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/callout.h> 51 #include <sys/filio.h> 52 #include <sys/hash.h> 53 #include <sys/osd.h> 54 #include <sys/sysctl.h> 55 #include <nlm/nlm_prot.h> 56 #include <nlm/nlm.h> 57 58 FEATURE(nfsd, "NFSv4 server"); 59 60 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 61 extern int nfsrv_useacl; 62 extern int newnfs_numnfsd; 63 extern int nfsrv_sessionhashsize; 64 extern struct nfslayouthash *nfslayouthash; 65 extern int nfsrv_layouthashsize; 66 extern struct mtx nfsrv_dslock_mtx; 67 extern int nfs_pnfsiothreads; 68 extern volatile int nfsrv_devidcnt; 69 extern int nfsrv_maxpnfsmirror; 70 extern uint32_t nfs_srvmaxio; 71 extern int nfs_bufpackets; 72 extern u_long sb_max_adj; 73 74 NFSD_VNET_DECLARE(int, nfsrv_numnfsd); 75 NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst); 76 NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool); 77 NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash); 78 NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash); 79 NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash); 80 NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock); 81 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p); 82 83 NFSDLOCKMUTEX; 84 NFSSTATESPINLOCK; 85 struct mtx nfsrc_udpmtx; 86 struct mtx nfs_v4root_mutex; 87 struct mtx nfsrv_dontlistlock_mtx; 88 struct mtx nfsrv_recalllock_mtx; 89 struct nfsrvfh nfs_pubfh; 90 int nfs_pubfhset = 0; 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 NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL; 99 NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl); 100 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table); 101 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table); 102 NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh); 103 NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0; 104 NFSD_VNET_DEFINE(struct callout, nfsd_callout); 105 NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt); 106 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt); 107 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt); 108 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false; 109 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false; 110 111 static void nfsrv_cleanup(struct prison *); 112 113 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, 114 struct ucred *); 115 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *); 116 117 int nfsrv_enable_crossmntpt = 1; 118 static int nfs_commit_blks; 119 static int nfs_commit_miss; 120 extern int nfsrv_issuedelegs; 121 extern int nfsrv_dolocallocks; 122 extern struct nfsdevicehead nfsrv_devidhead; 123 124 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, 125 struct iovec **); 126 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **, 127 struct mbuf **, struct iovec **); 128 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, 129 int *); 130 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, 131 NFSPROC_T *); 132 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, 133 int *, char *, fhandle_t *); 134 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, 135 NFSPROC_T *); 136 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, 137 struct thread *, int, struct mbuf **, char *, struct mbuf **, 138 struct nfsvattr *, struct acl *, off_t *, int, bool *); 139 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); 140 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, 141 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); 142 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, 143 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, 144 char *, int *); 145 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 146 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 147 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 148 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 149 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 150 struct vnode *, struct nfsmount **, int, struct acl *, int *); 151 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 152 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); 153 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 154 struct vnode *, struct nfsmount *, struct nfsvattr *); 155 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, 156 NFSPROC_T *, struct nfsmount *); 157 static int nfsrv_putfhname(fhandle_t *, char *); 158 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, 159 struct pnfsdsfile *, struct vnode **, NFSPROC_T *); 160 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, 161 struct vnode *, NFSPROC_T *); 162 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); 163 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, 164 NFSPROC_T *); 165 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); 166 167 int nfs_pnfsio(task_fn_t *, void *); 168 169 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 170 "NFS server"); 171 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, 172 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); 173 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 174 0, ""); 175 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 176 0, ""); 177 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, 178 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); 179 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, 180 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); 181 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 182 0, "Debug level for NFS server"); 183 NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid); 184 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, 185 CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid), 186 0, "Enable nfsd to accept numeric owner_names"); 187 static int nfsrv_pnfsgetdsattr = 1; 188 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, 189 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); 190 191 /* 192 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are 193 * not running. 194 * The dsN subdirectories for the increased values must have been created 195 * on all DS servers before this increase is done. 196 */ 197 u_int nfsrv_dsdirsize = 20; 198 static int 199 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) 200 { 201 int error, newdsdirsize; 202 203 newdsdirsize = nfsrv_dsdirsize; 204 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); 205 if (error != 0 || req->newptr == NULL) 206 return (error); 207 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || 208 newnfs_numnfsd != 0) 209 return (EINVAL); 210 nfsrv_dsdirsize = newdsdirsize; 211 return (0); 212 } 213 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, 214 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), 215 sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); 216 217 /* 218 * nfs_srvmaxio can only be increased and only when the nfsd threads are 219 * not running. The setting must be a power of 2, with the current limit of 220 * 1Mbyte. 221 */ 222 static int 223 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS) 224 { 225 int error; 226 u_int newsrvmaxio; 227 uint64_t tval; 228 229 newsrvmaxio = nfs_srvmaxio; 230 error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req); 231 if (error != 0 || req->newptr == NULL) 232 return (error); 233 if (newsrvmaxio == nfs_srvmaxio) 234 return (0); 235 if (newsrvmaxio < nfs_srvmaxio) { 236 printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n"); 237 return (EINVAL); 238 } 239 if (newsrvmaxio > 1048576) { 240 printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n"); 241 return (EINVAL); 242 } 243 if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) { 244 printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n"); 245 return (EINVAL); 246 } 247 248 /* 249 * Check that kern.ipc.maxsockbuf is large enough for 250 * newsrviomax, given the setting of vfs.nfs.bufpackets. 251 */ 252 if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets > 253 sb_max_adj) { 254 /* 255 * Suggest vfs.nfs.bufpackets * maximum RPC message for 256 * sb_max_adj. 257 */ 258 tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets; 259 260 /* 261 * Convert suggested sb_max_adj value to a suggested 262 * sb_max value, which is what is set via kern.ipc.maxsockbuf. 263 * Perform the inverse calculation of (from uipc_sockbuf.c): 264 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES / 265 * (MSIZE + MCLBYTES); 266 * XXX If the calculation of sb_max_adj from sb_max changes, 267 * this calculation must be changed as well. 268 */ 269 tval *= (MSIZE + MCLBYTES); /* Brackets for readability. */ 270 tval += MCLBYTES - 1; /* Round up divide. */ 271 tval /= MCLBYTES; 272 printf("nfsd: set kern.ipc.maxsockbuf to a minimum of " 273 "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval, 274 newsrvmaxio); 275 return (EINVAL); 276 } 277 278 NFSD_LOCK(); 279 if (newnfs_numnfsd != 0) { 280 NFSD_UNLOCK(); 281 printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd " 282 "threads are running\n"); 283 return (EINVAL); 284 } 285 286 287 nfs_srvmaxio = newsrvmaxio; 288 NFSD_UNLOCK(); 289 return (0); 290 } 291 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio, 292 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 293 sysctl_srvmaxio, "IU", "Maximum I/O size in bytes"); 294 295 #define MAX_REORDERED_RPC 16 296 #define NUM_HEURISTIC 1031 297 #define NHUSE_INIT 64 298 #define NHUSE_INC 16 299 #define NHUSE_MAX 2048 300 301 static struct nfsheur { 302 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 303 off_t nh_nextoff; /* next offset for sequential detection */ 304 int nh_use; /* use count for selection */ 305 int nh_seqcount; /* heuristic */ 306 } nfsheur[NUM_HEURISTIC]; 307 308 /* 309 * Heuristic to detect sequential operation. 310 */ 311 static struct nfsheur * 312 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 313 { 314 struct nfsheur *nh; 315 int hi, try; 316 317 /* Locate best candidate. */ 318 try = 32; 319 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 320 nh = &nfsheur[hi]; 321 while (try--) { 322 if (nfsheur[hi].nh_vp == vp) { 323 nh = &nfsheur[hi]; 324 break; 325 } 326 if (nfsheur[hi].nh_use > 0) 327 --nfsheur[hi].nh_use; 328 hi = (hi + 1) % NUM_HEURISTIC; 329 if (nfsheur[hi].nh_use < nh->nh_use) 330 nh = &nfsheur[hi]; 331 } 332 333 /* Initialize hint if this is a new file. */ 334 if (nh->nh_vp != vp) { 335 nh->nh_vp = vp; 336 nh->nh_nextoff = uio->uio_offset; 337 nh->nh_use = NHUSE_INIT; 338 if (uio->uio_offset == 0) 339 nh->nh_seqcount = 4; 340 else 341 nh->nh_seqcount = 1; 342 } 343 344 /* Calculate heuristic. */ 345 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 346 uio->uio_offset == nh->nh_nextoff) { 347 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 348 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 349 if (nh->nh_seqcount > IO_SEQMAX) 350 nh->nh_seqcount = IO_SEQMAX; 351 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 352 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 353 /* Probably a reordered RPC, leave seqcount alone. */ 354 } else if (nh->nh_seqcount > 1) { 355 nh->nh_seqcount /= 2; 356 } else { 357 nh->nh_seqcount = 0; 358 } 359 nh->nh_use += NHUSE_INC; 360 if (nh->nh_use > NHUSE_MAX) 361 nh->nh_use = NHUSE_MAX; 362 return (nh); 363 } 364 365 /* 366 * Get attributes into nfsvattr structure. 367 */ 368 int 369 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, 370 struct nfsrv_descript *nd, struct thread *p, int vpislocked, 371 nfsattrbit_t *attrbitp) 372 { 373 int error, gotattr, lockedit = 0; 374 struct nfsvattr na; 375 376 if (vpislocked == 0) { 377 /* 378 * When vpislocked == 0, the vnode is either exclusively 379 * locked by this thread or not locked by this thread. 380 * As such, shared lock it, if not exclusively locked. 381 */ 382 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 383 lockedit = 1; 384 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); 385 } 386 } 387 388 /* 389 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed 390 * attributes, as required. 391 * This needs to be done for regular files if: 392 * - non-NFSv4 RPCs or 393 * - when attrbitp == NULL or 394 * - an NFSv4 RPC with any of the above attributes in attrbitp. 395 * A return of 0 for nfsrv_proxyds() indicates that it has acquired 396 * these attributes. nfsrv_proxyds() will return an error if the 397 * server is not a pNFS one. 398 */ 399 gotattr = 0; 400 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || 401 (nd->nd_flag & ND_NFSV4) == 0 || 402 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || 403 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || 404 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || 405 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || 406 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { 407 error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, 408 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, 409 NULL); 410 if (error == 0) 411 gotattr = 1; 412 } 413 414 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); 415 if (lockedit != 0) 416 NFSVOPUNLOCK(vp); 417 418 /* 419 * If we got the Change, Size and Modify Time from the DS, 420 * replace them. 421 */ 422 if (gotattr != 0) { 423 nvap->na_atime = na.na_atime; 424 nvap->na_mtime = na.na_mtime; 425 nvap->na_filerev = na.na_filerev; 426 nvap->na_size = na.na_size; 427 nvap->na_bytes = na.na_bytes; 428 } 429 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, 430 error, (uintmax_t)na.na_filerev); 431 432 NFSEXITCODE(error); 433 return (error); 434 } 435 436 /* 437 * Get a file handle for a vnode. 438 */ 439 int 440 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) 441 { 442 int error; 443 444 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); 445 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 446 error = VOP_VPTOFH(vp, &fhp->fh_fid); 447 448 NFSEXITCODE(error); 449 return (error); 450 } 451 452 /* 453 * Perform access checking for vnodes obtained from file handles that would 454 * refer to files already opened by a Unix client. You cannot just use 455 * vn_writechk() and VOP_ACCESSX() for two reasons. 456 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 457 * case. 458 * 2 - The owner is to be given access irrespective of mode bits for some 459 * operations, so that processes that chmod after opening a file don't 460 * break. 461 */ 462 int 463 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, 464 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, 465 u_int32_t *supportedtypep) 466 { 467 struct vattr vattr; 468 int error = 0, getret = 0; 469 470 if (vpislocked == 0) { 471 if (NFSVOPLOCK(vp, LK_SHARED) != 0) { 472 error = EPERM; 473 goto out; 474 } 475 } 476 if (accmode & VWRITE) { 477 /* Just vn_writechk() changed to check rdonly */ 478 /* 479 * Disallow write attempts on read-only file systems; 480 * unless the file is a socket or a block or character 481 * device resident on the file system. 482 */ 483 if (NFSVNO_EXRDONLY(exp) || 484 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 485 switch (vp->v_type) { 486 case VREG: 487 case VDIR: 488 case VLNK: 489 error = EROFS; 490 default: 491 break; 492 } 493 } 494 /* 495 * If there's shared text associated with 496 * the inode, try to free it up once. If 497 * we fail, we can't allow writing. 498 */ 499 if (VOP_IS_TEXT(vp) && error == 0) 500 error = ETXTBSY; 501 } 502 if (error != 0) { 503 if (vpislocked == 0) 504 NFSVOPUNLOCK(vp); 505 goto out; 506 } 507 508 /* 509 * Should the override still be applied when ACLs are enabled? 510 */ 511 error = VOP_ACCESSX(vp, accmode, cred, p); 512 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { 513 /* 514 * Try again with VEXPLICIT_DENY, to see if the test for 515 * deletion is supported. 516 */ 517 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); 518 if (error == 0) { 519 if (vp->v_type == VDIR) { 520 accmode &= ~(VDELETE | VDELETE_CHILD); 521 accmode |= VWRITE; 522 error = VOP_ACCESSX(vp, accmode, cred, p); 523 } else if (supportedtypep != NULL) { 524 *supportedtypep &= ~NFSACCESS_DELETE; 525 } 526 } 527 } 528 529 /* 530 * Allow certain operations for the owner (reads and writes 531 * on files that are already open). 532 */ 533 if (override != NFSACCCHK_NOOVERRIDE && 534 (error == EPERM || error == EACCES)) { 535 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) 536 error = 0; 537 else if (override & NFSACCCHK_ALLOWOWNER) { 538 getret = VOP_GETATTR(vp, &vattr, cred); 539 if (getret == 0 && cred->cr_uid == vattr.va_uid) 540 error = 0; 541 } 542 } 543 if (vpislocked == 0) 544 NFSVOPUNLOCK(vp); 545 546 out: 547 NFSEXITCODE(error); 548 return (error); 549 } 550 551 /* 552 * Set attribute(s) vnop. 553 */ 554 int 555 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 556 struct thread *p, struct nfsexstuff *exp) 557 { 558 u_quad_t savsize = 0; 559 int error, savedit; 560 time_t savbtime; 561 562 /* 563 * If this is an exported file system and a pNFS service is running, 564 * don't VOP_SETATTR() of size for the MDS file system. 565 */ 566 savedit = 0; 567 error = 0; 568 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && 569 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && 570 nvap->na_vattr.va_size > 0) { 571 savsize = nvap->na_vattr.va_size; 572 nvap->na_vattr.va_size = VNOVAL; 573 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 574 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 575 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 576 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 577 nvap->na_vattr.va_mtime.tv_sec != VNOVAL) 578 savedit = 1; 579 else 580 savedit = 2; 581 } 582 if (savedit != 2) 583 error = VOP_SETATTR(vp, &nvap->na_vattr, cred); 584 if (savedit != 0) 585 nvap->na_vattr.va_size = savsize; 586 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 587 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 588 nvap->na_vattr.va_size != VNOVAL || 589 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 590 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 591 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { 592 /* Never modify birthtime on a DS file. */ 593 savbtime = nvap->na_vattr.va_birthtime.tv_sec; 594 nvap->na_vattr.va_birthtime.tv_sec = VNOVAL; 595 /* For a pNFS server, set the attributes on the DS file. */ 596 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, 597 NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); 598 nvap->na_vattr.va_birthtime.tv_sec = savbtime; 599 if (error == ENOENT) 600 error = 0; 601 } 602 NFSEXITCODE(error); 603 return (error); 604 } 605 606 /* 607 * Set up nameidata for a lookup() call and do it. 608 */ 609 int 610 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, 611 struct vnode *dp, int islocked, struct nfsexstuff *exp, 612 struct vnode **retdirp) 613 { 614 struct componentname *cnp = &ndp->ni_cnd; 615 int i; 616 struct iovec aiov; 617 struct uio auio; 618 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; 619 int error = 0; 620 char *cp; 621 622 *retdirp = NULL; 623 cnp->cn_nameptr = cnp->cn_pnbuf; 624 ndp->ni_lcf = 0; 625 /* 626 * Extract and set starting directory. 627 */ 628 if (dp->v_type != VDIR) { 629 if (islocked) 630 vput(dp); 631 else 632 vrele(dp); 633 nfsvno_relpathbuf(ndp); 634 error = ENOTDIR; 635 goto out1; 636 } 637 if (islocked) 638 NFSVOPUNLOCK(dp); 639 VREF(dp); 640 *retdirp = dp; 641 if (NFSVNO_EXRDONLY(exp)) 642 cnp->cn_flags |= RDONLY; 643 ndp->ni_segflg = UIO_SYSSPACE; 644 645 if (nd->nd_flag & ND_PUBLOOKUP) { 646 ndp->ni_loopcnt = 0; 647 if (cnp->cn_pnbuf[0] == '/') { 648 vrele(dp); 649 /* 650 * Check for degenerate pathnames here, since lookup() 651 * panics on them. 652 */ 653 for (i = 1; i < ndp->ni_pathlen; i++) 654 if (cnp->cn_pnbuf[i] != '/') 655 break; 656 if (i == ndp->ni_pathlen) { 657 error = NFSERR_ACCES; 658 goto out; 659 } 660 dp = rootvnode; 661 VREF(dp); 662 } 663 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || 664 (nd->nd_flag & ND_NFSV4) == 0) { 665 /* 666 * Only cross mount points for NFSv4 when doing a 667 * mount while traversing the file system above 668 * the mount point, unless nfsrv_enable_crossmntpt is set. 669 */ 670 cnp->cn_flags |= NOCROSSMOUNT; 671 } 672 673 /* 674 * Initialize for scan, set ni_startdir and bump ref on dp again 675 * because lookup() will dereference ni_startdir. 676 */ 677 678 ndp->ni_startdir = dp; 679 ndp->ni_rootdir = rootvnode; 680 ndp->ni_topdir = NULL; 681 682 if (!lockleaf) 683 cnp->cn_flags |= LOCKLEAF; 684 for (;;) { 685 cnp->cn_nameptr = cnp->cn_pnbuf; 686 /* 687 * Call lookup() to do the real work. If an error occurs, 688 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and 689 * we do not have to dereference anything before returning. 690 * In either case ni_startdir will be dereferenced and NULLed 691 * out. 692 */ 693 error = vfs_lookup(ndp); 694 if (error) 695 break; 696 697 /* 698 * Check for encountering a symbolic link. Trivial 699 * termination occurs if no symlink encountered. 700 */ 701 if ((cnp->cn_flags & ISSYMLINK) == 0) { 702 if (ndp->ni_vp && !lockleaf) 703 NFSVOPUNLOCK(ndp->ni_vp); 704 break; 705 } 706 707 /* 708 * Validate symlink 709 */ 710 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) 711 NFSVOPUNLOCK(ndp->ni_dvp); 712 if (!(nd->nd_flag & ND_PUBLOOKUP)) { 713 error = EINVAL; 714 goto badlink2; 715 } 716 717 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 718 error = ELOOP; 719 goto badlink2; 720 } 721 if (ndp->ni_pathlen > 1) 722 cp = uma_zalloc(namei_zone, M_WAITOK); 723 else 724 cp = cnp->cn_pnbuf; 725 aiov.iov_base = cp; 726 aiov.iov_len = MAXPATHLEN; 727 auio.uio_iov = &aiov; 728 auio.uio_iovcnt = 1; 729 auio.uio_offset = 0; 730 auio.uio_rw = UIO_READ; 731 auio.uio_segflg = UIO_SYSSPACE; 732 auio.uio_td = NULL; 733 auio.uio_resid = MAXPATHLEN; 734 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 735 if (error) { 736 badlink1: 737 if (ndp->ni_pathlen > 1) 738 uma_zfree(namei_zone, cp); 739 badlink2: 740 vrele(ndp->ni_dvp); 741 vput(ndp->ni_vp); 742 break; 743 } 744 linklen = MAXPATHLEN - auio.uio_resid; 745 if (linklen == 0) { 746 error = ENOENT; 747 goto badlink1; 748 } 749 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 750 error = ENAMETOOLONG; 751 goto badlink1; 752 } 753 754 /* 755 * Adjust or replace path 756 */ 757 if (ndp->ni_pathlen > 1) { 758 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 759 uma_zfree(namei_zone, cnp->cn_pnbuf); 760 cnp->cn_pnbuf = cp; 761 } else 762 cnp->cn_pnbuf[linklen] = '\0'; 763 ndp->ni_pathlen += linklen; 764 765 /* 766 * Cleanup refs for next loop and check if root directory 767 * should replace current directory. Normally ni_dvp 768 * becomes the new base directory and is cleaned up when 769 * we loop. Explicitly null pointers after invalidation 770 * to clarify operation. 771 */ 772 vput(ndp->ni_vp); 773 ndp->ni_vp = NULL; 774 775 if (cnp->cn_pnbuf[0] == '/') { 776 vrele(ndp->ni_dvp); 777 ndp->ni_dvp = ndp->ni_rootdir; 778 VREF(ndp->ni_dvp); 779 } 780 ndp->ni_startdir = ndp->ni_dvp; 781 ndp->ni_dvp = NULL; 782 } 783 if (!lockleaf) 784 cnp->cn_flags &= ~LOCKLEAF; 785 786 out: 787 if (error) { 788 nfsvno_relpathbuf(ndp); 789 ndp->ni_vp = NULL; 790 ndp->ni_dvp = NULL; 791 ndp->ni_startdir = NULL; 792 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { 793 ndp->ni_dvp = NULL; 794 } 795 796 out1: 797 NFSEXITCODE2(error, nd); 798 return (error); 799 } 800 801 /* 802 * Set up a pathname buffer and return a pointer to it and, optionally 803 * set a hash pointer. 804 */ 805 void 806 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) 807 { 808 struct componentname *cnp = &ndp->ni_cnd; 809 810 cnp->cn_flags |= (NOMACCHECK); 811 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 812 if (hashpp != NULL) 813 *hashpp = NULL; 814 *bufpp = cnp->cn_pnbuf; 815 } 816 817 /* 818 * Release the above path buffer, if not released by nfsvno_namei(). 819 */ 820 void 821 nfsvno_relpathbuf(struct nameidata *ndp) 822 { 823 824 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 825 ndp->ni_cnd.cn_pnbuf = NULL; 826 } 827 828 /* 829 * Readlink vnode op into an mbuf list. 830 */ 831 int 832 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz, 833 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 834 { 835 struct iovec *iv; 836 struct uio io, *uiop = &io; 837 struct mbuf *mp, *mp3; 838 int len, tlen, error = 0; 839 840 len = NFS_MAXPATHLEN; 841 if (maxextsiz > 0) 842 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 843 &mp3, &mp, &iv); 844 else 845 uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); 846 uiop->uio_iov = iv; 847 uiop->uio_offset = 0; 848 uiop->uio_resid = len; 849 uiop->uio_rw = UIO_READ; 850 uiop->uio_segflg = UIO_SYSSPACE; 851 uiop->uio_td = NULL; 852 error = VOP_READLINK(vp, uiop, cred); 853 free(iv, M_TEMP); 854 if (error) { 855 m_freem(mp3); 856 *lenp = 0; 857 goto out; 858 } 859 if (uiop->uio_resid > 0) { 860 len -= uiop->uio_resid; 861 tlen = NFSM_RNDUP(len); 862 if (tlen == 0) { 863 m_freem(mp3); 864 mp3 = mp = NULL; 865 } else if (tlen != NFS_MAXPATHLEN || tlen != len) 866 mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, 867 tlen - len); 868 } 869 *lenp = len; 870 *mpp = mp3; 871 *mpendp = mp; 872 873 out: 874 NFSEXITCODE(error); 875 return (error); 876 } 877 878 /* 879 * Create an mbuf chain and an associated iovec that can be used to Read 880 * or Getextattr of data. 881 * Upon success, return pointers to the first and last mbufs in the chain 882 * plus the malloc'd iovec and its iovlen. 883 */ 884 static int 885 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, 886 struct iovec **ivp) 887 { 888 struct mbuf *m, *m2 = NULL, *m3; 889 struct iovec *iv; 890 int i, left, siz; 891 892 left = len; 893 m3 = NULL; 894 /* 895 * Generate the mbuf list with the uio_iov ref. to it. 896 */ 897 i = 0; 898 while (left > 0) { 899 NFSMGET(m); 900 MCLGET(m, M_WAITOK); 901 m->m_len = 0; 902 siz = min(M_TRAILINGSPACE(m), left); 903 left -= siz; 904 i++; 905 if (m3) 906 m2->m_next = m; 907 else 908 m3 = m; 909 m2 = m; 910 } 911 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 912 m = m3; 913 left = len; 914 i = 0; 915 while (left > 0) { 916 if (m == NULL) 917 panic("nfsrv_createiovec iov"); 918 siz = min(M_TRAILINGSPACE(m), left); 919 if (siz > 0) { 920 iv->iov_base = mtod(m, caddr_t) + m->m_len; 921 iv->iov_len = siz; 922 m->m_len += siz; 923 left -= siz; 924 iv++; 925 i++; 926 } 927 m = m->m_next; 928 } 929 *mpp = m3; 930 *mpendp = m2; 931 return (i); 932 } 933 934 /* 935 * Create an mbuf chain and an associated iovec that can be used to Read 936 * or Getextattr of data. 937 * Upon success, return pointers to the first and last mbufs in the chain 938 * plus the malloc'd iovec and its iovlen. 939 * Same as above, but creates ext_pgs mbuf(s). 