1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 */ 35 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include <sys/capsicum.h> 40 #include <sys/extattr.h> 41 42 /* 43 * Functions that perform the vfs operations required by the routines in 44 * nfsd_serv.c. It is hoped that this change will make the server more 45 * portable. 46 */ 47 48 #include <fs/nfs/nfsport.h> 49 #include <security/mac/mac_framework.h> 50 #include <sys/filio.h> 51 #include <sys/hash.h> 52 #include <sys/sysctl.h> 53 #include <nlm/nlm_prot.h> 54 #include <nlm/nlm.h> 55 56 FEATURE(nfsd, "NFSv4 server"); 57 58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 59 extern int nfsrv_useacl; 60 extern int newnfs_numnfsd; 61 extern struct mount nfsv4root_mnt; 62 extern struct nfsrv_stablefirst nfsrv_stablefirst; 63 extern void (*nfsd_call_servertimer)(void); 64 extern SVCPOOL *nfsrvd_pool; 65 extern struct nfsv4lock nfsd_suspend_lock; 66 extern struct nfsclienthashhead *nfsclienthash; 67 extern struct nfslockhashhead *nfslockhash; 68 extern struct nfssessionhash *nfssessionhash; 69 extern int nfsrv_sessionhashsize; 70 extern struct nfsstatsv1 nfsstatsv1; 71 extern struct nfslayouthash *nfslayouthash; 72 extern int nfsrv_layouthashsize; 73 extern struct mtx nfsrv_dslock_mtx; 74 extern int nfs_pnfsiothreads; 75 extern struct nfsdontlisthead nfsrv_dontlisthead; 76 extern volatile int nfsrv_dontlistlen; 77 extern volatile int nfsrv_devidcnt; 78 extern int nfsrv_maxpnfsmirror; 79 extern uint32_t nfs_srvmaxio; 80 extern int nfs_bufpackets; 81 extern u_long sb_max_adj; 82 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt; 83 NFSDLOCKMUTEX; 84 NFSSTATESPINLOCK; 85 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE]; 86 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE]; 87 struct mtx nfsrc_udpmtx; 88 struct mtx nfs_v4root_mutex; 89 struct mtx nfsrv_dontlistlock_mtx; 90 struct mtx nfsrv_recalllock_mtx; 91 struct nfsrvfh nfs_rootfh, nfs_pubfh; 92 int nfs_pubfhset = 0, nfs_rootfhset = 0; 93 struct proc *nfsd_master_proc = NULL; 94 int nfsd_debuglevel = 0; 95 static pid_t nfsd_master_pid = (pid_t)-1; 96 static char nfsd_master_comm[MAXCOMLEN + 1]; 97 static struct timeval nfsd_master_start; 98 static uint32_t nfsv4_sysid = 0; 99 static fhandle_t zerofh; 100 101 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, 102 struct ucred *); 103 104 int nfsrv_enable_crossmntpt = 1; 105 static int nfs_commit_blks; 106 static int nfs_commit_miss; 107 extern int nfsrv_issuedelegs; 108 extern int nfsrv_dolocallocks; 109 extern int nfsd_enable_stringtouid; 110 extern struct nfsdevicehead nfsrv_devidhead; 111 112 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, 113 struct iovec **); 114 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **, 115 struct mbuf **, struct iovec **); 116 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, 117 int *); 118 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, 119 NFSPROC_T *); 120 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, 121 int *, char *, fhandle_t *); 122 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, 123 NFSPROC_T *); 124 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, 125 struct thread *, int, struct mbuf **, char *, struct mbuf **, 126 struct nfsvattr *, struct acl *, off_t *, int, bool *); 127 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); 128 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, 129 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); 130 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, 131 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, 132 char *, int *); 133 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 134 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 135 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 136 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 137 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 138 struct vnode *, struct nfsmount **, int, struct acl *, int *); 139 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 140 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); 141 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 142 struct vnode *, struct nfsmount *, struct nfsvattr *); 143 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, 144 NFSPROC_T *, struct nfsmount *); 145 static int nfsrv_putfhname(fhandle_t *, char *); 146 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, 147 struct pnfsdsfile *, struct vnode **, NFSPROC_T *); 148 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, 149 struct vnode *, NFSPROC_T *); 150 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); 151 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, 152 NFSPROC_T *); 153 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); 154 155 int nfs_pnfsio(task_fn_t *, void *); 156 157 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 158 "NFS server"); 159 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, 160 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); 161 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 162 0, ""); 163 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 164 0, ""); 165 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, 166 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); 167 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, 168 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); 169 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 170 0, "Debug level for NFS server"); 171 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW, 172 &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names"); 173 static int nfsrv_pnfsgetdsattr = 1; 174 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, 175 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); 176 177 /* 178 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are 179 * not running. 180 * The dsN subdirectories for the increased values must have been created 181 * on all DS servers before this increase is done. 182 */ 183 u_int nfsrv_dsdirsize = 20; 184 static int 185 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) 186 { 187 int error, newdsdirsize; 188 189 newdsdirsize = nfsrv_dsdirsize; 190 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); 191 if (error != 0 || req->newptr == NULL) 192 return (error); 193 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || 194 newnfs_numnfsd != 0) 195 return (EINVAL); 196 nfsrv_dsdirsize = newdsdirsize; 197 return (0); 198 } 199 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, 200 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), 201 sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); 202 203 /* 204 * nfs_srvmaxio can only be increased and only when the nfsd threads are 205 * not running. The setting must be a power of 2, with the current limit of 206 * 1Mbyte. 207 */ 208 static int 209 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS) 210 { 211 int error; 212 u_int newsrvmaxio; 213 uint64_t tval; 214 215 newsrvmaxio = nfs_srvmaxio; 216 error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req); 217 if (error != 0 || req->newptr == NULL) 218 return (error); 219 if (newsrvmaxio == nfs_srvmaxio) 220 return (0); 221 if (newsrvmaxio < nfs_srvmaxio) { 222 printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n"); 223 return (EINVAL); 224 } 225 if (newsrvmaxio > 1048576) { 226 printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n"); 227 return (EINVAL); 228 } 229 if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) { 230 printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n"); 231 return (EINVAL); 232 } 233 234 /* 235 * Check that kern.ipc.maxsockbuf is large enough for 236 * newsrviomax, given the setting of vfs.nfs.bufpackets. 237 */ 238 if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets > 239 sb_max_adj) { 240 /* 241 * Suggest vfs.nfs.bufpackets * maximum RPC message for 242 * sb_max_adj. 243 */ 244 tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets; 245 246 /* 247 * Convert suggested sb_max_adj value to a suggested 248 * sb_max value, which is what is set via kern.ipc.maxsockbuf. 