRBIT_SYSTEM: 2886 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2887 if (!nd->nd_repstat) 2888 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2889 attrsum += NFSX_UNSIGNED; 2890 break; 2891 case NFSATTRBIT_TIMEACCESSSET: 2892 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2893 attrsum += NFSX_UNSIGNED; 2894 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 2895 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2896 fxdr_nfsv4time(tl, &nvap->na_atime); 2897 toclient = 1; 2898 attrsum += NFSX_V4TIME; 2899 } else { 2900 vfs_timestamp(&nvap->na_atime); 2901 nvap->na_vaflags |= VA_UTIMES_NULL; 2902 } 2903 break; 2904 case NFSATTRBIT_TIMEBACKUP: 2905 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2906 if (!nd->nd_repstat) 2907 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2908 attrsum += NFSX_V4TIME; 2909 break; 2910 case NFSATTRBIT_TIMECREATE: 2911 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2912 if (!nd->nd_repstat) 2913 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2914 attrsum += NFSX_V4TIME; 2915 break; 2916 case NFSATTRBIT_TIMEMODIFYSET: 2917 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2918 attrsum += NFSX_UNSIGNED; 2919 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 2920 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2921 fxdr_nfsv4time(tl, &nvap->na_mtime); 2922 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2923 attrsum += NFSX_V4TIME; 2924 } else { 2925 vfs_timestamp(&nvap->na_mtime); 2926 if (!toclient) 2927 nvap->na_vaflags |= VA_UTIMES_NULL; 2928 } 2929 break; 2930 case NFSATTRBIT_MODESETMASKED: 2931 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 2932 mode = fxdr_unsigned(u_short, *tl++); 2933 mask = fxdr_unsigned(u_short, *tl); 2934 /* 2935 * vp == NULL implies an Open/Create operation. 2936 * This attribute can only be used for Setattr and 2937 * only for NFSv4.1 or higher. 2938 * If moderet != 0, a mode attribute has also been 2939 * specified and this attribute cannot be done in the 2940 * same Setattr operation. 2941 */ 2942 if ((nd->nd_flag & ND_NFSV41) == 0) 2943 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2944 else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 || 2945 vp == NULL) 2946 nd->nd_repstat = NFSERR_INVAL; 2947 else if (moderet == 0) 2948 moderet = VOP_GETATTR(vp, &va, nd->nd_cred); 2949 if (moderet == 0) 2950 nvap->na_mode = (mode & mask) | 2951 (va.va_mode & ~mask); 2952 else 2953 nd->nd_repstat = moderet; 2954 attrsum += 2 * NFSX_UNSIGNED; 2955 break; 2956 default: 2957 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2958 /* 2959 * set bitpos so we drop out of the loop. 2960 */ 2961 bitpos = NFSATTRBIT_MAX; 2962 break; 2963 } 2964 } 2965 2966 /* 2967 * some clients pad the attrlist, so we need to skip over the 2968 * padding. 2969 */ 2970 if (attrsum > attrsize) { 2971 error = NFSERR_BADXDR; 2972 } else { 2973 attrsize = NFSM_RNDUP(attrsize); 2974 if (attrsum < attrsize) 2975 error = nfsm_advance(nd, attrsize - attrsum, -1); 2976 } 2977 nfsmout: 2978 NFSEXITCODE2(error, nd); 2979 return (error); 2980 } 2981 2982 /* 2983 * Check/setup export credentials. 2984 */ 2985 int 2986 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 2987 struct ucred *credanon) 2988 { 2989 int error = 0; 2990 2991 /* 2992 * Check/setup credentials. 2993 */ 2994 if (nd->nd_flag & ND_GSS) 2995 exp->nes_exflag &= ~MNT_EXPORTANON; 2996 2997 /* 2998 * Check to see if the operation is allowed for this security flavor. 2999 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to 3000 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. 3001 * Also, allow Secinfo, so that it can acquire the correct flavor(s). 3002 */ 3003 if (nfsvno_testexp(nd, exp) && 3004 nd->nd_procnum != NFSV4OP_SECINFO && 3005 nd->nd_procnum != NFSPROC_FSINFO) { 3006 if (nd->nd_flag & ND_NFSV4) 3007 error = NFSERR_WRONGSEC; 3008 else 3009 error = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3010 goto out; 3011 } 3012 3013 /* 3014 * Check to see if the file system is exported V4 only. 3015 */ 3016 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 3017 error = NFSERR_PROGNOTV4; 3018 goto out; 3019 } 3020 3021 /* 3022 * Now, map the user credentials. 3023 * (Note that ND_AUTHNONE will only be set for an NFSv3 3024 * Fsinfo RPC. If set for anything else, this code might need 3025 * to change.) 3026 */ 3027 if (NFSVNO_EXPORTED(exp)) { 3028 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || 3029 NFSVNO_EXPORTANON(exp) || 3030 (nd->nd_flag & ND_AUTHNONE) != 0) { 3031 nd->nd_cred->cr_uid = credanon->cr_uid; 3032 nd->nd_cred->cr_gid = credanon->cr_gid; 3033 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 3034 credanon->cr_groups); 3035 } else if ((nd->nd_flag & ND_GSS) == 0) { 3036 /* 3037 * If using AUTH_SYS, call nfsrv_getgrpscred() to see 3038 * if there is a replacement credential with a group 3039 * list set up by "nfsuserd -manage-gids". 3040 * If there is no replacement, nfsrv_getgrpscred() 3041 * simply returns its argument. 3042 */ 3043 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); 3044 } 3045 } 3046 3047 out: 3048 NFSEXITCODE2(error, nd); 3049 return (error); 3050 } 3051 3052 /* 3053 * Check exports. 3054 */ 3055 int 3056 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 3057 struct ucred **credp) 3058 { 3059 int i, error, *secflavors; 3060 3061 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3062 &exp->nes_numsecflavor, &secflavors); 3063 if (error) { 3064 if (nfs_rootfhset) { 3065 exp->nes_exflag = 0; 3066 exp->nes_numsecflavor = 0; 3067 error = 0; 3068 } 3069 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3070 MAXSECFLAVORS) { 3071 printf("nfsvno_checkexp: numsecflavors out of range\n"); 3072 exp->nes_numsecflavor = 0; 3073 error = EACCES; 3074 } else { 3075 /* Copy the security flavors. */ 3076 for (i = 0; i < exp->nes_numsecflavor; i++) 3077 exp->nes_secflavors[i] = secflavors[i]; 3078 } 3079 NFSEXITCODE(error); 3080 return (error); 3081 } 3082 3083 /* 3084 * Get a vnode for a file handle and export stuff. 3085 */ 3086 int 3087 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 3088 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 3089 struct ucred **credp) 3090 { 3091 int i, error, *secflavors; 3092 3093 *credp = NULL; 3094 exp->nes_numsecflavor = 0; 3095 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); 3096 if (error != 0) 3097 /* Make sure the server replies ESTALE to the client. */ 3098 error = ESTALE; 3099 if (nam && !error) { 3100 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3101 &exp->nes_numsecflavor, &secflavors); 3102 if (error) { 3103 if (nfs_rootfhset) { 3104 exp->nes_exflag = 0; 3105 exp->nes_numsecflavor = 0; 3106 error = 0; 3107 } else { 3108 vput(*vpp); 3109 } 3110 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3111 MAXSECFLAVORS) { 3112 printf("nfsvno_fhtovp: numsecflavors out of range\n"); 3113 exp->nes_numsecflavor = 0; 3114 error = EACCES; 3115 vput(*vpp); 3116 } else { 3117 /* Copy the security flavors. */ 3118 for (i = 0; i < exp->nes_numsecflavor; i++) 3119 exp->nes_secflavors[i] = secflavors[i]; 3120 } 3121 } 3122 NFSEXITCODE(error); 3123 return (error); 3124 } 3125 3126 /* 3127 * nfsd_fhtovp() - convert a fh to a vnode ptr 3128 * - look up fsid in mount list (if not found ret error) 3129 * - get vp and export rights by calling nfsvno_fhtovp() 3130 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 3131 * for AUTH_SYS 3132 * - if mpp != NULL, return the mount point so that it can 3133 * be used for vn_finished_write() by the caller 3134 */ 3135 void 3136 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 3137 struct vnode **vpp, struct nfsexstuff *exp, 3138 struct mount **mpp, int startwrite) 3139 { 3140 struct mount *mp; 3141 struct ucred *credanon; 3142 fhandle_t *fhp; 3143 3144 fhp = (fhandle_t *)nfp->nfsrvfh_data; 3145 /* 3146 * Check for the special case of the nfsv4root_fh. 3147 */ 3148 mp = vfs_busyfs(&fhp->fh_fsid); 3149 if (mpp != NULL) 3150 *mpp = mp; 3151 if (mp == NULL) { 3152 *vpp = NULL; 3153 nd->nd_repstat = ESTALE; 3154 goto out; 3155 } 3156 3157 if (startwrite) { 3158 vn_start_write(NULL, mpp, V_WAIT); 3159 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) 3160 lktype = LK_EXCLUSIVE; 3161 } 3162 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 3163 &credanon); 3164 vfs_unbusy(mp); 3165 3166 /* 3167 * For NFSv4 without a pseudo root fs, unexported file handles 3168 * can be returned, so that Lookup works everywhere. 3169 */ 3170 if (!nd->nd_repstat && exp->nes_exflag == 0 && 3171 !(nd->nd_flag & ND_NFSV4)) { 3172 vput(*vpp); 3173 nd->nd_repstat = EACCES; 3174 } 3175 3176 /* 3177 * Personally, I've never seen any point in requiring a 3178 * reserved port#, since only in the rare case where the 3179 * clients are all boxes with secure system privileges, 3180 * does it provide any enhanced security, but... some people 3181 * believe it to be useful and keep putting this code back in. 3182 * (There is also some "security checker" out there that 3183 * complains if the nfs server doesn't enforce this.) 3184 * However, note the following: 3185 * RFC3530 (NFSv4) specifies that a reserved port# not be 3186 * required. 3187 * RFC2623 recommends that, if a reserved port# is checked for, 3188 * that there be a way to turn that off--> ifdef'd. 3189 */ 3190 #ifdef NFS_REQRSVPORT 3191 if (!nd->nd_repstat) { 3192 struct sockaddr_in *saddr; 3193 struct sockaddr_in6 *saddr6; 3194 3195 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 3196 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 3197 if (!(nd->nd_flag & ND_NFSV4) && 3198 ((saddr->sin_family == AF_INET && 3199 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 3200 (saddr6->sin6_family == AF_INET6 && 3201 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 3202 vput(*vpp); 3203 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3204 } 3205 } 3206 #endif /* NFS_REQRSVPORT */ 3207 3208 /* 3209 * Check/setup credentials. 3210 */ 3211 if (!nd->nd_repstat) { 3212 nd->nd_saveduid = nd->nd_cred->cr_uid; 3213 nd->nd_repstat = nfsd_excred(nd, exp, credanon); 3214 if (nd->nd_repstat) 3215 vput(*vpp); 3216 } 3217 if (credanon != NULL) 3218 crfree(credanon); 3219 if (nd->nd_repstat) { 3220 if (startwrite) 3221 vn_finished_write(mp); 3222 *vpp = NULL; 3223 if (mpp != NULL) 3224 *mpp = NULL; 3225 } 3226 3227 out: 3228 NFSEXITCODE2(0, nd); 3229 } 3230 3231 /* 3232 * glue for fp. 3233 */ 3234 static int 3235 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 3236 { 3237 struct filedesc *fdp; 3238 struct file *fp; 3239 int error = 0; 3240 3241 fdp = p->td_proc->p_fd; 3242 if (fd < 0 || fd >= fdp->fd_nfiles || 3243 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { 3244 error = EBADF; 3245 goto out; 3246 } 3247 *fpp = fp; 3248 3249 out: 3250 NFSEXITCODE(error); 3251 return (error); 3252 } 3253 3254 /* 3255 * Called from nfssvc() to update the exports list. Just call 3256 * vfs_export(). This has to be done, since the v4 root fake fs isn't 3257 * in the mount list. 3258 */ 3259 int 3260 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 3261 { 3262 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 3263 int error = 0; 3264 struct nameidata nd; 3265 fhandle_t fh; 3266 3267 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export); 3268 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 3269 nfs_rootfhset = 0; 3270 else if (error == 0) { 3271 if (nfsexargp->fspec == NULL) { 3272 error = EPERM; 3273 goto out; 3274 } 3275 /* 3276 * If fspec != NULL, this is the v4root path. 3277 */ 3278 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, 3279 nfsexargp->fspec, p); 3280 if ((error = namei(&nd)) != 0) 3281 goto out; 3282 error = nfsvno_getfh(nd.ni_vp, &fh, p); 3283 vrele(nd.ni_vp); 3284 if (!error) { 3285 nfs_rootfh.nfsrvfh_len = NFSX_MYFH; 3286 NFSBCOPY((caddr_t)&fh, 3287 nfs_rootfh.nfsrvfh_data, 3288 sizeof (fhandle_t)); 3289 nfs_rootfhset = 1; 3290 } 3291 } 3292 3293 out: 3294 NFSEXITCODE(error); 3295 return (error); 3296 } 3297 3298 /* 3299 * This function needs to test to see if the system is near its limit 3300 * for memory allocation via malloc() or mget() and return True iff 3301 * either of these resources are near their limit. 3302 * XXX (For now, this is just a stub.) 3303 */ 3304 int nfsrv_testmalloclimit = 0; 3305 int 3306 nfsrv_mallocmget_limit(void) 3307 { 3308 static int printmesg = 0; 3309 static int testval = 1; 3310 3311 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 3312 if ((printmesg++ % 100) == 0) 3313 printf("nfsd: malloc/mget near limit\n"); 3314 return (1); 3315 } 3316 return (0); 3317 } 3318 3319 /* 3320 * BSD specific initialization of a mount point. 3321 */ 3322 void 3323 nfsd_mntinit(void) 3324 { 3325 static int inited = 0; 3326 3327 if (inited) 3328 return; 3329 inited = 1; 3330 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 3331 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist); 3332 TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist); 3333 nfsv4root_mnt.mnt_export = NULL; 3334 TAILQ_INIT(&nfsv4root_opt); 3335 TAILQ_INIT(&nfsv4root_newopt); 3336 nfsv4root_mnt.mnt_opt = &nfsv4root_opt; 3337 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt; 3338 nfsv4root_mnt.mnt_nvnodelistsize = 0; 3339 nfsv4root_mnt.mnt_lazyvnodelistsize = 0; 3340 } 3341 3342 /* 3343 * Get a vnode for a file handle, without checking exports, etc. 3344 */ 3345 struct vnode * 3346 nfsvno_getvp(fhandle_t *fhp) 3347 { 3348 struct mount *mp; 3349 struct vnode *vp; 3350 int error; 3351 3352 mp = vfs_busyfs(&fhp->fh_fsid); 3353 if (mp == NULL) 3354 return (NULL); 3355 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 3356 vfs_unbusy(mp); 3357 if (error) 3358 return (NULL); 3359 return (vp); 3360 } 3361 3362 /* 3363 * Do a local VOP_ADVLOCK(). 