940 */ 941 static int 942 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp, 943 struct mbuf **mpendp, struct iovec **ivp) 944 { 945 struct mbuf *m, *m2 = NULL, *m3; 946 struct iovec *iv; 947 int i, left, pgno, siz; 948 949 left = len; 950 m3 = NULL; 951 /* 952 * Generate the mbuf list with the uio_iov ref. to it. 953 */ 954 i = 0; 955 while (left > 0) { 956 siz = min(left, maxextsiz); 957 m = mb_alloc_ext_plus_pages(siz, M_WAITOK); 958 left -= siz; 959 i += m->m_epg_npgs; 960 if (m3 != NULL) 961 m2->m_next = m; 962 else 963 m3 = m; 964 m2 = m; 965 } 966 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 967 m = m3; 968 left = len; 969 i = 0; 970 pgno = 0; 971 while (left > 0) { 972 if (m == NULL) 973 panic("nfsvno_createiovec_extpgs iov"); 974 siz = min(PAGE_SIZE, left); 975 if (siz > 0) { 976 iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); 977 iv->iov_len = siz; 978 m->m_len += siz; 979 if (pgno == m->m_epg_npgs - 1) 980 m->m_epg_last_len = siz; 981 left -= siz; 982 iv++; 983 i++; 984 pgno++; 985 } 986 if (pgno == m->m_epg_npgs && left > 0) { 987 m = m->m_next; 988 if (m == NULL) 989 panic("nfsvno_createiovec_extpgs iov"); 990 pgno = 0; 991 } 992 } 993 *mpp = m3; 994 *mpendp = m2; 995 return (i); 996 } 997 998 /* 999 * Read vnode op call into mbuf list. 1000 */ 1001 int 1002 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 1003 int maxextsiz, struct thread *p, struct mbuf **mpp, 1004 struct mbuf **mpendp) 1005 { 1006 struct mbuf *m; 1007 struct iovec *iv; 1008 int error = 0, len, tlen, ioflag = 0; 1009 struct mbuf *m3; 1010 struct uio io, *uiop = &io; 1011 struct nfsheur *nh; 1012 1013 /* 1014 * Attempt to read from a DS file. A return of ENOENT implies 1015 * there is no DS file to read. 1016 */ 1017 error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, 1018 NULL, mpendp, NULL, NULL, NULL, 0, NULL); 1019 if (error != ENOENT) 1020 return (error); 1021 1022 len = NFSM_RNDUP(cnt); 1023 if (maxextsiz > 0) 1024 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 1025 &m3, &m, &iv); 1026 else 1027 uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); 1028 uiop->uio_iov = iv; 1029 uiop->uio_offset = off; 1030 uiop->uio_resid = len; 1031 uiop->uio_rw = UIO_READ; 1032 uiop->uio_segflg = UIO_SYSSPACE; 1033 uiop->uio_td = NULL; 1034 nh = nfsrv_sequential_heuristic(uiop, vp); 1035 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 1036 /* XXX KDM make this more systematic? */ 1037 nfsstatsv1_p->srvbytes[NFSV4OP_READ] += uiop->uio_resid; 1038 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 1039 free(iv, M_TEMP); 1040 if (error) { 1041 m_freem(m3); 1042 *mpp = NULL; 1043 goto out; 1044 } 1045 nh->nh_nextoff = uiop->uio_offset; 1046 tlen = len - uiop->uio_resid; 1047 cnt = cnt < tlen ? cnt : tlen; 1048 tlen = NFSM_RNDUP(cnt); 1049 if (tlen == 0) { 1050 m_freem(m3); 1051 m3 = m = NULL; 1052 } else if (len != tlen || tlen != cnt) 1053 m = nfsrv_adj(m3, len - tlen, tlen - cnt); 1054 *mpp = m3; 1055 *mpendp = m; 1056 1057 out: 1058 NFSEXITCODE(error); 1059 return (error); 1060 } 1061 1062 /* 1063 * Create the iovec for the mbuf chain passed in as an argument. 1064 * The "cp" argument is where the data starts within the first mbuf in 1065 * the chain. It returns the iovec and the iovcnt. 1066 */ 1067 static int 1068 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, 1069 int *iovcntp) 1070 { 1071 struct mbuf *mp; 1072 struct iovec *ivp; 1073 int cnt, i, len; 1074 1075 /* 1076 * Loop through the mbuf chain, counting how many mbufs are a 1077 * part of this write operation, so the iovec size is known. 1078 */ 1079 cnt = 0; 1080 len = retlen; 1081 mp = m; 1082 i = mtod(mp, caddr_t) + mp->m_len - cp; 1083 while (len > 0) { 1084 if (i > 0) { 1085 len -= i; 1086 cnt++; 1087 } 1088 mp = mp->m_next; 1089 if (!mp) { 1090 if (len > 0) 1091 return (EBADRPC); 1092 } else 1093 i = mp->m_len; 1094 } 1095 1096 /* Now, create the iovec. */ 1097 mp = m; 1098 *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, 1099 M_WAITOK); 1100 *iovcntp = cnt; 1101 i = mtod(mp, caddr_t) + mp->m_len - cp; 1102 len = retlen; 1103 while (len > 0) { 1104 if (mp == NULL) 1105 panic("nfsrv_createiovecw"); 1106 if (i > 0) { 1107 i = min(i, len); 1108 ivp->iov_base = cp; 1109 ivp->iov_len = i; 1110 ivp++; 1111 len -= i; 1112 } 1113 mp = mp->m_next; 1114 if (mp) { 1115 i = mp->m_len; 1116 cp = mtod(mp, caddr_t); 1117 } 1118 } 1119 return (0); 1120 } 1121 1122 /* 1123 * Write vnode op from an mbuf list. 1124 */ 1125 int 1126 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, 1127 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) 1128 { 1129 struct iovec *iv; 1130 int cnt, ioflags, error; 1131 struct uio io, *uiop = &io; 1132 struct nfsheur *nh; 1133 1134 /* 1135 * Attempt to write to a DS file. A return of ENOENT implies 1136 * there is no DS file to write. 1137 */ 1138 error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, 1139 &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); 1140 if (error != ENOENT) { 1141 *stable = NFSWRITE_FILESYNC; 1142 return (error); 1143 } 1144 1145 if (*stable == NFSWRITE_UNSTABLE) 1146 ioflags = IO_NODELOCKED; 1147 else 1148 ioflags = (IO_SYNC | IO_NODELOCKED); 1149 error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); 1150 if (error != 0) 1151 return (error); 1152 uiop->uio_iov = iv; 1153 uiop->uio_iovcnt = cnt; 1154 uiop->uio_resid = retlen; 1155 uiop->uio_rw = UIO_WRITE; 1156 uiop->uio_segflg = UIO_SYSSPACE; 1157 NFSUIOPROC(uiop, p); 1158 uiop->uio_offset = off; 1159 nh = nfsrv_sequential_heuristic(uiop, vp); 1160 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 1161 /* XXX KDM make this more systematic? */ 1162 nfsstatsv1_p->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; 1163 error = VOP_WRITE(vp, uiop, ioflags, cred); 1164 if (error == 0) 1165 nh->nh_nextoff = uiop->uio_offset; 1166 free(iv, M_TEMP); 1167 1168 NFSEXITCODE(error); 1169 return (error); 1170 } 1171 1172 /* 1173 * Common code for creating a regular file (plus special files for V2). 1174 */ 1175 int 1176 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, 1177 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, 1178 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) 1179 { 1180 u_quad_t tempsize; 1181 int error; 1182 struct thread *p = curthread; 1183 1184 error = nd->nd_repstat; 1185 if (!error && ndp->ni_vp == NULL) { 1186 if (nvap->na_type == VREG || nvap->na_type == VSOCK) { 1187 error = VOP_CREATE(ndp->ni_dvp, 1188 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1189 /* For a pNFS server, create the data file on a DS. */ 1190 if (error == 0 && nvap->na_type == VREG) { 1191 /* 1192 * Create a data file on a DS for a pNFS server. 1193 * This function just returns if not 1194 * running a pNFS DS or the creation fails. 1195 */ 1196 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1197 nd->nd_cred, p); 1198 } 1199 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1200 NULL, false); 1201 nfsvno_relpathbuf(ndp); 1202 if (!error) { 1203 if (*exclusive_flagp) { 1204 *exclusive_flagp = 0; 1205 NFSVNO_ATTRINIT(nvap); 1206 nvap->na_atime.tv_sec = cverf[0]; 1207 nvap->na_atime.tv_nsec = cverf[1]; 1208 error = VOP_SETATTR(ndp->ni_vp, 1209 &nvap->na_vattr, nd->nd_cred); 1210 if (error != 0) { 1211 vput(ndp->ni_vp); 1212 ndp->ni_vp = NULL; 1213 error = NFSERR_NOTSUPP; 1214 } 1215 } 1216 } 1217 /* 1218 * NFS V2 Only. nfsrvd_mknod() does this for V3. 1219 * (This implies, just get out on an error.) 1220 */ 1221 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || 1222 nvap->na_type == VFIFO) { 1223 if (nvap->na_type == VCHR && rdev == 0xffffffff) 1224 nvap->na_type = VFIFO; 1225 if (nvap->na_type != VFIFO && 1226 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { 1227 nfsvno_relpathbuf(ndp); 1228 vput(ndp->ni_dvp); 1229 goto out; 1230 } 1231 nvap->na_rdev = rdev; 1232 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1233 &ndp->ni_cnd, &nvap->na_vattr); 1234 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1235 NULL, false); 1236 nfsvno_relpathbuf(ndp); 1237 if (error) 1238 goto out; 1239 } else { 1240 nfsvno_relpathbuf(ndp); 1241 vput(ndp->ni_dvp); 1242 error = ENXIO; 1243 goto out; 1244 } 1245 *vpp = ndp->ni_vp; 1246 } else { 1247 /* 1248 * Handle cases where error is already set and/or 1249 * the file exists. 1250 * 1 - clean up the lookup 1251 * 2 - iff !error and na_size set, truncate it 1252 */ 1253 nfsvno_relpathbuf(ndp); 1254 *vpp = ndp->ni_vp; 1255 if (ndp->ni_dvp == *vpp) 1256 vrele(ndp->ni_dvp); 1257 else 1258 vput(ndp->ni_dvp); 1259 if (!error && nvap->na_size != VNOVAL) { 1260 error = nfsvno_accchk(*vpp, VWRITE, 1261 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1262 NFSACCCHK_VPISLOCKED, NULL); 1263 if (!error) { 1264 tempsize = nvap->na_size; 1265 NFSVNO_ATTRINIT(nvap); 1266 nvap->na_size = tempsize; 1267 error = nfsvno_setattr(*vpp, nvap, 1268 nd->nd_cred, p, exp); 1269 } 1270 } 1271 if (error) 1272 vput(*vpp); 1273 } 1274 1275 out: 1276 NFSEXITCODE(error); 1277 return (error); 1278 } 1279 1280 /* 1281 * Do a mknod vnode op. 1282 */ 1283 int 1284 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, 1285 struct thread *p) 1286 { 1287 int error = 0; 1288 enum vtype vtyp; 1289 1290 vtyp = nvap->na_type; 1291 /* 1292 * Iff doesn't exist, create it. 1293 */ 1294 if (ndp->ni_vp) { 1295 nfsvno_relpathbuf(ndp); 1296 vput(ndp->ni_dvp); 1297 vrele(ndp->ni_vp); 1298 error = EEXIST; 1299 goto out; 1300 } 1301 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1302 nfsvno_relpathbuf(ndp); 1303 vput(ndp->ni_dvp); 1304 error = NFSERR_BADTYPE; 1305 goto out; 1306 } 1307 if (vtyp == VSOCK) { 1308 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 1309 &ndp->ni_cnd, &nvap->na_vattr); 1310 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1311 false); 1312 nfsvno_relpathbuf(ndp); 1313 } else { 1314 if (nvap->na_type != VFIFO && 1315 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { 1316 nfsvno_relpathbuf(ndp); 1317 vput(ndp->ni_dvp); 1318 goto out; 1319 } 1320 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1321 &ndp->ni_cnd, &nvap->na_vattr); 1322 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1323 false); 1324 nfsvno_relpathbuf(ndp); 1325 /* 1326 * Since VOP_MKNOD returns the ni_vp, I can't 1327 * see any reason to do the lookup. 1328 */ 1329 } 1330 1331 out: 1332 NFSEXITCODE(error); 1333 return (error); 1334 } 1335 1336 /* 1337 * Mkdir vnode op. 1338 */ 1339 int 1340 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, 1341 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 1342 { 1343 int error = 0; 1344 1345 if (ndp->ni_vp != NULL) { 1346 if (ndp->ni_dvp == ndp->ni_vp) 1347 vrele(ndp->ni_dvp); 1348 else 1349 vput(ndp->ni_dvp); 1350 vrele(ndp->ni_vp); 1351 nfsvno_relpathbuf(ndp); 1352 error = EEXIST; 1353 goto out; 1354 } 1355 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1356 &nvap->na_vattr); 1357 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); 1358 nfsvno_relpathbuf(ndp); 1359 1360 out: 1361 NFSEXITCODE(error); 1362 return (error); 1363 } 1364 1365 /* 1366 * symlink vnode op. 1367 */ 1368 int 1369 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, 1370 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, 1371 struct nfsexstuff *exp) 1372 { 1373 int error = 0; 1374 1375 if (ndp->ni_vp) { 1376 nfsvno_relpathbuf(ndp); 1377 if (ndp->ni_dvp == ndp->ni_vp) 1378 vrele(ndp->ni_dvp); 1379 else 1380 vput(ndp->ni_dvp); 1381 vrele(ndp->ni_vp); 1382 error = EEXIST; 1383 goto out; 1384 } 1385 1386 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1387 &nvap->na_vattr, pathcp); 1388 /* 1389 * Although FreeBSD still had the lookup code in 1390 * it for 7/current, there doesn't seem to be any 1391 * point, since VOP_SYMLINK() returns the ni_vp. 1392 * Just vput it for v2. 1393 */ 1394 VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0); 1395 nfsvno_relpathbuf(ndp); 1396 1397 out: 1398 NFSEXITCODE(error); 1399 return (error); 1400 } 1401 1402 /* 1403 * Parse symbolic link arguments. 1404 * This function has an ugly side effect. It will malloc() an area for 1405 * the symlink and set iov_base to point to it, only if it succeeds. 1406 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must 1407 * be FREE'd later. 1408 */ 1409 int 1410 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, 1411 struct thread *p, char **pathcpp, int *lenp) 1412 { 1413 u_int32_t *tl; 1414 char *pathcp = NULL; 1415 int error = 0, len; 1416 struct nfsv2_sattr *sp; 1417 1418 *pathcpp = NULL; 1419 *lenp = 0; 1420 if ((nd->nd_flag & ND_NFSV3) && 1421 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) 1422 goto nfsmout; 1423 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 1424 len = fxdr_unsigned(int, *tl); 1425 if (len > NFS_MAXPATHLEN || len <= 0) { 1426 error = EBADRPC; 1427 goto nfsmout; 1428 } 1429 pathcp = malloc(len + 1, M_TEMP, M_WAITOK); 1430 error = nfsrv_mtostr(nd, pathcp, len); 1431 if (error) 1432 goto nfsmout; 1433 if (nd->nd_flag & ND_NFSV2) { 1434 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1435 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 1436 } 1437 *pathcpp = pathcp; 1438 *lenp = len; 1439 NFSEXITCODE2(0, nd); 1440 return (0); 1441 nfsmout: 1442 if (pathcp) 1443 free(pathcp, M_TEMP); 1444 NFSEXITCODE2(error, nd); 1445 return (error); 1446 } 1447 1448 /* 1449 * Remove a non-directory object. 1450 */ 1451 int 1452 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1453 struct thread *p, struct nfsexstuff *exp) 1454 { 1455 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS]; 1456 int error = 0, mirrorcnt; 1457 char fname[PNFS_FILENAME_LEN + 1]; 1458 fhandle_t fh; 1459 1460 vp = ndp->ni_vp; 1461 dsdvp[0] = NULL; 1462 if (vp->v_type == VDIR) 1463 error = NFSERR_ISDIR; 1464 else if (is_v4) 1465 error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0), 1466 p); 1467 if (error == 0) 1468 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); 1469 if (!error) 1470 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); 1471 if (error == 0 && dsdvp[0] != NULL) 1472 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1473 if (ndp->ni_dvp == vp) 1474 vrele(ndp->ni_dvp); 1475 else 1476 vput(ndp->ni_dvp); 1477 vput(vp); 1478 nfsvno_relpathbuf(ndp); 1479 NFSEXITCODE(error); 1480 return (error); 1481 } 1482 1483 /* 1484 * Remove a directory. 1485 */ 1486 int 1487 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1488 struct thread *p, struct nfsexstuff *exp) 1489 { 1490 struct vnode *vp; 1491 int error = 0; 1492 1493 vp = ndp->ni_vp; 1494 if (vp->v_type != VDIR) { 1495 error = ENOTDIR; 1496 goto out; 1497 } 1498 /* 1499 * No rmdir "." please. 1500 */ 1501 if (ndp->ni_dvp == vp) { 1502 error = EINVAL; 1503 goto out; 1504 } 1505 /* 1506 * The root of a mounted filesystem cannot be deleted. 1507 */ 1508 if (vp->v_vflag & VV_ROOT) 1509 error = EBUSY; 1510 out: 1511 if (!error) 1512 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); 1513 if (ndp->ni_dvp == vp) 1514 vrele(ndp->ni_dvp); 1515 else 1516 vput(ndp->ni_dvp); 1517 vput(vp); 1518 nfsvno_relpathbuf(ndp); 1519 NFSEXITCODE(error); 1520 return (error); 1521 } 1522 1523 /* 1524 * Rename vnode op. 1525 */ 1526 int 1527 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, 1528 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) 1529 { 1530 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS]; 1531 int error = 0, mirrorcnt; 1532 char fname[PNFS_FILENAME_LEN + 1]; 1533 fhandle_t fh; 1534 1535 dsdvp[0] = NULL; 1536 fvp = fromndp->ni_vp; 1537 if (ndstat) { 1538 vrele(fromndp->ni_dvp); 1539 vrele(fvp); 1540 error = ndstat; 1541 goto out1; 1542 } 1543 tdvp = tondp->ni_dvp; 1544 tvp = tondp->ni_vp; 1545 if (tvp != NULL) { 1546 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1547 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; 1548 goto out; 1549 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1550 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; 1551 goto out; 1552 } 1553 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1554 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1555 goto out; 1556 } 1557 1558 /* 1559 * A rename to '.' or '..' results in a prematurely 1560 * unlocked vnode on FreeBSD5, so I'm just going to fail that 1561 * here. 1562 */ 1563 if ((tondp->ni_cnd.cn_namelen == 1 && 1564 tondp->ni_cnd.cn_nameptr[0] == '.') || 1565 (tondp->ni_cnd.cn_namelen == 2 && 1566 tondp->ni_cnd.cn_nameptr[0] == '.' && 1567 tondp->ni_cnd.cn_nameptr[1] == '.')) { 1568 error = EINVAL; 1569 goto out; 1570 } 1571 } 1572 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1573 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1574 goto out; 1575 } 1576 if (fvp->v_mount != tdvp->v_mount) { 1577 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1578 goto out; 1579 } 1580 if (fvp == tdvp) { 1581 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; 1582 goto out; 1583 } 1584 if (fvp == tvp) { 1585 /* 1586 * If source and destination are the same, there is nothing to 1587 * do. Set error to -1 to indicate this. 1588 */ 1589 error = -1; 1590 goto out; 1591 } 1592 if (ndflag & ND_NFSV4) { 1593 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { 1594 error = nfsrv_checkremove(fvp, 0, NULL, 1595 (nfsquad_t)((u_quad_t)0), p); 1596 NFSVOPUNLOCK(fvp); 1597 } else 1598 error = EPERM; 1599 if (tvp && !error) 1600 error = nfsrv_checkremove(tvp, 1, NULL, 1601 (nfsquad_t)((u_quad_t)0), p); 1602 } else { 1603 /* 1604 * For NFSv2 and NFSv3, try to get rid of the delegation, so 1605 * that the NFSv4 client won't be confused by the rename. 1606 * Since nfsd_recalldelegation() can only be called on an 1607 * unlocked vnode at this point and fvp is the file that will 1608 * still exist after the rename, just do fvp. 1609 */ 1610 nfsd_recalldelegation(fvp, p); 1611 } 1612 if (error == 0 && tvp != NULL) { 1613 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh); 1614 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" 1615 " dsdvp=%p\n", dsdvp[0]); 1616 } 1617 out: 1618 if (!error) { 1619 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, 1620 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, 1621 &tondp->ni_cnd); 1622 } else { 1623 if (tdvp == tvp) 1624 vrele(tdvp); 1625 else 1626 vput(tdvp); 1627 if (tvp) 1628 vput(tvp); 1629 vrele(fromndp->ni_dvp); 1630 vrele(fvp); 1631 if (error == -1) 1632 error = 0; 1633 } 1634 1635 /* 1636 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and 1637 * if the rename succeeded, the DS file for the tvp needs to be 1638 * removed. 1639 */ 1640 if (error == 0 && dsdvp[0] != NULL) { 1641 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1642 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); 1643 } 1644 1645 nfsvno_relpathbuf(tondp); 1646 out1: 1647 nfsvno_relpathbuf(fromndp); 1648 NFSEXITCODE(error); 1649 return (error); 1650 } 1651 1652 /* 1653 * Link vnode op. 1654 */ 1655 int 1656 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, 1657 struct thread *p, struct nfsexstuff *exp) 1658 { 1659 struct vnode *xp; 1660 int error = 0; 1661 1662 xp = ndp->ni_vp; 1663 if (xp != NULL) { 1664 error = EEXIST; 1665 } else { 1666 xp = ndp->ni_dvp; 1667 if (vp->v_mount != xp->v_mount) 1668 error = EXDEV; 1669 } 1670 if (!error) { 1671 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1672 if (!VN_IS_DOOMED(vp)) 1673 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); 1674 else 1675 error = EPERM; 1676 if (ndp->ni_dvp == vp) { 1677 vrele(ndp->ni_dvp); 1678 NFSVOPUNLOCK(vp); 1679 } else { 1680 vref(vp); 1681 VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true); 1682 } 1683 } else { 1684 if (ndp->ni_dvp == ndp->ni_vp) 1685 vrele(ndp->ni_dvp); 1686 else 1687 vput(ndp->ni_dvp); 1688 if (ndp->ni_vp) 1689 vrele(ndp->ni_vp); 1690 } 1691 nfsvno_relpathbuf(ndp); 1692 NFSEXITCODE(error); 1693 return (error); 1694 } 1695 1696 /* 1697 * Do the fsync() appropriate for the commit. 1698 */ 1699 int 1700 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, 1701 struct thread *td) 1702 { 1703 int error = 0; 1704 1705 /* 1706 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of 1707 * file is done. At this time VOP_FSYNC does not accept offset and 1708 * byte count parameters so call VOP_FSYNC the whole file for now. 1709 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. 1710 * File systems that do not use the buffer cache (as indicated 1711 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). 1712 */ 1713 if (cnt == 0 || cnt > MAX_COMMIT_COUNT || 1714 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { 1715 /* 1716 * Give up and do the whole thing 1717 */ 1718 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1719 VM_OBJECT_WLOCK(vp->v_object); 1720 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); 1721 VM_OBJECT_WUNLOCK(vp->v_object); 1722 } 1723 error = VOP_FSYNC(vp, MNT_WAIT, td); 1724 } else { 1725 /* 1726 * Locate and synchronously write any buffers that fall 1727 * into the requested range. Note: we are assuming that 1728 * f_iosize is a power of 2. 1729 */ 1730 int iosize = vp->v_mount->mnt_stat.f_iosize; 1731 int iomask = iosize - 1; 1732 struct bufobj *bo; 1733 daddr_t lblkno; 1734 1735 /* 1736 * Align to iosize boundary, super-align to page boundary. 1737 */ 1738 if (off & iomask) { 1739 cnt += off & iomask; 1740 off &= ~(u_quad_t)iomask; 1741 } 1742 if (off & PAGE_MASK) { 1743 cnt += off & PAGE_MASK; 1744 off &= ~(u_quad_t)PAGE_MASK; 1745 } 1746 lblkno = off / iosize; 1747 1748 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1749 VM_OBJECT_WLOCK(vp->v_object); 1750 vm_object_page_clean(vp->v_object, off, off + cnt, 1751 OBJPC_SYNC); 1752 VM_OBJECT_WUNLOCK(vp->v_object); 1753 } 1754 1755 bo = &vp->v_bufobj; 1756 BO_LOCK(bo); 1757 while (cnt > 0) { 1758 struct buf *bp; 1759 1760 /* 1761 * If we have a buffer and it is marked B_DELWRI we 1762 * have to lock and write it. Otherwise the prior 1763 * write is assumed to have already been committed. 1764 * 1765 * gbincore() can return invalid buffers now so we 1766 * have to check that bit as well (though B_DELWRI 1767 * should not be set if B_INVAL is set there could be 1768 * a race here since we haven't locked the buffer). 1769 */ 1770 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 1771 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 1772 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { 1773 BO_LOCK(bo); 1774 continue; /* retry */ 1775 } 1776 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 1777 B_DELWRI) { 1778 bremfree(bp); 1779 bp->b_flags &= ~B_ASYNC; 1780 bwrite(bp); 1781 ++nfs_commit_miss; 1782 } else 1783 BUF_UNLOCK(bp); 1784 BO_LOCK(bo); 1785 } 1786 ++nfs_commit_blks; 1787 if (cnt < iosize) 1788 break; 1789 cnt -= iosize; 1790 ++lblkno; 1791 } 1792 BO_UNLOCK(bo); 1793 } 1794 NFSEXITCODE(error); 1795 return (error); 1796 } 1797 1798 /* 1799 * Statfs vnode op. 1800 */ 1801 int 1802 nfsvno_statfs(struct vnode *vp, struct statfs *sf) 1803 { 1804 struct statfs *tsf; 1805 int error; 1806 1807 tsf = NULL; 1808 if (nfsrv_devidcnt > 0) { 1809 /* For a pNFS service, get the DS numbers. */ 1810 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); 1811 error = nfsrv_pnfsstatfs(tsf, vp->v_mount); 1812 if (error != 0) { 1813 free(tsf, M_TEMP); 1814 tsf = NULL; 1815 } 1816 } 1817 error = VFS_STATFS(vp->v_mount, sf); 1818 if (error == 0) { 1819 if (tsf != NULL) { 1820 sf->f_blocks = tsf->f_blocks; 1821 sf->f_bavail = tsf->f_bavail; 1822 sf->f_bfree = tsf->f_bfree; 1823 sf->f_bsize = tsf->f_bsize; 1824 } 1825 /* 1826 * Since NFS handles these values as unsigned on the 1827 * wire, there is no way to represent negative values, 1828 * so set them to 0. Without this, they will appear 1829 * to be very large positive values for clients like 1830 * Solaris10. 1831 */ 1832 if (sf->f_bavail < 0) 1833 sf->f_bavail = 0; 1834 if (sf->f_ffree < 0) 1835 sf->f_ffree = 0; 1836 } 1837 free(tsf, M_TEMP); 1838 NFSEXITCODE(error); 1839 return (error); 1840 } 1841 1842 /* 1843 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but 1844 * must handle nfsrv_opencheck() calls after any other access checks. 1845 */ 1846 void 1847 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, 1848 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, 1849 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, 1850 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, bool done_namei, 1851 struct nfsexstuff *exp, struct vnode **vpp) 1852 { 1853 struct vnode *vp = NULL; 1854 u_quad_t tempsize; 1855 struct nfsexstuff nes; 1856 struct thread *p = curthread; 1857 uint32_t oldrepstat; 1858 1859 if (ndp->ni_vp == NULL) { 1860 /* 1861 * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be 1862 * set true, since cleanup after nfsvno_namei() is needed. 1863 */ 1864 oldrepstat = nd->nd_repstat; 1865 nd->nd_repstat = nfsrv_opencheck(clientid, 1866 stateidp, stp, NULL, nd, p, nd->nd_repstat); 1867 if (nd->nd_repstat != 0 && oldrepstat == 0) 1868 done_namei = true; 1869 } 1870 if (!nd->nd_repstat) { 1871 if (ndp->ni_vp == NULL) { 1872 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, 1873 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1874 /* For a pNFS server, create the data file on a DS. */ 1875 if (nd->nd_repstat == 0) { 1876 /* 1877 * Create a data file on a DS for a pNFS server. 1878 * This function just returns if not 1879 * running a pNFS DS or the creation fails. 1880 */ 1881 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1882 cred, p); 1883 } 1884 VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ? 1885 &ndp->ni_vp : NULL, false); 1886 nfsvno_relpathbuf(ndp); 1887 if (!nd->nd_repstat) { 1888 if (*exclusive_flagp) { 1889 *exclusive_flagp = 0; 1890 NFSVNO_ATTRINIT(nvap); 1891 nvap->na_atime.tv_sec = cverf[0]; 1892 nvap->na_atime.tv_nsec = cverf[1]; 1893 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, 1894 &nvap->na_vattr, cred); 1895 if (nd->nd_repstat != 0) { 1896 vput(ndp->ni_vp); 1897 ndp->ni_vp = NULL; 1898 nd->nd_repstat = NFSERR_NOTSUPP; 1899 } else 1900 NFSSETBIT_ATTRBIT(attrbitp, 1901 NFSATTRBIT_TIMEACCESS); 1902 } else { 1903 nfsrv_fixattr(nd, ndp->ni_vp, nvap, 1904 aclp, p, attrbitp, exp); 1905 } 1906 } 1907 vp = ndp->ni_vp; 1908 } else { 1909 nfsvno_relpathbuf(ndp); 1910 vp = ndp->ni_vp; 1911 if (create == NFSV4OPEN_CREATE) { 1912 if (ndp->ni_dvp == vp) 1913 vrele(ndp->ni_dvp); 1914 else 1915 vput(ndp->ni_dvp); 1916 } 1917 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { 1918 if (ndp->ni_cnd.cn_flags & RDONLY) 1919 NFSVNO_SETEXRDONLY(&nes); 1920 else 1921 NFSVNO_EXINIT(&nes); 1922 nd->nd_repstat = nfsvno_accchk(vp, 1923 VWRITE, cred, &nes, p, 1924 NFSACCCHK_NOOVERRIDE, 1925 NFSACCCHK_VPISLOCKED, NULL); 1926 nd->nd_repstat = nfsrv_opencheck(clientid, 1927 stateidp, stp, vp, nd, p, nd->nd_repstat); 1928 if (!nd->nd_repstat) { 1929 tempsize = nvap->na_size; 1930 NFSVNO_ATTRINIT(nvap); 1931 nvap->na_size = tempsize; 1932 nd->nd_repstat = nfsvno_setattr(vp, 1933 nvap, cred, p, exp); 1934 } 1935 } else if (vp->v_type == VREG) { 1936 nd->nd_repstat = nfsrv_opencheck(clientid, 1937 stateidp, stp, vp, nd, p, nd->nd_repstat); 1938 } 1939 } 1940 } else if (done_namei) { 1941 KASSERT(create == NFSV4OPEN_CREATE, 1942 ("nfsvno_open: not create")); 1943 /* 1944 * done_namei is set when nfsvno_namei() has completed 1945 * successfully, but a subsequent error was set in 1946 * nd_repstat. As such, cleanup of the nfsvno_namei() 1947 * results is required. 