249 * Perform the inverse calculation of (from uipc_sockbuf.c): 250 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES / 251 * (MSIZE + MCLBYTES); 252 * XXX If the calculation of sb_max_adj from sb_max changes, 253 * this calculation must be changed as well. 254 */ 255 tval *= (MSIZE + MCLBYTES); /* Brackets for readability. */ 256 tval += MCLBYTES - 1; /* Round up divide. */ 257 tval /= MCLBYTES; 258 printf("nfsd: set kern.ipc.maxsockbuf to a minimum of " 259 "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval, 260 newsrvmaxio); 261 return (EINVAL); 262 } 263 264 NFSD_LOCK(); 265 if (newnfs_numnfsd != 0) { 266 NFSD_UNLOCK(); 267 printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd " 268 "threads are running\n"); 269 return (EINVAL); 270 } 271 272 273 nfs_srvmaxio = newsrvmaxio; 274 NFSD_UNLOCK(); 275 return (0); 276 } 277 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio, 278 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 279 sysctl_srvmaxio, "IU", "Maximum I/O size in bytes"); 280 281 #define MAX_REORDERED_RPC 16 282 #define NUM_HEURISTIC 1031 283 #define NHUSE_INIT 64 284 #define NHUSE_INC 16 285 #define NHUSE_MAX 2048 286 287 static struct nfsheur { 288 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 289 off_t nh_nextoff; /* next offset for sequential detection */ 290 int nh_use; /* use count for selection */ 291 int nh_seqcount; /* heuristic */ 292 } nfsheur[NUM_HEURISTIC]; 293 294 /* 295 * Heuristic to detect sequential operation. 296 */ 297 static struct nfsheur * 298 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 299 { 300 struct nfsheur *nh; 301 int hi, try; 302 303 /* Locate best candidate. */ 304 try = 32; 305 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 306 nh = &nfsheur[hi]; 307 while (try--) { 308 if (nfsheur[hi].nh_vp == vp) { 309 nh = &nfsheur[hi]; 310 break; 311 } 312 if (nfsheur[hi].nh_use > 0) 313 --nfsheur[hi].nh_use; 314 hi = (hi + 1) % NUM_HEURISTIC; 315 if (nfsheur[hi].nh_use < nh->nh_use) 316 nh = &nfsheur[hi]; 317 } 318 319 /* Initialize hint if this is a new file. */ 320 if (nh->nh_vp != vp) { 321 nh->nh_vp = vp; 322 nh->nh_nextoff = uio->uio_offset; 323 nh->nh_use = NHUSE_INIT; 324 if (uio->uio_offset == 0) 325 nh->nh_seqcount = 4; 326 else 327 nh->nh_seqcount = 1; 328 } 329 330 /* Calculate heuristic. */ 331 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 332 uio->uio_offset == nh->nh_nextoff) { 333 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 334 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 335 if (nh->nh_seqcount > IO_SEQMAX) 336 nh->nh_seqcount = IO_SEQMAX; 337 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 338 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 339 /* Probably a reordered RPC, leave seqcount alone. */ 340 } else if (nh->nh_seqcount > 1) { 341 nh->nh_seqcount /= 2; 342 } else { 343 nh->nh_seqcount = 0; 344 } 345 nh->nh_use += NHUSE_INC; 346 if (nh->nh_use > NHUSE_MAX) 347 nh->nh_use = NHUSE_MAX; 348 return (nh); 349 } 350 351 /* 352 * Get attributes into nfsvattr structure. 353 */ 354 int 355 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, 356 struct nfsrv_descript *nd, struct thread *p, int vpislocked, 357 nfsattrbit_t *attrbitp) 358 { 359 int error, gotattr, lockedit = 0; 360 struct nfsvattr na; 361 362 if (vpislocked == 0) { 363 /* 364 * When vpislocked == 0, the vnode is either exclusively 365 * locked by this thread or not locked by this thread. 366 * As such, shared lock it, if not exclusively locked. 367 */ 368 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 369 lockedit = 1; 370 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); 371 } 372 } 373 374 /* 375 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed 376 * attributes, as required. 377 * This needs to be done for regular files if: 378 * - non-NFSv4 RPCs or 379 * - when attrbitp == NULL or 380 * - an NFSv4 RPC with any of the above attributes in attrbitp. 381 * A return of 0 for nfsrv_proxyds() indicates that it has acquired 382 * these attributes. nfsrv_proxyds() will return an error if the 383 * server is not a pNFS one. 384 */ 385 gotattr = 0; 386 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || 387 (nd->nd_flag & ND_NFSV4) == 0 || 388 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || 389 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || 390 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || 391 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || 392 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { 393 error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, 394 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, 395 NULL); 396 if (error == 0) 397 gotattr = 1; 398 } 399 400 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); 401 if (lockedit != 0) 402 NFSVOPUNLOCK(vp); 403 404 /* 405 * If we got the Change, Size and Modify Time from the DS, 406 * replace them. 407 */ 408 if (gotattr != 0) { 409 nvap->na_atime = na.na_atime; 410 nvap->na_mtime = na.na_mtime; 411 nvap->na_filerev = na.na_filerev; 412 nvap->na_size = na.na_size; 413 nvap->na_bytes = na.na_bytes; 414 } 415 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, 416 error, (uintmax_t)na.na_filerev); 417 418 NFSEXITCODE(error); 419 return (error); 420 } 421 422 /* 423 * Get a file handle for a vnode. 424 */ 425 int 426 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) 427 { 428 int error; 429 430 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); 431 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 432 error = VOP_VPTOFH(vp, &fhp->fh_fid); 433 434 NFSEXITCODE(error); 435 return (error); 436 } 437 438 /* 439 * Perform access checking for vnodes obtained from file handles that would 440 * refer to files already opened by a Unix client. You cannot just use 441 * vn_writechk() and VOP_ACCESSX() for two reasons. 442 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 443 * case. 444 * 2 - The owner is to be given access irrespective of mode bits for some 445 * operations, so that processes that chmod after opening a file don't 446 * break. 447 */ 448 int 449 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, 450 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, 451 u_int32_t *supportedtypep) 452 { 453 struct vattr vattr; 454 int error = 0, getret = 0; 455 456 if (vpislocked == 0) { 457 if (NFSVOPLOCK(vp, LK_SHARED) != 0) { 458 error = EPERM; 459 goto out; 460 } 461 } 462 if (accmode & VWRITE) { 463 /* Just vn_writechk() changed to check rdonly */ 464 /* 465 * Disallow write attempts on read-only file systems; 466 * unless the file is a socket or a block or character 467 * device resident on the file system. 468 */ 469 if (NFSVNO_EXRDONLY(exp) || 470 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 471 switch (vp->v_type) { 472 case VREG: 473 case VDIR: 474 case VLNK: 475 error = EROFS; 476 default: 477 break; 478 } 479 } 480 /* 481 * If there's shared text associated with 482 * the inode, try to free it up once. If 483 * we fail, we can't allow writing. 484 */ 485 if (VOP_IS_TEXT(vp) && error == 0) 486 error = ETXTBSY; 487 } 488 if (error != 0) { 489 if (vpislocked == 0) 490 NFSVOPUNLOCK(vp); 491 goto out; 492 } 493 494 /* 495 * Should the override still be applied when ACLs are enabled? 496 */ 497 error = VOP_ACCESSX(vp, accmode, cred, p); 498 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { 499 /* 500 * Try again with VEXPLICIT_DENY, to see if the test for 501 * deletion is supported. 502 */ 503 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); 504 if (error == 0) { 505 if (vp->v_type == VDIR) { 506 accmode &= ~(VDELETE | VDELETE_CHILD); 507 accmode |= VWRITE; 508 error = VOP_ACCESSX(vp, accmode, cred, p); 509 } else if (supportedtypep != NULL) { 510 *supportedtypep &= ~NFSACCESS_DELETE; 511 } 512 } 513 } 514 515 /* 516 * Allow certain operations for the owner (reads and writes 517 * on files that are already open). 518 */ 519 if (override != NFSACCCHK_NOOVERRIDE && 520 (error == EPERM || error == EACCES)) { 521 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) 522 error = 0; 523 else if (override & NFSACCCHK_ALLOWOWNER) { 524 getret = VOP_GETATTR(vp, &vattr, cred); 525 if (getret == 0 && cred->cr_uid == vattr.va_uid) 526 error = 0; 527 } 528 } 529 if (vpislocked == 0) 530 NFSVOPUNLOCK(vp); 531 532 out: 533 NFSEXITCODE(error); 534 return (error); 535 } 536 537 /* 538 * Set attribute(s) vnop. 539 */ 540 int 541 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 542 struct thread *p, struct nfsexstuff *exp) 543 { 544 u_quad_t savsize = 0; 545 int error, savedit; 546 time_t savbtime; 547 548 /* 549 * If this is an exported file system and a pNFS service is running, 550 * don't VOP_SETATTR() of size for the MDS file system. 