3364 */ 3365 int 3366 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 3367 u_int64_t end, struct thread *td) 3368 { 3369 int error = 0; 3370 struct flock fl; 3371 u_int64_t tlen; 3372 3373 if (nfsrv_dolocallocks == 0) 3374 goto out; 3375 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); 3376 3377 fl.l_whence = SEEK_SET; 3378 fl.l_type = ftype; 3379 fl.l_start = (off_t)first; 3380 if (end == NFS64BITSSET) { 3381 fl.l_len = 0; 3382 } else { 3383 tlen = end - first; 3384 fl.l_len = (off_t)tlen; 3385 } 3386 /* 3387 * For FreeBSD8, the l_pid and l_sysid must be set to the same 3388 * values for all calls, so that all locks will be held by the 3389 * nfsd server. (The nfsd server handles conflicts between the 3390 * various clients.) 3391 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 3392 * bytes, so it can't be put in l_sysid. 3393 */ 3394 if (nfsv4_sysid == 0) 3395 nfsv4_sysid = nlm_acquire_next_sysid(); 3396 fl.l_pid = (pid_t)0; 3397 fl.l_sysid = (int)nfsv4_sysid; 3398 3399 if (ftype == F_UNLCK) 3400 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 3401 (F_POSIX | F_REMOTE)); 3402 else 3403 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 3404 (F_POSIX | F_REMOTE)); 3405 3406 out: 3407 NFSEXITCODE(error); 3408 return (error); 3409 } 3410 3411 /* 3412 * Check the nfsv4 root exports. 3413 */ 3414 int 3415 nfsvno_v4rootexport(struct nfsrv_descript *nd) 3416 { 3417 struct ucred *credanon; 3418 int exflags, error = 0, numsecflavor, *secflavors, i; 3419 3420 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags, 3421 &credanon, &numsecflavor, &secflavors); 3422 if (error) { 3423 error = NFSERR_PROGUNAVAIL; 3424 goto out; 3425 } 3426 if (credanon != NULL) 3427 crfree(credanon); 3428 for (i = 0; i < numsecflavor; i++) { 3429 if (secflavors[i] == AUTH_SYS) 3430 nd->nd_flag |= ND_EXAUTHSYS; 3431 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3432 nd->nd_flag |= ND_EXGSS; 3433 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3434 nd->nd_flag |= ND_EXGSSINTEGRITY; 3435 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3436 nd->nd_flag |= ND_EXGSSPRIVACY; 3437 } 3438 3439 out: 3440 NFSEXITCODE(error); 3441 return (error); 3442 } 3443 3444 /* 3445 * Nfs server pseudo system call for the nfsd's 3446 */ 3447 /* 3448 * MPSAFE 3449 */ 3450 static int 3451 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3452 { 3453 struct file *fp; 3454 struct nfsd_addsock_args sockarg; 3455 struct nfsd_nfsd_args nfsdarg; 3456 struct nfsd_nfsd_oargs onfsdarg; 3457 struct nfsd_pnfsd_args pnfsdarg; 3458 struct vnode *vp, *nvp, *curdvp; 3459 struct pnfsdsfile *pf; 3460 struct nfsdevice *ds, *fds; 3461 cap_rights_t rights; 3462 int buflen, error, ret; 3463 char *buf, *cp, *cp2, *cp3; 3464 char fname[PNFS_FILENAME_LEN + 1]; 3465 3466 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3467 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3468 if (error) 3469 goto out; 3470 /* 3471 * Since we don't know what rights might be required, 3472 * pretend that we need them all. It is better to be too 3473 * careful than too reckless. 3474 */ 3475 error = fget(td, sockarg.sock, 3476 cap_rights_init(&rights, CAP_SOCK_SERVER), &fp); 3477 if (error != 0) 3478 goto out; 3479 if (fp->f_type != DTYPE_SOCKET) { 3480 fdrop(fp, td); 3481 error = EPERM; 3482 goto out; 3483 } 3484 error = nfsrvd_addsock(fp); 3485 fdrop(fp, td); 3486 } else if (uap->flag & NFSSVC_NFSDNFSD) { 3487 if (uap->argp == NULL) { 3488 error = EINVAL; 3489 goto out; 3490 } 3491 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { 3492 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); 3493 if (error == 0) { 3494 nfsdarg.principal = onfsdarg.principal; 3495 nfsdarg.minthreads = onfsdarg.minthreads; 3496 nfsdarg.maxthreads = onfsdarg.maxthreads; 3497 nfsdarg.version = 1; 3498 nfsdarg.addr = NULL; 3499 nfsdarg.addrlen = 0; 3500 nfsdarg.dnshost = NULL; 3501 nfsdarg.dnshostlen = 0; 3502 nfsdarg.dspath = NULL; 3503 nfsdarg.dspathlen = 0; 3504 nfsdarg.mdspath = NULL; 3505 nfsdarg.mdspathlen = 0; 3506 nfsdarg.mirrorcnt = 1; 3507 } 3508 } else 3509 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); 3510 if (error) 3511 goto out; 3512 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && 3513 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && 3514 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && 3515 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && 3516 nfsdarg.mirrorcnt >= 1 && 3517 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && 3518 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && 3519 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { 3520 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" 3521 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, 3522 nfsdarg.dspathlen, nfsdarg.dnshostlen, 3523 nfsdarg.mdspathlen, nfsdarg.mirrorcnt); 3524 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); 3525 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); 3526 if (error != 0) { 3527 free(cp, M_TEMP); 3528 goto out; 3529 } 3530 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ 3531 nfsdarg.addr = cp; 3532 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); 3533 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); 3534 if (error != 0) { 3535 free(nfsdarg.addr, M_TEMP); 3536 free(cp, M_TEMP); 3537 goto out; 3538 } 3539 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ 3540 nfsdarg.dnshost = cp; 3541 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); 3542 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); 3543 if (error != 0) { 3544 free(nfsdarg.addr, M_TEMP); 3545 free(nfsdarg.dnshost, M_TEMP); 3546 free(cp, M_TEMP); 3547 goto out; 3548 } 3549 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ 3550 nfsdarg.dspath = cp; 3551 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); 3552 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); 3553 if (error != 0) { 3554 free(nfsdarg.addr, M_TEMP); 3555 free(nfsdarg.dnshost, M_TEMP); 3556 free(nfsdarg.dspath, M_TEMP); 3557 free(cp, M_TEMP); 3558 goto out; 3559 } 3560 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ 3561 nfsdarg.mdspath = cp; 3562 } else { 3563 nfsdarg.addr = NULL; 3564 nfsdarg.addrlen = 0; 3565 nfsdarg.dnshost = NULL; 3566 nfsdarg.dnshostlen = 0; 3567 nfsdarg.dspath = NULL; 3568 nfsdarg.dspathlen = 0; 3569 nfsdarg.mdspath = NULL; 3570 nfsdarg.mdspathlen = 0; 3571 nfsdarg.mirrorcnt = 1; 3572 } 3573 error = nfsrvd_nfsd(td, &nfsdarg); 3574 free(nfsdarg.addr, M_TEMP); 3575 free(nfsdarg.dnshost, M_TEMP); 3576 free(nfsdarg.dspath, M_TEMP); 3577 free(nfsdarg.mdspath, M_TEMP); 3578 } else if (uap->flag & NFSSVC_PNFSDS) { 3579 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); 3580 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || 3581 pnfsdarg.op == PNFSDOP_FORCEDELDS)) { 3582 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3583 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, 3584 NULL); 3585 if (error == 0) 3586 error = nfsrv_deldsserver(pnfsdarg.op, cp, td); 3587 free(cp, M_TEMP); 3588 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { 3589 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3590 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; 3591 buf = malloc(buflen, M_TEMP, M_WAITOK); 3592 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, 3593 NULL); 3594 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); 3595 if (error == 0 && pnfsdarg.dspath != NULL) { 3596 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3597 error = copyinstr(pnfsdarg.dspath, cp2, 3598 PATH_MAX + 1, NULL); 3599 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", 3600 error); 3601 } else 3602 cp2 = NULL; 3603 if (error == 0 && pnfsdarg.curdspath != NULL) { 3604 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3605 error = copyinstr(pnfsdarg.curdspath, cp3, 3606 PATH_MAX + 1, NULL); 3607 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", 3608 error); 3609 } else 3610 cp3 = NULL; 3611 curdvp = NULL; 3612 fds = NULL; 3613 if (error == 0) 3614 error = nfsrv_mdscopymr(cp, cp2, cp3, buf, 3615 &buflen, fname, td, &vp, &nvp, &pf, &ds, 3616 &fds); 3617 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); 3618 if (error == 0) { 3619 if (pf->dsf_dir >= nfsrv_dsdirsize) { 3620 printf("copymr: dsdir out of range\n"); 3621 pf->dsf_dir = 0; 3622 } 3623 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); 3624 error = nfsrv_copymr(vp, nvp, 3625 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, 3626 (struct pnfsdsfile *)buf, 3627 buflen / sizeof(*pf), td->td_ucred, td); 3628 vput(vp); 3629 vput(nvp); 3630 if (fds != NULL && error == 0) { 3631 curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; 3632 ret = vn_lock(curdvp, LK_EXCLUSIVE); 3633 if (ret == 0) { 3634 nfsrv_dsremove(curdvp, fname, 3635 td->td_ucred, td); 3636 NFSVOPUNLOCK(curdvp); 3637 } 3638 } 3639 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); 3640 } 3641 free(cp, M_TEMP); 3642 free(cp2, M_TEMP); 3643 free(cp3, M_TEMP); 3644 free(buf, M_TEMP); 3645 } 3646 } else { 3647 error = nfssvc_srvcall(td, uap, td->td_ucred); 3648 } 3649 3650 out: 3651 NFSEXITCODE(error); 3652 return (error); 3653 } 3654 3655 static int 3656 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 3657 { 3658 struct nfsex_args export; 3659 struct file *fp = NULL; 3660 int stablefd, len; 3661 struct nfsd_clid adminrevoke; 3662 struct nfsd_dumplist dumplist; 3663 struct nfsd_dumpclients *dumpclients; 3664 struct nfsd_dumplocklist dumplocklist; 3665 struct nfsd_dumplocks *dumplocks; 3666 struct nameidata nd; 3667 vnode_t vp; 3668 int error = EINVAL, igotlock; 3669 struct proc *procp; 3670 static int suspend_nfsd = 0; 3671 3672 if (uap->flag & NFSSVC_PUBLICFH) { 3673 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 3674 sizeof (fhandle_t)); 3675 error = copyin(uap->argp, 3676 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 3677 if (!error) 3678 nfs_pubfhset = 1; 3679 } else if (uap->flag & NFSSVC_V4ROOTEXPORT) { 3680 error = copyin(uap->argp,(caddr_t)&export, 3681 sizeof (struct nfsex_args)); 3682 if (!error) 3683 error = nfsrv_v4rootexport(&export, cred, p); 3684 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 3685 nfs_pubfhset = 0; 3686 error = 0; 3687 } else if (uap->flag & NFSSVC_STABLERESTART) { 3688 error = copyin(uap->argp, (caddr_t)&stablefd, 3689 sizeof (int)); 3690 if (!error) 3691 error = fp_getfvp(p, stablefd, &fp, &vp); 3692 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 3693 error = EBADF; 3694 if (!error && newnfs_numnfsd != 0) 3695 error = EPERM; 3696 if (!error) { 3697 nfsrv_stablefirst.nsf_fp = fp; 3698 nfsrv_setupstable(p); 3699 } 3700 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 3701 error = copyin(uap->argp, (caddr_t)&adminrevoke, 3702 sizeof (struct nfsd_clid)); 3703 if (!error) 3704 error = nfsrv_adminrevoke(&adminrevoke, p); 3705 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 3706 error = copyin(uap->argp, (caddr_t)&dumplist, 3707 sizeof (struct nfsd_dumplist)); 3708 if (!error && (dumplist.ndl_size < 1 || 3709 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 3710 error = EPERM; 3711 if (!error) { 3712 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 3713 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 3714 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 3715 error = copyout(dumpclients, dumplist.ndl_list, len); 3716 free(dumpclients, M_TEMP); 3717 } 3718 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 3719 error = copyin(uap->argp, (caddr_t)&dumplocklist, 3720 sizeof (struct nfsd_dumplocklist)); 3721 if (!error && (dumplocklist.ndllck_size < 1 || 3722 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 3723 error = EPERM; 3724 if (!error) 3725 error = nfsrv_lookupfilename(&nd, 3726 dumplocklist.ndllck_fname, p); 3727 if (!error) { 3728 len = sizeof (struct nfsd_dumplocks) * 3729 dumplocklist.ndllck_size; 3730 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 3731 nfsrv_dumplocks(nd.ni_vp, dumplocks, 3732 dumplocklist.ndllck_size, p); 3733 vput(nd.ni_vp); 3734 error = copyout(dumplocks, dumplocklist.ndllck_list, 3735 len); 3736 free(dumplocks, M_TEMP); 3737 } 3738 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 3739 procp = p->td_proc; 3740 PROC_LOCK(procp); 3741 nfsd_master_pid = procp->p_pid; 3742 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 3743 nfsd_master_start = procp->p_stats->p_start; 3744 nfsd_master_proc = procp; 3745 PROC_UNLOCK(procp); 3746 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { 3747 NFSLOCKV4ROOTMUTEX(); 3748 if (suspend_nfsd == 0) { 3749 /* Lock out all nfsd threads */ 3750 do { 3751 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1, 3752 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 3753 } while (igotlock == 0 && suspend_nfsd == 0); 3754 suspend_nfsd = 1; 3755 } 3756 NFSUNLOCKV4ROOTMUTEX(); 3757 error = 0; 3758 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { 3759 NFSLOCKV4ROOTMUTEX(); 3760 if (suspend_nfsd != 0) { 3761 nfsv4_unlock(&nfsd_suspend_lock, 0); 3762 suspend_nfsd = 0; 3763 } 3764 NFSUNLOCKV4ROOTMUTEX(); 3765 error = 0; 3766 } 3767 3768 NFSEXITCODE(error); 3769 return (error); 3770 } 3771 3772 /* 3773 * Check exports. 3774 * Returns 0 if ok, 1 otherwise. 