1948 */ 1949 nfsvno_relpathbuf(ndp); 1950 if (ndp->ni_dvp == ndp->ni_vp) 1951 vrele(ndp->ni_dvp); 1952 else 1953 vput(ndp->ni_dvp); 1954 if (ndp->ni_vp) 1955 vput(ndp->ni_vp); 1956 } 1957 *vpp = vp; 1958 1959 NFSEXITCODE2(0, nd); 1960 } 1961 1962 /* 1963 * Updates the file rev and sets the mtime and ctime 1964 * to the current clock time, returning the va_filerev and va_Xtime 1965 * values. 1966 * Return ESTALE to indicate the vnode is VIRF_DOOMED. 1967 */ 1968 int 1969 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, 1970 struct nfsrv_descript *nd, struct thread *p) 1971 { 1972 struct vattr va; 1973 1974 VATTR_NULL(&va); 1975 vfs_timestamp(&va.va_mtime); 1976 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 1977 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 1978 if (VN_IS_DOOMED(vp)) 1979 return (ESTALE); 1980 } 1981 (void) VOP_SETATTR(vp, &va, nd->nd_cred); 1982 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); 1983 return (0); 1984 } 1985 1986 /* 1987 * Glue routine to nfsv4_fillattr(). 1988 */ 1989 int 1990 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, 1991 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, 1992 struct ucred *cred, struct thread *p, int isdgram, int reterr, 1993 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) 1994 { 1995 struct statfs *sf; 1996 int error; 1997 1998 sf = NULL; 1999 if (nfsrv_devidcnt > 0 && 2000 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || 2001 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || 2002 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { 2003 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); 2004 error = nfsrv_pnfsstatfs(sf, mp); 2005 if (error != 0) { 2006 free(sf, M_TEMP); 2007 sf = NULL; 2008 } 2009 } 2010 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, 2011 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, 2012 mounted_on_fileno, sf); 2013 free(sf, M_TEMP); 2014 NFSEXITCODE2(0, nd); 2015 return (error); 2016 } 2017 2018 /* Since the Readdir vnode ops vary, put the entire functions in here. */ 2019 /* 2020 * nfs readdir service 2021 * - mallocs what it thinks is enough to read 2022 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR 2023 * - calls VOP_READDIR() 2024 * - loops around building the reply 2025 * if the output generated exceeds count break out of loop 2026 * The NFSM_CLGET macro is used here so that the reply will be packed 2027 * tightly in mbuf clusters. 2028 * - it trims out records with d_fileno == 0 2029 * this doesn't matter for Unix clients, but they might confuse clients 2030 * for other os'. 2031 * - it trims out records with d_type == DT_WHT 2032 * these cannot be seen through NFS (unless we extend the protocol) 2033 * The alternate call nfsrvd_readdirplus() does lookups as well. 2034 * PS: The NFS protocol spec. does not clarify what the "count" byte 2035 * argument is a count of.. just name strings and file id's or the 2036 * entire reply rpc or ... 2037 * I tried just file name and id sizes and it confused the Sun client, 2038 * so I am using the full rpc size now. The "paranoia.." comment refers 2039 * to including the status longwords that are not a part of the dir. 2040 * "entry" structures, but are in the rpc. 2041 */ 2042 int 2043 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, 2044 struct vnode *vp, struct nfsexstuff *exp) 2045 { 2046 struct dirent *dp; 2047 u_int32_t *tl; 2048 int dirlen; 2049 char *cpos, *cend, *rbuf; 2050 struct nfsvattr at; 2051 int nlen, error = 0, getret = 1; 2052 int siz, cnt, fullsiz, eofflag, ncookies; 2053 u_int64_t off, toff, verf __unused; 2054 uint64_t *cookies = NULL, *cookiep; 2055 struct uio io; 2056 struct iovec iv; 2057 int is_ufs; 2058 struct thread *p = curthread; 2059 2060 if (nd->nd_repstat) { 2061 nfsrv_postopattr(nd, getret, &at); 2062 goto out; 2063 } 2064 if (nd->nd_flag & ND_NFSV2) { 2065 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2066 off = fxdr_unsigned(u_quad_t, *tl++); 2067 } else { 2068 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2069 off = fxdr_hyper(tl); 2070 tl += 2; 2071 verf = fxdr_hyper(tl); 2072 tl += 2; 2073 } 2074 toff = off; 2075 cnt = fxdr_unsigned(int, *tl); 2076 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2077 cnt = NFS_SRVMAXDATA(nd); 2078 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2079 fullsiz = siz; 2080 if (nd->nd_flag & ND_NFSV3) { 2081 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, 2082 NULL); 2083 #if 0 2084 /* 2085 * va_filerev is not sufficient as a cookie verifier, 2086 * since it is not supposed to change when entries are 2087 * removed/added unless that offset cookies returned to 2088 * the client are no longer valid. 2089 */ 2090 if (!nd->nd_repstat && toff && verf != at.na_filerev) 2091 nd->nd_repstat = NFSERR_BAD_COOKIE; 2092 #endif 2093 } 2094 if (!nd->nd_repstat && vp->v_type != VDIR) 2095 nd->nd_repstat = NFSERR_NOTDIR; 2096 if (nd->nd_repstat == 0 && cnt == 0) { 2097 if (nd->nd_flag & ND_NFSV2) 2098 /* NFSv2 does not have NFSERR_TOOSMALL */ 2099 nd->nd_repstat = EPERM; 2100 else 2101 nd->nd_repstat = NFSERR_TOOSMALL; 2102 } 2103 if (!nd->nd_repstat) 2104 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2105 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2106 NFSACCCHK_VPISLOCKED, NULL); 2107 if (nd->nd_repstat) { 2108 vput(vp); 2109 if (nd->nd_flag & ND_NFSV3) 2110 nfsrv_postopattr(nd, getret, &at); 2111 goto out; 2112 } 2113 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2114 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2115 again: 2116 eofflag = 0; 2117 if (cookies) { 2118 free(cookies, M_TEMP); 2119 cookies = NULL; 2120 } 2121 2122 iv.iov_base = rbuf; 2123 iv.iov_len = siz; 2124 io.uio_iov = &iv; 2125 io.uio_iovcnt = 1; 2126 io.uio_offset = (off_t)off; 2127 io.uio_resid = siz; 2128 io.uio_segflg = UIO_SYSSPACE; 2129 io.uio_rw = UIO_READ; 2130 io.uio_td = NULL; 2131 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2132 &cookies); 2133 off = (u_int64_t)io.uio_offset; 2134 if (io.uio_resid) 2135 siz -= io.uio_resid; 2136 2137 if (!cookies && !nd->nd_repstat) 2138 nd->nd_repstat = NFSERR_PERM; 2139 if (nd->nd_flag & ND_NFSV3) { 2140 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2141 if (!nd->nd_repstat) 2142 nd->nd_repstat = getret; 2143 } 2144 2145 /* 2146 * Handles the failed cases. nd->nd_repstat == 0 past here. 2147 */ 2148 if (nd->nd_repstat) { 2149 vput(vp); 2150 free(rbuf, M_TEMP); 2151 if (cookies) 2152 free(cookies, M_TEMP); 2153 if (nd->nd_flag & ND_NFSV3) 2154 nfsrv_postopattr(nd, getret, &at); 2155 goto out; 2156 } 2157 /* 2158 * If nothing read, return eof 2159 * rpc reply 2160 */ 2161 if (siz == 0) { 2162 vput(vp); 2163 if (nd->nd_flag & ND_NFSV2) { 2164 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2165 } else { 2166 nfsrv_postopattr(nd, getret, &at); 2167 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2168 txdr_hyper(at.na_filerev, tl); 2169 tl += 2; 2170 } 2171 *tl++ = newnfs_false; 2172 *tl = newnfs_true; 2173 free(rbuf, M_TEMP); 2174 free(cookies, M_TEMP); 2175 goto out; 2176 } 2177 2178 /* 2179 * Check for degenerate cases of nothing useful read. 2180 * If so go try again 2181 */ 2182 cpos = rbuf; 2183 cend = rbuf + siz; 2184 dp = (struct dirent *)cpos; 2185 cookiep = cookies; 2186 2187 /* 2188 * For some reason FreeBSD's ufs_readdir() chooses to back the 2189 * directory offset up to a block boundary, so it is necessary to 2190 * skip over the records that precede the requested offset. This 2191 * requires the assumption that file offset cookies monotonically 2192 * increase. 2193 */ 2194 while (cpos < cend && ncookies > 0 && 2195 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2196 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { 2197 cpos += dp->d_reclen; 2198 dp = (struct dirent *)cpos; 2199 cookiep++; 2200 ncookies--; 2201 } 2202 if (cpos >= cend || ncookies == 0) { 2203 siz = fullsiz; 2204 toff = off; 2205 goto again; 2206 } 2207 vput(vp); 2208 2209 /* 2210 * If cnt > MCLBYTES and the reply will not be saved, use 2211 * ext_pgs mbufs for TLS. 2212 * For NFSv4.0, we do not know for sure if the reply will 2213 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2214 */ 2215 if (cnt > MCLBYTES && siz > MCLBYTES && 2216 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2217 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2218 nd->nd_flag |= ND_EXTPG; 2219 2220 /* 2221 * dirlen is the size of the reply, including all XDR and must 2222 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate 2223 * if the XDR should be included in "count", but to be safe, we do. 2224 * (Include the two booleans at the end of the reply in dirlen now.) 2225 */ 2226 if (nd->nd_flag & ND_NFSV3) { 2227 nfsrv_postopattr(nd, getret, &at); 2228 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2229 txdr_hyper(at.na_filerev, tl); 2230 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2231 } else { 2232 dirlen = 2 * NFSX_UNSIGNED; 2233 } 2234 2235 /* Loop through the records and build reply */ 2236 while (cpos < cend && ncookies > 0) { 2237 nlen = dp->d_namlen; 2238 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2239 nlen <= NFS_MAXNAMLEN) { 2240 if (nd->nd_flag & ND_NFSV3) 2241 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2242 else 2243 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2244 if (dirlen > cnt) { 2245 eofflag = 0; 2246 break; 2247 } 2248 2249 /* 2250 * Build the directory record xdr from 2251 * the dirent entry. 2252 */ 2253 if (nd->nd_flag & ND_NFSV3) { 2254 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2255 *tl++ = newnfs_true; 2256 *tl++ = 0; 2257 } else { 2258 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2259 *tl++ = newnfs_true; 2260 } 2261 *tl = txdr_unsigned(dp->d_fileno); 2262 (void) nfsm_strtom(nd, dp->d_name, nlen); 2263 if (nd->nd_flag & ND_NFSV3) { 2264 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2265 txdr_hyper(*cookiep, tl); 2266 } else { 2267 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 2268 *tl = txdr_unsigned(*cookiep); 2269 } 2270 } 2271 cpos += dp->d_reclen; 2272 dp = (struct dirent *)cpos; 2273 cookiep++; 2274 ncookies--; 2275 } 2276 if (cpos < cend) 2277 eofflag = 0; 2278 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2279 *tl++ = newnfs_false; 2280 if (eofflag) 2281 *tl = newnfs_true; 2282 else 2283 *tl = newnfs_false; 2284 free(rbuf, M_TEMP); 2285 free(cookies, M_TEMP); 2286 2287 out: 2288 NFSEXITCODE2(0, nd); 2289 return (0); 2290 nfsmout: 2291 vput(vp); 2292 NFSEXITCODE2(error, nd); 2293 return (error); 2294 } 2295 2296 /* 2297 * Readdirplus for V3 and Readdir for V4. 2298 */ 2299 int 2300 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, 2301 struct vnode *vp, struct nfsexstuff *exp) 2302 { 2303 struct dirent *dp; 2304 u_int32_t *tl; 2305 int dirlen; 2306 char *cpos, *cend, *rbuf; 2307 struct vnode *nvp; 2308 fhandle_t nfh; 2309 struct nfsvattr nva, at, *nvap = &nva; 2310 struct mbuf *mb0, *mb1; 2311 struct nfsreferral *refp; 2312 int nlen, r, error = 0, getret = 1, usevget = 1; 2313 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; 2314 caddr_t bpos0, bpos1; 2315 u_int64_t off, toff, verf __unused; 2316 uint64_t *cookies = NULL, *cookiep; 2317 nfsattrbit_t attrbits, rderrbits, savbits; 2318 struct uio io; 2319 struct iovec iv; 2320 struct componentname cn; 2321 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; 2322 struct mount *mp, *new_mp; 2323 uint64_t mounted_on_fileno; 2324 struct thread *p = curthread; 2325 int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1; 2326 2327 if (nd->nd_repstat) { 2328 nfsrv_postopattr(nd, getret, &at); 2329 goto out; 2330 } 2331 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2332 off = fxdr_hyper(tl); 2333 toff = off; 2334 tl += 2; 2335 verf = fxdr_hyper(tl); 2336 tl += 2; 2337 siz = fxdr_unsigned(int, *tl++); 2338 cnt = fxdr_unsigned(int, *tl); 2339 2340 /* 2341 * Use the server's maximum data transfer size as the upper bound 2342 * on reply datalen. 2343 */ 2344 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2345 cnt = NFS_SRVMAXDATA(nd); 2346 2347 /* 2348 * siz is a "hint" of how much directory information (name, fileid, 2349 * cookie) should be in the reply. At least one client "hints" 0, 2350 * so I set it to cnt for that case. I also round it up to the 2351 * next multiple of DIRBLKSIZ. 2352 * Since the size of a Readdirplus directory entry reply will always 2353 * be greater than a directory entry returned by VOP_READDIR(), it 2354 * does not make sense to read more than NFS_SRVMAXDATA() via 2355 * VOP_READDIR(). 2356 */ 2357 if (siz <= 0) 2358 siz = cnt; 2359 else if (siz > NFS_SRVMAXDATA(nd)) 2360 siz = NFS_SRVMAXDATA(nd); 2361 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2362 2363 if (nd->nd_flag & ND_NFSV4) { 2364 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 2365 if (error) 2366 goto nfsmout; 2367 NFSSET_ATTRBIT(&savbits, &attrbits); 2368 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); 2369 NFSZERO_ATTRBIT(&rderrbits); 2370 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); 2371 } else { 2372 NFSZERO_ATTRBIT(&attrbits); 2373 } 2374 fullsiz = siz; 2375 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2376 #if 0 2377 if (!nd->nd_repstat) { 2378 if (off && verf != at.na_filerev) { 2379 /* 2380 * va_filerev is not sufficient as a cookie verifier, 2381 * since it is not supposed to change when entries are 2382 * removed/added unless that offset cookies returned to 2383 * the client are no longer valid. 2384 */ 2385 if (nd->nd_flag & ND_NFSV4) { 2386 nd->nd_repstat = NFSERR_NOTSAME; 2387 } else { 2388 nd->nd_repstat = NFSERR_BAD_COOKIE; 2389 } 2390 } 2391 } 2392 #endif 2393 if (!nd->nd_repstat && vp->v_type != VDIR) 2394 nd->nd_repstat = NFSERR_NOTDIR; 2395 if (!nd->nd_repstat && cnt == 0) 2396 nd->nd_repstat = NFSERR_TOOSMALL; 2397 if (!nd->nd_repstat) 2398 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2399 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2400 NFSACCCHK_VPISLOCKED, NULL); 2401 if (nd->nd_repstat) { 2402 vput(vp); 2403 if (nd->nd_flag & ND_NFSV3) 2404 nfsrv_postopattr(nd, getret, &at); 2405 goto out; 2406 } 2407 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2408 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; 2409 2410 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2411 again: 2412 eofflag = 0; 2413 if (cookies) { 2414 free(cookies, M_TEMP); 2415 cookies = NULL; 2416 } 2417 2418 iv.iov_base = rbuf; 2419 iv.iov_len = siz; 2420 io.uio_iov = &iv; 2421 io.uio_iovcnt = 1; 2422 io.uio_offset = (off_t)off; 2423 io.uio_resid = siz; 2424 io.uio_segflg = UIO_SYSSPACE; 2425 io.uio_rw = UIO_READ; 2426 io.uio_td = NULL; 2427 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2428 &cookies); 2429 off = (u_int64_t)io.uio_offset; 2430 if (io.uio_resid) 2431 siz -= io.uio_resid; 2432 2433 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2434 2435 if (!cookies && !nd->nd_repstat) 2436 nd->nd_repstat = NFSERR_PERM; 2437 if (!nd->nd_repstat) 2438 nd->nd_repstat = getret; 2439 if (nd->nd_repstat) { 2440 vput(vp); 2441 if (cookies) 2442 free(cookies, M_TEMP); 2443 free(rbuf, M_TEMP); 2444 if (nd->nd_flag & ND_NFSV3) 2445 nfsrv_postopattr(nd, getret, &at); 2446 goto out; 2447 } 2448 /* 2449 * If nothing read, return eof 2450 * rpc reply 2451 */ 2452 if (siz == 0) { 2453 vput(vp); 2454 if (nd->nd_flag & ND_NFSV3) 2455 nfsrv_postopattr(nd, getret, &at); 2456 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2457 txdr_hyper(at.na_filerev, tl); 2458 tl += 2; 2459 *tl++ = newnfs_false; 2460 *tl = newnfs_true; 2461 free(cookies, M_TEMP); 2462 free(rbuf, M_TEMP); 2463 goto out; 2464 } 2465 2466 /* 2467 * Check for degenerate cases of nothing useful read. 2468 * If so go try again 2469 */ 2470 cpos = rbuf; 2471 cend = rbuf + siz; 2472 dp = (struct dirent *)cpos; 2473 cookiep = cookies; 2474 2475 /* 2476 * For some reason FreeBSD's ufs_readdir() chooses to back the 2477 * directory offset up to a block boundary, so it is necessary to 2478 * skip over the records that precede the requested offset. This 2479 * requires the assumption that file offset cookies monotonically 2480 * increase. 2481 */ 2482 while (cpos < cend && ncookies > 0 && 2483 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2484 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || 2485 ((nd->nd_flag & ND_NFSV4) && 2486 ((dp->d_namlen == 1 && dp->d_name[0] == '.') || 2487 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { 2488 cpos += dp->d_reclen; 2489 dp = (struct dirent *)cpos; 2490 cookiep++; 2491 ncookies--; 2492 } 2493 if (cpos >= cend || ncookies == 0) { 2494 siz = fullsiz; 2495 toff = off; 2496 goto again; 2497 } 2498 2499 /* 2500 * Busy the file system so that the mount point won't go away 2501 * and, as such, VFS_VGET() can be used safely. 2502 */ 2503 mp = vp->v_mount; 2504 vfs_ref(mp); 2505 NFSVOPUNLOCK(vp); 2506 nd->nd_repstat = vfs_busy(mp, 0); 2507 vfs_rel(mp); 2508 if (nd->nd_repstat != 0) { 2509 vrele(vp); 2510 free(cookies, M_TEMP); 2511 free(rbuf, M_TEMP); 2512 if (nd->nd_flag & ND_NFSV3) 2513 nfsrv_postopattr(nd, getret, &at); 2514 goto out; 2515 } 2516 2517 /* 2518 * Check to see if entries in this directory can be safely acquired 2519 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. 2520 * ZFS snapshot directories need VOP_LOOKUP(), so that any 2521 * automount of the snapshot directory that is required will 2522 * be done. 2523 * This needs to be done here for NFSv4, since NFSv4 never does 2524 * a VFS_VGET() for "." or "..". 2525 */ 2526 if (is_zfs == 1) { 2527 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); 2528 if (r == EOPNOTSUPP) { 2529 usevget = 0; 2530 cn.cn_nameiop = LOOKUP; 2531 cn.cn_lkflags = LK_SHARED | LK_RETRY; 2532 cn.cn_cred = nd->nd_cred; 2533 } else if (r == 0) 2534 vput(nvp); 2535 } 2536 2537 /* 2538 * If the reply is likely to exceed MCLBYTES and the reply will 2539 * not be saved, use ext_pgs mbufs for TLS. 2540 * It is difficult to predict how large each entry will be and 2541 * how many entries have been read, so just assume the directory 2542 * entries grow by a factor of 4 when attributes are included. 2543 * For NFSv4.0, we do not know for sure if the reply will 2544 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2545 */ 2546 if (cnt > MCLBYTES && siz > MCLBYTES / 4 && 2547 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2548 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2549 nd->nd_flag |= ND_EXTPG; 2550 2551 /* 2552 * Save this position, in case there is an error before one entry 2553 * is created. 2554 */ 2555 mb0 = nd->nd_mb; 2556 bpos0 = nd->nd_bpos; 2557 bextpg0 = nd->nd_bextpg; 2558 bextpgsiz0 = nd->nd_bextpgsiz; 2559 2560 /* 2561 * Fill in the first part of the reply. 2562 * dirlen is the reply length in bytes and cannot exceed cnt. 2563 * (Include the two booleans at the end of the reply in dirlen now, 2564 * so we recognize when we have exceeded cnt.) 2565 */ 2566 if (nd->nd_flag & ND_NFSV3) { 2567 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2568 nfsrv_postopattr(nd, getret, &at); 2569 } else { 2570 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; 2571 } 2572 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); 2573 txdr_hyper(at.na_filerev, tl); 2574 2575 /* 2576 * Save this position, in case there is an empty reply needed. 2577 */ 2578 mb1 = nd->nd_mb; 2579 bpos1 = nd->nd_bpos; 2580 bextpg1 = nd->nd_bextpg; 2581 bextpgsiz1 = nd->nd_bextpgsiz; 2582 2583 /* Loop through the records and build reply */ 2584 entrycnt = 0; 2585 while (cpos < cend && ncookies > 0 && dirlen < cnt) { 2586 nlen = dp->d_namlen; 2587 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2588 nlen <= NFS_MAXNAMLEN && 2589 ((nd->nd_flag & ND_NFSV3) || nlen > 2 || 2590 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) 2591 || (nlen == 1 && dp->d_name[0] != '.'))) { 2592 /* 2593 * Save the current position in the reply, in case 2594 * this entry exceeds cnt. 2595 */ 2596 mb1 = nd->nd_mb; 2597 bpos1 = nd->nd_bpos; 2598 bextpg1 = nd->nd_bextpg; 2599 bextpgsiz1 = nd->nd_bextpgsiz; 2600 2601 /* 2602 * For readdir_and_lookup get the vnode using 2603 * the file number. 2604 */ 2605 nvp = NULL; 2606 refp = NULL; 2607 r = 0; 2608 at_root = 0; 2609 needs_unbusy = 0; 2610 new_mp = mp; 2611 mounted_on_fileno = (uint64_t)dp->d_fileno; 2612 if ((nd->nd_flag & ND_NFSV3) || 2613 NFSNONZERO_ATTRBIT(&savbits)) { 2614 if (nd->nd_flag & ND_NFSV4) 2615 refp = nfsv4root_getreferral(NULL, 2616 vp, dp->d_fileno); 2617 if (refp == NULL) { 2618 if (usevget) 2619 r = VFS_VGET(mp, dp->d_fileno, 2620 LK_SHARED, &nvp); 2621 else 2622 r = EOPNOTSUPP; 2623 if (r == EOPNOTSUPP) { 2624 if (usevget) { 2625 usevget = 0; 2626 cn.cn_nameiop = LOOKUP; 2627 cn.cn_lkflags = 2628 LK_SHARED | 2629 LK_RETRY; 2630 cn.cn_cred = 2631 nd->nd_cred; 2632 } 2633 cn.cn_nameptr = dp->d_name; 2634 cn.cn_namelen = nlen; 2635 cn.cn_flags = ISLASTCN | 2636 NOFOLLOW | LOCKLEAF; 2637 if (nlen == 2 && 2638 dp->d_name[0] == '.' && 2639 dp->d_name[1] == '.') 2640 cn.cn_flags |= 2641 ISDOTDOT; 2642 if (NFSVOPLOCK(vp, LK_SHARED) 2643 != 0) { 2644 nd->nd_repstat = EPERM; 2645 break; 2646 } 2647 if ((vp->v_vflag & VV_ROOT) != 0 2648 && (cn.cn_flags & ISDOTDOT) 2649 != 0) { 2650 vref(vp); 2651 nvp = vp; 2652 r = 0; 2653 } else { 2654 r = VOP_LOOKUP(vp, &nvp, 2655 &cn); 2656 if (vp != nvp) 2657 NFSVOPUNLOCK(vp); 2658 } 2659 } 2660 2661 /* 2662 * For NFSv4, check to see if nvp is 2663 * a mount point and get the mount 2664 * point vnode, as required. 2665 */ 2666 if (r == 0 && 2667 nfsrv_enable_crossmntpt != 0 && 2668 (nd->nd_flag & ND_NFSV4) != 0 && 2669 nvp->v_type == VDIR && 2670 nvp->v_mountedhere != NULL) { 2671 new_mp = nvp->v_mountedhere; 2672 r = vfs_busy(new_mp, 0); 2673 vput(nvp); 2674 nvp = NULL; 2675 if (r == 0) { 2676 r = VFS_ROOT(new_mp, 2677 LK_SHARED, &nvp); 2678 needs_unbusy = 1; 2679 if (r == 0) 2680 at_root = 1; 2681 } 2682 } 2683 } 2684 2685 /* 2686 * If we failed to look up the entry, then it 2687 * has become invalid, most likely removed. 2688 */ 2689 if (r != 0) { 2690 if (needs_unbusy) 2691 vfs_unbusy(new_mp); 2692 goto invalid; 2693 } 2694 KASSERT(refp != NULL || nvp != NULL, 2695 ("%s: undetected lookup error", __func__)); 2696 2697 if (refp == NULL && 2698 ((nd->nd_flag & ND_NFSV3) || 2699 NFSNONZERO_ATTRBIT(&attrbits))) { 2700 r = nfsvno_getfh(nvp, &nfh, p); 2701 if (!r) 2702 r = nfsvno_getattr(nvp, nvap, nd, p, 2703 1, &attrbits); 2704 if (r == 0 && is_zfs == 1 && 2705 nfsrv_enable_crossmntpt != 0 && 2706 (nd->nd_flag & ND_NFSV4) != 0 && 2707 nvp->v_type == VDIR && 2708 vp->v_mount != nvp->v_mount) { 2709 /* 2710 * For a ZFS snapshot, there is a 2711 * pseudo mount that does not set 2712 * v_mountedhere, so it needs to 2713 * be detected via a different 2714 * mount structure. 2715 */ 2716 at_root = 1; 2717 if (new_mp == mp) 2718 new_mp = nvp->v_mount; 2719 } 2720 } 2721 2722 /* 2723 * If we failed to get attributes of the entry, 2724 * then just skip it for NFSv3 (the traditional 2725 * behavior in the old NFS server). 2726 * For NFSv4 the behavior is controlled by 2727 * RDATTRERROR: we either ignore the error or 2728 * fail the request. 2729 * The exception is EOPNOTSUPP, which can be 2730 * returned by nfsvno_getfh() for certain 2731 * file systems, such as devfs. This indicates 2732 * that the file system cannot be exported, 2733 * so just skip over the entry. 2734 * Note that RDATTRERROR is never set for NFSv3. 2735 */ 2736 if (r != 0) { 2737 if (!NFSISSET_ATTRBIT(&attrbits, 2738 NFSATTRBIT_RDATTRERROR) || 2739 r == EOPNOTSUPP) { 2740 vput(nvp); 2741 if (needs_unbusy != 0) 2742 vfs_unbusy(new_mp); 2743 if ((nd->nd_flag & ND_NFSV3) || 2744 r == EOPNOTSUPP) 2745 goto invalid; 2746 nd->nd_repstat = r; 2747 break; 2748 } 2749 } 2750 } 2751 2752 /* 2753 * Build the directory record xdr 2754 */ 2755 if (nd->nd_flag & ND_NFSV3) { 2756 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2757 *tl++ = newnfs_true; 2758 *tl++ = 0; 2759 *tl = txdr_unsigned(dp->d_fileno); 2760 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2761 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2762 txdr_hyper(*cookiep, tl); 2763 nfsrv_postopattr(nd, 0, nvap); 2764 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1); 2765 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); 2766 if (nvp != NULL) 2767 vput(nvp); 2768 } else { 2769 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2770 *tl++ = newnfs_true; 2771 txdr_hyper(*cookiep, tl); 2772 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2773 if (nvp != NULL) { 2774 supports_nfsv4acls = 2775 nfs_supportsnfsv4acls(nvp); 2776 NFSVOPUNLOCK(nvp); 2777 } else 2778 supports_nfsv4acls = 0; 2779 if (refp != NULL) { 2780 dirlen += nfsrv_putreferralattr(nd, 2781 &savbits, refp, 0, 2782 &nd->nd_repstat); 2783 if (nd->nd_repstat) { 2784 if (nvp != NULL) 2785 vrele(nvp); 2786 if (needs_unbusy != 0) 2787 vfs_unbusy(new_mp); 2788 break; 2789 } 2790 } else if (r) { 2791 dirlen += nfsvno_fillattr(nd, new_mp, 2792 nvp, nvap, &nfh, r, &rderrbits, 2793 nd->nd_cred, p, isdgram, 0, 2794 supports_nfsv4acls, at_root, 2795 mounted_on_fileno); 2796 } else { 2797 dirlen += nfsvno_fillattr(nd, new_mp, 2798 nvp, nvap, &nfh, r, &attrbits, 2799 nd->nd_cred, p, isdgram, 0, 2800 supports_nfsv4acls, at_root, 2801 mounted_on_fileno); 2802 } 2803 if (nvp != NULL) 2804 vrele(nvp); 2805 dirlen += (3 * NFSX_UNSIGNED); 2806 } 2807 if (needs_unbusy != 0) 2808 vfs_unbusy(new_mp); 2809 if (dirlen <= cnt) 2810 entrycnt++; 2811 } 2812 invalid: 2813 cpos += dp->d_reclen; 2814 dp = (struct dirent *)cpos; 2815 cookiep++; 2816 ncookies--; 2817 } 2818 vrele(vp); 2819 vfs_unbusy(mp); 2820 2821 /* 2822 * If dirlen > cnt, we must strip off the last entry. If that 2823 * results in an empty reply, report NFSERR_TOOSMALL. 