551 */ 552 savedit = 0; 553 error = 0; 554 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && 555 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && 556 nvap->na_vattr.va_size > 0) { 557 savsize = nvap->na_vattr.va_size; 558 nvap->na_vattr.va_size = VNOVAL; 559 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 560 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 561 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 562 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 563 nvap->na_vattr.va_mtime.tv_sec != VNOVAL) 564 savedit = 1; 565 else 566 savedit = 2; 567 } 568 if (savedit != 2) 569 error = VOP_SETATTR(vp, &nvap->na_vattr, cred); 570 if (savedit != 0) 571 nvap->na_vattr.va_size = savsize; 572 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 573 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 574 nvap->na_vattr.va_size != 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 /* Never modify birthtime on a DS file. */ 579 savbtime = nvap->na_vattr.va_birthtime.tv_sec; 580 nvap->na_vattr.va_birthtime.tv_sec = VNOVAL; 581 /* For a pNFS server, set the attributes on the DS file. */ 582 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, 583 NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); 584 nvap->na_vattr.va_birthtime.tv_sec = savbtime; 585 if (error == ENOENT) 586 error = 0; 587 } 588 NFSEXITCODE(error); 589 return (error); 590 } 591 592 /* 593 * Set up nameidata for a lookup() call and do it. 594 */ 595 int 596 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, 597 struct vnode *dp, int islocked, struct nfsexstuff *exp, 598 struct vnode **retdirp) 599 { 600 struct componentname *cnp = &ndp->ni_cnd; 601 int i; 602 struct iovec aiov; 603 struct uio auio; 604 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; 605 int error = 0; 606 char *cp; 607 608 *retdirp = NULL; 609 cnp->cn_nameptr = cnp->cn_pnbuf; 610 ndp->ni_lcf = 0; 611 /* 612 * Extract and set starting directory. 613 */ 614 if (dp->v_type != VDIR) { 615 if (islocked) 616 vput(dp); 617 else 618 vrele(dp); 619 nfsvno_relpathbuf(ndp); 620 error = ENOTDIR; 621 goto out1; 622 } 623 if (islocked) 624 NFSVOPUNLOCK(dp); 625 VREF(dp); 626 *retdirp = dp; 627 if (NFSVNO_EXRDONLY(exp)) 628 cnp->cn_flags |= RDONLY; 629 ndp->ni_segflg = UIO_SYSSPACE; 630 631 if (nd->nd_flag & ND_PUBLOOKUP) { 632 ndp->ni_loopcnt = 0; 633 if (cnp->cn_pnbuf[0] == '/') { 634 vrele(dp); 635 /* 636 * Check for degenerate pathnames here, since lookup() 637 * panics on them. 638 */ 639 for (i = 1; i < ndp->ni_pathlen; i++) 640 if (cnp->cn_pnbuf[i] != '/') 641 break; 642 if (i == ndp->ni_pathlen) { 643 error = NFSERR_ACCES; 644 goto out; 645 } 646 dp = rootvnode; 647 VREF(dp); 648 } 649 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || 650 (nd->nd_flag & ND_NFSV4) == 0) { 651 /* 652 * Only cross mount points for NFSv4 when doing a 653 * mount while traversing the file system above 654 * the mount point, unless nfsrv_enable_crossmntpt is set. 655 */ 656 cnp->cn_flags |= NOCROSSMOUNT; 657 } 658 659 /* 660 * Initialize for scan, set ni_startdir and bump ref on dp again 661 * because lookup() will dereference ni_startdir. 662 */ 663 664 ndp->ni_startdir = dp; 665 ndp->ni_rootdir = rootvnode; 666 ndp->ni_topdir = NULL; 667 668 if (!lockleaf) 669 cnp->cn_flags |= LOCKLEAF; 670 for (;;) { 671 cnp->cn_nameptr = cnp->cn_pnbuf; 672 /* 673 * Call lookup() to do the real work. If an error occurs, 674 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and 675 * we do not have to dereference anything before returning. 676 * In either case ni_startdir will be dereferenced and NULLed 677 * out. 678 */ 679 error = lookup(ndp); 680 if (error) 681 break; 682 683 /* 684 * Check for encountering a symbolic link. Trivial 685 * termination occurs if no symlink encountered. 686 */ 687 if ((cnp->cn_flags & ISSYMLINK) == 0) { 688 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) 689 nfsvno_relpathbuf(ndp); 690 if (ndp->ni_vp && !lockleaf) 691 NFSVOPUNLOCK(ndp->ni_vp); 692 break; 693 } 694 695 /* 696 * Validate symlink 697 */ 698 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) 699 NFSVOPUNLOCK(ndp->ni_dvp); 700 if (!(nd->nd_flag & ND_PUBLOOKUP)) { 701 error = EINVAL; 702 goto badlink2; 703 } 704 705 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 706 error = ELOOP; 707 goto badlink2; 708 } 709 if (ndp->ni_pathlen > 1) 710 cp = uma_zalloc(namei_zone, M_WAITOK); 711 else 712 cp = cnp->cn_pnbuf; 713 aiov.iov_base = cp; 714 aiov.iov_len = MAXPATHLEN; 715 auio.uio_iov = &aiov; 716 auio.uio_iovcnt = 1; 717 auio.uio_offset = 0; 718 auio.uio_rw = UIO_READ; 719 auio.uio_segflg = UIO_SYSSPACE; 720 auio.uio_td = NULL; 721 auio.uio_resid = MAXPATHLEN; 722 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 723 if (error) { 724 badlink1: 725 if (ndp->ni_pathlen > 1) 726 uma_zfree(namei_zone, cp); 727 badlink2: 728 vrele(ndp->ni_dvp); 729 vput(ndp->ni_vp); 730 break; 731 } 732 linklen = MAXPATHLEN - auio.uio_resid; 733 if (linklen == 0) { 734 error = ENOENT; 735 goto badlink1; 736 } 737 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 738 error = ENAMETOOLONG; 739 goto badlink1; 740 } 741 742 /* 743 * Adjust or replace path 744 */ 745 if (ndp->ni_pathlen > 1) { 746 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 747 uma_zfree(namei_zone, cnp->cn_pnbuf); 748 cnp->cn_pnbuf = cp; 749 } else 750 cnp->cn_pnbuf[linklen] = '\0'; 751 ndp->ni_pathlen += linklen; 752 753 /* 754 * Cleanup refs for next loop and check if root directory 755 * should replace current directory. Normally ni_dvp 756 * becomes the new base directory and is cleaned up when 757 * we loop. Explicitly null pointers after invalidation 758 * to clarify operation. 759 */ 760 vput(ndp->ni_vp); 761 ndp->ni_vp = NULL; 762 763 if (cnp->cn_pnbuf[0] == '/') { 764 vrele(ndp->ni_dvp); 765 ndp->ni_dvp = ndp->ni_rootdir; 766 VREF(ndp->ni_dvp); 767 } 768 ndp->ni_startdir = ndp->ni_dvp; 769 ndp->ni_dvp = NULL; 770 } 771 if (!lockleaf) 772 cnp->cn_flags &= ~LOCKLEAF; 773 774 out: 775 if (error) { 776 nfsvno_relpathbuf(ndp); 777 ndp->ni_vp = NULL; 778 ndp->ni_dvp = NULL; 779 ndp->ni_startdir = NULL; 780 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { 781 ndp->ni_dvp = NULL; 782 } 783 784 out1: 785 NFSEXITCODE2(error, nd); 786 return (error); 787 } 788 789 /* 790 * Set up a pathname buffer and return a pointer to it and, optionally 791 * set a hash pointer. 792 */ 793 void 794 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) 795 { 796 struct componentname *cnp = &ndp->ni_cnd; 797 798 cnp->cn_flags |= (NOMACCHECK | HASBUF); 799 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 800 if (hashpp != NULL) 801 *hashpp = NULL; 802 *bufpp = cnp->cn_pnbuf; 803 } 804 805 /* 806 * Release the above path buffer, if not released by nfsvno_namei(). 807 */ 808 void 809 nfsvno_relpathbuf(struct nameidata *ndp) 810 { 811 812 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0) 813 panic("nfsrelpath"); 814 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 815 ndp->ni_cnd.cn_flags &= ~HASBUF; 816 } 817 818 /* 819 * Readlink vnode op into an mbuf list. 820 */ 821 int 822 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz, 823 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 824 { 825 struct iovec *iv; 826 struct uio io, *uiop = &io; 827 struct mbuf *mp, *mp3; 828 int len, tlen, error = 0; 829 830 len = NFS_MAXPATHLEN; 831 if (maxextsiz > 0) 832 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 833 &mp3, &mp, &iv); 834 else 835 uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); 836 uiop->uio_iov = iv; 837 uiop->uio_offset = 0; 838 uiop->uio_resid = len; 839 uiop->uio_rw = UIO_READ; 840 uiop->uio_segflg = UIO_SYSSPACE; 841 uiop->uio_td = NULL; 842 error = VOP_READLINK(vp, uiop, cred); 843 free(iv, M_TEMP); 844 if (error) { 845 m_freem(mp3); 846 *lenp = 0; 847 goto out; 848 } 849 if (uiop->uio_resid > 0) { 850 len -= uiop->uio_resid; 851 tlen = NFSM_RNDUP(len); 852 if (tlen == 0) { 853 m_freem(mp3); 854 mp3 = mp = NULL; 855 } else if (tlen != NFS_MAXPATHLEN || tlen != len) 856 mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, 857 tlen - len); 858 } 859 *lenp = len; 860 *mpp = mp3; 861 *mpendp = mp; 862 863 out: 864 NFSEXITCODE(error); 865 return (error); 866 } 867 868 /* 869 * Create an mbuf chain and an associated iovec that can be used to Read 870 * or Getextattr of data. 871 * Upon success, return pointers to the first and last mbufs in the chain 872 * plus the malloc'd iovec and its iovlen. 