3775 */ 3776 int 3777 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 3778 { 3779 int i; 3780 3781 /* 3782 * This seems odd, but allow the case where the security flavor 3783 * list is empty. This happens when NFSv4 is traversing non-exported 3784 * file systems. Exported file systems should always have a non-empty 3785 * security flavor list. 3786 */ 3787 if (exp->nes_numsecflavor == 0) 3788 return (0); 3789 3790 for (i = 0; i < exp->nes_numsecflavor; i++) { 3791 /* 3792 * The tests for privacy and integrity must be first, 3793 * since ND_GSS is set for everything but AUTH_SYS. 3794 */ 3795 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 3796 (nd->nd_flag & ND_GSSPRIVACY)) 3797 return (0); 3798 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 3799 (nd->nd_flag & ND_GSSINTEGRITY)) 3800 return (0); 3801 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 3802 (nd->nd_flag & ND_GSS)) 3803 return (0); 3804 if (exp->nes_secflavors[i] == AUTH_SYS && 3805 (nd->nd_flag & ND_GSS) == 0) 3806 return (0); 3807 } 3808 return (1); 3809 } 3810 3811 /* 3812 * Calculate a hash value for the fid in a file handle. 3813 */ 3814 uint32_t 3815 nfsrv_hashfh(fhandle_t *fhp) 3816 { 3817 uint32_t hashval; 3818 3819 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 3820 return (hashval); 3821 } 3822 3823 /* 3824 * Calculate a hash value for the sessionid. 3825 */ 3826 uint32_t 3827 nfsrv_hashsessionid(uint8_t *sessionid) 3828 { 3829 uint32_t hashval; 3830 3831 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); 3832 return (hashval); 3833 } 3834 3835 /* 3836 * Signal the userland master nfsd to backup the stable restart file. 3837 */ 3838 void 3839 nfsrv_backupstable(void) 3840 { 3841 struct proc *procp; 3842 3843 if (nfsd_master_proc != NULL) { 3844 procp = pfind(nfsd_master_pid); 3845 /* Try to make sure it is the correct process. */ 3846 if (procp == nfsd_master_proc && 3847 procp->p_stats->p_start.tv_sec == 3848 nfsd_master_start.tv_sec && 3849 procp->p_stats->p_start.tv_usec == 3850 nfsd_master_start.tv_usec && 3851 strcmp(procp->p_comm, nfsd_master_comm) == 0) 3852 kern_psignal(procp, SIGUSR2); 3853 else 3854 nfsd_master_proc = NULL; 3855 3856 if (procp != NULL) 3857 PROC_UNLOCK(procp); 3858 } 3859 } 3860 3861 /* 3862 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. 3863 * The arguments are in a structure, so that they can be passed through 3864 * taskqueue for a kernel process to execute this function. 3865 */ 3866 struct nfsrvdscreate { 3867 int done; 3868 int inprog; 3869 struct task tsk; 3870 struct ucred *tcred; 3871 struct vnode *dvp; 3872 NFSPROC_T *p; 3873 struct pnfsdsfile *pf; 3874 int err; 3875 fhandle_t fh; 3876 struct vattr va; 3877 struct vattr createva; 3878 }; 3879 3880 int 3881 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, 3882 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, 3883 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) 3884 { 3885 struct vnode *nvp; 3886 struct nameidata named; 3887 struct vattr va; 3888 char *bufp; 3889 u_long *hashp; 3890 struct nfsnode *np; 3891 struct nfsmount *nmp; 3892 int error; 3893 3894 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, 3895 LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE); 3896 nfsvno_setpathbuf(&named, &bufp, &hashp); 3897 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 3898 named.ni_cnd.cn_thread = p; 3899 named.ni_cnd.cn_nameptr = bufp; 3900 if (fnamep != NULL) { 3901 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); 3902 named.ni_cnd.cn_namelen = strlen(bufp); 3903 } else 3904 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); 3905 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); 3906 3907 /* Create the date file in the DS mount. */ 3908 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 3909 if (error == 0) { 3910 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); 3911 NFSVOPUNLOCK(dvp); 3912 if (error == 0) { 3913 /* Set the ownership of the file. */ 3914 error = VOP_SETATTR(nvp, nvap, tcred); 3915 NFSD_DEBUG(4, "nfsrv_dscreate:" 3916 " setattr-uid=%d\n", error); 3917 if (error != 0) 3918 vput(nvp); 3919 } 3920 if (error != 0) 3921 printf("pNFS: pnfscreate failed=%d\n", error); 3922 } else 3923 printf("pNFS: pnfscreate vnlock=%d\n", error); 3924 if (error == 0) { 3925 np = VTONFS(nvp); 3926 nmp = VFSTONFS(nvp->v_mount); 3927 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") 3928 != 0 || nmp->nm_nam->sa_len > sizeof( 3929 struct sockaddr_in6) || 3930 np->n_fhp->nfh_len != NFSX_MYFH) { 3931 printf("Bad DS file: fstype=%s salen=%d" 3932 " fhlen=%d\n", 3933 nvp->v_mount->mnt_vfc->vfc_name, 3934 nmp->nm_nam->sa_len, np->n_fhp->nfh_len); 3935 error = ENOENT; 3936 } 3937 3938 /* Set extattrs for the DS on the MDS file. */ 3939 if (error == 0) { 3940 if (dsa != NULL) { 3941 error = VOP_GETATTR(nvp, &va, tcred); 3942 if (error == 0) { 3943 dsa->dsa_filerev = va.va_filerev; 3944 dsa->dsa_size = va.va_size; 3945 dsa->dsa_atime = va.va_atime; 3946 dsa->dsa_mtime = va.va_mtime; 3947 dsa->dsa_bytes = va.va_bytes; 3948 } 3949 } 3950 if (error == 0) { 3951 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, 3952 NFSX_MYFH); 3953 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, 3954 nmp->nm_nam->sa_len); 3955 NFSBCOPY(named.ni_cnd.cn_nameptr, 3956 pf->dsf_filename, 3957 sizeof(pf->dsf_filename)); 3958 } 3959 } else 3960 printf("pNFS: pnfscreate can't get DS" 3961 " attr=%d\n", error); 3962 if (nvpp != NULL && error == 0) 3963 *nvpp = nvp; 3964 else 3965 vput(nvp); 3966 } 3967 nfsvno_relpathbuf(&named); 3968 return (error); 3969 } 3970 3971 /* 3972 * Start up the thread that will execute nfsrv_dscreate(). 3973 */ 3974 static void 3975 start_dscreate(void *arg, int pending) 3976 { 3977 struct nfsrvdscreate *dsc; 3978 3979 dsc = (struct nfsrvdscreate *)arg; 3980 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, 3981 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); 3982 dsc->done = 1; 3983 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); 3984 } 3985 3986 /* 3987 * Create a pNFS data file on the Data Server(s). 3988 */ 3989 static void 3990 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, 3991 NFSPROC_T *p) 3992 { 3993 struct nfsrvdscreate *dsc, *tdsc = NULL; 3994 struct nfsdevice *ds, *tds, *fds; 3995 struct mount *mp; 3996 struct pnfsdsfile *pf, *tpf; 3997 struct pnfsdsattr dsattr; 3998 struct vattr va; 3999 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4000 struct nfsmount *nmp; 4001 fhandle_t fh; 4002 uid_t vauid; 4003 gid_t vagid; 4004 u_short vamode; 4005 struct ucred *tcred; 4006 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; 4007 int failpos, timo; 4008 4009 /* Get a DS server directory in a round-robin order. */ 4010 mirrorcnt = 1; 4011 mp = vp->v_mount; 4012 ds = fds = NULL; 4013 NFSDDSLOCK(); 4014 /* 4015 * Search for the first entry that handles this MDS fs, but use the 4016 * first entry for all MDS fs's otherwise. 4017 */ 4018 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { 4019 if (tds->nfsdev_nmp != NULL) { 4020 if (tds->nfsdev_mdsisset == 0 && ds == NULL) 4021 ds = tds; 4022 else if (tds->nfsdev_mdsisset != 0 && fsidcmp( 4023 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { 4024 ds = fds = tds; 4025 break; 4026 } 4027 } 4028 } 4029 if (ds == NULL) { 4030 NFSDDSUNLOCK(); 4031 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); 4032 return; 4033 } 4034 i = dsdir[0] = ds->nfsdev_nextdir; 4035 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; 4036 dvp[0] = ds->nfsdev_dsdir[i]; 4037 tds = TAILQ_NEXT(ds, nfsdev_list); 4038 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { 4039 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { 4040 if (tds->nfsdev_nmp != NULL && 4041 ((tds->nfsdev_mdsisset == 0 && fds == NULL) || 4042 (tds->nfsdev_mdsisset != 0 && fds != NULL && 4043 fsidcmp(&mp->mnt_stat.f_fsid, 4044 &tds->nfsdev_mdsfsid) == 0))) { 4045 dsdir[mirrorcnt] = i; 4046 dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; 4047 mirrorcnt++; 4048 if (mirrorcnt >= nfsrv_maxpnfsmirror) 4049 break; 4050 } 4051 } 4052 } 4053 /* Put at end of list to implement round-robin usage. */ 4054 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 4055 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 4056 NFSDDSUNLOCK(); 4057 dsc = NULL; 4058 if (mirrorcnt > 1) 4059 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, 4060 M_WAITOK | M_ZERO); 4061 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | 4062 M_ZERO); 4063 4064 error = nfsvno_getfh(vp, &fh, p); 4065 if (error == 0) 4066 error = VOP_GETATTR(vp, &va, cred); 4067 if (error == 0) { 4068 /* Set the attributes for "vp" to Setattr the DS vp. */ 4069 vauid = va.va_uid; 4070 vagid = va.va_gid; 4071 vamode = va.va_mode; 4072 VATTR_NULL(&va); 4073 va.va_uid = vauid; 4074 va.va_gid = vagid; 4075 va.va_mode = vamode; 4076 va.va_size = 0; 4077 } else 4078 printf("pNFS: pnfscreate getfh+attr=%d\n", error); 4079 4080 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, 4081 cred->cr_gid); 4082 /* Make data file name based on FH. */ 4083 tcred = newnfs_getcred(); 4084 4085 /* 4086 * Create the file on each DS mirror, using kernel process(es) for the 4087 * additional mirrors. 4088 */ 4089 failpos = -1; 4090 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { 4091 tpf->dsf_dir = dsdir[i]; 4092 tdsc->tcred = tcred; 4093 tdsc->p = p; 4094 tdsc->pf = tpf; 4095 tdsc->createva = *vap; 4096 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); 4097 tdsc->va = va; 4098 tdsc->dvp = dvp[i]; 4099 tdsc->done = 0; 4100 tdsc->inprog = 0; 4101 tdsc->err = 0; 4102 ret = EIO; 4103 if (nfs_pnfsiothreads != 0) { 4104 ret = nfs_pnfsio(start_dscreate, tdsc); 4105 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); 4106 } 4107 if (ret != 0) { 4108 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, 4109 NULL, tcred, p, NULL); 4110 if (ret != 0) { 4111 KASSERT(error == 0, ("nfsrv_dscreate err=%d", 4112 error)); 4113 if (failpos == -1 && nfsds_failerr(ret)) 4114 failpos = i; 4115 else 4116 error = ret; 4117 } 4118 } 4119 } 4120 if (error == 0) { 4121 tpf->dsf_dir = dsdir[mirrorcnt - 1]; 4122 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, 4123 &dsattr, NULL, tcred, p, NULL); 4124 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { 4125 failpos = mirrorcnt - 1; 4126 error = 0; 4127 } 4128 } 4129 timo = hz / 50; /* Wait for 20msec. */ 4130 if (timo < 1) 4131 timo = 1; 4132 /* Wait for kernel task(s) to complete. */ 4133 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { 4134 while (tdsc->inprog != 0 && tdsc->done == 0) 4135 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); 4136 if (tdsc->err != 0) { 4137 if (failpos == -1 && nfsds_failerr(tdsc->err)) 4138 failpos = i; 4139 else if (error == 0) 4140 error = tdsc->err; 4141 } 4142 } 4143 4144 /* 4145 * If failpos has been set, that mirror has failed, so it needs 4146 * to be disabled. 4147 */ 4148 if (failpos >= 0) { 4149 nmp = VFSTONFS(dvp[failpos]->v_mount); 4150 NFSLOCKMNT(nmp); 4151 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4152 NFSMNTP_CANCELRPCS)) == 0) { 4153 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4154 NFSUNLOCKMNT(nmp); 4155 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4156 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, 4157 ds); 4158 if (ds != NULL) 4159 nfsrv_killrpcs(nmp); 4160 NFSLOCKMNT(nmp); 4161 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4162 wakeup(nmp); 4163 } 4164 NFSUNLOCKMNT(nmp); 4165 } 4166 4167 NFSFREECRED(tcred); 4168 if (error == 0) { 4169 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); 4170 4171 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", 4172 mirrorcnt, nfsrv_maxpnfsmirror); 4173 /* 4174 * For all mirrors that couldn't be created, fill in the 4175 * *pf structure, but with an IP address == 0.0.0.0. 4176 */ 4177 tpf = pf + mirrorcnt; 4178 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { 4179 *tpf = *pf; 4180 tpf->dsf_sin.sin_family = AF_INET; 4181 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); 4182 tpf->dsf_sin.sin_addr.s_addr = 0; 4183 tpf->dsf_sin.sin_port = 0; 4184 } 4185 4186 error = vn_extattr_set(vp, IO_NODELOCKED, 4187 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 4188 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); 4189 if (error == 0) 4190 error = vn_extattr_set(vp, IO_NODELOCKED, 4191 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", 4192 sizeof(dsattr), (char *)&dsattr, p); 4193 if (error != 0) 4194 printf("pNFS: pnfscreate setextattr=%d\n", 4195 error); 4196 } else 4197 printf("pNFS: pnfscreate=%d\n", error); 4198 free(pf, M_TEMP); 4199 free(dsc, M_TEMP); 4200 } 4201 4202 /* 4203 * Get the information needed to remove the pNFS Data Server file from the 4204 * Metadata file. Upon success, ddvp is set non-NULL to the locked 4205 * DS directory vnode. The caller must unlock *ddvp when done with it. 4206 */ 4207 static void 4208 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, 4209 int *mirrorcntp, char *fname, fhandle_t *fhp) 4210 { 4211 struct vattr va; 4212 struct ucred *tcred; 4213 char *buf; 4214 int buflen, error; 4215 4216 dvpp[0] = NULL; 4217 /* If not an exported regular file or not a pNFS server, just return. */ 4218 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4219 nfsrv_devidcnt == 0) 4220 return; 4221 4222 /* Check to see if this is the last hard link. */ 4223 tcred = newnfs_getcred(); 4224 error = VOP_GETATTR(vp, &va, tcred); 4225 NFSFREECRED(tcred); 4226 if (error != 0) { 4227 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); 4228 return; 4229 } 4230 if (va.va_nlink > 1) 4231 return; 4232 4233 error = nfsvno_getfh(vp, fhp, p); 4234 if (error != 0) { 4235 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); 4236 return; 4237 } 4238 4239 buflen = 1024; 4240 buf = malloc(buflen, M_TEMP, M_WAITOK); 4241 /* Get the directory vnode for the DS mount and the file handle. */ 4242 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, 4243 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); 4244 free(buf, M_TEMP); 4245 if (error != 0) 4246 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); 4247 } 4248 4249 /* 4250 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. 