2824 */ 2825 if (dirlen > cnt || nd->nd_repstat) { 2826 if (!nd->nd_repstat && entrycnt == 0) 2827 nd->nd_repstat = NFSERR_TOOSMALL; 2828 if (nd->nd_repstat) { 2829 nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); 2830 if (nd->nd_flag & ND_NFSV3) 2831 nfsrv_postopattr(nd, getret, &at); 2832 } else 2833 nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); 2834 eofflag = 0; 2835 } else if (cpos < cend) 2836 eofflag = 0; 2837 if (!nd->nd_repstat) { 2838 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2839 *tl++ = newnfs_false; 2840 if (eofflag) 2841 *tl = newnfs_true; 2842 else 2843 *tl = newnfs_false; 2844 } 2845 free(cookies, M_TEMP); 2846 free(rbuf, M_TEMP); 2847 2848 out: 2849 NFSEXITCODE2(0, nd); 2850 return (0); 2851 nfsmout: 2852 vput(vp); 2853 NFSEXITCODE2(error, nd); 2854 return (error); 2855 } 2856 2857 /* 2858 * Get the settable attributes out of the mbuf list. 2859 * (Return 0 or EBADRPC) 2860 */ 2861 int 2862 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2863 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2864 { 2865 u_int32_t *tl; 2866 struct nfsv2_sattr *sp; 2867 int error = 0, toclient = 0; 2868 2869 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 2870 case ND_NFSV2: 2871 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2872 /* 2873 * Some old clients didn't fill in the high order 16bits. 2874 * --> check the low order 2 bytes for 0xffff 2875 */ 2876 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 2877 nvap->na_mode = nfstov_mode(sp->sa_mode); 2878 if (sp->sa_uid != newnfs_xdrneg1) 2879 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 2880 if (sp->sa_gid != newnfs_xdrneg1) 2881 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 2882 if (sp->sa_size != newnfs_xdrneg1) 2883 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 2884 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 2885 #ifdef notyet 2886 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 2887 #else 2888 nvap->na_atime.tv_sec = 2889 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 2890 nvap->na_atime.tv_nsec = 0; 2891 #endif 2892 } 2893 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 2894 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 2895 break; 2896 case ND_NFSV3: 2897 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2898 if (*tl == newnfs_true) { 2899 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2900 nvap->na_mode = nfstov_mode(*tl); 2901 } 2902 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2903 if (*tl == newnfs_true) { 2904 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2905 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 2906 } 2907 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2908 if (*tl == newnfs_true) { 2909 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2910 nvap->na_gid = fxdr_unsigned(gid_t, *tl); 2911 } 2912 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2913 if (*tl == newnfs_true) { 2914 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2915 nvap->na_size = fxdr_hyper(tl); 2916 } 2917 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2918 switch (fxdr_unsigned(int, *tl)) { 2919 case NFSV3SATTRTIME_TOCLIENT: 2920 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2921 fxdr_nfsv3time(tl, &nvap->na_atime); 2922 toclient = 1; 2923 break; 2924 case NFSV3SATTRTIME_TOSERVER: 2925 vfs_timestamp(&nvap->na_atime); 2926 nvap->na_vaflags |= VA_UTIMES_NULL; 2927 break; 2928 } 2929 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2930 switch (fxdr_unsigned(int, *tl)) { 2931 case NFSV3SATTRTIME_TOCLIENT: 2932 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2933 fxdr_nfsv3time(tl, &nvap->na_mtime); 2934 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2935 break; 2936 case NFSV3SATTRTIME_TOSERVER: 2937 vfs_timestamp(&nvap->na_mtime); 2938 if (!toclient) 2939 nvap->na_vaflags |= VA_UTIMES_NULL; 2940 break; 2941 } 2942 break; 2943 case ND_NFSV4: 2944 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); 2945 } 2946 nfsmout: 2947 NFSEXITCODE2(error, nd); 2948 return (error); 2949 } 2950 2951 /* 2952 * Handle the setable attributes for V4. 2953 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 2954 */ 2955 int 2956 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2957 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2958 { 2959 u_int32_t *tl; 2960 int attrsum = 0; 2961 int i, j; 2962 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 2963 int moderet, toclient = 0; 2964 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 2965 uid_t uid; 2966 gid_t gid; 2967 u_short mode, mask; /* Same type as va_mode. */ 2968 struct vattr va; 2969 2970 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 2971 if (error) 2972 goto nfsmout; 2973 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2974 attrsize = fxdr_unsigned(int, *tl); 2975 2976 /* 2977 * Loop around getting the setable attributes. If an unsupported 2978 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 2979 */ 2980 if (retnotsup) { 2981 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2982 bitpos = NFSATTRBIT_MAX; 2983 } else { 2984 bitpos = 0; 2985 } 2986 moderet = 0; 2987 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 2988 if (attrsum > attrsize) { 2989 error = NFSERR_BADXDR; 2990 goto nfsmout; 2991 } 2992 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 2993 switch (bitpos) { 2994 case NFSATTRBIT_SIZE: 2995 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 2996 if (vp != NULL && vp->v_type != VREG) { 2997 error = (vp->v_type == VDIR) ? NFSERR_ISDIR : 2998 NFSERR_INVAL; 2999 goto nfsmout; 3000 } 3001 nvap->na_size = fxdr_hyper(tl); 3002 attrsum += NFSX_HYPER; 3003 break; 3004 case NFSATTRBIT_ACL: 3005 error = nfsrv_dissectacl(nd, aclp, true, &aceerr, 3006 &aclsize, p); 3007 if (error) 3008 goto nfsmout; 3009 if (aceerr && !nd->nd_repstat) 3010 nd->nd_repstat = aceerr; 3011 attrsum += aclsize; 3012 break; 3013 case NFSATTRBIT_ARCHIVE: 3014 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3015 if (!nd->nd_repstat) 3016 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3017 attrsum += NFSX_UNSIGNED; 3018 break; 3019 case NFSATTRBIT_HIDDEN: 3020 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3021 if (!nd->nd_repstat) 3022 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3023 attrsum += NFSX_UNSIGNED; 3024 break; 3025 case NFSATTRBIT_MIMETYPE: 3026 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3027 i = fxdr_unsigned(int, *tl); 3028 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 3029 if (error) 3030 goto nfsmout; 3031 if (!nd->nd_repstat) 3032 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3033 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 3034 break; 3035 case NFSATTRBIT_MODE: 3036 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ 3037 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3038 nvap->na_mode = nfstov_mode(*tl); 3039 attrsum += NFSX_UNSIGNED; 3040 break; 3041 case NFSATTRBIT_OWNER: 3042 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3043 j = fxdr_unsigned(int, *tl); 3044 if (j < 0) { 3045 error = NFSERR_BADXDR; 3046 goto nfsmout; 3047 } 3048 if (j > NFSV4_SMALLSTR) 3049 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3050 else 3051 cp = namestr; 3052 error = nfsrv_mtostr(nd, cp, j); 3053 if (error) { 3054 if (j > NFSV4_SMALLSTR) 3055 free(cp, M_NFSSTRING); 3056 goto nfsmout; 3057 } 3058 if (!nd->nd_repstat) { 3059 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, 3060 &uid); 3061 if (!nd->nd_repstat) 3062 nvap->na_uid = uid; 3063 } 3064 if (j > NFSV4_SMALLSTR) 3065 free(cp, M_NFSSTRING); 3066 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3067 break; 3068 case NFSATTRBIT_OWNERGROUP: 3069 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3070 j = fxdr_unsigned(int, *tl); 3071 if (j < 0) { 3072 error = NFSERR_BADXDR; 3073 goto nfsmout; 3074 } 3075 if (j > NFSV4_SMALLSTR) 3076 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3077 else 3078 cp = namestr; 3079 error = nfsrv_mtostr(nd, cp, j); 3080 if (error) { 3081 if (j > NFSV4_SMALLSTR) 3082 free(cp, M_NFSSTRING); 3083 goto nfsmout; 3084 } 3085 if (!nd->nd_repstat) { 3086 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, 3087 &gid); 3088 if (!nd->nd_repstat) 3089 nvap->na_gid = gid; 3090 } 3091 if (j > NFSV4_SMALLSTR) 3092 free(cp, M_NFSSTRING); 3093 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3094 break; 3095 case NFSATTRBIT_SYSTEM: 3096 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3097 if (!nd->nd_repstat) 3098 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3099 attrsum += NFSX_UNSIGNED; 3100 break; 3101 case NFSATTRBIT_TIMEACCESSSET: 3102 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3103 attrsum += NFSX_UNSIGNED; 3104 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3105 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3106 fxdr_nfsv4time(tl, &nvap->na_atime); 3107 toclient = 1; 3108 attrsum += NFSX_V4TIME; 3109 } else { 3110 vfs_timestamp(&nvap->na_atime); 3111 nvap->na_vaflags |= VA_UTIMES_NULL; 3112 } 3113 break; 3114 case NFSATTRBIT_TIMEBACKUP: 3115 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3116 if (!nd->nd_repstat) 3117 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3118 attrsum += NFSX_V4TIME; 3119 break; 3120 case NFSATTRBIT_TIMECREATE: 3121 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3122 fxdr_nfsv4time(tl, &nvap->na_btime); 3123 attrsum += NFSX_V4TIME; 3124 break; 3125 case NFSATTRBIT_TIMEMODIFYSET: 3126 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3127 attrsum += NFSX_UNSIGNED; 3128 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3129 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3130 fxdr_nfsv4time(tl, &nvap->na_mtime); 3131 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3132 attrsum += NFSX_V4TIME; 3133 } else { 3134 vfs_timestamp(&nvap->na_mtime); 3135 if (!toclient) 3136 nvap->na_vaflags |= VA_UTIMES_NULL; 3137 } 3138 break; 3139 case NFSATTRBIT_MODESETMASKED: 3140 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3141 mode = fxdr_unsigned(u_short, *tl++); 3142 mask = fxdr_unsigned(u_short, *tl); 3143 /* 3144 * vp == NULL implies an Open/Create operation. 3145 * This attribute can only be used for Setattr and 3146 * only for NFSv4.1 or higher. 3147 * If moderet != 0, a mode attribute has also been 3148 * specified and this attribute cannot be done in the 3149 * same Setattr operation. 3150 */ 3151 if ((nd->nd_flag & ND_NFSV41) == 0) 3152 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3153 else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 || 3154 vp == NULL) 3155 nd->nd_repstat = NFSERR_INVAL; 3156 else if (moderet == 0) 3157 moderet = VOP_GETATTR(vp, &va, nd->nd_cred); 3158 if (moderet == 0) 3159 nvap->na_mode = (mode & mask) | 3160 (va.va_mode & ~mask); 3161 else 3162 nd->nd_repstat = moderet; 3163 attrsum += 2 * NFSX_UNSIGNED; 3164 break; 3165 default: 3166 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3167 /* 3168 * set bitpos so we drop out of the loop. 3169 */ 3170 bitpos = NFSATTRBIT_MAX; 3171 break; 3172 } 3173 } 3174 3175 /* 3176 * some clients pad the attrlist, so we need to skip over the 3177 * padding. 3178 */ 3179 if (attrsum > attrsize) { 3180 error = NFSERR_BADXDR; 3181 } else { 3182 attrsize = NFSM_RNDUP(attrsize); 3183 if (attrsum < attrsize) 3184 error = nfsm_advance(nd, attrsize - attrsum, -1); 3185 } 3186 nfsmout: 3187 NFSEXITCODE2(error, nd); 3188 return (error); 3189 } 3190 3191 /* 3192 * Check/setup export credentials. 3193 */ 3194 int 3195 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 3196 struct ucred *credanon, bool testsec) 3197 { 3198 int error; 3199 3200 /* 3201 * Check/setup credentials. 3202 */ 3203 if (nd->nd_flag & ND_GSS) 3204 exp->nes_exflag &= ~MNT_EXPORTANON; 3205 3206 /* 3207 * Check to see if the operation is allowed for this security flavor. 3208 */ 3209 error = 0; 3210 if (testsec) { 3211 error = nfsvno_testexp(nd, exp); 3212 if (error != 0) 3213 goto out; 3214 } 3215 3216 /* 3217 * Check to see if the file system is exported V4 only. 3218 */ 3219 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 3220 error = NFSERR_PROGNOTV4; 3221 goto out; 3222 } 3223 3224 /* 3225 * Now, map the user credentials. 3226 * (Note that ND_AUTHNONE will only be set for an NFSv3 3227 * Fsinfo RPC. If set for anything else, this code might need 3228 * to change.) 3229 */ 3230 if (NFSVNO_EXPORTED(exp)) { 3231 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || 3232 NFSVNO_EXPORTANON(exp) || 3233 (nd->nd_flag & ND_AUTHNONE) != 0) { 3234 nd->nd_cred->cr_uid = credanon->cr_uid; 3235 nd->nd_cred->cr_gid = credanon->cr_gid; 3236 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 3237 credanon->cr_groups); 3238 } else if ((nd->nd_flag & ND_GSS) == 0) { 3239 /* 3240 * If using AUTH_SYS, call nfsrv_getgrpscred() to see 3241 * if there is a replacement credential with a group 3242 * list set up by "nfsuserd -manage-gids". 3243 * If there is no replacement, nfsrv_getgrpscred() 3244 * simply returns its argument. 3245 */ 3246 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); 3247 } 3248 } 3249 3250 out: 3251 NFSEXITCODE2(error, nd); 3252 return (error); 3253 } 3254 3255 /* 3256 * Check exports. 3257 */ 3258 int 3259 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 3260 struct ucred **credp) 3261 { 3262 int error; 3263 3264 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3265 &exp->nes_numsecflavor, exp->nes_secflavors); 3266 if (error) { 3267 if (NFSD_VNET(nfs_rootfhset)) { 3268 exp->nes_exflag = 0; 3269 exp->nes_numsecflavor = 0; 3270 error = 0; 3271 } 3272 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3273 MAXSECFLAVORS) { 3274 printf("nfsvno_checkexp: numsecflavors out of range\n"); 3275 exp->nes_numsecflavor = 0; 3276 error = EACCES; 3277 } 3278 NFSEXITCODE(error); 3279 return (error); 3280 } 3281 3282 /* 3283 * Get a vnode for a file handle and export stuff. 3284 */ 3285 int 3286 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 3287 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 3288 struct ucred **credp) 3289 { 3290 int error; 3291 3292 *credp = NULL; 3293 exp->nes_numsecflavor = 0; 3294 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); 3295 if (error != 0) 3296 /* Make sure the server replies ESTALE to the client. */ 3297 error = ESTALE; 3298 if (nam && !error) { 3299 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3300 &exp->nes_numsecflavor, exp->nes_secflavors); 3301 if (error) { 3302 if (NFSD_VNET(nfs_rootfhset)) { 3303 exp->nes_exflag = 0; 3304 exp->nes_numsecflavor = 0; 3305 error = 0; 3306 } else { 3307 vput(*vpp); 3308 } 3309 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3310 MAXSECFLAVORS) { 3311 printf("nfsvno_fhtovp: numsecflavors out of range\n"); 3312 exp->nes_numsecflavor = 0; 3313 error = EACCES; 3314 vput(*vpp); 3315 } 3316 } 3317 NFSEXITCODE(error); 3318 return (error); 3319 } 3320 3321 /* 3322 * nfsd_fhtovp() - convert a fh to a vnode ptr 3323 * - look up fsid in mount list (if not found ret error) 3324 * - get vp and export rights by calling nfsvno_fhtovp() 3325 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 3326 * for AUTH_SYS 3327 * - if mpp != NULL, return the mount point so that it can 3328 * be used for vn_finished_write() by the caller 3329 */ 3330 void 3331 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 3332 struct vnode **vpp, struct nfsexstuff *exp, 3333 struct mount **mpp, int startwrite, int nextop) 3334 { 3335 struct mount *mp, *mpw; 3336 struct ucred *credanon; 3337 fhandle_t *fhp; 3338 int error; 3339 3340 if (mpp != NULL) 3341 *mpp = NULL; 3342 *vpp = NULL; 3343 fhp = (fhandle_t *)nfp->nfsrvfh_data; 3344 mp = vfs_busyfs(&fhp->fh_fsid); 3345 if (mp == NULL) { 3346 nd->nd_repstat = ESTALE; 3347 goto out; 3348 } 3349 3350 if (startwrite) { 3351 mpw = mp; 3352 error = vn_start_write(NULL, &mpw, V_WAIT); 3353 if (error != 0) { 3354 mpw = NULL; 3355 vfs_unbusy(mp); 3356 nd->nd_repstat = ESTALE; 3357 goto out; 3358 } 3359 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) 3360 lktype = LK_EXCLUSIVE; 3361 } else 3362 mpw = NULL; 3363 3364 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 3365 &credanon); 3366 vfs_unbusy(mp); 3367 3368 /* 3369 * For NFSv4 without a pseudo root fs, unexported file handles 3370 * can be returned, so that Lookup works everywhere. 3371 */ 3372 if (!nd->nd_repstat && exp->nes_exflag == 0 && 3373 !(nd->nd_flag & ND_NFSV4)) { 3374 vput(*vpp); 3375 *vpp = NULL; 3376 nd->nd_repstat = EACCES; 3377 } 3378 3379 /* 3380 * Personally, I've never seen any point in requiring a 3381 * reserved port#, since only in the rare case where the 3382 * clients are all boxes with secure system privileges, 3383 * does it provide any enhanced security, but... some people 3384 * believe it to be useful and keep putting this code back in. 3385 * (There is also some "security checker" out there that 3386 * complains if the nfs server doesn't enforce this.) 3387 * However, note the following: 3388 * RFC3530 (NFSv4) specifies that a reserved port# not be 3389 * required. 3390 * RFC2623 recommends that, if a reserved port# is checked for, 3391 * that there be a way to turn that off--> ifdef'd. 3392 */ 3393 #ifdef NFS_REQRSVPORT 3394 if (!nd->nd_repstat) { 3395 struct sockaddr_in *saddr; 3396 struct sockaddr_in6 *saddr6; 3397 3398 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 3399 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 3400 if (!(nd->nd_flag & ND_NFSV4) && 3401 ((saddr->sin_family == AF_INET && 3402 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 3403 (saddr6->sin6_family == AF_INET6 && 3404 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 3405 vput(*vpp); 3406 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3407 } 3408 } 3409 #endif /* NFS_REQRSVPORT */ 3410 3411 /* 3412 * Check/setup credentials. 3413 */ 3414 if (!nd->nd_repstat) { 3415 nd->nd_saveduid = nd->nd_cred->cr_uid; 3416 nd->nd_repstat = nfsd_excred(nd, exp, credanon, 3417 nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type)); 3418 if (nd->nd_repstat) 3419 vput(*vpp); 3420 } 3421 if (credanon != NULL) 3422 crfree(credanon); 3423 if (nd->nd_repstat) { 3424 vn_finished_write(mpw); 3425 *vpp = NULL; 3426 } else if (mpp != NULL) { 3427 *mpp = mpw; 3428 } 3429 3430 out: 3431 NFSEXITCODE2(0, nd); 3432 } 3433 3434 /* 3435 * glue for fp. 3436 */ 3437 static int 3438 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 3439 { 3440 struct filedesc *fdp; 3441 struct file *fp; 3442 int error = 0; 3443 3444 fdp = p->td_proc->p_fd; 3445 if (fd < 0 || fd >= fdp->fd_nfiles || 3446 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { 3447 error = EBADF; 3448 goto out; 3449 } 3450 *fpp = fp; 3451 3452 out: 3453 NFSEXITCODE(error); 3454 return (error); 3455 } 3456 3457 /* 3458 * Called from nfssvc() to update the exports list. Just call 3459 * vfs_export(). This has to be done, since the v4 root fake fs isn't 3460 * in the mount list. 3461 */ 3462 int 3463 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 3464 { 3465 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 3466 int error = 0; 3467 struct nameidata nd; 3468 fhandle_t fh; 3469 3470 error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export); 3471 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 3472 NFSD_VNET(nfs_rootfhset) = 0; 3473 else if (error == 0) { 3474 if (nfsexargp->fspec == NULL) { 3475 error = EPERM; 3476 goto out; 3477 } 3478 /* 3479 * If fspec != NULL, this is the v4root path. 3480 */ 3481 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec); 3482 if ((error = namei(&nd)) != 0) 3483 goto out; 3484 error = nfsvno_getfh(nd.ni_vp, &fh, p); 3485 vrele(nd.ni_vp); 3486 if (!error) { 3487 NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH; 3488 NFSBCOPY((caddr_t)&fh, 3489 NFSD_VNET(nfs_rootfh).nfsrvfh_data, 3490 sizeof (fhandle_t)); 3491 NFSD_VNET(nfs_rootfhset) = 1; 3492 } 3493 } 3494 3495 out: 3496 NFSEXITCODE(error); 3497 return (error); 3498 } 3499 3500 /* 3501 * This function needs to test to see if the system is near its limit 3502 * for memory allocation via malloc() or mget() and return True iff 3503 * either of these resources are near their limit. 3504 * XXX (For now, this is just a stub.) 3505 */ 3506 int nfsrv_testmalloclimit = 0; 3507 int 3508 nfsrv_mallocmget_limit(void) 3509 { 3510 static int printmesg = 0; 3511 static int testval = 1; 3512 3513 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 3514 if ((printmesg++ % 100) == 0) 3515 printf("nfsd: malloc/mget near limit\n"); 3516 return (1); 3517 } 3518 return (0); 3519 } 3520 3521 /* 3522 * BSD specific initialization of a mount point. 3523 */ 3524 void 3525 nfsd_mntinit(void) 3526 { 3527 3528 NFSD_LOCK(); 3529 if (NFSD_VNET(nfsrv_mntinited)) { 3530 NFSD_UNLOCK(); 3531 return; 3532 } 3533 NFSD_VNET(nfsrv_mntinited) = true; 3534 nfsrvd_init(0); 3535 NFSD_UNLOCK(); 3536 3537 NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP, 3538 M_WAITOK | M_ZERO); 3539 NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 3540 mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF); 3541 lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0); 3542 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist); 3543 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist); 3544 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 3545 TAILQ_INIT(&NFSD_VNET(nfsv4root_opt)); 3546 TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt)); 3547 NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt); 3548 NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt); 3549 NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0; 3550 NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0; 3551 callout_init(&NFSD_VNET(nfsd_callout), 1); 3552 3553 nfsrvd_initcache(); 3554 nfsd_init(); 3555 } 3556 3557 static void 3558 nfsd_timer(void *arg) 3559 { 3560 struct vnet *vnetp; 3561 3562 vnetp = (struct vnet *)arg; 3563 NFSD_CURVNET_SET_QUIET(vnetp); 3564 nfsrv_servertimer(vnetp); 3565 callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer, 3566 arg, 0); 3567 NFSD_CURVNET_RESTORE(); 3568 } 3569 3570 /* 3571 * Get a vnode for a file handle, without checking exports, etc. 3572 */ 3573 struct vnode * 3574 nfsvno_getvp(fhandle_t *fhp) 3575 { 3576 struct mount *mp; 3577 struct vnode *vp; 3578 int error; 3579 3580 mp = vfs_busyfs(&fhp->fh_fsid); 3581 if (mp == NULL) 3582 return (NULL); 3583 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 3584 vfs_unbusy(mp); 3585 if (error) 3586 return (NULL); 3587 return (vp); 3588 } 3589 3590 /* 3591 * Do a local VOP_ADVLOCK(). 3592 */ 3593 int 3594 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 3595 u_int64_t end, struct thread *td) 3596 { 3597 int error = 0; 3598 struct flock fl; 3599 u_int64_t tlen; 3600 3601 if (nfsrv_dolocallocks == 0) 3602 goto out; 3603 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); 3604 3605 fl.l_whence = SEEK_SET; 3606 fl.l_type = ftype; 3607 fl.l_start = (off_t)first; 3608 if (end == NFS64BITSSET) { 3609 fl.l_len = 0; 3610 } else { 3611 tlen = end - first; 3612 fl.l_len = (off_t)tlen; 3613 } 3614 /* 3615 * For FreeBSD8, the l_pid and l_sysid must be set to the same 3616 * values for all calls, so that all locks will be held by the 3617 * nfsd server. (The nfsd server handles conflicts between the 3618 * various clients.) 3619 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 3620 * bytes, so it can't be put in l_sysid. 3621 */ 3622 if (nfsv4_sysid == 0) 3623 nfsv4_sysid = nlm_acquire_next_sysid(); 3624 fl.l_pid = (pid_t)0; 3625 fl.l_sysid = (int)nfsv4_sysid; 3626 3627 if (ftype == F_UNLCK) 3628 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 3629 (F_POSIX | F_REMOTE)); 3630 else 3631 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 3632 (F_POSIX | F_REMOTE)); 3633 3634 out: 3635 NFSEXITCODE(error); 3636 return (error); 3637 } 3638 3639 /* 3640 * Check the nfsv4 root exports. 3641 */ 3642 int 3643 nfsvno_v4rootexport(struct nfsrv_descript *nd) 3644 { 3645 struct ucred *credanon; 3646 int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; 3647 uint64_t exflags; 3648 3649 error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags, 3650 &credanon, &numsecflavor, secflavors); 3651 if (error) { 3652 error = NFSERR_PROGUNAVAIL; 3653 goto out; 3654 } 3655 if (credanon != NULL) 3656 crfree(credanon); 3657 for (i = 0; i < numsecflavor; i++) { 3658 if (secflavors[i] == AUTH_SYS) 3659 nd->nd_flag |= ND_EXAUTHSYS; 3660 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3661 nd->nd_flag |= ND_EXGSS; 3662 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3663 nd->nd_flag |= ND_EXGSSINTEGRITY; 3664 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3665 nd->nd_flag |= ND_EXGSSPRIVACY; 3666 } 3667 3668 /* And set ND_EXxx flags for TLS. */ 3669 if ((exflags & MNT_EXTLS) != 0) { 3670 nd->nd_flag |= ND_EXTLS; 3671 if ((exflags & MNT_EXTLSCERT) != 0) 3672 nd->nd_flag |= ND_EXTLSCERT; 3673 if ((exflags & MNT_EXTLSCERTUSER) != 0) 3674 nd->nd_flag |= ND_EXTLSCERTUSER; 3675 } 3676 3677 out: 3678 NFSEXITCODE(error); 3679 return (error); 3680 } 3681 3682 /* 3683 * Nfs server pseudo system call for the nfsd's 3684 */ 3685 /* 3686 * MPSAFE 3687 */ 3688 static int 3689 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3690 { 3691 struct file *fp; 3692 struct nfsd_addsock_args sockarg; 3693 struct nfsd_nfsd_args nfsdarg; 3694 struct nfsd_nfsd_oargs onfsdarg; 3695 struct nfsd_pnfsd_args pnfsdarg; 3696 struct vnode *vp, *nvp, *curdvp; 3697 struct pnfsdsfile *pf; 3698 struct nfsdevice *ds, *fds; 3699 cap_rights_t rights; 3700 int buflen, error, ret; 3701 char *buf, *cp, *cp2, *cp3; 3702 char fname[PNFS_FILENAME_LEN + 1]; 3703 3704 NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td)); 3705 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3706 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3707 if (error) 3708 goto out; 3709 /* 3710 * Since we don't know what rights might be required, 3711 * pretend that we need them all. It is better to be too 3712 * careful than too reckless. 