873 */ 874 static int 875 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, 876 struct iovec **ivp) 877 { 878 struct mbuf *m, *m2 = NULL, *m3; 879 struct iovec *iv; 880 int i, left, siz; 881 882 left = len; 883 m3 = NULL; 884 /* 885 * Generate the mbuf list with the uio_iov ref. to it. 886 */ 887 i = 0; 888 while (left > 0) { 889 NFSMGET(m); 890 MCLGET(m, M_WAITOK); 891 m->m_len = 0; 892 siz = min(M_TRAILINGSPACE(m), left); 893 left -= siz; 894 i++; 895 if (m3) 896 m2->m_next = m; 897 else 898 m3 = m; 899 m2 = m; 900 } 901 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 902 m = m3; 903 left = len; 904 i = 0; 905 while (left > 0) { 906 if (m == NULL) 907 panic("nfsrv_createiovec iov"); 908 siz = min(M_TRAILINGSPACE(m), left); 909 if (siz > 0) { 910 iv->iov_base = mtod(m, caddr_t) + m->m_len; 911 iv->iov_len = siz; 912 m->m_len += siz; 913 left -= siz; 914 iv++; 915 i++; 916 } 917 m = m->m_next; 918 } 919 *mpp = m3; 920 *mpendp = m2; 921 return (i); 922 } 923 924 /* 925 * Create an mbuf chain and an associated iovec that can be used to Read 926 * or Getextattr of data. 927 * Upon success, return pointers to the first and last mbufs in the chain 928 * plus the malloc'd iovec and its iovlen. 929 * Same as above, but creates ext_pgs mbuf(s). 930 */ 931 static int 932 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp, 933 struct mbuf **mpendp, struct iovec **ivp) 934 { 935 struct mbuf *m, *m2 = NULL, *m3; 936 struct iovec *iv; 937 int i, left, pgno, siz; 938 939 left = len; 940 m3 = NULL; 941 /* 942 * Generate the mbuf list with the uio_iov ref. to it. 943 */ 944 i = 0; 945 while (left > 0) { 946 siz = min(left, maxextsiz); 947 m = mb_alloc_ext_plus_pages(siz, M_WAITOK); 948 left -= siz; 949 i += m->m_epg_npgs; 950 if (m3 != NULL) 951 m2->m_next = m; 952 else 953 m3 = m; 954 m2 = m; 955 } 956 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 957 m = m3; 958 left = len; 959 i = 0; 960 pgno = 0; 961 while (left > 0) { 962 if (m == NULL) 963 panic("nfsvno_createiovec_extpgs iov"); 964 siz = min(PAGE_SIZE, left); 965 if (siz > 0) { 966 iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); 967 iv->iov_len = siz; 968 m->m_len += siz; 969 if (pgno == m->m_epg_npgs - 1) 970 m->m_epg_last_len = siz; 971 left -= siz; 972 iv++; 973 i++; 974 pgno++; 975 } 976 if (pgno == m->m_epg_npgs && left > 0) { 977 m = m->m_next; 978 if (m == NULL) 979 panic("nfsvno_createiovec_extpgs iov"); 980 pgno = 0; 981 } 982 } 983 *mpp = m3; 984 *mpendp = m2; 985 return (i); 986 } 987 988 /* 989 * Read vnode op call into mbuf list. 990 */ 991 int 992 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 993 int maxextsiz, struct thread *p, struct mbuf **mpp, 994 struct mbuf **mpendp) 995 { 996 struct mbuf *m; 997 struct iovec *iv; 998 int error = 0, len, tlen, ioflag = 0; 999 struct mbuf *m3; 1000 struct uio io, *uiop = &io; 1001 struct nfsheur *nh; 1002 1003 /* 1004 * Attempt to read from a DS file. A return of ENOENT implies 1005 * there is no DS file to read. 1006 */ 1007 error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, 1008 NULL, mpendp, NULL, NULL, NULL, 0, NULL); 1009 if (error != ENOENT) 1010 return (error); 1011 1012 len = NFSM_RNDUP(cnt); 1013 if (maxextsiz > 0) 1014 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 1015 &m3, &m, &iv); 1016 else 1017 uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); 1018 uiop->uio_iov = iv; 1019 uiop->uio_offset = off; 1020 uiop->uio_resid = len; 1021 uiop->uio_rw = UIO_READ; 1022 uiop->uio_segflg = UIO_SYSSPACE; 1023 uiop->uio_td = NULL; 1024 nh = nfsrv_sequential_heuristic(uiop, vp); 1025 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 1026 /* XXX KDM make this more systematic? */ 1027 nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid; 1028 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 1029 free(iv, M_TEMP); 1030 if (error) { 1031 m_freem(m3); 1032 *mpp = NULL; 1033 goto out; 1034 } 1035 nh->nh_nextoff = uiop->uio_offset; 1036 tlen = len - uiop->uio_resid; 1037 cnt = cnt < tlen ? cnt : tlen; 1038 tlen = NFSM_RNDUP(cnt); 1039 if (tlen == 0) { 1040 m_freem(m3); 1041 m3 = m = NULL; 1042 } else if (len != tlen || tlen != cnt) 1043 m = nfsrv_adj(m3, len - tlen, tlen - cnt); 1044 *mpp = m3; 1045 *mpendp = m; 1046 1047 out: 1048 NFSEXITCODE(error); 1049 return (error); 1050 } 1051 1052 /* 1053 * Create the iovec for the mbuf chain passed in as an argument. 1054 * The "cp" argument is where the data starts within the first mbuf in 1055 * the chain. It returns the iovec and the iovcnt. 1056 */ 1057 static int 1058 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, 1059 int *iovcntp) 1060 { 1061 struct mbuf *mp; 1062 struct iovec *ivp; 1063 int cnt, i, len; 1064 1065 /* 1066 * Loop through the mbuf chain, counting how many mbufs are a 1067 * part of this write operation, so the iovec size is known. 1068 */ 1069 cnt = 0; 1070 len = retlen; 1071 mp = m; 1072 i = mtod(mp, caddr_t) + mp->m_len - cp; 1073 while (len > 0) { 1074 if (i > 0) { 1075 len -= i; 1076 cnt++; 1077 } 1078 mp = mp->m_next; 1079 if (!mp) { 1080 if (len > 0) 1081 return (EBADRPC); 1082 } else 1083 i = mp->m_len; 1084 } 1085 1086 /* Now, create the iovec. */ 1087 mp = m; 1088 *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, 1089 M_WAITOK); 1090 *iovcntp = cnt; 1091 i = mtod(mp, caddr_t) + mp->m_len - cp; 1092 len = retlen; 1093 while (len > 0) { 1094 if (mp == NULL) 1095 panic("nfsrv_createiovecw"); 1096 if (i > 0) { 1097 i = min(i, len); 1098 ivp->iov_base = cp; 1099 ivp->iov_len = i; 1100 ivp++; 1101 len -= i; 1102 } 1103 mp = mp->m_next; 1104 if (mp) { 1105 i = mp->m_len; 1106 cp = mtod(mp, caddr_t); 1107 } 1108 } 1109 return (0); 1110 } 1111 1112 /* 1113 * Write vnode op from an mbuf list. 1114 */ 1115 int 1116 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, 1117 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) 1118 { 1119 struct iovec *iv; 1120 int cnt, ioflags, error; 1121 struct uio io, *uiop = &io; 1122 struct nfsheur *nh; 1123 1124 /* 1125 * Attempt to write to a DS file. A return of ENOENT implies 1126 * there is no DS file to write. 1127 */ 1128 error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, 1129 &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); 1130 if (error != ENOENT) { 1131 *stable = NFSWRITE_FILESYNC; 1132 return (error); 1133 } 1134 1135 if (*stable == NFSWRITE_UNSTABLE) 1136 ioflags = IO_NODELOCKED; 1137 else 1138 ioflags = (IO_SYNC | IO_NODELOCKED); 1139 error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); 1140 if (error != 0) 1141 return (error); 1142 uiop->uio_iov = iv; 1143 uiop->uio_iovcnt = cnt; 1144 uiop->uio_resid = retlen; 1145 uiop->uio_rw = UIO_WRITE; 1146 uiop->uio_segflg = UIO_SYSSPACE; 1147 NFSUIOPROC(uiop, p); 1148 uiop->uio_offset = off; 1149 nh = nfsrv_sequential_heuristic(uiop, vp); 1150 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 1151 /* XXX KDM make this more systematic? */ 1152 nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; 1153 error = VOP_WRITE(vp, uiop, ioflags, cred); 1154 if (error == 0) 1155 nh->nh_nextoff = uiop->uio_offset; 1156 free(iv, M_TEMP); 1157 1158 NFSEXITCODE(error); 1159 return (error); 1160 } 1161 1162 /* 1163 * Common code for creating a regular file (plus special files for V2). 1164 */ 1165 int 1166 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, 1167 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, 1168 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) 1169 { 1170 u_quad_t tempsize; 1171 int error; 1172 struct thread *p = curthread; 1173 1174 error = nd->nd_repstat; 1175 if (!error && ndp->ni_vp == NULL) { 1176 if (nvap->na_type == VREG || nvap->na_type == VSOCK) { 1177 vrele(ndp->ni_startdir); 1178 error = VOP_CREATE(ndp->ni_dvp, 1179 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1180 /* For a pNFS server, create the data file on a DS. */ 1181 if (error == 0 && nvap->na_type == VREG) { 1182 /* 1183 * Create a data file on a DS for a pNFS server. 