4251 * The arguments are in a structure, so that they can be passed through 4252 * taskqueue for a kernel process to execute this function. 4253 */ 4254 struct nfsrvdsremove { 4255 int done; 4256 int inprog; 4257 struct task tsk; 4258 struct ucred *tcred; 4259 struct vnode *dvp; 4260 NFSPROC_T *p; 4261 int err; 4262 char fname[PNFS_FILENAME_LEN + 1]; 4263 }; 4264 4265 static int 4266 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, 4267 NFSPROC_T *p) 4268 { 4269 struct nameidata named; 4270 struct vnode *nvp; 4271 char *bufp; 4272 u_long *hashp; 4273 int error; 4274 4275 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4276 if (error != 0) 4277 return (error); 4278 named.ni_cnd.cn_nameiop = DELETE; 4279 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 4280 named.ni_cnd.cn_cred = tcred; 4281 named.ni_cnd.cn_thread = p; 4282 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; 4283 nfsvno_setpathbuf(&named, &bufp, &hashp); 4284 named.ni_cnd.cn_nameptr = bufp; 4285 named.ni_cnd.cn_namelen = strlen(fname); 4286 strlcpy(bufp, fname, NAME_MAX); 4287 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); 4288 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 4289 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); 4290 if (error == 0) { 4291 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); 4292 vput(nvp); 4293 } 4294 NFSVOPUNLOCK(dvp); 4295 nfsvno_relpathbuf(&named); 4296 if (error != 0) 4297 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); 4298 return (error); 4299 } 4300 4301 /* 4302 * Start up the thread that will execute nfsrv_dsremove(). 4303 */ 4304 static void 4305 start_dsremove(void *arg, int pending) 4306 { 4307 struct nfsrvdsremove *dsrm; 4308 4309 dsrm = (struct nfsrvdsremove *)arg; 4310 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, 4311 dsrm->p); 4312 dsrm->done = 1; 4313 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); 4314 } 4315 4316 /* 4317 * Remove a pNFS data file from a Data Server. 4318 * nfsrv_pnfsremovesetup() must have been called before the MDS file was 4319 * removed to set up the dvp and fill in the FH. 4320 */ 4321 static void 4322 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, 4323 NFSPROC_T *p) 4324 { 4325 struct ucred *tcred; 4326 struct nfsrvdsremove *dsrm, *tdsrm; 4327 struct nfsdevice *ds; 4328 struct nfsmount *nmp; 4329 int failpos, i, ret, timo; 4330 4331 tcred = newnfs_getcred(); 4332 dsrm = NULL; 4333 if (mirrorcnt > 1) 4334 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); 4335 /* 4336 * Remove the file on each DS mirror, using kernel process(es) for the 4337 * additional mirrors. 4338 */ 4339 failpos = -1; 4340 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4341 tdsrm->tcred = tcred; 4342 tdsrm->p = p; 4343 tdsrm->dvp = dvp[i]; 4344 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); 4345 tdsrm->inprog = 0; 4346 tdsrm->done = 0; 4347 tdsrm->err = 0; 4348 ret = EIO; 4349 if (nfs_pnfsiothreads != 0) { 4350 ret = nfs_pnfsio(start_dsremove, tdsrm); 4351 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); 4352 } 4353 if (ret != 0) { 4354 ret = nfsrv_dsremove(dvp[i], fname, tcred, p); 4355 if (failpos == -1 && nfsds_failerr(ret)) 4356 failpos = i; 4357 } 4358 } 4359 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); 4360 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) 4361 failpos = mirrorcnt - 1; 4362 timo = hz / 50; /* Wait for 20msec. */ 4363 if (timo < 1) 4364 timo = 1; 4365 /* Wait for kernel task(s) to complete. */ 4366 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4367 while (tdsrm->inprog != 0 && tdsrm->done == 0) 4368 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); 4369 if (failpos == -1 && nfsds_failerr(tdsrm->err)) 4370 failpos = i; 4371 } 4372 4373 /* 4374 * If failpos has been set, that mirror has failed, so it needs 4375 * to be disabled. 4376 */ 4377 if (failpos >= 0) { 4378 nmp = VFSTONFS(dvp[failpos]->v_mount); 4379 NFSLOCKMNT(nmp); 4380 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4381 NFSMNTP_CANCELRPCS)) == 0) { 4382 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4383 NFSUNLOCKMNT(nmp); 4384 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4385 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, 4386 ds); 4387 if (ds != NULL) 4388 nfsrv_killrpcs(nmp); 4389 NFSLOCKMNT(nmp); 4390 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4391 wakeup(nmp); 4392 } 4393 NFSUNLOCKMNT(nmp); 4394 } 4395 4396 /* Get rid all layouts for the file. */ 4397 nfsrv_freefilelayouts(fhp); 4398 4399 NFSFREECRED(tcred); 4400 free(dsrm, M_TEMP); 4401 } 4402 4403 /* 4404 * Generate a file name based on the file handle and put it in *bufp. 4405 * Return the number of bytes generated. 4406 */ 4407 static int 4408 nfsrv_putfhname(fhandle_t *fhp, char *bufp) 4409 { 4410 int i; 4411 uint8_t *cp; 4412 const uint8_t *hexdigits = "0123456789abcdef"; 4413 4414 cp = (uint8_t *)fhp; 4415 for (i = 0; i < sizeof(*fhp); i++) { 4416 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; 4417 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; 4418 } 4419 bufp[2 * i] = '\0'; 4420 return (2 * i); 4421 } 4422 4423 /* 4424 * Update the Metadata file's attributes from the DS file when a Read/Write 4425 * layout is returned. 4426 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN 4427 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. 4428 */ 4429 int 4430 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 4431 { 4432 struct ucred *tcred; 4433 int error; 4434 4435 /* Do this as root so that it won't fail with EACCES. */ 4436 tcred = newnfs_getcred(); 4437 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, 4438 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); 4439 NFSFREECRED(tcred); 4440 return (error); 4441 } 4442 4443 /* 4444 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. 4445 */ 4446 static int 4447 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, 4448 NFSPROC_T *p) 4449 { 4450 int error; 4451 4452 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, 4453 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); 4454 return (error); 4455 } 4456 4457 static int 4458 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 4459 struct thread *p, int ioproc, struct mbuf **mpp, char *cp, 4460 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, 4461 off_t *offp, int content, bool *eofp) 4462 { 4463 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; 4464 fhandle_t fh[NFSDEV_MAXMIRRORS]; 4465 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4466 struct nfsdevice *ds; 4467 struct pnfsdsattr dsattr; 4468 struct opnfsdsattr odsattr; 4469 char *buf; 4470 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; 4471 4472 NFSD_DEBUG(4, "in nfsrv_proxyds\n"); 4473 /* 4474 * If not a regular file, not exported or not a pNFS server, 4475 * just return ENOENT. 4476 */ 4477 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4478 nfsrv_devidcnt == 0) 4479 return (ENOENT); 4480 4481 buflen = 1024; 4482 buf = malloc(buflen, M_TEMP, M_WAITOK); 4483 error = 0; 4484 4485 /* 4486 * For Getattr, get the Change attribute (va_filerev) and size (va_size) 4487 * from the MetaData file's extended attribute. 4488 */ 4489 if (ioproc == NFSPROC_GETATTR) { 4490 error = vn_extattr_get(vp, IO_NODELOCKED, 4491 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, 4492 p); 4493 if (error == 0) { 4494 if (buflen == sizeof(odsattr)) { 4495 NFSBCOPY(buf, &odsattr, buflen); 4496 nap->na_filerev = odsattr.dsa_filerev; 4497 nap->na_size = odsattr.dsa_size; 4498 nap->na_atime = odsattr.dsa_atime; 4499 nap->na_mtime = odsattr.dsa_mtime; 4500 /* 4501 * Fake na_bytes by rounding up na_size. 4502 * Since we don't know the block size, just 4503 * use BLKDEV_IOSIZE. 4504 */ 4505 nap->na_bytes = (odsattr.dsa_size + 4506 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); 4507 } else if (buflen == sizeof(dsattr)) { 4508 NFSBCOPY(buf, &dsattr, buflen); 4509 nap->na_filerev = dsattr.dsa_filerev; 4510 nap->na_size = dsattr.dsa_size; 4511 nap->na_atime = dsattr.dsa_atime; 4512 nap->na_mtime = dsattr.dsa_mtime; 4513 nap->na_bytes = dsattr.dsa_bytes; 4514 } else 4515 error = ENXIO; 4516 } 4517 if (error == 0) { 4518 /* 4519 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() 4520 * returns 0, just return now. nfsrv_checkdsattr() 4521 * returns 0 if there is no Read/Write layout 4522 * plus either an Open/Write_access or Write 4523 * delegation issued to a client for the file. 4524 */ 4525 if (nfsrv_pnfsgetdsattr == 0 || 4526 nfsrv_checkdsattr(vp, p) == 0) { 4527 free(buf, M_TEMP); 4528 return (error); 4529 } 4530 } 4531 4532 /* 4533 * Clear ENOATTR so the code below will attempt to do a 4534 * nfsrv_getattrdsrpc() to get the attributes and (re)create 4535 * the extended attribute. 4536 */ 4537 if (error == ENOATTR) 4538 error = 0; 4539 } 4540 4541 origmircnt = -1; 4542 trycnt = 0; 4543 tryagain: 4544 if (error == 0) { 4545 buflen = 1024; 4546 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == 4547 LK_EXCLUSIVE) 4548 printf("nfsrv_proxyds: Readds vp exclusively locked\n"); 4549 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, 4550 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, 4551 NULL, NULL); 4552 if (error == 0) { 4553 for (i = 0; i < mirrorcnt; i++) 4554 nmp[i] = VFSTONFS(dvp[i]->v_mount); 4555 } else 4556 printf("pNFS: proxy getextattr sockaddr=%d\n", error); 4557 } else 4558 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); 4559 if (error == 0) { 4560 failpos = -1; 4561 if (origmircnt == -1) 4562 origmircnt = mirrorcnt; 4563 /* 4564 * If failpos is set to a mirror#, then that mirror has 4565 * failed and will be disabled. For Read, Getattr and Seek, the 4566 * function only tries one mirror, so if that mirror has 4567 * failed, it will need to be retried. As such, increment 4568 * tryitagain for these cases. 4569 * For Write, Setattr and Setacl, the function tries all 4570 * mirrors and will not return an error for the case where 4571 * one mirror has failed. For these cases, the functioning 4572 * mirror(s) will have been modified, so a retry isn't 4573 * necessary. These functions will set failpos for the 4574 * failed mirror#. 4575 */ 4576 if (ioproc == NFSPROC_READDS) { 4577 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], 4578 mpp, mpp2); 4579 if (nfsds_failerr(error) && mirrorcnt > 1) { 4580 /* 4581 * Setting failpos will cause the mirror 4582 * to be disabled and then a retry of this 4583 * read is required. 4584 */ 4585 failpos = 0; 4586 error = 0; 4587 trycnt++; 4588 } 4589 } else if (ioproc == NFSPROC_WRITEDS) 4590 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, 4591 &nmp[0], mirrorcnt, mpp, cp, &failpos); 4592 else if (ioproc == NFSPROC_SETATTR) 4593 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], 4594 mirrorcnt, nap, &failpos); 4595 else if (ioproc == NFSPROC_SETACL) 4596 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], 4597 mirrorcnt, aclp, &failpos); 4598 else if (ioproc == NFSPROC_SEEKDS) { 4599 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, 4600 p, nmp[0]); 4601 if (nfsds_failerr(error) && mirrorcnt > 1) { 4602 /* 4603 * Setting failpos will cause the mirror 4604 * to be disabled and then a retry of this 4605 * read is required. 4606 */ 4607 failpos = 0; 4608 error = 0; 4609 trycnt++; 4610 } 4611 } else if (ioproc == NFSPROC_ALLOCATE) 4612 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, 4613 &nmp[0], mirrorcnt, &failpos); 4614 else { 4615 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, 4616 vp, nmp[mirrorcnt - 1], nap); 4617 if (nfsds_failerr(error) && mirrorcnt > 1) { 4618 /* 4619 * Setting failpos will cause the mirror 4620 * to be disabled and then a retry of this 4621 * getattr is required. 4622 */ 4623 failpos = mirrorcnt - 1; 4624 error = 0; 4625 trycnt++; 4626 } 4627 } 4628 ds = NULL; 4629 if (failpos >= 0) { 4630 failnmp = nmp[failpos]; 4631 NFSLOCKMNT(failnmp); 4632 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | 4633 NFSMNTP_CANCELRPCS)) == 0) { 4634 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4635 NFSUNLOCKMNT(failnmp); 4636 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, 4637 failnmp, p); 4638 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", 4639 failpos, ds); 4640 if (ds != NULL) 4641 nfsrv_killrpcs(failnmp); 4642 NFSLOCKMNT(failnmp); 4643 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4644 wakeup(failnmp); 4645 } 4646 NFSUNLOCKMNT(failnmp); 4647 } 4648 for (i = 0; i < mirrorcnt; i++) 4649 NFSVOPUNLOCK(dvp[i]); 4650 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, 4651 trycnt); 4652 /* Try the Read/Getattr again if a mirror was deleted. */ 4653 if (ds != NULL && trycnt > 0 && trycnt < origmircnt) 4654 goto tryagain; 4655 } else { 4656 /* Return ENOENT for any Extended Attribute error. */ 4657 error = ENOENT; 4658 } 4659 free(buf, M_TEMP); 4660 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); 4661 return (error); 4662 } 4663 4664 /* 4665 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended 4666 * attribute. 