3713 */ 3714 error = fget(td, sockarg.sock, 3715 cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); 3716 if (error != 0) 3717 goto out; 3718 if (fp->f_type != DTYPE_SOCKET) { 3719 fdrop(fp, td); 3720 error = EPERM; 3721 goto out; 3722 } 3723 error = nfsrvd_addsock(fp); 3724 fdrop(fp, td); 3725 } else if (uap->flag & NFSSVC_NFSDNFSD) { 3726 if (uap->argp == NULL) { 3727 error = EINVAL; 3728 goto out; 3729 } 3730 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { 3731 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); 3732 if (error == 0) { 3733 nfsdarg.principal = onfsdarg.principal; 3734 nfsdarg.minthreads = onfsdarg.minthreads; 3735 nfsdarg.maxthreads = onfsdarg.maxthreads; 3736 nfsdarg.version = 1; 3737 nfsdarg.addr = NULL; 3738 nfsdarg.addrlen = 0; 3739 nfsdarg.dnshost = NULL; 3740 nfsdarg.dnshostlen = 0; 3741 nfsdarg.dspath = NULL; 3742 nfsdarg.dspathlen = 0; 3743 nfsdarg.mdspath = NULL; 3744 nfsdarg.mdspathlen = 0; 3745 nfsdarg.mirrorcnt = 1; 3746 } 3747 } else 3748 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); 3749 if (error) 3750 goto out; 3751 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && 3752 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && 3753 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && 3754 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && 3755 nfsdarg.mirrorcnt >= 1 && 3756 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && 3757 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && 3758 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { 3759 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" 3760 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, 3761 nfsdarg.dspathlen, nfsdarg.dnshostlen, 3762 nfsdarg.mdspathlen, nfsdarg.mirrorcnt); 3763 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); 3764 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); 3765 if (error != 0) { 3766 free(cp, M_TEMP); 3767 goto out; 3768 } 3769 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ 3770 nfsdarg.addr = cp; 3771 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); 3772 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); 3773 if (error != 0) { 3774 free(nfsdarg.addr, M_TEMP); 3775 free(cp, M_TEMP); 3776 goto out; 3777 } 3778 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ 3779 nfsdarg.dnshost = cp; 3780 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); 3781 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); 3782 if (error != 0) { 3783 free(nfsdarg.addr, M_TEMP); 3784 free(nfsdarg.dnshost, M_TEMP); 3785 free(cp, M_TEMP); 3786 goto out; 3787 } 3788 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ 3789 nfsdarg.dspath = cp; 3790 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); 3791 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); 3792 if (error != 0) { 3793 free(nfsdarg.addr, M_TEMP); 3794 free(nfsdarg.dnshost, M_TEMP); 3795 free(nfsdarg.dspath, M_TEMP); 3796 free(cp, M_TEMP); 3797 goto out; 3798 } 3799 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ 3800 nfsdarg.mdspath = cp; 3801 } else { 3802 nfsdarg.addr = NULL; 3803 nfsdarg.addrlen = 0; 3804 nfsdarg.dnshost = NULL; 3805 nfsdarg.dnshostlen = 0; 3806 nfsdarg.dspath = NULL; 3807 nfsdarg.dspathlen = 0; 3808 nfsdarg.mdspath = NULL; 3809 nfsdarg.mdspathlen = 0; 3810 nfsdarg.mirrorcnt = 1; 3811 } 3812 nfsd_timer(NFSD_TD_TO_VNET(td)); 3813 error = nfsrvd_nfsd(td, &nfsdarg); 3814 callout_drain(&NFSD_VNET(nfsd_callout)); 3815 free(nfsdarg.addr, M_TEMP); 3816 free(nfsdarg.dnshost, M_TEMP); 3817 free(nfsdarg.dspath, M_TEMP); 3818 free(nfsdarg.mdspath, M_TEMP); 3819 } else if (uap->flag & NFSSVC_PNFSDS) { 3820 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); 3821 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || 3822 pnfsdarg.op == PNFSDOP_FORCEDELDS)) { 3823 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3824 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, 3825 NULL); 3826 if (error == 0) 3827 error = nfsrv_deldsserver(pnfsdarg.op, cp, td); 3828 free(cp, M_TEMP); 3829 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { 3830 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3831 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; 3832 buf = malloc(buflen, M_TEMP, M_WAITOK); 3833 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, 3834 NULL); 3835 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); 3836 if (error == 0 && pnfsdarg.dspath != NULL) { 3837 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3838 error = copyinstr(pnfsdarg.dspath, cp2, 3839 PATH_MAX + 1, NULL); 3840 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", 3841 error); 3842 } else 3843 cp2 = NULL; 3844 if (error == 0 && pnfsdarg.curdspath != NULL) { 3845 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3846 error = copyinstr(pnfsdarg.curdspath, cp3, 3847 PATH_MAX + 1, NULL); 3848 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", 3849 error); 3850 } else 3851 cp3 = NULL; 3852 curdvp = NULL; 3853 fds = NULL; 3854 if (error == 0) 3855 error = nfsrv_mdscopymr(cp, cp2, cp3, buf, 3856 &buflen, fname, td, &vp, &nvp, &pf, &ds, 3857 &fds); 3858 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); 3859 if (error == 0) { 3860 if (pf->dsf_dir >= nfsrv_dsdirsize) { 3861 printf("copymr: dsdir out of range\n"); 3862 pf->dsf_dir = 0; 3863 } 3864 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); 3865 error = nfsrv_copymr(vp, nvp, 3866 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, 3867 (struct pnfsdsfile *)buf, 3868 buflen / sizeof(*pf), td->td_ucred, td); 3869 vput(vp); 3870 vput(nvp); 3871 if (fds != NULL && error == 0) { 3872 curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; 3873 ret = vn_lock(curdvp, LK_EXCLUSIVE); 3874 if (ret == 0) { 3875 nfsrv_dsremove(curdvp, fname, 3876 td->td_ucred, td); 3877 NFSVOPUNLOCK(curdvp); 3878 } 3879 } 3880 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); 3881 } 3882 free(cp, M_TEMP); 3883 free(cp2, M_TEMP); 3884 free(cp3, M_TEMP); 3885 free(buf, M_TEMP); 3886 } 3887 } else { 3888 error = nfssvc_srvcall(td, uap, td->td_ucred); 3889 } 3890 3891 out: 3892 NFSD_CURVNET_RESTORE(); 3893 NFSEXITCODE(error); 3894 return (error); 3895 } 3896 3897 static int 3898 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 3899 { 3900 struct nfsex_args export; 3901 struct nfsex_oldargs oexp; 3902 struct file *fp = NULL; 3903 int stablefd, i, len; 3904 struct nfsd_clid adminrevoke; 3905 struct nfsd_dumplist dumplist; 3906 struct nfsd_dumpclients *dumpclients; 3907 struct nfsd_dumplocklist dumplocklist; 3908 struct nfsd_dumplocks *dumplocks; 3909 struct nameidata nd; 3910 vnode_t vp; 3911 int error = EINVAL, igotlock; 3912 struct proc *procp; 3913 gid_t *grps; 3914 3915 if (uap->flag & NFSSVC_PUBLICFH) { 3916 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 3917 sizeof (fhandle_t)); 3918 error = copyin(uap->argp, 3919 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 3920 if (!error) 3921 nfs_pubfhset = 1; 3922 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 3923 (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { 3924 error = copyin(uap->argp,(caddr_t)&export, 3925 sizeof (struct nfsex_args)); 3926 if (!error) { 3927 grps = NULL; 3928 if (export.export.ex_ngroups > NGROUPS_MAX || 3929 export.export.ex_ngroups < 0) 3930 error = EINVAL; 3931 else if (export.export.ex_ngroups > 0) { 3932 grps = malloc(export.export.ex_ngroups * 3933 sizeof(gid_t), M_TEMP, M_WAITOK); 3934 error = copyin(export.export.ex_groups, grps, 3935 export.export.ex_ngroups * sizeof(gid_t)); 3936 export.export.ex_groups = grps; 3937 } else 3938 export.export.ex_groups = NULL; 3939 if (!error) 3940 error = nfsrv_v4rootexport(&export, cred, p); 3941 free(grps, M_TEMP); 3942 } 3943 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 3944 NFSSVC_V4ROOTEXPORT) { 3945 error = copyin(uap->argp,(caddr_t)&oexp, 3946 sizeof (struct nfsex_oldargs)); 3947 if (!error) { 3948 memset(&export.export, 0, sizeof(export.export)); 3949 export.export.ex_flags = (uint64_t)oexp.export.ex_flags; 3950 export.export.ex_root = oexp.export.ex_root; 3951 export.export.ex_uid = oexp.export.ex_anon.cr_uid; 3952 export.export.ex_ngroups = 3953 oexp.export.ex_anon.cr_ngroups; 3954 export.export.ex_groups = NULL; 3955 if (export.export.ex_ngroups > XU_NGROUPS || 3956 export.export.ex_ngroups < 0) 3957 error = EINVAL; 3958 else if (export.export.ex_ngroups > 0) { 3959 export.export.ex_groups = malloc( 3960 export.export.ex_ngroups * sizeof(gid_t), 3961 M_TEMP, M_WAITOK); 3962 for (i = 0; i < export.export.ex_ngroups; i++) 3963 export.export.ex_groups[i] = 3964 oexp.export.ex_anon.cr_groups[i]; 3965 } 3966 export.export.ex_addr = oexp.export.ex_addr; 3967 export.export.ex_addrlen = oexp.export.ex_addrlen; 3968 export.export.ex_mask = oexp.export.ex_mask; 3969 export.export.ex_masklen = oexp.export.ex_masklen; 3970 export.export.ex_indexfile = oexp.export.ex_indexfile; 3971 export.export.ex_numsecflavors = 3972 oexp.export.ex_numsecflavors; 3973 if (export.export.ex_numsecflavors >= MAXSECFLAVORS || 3974 export.export.ex_numsecflavors < 0) 3975 error = EINVAL; 3976 else { 3977 for (i = 0; i < export.export.ex_numsecflavors; 3978 i++) 3979 export.export.ex_secflavors[i] = 3980 oexp.export.ex_secflavors[i]; 3981 } 3982 export.fspec = oexp.fspec; 3983 if (error == 0) 3984 error = nfsrv_v4rootexport(&export, cred, p); 3985 free(export.export.ex_groups, M_TEMP); 3986 } 3987 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 3988 nfs_pubfhset = 0; 3989 error = 0; 3990 } else if (uap->flag & NFSSVC_STABLERESTART) { 3991 error = copyin(uap->argp, (caddr_t)&stablefd, 3992 sizeof (int)); 3993 if (!error) 3994 error = fp_getfvp(p, stablefd, &fp, &vp); 3995 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 3996 error = EBADF; 3997 if (!error && NFSD_VNET(nfsrv_numnfsd) != 0) 3998 error = EPERM; 3999 if (!error) { 4000 NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp; 4001 nfsrv_setupstable(p); 4002 } 4003 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 4004 error = copyin(uap->argp, (caddr_t)&adminrevoke, 4005 sizeof (struct nfsd_clid)); 4006 if (!error) 4007 error = nfsrv_adminrevoke(&adminrevoke, p); 4008 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 4009 error = copyin(uap->argp, (caddr_t)&dumplist, 4010 sizeof (struct nfsd_dumplist)); 4011 if (!error && (dumplist.ndl_size < 1 || 4012 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 4013 error = EPERM; 4014 if (!error) { 4015 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 4016 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4017 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 4018 error = copyout(dumpclients, dumplist.ndl_list, len); 4019 free(dumpclients, M_TEMP); 4020 } 4021 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 4022 error = copyin(uap->argp, (caddr_t)&dumplocklist, 4023 sizeof (struct nfsd_dumplocklist)); 4024 if (!error && (dumplocklist.ndllck_size < 1 || 4025 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 4026 error = EPERM; 4027 if (!error) 4028 error = nfsrv_lookupfilename(&nd, 4029 dumplocklist.ndllck_fname, p); 4030 if (!error) { 4031 len = sizeof (struct nfsd_dumplocks) * 4032 dumplocklist.ndllck_size; 4033 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4034 nfsrv_dumplocks(nd.ni_vp, dumplocks, 4035 dumplocklist.ndllck_size, p); 4036 vput(nd.ni_vp); 4037 error = copyout(dumplocks, dumplocklist.ndllck_list, 4038 len); 4039 free(dumplocks, M_TEMP); 4040 } 4041 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 4042 procp = p->td_proc; 4043 PROC_LOCK(procp); 4044 nfsd_master_pid = procp->p_pid; 4045 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 4046 nfsd_master_start = procp->p_stats->p_start; 4047 NFSD_VNET(nfsd_master_proc) = procp; 4048 PROC_UNLOCK(procp); 4049 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { 4050 NFSLOCKV4ROOTMUTEX(); 4051 if (!NFSD_VNET(nfsrv_suspend_nfsd)) { 4052 /* Lock out all nfsd threads */ 4053 do { 4054 igotlock = nfsv4_lock( 4055 &NFSD_VNET(nfsd_suspend_lock), 1, NULL, 4056 NFSV4ROOTLOCKMUTEXPTR, NULL); 4057 } while (igotlock == 0 && 4058 !NFSD_VNET(nfsrv_suspend_nfsd)); 4059 NFSD_VNET(nfsrv_suspend_nfsd) = true; 4060 } 4061 NFSUNLOCKV4ROOTMUTEX(); 4062 error = 0; 4063 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { 4064 NFSLOCKV4ROOTMUTEX(); 4065 if (NFSD_VNET(nfsrv_suspend_nfsd)) { 4066 nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0); 4067 NFSD_VNET(nfsrv_suspend_nfsd) = false; 4068 } 4069 NFSUNLOCKV4ROOTMUTEX(); 4070 error = 0; 4071 } 4072 4073 NFSEXITCODE(error); 4074 return (error); 4075 } 4076 4077 /* 4078 * Check exports. 4079 * Returns 0 if ok, 1 otherwise. 4080 */ 4081 int 4082 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 4083 { 4084 int i; 4085 4086 if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) || 4087 (NFSVNO_EXTLSCERT(exp) && 4088 (nd->nd_flag & ND_TLSCERT) == 0) || 4089 (NFSVNO_EXTLSCERTUSER(exp) && 4090 (nd->nd_flag & ND_TLSCERTUSER) == 0)) { 4091 if ((nd->nd_flag & ND_NFSV4) != 0) 4092 return (NFSERR_WRONGSEC); 4093 #ifdef notnow 4094 /* There is currently no auth_stat for this. */ 4095 else if ((nd->nd_flag & ND_TLS) == 0) 4096 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS); 4097 else 4098 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST); 4099 #endif 4100 else 4101 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4102 } 4103 4104 /* 4105 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use 4106 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. 4107 */ 4108 if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO) 4109 return (0); 4110 4111 /* 4112 * This seems odd, but allow the case where the security flavor 4113 * list is empty. This happens when NFSv4 is traversing non-exported 4114 * file systems. Exported file systems should always have a non-empty 4115 * security flavor list. 4116 */ 4117 if (exp->nes_numsecflavor == 0) 4118 return (0); 4119 4120 for (i = 0; i < exp->nes_numsecflavor; i++) { 4121 /* 4122 * The tests for privacy and integrity must be first, 4123 * since ND_GSS is set for everything but AUTH_SYS. 4124 */ 4125 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 4126 (nd->nd_flag & ND_GSSPRIVACY)) 4127 return (0); 4128 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 4129 (nd->nd_flag & ND_GSSINTEGRITY)) 4130 return (0); 4131 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 4132 (nd->nd_flag & ND_GSS)) 4133 return (0); 4134 if (exp->nes_secflavors[i] == AUTH_SYS && 4135 (nd->nd_flag & ND_GSS) == 0) 4136 return (0); 4137 } 4138 if ((nd->nd_flag & ND_NFSV4) != 0) 4139 return (NFSERR_WRONGSEC); 4140 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4141 } 4142 4143 /* 4144 * Calculate a hash value for the fid in a file handle. 4145 */ 4146 uint32_t 4147 nfsrv_hashfh(fhandle_t *fhp) 4148 { 4149 uint32_t hashval; 4150 4151 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 4152 return (hashval); 4153 } 4154 4155 /* 4156 * Calculate a hash value for the sessionid. 4157 */ 4158 uint32_t 4159 nfsrv_hashsessionid(uint8_t *sessionid) 4160 { 4161 uint32_t hashval; 4162 4163 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); 4164 return (hashval); 4165 } 4166 4167 /* 4168 * Signal the userland master nfsd to backup the stable restart file. 4169 */ 4170 void 4171 nfsrv_backupstable(void) 4172 { 4173 struct proc *procp; 4174 4175 if (NFSD_VNET(nfsd_master_proc) != NULL) { 4176 procp = pfind(nfsd_master_pid); 4177 /* Try to make sure it is the correct process. */ 4178 if (procp == NFSD_VNET(nfsd_master_proc) && 4179 procp->p_stats->p_start.tv_sec == 4180 nfsd_master_start.tv_sec && 4181 procp->p_stats->p_start.tv_usec == 4182 nfsd_master_start.tv_usec && 4183 strcmp(procp->p_comm, nfsd_master_comm) == 0) 4184 kern_psignal(procp, SIGUSR2); 4185 else 4186 NFSD_VNET(nfsd_master_proc) = NULL; 4187 4188 if (procp != NULL) 4189 PROC_UNLOCK(procp); 4190 } 4191 } 4192 4193 /* 4194 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. 4195 * The arguments are in a structure, so that they can be passed through 4196 * taskqueue for a kernel process to execute this function. 4197 */ 4198 struct nfsrvdscreate { 4199 int done; 4200 int inprog; 4201 struct task tsk; 4202 struct ucred *tcred; 4203 struct vnode *dvp; 4204 NFSPROC_T *p; 4205 struct pnfsdsfile *pf; 4206 int err; 4207 fhandle_t fh; 4208 struct vattr va; 4209 struct vattr createva; 4210 }; 4211 4212 int 4213 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, 4214 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, 4215 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) 4216 { 4217 struct vnode *nvp; 4218 struct nameidata named; 4219 struct vattr va; 4220 char *bufp; 4221 u_long *hashp; 4222 struct nfsnode *np; 4223 struct nfsmount *nmp; 4224 int error; 4225 4226 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, 4227 LOCKPARENT | LOCKLEAF | NOCACHE); 4228 nfsvno_setpathbuf(&named, &bufp, &hashp); 4229 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 4230 named.ni_cnd.cn_nameptr = bufp; 4231 if (fnamep != NULL) { 4232 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); 4233 named.ni_cnd.cn_namelen = strlen(bufp); 4234 } else 4235 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); 4236 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); 4237 4238 /* Create the date file in the DS mount. */ 4239 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4240 if (error == 0) { 4241 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); 4242 vref(dvp); 4243 VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false); 4244 if (error == 0) { 4245 /* Set the ownership of the file. */ 4246 error = VOP_SETATTR(nvp, nvap, tcred); 4247 NFSD_DEBUG(4, "nfsrv_dscreate:" 4248 " setattr-uid=%d\n", error); 4249 if (error != 0) 4250 vput(nvp); 4251 } 4252 if (error != 0) 4253 printf("pNFS: pnfscreate failed=%d\n", error); 4254 } else 4255 printf("pNFS: pnfscreate vnlock=%d\n", error); 4256 if (error == 0) { 4257 np = VTONFS(nvp); 4258 nmp = VFSTONFS(nvp->v_mount); 4259 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") 4260 != 0 || nmp->nm_nam->sa_len > sizeof( 4261 struct sockaddr_in6) || 4262 np->n_fhp->nfh_len != NFSX_MYFH) { 4263 printf("Bad DS file: fstype=%s salen=%d" 4264 " fhlen=%d\n", 4265 nvp->v_mount->mnt_vfc->vfc_name, 4266 nmp->nm_nam->sa_len, np->n_fhp->nfh_len); 4267 error = ENOENT; 4268 } 4269 4270 /* Set extattrs for the DS on the MDS file. */ 4271 if (error == 0) { 4272 if (dsa != NULL) { 4273 error = VOP_GETATTR(nvp, &va, tcred); 4274 if (error == 0) { 4275 dsa->dsa_filerev = va.va_filerev; 4276 dsa->dsa_size = va.va_size; 4277 dsa->dsa_atime = va.va_atime; 4278 dsa->dsa_mtime = va.va_mtime; 4279 dsa->dsa_bytes = va.va_bytes; 4280 } 4281 } 4282 if (error == 0) { 4283 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, 4284 NFSX_MYFH); 4285 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, 4286 nmp->nm_nam->sa_len); 4287 NFSBCOPY(named.ni_cnd.cn_nameptr, 4288 pf->dsf_filename, 4289 sizeof(pf->dsf_filename)); 4290 } 4291 } else 4292 printf("pNFS: pnfscreate can't get DS" 4293 " attr=%d\n", error); 4294 if (nvpp != NULL && error == 0) 4295 *nvpp = nvp; 4296 else 4297 vput(nvp); 4298 } 4299 nfsvno_relpathbuf(&named); 4300 return (error); 4301 } 4302 4303 /* 4304 * Start up the thread that will execute nfsrv_dscreate(). 4305 */ 4306 static void 4307 start_dscreate(void *arg, int pending) 4308 { 4309 struct nfsrvdscreate *dsc; 4310 4311 dsc = (struct nfsrvdscreate *)arg; 4312 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, 4313 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); 4314 dsc->done = 1; 4315 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); 4316 } 4317 4318 /* 4319 * Create a pNFS data file on the Data Server(s). 4320 */ 4321 static void 4322 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, 4323 NFSPROC_T *p) 4324 { 4325 struct nfsrvdscreate *dsc, *tdsc = NULL; 4326 struct nfsdevice *ds, *tds, *fds; 4327 struct mount *mp; 4328 struct pnfsdsfile *pf, *tpf; 4329 struct pnfsdsattr dsattr; 4330 struct vattr va; 4331 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4332 struct nfsmount *nmp; 4333 fhandle_t fh; 4334 uid_t vauid; 4335 gid_t vagid; 4336 u_short vamode; 4337 struct ucred *tcred; 4338 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; 4339 int failpos, timo; 4340 4341 /* Get a DS server directory in a round-robin order. */ 4342 mirrorcnt = 1; 4343 mp = vp->v_mount; 4344 ds = fds = NULL; 4345 NFSDDSLOCK(); 4346 /* 4347 * Search for the first entry that handles this MDS fs, but use the 4348 * first entry for all MDS fs's otherwise. 4349 */ 4350 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { 4351 if (tds->nfsdev_nmp != NULL) { 4352 if (tds->nfsdev_mdsisset == 0 && ds == NULL) 4353 ds = tds; 4354 else if (tds->nfsdev_mdsisset != 0 && fsidcmp( 4355 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { 4356 ds = fds = tds; 4357 break; 4358 } 4359 } 4360 } 4361 if (ds == NULL) { 4362 NFSDDSUNLOCK(); 4363 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); 4364 return; 4365 } 4366 i = dsdir[0] = ds->nfsdev_nextdir; 4367 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; 4368 dvp[0] = ds->nfsdev_dsdir[i]; 4369 tds = TAILQ_NEXT(ds, nfsdev_list); 4370 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { 4371 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { 4372 if (tds->nfsdev_nmp != NULL && 4373 ((tds->nfsdev_mdsisset == 0 && fds == NULL) || 4374 (tds->nfsdev_mdsisset != 0 && fds != NULL && 4375 fsidcmp(&mp->mnt_stat.f_fsid, 4376 &tds->nfsdev_mdsfsid) == 0))) { 4377 dsdir[mirrorcnt] = i; 4378 dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; 4379 mirrorcnt++; 4380 if (mirrorcnt >= nfsrv_maxpnfsmirror) 4381 break; 4382 } 4383 } 4384 } 4385 /* Put at end of list to implement round-robin usage. */ 4386 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 4387 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 4388 NFSDDSUNLOCK(); 4389 dsc = NULL; 4390 if (mirrorcnt > 1) 4391 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, 4392 M_WAITOK | M_ZERO); 4393 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | 4394 M_ZERO); 4395 4396 error = nfsvno_getfh(vp, &fh, p); 4397 if (error == 0) 4398 error = VOP_GETATTR(vp, &va, cred); 4399 if (error == 0) { 4400 /* Set the attributes for "vp" to Setattr the DS vp. */ 4401 vauid = va.va_uid; 4402 vagid = va.va_gid; 4403 vamode = va.va_mode; 4404 VATTR_NULL(&va); 4405 va.va_uid = vauid; 4406 va.va_gid = vagid; 4407 va.va_mode = vamode; 4408 va.va_size = 0; 4409 } else 4410 printf("pNFS: pnfscreate getfh+attr=%d\n", error); 4411 4412 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, 4413 cred->cr_gid); 4414 /* Make data file name based on FH. */ 4415 tcred = newnfs_getcred(); 4416 4417 /* 4418 * Create the file on each DS mirror, using kernel process(es) for the 4419 * additional mirrors. 4420 */ 4421 failpos = -1; 4422 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { 4423 tpf->dsf_dir = dsdir[i]; 4424 tdsc->tcred = tcred; 4425 tdsc->p = p; 4426 tdsc->pf = tpf; 4427 tdsc->createva = *vap; 4428 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); 4429 tdsc->va = va; 4430 tdsc->dvp = dvp[i]; 4431 tdsc->done = 0; 4432 tdsc->inprog = 0; 4433 tdsc->err = 0; 4434 ret = EIO; 4435 if (nfs_pnfsiothreads != 0) { 4436 ret = nfs_pnfsio(start_dscreate, tdsc); 4437 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); 4438 } 4439 if (ret != 0) { 4440 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, 4441 NULL, tcred, p, NULL); 4442 if (ret != 0) { 4443 KASSERT(error == 0, ("nfsrv_dscreate err=%d", 4444 error)); 4445 if (failpos == -1 && nfsds_failerr(ret)) 4446 failpos = i; 4447 else 4448 error = ret; 4449 } 4450 } 4451 } 4452 if (error == 0) { 4453 tpf->dsf_dir = dsdir[mirrorcnt - 1]; 4454 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, 4455 &dsattr, NULL, tcred, p, NULL); 4456 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { 4457 failpos = mirrorcnt - 1; 4458 error = 0; 4459 } 4460 } 4461 timo = hz / 50; /* Wait for 20msec. */ 4462 if (timo < 1) 4463 timo = 1; 4464 /* Wait for kernel task(s) to complete. */ 4465 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { 4466 while (tdsc->inprog != 0 && tdsc->done == 0) 4467 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); 4468 if (tdsc->err != 0) { 4469 if (failpos == -1 && nfsds_failerr(tdsc->err)) 4470 failpos = i; 4471 else if (error == 0) 4472 error = tdsc->err; 4473 } 4474 } 4475 4476 /* 4477 * If failpos has been set, that mirror has failed, so it needs 4478 * to be disabled. 4479 */ 4480 if (failpos >= 0) { 4481 nmp = VFSTONFS(dvp[failpos]->v_mount); 4482 NFSLOCKMNT(nmp); 4483 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4484 NFSMNTP_CANCELRPCS)) == 0) { 4485 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4486 NFSUNLOCKMNT(nmp); 4487 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4488 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, 4489 ds); 4490 if (ds != NULL) 4491 nfsrv_killrpcs(nmp); 4492 NFSLOCKMNT(nmp); 4493 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4494 wakeup(nmp); 4495 } 4496 NFSUNLOCKMNT(nmp); 4497 } 4498 4499 NFSFREECRED(tcred); 4500 if (error == 0) { 4501 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); 4502 4503 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", 4504 mirrorcnt, nfsrv_maxpnfsmirror); 4505 /* 4506 * For all mirrors that couldn't be created, fill in the 4507 * *pf structure, but with an IP address == 0.0.0.0. 4508 */ 4509 tpf = pf + mirrorcnt; 4510 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { 4511 *tpf = *pf; 4512 tpf->dsf_sin.sin_family = AF_INET; 4513 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); 4514 tpf->dsf_sin.sin_addr.s_addr = 0; 4515 tpf->dsf_sin.sin_port = 0; 4516 } 4517 4518 error = vn_extattr_set(vp, IO_NODELOCKED, 4519 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 4520 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); 4521 if (error == 0) 4522 error = vn_extattr_set(vp, IO_NODELOCKED, 4523 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", 4524 sizeof(dsattr), (char *)&dsattr, p); 4525 if (error != 0) 4526 printf("pNFS: pnfscreate setextattr=%d\n", 4527 error); 4528 } else 4529 printf("pNFS: pnfscreate=%d\n", error); 4530 free(pf, M_TEMP); 4531 free(dsc, M_TEMP); 4532 } 4533 4534 /* 4535 * Get the information needed to remove the pNFS Data Server file from the 4536 * Metadata file. Upon success, ddvp is set non-NULL to the locked 4537 * DS directory vnode. The caller must unlock *ddvp when done with it. 4538 */ 4539 static void 4540 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, 4541 int *mirrorcntp, char *fname, fhandle_t *fhp) 4542 { 4543 struct vattr va; 4544 struct ucred *tcred; 4545 char *buf; 4546 int buflen, error; 4547 4548 dvpp[0] = NULL; 4549 /* If not an exported regular file or not a pNFS server, just return. */ 4550 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4551 nfsrv_devidcnt == 0) 4552 return; 4553 4554 /* Check to see if this is the last hard link. */ 4555 tcred = newnfs_getcred(); 4556 error = VOP_GETATTR(vp, &va, tcred); 4557 NFSFREECRED(tcred); 4558 if (error != 0) { 4559 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); 4560 return; 4561 } 4562 if (va.va_nlink > 1) 4563 return; 4564 4565 error = nfsvno_getfh(vp, fhp, p); 4566 if (error != 0) { 4567 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); 4568 return; 4569 } 4570 4571 buflen = 1024; 4572 buf = malloc(buflen, M_TEMP, M_WAITOK); 4573 /* Get the directory vnode for the DS mount and the file handle. */ 4574 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, 4575 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); 4576 free(buf, M_TEMP); 4577 if (error != 0) 4578 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); 4579 } 4580 4581 /* 4582 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. 4583 * The arguments are in a structure, so that they can be passed through 4584 * taskqueue for a kernel process to execute this function. 4585 */ 4586 struct nfsrvdsremove { 4587 int done; 4588 int inprog; 4589 struct task tsk; 4590 struct ucred *tcred; 4591 struct vnode *dvp; 4592 NFSPROC_T *p; 4593 int err; 4594 char fname[PNFS_FILENAME_LEN + 1]; 4595 }; 4596 4597 static int 4598 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, 4599 NFSPROC_T *p) 4600 { 4601 struct nameidata named; 4602 struct vnode *nvp; 4603 char *bufp; 4604 u_long *hashp; 4605 int error; 4606 4607 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4608 if (error != 0) 4609 return (error); 4610 named.ni_cnd.cn_nameiop = DELETE; 4611 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 4612 named.ni_cnd.cn_cred = tcred; 4613 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 4614 nfsvno_setpathbuf(&named, &bufp, &hashp); 4615 named.ni_cnd.cn_nameptr = bufp; 4616 named.ni_cnd.cn_namelen = strlen(fname); 4617 strlcpy(bufp, fname, NAME_MAX); 4618 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); 4619 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 4620 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); 4621 if (error == 0) { 4622 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); 4623 vput(nvp); 4624 } 4625 NFSVOPUNLOCK(dvp); 4626 nfsvno_relpathbuf(&named); 4627 if (error != 0) 4628 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); 4629 return (error); 4630 } 4631 4632 /* 4633 * Start up the thread that will execute nfsrv_dsremove(). 