1184 * This function just returns if not 1185 * running a pNFS DS or the creation fails. 1186 */ 1187 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1188 nd->nd_cred, p); 1189 } 1190 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1191 NULL, false); 1192 nfsvno_relpathbuf(ndp); 1193 if (!error) { 1194 if (*exclusive_flagp) { 1195 *exclusive_flagp = 0; 1196 NFSVNO_ATTRINIT(nvap); 1197 nvap->na_atime.tv_sec = cverf[0]; 1198 nvap->na_atime.tv_nsec = cverf[1]; 1199 error = VOP_SETATTR(ndp->ni_vp, 1200 &nvap->na_vattr, nd->nd_cred); 1201 if (error != 0) { 1202 vput(ndp->ni_vp); 1203 ndp->ni_vp = NULL; 1204 error = NFSERR_NOTSUPP; 1205 } 1206 } 1207 } 1208 /* 1209 * NFS V2 Only. nfsrvd_mknod() does this for V3. 1210 * (This implies, just get out on an error.) 1211 */ 1212 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || 1213 nvap->na_type == VFIFO) { 1214 if (nvap->na_type == VCHR && rdev == 0xffffffff) 1215 nvap->na_type = VFIFO; 1216 if (nvap->na_type != VFIFO && 1217 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { 1218 vrele(ndp->ni_startdir); 1219 nfsvno_relpathbuf(ndp); 1220 vput(ndp->ni_dvp); 1221 goto out; 1222 } 1223 nvap->na_rdev = rdev; 1224 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1225 &ndp->ni_cnd, &nvap->na_vattr); 1226 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1227 NULL, false); 1228 nfsvno_relpathbuf(ndp); 1229 vrele(ndp->ni_startdir); 1230 if (error) 1231 goto out; 1232 } else { 1233 vrele(ndp->ni_startdir); 1234 nfsvno_relpathbuf(ndp); 1235 vput(ndp->ni_dvp); 1236 error = ENXIO; 1237 goto out; 1238 } 1239 *vpp = ndp->ni_vp; 1240 } else { 1241 /* 1242 * Handle cases where error is already set and/or 1243 * the file exists. 1244 * 1 - clean up the lookup 1245 * 2 - iff !error and na_size set, truncate it 1246 */ 1247 vrele(ndp->ni_startdir); 1248 nfsvno_relpathbuf(ndp); 1249 *vpp = ndp->ni_vp; 1250 if (ndp->ni_dvp == *vpp) 1251 vrele(ndp->ni_dvp); 1252 else 1253 vput(ndp->ni_dvp); 1254 if (!error && nvap->na_size != VNOVAL) { 1255 error = nfsvno_accchk(*vpp, VWRITE, 1256 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1257 NFSACCCHK_VPISLOCKED, NULL); 1258 if (!error) { 1259 tempsize = nvap->na_size; 1260 NFSVNO_ATTRINIT(nvap); 1261 nvap->na_size = tempsize; 1262 error = VOP_SETATTR(*vpp, 1263 &nvap->na_vattr, nd->nd_cred); 1264 } 1265 } 1266 if (error) 1267 vput(*vpp); 1268 } 1269 1270 out: 1271 NFSEXITCODE(error); 1272 return (error); 1273 } 1274 1275 /* 1276 * Do a mknod vnode op. 1277 */ 1278 int 1279 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, 1280 struct thread *p) 1281 { 1282 int error = 0; 1283 enum vtype vtyp; 1284 1285 vtyp = nvap->na_type; 1286 /* 1287 * Iff doesn't exist, create it. 1288 */ 1289 if (ndp->ni_vp) { 1290 vrele(ndp->ni_startdir); 1291 nfsvno_relpathbuf(ndp); 1292 vput(ndp->ni_dvp); 1293 vrele(ndp->ni_vp); 1294 error = EEXIST; 1295 goto out; 1296 } 1297 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1298 vrele(ndp->ni_startdir); 1299 nfsvno_relpathbuf(ndp); 1300 vput(ndp->ni_dvp); 1301 error = NFSERR_BADTYPE; 1302 goto out; 1303 } 1304 if (vtyp == VSOCK) { 1305 vrele(ndp->ni_startdir); 1306 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 1307 &ndp->ni_cnd, &nvap->na_vattr); 1308 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1309 false); 1310 nfsvno_relpathbuf(ndp); 1311 } else { 1312 if (nvap->na_type != VFIFO && 1313 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { 1314 vrele(ndp->ni_startdir); 1315 nfsvno_relpathbuf(ndp); 1316 vput(ndp->ni_dvp); 1317 goto out; 1318 } 1319 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1320 &ndp->ni_cnd, &nvap->na_vattr); 1321 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1322 false); 1323 nfsvno_relpathbuf(ndp); 1324 vrele(ndp->ni_startdir); 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 vrele(ndp->ni_startdir); 1377 nfsvno_relpathbuf(ndp); 1378 if (ndp->ni_dvp == ndp->ni_vp) 1379 vrele(ndp->ni_dvp); 1380 else 1381 vput(ndp->ni_dvp); 1382 vrele(ndp->ni_vp); 1383 error = EEXIST; 1384 goto out; 1385 } 1386 1387 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1388 &nvap->na_vattr, pathcp); 1389 /* 1390 * Although FreeBSD still had the lookup code in 1391 * it for 7/current, there doesn't seem to be any 1392 * point, since VOP_SYMLINK() returns the ni_vp. 1393 * Just vput it for v2. 1394 */ 1395 VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0); 1396 vrele(ndp->ni_startdir); 1397 nfsvno_relpathbuf(ndp); 1398 1399 out: 1400 NFSEXITCODE(error); 1401 return (error); 1402 } 1403 1404 /* 1405 * Parse symbolic link arguments. 1406 * This function has an ugly side effect. It will malloc() an area for 1407 * the symlink and set iov_base to point to it, only if it succeeds. 1408 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must 1409 * be FREE'd later. 1410 */ 1411 int 1412 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, 1413 struct thread *p, char **pathcpp, int *lenp) 1414 { 1415 u_int32_t *tl; 1416 char *pathcp = NULL; 1417 int error = 0, len; 1418 struct nfsv2_sattr *sp; 1419 1420 *pathcpp = NULL; 1421 *lenp = 0; 1422 if ((nd->nd_flag & ND_NFSV3) && 1423 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) 1424 goto nfsmout; 1425 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 1426 len = fxdr_unsigned(int, *tl); 1427 if (len > NFS_MAXPATHLEN || len <= 0) { 1428 error = EBADRPC; 1429 goto nfsmout; 1430 } 1431 pathcp = malloc(len + 1, M_TEMP, M_WAITOK); 1432 error = nfsrv_mtostr(nd, pathcp, len); 1433 if (error) 1434 goto nfsmout; 1435 if (nd->nd_flag & ND_NFSV2) { 1436 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1437 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 1438 } 1439 *pathcpp = pathcp; 1440 *lenp = len; 1441 NFSEXITCODE2(0, nd); 1442 return (0); 1443 nfsmout: 1444 if (pathcp) 1445 free(pathcp, M_TEMP); 1446 NFSEXITCODE2(error, nd); 1447 return (error); 1448 } 1449 1450 /* 1451 * Remove a non-directory object. 1452 */ 1453 int 1454 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1455 struct thread *p, struct nfsexstuff *exp) 1456 { 1457 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS]; 1458 int error = 0, mirrorcnt; 1459 char fname[PNFS_FILENAME_LEN + 1]; 1460 fhandle_t fh; 1461 1462 vp = ndp->ni_vp; 1463 dsdvp[0] = NULL; 1464 if (vp->v_type == VDIR) 1465 error = NFSERR_ISDIR; 1466 else if (is_v4) 1467 error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0), 1468 p); 1469 if (error == 0) 1470 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); 1471 if (!error) 1472 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); 1473 if (error == 0 && dsdvp[0] != NULL) 1474 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1475 if (ndp->ni_dvp == vp) 1476 vrele(ndp->ni_dvp); 1477 else 1478 vput(ndp->ni_dvp); 1479 vput(vp); 1480 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) 1481 nfsvno_relpathbuf(ndp); 1482 NFSEXITCODE(error); 1483 return (error); 1484 } 1485 1486 /* 1487 * Remove a directory. 1488 */ 1489 int 1490 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1491 struct thread *p, struct nfsexstuff *exp) 1492 { 1493 struct vnode *vp; 1494 int error = 0; 1495 1496 vp = ndp->ni_vp; 1497 if (vp->v_type != VDIR) { 1498 error = ENOTDIR; 1499 goto out; 1500 } 1501 /* 1502 * No rmdir "." please. 1503 */ 1504 if (ndp->ni_dvp == vp) { 1505 error = EINVAL; 1506 goto out; 1507 } 1508 /* 1509 * The root of a mounted filesystem cannot be deleted. 