4667 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs 4668 * to be checked. If it points to a NULL nmp, then it returns 4669 * a suitable destination. 4670 * curnmp - If non-NULL, it is the source mount for the copy. 4671 */ 4672 int 4673 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, 4674 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, 4675 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, 4676 struct nfsmount *curnmp, int *ippos, int *dsdirp) 4677 { 4678 struct vnode *dvp, *nvp = NULL, **tdvpp; 4679 struct mount *mp; 4680 struct nfsmount *nmp, *newnmp; 4681 struct sockaddr *sad; 4682 struct sockaddr_in *sin; 4683 struct nfsdevice *ds, *tds, *fndds; 4684 struct pnfsdsfile *pf; 4685 uint32_t dsdir; 4686 int error, fhiszero, fnd, gotone, i, mirrorcnt; 4687 4688 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); 4689 *mirrorcntp = 1; 4690 tdvpp = dvpp; 4691 if (nvpp != NULL) 4692 *nvpp = NULL; 4693 if (dvpp != NULL) 4694 *dvpp = NULL; 4695 if (ippos != NULL) 4696 *ippos = -1; 4697 if (newnmpp != NULL) 4698 newnmp = *newnmpp; 4699 else 4700 newnmp = NULL; 4701 mp = vp->v_mount; 4702 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 4703 "pnfsd.dsfile", buflenp, buf, p); 4704 mirrorcnt = *buflenp / sizeof(*pf); 4705 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || 4706 *buflenp != sizeof(*pf) * mirrorcnt)) 4707 error = ENOATTR; 4708 4709 pf = (struct pnfsdsfile *)buf; 4710 /* If curnmp != NULL, check for a match in the mirror list. */ 4711 if (curnmp != NULL && error == 0) { 4712 fnd = 0; 4713 for (i = 0; i < mirrorcnt; i++, pf++) { 4714 sad = (struct sockaddr *)&pf->dsf_sin; 4715 if (nfsaddr2_match(sad, curnmp->nm_nam)) { 4716 if (ippos != NULL) 4717 *ippos = i; 4718 fnd = 1; 4719 break; 4720 } 4721 } 4722 if (fnd == 0) 4723 error = ENXIO; 4724 } 4725 4726 gotone = 0; 4727 pf = (struct pnfsdsfile *)buf; 4728 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, 4729 error); 4730 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { 4731 fhiszero = 0; 4732 sad = (struct sockaddr *)&pf->dsf_sin; 4733 sin = &pf->dsf_sin; 4734 dsdir = pf->dsf_dir; 4735 if (dsdir >= nfsrv_dsdirsize) { 4736 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); 4737 error = ENOATTR; 4738 } else if (nvpp != NULL && newnmp != NULL && 4739 nfsaddr2_match(sad, newnmp->nm_nam)) 4740 error = EEXIST; 4741 if (error == 0) { 4742 if (ippos != NULL && curnmp == NULL && 4743 sad->sa_family == AF_INET && 4744 sin->sin_addr.s_addr == 0) 4745 *ippos = i; 4746 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) 4747 fhiszero = 1; 4748 /* Use the socket address to find the mount point. */ 4749 fndds = NULL; 4750 NFSDDSLOCK(); 4751 /* Find a match for the IP address. */ 4752 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 4753 if (ds->nfsdev_nmp != NULL) { 4754 dvp = ds->nfsdev_dvp; 4755 nmp = VFSTONFS(dvp->v_mount); 4756 if (nmp != ds->nfsdev_nmp) 4757 printf("different2 nmp %p %p\n", 4758 nmp, ds->nfsdev_nmp); 4759 if (nfsaddr2_match(sad, nmp->nm_nam)) { 4760 fndds = ds; 4761 break; 4762 } 4763 } 4764 } 4765 if (fndds != NULL && newnmpp != NULL && 4766 newnmp == NULL) { 4767 /* Search for a place to make a mirror copy. */ 4768 TAILQ_FOREACH(tds, &nfsrv_devidhead, 4769 nfsdev_list) { 4770 if (tds->nfsdev_nmp != NULL && 4771 fndds != tds && 4772 ((tds->nfsdev_mdsisset == 0 && 4773 fndds->nfsdev_mdsisset == 0) || 4774 (tds->nfsdev_mdsisset != 0 && 4775 fndds->nfsdev_mdsisset != 0 && 4776 fsidcmp(&tds->nfsdev_mdsfsid, 4777 &mp->mnt_stat.f_fsid) == 0))) { 4778 *newnmpp = tds->nfsdev_nmp; 4779 break; 4780 } 4781 } 4782 if (tds != NULL) { 4783 /* 4784 * Move this entry to the end of the 4785 * list, so it won't be selected as 4786 * easily the next time. 4787 */ 4788 TAILQ_REMOVE(&nfsrv_devidhead, tds, 4789 nfsdev_list); 4790 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, 4791 nfsdev_list); 4792 } 4793 } 4794 NFSDDSUNLOCK(); 4795 if (fndds != NULL) { 4796 dvp = fndds->nfsdev_dsdir[dsdir]; 4797 if (lktype != 0 || fhiszero != 0 || 4798 (nvpp != NULL && *nvpp == NULL)) { 4799 if (fhiszero != 0) 4800 error = vn_lock(dvp, 4801 LK_EXCLUSIVE); 4802 else if (lktype != 0) 4803 error = vn_lock(dvp, lktype); 4804 else 4805 error = vn_lock(dvp, LK_SHARED); 4806 /* 4807 * If the file handle is all 0's, try to 4808 * do a Lookup against the DS to acquire 4809 * it. 4810 * If dvpp == NULL or the Lookup fails, 4811 * unlock dvp after the call. 4812 */ 4813 if (error == 0 && (fhiszero != 0 || 4814 (nvpp != NULL && *nvpp == NULL))) { 4815 error = nfsrv_pnfslookupds(vp, 4816 dvp, pf, &nvp, p); 4817 if (error == 0) { 4818 if (fhiszero != 0) 4819 nfsrv_pnfssetfh( 4820 vp, pf, 4821 devid, 4822 fnamep, 4823 nvp, p); 4824 if (nvpp != NULL && 4825 *nvpp == NULL) { 4826 *nvpp = nvp; 4827 *dsdirp = dsdir; 4828 } else 4829 vput(nvp); 4830 } 4831 if (error != 0 || lktype == 0) 4832 NFSVOPUNLOCK(dvp); 4833 } 4834 } 4835 if (error == 0) { 4836 gotone++; 4837 NFSD_DEBUG(4, "gotone=%d\n", gotone); 4838 if (devid != NULL) { 4839 NFSBCOPY(fndds->nfsdev_deviceid, 4840 devid, NFSX_V4DEVICEID); 4841 devid += NFSX_V4DEVICEID; 4842 } 4843 if (dvpp != NULL) 4844 *tdvpp++ = dvp; 4845 if (fhp != NULL) 4846 NFSBCOPY(&pf->dsf_fh, fhp++, 4847 NFSX_MYFH); 4848 if (fnamep != NULL && gotone == 1) 4849 strlcpy(fnamep, 4850 pf->dsf_filename, 4851 sizeof(pf->dsf_filename)); 4852 } else 4853 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " 4854 "err=%d\n", error); 4855 } 4856 } 4857 } 4858 if (error == 0 && gotone == 0) 4859 error = ENOENT; 4860 4861 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, 4862 error); 4863 if (error == 0) 4864 *mirrorcntp = gotone; 4865 else { 4866 if (gotone > 0 && dvpp != NULL) { 4867 /* 4868 * If the error didn't occur on the first one and 4869 * dvpp != NULL, the one(s) prior to the failure will 4870 * have locked dvp's that need to be unlocked. 4871 */ 4872 for (i = 0; i < gotone; i++) { 4873 NFSVOPUNLOCK(*dvpp); 4874 *dvpp++ = NULL; 4875 } 4876 } 4877 /* 4878 * If it found the vnode to be copied from before a failure, 4879 * it needs to be vput()'d. 4880 */ 4881 if (nvpp != NULL && *nvpp != NULL) { 4882 vput(*nvpp); 4883 *nvpp = NULL; 4884 } 4885 } 4886 return (error); 4887 } 4888 4889 /* 4890 * Set the extended attribute for the Change attribute. 4891 */ 4892 static int 4893 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 4894 { 4895 struct pnfsdsattr dsattr; 4896 int error; 4897 4898 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); 4899 dsattr.dsa_filerev = nap->na_filerev; 4900 dsattr.dsa_size = nap->na_size; 4901 dsattr.dsa_atime = nap->na_atime; 4902 dsattr.dsa_mtime = nap->na_mtime; 4903 dsattr.dsa_bytes = nap->na_bytes; 4904 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 4905 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); 4906 if (error != 0) 4907 printf("pNFS: setextattr=%d\n", error); 4908 return (error); 4909 } 4910 4911 static int 4912 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 4913 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) 4914 { 4915 uint32_t *tl; 4916 struct nfsrv_descript *nd; 4917 nfsv4stateid_t st; 4918 struct mbuf *m, *m2; 4919 int error = 0, retlen, tlen, trimlen; 4920 4921 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); 4922 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 4923 *mpp = NULL; 4924 /* 4925 * Use a stateid where other is an alternating 01010 pattern and 4926 * seqid is 0xffffffff. This value is not defined as special by 4927 * the RFC and is used by the FreeBSD NFS server to indicate an 4928 * MDS->DS proxy operation. 4929 */ 4930 st.other[0] = 0x55555555; 4931 st.other[1] = 0x55555555; 4932 st.other[2] = 0x55555555; 4933 st.seqid = 0xffffffff; 4934 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), 4935 NULL, NULL, 0, 0); 4936 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 4937 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); 4938 txdr_hyper(off, tl); 4939 *(tl + 2) = txdr_unsigned(len); 4940 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 4941 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 4942 if (error != 0) { 4943 free(nd, M_TEMP); 4944 return (error); 4945 } 4946 if (nd->nd_repstat == 0) { 4947 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 4948 NFSM_STRSIZ(retlen, len); 4949 if (retlen > 0) { 4950 /* Trim off the pre-data XDR from the mbuf chain. */ 4951 m = nd->nd_mrep; 4952 while (m != NULL && m != nd->nd_md) { 4953 if (m->m_next == nd->nd_md) { 4954 m->m_next = NULL; 4955 m_freem(nd->nd_mrep); 4956 nd->nd_mrep = m = nd->nd_md; 4957 } else 4958 m = m->m_next; 4959 } 4960 if (m == NULL) { 4961 printf("nfsrv_readdsrpc: busted mbuf list\n"); 4962 error = ENOENT; 4963 goto nfsmout; 4964 } 4965 4966 /* 4967 * Now, adjust first mbuf so that any XDR before the 4968 * read data is skipped over. 4969 */ 4970 trimlen = nd->nd_dpos - mtod(m, char *); 4971 if (trimlen > 0) { 4972 m->m_len -= trimlen; 4973 NFSM_DATAP(m, trimlen); 4974 } 4975 4976 /* 4977 * Truncate the mbuf chain at retlen bytes of data, 4978 * plus XDR padding that brings the length up to a 4979 * multiple of 4. 4980 */ 4981 tlen = NFSM_RNDUP(retlen); 4982 do { 4983 if (m->m_len >= tlen) { 4984 m->m_len = tlen; 4985 tlen = 0; 4986 m2 = m->m_next; 4987 m->m_next = NULL; 4988 m_freem(m2); 4989 break; 4990 } 4991 tlen -= m->m_len; 4992 m = m->m_next; 4993 } while (m != NULL); 4994 if (tlen > 0) { 4995 printf("nfsrv_readdsrpc: busted mbuf list\n"); 4996 error = ENOENT; 4997 goto nfsmout; 4998 } 4999 *mpp = nd->nd_mrep; 5000 *mpendp = m; 5001 nd->nd_mrep = NULL; 5002 } 5003 } else 5004 error = nd->nd_repstat; 5005 nfsmout: 5006 /* If nd->nd_mrep is already NULL, this is a no-op. */ 5007 m_freem(nd->nd_mrep); 5008 free(nd, M_TEMP); 5009 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); 5010 return (error); 5011 } 5012 5013 /* 5014 * Do a write RPC on a DS data file, using this structure for the arguments, 5015 * so that this function can be executed by a separate kernel process. 5016 */ 5017 struct nfsrvwritedsdorpc { 5018 int done; 5019 int inprog; 5020 struct task tsk; 5021 fhandle_t fh; 5022 off_t off; 5023 int len; 5024 struct nfsmount *nmp; 5025 struct ucred *cred; 5026 NFSPROC_T *p; 5027 struct mbuf *m; 5028 int err; 5029 }; 5030 5031 static int 5032 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, 5033 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) 5034 { 5035 uint32_t *tl; 5036 struct nfsrv_descript *nd; 5037 nfsattrbit_t attrbits; 5038 nfsv4stateid_t st; 5039 int commit, error, retlen; 5040 5041 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5042 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, 5043 sizeof(fhandle_t), NULL, NULL, 0, 0); 5044 5045 /* 5046 * Use a stateid where other is an alternating 01010 pattern and 5047 * seqid is 0xffffffff. This value is not defined as special by 5048 * the RFC and is used by the FreeBSD NFS server to indicate an 5049 * MDS->DS proxy operation. 5050 */ 5051 st.other[0] = 0x55555555; 5052 st.other[1] = 0x55555555; 5053 st.other[2] = 0x55555555; 5054 st.seqid = 0xffffffff; 5055 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5056 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); 5057 txdr_hyper(off, tl); 5058 tl += 2; 5059 /* 5060 * Do all writes FileSync, since the server doesn't hold onto dirty 5061 * buffers. Since clients should be accessing the DS servers directly 5062 * using the pNFS layouts, this just needs to work correctly as a 5063 * fallback. 5064 */ 5065 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); 5066 *tl = txdr_unsigned(len); 5067 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); 5068 5069 /* Put data in mbuf chain. */ 5070 nd->nd_mb->m_next = m; 5071 5072 /* Set nd_mb and nd_bpos to end of data. */ 5073 while (m->m_next != NULL) 5074 m = m->m_next; 5075 nd->nd_mb = m; 5076 nfsm_set(nd, m->m_len); 5077 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); 5078 5079 /* Do a Getattr for the attributes that change upon writing. */ 5080 NFSZERO_ATTRBIT(&attrbits); 5081 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5082 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5083 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5084 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5085 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5086 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5087 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5088 (void) nfsrv_putattrbit(nd, &attrbits); 5089 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5090 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5091 if (error != 0) { 5092 free(nd, M_TEMP); 5093 return (error); 5094 } 5095 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); 5096 /* Get rid of weak cache consistency data for now. */ 5097 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5098 (ND_NFSV4 | ND_V4WCCATTR)) { 5099 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5100 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5101 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); 5102 if (error != 0) 5103 goto nfsmout; 5104 /* 5105 * Get rid of Op# and status for next op. 5106 */ 5107 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5108 if (*++tl != 0) 5109 nd->nd_flag |= ND_NOMOREDATA; 5110 } 5111 if (nd->nd_repstat == 0) { 5112 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); 5113 retlen = fxdr_unsigned(int, *tl++); 5114 commit = fxdr_unsigned(int, *tl); 5115 if (commit != NFSWRITE_FILESYNC) 5116 error = NFSERR_IO; 5117 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", 5118 retlen, commit, error); 5119 } else 5120 error = nd->nd_repstat; 5121 /* We have no use for the Write Verifier since we use FileSync. */ 5122 5123 /* 5124 * Get the Change, Size, Access Time and Modify Time attributes and set 5125 * on the Metadata file, so its attributes will be what the file's 5126 * would be if it had been written. 