4634 */ 4635 static void 4636 start_dsremove(void *arg, int pending) 4637 { 4638 struct nfsrvdsremove *dsrm; 4639 4640 dsrm = (struct nfsrvdsremove *)arg; 4641 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, 4642 dsrm->p); 4643 dsrm->done = 1; 4644 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); 4645 } 4646 4647 /* 4648 * Remove a pNFS data file from a Data Server. 4649 * nfsrv_pnfsremovesetup() must have been called before the MDS file was 4650 * removed to set up the dvp and fill in the FH. 4651 */ 4652 static void 4653 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, 4654 NFSPROC_T *p) 4655 { 4656 struct ucred *tcred; 4657 struct nfsrvdsremove *dsrm, *tdsrm; 4658 struct nfsdevice *ds; 4659 struct nfsmount *nmp; 4660 int failpos, i, ret, timo; 4661 4662 tcred = newnfs_getcred(); 4663 dsrm = NULL; 4664 if (mirrorcnt > 1) 4665 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); 4666 /* 4667 * Remove the file on each DS mirror, using kernel process(es) for the 4668 * additional mirrors. 4669 */ 4670 failpos = -1; 4671 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4672 tdsrm->tcred = tcred; 4673 tdsrm->p = p; 4674 tdsrm->dvp = dvp[i]; 4675 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); 4676 tdsrm->inprog = 0; 4677 tdsrm->done = 0; 4678 tdsrm->err = 0; 4679 ret = EIO; 4680 if (nfs_pnfsiothreads != 0) { 4681 ret = nfs_pnfsio(start_dsremove, tdsrm); 4682 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); 4683 } 4684 if (ret != 0) { 4685 ret = nfsrv_dsremove(dvp[i], fname, tcred, p); 4686 if (failpos == -1 && nfsds_failerr(ret)) 4687 failpos = i; 4688 } 4689 } 4690 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); 4691 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) 4692 failpos = mirrorcnt - 1; 4693 timo = hz / 50; /* Wait for 20msec. */ 4694 if (timo < 1) 4695 timo = 1; 4696 /* Wait for kernel task(s) to complete. */ 4697 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4698 while (tdsrm->inprog != 0 && tdsrm->done == 0) 4699 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); 4700 if (failpos == -1 && nfsds_failerr(tdsrm->err)) 4701 failpos = i; 4702 } 4703 4704 /* 4705 * If failpos has been set, that mirror has failed, so it needs 4706 * to be disabled. 4707 */ 4708 if (failpos >= 0) { 4709 nmp = VFSTONFS(dvp[failpos]->v_mount); 4710 NFSLOCKMNT(nmp); 4711 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4712 NFSMNTP_CANCELRPCS)) == 0) { 4713 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4714 NFSUNLOCKMNT(nmp); 4715 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4716 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, 4717 ds); 4718 if (ds != NULL) 4719 nfsrv_killrpcs(nmp); 4720 NFSLOCKMNT(nmp); 4721 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4722 wakeup(nmp); 4723 } 4724 NFSUNLOCKMNT(nmp); 4725 } 4726 4727 /* Get rid all layouts for the file. */ 4728 nfsrv_freefilelayouts(fhp); 4729 4730 NFSFREECRED(tcred); 4731 free(dsrm, M_TEMP); 4732 } 4733 4734 /* 4735 * Generate a file name based on the file handle and put it in *bufp. 4736 * Return the number of bytes generated. 4737 */ 4738 static int 4739 nfsrv_putfhname(fhandle_t *fhp, char *bufp) 4740 { 4741 int i; 4742 uint8_t *cp; 4743 const uint8_t *hexdigits = "0123456789abcdef"; 4744 4745 cp = (uint8_t *)fhp; 4746 for (i = 0; i < sizeof(*fhp); i++) { 4747 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; 4748 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; 4749 } 4750 bufp[2 * i] = '\0'; 4751 return (2 * i); 4752 } 4753 4754 /* 4755 * Update the Metadata file's attributes from the DS file when a Read/Write 4756 * layout is returned. 4757 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN 4758 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. 4759 */ 4760 int 4761 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 4762 { 4763 struct ucred *tcred; 4764 int error; 4765 4766 /* Do this as root so that it won't fail with EACCES. */ 4767 tcred = newnfs_getcred(); 4768 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, 4769 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); 4770 NFSFREECRED(tcred); 4771 return (error); 4772 } 4773 4774 /* 4775 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. 4776 */ 4777 static int 4778 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, 4779 NFSPROC_T *p) 4780 { 4781 int error; 4782 4783 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, 4784 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); 4785 return (error); 4786 } 4787 4788 static int 4789 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 4790 struct thread *p, int ioproc, struct mbuf **mpp, char *cp, 4791 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, 4792 off_t *offp, int content, bool *eofp) 4793 { 4794 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; 4795 fhandle_t fh[NFSDEV_MAXMIRRORS]; 4796 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4797 struct nfsdevice *ds; 4798 struct pnfsdsattr dsattr; 4799 struct opnfsdsattr odsattr; 4800 char *buf; 4801 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; 4802 4803 NFSD_DEBUG(4, "in nfsrv_proxyds\n"); 4804 /* 4805 * If not a regular file, not exported or not a pNFS server, 4806 * just return ENOENT. 4807 */ 4808 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4809 nfsrv_devidcnt == 0) 4810 return (ENOENT); 4811 4812 buflen = 1024; 4813 buf = malloc(buflen, M_TEMP, M_WAITOK); 4814 error = 0; 4815 4816 /* 4817 * For Getattr, get the Change attribute (va_filerev) and size (va_size) 4818 * from the MetaData file's extended attribute. 4819 */ 4820 if (ioproc == NFSPROC_GETATTR) { 4821 error = vn_extattr_get(vp, IO_NODELOCKED, 4822 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, 4823 p); 4824 if (error == 0) { 4825 if (buflen == sizeof(odsattr)) { 4826 NFSBCOPY(buf, &odsattr, buflen); 4827 nap->na_filerev = odsattr.dsa_filerev; 4828 nap->na_size = odsattr.dsa_size; 4829 nap->na_atime = odsattr.dsa_atime; 4830 nap->na_mtime = odsattr.dsa_mtime; 4831 /* 4832 * Fake na_bytes by rounding up na_size. 4833 * Since we don't know the block size, just 4834 * use BLKDEV_IOSIZE. 4835 */ 4836 nap->na_bytes = (odsattr.dsa_size + 4837 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); 4838 } else if (buflen == sizeof(dsattr)) { 4839 NFSBCOPY(buf, &dsattr, buflen); 4840 nap->na_filerev = dsattr.dsa_filerev; 4841 nap->na_size = dsattr.dsa_size; 4842 nap->na_atime = dsattr.dsa_atime; 4843 nap->na_mtime = dsattr.dsa_mtime; 4844 nap->na_bytes = dsattr.dsa_bytes; 4845 } else 4846 error = ENXIO; 4847 } 4848 if (error == 0) { 4849 /* 4850 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() 4851 * returns 0, just return now. nfsrv_checkdsattr() 4852 * returns 0 if there is no Read/Write layout 4853 * plus either an Open/Write_access or Write 4854 * delegation issued to a client for the file. 4855 */ 4856 if (nfsrv_pnfsgetdsattr == 0 || 4857 nfsrv_checkdsattr(vp, p) == 0) { 4858 free(buf, M_TEMP); 4859 return (error); 4860 } 4861 } 4862 4863 /* 4864 * Clear ENOATTR so the code below will attempt to do a 4865 * nfsrv_getattrdsrpc() to get the attributes and (re)create 4866 * the extended attribute. 4867 */ 4868 if (error == ENOATTR) 4869 error = 0; 4870 } 4871 4872 origmircnt = -1; 4873 trycnt = 0; 4874 tryagain: 4875 if (error == 0) { 4876 buflen = 1024; 4877 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == 4878 LK_EXCLUSIVE) 4879 printf("nfsrv_proxyds: Readds vp exclusively locked\n"); 4880 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, 4881 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, 4882 NULL, NULL); 4883 if (error == 0) { 4884 for (i = 0; i < mirrorcnt; i++) 4885 nmp[i] = VFSTONFS(dvp[i]->v_mount); 4886 } else 4887 printf("pNFS: proxy getextattr sockaddr=%d\n", error); 4888 } else 4889 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); 4890 if (error == 0) { 4891 failpos = -1; 4892 if (origmircnt == -1) 4893 origmircnt = mirrorcnt; 4894 /* 4895 * If failpos is set to a mirror#, then that mirror has 4896 * failed and will be disabled. For Read, Getattr and Seek, the 4897 * function only tries one mirror, so if that mirror has 4898 * failed, it will need to be retried. As such, increment 4899 * tryitagain for these cases. 4900 * For Write, Setattr and Setacl, the function tries all 4901 * mirrors and will not return an error for the case where 4902 * one mirror has failed. For these cases, the functioning 4903 * mirror(s) will have been modified, so a retry isn't 4904 * necessary. These functions will set failpos for the 4905 * failed mirror#. 4906 */ 4907 if (ioproc == NFSPROC_READDS) { 4908 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], 4909 mpp, mpp2); 4910 if (nfsds_failerr(error) && mirrorcnt > 1) { 4911 /* 4912 * Setting failpos will cause the mirror 4913 * to be disabled and then a retry of this 4914 * read is required. 4915 */ 4916 failpos = 0; 4917 error = 0; 4918 trycnt++; 4919 } 4920 } else if (ioproc == NFSPROC_WRITEDS) 4921 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, 4922 &nmp[0], mirrorcnt, mpp, cp, &failpos); 4923 else if (ioproc == NFSPROC_SETATTR) 4924 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], 4925 mirrorcnt, nap, &failpos); 4926 else if (ioproc == NFSPROC_SETACL) 4927 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], 4928 mirrorcnt, aclp, &failpos); 4929 else if (ioproc == NFSPROC_SEEKDS) { 4930 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, 4931 p, nmp[0]); 4932 if (nfsds_failerr(error) && mirrorcnt > 1) { 4933 /* 4934 * Setting failpos will cause the mirror 4935 * to be disabled and then a retry of this 4936 * read is required. 4937 */ 4938 failpos = 0; 4939 error = 0; 4940 trycnt++; 4941 } 4942 } else if (ioproc == NFSPROC_ALLOCATE) 4943 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, 4944 &nmp[0], mirrorcnt, &failpos); 4945 else if (ioproc == NFSPROC_DEALLOCATE) 4946 error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p, 4947 vp, &nmp[0], mirrorcnt, &failpos); 4948 else { 4949 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, 4950 vp, nmp[mirrorcnt - 1], nap); 4951 if (nfsds_failerr(error) && mirrorcnt > 1) { 4952 /* 4953 * Setting failpos will cause the mirror 4954 * to be disabled and then a retry of this 4955 * getattr is required. 4956 */ 4957 failpos = mirrorcnt - 1; 4958 error = 0; 4959 trycnt++; 4960 } 4961 } 4962 ds = NULL; 4963 if (failpos >= 0) { 4964 failnmp = nmp[failpos]; 4965 NFSLOCKMNT(failnmp); 4966 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | 4967 NFSMNTP_CANCELRPCS)) == 0) { 4968 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4969 NFSUNLOCKMNT(failnmp); 4970 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, 4971 failnmp, p); 4972 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", 4973 failpos, ds); 4974 if (ds != NULL) 4975 nfsrv_killrpcs(failnmp); 4976 NFSLOCKMNT(failnmp); 4977 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4978 wakeup(failnmp); 4979 } 4980 NFSUNLOCKMNT(failnmp); 4981 } 4982 for (i = 0; i < mirrorcnt; i++) 4983 NFSVOPUNLOCK(dvp[i]); 4984 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, 4985 trycnt); 4986 /* Try the Read/Getattr again if a mirror was deleted. */ 4987 if (ds != NULL && trycnt > 0 && trycnt < origmircnt) 4988 goto tryagain; 4989 } else { 4990 /* Return ENOENT for any Extended Attribute error. */ 4991 error = ENOENT; 4992 } 4993 free(buf, M_TEMP); 4994 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); 4995 return (error); 4996 } 4997 4998 /* 4999 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended 5000 * attribute. 5001 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs 5002 * to be checked. If it points to a NULL nmp, then it returns 5003 * a suitable destination. 5004 * curnmp - If non-NULL, it is the source mount for the copy. 5005 */ 5006 int 5007 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, 5008 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, 5009 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, 5010 struct nfsmount *curnmp, int *ippos, int *dsdirp) 5011 { 5012 struct vnode *dvp, *nvp = NULL, **tdvpp; 5013 struct mount *mp; 5014 struct nfsmount *nmp, *newnmp; 5015 struct sockaddr *sad; 5016 struct sockaddr_in *sin; 5017 struct nfsdevice *ds, *tds, *fndds; 5018 struct pnfsdsfile *pf; 5019 uint32_t dsdir; 5020 int error, fhiszero, fnd, gotone, i, mirrorcnt; 5021 5022 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); 5023 *mirrorcntp = 1; 5024 tdvpp = dvpp; 5025 if (nvpp != NULL) 5026 *nvpp = NULL; 5027 if (dvpp != NULL) 5028 *dvpp = NULL; 5029 if (ippos != NULL) 5030 *ippos = -1; 5031 if (newnmpp != NULL) 5032 newnmp = *newnmpp; 5033 else 5034 newnmp = NULL; 5035 mp = vp->v_mount; 5036 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5037 "pnfsd.dsfile", buflenp, buf, p); 5038 mirrorcnt = *buflenp / sizeof(*pf); 5039 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || 5040 *buflenp != sizeof(*pf) * mirrorcnt)) 5041 error = ENOATTR; 5042 5043 pf = (struct pnfsdsfile *)buf; 5044 /* If curnmp != NULL, check for a match in the mirror list. */ 5045 if (curnmp != NULL && error == 0) { 5046 fnd = 0; 5047 for (i = 0; i < mirrorcnt; i++, pf++) { 5048 sad = (struct sockaddr *)&pf->dsf_sin; 5049 if (nfsaddr2_match(sad, curnmp->nm_nam)) { 5050 if (ippos != NULL) 5051 *ippos = i; 5052 fnd = 1; 5053 break; 5054 } 5055 } 5056 if (fnd == 0) 5057 error = ENXIO; 5058 } 5059 5060 gotone = 0; 5061 pf = (struct pnfsdsfile *)buf; 5062 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, 5063 error); 5064 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { 5065 fhiszero = 0; 5066 sad = (struct sockaddr *)&pf->dsf_sin; 5067 sin = &pf->dsf_sin; 5068 dsdir = pf->dsf_dir; 5069 if (dsdir >= nfsrv_dsdirsize) { 5070 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); 5071 error = ENOATTR; 5072 } else if (nvpp != NULL && newnmp != NULL && 5073 nfsaddr2_match(sad, newnmp->nm_nam)) 5074 error = EEXIST; 5075 if (error == 0) { 5076 if (ippos != NULL && curnmp == NULL && 5077 sad->sa_family == AF_INET && 5078 sin->sin_addr.s_addr == 0) 5079 *ippos = i; 5080 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) 5081 fhiszero = 1; 5082 /* Use the socket address to find the mount point. */ 5083 fndds = NULL; 5084 NFSDDSLOCK(); 5085 /* Find a match for the IP address. */ 5086 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5087 if (ds->nfsdev_nmp != NULL) { 5088 dvp = ds->nfsdev_dvp; 5089 nmp = VFSTONFS(dvp->v_mount); 5090 if (nmp != ds->nfsdev_nmp) 5091 printf("different2 nmp %p %p\n", 5092 nmp, ds->nfsdev_nmp); 5093 if (nfsaddr2_match(sad, nmp->nm_nam)) { 5094 fndds = ds; 5095 break; 5096 } 5097 } 5098 } 5099 if (fndds != NULL && newnmpp != NULL && 5100 newnmp == NULL) { 5101 /* Search for a place to make a mirror copy. */ 5102 TAILQ_FOREACH(tds, &nfsrv_devidhead, 5103 nfsdev_list) { 5104 if (tds->nfsdev_nmp != NULL && 5105 fndds != tds && 5106 ((tds->nfsdev_mdsisset == 0 && 5107 fndds->nfsdev_mdsisset == 0) || 5108 (tds->nfsdev_mdsisset != 0 && 5109 fndds->nfsdev_mdsisset != 0 && 5110 fsidcmp(&tds->nfsdev_mdsfsid, 5111 &mp->mnt_stat.f_fsid) == 0))) { 5112 *newnmpp = tds->nfsdev_nmp; 5113 break; 5114 } 5115 } 5116 if (tds != NULL) { 5117 /* 5118 * Move this entry to the end of the 5119 * list, so it won't be selected as 5120 * easily the next time. 5121 */ 5122 TAILQ_REMOVE(&nfsrv_devidhead, tds, 5123 nfsdev_list); 5124 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, 5125 nfsdev_list); 5126 } 5127 } 5128 NFSDDSUNLOCK(); 5129 if (fndds != NULL) { 5130 dvp = fndds->nfsdev_dsdir[dsdir]; 5131 if (lktype != 0 || fhiszero != 0 || 5132 (nvpp != NULL && *nvpp == NULL)) { 5133 if (fhiszero != 0) 5134 error = vn_lock(dvp, 5135 LK_EXCLUSIVE); 5136 else if (lktype != 0) 5137 error = vn_lock(dvp, lktype); 5138 else 5139 error = vn_lock(dvp, LK_SHARED); 5140 /* 5141 * If the file handle is all 0's, try to 5142 * do a Lookup against the DS to acquire 5143 * it. 5144 * If dvpp == NULL or the Lookup fails, 5145 * unlock dvp after the call. 5146 */ 5147 if (error == 0 && (fhiszero != 0 || 5148 (nvpp != NULL && *nvpp == NULL))) { 5149 error = nfsrv_pnfslookupds(vp, 5150 dvp, pf, &nvp, p); 5151 if (error == 0) { 5152 if (fhiszero != 0) 5153 nfsrv_pnfssetfh( 5154 vp, pf, 5155 devid, 5156 fnamep, 5157 nvp, p); 5158 if (nvpp != NULL && 5159 *nvpp == NULL) { 5160 *nvpp = nvp; 5161 *dsdirp = dsdir; 5162 } else 5163 vput(nvp); 5164 } 5165 if (error != 0 || lktype == 0) 5166 NFSVOPUNLOCK(dvp); 5167 } 5168 } 5169 if (error == 0) { 5170 gotone++; 5171 NFSD_DEBUG(4, "gotone=%d\n", gotone); 5172 if (devid != NULL) { 5173 NFSBCOPY(fndds->nfsdev_deviceid, 5174 devid, NFSX_V4DEVICEID); 5175 devid += NFSX_V4DEVICEID; 5176 } 5177 if (dvpp != NULL) 5178 *tdvpp++ = dvp; 5179 if (fhp != NULL) 5180 NFSBCOPY(&pf->dsf_fh, fhp++, 5181 NFSX_MYFH); 5182 if (fnamep != NULL && gotone == 1) 5183 strlcpy(fnamep, 5184 pf->dsf_filename, 5185 sizeof(pf->dsf_filename)); 5186 } else 5187 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " 5188 "err=%d\n", error); 5189 } 5190 } 5191 } 5192 if (error == 0 && gotone == 0) 5193 error = ENOENT; 5194 5195 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, 5196 error); 5197 if (error == 0) 5198 *mirrorcntp = gotone; 5199 else { 5200 if (gotone > 0 && dvpp != NULL) { 5201 /* 5202 * If the error didn't occur on the first one and 5203 * dvpp != NULL, the one(s) prior to the failure will 5204 * have locked dvp's that need to be unlocked. 5205 */ 5206 for (i = 0; i < gotone; i++) { 5207 NFSVOPUNLOCK(*dvpp); 5208 *dvpp++ = NULL; 5209 } 5210 } 5211 /* 5212 * If it found the vnode to be copied from before a failure, 5213 * it needs to be vput()'d. 5214 */ 5215 if (nvpp != NULL && *nvpp != NULL) { 5216 vput(*nvpp); 5217 *nvpp = NULL; 5218 } 5219 } 5220 return (error); 5221 } 5222 5223 /* 5224 * Set the extended attribute for the Change attribute. 5225 */ 5226 static int 5227 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 5228 { 5229 struct pnfsdsattr dsattr; 5230 int error; 5231 5232 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); 5233 dsattr.dsa_filerev = nap->na_filerev; 5234 dsattr.dsa_size = nap->na_size; 5235 dsattr.dsa_atime = nap->na_atime; 5236 dsattr.dsa_mtime = nap->na_mtime; 5237 dsattr.dsa_bytes = nap->na_bytes; 5238 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5239 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); 5240 if (error != 0) 5241 printf("pNFS: setextattr=%d\n", error); 5242 return (error); 5243 } 5244 5245 static int 5246 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5247 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) 5248 { 5249 uint32_t *tl; 5250 struct nfsrv_descript *nd; 5251 nfsv4stateid_t st; 5252 struct mbuf *m, *m2; 5253 int error = 0, retlen, tlen, trimlen; 5254 5255 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); 5256 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5257 *mpp = NULL; 5258 /* 5259 * Use a stateid where other is an alternating 01010 pattern and 5260 * seqid is 0xffffffff. This value is not defined as special by 5261 * the RFC and is used by the FreeBSD NFS server to indicate an 5262 * MDS->DS proxy operation. 5263 */ 5264 st.other[0] = 0x55555555; 5265 st.other[1] = 0x55555555; 5266 st.other[2] = 0x55555555; 5267 st.seqid = 0xffffffff; 5268 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5269 NULL, NULL, 0, 0, cred); 5270 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5271 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); 5272 txdr_hyper(off, tl); 5273 *(tl + 2) = txdr_unsigned(len); 5274 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5275 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5276 if (error != 0) { 5277 free(nd, M_TEMP); 5278 return (error); 5279 } 5280 if (nd->nd_repstat == 0) { 5281 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 5282 NFSM_STRSIZ(retlen, len); 5283 if (retlen > 0) { 5284 /* Trim off the pre-data XDR from the mbuf chain. */ 5285 m = nd->nd_mrep; 5286 while (m != NULL && m != nd->nd_md) { 5287 if (m->m_next == nd->nd_md) { 5288 m->m_next = NULL; 5289 m_freem(nd->nd_mrep); 5290 nd->nd_mrep = m = nd->nd_md; 5291 } else 5292 m = m->m_next; 5293 } 5294 if (m == NULL) { 5295 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5296 error = ENOENT; 5297 goto nfsmout; 5298 } 5299 5300 /* 5301 * Now, adjust first mbuf so that any XDR before the 5302 * read data is skipped over. 5303 */ 5304 trimlen = nd->nd_dpos - mtod(m, char *); 5305 if (trimlen > 0) { 5306 m->m_len -= trimlen; 5307 NFSM_DATAP(m, trimlen); 5308 } 5309 5310 /* 5311 * Truncate the mbuf chain at retlen bytes of data, 5312 * plus XDR padding that brings the length up to a 5313 * multiple of 4. 5314 */ 5315 tlen = NFSM_RNDUP(retlen); 5316 do { 5317 if (m->m_len >= tlen) { 5318 m->m_len = tlen; 5319 tlen = 0; 5320 m2 = m->m_next; 5321 m->m_next = NULL; 5322 m_freem(m2); 5323 break; 5324 } 5325 tlen -= m->m_len; 5326 m = m->m_next; 5327 } while (m != NULL); 5328 if (tlen > 0) { 5329 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5330 error = ENOENT; 5331 goto nfsmout; 5332 } 5333 *mpp = nd->nd_mrep; 5334 *mpendp = m; 5335 nd->nd_mrep = NULL; 5336 } 5337 } else 5338 error = nd->nd_repstat; 5339 nfsmout: 5340 /* If nd->nd_mrep is already NULL, this is a no-op. */ 5341 m_freem(nd->nd_mrep); 5342 free(nd, M_TEMP); 5343 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); 5344 return (error); 5345 } 5346 5347 /* 5348 * Do a write RPC on a DS data file, using this structure for the arguments, 5349 * so that this function can be executed by a separate kernel process. 5350 */ 5351 struct nfsrvwritedsdorpc { 5352 int done; 5353 int inprog; 5354 struct task tsk; 5355 fhandle_t fh; 5356 off_t off; 5357 int len; 5358 struct nfsmount *nmp; 5359 struct ucred *cred; 5360 NFSPROC_T *p; 5361 struct mbuf *m; 5362 int err; 5363 }; 5364 5365 static int 5366 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, 5367 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) 5368 { 5369 uint32_t *tl; 5370 struct nfsrv_descript *nd; 5371 nfsattrbit_t attrbits; 5372 nfsv4stateid_t st; 5373 int commit, error, retlen; 5374 5375 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5376 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, 5377 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5378 5379 /* 5380 * Use a stateid where other is an alternating 01010 pattern and 5381 * seqid is 0xffffffff. This value is not defined as special by 5382 * the RFC and is used by the FreeBSD NFS server to indicate an 5383 * MDS->DS proxy operation. 5384 */ 5385 st.other[0] = 0x55555555; 5386 st.other[1] = 0x55555555; 5387 st.other[2] = 0x55555555; 5388 st.seqid = 0xffffffff; 5389 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5390 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); 5391 txdr_hyper(off, tl); 5392 tl += 2; 5393 /* 5394 * Do all writes FileSync, since the server doesn't hold onto dirty 5395 * buffers. Since clients should be accessing the DS servers directly 5396 * using the pNFS layouts, this just needs to work correctly as a 5397 * fallback. 5398 */ 5399 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); 5400 *tl = txdr_unsigned(len); 5401 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); 5402 5403 /* Put data in mbuf chain. */ 5404 nd->nd_mb->m_next = m; 5405 5406 /* Set nd_mb and nd_bpos to end of data. */ 5407 while (m->m_next != NULL) 5408 m = m->m_next; 5409 nd->nd_mb = m; 5410 nfsm_set(nd, m->m_len); 5411 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); 5412 5413 /* Do a Getattr for the attributes that change upon writing. */ 5414 NFSZERO_ATTRBIT(&attrbits); 5415 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5416 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5417 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5418 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5419 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5420 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5421 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5422 (void) nfsrv_putattrbit(nd, &attrbits); 5423 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5424 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5425 if (error != 0) { 5426 free(nd, M_TEMP); 5427 return (error); 5428 } 5429 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); 5430 /* Get rid of weak cache consistency data for now. */ 5431 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5432 (ND_NFSV4 | ND_V4WCCATTR)) { 5433 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5434 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5435 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); 5436 if (error != 0) 5437 goto nfsmout; 5438 /* 5439 * Get rid of Op# and status for next op. 5440 */ 5441 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5442 if (*++tl != 0) 5443 nd->nd_flag |= ND_NOMOREDATA; 5444 } 5445 if (nd->nd_repstat == 0) { 5446 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); 5447 retlen = fxdr_unsigned(int, *tl++); 5448 commit = fxdr_unsigned(int, *tl); 5449 if (commit != NFSWRITE_FILESYNC) 5450 error = NFSERR_IO; 5451 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", 5452 retlen, commit, error); 5453 } else 5454 error = nd->nd_repstat; 5455 /* We have no use for the Write Verifier since we use FileSync. */ 5456 5457 /* 5458 * Get the Change, Size, Access Time and Modify Time attributes and set 5459 * on the Metadata file, so its attributes will be what the file's 5460 * would be if it had been written. 5461 */ 5462 if (error == 0) { 5463 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5464 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5465 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5466 } 5467 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); 5468 nfsmout: 5469 m_freem(nd->nd_mrep); 5470 free(nd, M_TEMP); 5471 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); 5472 return (error); 5473 } 5474 5475 /* 5476 * Start up the thread that will execute nfsrv_writedsdorpc(). 