1510 */ 1511 if (vp->v_vflag & VV_ROOT) 1512 error = EBUSY; 1513 out: 1514 if (!error) 1515 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); 1516 if (ndp->ni_dvp == vp) 1517 vrele(ndp->ni_dvp); 1518 else 1519 vput(ndp->ni_dvp); 1520 vput(vp); 1521 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0) 1522 nfsvno_relpathbuf(ndp); 1523 NFSEXITCODE(error); 1524 return (error); 1525 } 1526 1527 /* 1528 * Rename vnode op. 1529 */ 1530 int 1531 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, 1532 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) 1533 { 1534 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS]; 1535 int error = 0, mirrorcnt; 1536 char fname[PNFS_FILENAME_LEN + 1]; 1537 fhandle_t fh; 1538 1539 dsdvp[0] = NULL; 1540 fvp = fromndp->ni_vp; 1541 if (ndstat) { 1542 vrele(fromndp->ni_dvp); 1543 vrele(fvp); 1544 error = ndstat; 1545 goto out1; 1546 } 1547 tdvp = tondp->ni_dvp; 1548 tvp = tondp->ni_vp; 1549 if (tvp != NULL) { 1550 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1551 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; 1552 goto out; 1553 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1554 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; 1555 goto out; 1556 } 1557 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1558 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1559 goto out; 1560 } 1561 1562 /* 1563 * A rename to '.' or '..' results in a prematurely 1564 * unlocked vnode on FreeBSD5, so I'm just going to fail that 1565 * here. 1566 */ 1567 if ((tondp->ni_cnd.cn_namelen == 1 && 1568 tondp->ni_cnd.cn_nameptr[0] == '.') || 1569 (tondp->ni_cnd.cn_namelen == 2 && 1570 tondp->ni_cnd.cn_nameptr[0] == '.' && 1571 tondp->ni_cnd.cn_nameptr[1] == '.')) { 1572 error = EINVAL; 1573 goto out; 1574 } 1575 } 1576 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1577 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1578 goto out; 1579 } 1580 if (fvp->v_mount != tdvp->v_mount) { 1581 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1582 goto out; 1583 } 1584 if (fvp == tdvp) { 1585 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; 1586 goto out; 1587 } 1588 if (fvp == tvp) { 1589 /* 1590 * If source and destination are the same, there is nothing to 1591 * do. Set error to -1 to indicate this. 1592 */ 1593 error = -1; 1594 goto out; 1595 } 1596 if (ndflag & ND_NFSV4) { 1597 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { 1598 error = nfsrv_checkremove(fvp, 0, NULL, 1599 (nfsquad_t)((u_quad_t)0), p); 1600 NFSVOPUNLOCK(fvp); 1601 } else 1602 error = EPERM; 1603 if (tvp && !error) 1604 error = nfsrv_checkremove(tvp, 1, NULL, 1605 (nfsquad_t)((u_quad_t)0), p); 1606 } else { 1607 /* 1608 * For NFSv2 and NFSv3, try to get rid of the delegation, so 1609 * that the NFSv4 client won't be confused by the rename. 1610 * Since nfsd_recalldelegation() can only be called on an 1611 * unlocked vnode at this point and fvp is the file that will 1612 * still exist after the rename, just do fvp. 1613 */ 1614 nfsd_recalldelegation(fvp, p); 1615 } 1616 if (error == 0 && tvp != NULL) { 1617 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh); 1618 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" 1619 " dsdvp=%p\n", dsdvp[0]); 1620 } 1621 out: 1622 if (!error) { 1623 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, 1624 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, 1625 &tondp->ni_cnd); 1626 } else { 1627 if (tdvp == tvp) 1628 vrele(tdvp); 1629 else 1630 vput(tdvp); 1631 if (tvp) 1632 vput(tvp); 1633 vrele(fromndp->ni_dvp); 1634 vrele(fvp); 1635 if (error == -1) 1636 error = 0; 1637 } 1638 1639 /* 1640 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and 1641 * if the rename succeeded, the DS file for the tvp needs to be 1642 * removed. 1643 */ 1644 if (error == 0 && dsdvp[0] != NULL) { 1645 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1646 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); 1647 } 1648 1649 vrele(tondp->ni_startdir); 1650 nfsvno_relpathbuf(tondp); 1651 out1: 1652 vrele(fromndp->ni_startdir); 1653 nfsvno_relpathbuf(fromndp); 1654 NFSEXITCODE(error); 1655 return (error); 1656 } 1657 1658 /* 1659 * Link vnode op. 1660 */ 1661 int 1662 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, 1663 struct thread *p, struct nfsexstuff *exp) 1664 { 1665 struct vnode *xp; 1666 int error = 0; 1667 1668 xp = ndp->ni_vp; 1669 if (xp != NULL) { 1670 error = EEXIST; 1671 } else { 1672 xp = ndp->ni_dvp; 1673 if (vp->v_mount != xp->v_mount) 1674 error = EXDEV; 1675 } 1676 if (!error) { 1677 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1678 if (!VN_IS_DOOMED(vp)) 1679 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); 1680 else 1681 error = EPERM; 1682 if (ndp->ni_dvp == vp) { 1683 vrele(ndp->ni_dvp); 1684 NFSVOPUNLOCK(vp); 1685 } else { 1686 vref(vp); 1687 VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true); 1688 } 1689 } else { 1690 if (ndp->ni_dvp == ndp->ni_vp) 1691 vrele(ndp->ni_dvp); 1692 else 1693 vput(ndp->ni_dvp); 1694 if (ndp->ni_vp) 1695 vrele(ndp->ni_vp); 1696 } 1697 nfsvno_relpathbuf(ndp); 1698 NFSEXITCODE(error); 1699 return (error); 1700 } 1701 1702 /* 1703 * Do the fsync() appropriate for the commit. 1704 */ 1705 int 1706 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, 1707 struct thread *td) 1708 { 1709 int error = 0; 1710 1711 /* 1712 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of 1713 * file is done. At this time VOP_FSYNC does not accept offset and 1714 * byte count parameters so call VOP_FSYNC the whole file for now. 1715 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. 1716 * File systems that do not use the buffer cache (as indicated 1717 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). 1718 */ 1719 if (cnt == 0 || cnt > MAX_COMMIT_COUNT || 1720 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { 1721 /* 1722 * Give up and do the whole thing 1723 */ 1724 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1725 VM_OBJECT_WLOCK(vp->v_object); 1726 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); 1727 VM_OBJECT_WUNLOCK(vp->v_object); 1728 } 1729 error = VOP_FSYNC(vp, MNT_WAIT, td); 1730 } else { 1731 /* 1732 * Locate and synchronously write any buffers that fall 1733 * into the requested range. Note: we are assuming that 1734 * f_iosize is a power of 2. 1735 */ 1736 int iosize = vp->v_mount->mnt_stat.f_iosize; 1737 int iomask = iosize - 1; 1738 struct bufobj *bo; 1739 daddr_t lblkno; 1740 1741 /* 1742 * Align to iosize boundary, super-align to page boundary. 1743 */ 1744 if (off & iomask) { 1745 cnt += off & iomask; 1746 off &= ~(u_quad_t)iomask; 1747 } 1748 if (off & PAGE_MASK) { 1749 cnt += off & PAGE_MASK; 1750 off &= ~(u_quad_t)PAGE_MASK; 1751 } 1752 lblkno = off / iosize; 1753 1754 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1755 VM_OBJECT_WLOCK(vp->v_object); 1756 vm_object_page_clean(vp->v_object, off, off + cnt, 1757 OBJPC_SYNC); 1758 VM_OBJECT_WUNLOCK(vp->v_object); 1759 } 1760 1761 bo = &vp->v_bufobj; 1762 BO_LOCK(bo); 1763 while (cnt > 0) { 1764 struct buf *bp; 1765 1766 /* 1767 * If we have a buffer and it is marked B_DELWRI we 1768 * have to lock and write it. Otherwise the prior 1769 * write is assumed to have already been committed. 1770 * 1771 * gbincore() can return invalid buffers now so we 1772 * have to check that bit as well (though B_DELWRI 1773 * should not be set if B_INVAL is set there could be 1774 * a race here since we haven't locked the buffer). 1775 */ 1776 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 1777 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 1778 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { 1779 BO_LOCK(bo); 1780 continue; /* retry */ 1781 } 1782 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 1783 B_DELWRI) { 1784 bremfree(bp); 1785 bp->b_flags &= ~B_ASYNC; 1786 bwrite(bp); 1787 ++nfs_commit_miss; 1788 } else 1789 BUF_UNLOCK(bp); 1790 BO_LOCK(bo); 1791 } 1792 ++nfs_commit_blks; 1793 if (cnt < iosize) 1794 break; 1795 cnt -= iosize; 1796 ++lblkno; 1797 } 1798 BO_UNLOCK(bo); 1799 } 1800 NFSEXITCODE(error); 1801 return (error); 1802 } 1803 1804 /* 1805 * Statfs vnode op. 1806 */ 1807 int 1808 nfsvno_statfs(struct vnode *vp, struct statfs *sf) 1809 { 1810 struct statfs *tsf; 1811 int error; 1812 1813 tsf = NULL; 1814 if (nfsrv_devidcnt > 0) { 1815 /* For a pNFS service, get the DS numbers. */ 1816 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); 1817 error = nfsrv_pnfsstatfs(tsf, vp->v_mount); 1818 if (error != 0) { 1819 free(tsf, M_TEMP); 1820 tsf = NULL; 1821 } 1822 } 1823 error = VFS_STATFS(vp->v_mount, sf); 1824 if (error == 0) { 1825 if (tsf != NULL) { 1826 sf->f_blocks = tsf->f_blocks; 1827 sf->f_bavail = tsf->f_bavail; 1828 sf->f_bfree = tsf->f_bfree; 1829 sf->f_bsize = tsf->f_bsize; 1830 } 1831 /* 1832 * Since NFS handles these values as unsigned on the 1833 * wire, there is no way to represent negative values, 1834 * so set them to 0. Without this, they will appear 1835 * to be very large positive values for clients like 1836 * Solaris10. 