5127 */ 5128 if (error == 0) { 5129 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5130 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5131 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5132 } 5133 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); 5134 nfsmout: 5135 m_freem(nd->nd_mrep); 5136 free(nd, M_TEMP); 5137 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); 5138 return (error); 5139 } 5140 5141 /* 5142 * Start up the thread that will execute nfsrv_writedsdorpc(). 5143 */ 5144 static void 5145 start_writedsdorpc(void *arg, int pending) 5146 { 5147 struct nfsrvwritedsdorpc *drpc; 5148 5149 drpc = (struct nfsrvwritedsdorpc *)arg; 5150 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5151 drpc->len, NULL, drpc->m, drpc->cred, drpc->p); 5152 drpc->done = 1; 5153 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); 5154 } 5155 5156 static int 5157 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5158 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5159 struct mbuf **mpp, char *cp, int *failposp) 5160 { 5161 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; 5162 struct nfsvattr na; 5163 struct mbuf *m; 5164 int error, i, offs, ret, timo; 5165 5166 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); 5167 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); 5168 drpc = NULL; 5169 if (mirrorcnt > 1) 5170 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5171 M_WAITOK); 5172 5173 /* Calculate offset in mbuf chain that data starts. */ 5174 offs = cp - mtod(*mpp, char *); 5175 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); 5176 5177 /* 5178 * Do the write RPC for every DS, using a separate kernel process 5179 * for every DS except the last one. 5180 */ 5181 error = 0; 5182 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5183 tdrpc->done = 0; 5184 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5185 tdrpc->off = off; 5186 tdrpc->len = len; 5187 tdrpc->nmp = *nmpp; 5188 tdrpc->cred = cred; 5189 tdrpc->p = p; 5190 tdrpc->inprog = 0; 5191 tdrpc->err = 0; 5192 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5193 ret = EIO; 5194 if (nfs_pnfsiothreads != 0) { 5195 ret = nfs_pnfsio(start_writedsdorpc, tdrpc); 5196 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", 5197 ret); 5198 } 5199 if (ret != 0) { 5200 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, 5201 tdrpc->m, cred, p); 5202 if (nfsds_failerr(ret) && *failposp == -1) 5203 *failposp = i; 5204 else if (error == 0 && ret != 0) 5205 error = ret; 5206 } 5207 nmpp++; 5208 fhp++; 5209 } 5210 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5211 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); 5212 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5213 *failposp = mirrorcnt - 1; 5214 else if (error == 0 && ret != 0) 5215 error = ret; 5216 if (error == 0) 5217 error = nfsrv_setextattr(vp, &na, p); 5218 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); 5219 tdrpc = drpc; 5220 timo = hz / 50; /* Wait for 20msec. */ 5221 if (timo < 1) 5222 timo = 1; 5223 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5224 /* Wait for RPCs on separate threads to complete. */ 5225 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5226 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); 5227 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5228 *failposp = i; 5229 else if (error == 0 && tdrpc->err != 0) 5230 error = tdrpc->err; 5231 } 5232 free(drpc, M_TEMP); 5233 return (error); 5234 } 5235 5236 /* 5237 * Do a allocate RPC on a DS data file, using this structure for the arguments, 5238 * so that this function can be executed by a separate kernel process. 5239 */ 5240 struct nfsrvallocatedsdorpc { 5241 int done; 5242 int inprog; 5243 struct task tsk; 5244 fhandle_t fh; 5245 off_t off; 5246 off_t len; 5247 struct nfsmount *nmp; 5248 struct ucred *cred; 5249 NFSPROC_T *p; 5250 int err; 5251 }; 5252 5253 static int 5254 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5255 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5256 { 5257 uint32_t *tl; 5258 struct nfsrv_descript *nd; 5259 nfsattrbit_t attrbits; 5260 nfsv4stateid_t st; 5261 int error; 5262 5263 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5264 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, 5265 sizeof(fhandle_t), NULL, NULL, 0, 0); 5266 5267 /* 5268 * Use a stateid where other is an alternating 01010 pattern and 5269 * seqid is 0xffffffff. This value is not defined as special by 5270 * the RFC and is used by the FreeBSD NFS server to indicate an 5271 * MDS->DS proxy operation. 5272 */ 5273 st.other[0] = 0x55555555; 5274 st.other[1] = 0x55555555; 5275 st.other[2] = 0x55555555; 5276 st.seqid = 0xffffffff; 5277 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5278 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5279 txdr_hyper(off, tl); tl += 2; 5280 txdr_hyper(len, tl); tl += 2; 5281 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); 5282 5283 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5284 NFSGETATTR_ATTRBIT(&attrbits); 5285 nfsrv_putattrbit(nd, &attrbits); 5286 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5287 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5288 if (error != 0) { 5289 free(nd, M_TEMP); 5290 return (error); 5291 } 5292 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", 5293 nd->nd_repstat); 5294 if (nd->nd_repstat == 0) { 5295 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5296 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5297 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5298 } else 5299 error = nd->nd_repstat; 5300 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); 5301 nfsmout: 5302 m_freem(nd->nd_mrep); 5303 free(nd, M_TEMP); 5304 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); 5305 return (error); 5306 } 5307 5308 /* 5309 * Start up the thread that will execute nfsrv_allocatedsdorpc(). 5310 */ 5311 static void 5312 start_allocatedsdorpc(void *arg, int pending) 5313 { 5314 struct nfsrvallocatedsdorpc *drpc; 5315 5316 drpc = (struct nfsrvallocatedsdorpc *)arg; 5317 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5318 drpc->len, NULL, drpc->cred, drpc->p); 5319 drpc->done = 1; 5320 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); 5321 } 5322 5323 static int 5324 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5325 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5326 int *failposp) 5327 { 5328 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; 5329 struct nfsvattr na; 5330 int error, i, ret, timo; 5331 5332 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); 5333 drpc = NULL; 5334 if (mirrorcnt > 1) 5335 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5336 M_WAITOK); 5337 5338 /* 5339 * Do the allocate RPC for every DS, using a separate kernel process 5340 * for every DS except the last one. 5341 */ 5342 error = 0; 5343 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5344 tdrpc->done = 0; 5345 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5346 tdrpc->off = off; 5347 tdrpc->len = len; 5348 tdrpc->nmp = *nmpp; 5349 tdrpc->cred = cred; 5350 tdrpc->p = p; 5351 tdrpc->inprog = 0; 5352 tdrpc->err = 0; 5353 ret = EIO; 5354 if (nfs_pnfsiothreads != 0) { 5355 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); 5356 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", 5357 ret); 5358 } 5359 if (ret != 0) { 5360 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, 5361 cred, p); 5362 if (nfsds_failerr(ret) && *failposp == -1) 5363 *failposp = i; 5364 else if (error == 0 && ret != 0) 5365 error = ret; 5366 } 5367 nmpp++; 5368 fhp++; 5369 } 5370 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5371 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5372 *failposp = mirrorcnt - 1; 5373 else if (error == 0 && ret != 0) 5374 error = ret; 5375 if (error == 0) 5376 error = nfsrv_setextattr(vp, &na, p); 5377 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); 5378 tdrpc = drpc; 5379 timo = hz / 50; /* Wait for 20msec. */ 5380 if (timo < 1) 5381 timo = 1; 5382 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5383 /* Wait for RPCs on separate threads to complete. */ 5384 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5385 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5386 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5387 *failposp = i; 5388 else if (error == 0 && tdrpc->err != 0) 5389 error = tdrpc->err; 5390 } 5391 free(drpc, M_TEMP); 5392 return (error); 5393 } 5394 5395 static int 5396 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5397 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, 5398 struct nfsvattr *dsnap) 5399 { 5400 uint32_t *tl; 5401 struct nfsrv_descript *nd; 5402 nfsv4stateid_t st; 5403 nfsattrbit_t attrbits; 5404 int error; 5405 5406 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); 5407 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5408 /* 5409 * Use a stateid where other is an alternating 01010 pattern and 5410 * seqid is 0xffffffff. This value is not defined as special by 5411 * the RFC and is used by the FreeBSD NFS server to indicate an 5412 * MDS->DS proxy operation. 5413 */ 5414 st.other[0] = 0x55555555; 5415 st.other[1] = 0x55555555; 5416 st.other[2] = 0x55555555; 5417 st.seqid = 0xffffffff; 5418 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5419 NULL, NULL, 0, 0); 5420 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5421 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); 5422 5423 /* Do a Getattr for the attributes that change due to writing. */ 5424 NFSZERO_ATTRBIT(&attrbits); 5425 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5426 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5427 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5428 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5429 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5430 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5431 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5432 (void) nfsrv_putattrbit(nd, &attrbits); 5433 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5434 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5435 if (error != 0) { 5436 free(nd, M_TEMP); 5437 return (error); 5438 } 5439 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", 5440 nd->nd_repstat); 5441 /* Get rid of weak cache consistency data for now. */ 5442 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5443 (ND_NFSV4 | ND_V4WCCATTR)) { 5444 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5445 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5446 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); 5447 if (error != 0) 5448 goto nfsmout; 5449 /* 5450 * Get rid of Op# and status for next op. 5451 */ 5452 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5453 if (*++tl != 0) 5454 nd->nd_flag |= ND_NOMOREDATA; 5455 } 5456 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 5457 if (error != 0) 5458 goto nfsmout; 5459 if (nd->nd_repstat != 0) 5460 error = nd->nd_repstat; 5461 /* 5462 * Get the Change, Size, Access Time and Modify Time attributes and set 5463 * on the Metadata file, so its attributes will be what the file's 5464 * would be if it had been written. 5465 */ 5466 if (error == 0) { 5467 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5468 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5469 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5470 } 5471 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); 5472 nfsmout: 5473 m_freem(nd->nd_mrep); 5474 free(nd, M_TEMP); 5475 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); 5476 return (error); 5477 } 5478 5479 struct nfsrvsetattrdsdorpc { 5480 int done; 5481 int inprog; 5482 struct task tsk; 5483 fhandle_t fh; 5484 struct nfsmount *nmp; 5485 struct vnode *vp; 5486 struct ucred *cred; 5487 NFSPROC_T *p; 5488 struct nfsvattr na; 5489 struct nfsvattr dsna; 5490 int err; 5491 }; 5492 5493 /* 5494 * Start up the thread that will execute nfsrv_setattrdsdorpc(). 