5477 */ 5478 static void 5479 start_writedsdorpc(void *arg, int pending) 5480 { 5481 struct nfsrvwritedsdorpc *drpc; 5482 5483 drpc = (struct nfsrvwritedsdorpc *)arg; 5484 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5485 drpc->len, NULL, drpc->m, drpc->cred, drpc->p); 5486 drpc->done = 1; 5487 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); 5488 } 5489 5490 static int 5491 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5492 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5493 struct mbuf **mpp, char *cp, int *failposp) 5494 { 5495 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; 5496 struct nfsvattr na; 5497 struct mbuf *m; 5498 int error, i, offs, ret, timo; 5499 5500 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); 5501 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); 5502 drpc = NULL; 5503 if (mirrorcnt > 1) 5504 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5505 M_WAITOK); 5506 5507 /* Calculate offset in mbuf chain that data starts. */ 5508 offs = cp - mtod(*mpp, char *); 5509 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); 5510 5511 /* 5512 * Do the write RPC for every DS, using a separate kernel process 5513 * for every DS except the last one. 5514 */ 5515 error = 0; 5516 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5517 tdrpc->done = 0; 5518 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5519 tdrpc->off = off; 5520 tdrpc->len = len; 5521 tdrpc->nmp = *nmpp; 5522 tdrpc->cred = cred; 5523 tdrpc->p = p; 5524 tdrpc->inprog = 0; 5525 tdrpc->err = 0; 5526 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5527 ret = EIO; 5528 if (nfs_pnfsiothreads != 0) { 5529 ret = nfs_pnfsio(start_writedsdorpc, tdrpc); 5530 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", 5531 ret); 5532 } 5533 if (ret != 0) { 5534 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, 5535 tdrpc->m, cred, p); 5536 if (nfsds_failerr(ret) && *failposp == -1) 5537 *failposp = i; 5538 else if (error == 0 && ret != 0) 5539 error = ret; 5540 } 5541 nmpp++; 5542 fhp++; 5543 } 5544 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5545 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); 5546 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5547 *failposp = mirrorcnt - 1; 5548 else if (error == 0 && ret != 0) 5549 error = ret; 5550 if (error == 0) 5551 error = nfsrv_setextattr(vp, &na, p); 5552 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); 5553 tdrpc = drpc; 5554 timo = hz / 50; /* Wait for 20msec. */ 5555 if (timo < 1) 5556 timo = 1; 5557 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5558 /* Wait for RPCs on separate threads to complete. */ 5559 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5560 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); 5561 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5562 *failposp = i; 5563 else if (error == 0 && tdrpc->err != 0) 5564 error = tdrpc->err; 5565 } 5566 free(drpc, M_TEMP); 5567 return (error); 5568 } 5569 5570 /* 5571 * Do a allocate RPC on a DS data file, using this structure for the arguments, 5572 * so that this function can be executed by a separate kernel process. 5573 */ 5574 struct nfsrvallocatedsdorpc { 5575 int done; 5576 int inprog; 5577 struct task tsk; 5578 fhandle_t fh; 5579 off_t off; 5580 off_t len; 5581 struct nfsmount *nmp; 5582 struct ucred *cred; 5583 NFSPROC_T *p; 5584 int err; 5585 }; 5586 5587 static int 5588 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5589 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5590 { 5591 uint32_t *tl; 5592 struct nfsrv_descript *nd; 5593 nfsattrbit_t attrbits; 5594 nfsv4stateid_t st; 5595 int error; 5596 5597 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5598 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, 5599 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5600 5601 /* 5602 * Use a stateid where other is an alternating 01010 pattern and 5603 * seqid is 0xffffffff. This value is not defined as special by 5604 * the RFC and is used by the FreeBSD NFS server to indicate an 5605 * MDS->DS proxy operation. 5606 */ 5607 st.other[0] = 0x55555555; 5608 st.other[1] = 0x55555555; 5609 st.other[2] = 0x55555555; 5610 st.seqid = 0xffffffff; 5611 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5612 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5613 txdr_hyper(off, tl); tl += 2; 5614 txdr_hyper(len, tl); tl += 2; 5615 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); 5616 5617 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5618 NFSGETATTR_ATTRBIT(&attrbits); 5619 nfsrv_putattrbit(nd, &attrbits); 5620 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5621 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5622 if (error != 0) { 5623 free(nd, M_TEMP); 5624 return (error); 5625 } 5626 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", 5627 nd->nd_repstat); 5628 if (nd->nd_repstat == 0) { 5629 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5630 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5631 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5632 } else 5633 error = nd->nd_repstat; 5634 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); 5635 nfsmout: 5636 m_freem(nd->nd_mrep); 5637 free(nd, M_TEMP); 5638 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); 5639 return (error); 5640 } 5641 5642 /* 5643 * Start up the thread that will execute nfsrv_allocatedsdorpc(). 5644 */ 5645 static void 5646 start_allocatedsdorpc(void *arg, int pending) 5647 { 5648 struct nfsrvallocatedsdorpc *drpc; 5649 5650 drpc = (struct nfsrvallocatedsdorpc *)arg; 5651 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5652 drpc->len, NULL, drpc->cred, drpc->p); 5653 drpc->done = 1; 5654 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); 5655 } 5656 5657 static int 5658 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5659 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5660 int *failposp) 5661 { 5662 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; 5663 struct nfsvattr na; 5664 int error, i, ret, timo; 5665 5666 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); 5667 drpc = NULL; 5668 if (mirrorcnt > 1) 5669 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5670 M_WAITOK); 5671 5672 /* 5673 * Do the allocate RPC for every DS, using a separate kernel process 5674 * for every DS except the last one. 5675 */ 5676 error = 0; 5677 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5678 tdrpc->done = 0; 5679 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5680 tdrpc->off = off; 5681 tdrpc->len = len; 5682 tdrpc->nmp = *nmpp; 5683 tdrpc->cred = cred; 5684 tdrpc->p = p; 5685 tdrpc->inprog = 0; 5686 tdrpc->err = 0; 5687 ret = EIO; 5688 if (nfs_pnfsiothreads != 0) { 5689 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); 5690 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", 5691 ret); 5692 } 5693 if (ret != 0) { 5694 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, 5695 cred, p); 5696 if (nfsds_failerr(ret) && *failposp == -1) 5697 *failposp = i; 5698 else if (error == 0 && ret != 0) 5699 error = ret; 5700 } 5701 nmpp++; 5702 fhp++; 5703 } 5704 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5705 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5706 *failposp = mirrorcnt - 1; 5707 else if (error == 0 && ret != 0) 5708 error = ret; 5709 if (error == 0) 5710 error = nfsrv_setextattr(vp, &na, p); 5711 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); 5712 tdrpc = drpc; 5713 timo = hz / 50; /* Wait for 20msec. */ 5714 if (timo < 1) 5715 timo = 1; 5716 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5717 /* Wait for RPCs on separate threads to complete. */ 5718 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5719 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5720 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5721 *failposp = i; 5722 else if (error == 0 && tdrpc->err != 0) 5723 error = tdrpc->err; 5724 } 5725 free(drpc, M_TEMP); 5726 return (error); 5727 } 5728 5729 /* 5730 * Do a deallocate RPC on a DS data file, using this structure for the 5731 * arguments, so that this function can be executed by a separate kernel 5732 * process. 5733 */ 5734 struct nfsrvdeallocatedsdorpc { 5735 int done; 5736 int inprog; 5737 struct task tsk; 5738 fhandle_t fh; 5739 off_t off; 5740 off_t len; 5741 struct nfsmount *nmp; 5742 struct ucred *cred; 5743 NFSPROC_T *p; 5744 int err; 5745 }; 5746 5747 static int 5748 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5749 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5750 { 5751 uint32_t *tl; 5752 struct nfsrv_descript *nd; 5753 nfsattrbit_t attrbits; 5754 nfsv4stateid_t st; 5755 int error; 5756 5757 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5758 nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp, 5759 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5760 5761 /* 5762 * Use a stateid where other is an alternating 01010 pattern and 5763 * seqid is 0xffffffff. This value is not defined as special by 5764 * the RFC and is used by the FreeBSD NFS server to indicate an 5765 * MDS->DS proxy operation. 5766 */ 5767 st.other[0] = 0x55555555; 5768 st.other[1] = 0x55555555; 5769 st.other[2] = 0x55555555; 5770 st.seqid = 0xffffffff; 5771 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5772 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5773 txdr_hyper(off, tl); tl += 2; 5774 txdr_hyper(len, tl); tl += 2; 5775 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len); 5776 5777 /* Do a Getattr for the attributes that change upon writing. */ 5778 NFSZERO_ATTRBIT(&attrbits); 5779 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5780 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5781 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5782 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5783 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5784 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5785 nfsrv_putattrbit(nd, &attrbits); 5786 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5787 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5788 if (error != 0) { 5789 free(nd, M_TEMP); 5790 return (error); 5791 } 5792 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n", 5793 nd->nd_repstat); 5794 /* Get rid of weak cache consistency data for now. */ 5795 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5796 (ND_NFSV4 | ND_V4WCCATTR)) { 5797 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5798 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5799 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error); 5800 if (error != 0) 5801 goto nfsmout; 5802 /* 5803 * Get rid of Op# and status for next op. 5804 */ 5805 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5806 if (*++tl != 0) 5807 nd->nd_flag |= ND_NOMOREDATA; 5808 } 5809 if (nd->nd_repstat == 0) { 5810 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5811 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5812 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5813 } else 5814 error = nd->nd_repstat; 5815 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error); 5816 nfsmout: 5817 m_freem(nd->nd_mrep); 5818 free(nd, M_TEMP); 5819 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error); 5820 return (error); 5821 } 5822 5823 /* 5824 * Start up the thread that will execute nfsrv_deallocatedsdorpc(). 5825 */ 5826 static void 5827 start_deallocatedsdorpc(void *arg, int pending) 5828 { 5829 struct nfsrvdeallocatedsdorpc *drpc; 5830 5831 drpc = (struct nfsrvdeallocatedsdorpc *)arg; 5832 drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5833 drpc->len, NULL, drpc->cred, drpc->p); 5834 drpc->done = 1; 5835 NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err); 5836 } 5837 5838 static int 5839 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5840 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5841 int *failposp) 5842 { 5843 struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL; 5844 struct nfsvattr na; 5845 int error, i, ret, timo; 5846 5847 NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n"); 5848 drpc = NULL; 5849 if (mirrorcnt > 1) 5850 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5851 M_WAITOK); 5852 5853 /* 5854 * Do the deallocate RPC for every DS, using a separate kernel process 5855 * for every DS except the last one. 5856 */ 5857 error = 0; 5858 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5859 tdrpc->done = 0; 5860 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5861 tdrpc->off = off; 5862 tdrpc->len = len; 5863 tdrpc->nmp = *nmpp; 5864 tdrpc->cred = cred; 5865 tdrpc->p = p; 5866 tdrpc->inprog = 0; 5867 tdrpc->err = 0; 5868 ret = EIO; 5869 if (nfs_pnfsiothreads != 0) { 5870 ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc); 5871 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n", 5872 ret); 5873 } 5874 if (ret != 0) { 5875 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, 5876 NULL, cred, p); 5877 if (nfsds_failerr(ret) && *failposp == -1) 5878 *failposp = i; 5879 else if (error == 0 && ret != 0) 5880 error = ret; 5881 } 5882 nmpp++; 5883 fhp++; 5884 } 5885 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5886 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5887 *failposp = mirrorcnt - 1; 5888 else if (error == 0 && ret != 0) 5889 error = ret; 5890 if (error == 0) 5891 error = nfsrv_setextattr(vp, &na, p); 5892 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error); 5893 tdrpc = drpc; 5894 timo = hz / 50; /* Wait for 20msec. */ 5895 if (timo < 1) 5896 timo = 1; 5897 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5898 /* Wait for RPCs on separate threads to complete. */ 5899 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5900 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5901 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5902 *failposp = i; 5903 else if (error == 0 && tdrpc->err != 0) 5904 error = tdrpc->err; 5905 } 5906 free(drpc, M_TEMP); 5907 return (error); 5908 } 5909 5910 static int 5911 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5912 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, 5913 struct nfsvattr *dsnap) 5914 { 5915 uint32_t *tl; 5916 struct nfsrv_descript *nd; 5917 nfsv4stateid_t st; 5918 nfsattrbit_t attrbits; 5919 int error; 5920 5921 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); 5922 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5923 /* 5924 * Use a stateid where other is an alternating 01010 pattern and 5925 * seqid is 0xffffffff. This value is not defined as special by 5926 * the RFC and is used by the FreeBSD NFS server to indicate an 5927 * MDS->DS proxy operation. 5928 */ 5929 st.other[0] = 0x55555555; 5930 st.other[1] = 0x55555555; 5931 st.other[2] = 0x55555555; 5932 st.seqid = 0xffffffff; 5933 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5934 NULL, NULL, 0, 0, cred); 5935 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5936 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); 5937 5938 /* Do a Getattr for the attributes that change due to writing. */ 5939 NFSZERO_ATTRBIT(&attrbits); 5940 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5941 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5942 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5943 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5944 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5945 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5946 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5947 (void) nfsrv_putattrbit(nd, &attrbits); 5948 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5949 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5950 if (error != 0) { 5951 free(nd, M_TEMP); 5952 return (error); 5953 } 5954 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", 5955 nd->nd_repstat); 5956 /* Get rid of weak cache consistency data for now. */ 5957 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5958 (ND_NFSV4 | ND_V4WCCATTR)) { 5959 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5960 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5961 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); 5962 if (error != 0) 5963 goto nfsmout; 5964 /* 5965 * Get rid of Op# and status for next op. 5966 */ 5967 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5968 if (*++tl != 0) 5969 nd->nd_flag |= ND_NOMOREDATA; 5970 } 5971 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 5972 if (error != 0) 5973 goto nfsmout; 5974 if (nd->nd_repstat != 0) 5975 error = nd->nd_repstat; 5976 /* 5977 * Get the Change, Size, Access Time and Modify Time attributes and set 5978 * on the Metadata file, so its attributes will be what the file's 5979 * would be if it had been written. 5980 */ 5981 if (error == 0) { 5982 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5983 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5984 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5985 } 5986 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); 5987 nfsmout: 5988 m_freem(nd->nd_mrep); 5989 free(nd, M_TEMP); 5990 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); 5991 return (error); 5992 } 5993 5994 struct nfsrvsetattrdsdorpc { 5995 int done; 5996 int inprog; 5997 struct task tsk; 5998 fhandle_t fh; 5999 struct nfsmount *nmp; 6000 struct vnode *vp; 6001 struct ucred *cred; 6002 NFSPROC_T *p; 6003 struct nfsvattr na; 6004 struct nfsvattr dsna; 6005 int err; 6006 }; 6007 6008 /* 6009 * Start up the thread that will execute nfsrv_setattrdsdorpc(). 6010 */ 6011 static void 6012 start_setattrdsdorpc(void *arg, int pending) 6013 { 6014 struct nfsrvsetattrdsdorpc *drpc; 6015 6016 drpc = (struct nfsrvsetattrdsdorpc *)arg; 6017 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, 6018 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); 6019 drpc->done = 1; 6020 } 6021 6022 static int 6023 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6024 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 6025 struct nfsvattr *nap, int *failposp) 6026 { 6027 struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; 6028 struct nfsvattr na; 6029 int error, i, ret, timo; 6030 6031 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); 6032 drpc = NULL; 6033 if (mirrorcnt > 1) 6034 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6035 M_WAITOK); 6036 6037 /* 6038 * Do the setattr RPC for every DS, using a separate kernel process 6039 * for every DS except the last one. 6040 */ 6041 error = 0; 6042 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6043 tdrpc->done = 0; 6044 tdrpc->inprog = 0; 6045 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6046 tdrpc->nmp = *nmpp; 6047 tdrpc->vp = vp; 6048 tdrpc->cred = cred; 6049 tdrpc->p = p; 6050 tdrpc->na = *nap; 6051 tdrpc->err = 0; 6052 ret = EIO; 6053 if (nfs_pnfsiothreads != 0) { 6054 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); 6055 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", 6056 ret); 6057 } 6058 if (ret != 0) { 6059 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, 6060 &na); 6061 if (nfsds_failerr(ret) && *failposp == -1) 6062 *failposp = i; 6063 else if (error == 0 && ret != 0) 6064 error = ret; 6065 } 6066 nmpp++; 6067 fhp++; 6068 } 6069 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); 6070 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6071 *failposp = mirrorcnt - 1; 6072 else if (error == 0 && ret != 0) 6073 error = ret; 6074 if (error == 0) 6075 error = nfsrv_setextattr(vp, &na, p); 6076 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); 6077 tdrpc = drpc; 6078 timo = hz / 50; /* Wait for 20msec. */ 6079 if (timo < 1) 6080 timo = 1; 6081 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6082 /* Wait for RPCs on separate threads to complete. */ 6083 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6084 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); 6085 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6086 *failposp = i; 6087 else if (error == 0 && tdrpc->err != 0) 6088 error = tdrpc->err; 6089 } 6090 free(drpc, M_TEMP); 6091 return (error); 6092 } 6093 6094 /* 6095 * Do a Setattr of an NFSv4 ACL on the DS file. 6096 */ 6097 static int 6098 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6099 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) 6100 { 6101 struct nfsrv_descript *nd; 6102 nfsv4stateid_t st; 6103 nfsattrbit_t attrbits; 6104 int error; 6105 6106 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); 6107 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6108 /* 6109 * Use a stateid where other is an alternating 01010 pattern and 6110 * seqid is 0xffffffff. This value is not defined as special by 6111 * the RFC and is used by the FreeBSD NFS server to indicate an 6112 * MDS->DS proxy operation. 6113 */ 6114 st.other[0] = 0x55555555; 6115 st.other[1] = 0x55555555; 6116 st.other[2] = 0x55555555; 6117 st.seqid = 0xffffffff; 6118 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), 6119 NULL, NULL, 0, 0, cred); 6120 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6121 NFSZERO_ATTRBIT(&attrbits); 6122 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); 6123 /* 6124 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), 6125 * so passing in the metadata "vp" will be ok, since it is of 6126 * the same type (VREG). 6127 */ 6128 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, 6129 NULL, 0, 0, 0, 0, 0, NULL); 6130 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6131 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6132 if (error != 0) { 6133 free(nd, M_TEMP); 6134 return (error); 6135 } 6136 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", 6137 nd->nd_repstat); 6138 error = nd->nd_repstat; 6139 m_freem(nd->nd_mrep); 6140 free(nd, M_TEMP); 6141 return (error); 6142 } 6143 6144 struct nfsrvsetacldsdorpc { 6145 int done; 6146 int inprog; 6147 struct task tsk; 6148 fhandle_t fh; 6149 struct nfsmount *nmp; 6150 struct vnode *vp; 6151 struct ucred *cred; 6152 NFSPROC_T *p; 6153 struct acl *aclp; 6154 int err; 6155 }; 6156 6157 /* 6158 * Start up the thread that will execute nfsrv_setacldsdorpc(). 6159 */ 6160 static void 6161 start_setacldsdorpc(void *arg, int pending) 6162 { 6163 struct nfsrvsetacldsdorpc *drpc; 6164 6165 drpc = (struct nfsrvsetacldsdorpc *)arg; 6166 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, 6167 drpc->vp, drpc->nmp, drpc->aclp); 6168 drpc->done = 1; 6169 } 6170 6171 static int 6172 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6173 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, 6174 int *failposp) 6175 { 6176 struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; 6177 int error, i, ret, timo; 6178 6179 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); 6180 drpc = NULL; 6181 if (mirrorcnt > 1) 6182 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6183 M_WAITOK); 6184 6185 /* 6186 * Do the setattr RPC for every DS, using a separate kernel process 6187 * for every DS except the last one. 6188 */ 6189 error = 0; 6190 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6191 tdrpc->done = 0; 6192 tdrpc->inprog = 0; 6193 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6194 tdrpc->nmp = *nmpp; 6195 tdrpc->vp = vp; 6196 tdrpc->cred = cred; 6197 tdrpc->p = p; 6198 tdrpc->aclp = aclp; 6199 tdrpc->err = 0; 6200 ret = EIO; 6201 if (nfs_pnfsiothreads != 0) { 6202 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); 6203 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", 6204 ret); 6205 } 6206 if (ret != 0) { 6207 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, 6208 aclp); 6209 if (nfsds_failerr(ret) && *failposp == -1) 6210 *failposp = i; 6211 else if (error == 0 && ret != 0) 6212 error = ret; 6213 } 6214 nmpp++; 6215 fhp++; 6216 } 6217 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); 6218 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6219 *failposp = mirrorcnt - 1; 6220 else if (error == 0 && ret != 0) 6221 error = ret; 6222 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); 6223 tdrpc = drpc; 6224 timo = hz / 50; /* Wait for 20msec. */ 6225 if (timo < 1) 6226 timo = 1; 6227 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6228 /* Wait for RPCs on separate threads to complete. */ 6229 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6230 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); 6231 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6232 *failposp = i; 6233 else if (error == 0 && tdrpc->err != 0) 6234 error = tdrpc->err; 6235 } 6236 free(drpc, M_TEMP); 6237 return (error); 6238 } 6239 6240 /* 6241 * Getattr call to the DS for the attributes that change due to writing. 6242 */ 6243 static int 6244 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6245 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) 6246 { 6247 struct nfsrv_descript *nd; 6248 int error; 6249 nfsattrbit_t attrbits; 6250 6251 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); 6252 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6253 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, 6254 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6255 NFSZERO_ATTRBIT(&attrbits); 6256 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6257 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6258 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6259 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6260 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6261 (void) nfsrv_putattrbit(nd, &attrbits); 6262 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6263 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6264 if (error != 0) { 6265 free(nd, M_TEMP); 6266 return (error); 6267 } 6268 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", 6269 nd->nd_repstat); 6270 if (nd->nd_repstat == 0) { 6271 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, 6272 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, 6273 NULL, NULL); 6274 /* 6275 * We can only save the updated values in the extended 6276 * attribute if the vp is exclusively locked. 6277 * This should happen when any of the following operations 6278 * occur on the vnode: 6279 * Close, Delegreturn, LayoutCommit, LayoutReturn 6280 * As such, the updated extended attribute should get saved 6281 * before nfsrv_checkdsattr() returns 0 and allows the cached 6282 * attributes to be returned without calling this function. 6283 */ 6284 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { 6285 error = nfsrv_setextattr(vp, nap, p); 6286 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", 6287 error); 6288 } 6289 } else 6290 error = nd->nd_repstat; 6291 m_freem(nd->nd_mrep); 6292 free(nd, M_TEMP); 6293 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); 6294 return (error); 6295 } 6296 6297 /* 6298 * Seek call to a DS. 6299 */ 6300 static int 6301 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, 6302 struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) 6303 { 6304 uint32_t *tl; 6305 struct nfsrv_descript *nd; 6306 nfsv4stateid_t st; 6307 int error; 6308 6309 NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); 6310 /* 6311 * Use a stateid where other is an alternating 01010 pattern and 6312 * seqid is 0xffffffff. This value is not defined as special by 6313 * the RFC and is used by the FreeBSD NFS server to indicate an 6314 * MDS->DS proxy operation. 