1837 */ 1838 if (sf->f_bavail < 0) 1839 sf->f_bavail = 0; 1840 if (sf->f_ffree < 0) 1841 sf->f_ffree = 0; 1842 } 1843 free(tsf, M_TEMP); 1844 NFSEXITCODE(error); 1845 return (error); 1846 } 1847 1848 /* 1849 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but 1850 * must handle nfsrv_opencheck() calls after any other access checks. 1851 */ 1852 void 1853 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, 1854 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, 1855 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, 1856 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, 1857 struct nfsexstuff *exp, struct vnode **vpp) 1858 { 1859 struct vnode *vp = NULL; 1860 u_quad_t tempsize; 1861 struct nfsexstuff nes; 1862 struct thread *p = curthread; 1863 1864 if (ndp->ni_vp == NULL) 1865 nd->nd_repstat = nfsrv_opencheck(clientid, 1866 stateidp, stp, NULL, nd, p, nd->nd_repstat); 1867 if (!nd->nd_repstat) { 1868 if (ndp->ni_vp == NULL) { 1869 vrele(ndp->ni_startdir); 1870 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, 1871 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1872 /* For a pNFS server, create the data file on a DS. */ 1873 if (nd->nd_repstat == 0) { 1874 /* 1875 * Create a data file on a DS for a pNFS server. 1876 * This function just returns if not 1877 * running a pNFS DS or the creation fails. 1878 */ 1879 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1880 cred, p); 1881 } 1882 VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ? 1883 &ndp->ni_vp : NULL, false); 1884 nfsvno_relpathbuf(ndp); 1885 if (!nd->nd_repstat) { 1886 if (*exclusive_flagp) { 1887 *exclusive_flagp = 0; 1888 NFSVNO_ATTRINIT(nvap); 1889 nvap->na_atime.tv_sec = cverf[0]; 1890 nvap->na_atime.tv_nsec = cverf[1]; 1891 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, 1892 &nvap->na_vattr, cred); 1893 if (nd->nd_repstat != 0) { 1894 vput(ndp->ni_vp); 1895 ndp->ni_vp = NULL; 1896 nd->nd_repstat = NFSERR_NOTSUPP; 1897 } else 1898 NFSSETBIT_ATTRBIT(attrbitp, 1899 NFSATTRBIT_TIMEACCESS); 1900 } else { 1901 nfsrv_fixattr(nd, ndp->ni_vp, nvap, 1902 aclp, p, attrbitp, exp); 1903 } 1904 } 1905 vp = ndp->ni_vp; 1906 } else { 1907 if (ndp->ni_startdir) 1908 vrele(ndp->ni_startdir); 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 = VOP_SETATTR(vp, 1933 &nvap->na_vattr, cred); 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 { 1941 if (ndp->ni_cnd.cn_flags & HASBUF) 1942 nfsvno_relpathbuf(ndp); 1943 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) { 1944 vrele(ndp->ni_startdir); 1945 if (ndp->ni_dvp == ndp->ni_vp) 1946 vrele(ndp->ni_dvp); 1947 else 1948 vput(ndp->ni_dvp); 1949 if (ndp->ni_vp) 1950 vput(ndp->ni_vp); 1951 } 1952 } 1953 *vpp = vp; 1954 1955 NFSEXITCODE2(0, nd); 1956 } 1957 1958 /* 1959 * Updates the file rev and sets the mtime and ctime 1960 * to the current clock time, returning the va_filerev and va_Xtime 1961 * values. 1962 * Return ESTALE to indicate the vnode is VIRF_DOOMED. 1963 */ 1964 int 1965 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, 1966 struct nfsrv_descript *nd, struct thread *p) 1967 { 1968 struct vattr va; 1969 1970 VATTR_NULL(&va); 1971 vfs_timestamp(&va.va_mtime); 1972 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 1973 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 1974 if (VN_IS_DOOMED(vp)) 1975 return (ESTALE); 1976 } 1977 (void) VOP_SETATTR(vp, &va, nd->nd_cred); 1978 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); 1979 return (0); 1980 } 1981 1982 /* 1983 * Glue routine to nfsv4_fillattr(). 1984 */ 1985 int 1986 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, 1987 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, 1988 struct ucred *cred, struct thread *p, int isdgram, int reterr, 1989 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) 1990 { 1991 struct statfs *sf; 1992 int error; 1993 1994 sf = NULL; 1995 if (nfsrv_devidcnt > 0 && 1996 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || 1997 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || 1998 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { 1999 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); 2000 error = nfsrv_pnfsstatfs(sf, mp); 2001 if (error != 0) { 2002 free(sf, M_TEMP); 2003 sf = NULL; 2004 } 2005 } 2006 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, 2007 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, 2008 mounted_on_fileno, sf); 2009 free(sf, M_TEMP); 2010 NFSEXITCODE2(0, nd); 2011 return (error); 2012 } 2013 2014 /* Since the Readdir vnode ops vary, put the entire functions in here. */ 2015 /* 2016 * nfs readdir service 2017 * - mallocs what it thinks is enough to read 2018 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR 2019 * - calls VOP_READDIR() 2020 * - loops around building the reply 2021 * if the output generated exceeds count break out of loop 2022 * The NFSM_CLGET macro is used here so that the reply will be packed 2023 * tightly in mbuf clusters. 2024 * - it trims out records with d_fileno == 0 2025 * this doesn't matter for Unix clients, but they might confuse clients 2026 * for other os'. 2027 * - it trims out records with d_type == DT_WHT 2028 * these cannot be seen through NFS (unless we extend the protocol) 2029 * The alternate call nfsrvd_readdirplus() does lookups as well. 2030 * PS: The NFS protocol spec. does not clarify what the "count" byte 2031 * argument is a count of.. just name strings and file id's or the 2032 * entire reply rpc or ... 2033 * I tried just file name and id sizes and it confused the Sun client, 2034 * so I am using the full rpc size now. The "paranoia.." comment refers 2035 * to including the status longwords that are not a part of the dir. 2036 * "entry" structures, but are in the rpc. 2037 */ 2038 int 2039 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, 2040 struct vnode *vp, struct nfsexstuff *exp) 2041 { 2042 struct dirent *dp; 2043 u_int32_t *tl; 2044 int dirlen; 2045 char *cpos, *cend, *rbuf; 2046 struct nfsvattr at; 2047 int nlen, error = 0, getret = 1; 2048 int siz, cnt, fullsiz, eofflag, ncookies; 2049 u_int64_t off, toff, verf __unused; 2050 u_long *cookies = NULL, *cookiep; 2051 struct uio io; 2052 struct iovec iv; 2053 int is_ufs; 2054 struct thread *p = curthread; 2055 2056 if (nd->nd_repstat) { 2057 nfsrv_postopattr(nd, getret, &at); 2058 goto out; 2059 } 2060 if (nd->nd_flag & ND_NFSV2) { 2061 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2062 off = fxdr_unsigned(u_quad_t, *tl++); 2063 } else { 2064 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2065 off = fxdr_hyper(tl); 2066 tl += 2; 2067 verf = fxdr_hyper(tl); 2068 tl += 2; 2069 } 2070 toff = off; 2071 cnt = fxdr_unsigned(int, *tl); 2072 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2073 cnt = NFS_SRVMAXDATA(nd); 2074 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2075 fullsiz = siz; 2076 if (nd->nd_flag & ND_NFSV3) { 2077 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, 2078 NULL); 2079 #if 0 2080 /* 2081 * va_filerev is not sufficient as a cookie verifier, 2082 * since it is not supposed to change when entries are 2083 * removed/added unless that offset cookies returned to 2084 * the client are no longer valid. 