5495 */ 5496 static void 5497 start_setattrdsdorpc(void *arg, int pending) 5498 { 5499 struct nfsrvsetattrdsdorpc *drpc; 5500 5501 drpc = (struct nfsrvsetattrdsdorpc *)arg; 5502 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, 5503 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); 5504 drpc->done = 1; 5505 } 5506 5507 static int 5508 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5509 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5510 struct nfsvattr *nap, int *failposp) 5511 { 5512 struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; 5513 struct nfsvattr na; 5514 int error, i, ret, timo; 5515 5516 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); 5517 drpc = NULL; 5518 if (mirrorcnt > 1) 5519 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5520 M_WAITOK); 5521 5522 /* 5523 * Do the setattr RPC for every DS, using a separate kernel process 5524 * for every DS except the last one. 5525 */ 5526 error = 0; 5527 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5528 tdrpc->done = 0; 5529 tdrpc->inprog = 0; 5530 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5531 tdrpc->nmp = *nmpp; 5532 tdrpc->vp = vp; 5533 tdrpc->cred = cred; 5534 tdrpc->p = p; 5535 tdrpc->na = *nap; 5536 tdrpc->err = 0; 5537 ret = EIO; 5538 if (nfs_pnfsiothreads != 0) { 5539 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); 5540 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", 5541 ret); 5542 } 5543 if (ret != 0) { 5544 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, 5545 &na); 5546 if (nfsds_failerr(ret) && *failposp == -1) 5547 *failposp = i; 5548 else if (error == 0 && ret != 0) 5549 error = ret; 5550 } 5551 nmpp++; 5552 fhp++; 5553 } 5554 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); 5555 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5556 *failposp = mirrorcnt - 1; 5557 else if (error == 0 && ret != 0) 5558 error = ret; 5559 if (error == 0) 5560 error = nfsrv_setextattr(vp, &na, p); 5561 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); 5562 tdrpc = drpc; 5563 timo = hz / 50; /* Wait for 20msec. */ 5564 if (timo < 1) 5565 timo = 1; 5566 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5567 /* Wait for RPCs on separate threads to complete. */ 5568 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5569 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); 5570 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5571 *failposp = i; 5572 else if (error == 0 && tdrpc->err != 0) 5573 error = tdrpc->err; 5574 } 5575 free(drpc, M_TEMP); 5576 return (error); 5577 } 5578 5579 /* 5580 * Do a Setattr of an NFSv4 ACL on the DS file. 5581 */ 5582 static int 5583 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5584 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) 5585 { 5586 struct nfsrv_descript *nd; 5587 nfsv4stateid_t st; 5588 nfsattrbit_t attrbits; 5589 int error; 5590 5591 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); 5592 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5593 /* 5594 * Use a stateid where other is an alternating 01010 pattern and 5595 * seqid is 0xffffffff. This value is not defined as special by 5596 * the RFC and is used by the FreeBSD NFS server to indicate an 5597 * MDS->DS proxy operation. 5598 */ 5599 st.other[0] = 0x55555555; 5600 st.other[1] = 0x55555555; 5601 st.other[2] = 0x55555555; 5602 st.seqid = 0xffffffff; 5603 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5604 NULL, NULL, 0, 0); 5605 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5606 NFSZERO_ATTRBIT(&attrbits); 5607 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); 5608 /* 5609 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), 5610 * so passing in the metadata "vp" will be ok, since it is of 5611 * the same type (VREG). 5612 */ 5613 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, 5614 NULL, 0, 0, 0, 0, 0, NULL); 5615 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5616 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5617 if (error != 0) { 5618 free(nd, M_TEMP); 5619 return (error); 5620 } 5621 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", 5622 nd->nd_repstat); 5623 error = nd->nd_repstat; 5624 m_freem(nd->nd_mrep); 5625 free(nd, M_TEMP); 5626 return (error); 5627 } 5628 5629 struct nfsrvsetacldsdorpc { 5630 int done; 5631 int inprog; 5632 struct task tsk; 5633 fhandle_t fh; 5634 struct nfsmount *nmp; 5635 struct vnode *vp; 5636 struct ucred *cred; 5637 NFSPROC_T *p; 5638 struct acl *aclp; 5639 int err; 5640 }; 5641 5642 /* 5643 * Start up the thread that will execute nfsrv_setacldsdorpc(). 5644 */ 5645 static void 5646 start_setacldsdorpc(void *arg, int pending) 5647 { 5648 struct nfsrvsetacldsdorpc *drpc; 5649 5650 drpc = (struct nfsrvsetacldsdorpc *)arg; 5651 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, 5652 drpc->vp, drpc->nmp, drpc->aclp); 5653 drpc->done = 1; 5654 } 5655 5656 static int 5657 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5658 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, 5659 int *failposp) 5660 { 5661 struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; 5662 int error, i, ret, timo; 5663 5664 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); 5665 drpc = NULL; 5666 if (mirrorcnt > 1) 5667 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5668 M_WAITOK); 5669 5670 /* 5671 * Do the setattr RPC for every DS, using a separate kernel process 5672 * for every DS except the last one. 5673 */ 5674 error = 0; 5675 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5676 tdrpc->done = 0; 5677 tdrpc->inprog = 0; 5678 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5679 tdrpc->nmp = *nmpp; 5680 tdrpc->vp = vp; 5681 tdrpc->cred = cred; 5682 tdrpc->p = p; 5683 tdrpc->aclp = aclp; 5684 tdrpc->err = 0; 5685 ret = EIO; 5686 if (nfs_pnfsiothreads != 0) { 5687 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); 5688 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", 5689 ret); 5690 } 5691 if (ret != 0) { 5692 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, 5693 aclp); 5694 if (nfsds_failerr(ret) && *failposp == -1) 5695 *failposp = i; 5696 else if (error == 0 && ret != 0) 5697 error = ret; 5698 } 5699 nmpp++; 5700 fhp++; 5701 } 5702 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); 5703 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5704 *failposp = mirrorcnt - 1; 5705 else if (error == 0 && ret != 0) 5706 error = ret; 5707 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); 5708 tdrpc = drpc; 5709 timo = hz / 50; /* Wait for 20msec. */ 5710 if (timo < 1) 5711 timo = 1; 5712 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5713 /* Wait for RPCs on separate threads to complete. */ 5714 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5715 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); 5716 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5717 *failposp = i; 5718 else if (error == 0 && tdrpc->err != 0) 5719 error = tdrpc->err; 5720 } 5721 free(drpc, M_TEMP); 5722 return (error); 5723 } 5724 5725 /* 5726 * Getattr call to the DS for the attributes that change due to writing. 5727 */ 5728 static int 5729 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5730 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) 5731 { 5732 struct nfsrv_descript *nd; 5733 int error; 5734 nfsattrbit_t attrbits; 5735 5736 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); 5737 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5738 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, 5739 sizeof(fhandle_t), NULL, NULL, 0, 0); 5740 NFSZERO_ATTRBIT(&attrbits); 5741 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5742 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5743 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5744 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5745 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5746 (void) nfsrv_putattrbit(nd, &attrbits); 5747 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5748 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5749 if (error != 0) { 5750 free(nd, M_TEMP); 5751 return (error); 5752 } 5753 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", 5754 nd->nd_repstat); 5755 if (nd->nd_repstat == 0) { 5756 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, 5757 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, 5758 NULL, NULL); 5759 /* 5760 * We can only save the updated values in the extended 5761 * attribute if the vp is exclusively locked. 5762 * This should happen when any of the following operations 5763 * occur on the vnode: 5764 * Close, Delegreturn, LayoutCommit, LayoutReturn 5765 * As such, the updated extended attribute should get saved 5766 * before nfsrv_checkdsattr() returns 0 and allows the cached 5767 * attributes to be returned without calling this function. 5768 */ 5769 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { 5770 error = nfsrv_setextattr(vp, nap, p); 5771 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", 5772 error); 5773 } 5774 } else 5775 error = nd->nd_repstat; 5776 m_freem(nd->nd_mrep); 5777 free(nd, M_TEMP); 5778 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); 5779 return (error); 5780 } 5781 5782 /* 5783 * Seek call to a DS. 5784 */ 5785 static int 5786 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, 5787 struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) 5788 { 5789 uint32_t *tl; 5790 struct nfsrv_descript *nd; 5791 nfsv4stateid_t st; 5792 int error; 5793 5794 NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); 5795 /* 5796 * Use a stateid where other is an alternating 01010 pattern and 5797 * seqid is 0xffffffff. This value is not defined as special by 5798 * the RFC and is used by the FreeBSD NFS server to indicate an 5799 * MDS->DS proxy operation. 5800 */ 5801 st.other[0] = 0x55555555; 5802 st.other[1] = 0x55555555; 5803 st.other[2] = 0x55555555; 5804 st.seqid = 0xffffffff; 5805 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5806 nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, 5807 sizeof(fhandle_t), NULL, NULL, 0, 0); 5808 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5809 NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); 5810 txdr_hyper(*offp, tl); tl += 2; 5811 *tl = txdr_unsigned(content); 5812 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5813 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5814 if (error != 0) { 5815 free(nd, M_TEMP); 5816 return (error); 5817 } 5818 NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); 5819 if (nd->nd_repstat == 0) { 5820 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); 5821 if (*tl++ == newnfs_true) 5822 *eofp = true; 5823 else 5824 *eofp = false; 5825 *offp = fxdr_hyper(tl); 5826 } else 5827 error = nd->nd_repstat; 5828 nfsmout: 5829 m_freem(nd->nd_mrep); 5830 free(nd, M_TEMP); 5831 NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); 5832 return (error); 5833 } 5834 5835 /* 5836 * Get the device id and file handle for a DS file. 5837 */ 5838 int 5839 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, 5840 fhandle_t *fhp, char *devid) 5841 { 5842 int buflen, error; 5843 char *buf; 5844 5845 buflen = 1024; 5846 buf = malloc(buflen, M_TEMP, M_WAITOK); 5847 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, 5848 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); 5849 free(buf, M_TEMP); 5850 return (error); 5851 } 5852 5853 /* 5854 * Do a Lookup against the DS for the filename. 5855 */ 5856 static int 5857 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, 5858 struct vnode **nvpp, NFSPROC_T *p) 5859 { 5860 struct nameidata named; 5861 struct ucred *tcred; 5862 char *bufp; 5863 u_long *hashp; 5864 struct vnode *nvp; 5865 int error; 5866 5867 tcred = newnfs_getcred(); 5868 named.ni_cnd.cn_nameiop = LOOKUP; 5869 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; 5870 named.ni_cnd.cn_cred = tcred; 5871 named.ni_cnd.cn_thread = p; 5872 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; 5873 nfsvno_setpathbuf(&named, &bufp, &hashp); 5874 named.ni_cnd.cn_nameptr = bufp; 5875 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); 5876 strlcpy(bufp, pf->dsf_filename, NAME_MAX); 5877 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); 5878 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 5879 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); 5880 NFSFREECRED(tcred); 5881 nfsvno_relpathbuf(&named); 5882 if (error == 0) 5883 *nvpp = nvp; 5884 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); 5885 return (error); 5886 } 5887 5888 /* 5889 * Set the file handle to the correct one. 5890 */ 5891 static void 5892 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, 5893 char *fnamep, struct vnode *nvp, NFSPROC_T *p) 5894 { 5895 struct nfsnode *np; 5896 int ret = 0; 5897 5898 np = VTONFS(nvp); 5899 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); 5900 /* 5901 * We can only do a vn_set_extattr() if the vnode is exclusively 5902 * locked and vn_start_write() has been done. If devid != NULL or 5903 * fnamep != NULL or the vnode is shared locked, vn_start_write() 5904 * may not have been done. 5905 * If not done now, it will be done on a future call. 5906 */ 5907 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == 5908 LK_EXCLUSIVE) 5909 ret = vn_extattr_set(vp, IO_NODELOCKED, 5910 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), 5911 (char *)pf, p); 5912 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); 5913 } 5914 5915 /* 5916 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point 5917 * when the DS has failed. 5918 */ 5919 void 5920 nfsrv_killrpcs(struct nfsmount *nmp) 5921 { 5922 5923 /* 5924 * Call newnfs_nmcancelreqs() to cause 5925 * any RPCs in progress on the mount point to 5926 * fail. 5927 * This will cause any process waiting for an 5928 * RPC to complete while holding a vnode lock 5929 * on the mounted-on vnode (such as "df" or 5930 * a non-forced "umount") to fail. 5931 * This will unlock the mounted-on vnode so 5932 * a forced dismount can succeed. 5933 * The NFSMNTP_CANCELRPCS flag should be set when this function is 5934 * called. 5935 */ 5936 newnfs_nmcancelreqs(nmp); 5937 } 5938 5939 /* 5940 * Sum up the statfs info for each of the DSs, so that the client will 5941 * receive the total for all DSs. 5942 */ 5943 static int 5944 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) 5945 { 5946 struct statfs *tsf; 5947 struct nfsdevice *ds; 5948 struct vnode **dvpp, **tdvpp, *dvp; 5949 uint64_t tot; 5950 int cnt, error = 0, i; 5951 5952 if (nfsrv_devidcnt <= 0) 5953 return (ENXIO); 5954 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 5955 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 5956 5957 /* Get an array of the dvps for the DSs. */ 5958 tdvpp = dvpp; 5959 i = 0; 5960 NFSDDSLOCK(); 5961 /* First, search for matches for same file system. */ 5962 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5963 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && 5964 fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { 5965 if (++i > nfsrv_devidcnt) 5966 break; 5967 *tdvpp++ = ds->nfsdev_dvp; 5968 } 5969 } 5970 /* 5971 * If no matches for same file system, total all servers not assigned 5972 * to a file system. 