6315 */ 6316 st.other[0] = 0x55555555; 6317 st.other[1] = 0x55555555; 6318 st.other[2] = 0x55555555; 6319 st.seqid = 0xffffffff; 6320 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6321 nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, 6322 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6323 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6324 NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); 6325 txdr_hyper(*offp, tl); tl += 2; 6326 *tl = txdr_unsigned(content); 6327 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6328 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6329 if (error != 0) { 6330 free(nd, M_TEMP); 6331 return (error); 6332 } 6333 NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); 6334 if (nd->nd_repstat == 0) { 6335 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); 6336 if (*tl++ == newnfs_true) 6337 *eofp = true; 6338 else 6339 *eofp = false; 6340 *offp = fxdr_hyper(tl); 6341 } else 6342 error = nd->nd_repstat; 6343 nfsmout: 6344 m_freem(nd->nd_mrep); 6345 free(nd, M_TEMP); 6346 NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); 6347 return (error); 6348 } 6349 6350 /* 6351 * Get the device id and file handle for a DS file. 6352 */ 6353 int 6354 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, 6355 fhandle_t *fhp, char *devid) 6356 { 6357 int buflen, error; 6358 char *buf; 6359 6360 buflen = 1024; 6361 buf = malloc(buflen, M_TEMP, M_WAITOK); 6362 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, 6363 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); 6364 free(buf, M_TEMP); 6365 return (error); 6366 } 6367 6368 /* 6369 * Do a Lookup against the DS for the filename. 6370 */ 6371 static int 6372 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, 6373 struct vnode **nvpp, NFSPROC_T *p) 6374 { 6375 struct nameidata named; 6376 struct ucred *tcred; 6377 char *bufp; 6378 u_long *hashp; 6379 struct vnode *nvp; 6380 int error; 6381 6382 tcred = newnfs_getcred(); 6383 named.ni_cnd.cn_nameiop = LOOKUP; 6384 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; 6385 named.ni_cnd.cn_cred = tcred; 6386 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 6387 nfsvno_setpathbuf(&named, &bufp, &hashp); 6388 named.ni_cnd.cn_nameptr = bufp; 6389 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); 6390 strlcpy(bufp, pf->dsf_filename, NAME_MAX); 6391 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); 6392 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 6393 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); 6394 NFSFREECRED(tcred); 6395 nfsvno_relpathbuf(&named); 6396 if (error == 0) 6397 *nvpp = nvp; 6398 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); 6399 return (error); 6400 } 6401 6402 /* 6403 * Set the file handle to the correct one. 6404 */ 6405 static void 6406 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, 6407 char *fnamep, struct vnode *nvp, NFSPROC_T *p) 6408 { 6409 struct nfsnode *np; 6410 int ret = 0; 6411 6412 np = VTONFS(nvp); 6413 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); 6414 /* 6415 * We can only do a vn_set_extattr() if the vnode is exclusively 6416 * locked and vn_start_write() has been done. If devid != NULL or 6417 * fnamep != NULL or the vnode is shared locked, vn_start_write() 6418 * may not have been done. 6419 * If not done now, it will be done on a future call. 6420 */ 6421 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == 6422 LK_EXCLUSIVE) 6423 ret = vn_extattr_set(vp, IO_NODELOCKED, 6424 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), 6425 (char *)pf, p); 6426 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); 6427 } 6428 6429 /* 6430 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point 6431 * when the DS has failed. 6432 */ 6433 void 6434 nfsrv_killrpcs(struct nfsmount *nmp) 6435 { 6436 6437 /* 6438 * Call newnfs_nmcancelreqs() to cause 6439 * any RPCs in progress on the mount point to 6440 * fail. 6441 * This will cause any process waiting for an 6442 * RPC to complete while holding a vnode lock 6443 * on the mounted-on vnode (such as "df" or 6444 * a non-forced "umount") to fail. 6445 * This will unlock the mounted-on vnode so 6446 * a forced dismount can succeed. 6447 * The NFSMNTP_CANCELRPCS flag should be set when this function is 6448 * called. 6449 */ 6450 newnfs_nmcancelreqs(nmp); 6451 } 6452 6453 /* 6454 * Sum up the statfs info for each of the DSs, so that the client will 6455 * receive the total for all DSs. 6456 */ 6457 static int 6458 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) 6459 { 6460 struct statfs *tsf; 6461 struct nfsdevice *ds; 6462 struct vnode **dvpp, **tdvpp, *dvp; 6463 uint64_t tot; 6464 int cnt, error = 0, i; 6465 6466 if (nfsrv_devidcnt <= 0) 6467 return (ENXIO); 6468 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 6469 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 6470 6471 /* Get an array of the dvps for the DSs. */ 6472 tdvpp = dvpp; 6473 i = 0; 6474 NFSDDSLOCK(); 6475 /* First, search for matches for same file system. */ 6476 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6477 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && 6478 fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { 6479 if (++i > nfsrv_devidcnt) 6480 break; 6481 *tdvpp++ = ds->nfsdev_dvp; 6482 } 6483 } 6484 /* 6485 * If no matches for same file system, total all servers not assigned 6486 * to a file system. 6487 */ 6488 if (i == 0) { 6489 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6490 if (ds->nfsdev_nmp != NULL && 6491 ds->nfsdev_mdsisset == 0) { 6492 if (++i > nfsrv_devidcnt) 6493 break; 6494 *tdvpp++ = ds->nfsdev_dvp; 6495 } 6496 } 6497 } 6498 NFSDDSUNLOCK(); 6499 cnt = i; 6500 6501 /* Do a VFS_STATFS() for each of the DSs and sum them up. */ 6502 tdvpp = dvpp; 6503 for (i = 0; i < cnt && error == 0; i++) { 6504 dvp = *tdvpp++; 6505 error = VFS_STATFS(dvp->v_mount, tsf); 6506 if (error == 0) { 6507 if (sf->f_bsize == 0) { 6508 if (tsf->f_bsize > 0) 6509 sf->f_bsize = tsf->f_bsize; 6510 else 6511 sf->f_bsize = 8192; 6512 } 6513 if (tsf->f_blocks > 0) { 6514 if (sf->f_bsize != tsf->f_bsize) { 6515 tot = tsf->f_blocks * tsf->f_bsize; 6516 sf->f_blocks += (tot / sf->f_bsize); 6517 } else 6518 sf->f_blocks += tsf->f_blocks; 6519 } 6520 if (tsf->f_bfree > 0) { 6521 if (sf->f_bsize != tsf->f_bsize) { 6522 tot = tsf->f_bfree * tsf->f_bsize; 6523 sf->f_bfree += (tot / sf->f_bsize); 6524 } else 6525 sf->f_bfree += tsf->f_bfree; 6526 } 6527 if (tsf->f_bavail > 0) { 6528 if (sf->f_bsize != tsf->f_bsize) { 6529 tot = tsf->f_bavail * tsf->f_bsize; 6530 sf->f_bavail += (tot / sf->f_bsize); 6531 } else 6532 sf->f_bavail += tsf->f_bavail; 6533 } 6534 } 6535 } 6536 free(tsf, M_TEMP); 6537 free(dvpp, M_TEMP); 6538 return (error); 6539 } 6540 6541 /* 6542 * Set an NFSv4 acl. 6543 */ 6544 int 6545 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) 6546 { 6547 int error; 6548 6549 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { 6550 error = NFSERR_ATTRNOTSUPP; 6551 goto out; 6552 } 6553 /* 6554 * With NFSv4 ACLs, chmod(2) may need to add additional entries. 6555 * Make sure it has enough room for that - splitting every entry 6556 * into two and appending "canonical six" entries at the end. 6557 * Cribbed out of kern/vfs_acl.c - Rick M. 6558 */ 6559 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { 6560 error = NFSERR_ATTRNOTSUPP; 6561 goto out; 6562 } 6563 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 6564 if (error == 0) { 6565 error = nfsrv_dssetacl(vp, aclp, cred, p); 6566 if (error == ENOENT) 6567 error = 0; 6568 } 6569 6570 out: 6571 NFSEXITCODE(error); 6572 return (error); 6573 } 6574 6575 /* 6576 * Seek vnode op call (actually it is a VOP_IOCTL()). 6577 * This function is called with the vnode locked, but unlocks and vrele()s 6578 * the vp before returning. 6579 */ 6580 int 6581 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, 6582 off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) 6583 { 6584 struct nfsvattr at; 6585 int error, ret; 6586 6587 ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); 6588 /* 6589 * Attempt to seek on a DS file. A return of ENOENT implies 6590 * there is no DS file to seek on. 6591 */ 6592 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, 6593 NULL, NULL, NULL, NULL, offp, content, eofp); 6594 if (error != ENOENT) { 6595 vput(vp); 6596 return (error); 6597 } 6598 6599 /* 6600 * Do the VOP_IOCTL() call. For the case where *offp == file_size, 6601 * VOP_IOCTL() will return ENXIO. However, the correct reply for 6602 * NFSv4.2 is *eofp == true and error == 0 for this case. 6603 */ 6604 NFSVOPUNLOCK(vp); 6605 error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); 6606 *eofp = false; 6607 if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { 6608 /* Handle the cases where we might be at EOF. */ 6609 ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); 6610 if (ret == 0 && *offp == at.na_size) { 6611 *eofp = true; 6612 error = 0; 6613 } 6614 if (ret != 0 && error == 0) 6615 error = ret; 6616 } 6617 vrele(vp); 6618 NFSEXITCODE(error); 6619 return (error); 6620 } 6621 6622 /* 6623 * Allocate vnode op call. 6624 */ 6625 int 6626 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6627 NFSPROC_T *p) 6628 { 6629 int error; 6630 off_t olen; 6631 6632 ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); 6633 /* 6634 * Attempt to allocate on a DS file. A return of ENOENT implies 6635 * there is no DS file to allocate on. 6636 */ 6637 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, 6638 NULL, NULL, NULL, NULL, &len, 0, NULL); 6639 if (error != ENOENT) 6640 return (error); 6641 6642 /* 6643 * Do the actual VOP_ALLOCATE(), looping so long as 6644 * progress is being made, to achieve completion. 6645 */ 6646 do { 6647 olen = len; 6648 error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred); 6649 if (error == 0 && len > 0 && olen > len) 6650 maybe_yield(); 6651 } while (error == 0 && len > 0 && olen > len); 6652 if (error == 0 && len > 0) 6653 error = NFSERR_IO; 6654 NFSEXITCODE(error); 6655 return (error); 6656 } 6657 6658 /* 6659 * Deallocate vnode op call. 6660 */ 6661 int 6662 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6663 NFSPROC_T *p) 6664 { 6665 int error; 6666 off_t olen; 6667 6668 ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp"); 6669 /* 6670 * Attempt to deallocate on a DS file. A return of ENOENT implies 6671 * there is no DS file to deallocate on. 6672 */ 6673 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL, 6674 NULL, NULL, NULL, NULL, &len, 0, NULL); 6675 if (error != ENOENT) 6676 return (error); 6677 6678 /* 6679 * Do the actual VOP_DEALLOCATE(), looping so long as 6680 * progress is being made, to achieve completion. 6681 */ 6682 do { 6683 olen = len; 6684 error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred); 6685 if (error == 0 && len > 0 && olen > len) 6686 maybe_yield(); 6687 } while (error == 0 && len > 0 && olen > len); 6688 if (error == 0 && len > 0) 6689 error = NFSERR_IO; 6690 NFSEXITCODE(error); 6691 return (error); 6692 } 6693 6694 /* 6695 * Get Extended Atribute vnode op into an mbuf list. 6696 */ 6697 int 6698 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, 6699 struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p, 6700 struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 6701 { 6702 struct iovec *iv; 6703 struct uio io, *uiop = &io; 6704 struct mbuf *m, *m2; 6705 int alen, error, len, tlen; 6706 size_t siz; 6707 6708 /* First, find out the size of the extended attribute. */ 6709 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6710 &siz, cred, p); 6711 if (error != 0) 6712 return (NFSERR_NOXATTR); 6713 if (siz > maxresp - NFS_MAXXDR) 6714 return (NFSERR_XATTR2BIG); 6715 len = siz; 6716 tlen = NFSM_RNDUP(len); 6717 if (tlen > 0) { 6718 /* 6719 * If cnt > MCLBYTES and the reply will not be saved, use 6720 * ext_pgs mbufs for TLS. 6721 * For NFSv4.0, we do not know for sure if the reply will 6722 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 6723 * Always use ext_pgs mbufs if ND_EXTPG is set. 6724 */ 6725 if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES && 6726 (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS && 6727 (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)) 6728 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen, 6729 maxextsiz, &m, &m2, &iv); 6730 else 6731 uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, 6732 &iv); 6733 uiop->uio_iov = iv; 6734 } else { 6735 uiop->uio_iovcnt = 0; 6736 uiop->uio_iov = iv = NULL; 6737 m = m2 = NULL; 6738 } 6739 uiop->uio_offset = 0; 6740 uiop->uio_resid = tlen; 6741 uiop->uio_rw = UIO_READ; 6742 uiop->uio_segflg = UIO_SYSSPACE; 6743 uiop->uio_td = p; 6744 #ifdef MAC 6745 error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6746 name); 6747 if (error != 0) 6748 goto out; 6749 #endif 6750 6751 if (tlen > 0) 6752 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 6753 NULL, cred, p); 6754 if (error != 0) 6755 goto out; 6756 if (uiop->uio_resid > 0) { 6757 alen = tlen; 6758 len = tlen - uiop->uio_resid; 6759 tlen = NFSM_RNDUP(len); 6760 if (alen != tlen) 6761 printf("nfsvno_getxattr: weird size read\n"); 6762 if (tlen == 0) { 6763 m_freem(m); 6764 m = m2 = NULL; 6765 } else if (alen != tlen || tlen != len) 6766 m2 = nfsrv_adj(m, alen - tlen, tlen - len); 6767 } 6768 *lenp = len; 6769 *mpp = m; 6770 *mpendp = m2; 6771 6772 out: 6773 if (error != 0) { 6774 if (m != NULL) 6775 m_freem(m); 6776 *lenp = 0; 6777 } 6778 free(iv, M_TEMP); 6779 NFSEXITCODE(error); 6780 return (error); 6781 } 6782 6783 /* 6784 * Set Extended attribute vnode op from an mbuf list. 6785 */ 6786 int 6787 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, 6788 char *cp, struct ucred *cred, struct thread *p) 6789 { 6790 struct iovec *iv; 6791 struct uio uio, *uiop = &uio; 6792 int cnt, error; 6793 6794 error = 0; 6795 #ifdef MAC 6796 error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6797 name); 6798 #endif 6799 if (error != 0) 6800 goto out; 6801 6802 uiop->uio_rw = UIO_WRITE; 6803 uiop->uio_segflg = UIO_SYSSPACE; 6804 uiop->uio_td = p; 6805 uiop->uio_offset = 0; 6806 uiop->uio_resid = len; 6807 if (len > 0) { 6808 error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); 6809 uiop->uio_iov = iv; 6810 uiop->uio_iovcnt = cnt; 6811 } else { 6812 uiop->uio_iov = iv = NULL; 6813 uiop->uio_iovcnt = 0; 6814 } 6815 if (error == 0) { 6816 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 6817 cred, p); 6818 if (error == 0) { 6819 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 6820 nfsvno_updateds(vp, cred, p); 6821 error = VOP_FSYNC(vp, MNT_WAIT, p); 6822 } 6823 free(iv, M_TEMP); 6824 } 6825 6826 out: 6827 NFSEXITCODE(error); 6828 return (error); 6829 } 6830 6831 /* 6832 * For a pNFS server, the DS file's ctime and 6833 * va_filerev (TimeMetadata and Change) needs to 6834 * be updated. This is a hack, but works by 6835 * flipping the S_ISGID bit in va_mode and then 6836 * flipping it back. 6837 * It does result in two MDS->DS RPCs, but creating 6838 * a custom RPC just to do this seems overkill, since 6839 * Setxattr/Rmxattr will not be done that frequently. 6840 * If it fails part way through, that is not too 6841 * serious, since the DS file is never executed. 6842 */ 6843 static void 6844 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p) 6845 { 6846 struct nfsvattr nva; 6847 int ret; 6848 u_short tmode; 6849 6850 ret = VOP_GETATTR(vp, &nva.na_vattr, cred); 6851 if (ret == 0) { 6852 tmode = nva.na_mode; 6853 NFSVNO_ATTRINIT(&nva); 6854 tmode ^= S_ISGID; 6855 NFSVNO_SETATTRVAL(&nva, mode, tmode); 6856 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 6857 NFSPROC_SETATTR, NULL, NULL, NULL, &nva, 6858 NULL, NULL, 0, NULL); 6859 if (ret == 0) { 6860 tmode ^= S_ISGID; 6861 NFSVNO_SETATTRVAL(&nva, mode, tmode); 6862 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 6863 NFSPROC_SETATTR, NULL, NULL, NULL, 6864 &nva, NULL, NULL, 0, NULL); 6865 } 6866 } 6867 } 6868 6869 /* 6870 * Remove Extended attribute vnode op. 6871 */ 6872 int 6873 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, 6874 struct ucred *cred, struct thread *p) 6875 { 6876 int error; 6877 6878 /* 6879 * Get rid of any delegations. I am not sure why this is required, 6880 * but RFC-8276 says so. 6881 */ 6882 error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); 6883 if (error != 0) 6884 goto out; 6885 #ifdef MAC 6886 error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6887 name); 6888 if (error != 0) 6889 goto out; 6890 #endif 6891 6892 error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); 6893 if (error == EOPNOTSUPP) 6894 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6895 cred, p); 6896 if (error == 0) { 6897 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 6898 nfsvno_updateds(vp, cred, p); 6899 error = VOP_FSYNC(vp, MNT_WAIT, p); 6900 } 6901 out: 6902 NFSEXITCODE(error); 6903 return (error); 6904 } 6905 6906 /* 6907 * List Extended Atribute vnode op into an mbuf list. 6908 */ 6909 int 6910 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, 6911 struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) 6912 { 6913 struct iovec iv; 6914 struct uio io; 6915 int error; 6916 size_t siz; 6917 6918 *bufp = NULL; 6919 /* First, find out the size of the extended attribute. */ 6920 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, 6921 p); 6922 if (error != 0) 6923 return (NFSERR_NOXATTR); 6924 if (siz <= cookie) { 6925 *lenp = 0; 6926 *eofp = true; 6927 goto out; 6928 } 6929 if (siz > cookie + *lenp) { 6930 siz = cookie + *lenp; 6931 *eofp = false; 6932 } else 6933 *eofp = true; 6934 /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ 6935 if (siz > 10 * 1024 * 1024) { 6936 error = NFSERR_XATTR2BIG; 6937 goto out; 6938 } 6939 *bufp = malloc(siz, M_TEMP, M_WAITOK); 6940 iv.iov_base = *bufp; 6941 iv.iov_len = siz; 6942 io.uio_iovcnt = 1; 6943 io.uio_iov = &iv; 6944 io.uio_offset = 0; 6945 io.uio_resid = siz; 6946 io.uio_rw = UIO_READ; 6947 io.uio_segflg = UIO_SYSSPACE; 6948 io.uio_td = p; 6949 #ifdef MAC 6950 error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); 6951 if (error != 0) 6952 goto out; 6953 #endif 6954 6955 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, 6956 p); 6957 if (error != 0) 6958 goto out; 6959 if (io.uio_resid > 0) 6960 siz -= io.uio_resid; 6961 *lenp = siz; 6962 6963 out: 6964 if (error != 0) { 6965 free(*bufp, M_TEMP); 6966 *bufp = NULL; 6967 } 6968 NFSEXITCODE(error); 6969 return (error); 6970 } 6971 6972 /* 6973 * Trim trailing data off the mbuf list being built. 6974 */ 6975 void 6976 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos, 6977 int bextpg, int bextpgsiz) 6978 { 6979 vm_page_t pg; 6980 int fullpgsiz, i; 6981 6982 if (mb->m_next != NULL) { 6983 m_freem(mb->m_next); 6984 mb->m_next = NULL; 6985 } 6986 if ((mb->m_flags & M_EXTPG) != 0) { 6987 KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs, 6988 ("nfsm_trimtrailing: bextpg out of range")); 6989 KASSERT(bpos == (char *)(void *) 6990 PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz, 6991 ("nfsm_trimtrailing: bextpgsiz bad!")); 6992 6993 /* First, get rid of any pages after this position. */ 6994 for (i = mb->m_epg_npgs - 1; i > bextpg; i--) { 6995 pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]); 6996 vm_page_unwire_noq(pg); 6997 vm_page_free(pg); 6998 } 6999 mb->m_epg_npgs = bextpg + 1; 7000 if (bextpg == 0) 7001 fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off; 7002 else 7003 fullpgsiz = PAGE_SIZE; 7004 mb->m_epg_last_len = fullpgsiz - bextpgsiz; 7005 mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off); 7006 for (i = 1; i < mb->m_epg_npgs; i++) 7007 mb->m_len += m_epg_pagelen(mb, i, 0); 7008 nd->nd_bextpgsiz = bextpgsiz; 7009 nd->nd_bextpg = bextpg; 7010 } else 7011 mb->m_len = bpos - mtod(mb, char *); 7012 nd->nd_mb = mb; 7013 nd->nd_bpos = bpos; 7014 } 7015 7016 7017 /* 7018 * Check to see if a put file handle operation should test for 7019 * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR. 7020 * When Open is the next operation, NFSERR_WRONGSEC cannot be 7021 * replied for the Open cases that use a component. This can 7022 * be identified by the fact that the file handle's type is VDIR. 7023 */ 7024 bool 7025 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp) 7026 { 7027 7028 if ((nd->nd_flag & ND_NFSV4) == 0) 7029 return (true); 7030 7031 if ((nd->nd_flag & ND_LASTOP) != 0) 7032 return (false); 7033 7034 if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH || 7035 nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH || 7036 nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP || 7037 nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME) 7038 return (false); 7039 if (nextop == NFSV4OP_OPEN && vtyp == VDIR) 7040 return (false); 7041 return (true); 7042 } 7043 7044 /* 7045 * Check DSs marked no space. 7046 */ 7047 void 7048 nfsrv_checknospc(void) 7049 { 7050 struct statfs *tsf; 7051 struct nfsdevice *ds; 7052 struct vnode **dvpp, **tdvpp, *dvp; 7053 char *devid, *tdevid; 7054 int cnt, error = 0, i; 7055 7056 if (nfsrv_devidcnt <= 0) 7057 return; 7058 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 7059 devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK); 7060 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 7061 7062 /* Get an array of the dvps for the DSs. */ 7063 tdvpp = dvpp; 7064 tdevid = devid; 7065 i = 0; 7066 NFSDDSLOCK(); 7067 /* First, search for matches for same file system. */ 7068 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7069 if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) { 7070 if (++i > nfsrv_devidcnt) 7071 break; 7072 *tdvpp++ = ds->nfsdev_dvp; 7073 NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID); 7074 tdevid += NFSX_V4DEVICEID; 7075 } 7076 } 7077 NFSDDSUNLOCK(); 7078 7079 /* Do a VFS_STATFS() for each of the DSs and clear no space. */ 7080 cnt = i; 7081 tdvpp = dvpp; 7082 tdevid = devid; 7083 for (i = 0; i < cnt && error == 0; i++) { 7084 dvp = *tdvpp++; 7085 error = VFS_STATFS(dvp->v_mount, tsf); 7086 if (error == 0 && tsf->f_bavail > 0) { 7087 NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n"); 7088 nfsrv_marknospc(tdevid, false); 7089 } 7090 tdevid += NFSX_V4DEVICEID; 7091 } 7092 free(tsf, M_TEMP); 7093 free(dvpp, M_TEMP); 7094 free(devid, M_TEMP); 7095 } 7096 7097 /* 7098 * Initialize everything that needs to be initialized for a vnet. 7099 */ 7100 static void 7101 nfsrv_vnetinit(const void *unused __unused) 7102 { 7103 7104 nfsd_mntinit(); 7105 } 7106 SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY, 7107 nfsrv_vnetinit, NULL); 7108 7109 /* 7110 * Clean up everything that is in a vnet and needs to be 7111 * done when the jail is destroyed or the module unloaded. 7112 */ 7113 static void 7114 nfsrv_cleanup(struct prison *pr) 7115 { 7116 int i; 7117 7118 NFSD_CURVNET_SET(pr->pr_vnet); 7119 NFSD_LOCK(); 7120 if (!NFSD_VNET(nfsrv_mntinited)) { 7121 NFSD_UNLOCK(); 7122 NFSD_CURVNET_RESTORE(); 7123 return; 7124 } 7125 NFSD_VNET(nfsrv_mntinited) = false; 7126 NFSD_UNLOCK(); 7127 7128 /* Clean out all NFSv4 state. */ 7129 nfsrv_throwawayallstate(curthread); 7130 7131 /* Clean the NFS server reply cache */ 7132 nfsrvd_cleancache(); 7133 7134 /* Clean out v4root exports. */ 7135 if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) { 7136 vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export); 7137 free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT); 7138 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 7139 } 7140 7141 /* Free up the krpc server pool. */ 7142 if (NFSD_VNET(nfsrvd_pool) != NULL) 7143 svcpool_destroy(NFSD_VNET(nfsrvd_pool)); 7144 7145 /* and get rid of the locks */ 7146 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 7147 mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx); 7148 mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx); 7149 } 7150 mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx); 7151 for (i = 0; i < nfsrv_sessionhashsize; i++) 7152 mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx); 7153 lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock); 7154 free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE); 7155 free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE); 7156 free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE); 7157 free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT); 7158 free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE); 7159 free(NFSD_VNET(nfssessionhash), M_NFSDSESSION); 7160 free(NFSD_VNET(nfsv4root_mnt), M_TEMP); 7161 NFSD_VNET(nfsv4root_mnt) = NULL; 7162 NFSD_CURVNET_RESTORE(); 7163 } 7164 7165 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); 7166 7167 /* 7168 * Called once to initialize data structures... 7169 */ 7170 static int 7171 nfsd_modevent(module_t mod, int type, void *data) 7172 { 7173 int error = 0, i; 7174 static int loaded = 0; 7175 7176 switch (type) { 7177 case MOD_LOAD: 7178 if (loaded) 7179 goto out; 7180 newnfs_portinit(); 7181 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); 7182 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); 7183 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); 7184 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); 7185 #ifdef VV_DISABLEDELEG 7186 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; 7187 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; 7188 #endif 7189 nfsd_call_nfsd = nfssvc_nfsd; 7190 loaded = 1; 7191 break; 7192 7193 case MOD_UNLOAD: 7194 if (newnfs_numnfsd != 0) { 7195 error = EBUSY; 7196 break; 7197 } 7198 7199 #ifdef VV_DISABLEDELEG 7200 vn_deleg_ops.vndeleg_recall = NULL; 7201 vn_deleg_ops.vndeleg_disable = NULL; 7202 #endif 7203 nfsd_call_nfsd = NULL; 7204 nfsrv_cleanup(&prison0); 7205 mtx_destroy(&nfsrc_udpmtx); 7206 mtx_destroy(&nfs_v4root_mutex); 7207 mtx_destroy(&nfsrv_dontlistlock_mtx); 7208 mtx_destroy(&nfsrv_recalllock_mtx); 7209 if (nfslayouthash != NULL) { 7210 for (i = 0; i < nfsrv_layouthashsize; i++) 7211 mtx_destroy(&nfslayouthash[i].mtx); 7212 free(nfslayouthash, M_NFSDSESSION); 7213 } 7214 loaded = 0; 7215 break; 7216 default: 7217 error = EOPNOTSUPP; 7218 break; 7219 } 7220 7221 out: 7222 NFSEXITCODE(error); 7223 return (error); 7224 } 7225 static moduledata_t nfsd_mod = { 7226 "nfsd", 7227 nfsd_modevent, 7228 NULL, 7229 }; 7230 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); 7231 7232 /* So that loader and kldload(2) can find us, wherever we are.. */ 7233 MODULE_VERSION(nfsd, 1); 7234 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); 7235 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); 7236 MODULE_DEPEND(nfsd, krpc, 1, 1, 1); 7237 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); 7238