5973 */ 5974 if (i == 0) { 5975 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5976 if (ds->nfsdev_nmp != NULL && 5977 ds->nfsdev_mdsisset == 0) { 5978 if (++i > nfsrv_devidcnt) 5979 break; 5980 *tdvpp++ = ds->nfsdev_dvp; 5981 } 5982 } 5983 } 5984 NFSDDSUNLOCK(); 5985 cnt = i; 5986 5987 /* Do a VFS_STATFS() for each of the DSs and sum them up. */ 5988 tdvpp = dvpp; 5989 for (i = 0; i < cnt && error == 0; i++) { 5990 dvp = *tdvpp++; 5991 error = VFS_STATFS(dvp->v_mount, tsf); 5992 if (error == 0) { 5993 if (sf->f_bsize == 0) { 5994 if (tsf->f_bsize > 0) 5995 sf->f_bsize = tsf->f_bsize; 5996 else 5997 sf->f_bsize = 8192; 5998 } 5999 if (tsf->f_blocks > 0) { 6000 if (sf->f_bsize != tsf->f_bsize) { 6001 tot = tsf->f_blocks * tsf->f_bsize; 6002 sf->f_blocks += (tot / sf->f_bsize); 6003 } else 6004 sf->f_blocks += tsf->f_blocks; 6005 } 6006 if (tsf->f_bfree > 0) { 6007 if (sf->f_bsize != tsf->f_bsize) { 6008 tot = tsf->f_bfree * tsf->f_bsize; 6009 sf->f_bfree += (tot / sf->f_bsize); 6010 } else 6011 sf->f_bfree += tsf->f_bfree; 6012 } 6013 if (tsf->f_bavail > 0) { 6014 if (sf->f_bsize != tsf->f_bsize) { 6015 tot = tsf->f_bavail * tsf->f_bsize; 6016 sf->f_bavail += (tot / sf->f_bsize); 6017 } else 6018 sf->f_bavail += tsf->f_bavail; 6019 } 6020 } 6021 } 6022 free(tsf, M_TEMP); 6023 free(dvpp, M_TEMP); 6024 return (error); 6025 } 6026 6027 /* 6028 * Set an NFSv4 acl. 6029 */ 6030 int 6031 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) 6032 { 6033 int error; 6034 6035 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { 6036 error = NFSERR_ATTRNOTSUPP; 6037 goto out; 6038 } 6039 /* 6040 * With NFSv4 ACLs, chmod(2) may need to add additional entries. 6041 * Make sure it has enough room for that - splitting every entry 6042 * into two and appending "canonical six" entries at the end. 6043 * Cribbed out of kern/vfs_acl.c - Rick M. 6044 */ 6045 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { 6046 error = NFSERR_ATTRNOTSUPP; 6047 goto out; 6048 } 6049 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 6050 if (error == 0) { 6051 error = nfsrv_dssetacl(vp, aclp, cred, p); 6052 if (error == ENOENT) 6053 error = 0; 6054 } 6055 6056 out: 6057 NFSEXITCODE(error); 6058 return (error); 6059 } 6060 6061 /* 6062 * Seek vnode op call (actually it is a VOP_IOCTL()). 6063 * This function is called with the vnode locked, but unlocks and vrele()s 6064 * the vp before returning. 6065 */ 6066 int 6067 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, 6068 off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) 6069 { 6070 struct nfsvattr at; 6071 int error, ret; 6072 6073 ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); 6074 /* 6075 * Attempt to seek on a DS file. A return of ENOENT implies 6076 * there is no DS file to seek on. 6077 */ 6078 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, 6079 NULL, NULL, NULL, NULL, offp, content, eofp); 6080 if (error != ENOENT) { 6081 vput(vp); 6082 return (error); 6083 } 6084 6085 /* 6086 * Do the VOP_IOCTL() call. For the case where *offp == file_size, 6087 * VOP_IOCTL() will return ENXIO. However, the correct reply for 6088 * NFSv4.2 is *eofp == true and error == 0 for this case. 6089 */ 6090 NFSVOPUNLOCK(vp); 6091 error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); 6092 *eofp = false; 6093 if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { 6094 /* Handle the cases where we might be at EOF. */ 6095 ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); 6096 if (ret == 0 && *offp == at.na_size) { 6097 *eofp = true; 6098 error = 0; 6099 } 6100 if (ret != 0 && error == 0) 6101 error = ret; 6102 } 6103 vrele(vp); 6104 NFSEXITCODE(error); 6105 return (error); 6106 } 6107 6108 /* 6109 * Allocate vnode op call. 6110 */ 6111 int 6112 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6113 NFSPROC_T *p) 6114 { 6115 int error, trycnt; 6116 6117 ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); 6118 /* 6119 * Attempt to allocate on a DS file. A return of ENOENT implies 6120 * there is no DS file to allocate on. 6121 */ 6122 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, 6123 NULL, NULL, NULL, NULL, &len, 0, NULL); 6124 if (error != ENOENT) 6125 return (error); 6126 error = 0; 6127 6128 /* 6129 * Do the actual VOP_ALLOCATE(), looping a reasonable number of 6130 * times to achieve completion. 6131 */ 6132 trycnt = 0; 6133 while (error == 0 && len > 0 && trycnt++ < 20) 6134 error = VOP_ALLOCATE(vp, &off, &len); 6135 if (error == 0 && len > 0) 6136 error = NFSERR_IO; 6137 NFSEXITCODE(error); 6138 return (error); 6139 } 6140 6141 /* 6142 * Get Extended Atribute vnode op into an mbuf list. 6143 */ 6144 int 6145 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, 6146 struct ucred *cred, struct thread *p, struct mbuf **mpp, 6147 struct mbuf **mpendp, int *lenp) 6148 { 6149 struct iovec *iv; 6150 struct uio io, *uiop = &io; 6151 struct mbuf *m, *m2; 6152 int alen, error, len, tlen; 6153 size_t siz; 6154 6155 /* First, find out the size of the extended attribute. */ 6156 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6157 &siz, cred, p); 6158 if (error != 0) 6159 return (NFSERR_NOXATTR); 6160 if (siz > maxresp - NFS_MAXXDR) 6161 return (NFSERR_XATTR2BIG); 6162 len = siz; 6163 tlen = NFSM_RNDUP(len); 6164 if (tlen > 0) { 6165 uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, &iv); 6166 uiop->uio_iov = iv; 6167 } else { 6168 uiop->uio_iovcnt = 0; 6169 uiop->uio_iov = iv = NULL; 6170 m = m2 = NULL; 6171 } 6172 uiop->uio_offset = 0; 6173 uiop->uio_resid = tlen; 6174 uiop->uio_rw = UIO_READ; 6175 uiop->uio_segflg = UIO_SYSSPACE; 6176 uiop->uio_td = p; 6177 #ifdef MAC 6178 error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6179 name); 6180 if (error != 0) 6181 goto out; 6182 #endif 6183 6184 if (tlen > 0) 6185 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 6186 NULL, cred, p); 6187 if (error != 0) 6188 goto out; 6189 if (uiop->uio_resid > 0) { 6190 alen = tlen; 6191 len = tlen - uiop->uio_resid; 6192 tlen = NFSM_RNDUP(len); 6193 if (alen != tlen) 6194 printf("nfsvno_getxattr: weird size read\n"); 6195 nfsrv_adj(m, alen - tlen, tlen - len); 6196 } 6197 *lenp = len; 6198 *mpp = m; 6199 *mpendp = m2; 6200 6201 out: 6202 if (error != 0) { 6203 if (m != NULL) 6204 m_freem(m); 6205 *lenp = 0; 6206 } 6207 free(iv, M_TEMP); 6208 NFSEXITCODE(error); 6209 return (error); 6210 } 6211 6212 /* 6213 * Set Extended attribute vnode op from an mbuf list. 6214 */ 6215 int 6216 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, 6217 char *cp, struct ucred *cred, struct thread *p) 6218 { 6219 struct iovec *iv; 6220 struct uio uio, *uiop = &uio; 6221 int cnt, error; 6222 6223 error = 0; 6224 #ifdef MAC 6225 error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6226 name); 6227 #endif 6228 if (error != 0) 6229 goto out; 6230 6231 uiop->uio_rw = UIO_WRITE; 6232 uiop->uio_segflg = UIO_SYSSPACE; 6233 uiop->uio_td = p; 6234 uiop->uio_offset = 0; 6235 uiop->uio_resid = len; 6236 if (len > 0) { 6237 error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); 6238 uiop->uio_iov = iv; 6239 uiop->uio_iovcnt = cnt; 6240 } else { 6241 uiop->uio_iov = iv = NULL; 6242 uiop->uio_iovcnt = 0; 6243 } 6244 if (error == 0) { 6245 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 6246 cred, p); 6247 free(iv, M_TEMP); 6248 } 6249 6250 out: 6251 NFSEXITCODE(error); 6252 return (error); 6253 } 6254 6255 /* 6256 * Remove Extended attribute vnode op. 6257 */ 6258 int 6259 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, 6260 struct ucred *cred, struct thread *p) 6261 { 6262 int error; 6263 6264 /* 6265 * Get rid of any delegations. I am not sure why this is required, 6266 * but RFC-8276 says so. 6267 */ 6268 error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); 6269 if (error != 0) 6270 goto out; 6271 #ifdef MAC 6272 error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6273 name); 6274 if (error != 0) 6275 goto out; 6276 #endif 6277 6278 error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); 6279 if (error == EOPNOTSUPP) 6280 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6281 cred, p); 6282 out: 6283 NFSEXITCODE(error); 6284 return (error); 6285 } 6286 6287 /* 6288 * List Extended Atribute vnode op into an mbuf list. 6289 */ 6290 int 6291 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, 6292 struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) 6293 { 6294 struct iovec iv; 6295 struct uio io; 6296 int error; 6297 size_t siz; 6298 6299 *bufp = NULL; 6300 /* First, find out the size of the extended attribute. */ 6301 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, 6302 p); 6303 if (error != 0) 6304 return (NFSERR_NOXATTR); 6305 if (siz <= cookie) { 6306 *lenp = 0; 6307 *eofp = true; 6308 goto out; 6309 } 6310 if (siz > cookie + *lenp) { 6311 siz = cookie + *lenp; 6312 *eofp = false; 6313 } else 6314 *eofp = true; 6315 /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ 6316 if (siz > 10 * 1024 * 1024) { 6317 error = NFSERR_XATTR2BIG; 6318 goto out; 6319 } 6320 *bufp = malloc(siz, M_TEMP, M_WAITOK); 6321 iv.iov_base = *bufp; 6322 iv.iov_len = siz; 6323 io.uio_iovcnt = 1; 6324 io.uio_iov = &iv; 6325 io.uio_offset = 0; 6326 io.uio_resid = siz; 6327 io.uio_rw = UIO_READ; 6328 io.uio_segflg = UIO_SYSSPACE; 6329 io.uio_td = p; 6330 #ifdef MAC 6331 error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); 6332 if (error != 0) 6333 goto out; 6334 #endif 6335 6336 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, 6337 p); 6338 if (error != 0) 6339 goto out; 6340 if (io.uio_resid > 0) 6341 siz -= io.uio_resid; 6342 *lenp = siz; 6343 6344 out: 6345 if (error != 0) { 6346 free(*bufp, M_TEMP); 6347 *bufp = NULL; 6348 } 6349 NFSEXITCODE(error); 6350 return (error); 6351 } 6352 6353 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); 6354 6355 /* 6356 * Called once to initialize data structures... 6357 */ 6358 static int 6359 nfsd_modevent(module_t mod, int type, void *data) 6360 { 6361 int error = 0, i; 6362 static int loaded = 0; 6363 6364 switch (type) { 6365 case MOD_LOAD: 6366 if (loaded) 6367 goto out; 6368 newnfs_portinit(); 6369 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 6370 mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL, 6371 MTX_DEF); 6372 mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL, 6373 MTX_DEF); 6374 } 6375 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); 6376 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); 6377 mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF); 6378 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); 6379 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); 6380 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0); 6381 nfsrvd_initcache(); 6382 nfsd_init(); 6383 NFSD_LOCK(); 6384 nfsrvd_init(0); 6385 NFSD_UNLOCK(); 6386 nfsd_mntinit(); 6387 #ifdef VV_DISABLEDELEG 6388 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; 6389 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; 6390 #endif 6391 nfsd_call_servertimer = nfsrv_servertimer; 6392 nfsd_call_nfsd = nfssvc_nfsd; 6393 loaded = 1; 6394 break; 6395 6396 case MOD_UNLOAD: 6397 if (newnfs_numnfsd != 0) { 6398 error = EBUSY; 6399 break; 6400 } 6401 6402 #ifdef VV_DISABLEDELEG 6403 vn_deleg_ops.vndeleg_recall = NULL; 6404 vn_deleg_ops.vndeleg_disable = NULL; 6405 #endif 6406 nfsd_call_servertimer = NULL; 6407 nfsd_call_nfsd = NULL; 6408 6409 /* Clean out all NFSv4 state. */ 6410 nfsrv_throwawayallstate(curthread); 6411 6412 /* Clean the NFS server reply cache */ 6413 nfsrvd_cleancache(); 6414 6415 /* Free up the krpc server pool. */ 6416 if (nfsrvd_pool != NULL) 6417 svcpool_destroy(nfsrvd_pool); 6418 6419 /* and get rid of the locks */ 6420 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 6421 mtx_destroy(&nfsrchash_table[i].mtx); 6422 mtx_destroy(&nfsrcahash_table[i].mtx); 6423 } 6424 mtx_destroy(&nfsrc_udpmtx); 6425 mtx_destroy(&nfs_v4root_mutex); 6426 mtx_destroy(&nfsv4root_mnt.mnt_mtx); 6427 mtx_destroy(&nfsrv_dontlistlock_mtx); 6428 mtx_destroy(&nfsrv_recalllock_mtx); 6429 for (i = 0; i < nfsrv_sessionhashsize; i++) 6430 mtx_destroy(&nfssessionhash[i].mtx); 6431 if (nfslayouthash != NULL) { 6432 for (i = 0; i < nfsrv_layouthashsize; i++) 6433 mtx_destroy(&nfslayouthash[i].mtx); 6434 free(nfslayouthash, M_NFSDSESSION); 6435 } 6436 lockdestroy(&nfsv4root_mnt.mnt_explock); 6437 free(nfsclienthash, M_NFSDCLIENT); 6438 free(nfslockhash, M_NFSDLOCKFILE); 6439 free(nfssessionhash, M_NFSDSESSION); 6440 loaded = 0; 6441 break; 6442 default: 6443 error = EOPNOTSUPP; 6444 break; 6445 } 6446 6447 out: 6448 NFSEXITCODE(error); 6449 return (error); 6450 } 6451 static moduledata_t nfsd_mod = { 6452 "nfsd", 6453 nfsd_modevent, 6454 NULL, 6455 }; 6456 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); 6457 6458 /* So that loader and kldload(2) can find us, wherever we are.. */ 6459 MODULE_VERSION(nfsd, 1); 6460 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); 6461 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); 6462 MODULE_DEPEND(nfsd, krpc, 1, 1, 1); 6463 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); 6464 6465