2085 */ 2086 if (!nd->nd_repstat && toff && verf != at.na_filerev) 2087 nd->nd_repstat = NFSERR_BAD_COOKIE; 2088 #endif 2089 } 2090 if (!nd->nd_repstat && vp->v_type != VDIR) 2091 nd->nd_repstat = NFSERR_NOTDIR; 2092 if (nd->nd_repstat == 0 && cnt == 0) { 2093 if (nd->nd_flag & ND_NFSV2) 2094 /* NFSv2 does not have NFSERR_TOOSMALL */ 2095 nd->nd_repstat = EPERM; 2096 else 2097 nd->nd_repstat = NFSERR_TOOSMALL; 2098 } 2099 if (!nd->nd_repstat) 2100 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2101 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2102 NFSACCCHK_VPISLOCKED, NULL); 2103 if (nd->nd_repstat) { 2104 vput(vp); 2105 if (nd->nd_flag & ND_NFSV3) 2106 nfsrv_postopattr(nd, getret, &at); 2107 goto out; 2108 } 2109 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2110 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2111 again: 2112 eofflag = 0; 2113 if (cookies) { 2114 free(cookies, M_TEMP); 2115 cookies = NULL; 2116 } 2117 2118 iv.iov_base = rbuf; 2119 iv.iov_len = siz; 2120 io.uio_iov = &iv; 2121 io.uio_iovcnt = 1; 2122 io.uio_offset = (off_t)off; 2123 io.uio_resid = siz; 2124 io.uio_segflg = UIO_SYSSPACE; 2125 io.uio_rw = UIO_READ; 2126 io.uio_td = NULL; 2127 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2128 &cookies); 2129 off = (u_int64_t)io.uio_offset; 2130 if (io.uio_resid) 2131 siz -= io.uio_resid; 2132 2133 if (!cookies && !nd->nd_repstat) 2134 nd->nd_repstat = NFSERR_PERM; 2135 if (nd->nd_flag & ND_NFSV3) { 2136 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2137 if (!nd->nd_repstat) 2138 nd->nd_repstat = getret; 2139 } 2140 2141 /* 2142 * Handles the failed cases. nd->nd_repstat == 0 past here. 2143 */ 2144 if (nd->nd_repstat) { 2145 vput(vp); 2146 free(rbuf, M_TEMP); 2147 if (cookies) 2148 free(cookies, M_TEMP); 2149 if (nd->nd_flag & ND_NFSV3) 2150 nfsrv_postopattr(nd, getret, &at); 2151 goto out; 2152 } 2153 /* 2154 * If nothing read, return eof 2155 * rpc reply 2156 */ 2157 if (siz == 0) { 2158 vput(vp); 2159 if (nd->nd_flag & ND_NFSV2) { 2160 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2161 } else { 2162 nfsrv_postopattr(nd, getret, &at); 2163 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2164 txdr_hyper(at.na_filerev, tl); 2165 tl += 2; 2166 } 2167 *tl++ = newnfs_false; 2168 *tl = newnfs_true; 2169 free(rbuf, M_TEMP); 2170 free(cookies, M_TEMP); 2171 goto out; 2172 } 2173 2174 /* 2175 * Check for degenerate cases of nothing useful read. 2176 * If so go try again 2177 */ 2178 cpos = rbuf; 2179 cend = rbuf + siz; 2180 dp = (struct dirent *)cpos; 2181 cookiep = cookies; 2182 2183 /* 2184 * For some reason FreeBSD's ufs_readdir() chooses to back the 2185 * directory offset up to a block boundary, so it is necessary to 2186 * skip over the records that precede the requested offset. This 2187 * requires the assumption that file offset cookies monotonically 2188 * increase. 2189 */ 2190 while (cpos < cend && ncookies > 0 && 2191 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2192 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { 2193 cpos += dp->d_reclen; 2194 dp = (struct dirent *)cpos; 2195 cookiep++; 2196 ncookies--; 2197 } 2198 if (cpos >= cend || ncookies == 0) { 2199 siz = fullsiz; 2200 toff = off; 2201 goto again; 2202 } 2203 vput(vp); 2204 2205 /* 2206 * If cnt > MCLBYTES and the reply will not be saved, use 2207 * ext_pgs mbufs for TLS. 2208 * For NFSv4.0, we do not know for sure if the reply will 2209 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2210 */ 2211 if (cnt > MCLBYTES && siz > MCLBYTES && 2212 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2213 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2214 nd->nd_flag |= ND_EXTPG; 2215 2216 /* 2217 * dirlen is the size of the reply, including all XDR and must 2218 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate 2219 * if the XDR should be included in "count", but to be safe, we do. 2220 * (Include the two booleans at the end of the reply in dirlen now.) 2221 */ 2222 if (nd->nd_flag & ND_NFSV3) { 2223 nfsrv_postopattr(nd, getret, &at); 2224 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2225 txdr_hyper(at.na_filerev, tl); 2226 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2227 } else { 2228 dirlen = 2 * NFSX_UNSIGNED; 2229 } 2230 2231 /* Loop through the records and build reply */ 2232 while (cpos < cend && ncookies > 0) { 2233 nlen = dp->d_namlen; 2234 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2235 nlen <= NFS_MAXNAMLEN) { 2236 if (nd->nd_flag & ND_NFSV3) 2237 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2238 else 2239 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2240 if (dirlen > cnt) { 2241 eofflag = 0; 2242 break; 2243 } 2244 2245 /* 2246 * Build the directory record xdr from 2247 * the dirent entry. 2248 */ 2249 if (nd->nd_flag & ND_NFSV3) { 2250 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2251 *tl++ = newnfs_true; 2252 *tl++ = 0; 2253 } else { 2254 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2255 *tl++ = newnfs_true; 2256 } 2257 *tl = txdr_unsigned(dp->d_fileno); 2258 (void) nfsm_strtom(nd, dp->d_name, nlen); 2259 if (nd->nd_flag & ND_NFSV3) { 2260 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2261 *tl++ = 0; 2262 } else 2263 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 2264 *tl = txdr_unsigned(*cookiep); 2265 } 2266 cpos += dp->d_reclen; 2267 dp = (struct dirent *)cpos; 2268 cookiep++; 2269 ncookies--; 2270 } 2271 if (cpos < cend) 2272 eofflag = 0; 2273 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2274 *tl++ = newnfs_false; 2275 if (eofflag) 2276 *tl = newnfs_true; 2277 else 2278 *tl = newnfs_false; 2279 free(rbuf, M_TEMP); 2280 free(cookies, M_TEMP); 2281 2282 out: 2283 NFSEXITCODE2(0, nd); 2284 return (0); 2285 nfsmout: 2286 vput(vp); 2287 NFSEXITCODE2(error, nd); 2288 return (error); 2289 } 2290 2291 /* 2292 * Readdirplus for V3 and Readdir for V4. 2293 */ 2294 int 2295 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, 2296 struct vnode *vp, struct nfsexstuff *exp) 2297 { 2298 struct dirent *dp; 2299 u_int32_t *tl; 2300 int dirlen; 2301 char *cpos, *cend, *rbuf; 2302 struct vnode *nvp; 2303 fhandle_t nfh; 2304 struct nfsvattr nva, at, *nvap = &nva; 2305 struct mbuf *mb0, *mb1; 2306 struct nfsreferral *refp; 2307 int nlen, r, error = 0, getret = 1, usevget = 1; 2308 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; 2309 caddr_t bpos0, bpos1; 2310 u_int64_t off, toff, verf; 2311 u_long *cookies = NULL, *cookiep; 2312 nfsattrbit_t attrbits, rderrbits, savbits; 2313 struct uio io; 2314 struct iovec iv; 2315 struct componentname cn; 2316 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; 2317 struct mount *mp, *new_mp; 2318 uint64_t mounted_on_fileno; 2319 struct thread *p = curthread; 2320 int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1; 2321 2322 if (nd->nd_repstat) { 2323 nfsrv_postopattr(nd, getret, &at); 2324 goto out; 2325 } 2326 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2327 off = fxdr_hyper(tl); 2328 toff = off; 2329 tl += 2; 2330 verf = fxdr_hyper(tl); 2331 tl += 2; 2332 siz = fxdr_unsigned(int, *tl++); 2333 cnt = fxdr_unsigned(int, *tl); 2334 2335 /* 2336 * Use the server's maximum data transfer size as the upper bound 2337 * on reply datalen. 2338 */ 2339 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2340 cnt = NFS_SRVMAXDATA(nd); 2341 2342 /* 2343 * siz is a "hint" of how much directory information (name, fileid, 2344 * cookie) should be in the reply. At least one client "hints" 0, 2345 * so I set it to cnt for that case. I also round it up to the 2346 * next multiple of DIRBLKSIZ. 2347 * Since the size of a Readdirplus directory entry reply will always 2348 * be greater than a directory entry returned by VOP_READDIR(), it 2349 * does not make sense to read more than NFS_SRVMAXDATA() via 2350 * VOP_READDIR(). 2351 */ 2352 if (siz <= 0) 2353 siz = cnt; 2354 else if (siz > NFS_SRVMAXDATA(nd)) 2355 siz = NFS_SRVMAXDATA(nd); 2356 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2357 2358 if (nd->nd_flag & ND_NFSV4) { 2359 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 2360 if (error) 2361 goto nfsmout; 2362 NFSSET_ATTRBIT(&savbits, &attrbits); 2363 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); 2364 NFSZERO_ATTRBIT(&rderrbits); 2365 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); 2366 } else { 2367 NFSZERO_ATTRBIT(&attrbits); 2368 } 2369 fullsiz = siz; 2370 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2371 #if 0 2372 if (!nd->nd_repstat) { 2373 if (off && verf != at.na_filerev) { 2374 /* 2375 * va_filerev is not sufficient as a cookie verifier, 2376 * since it is not supposed to change when entries are 2377 * removed/added unless that offset cookies returned to 2378 * the client are no longer valid. 2379 */ 2380 if (nd->nd_flag & ND_NFSV4) { 2381 nd->nd_repstat = NFSERR_NOTSAME; 2382 } else { 2383 nd->nd_repstat = NFSERR_BAD_COOKIE; 2384 } 2385 } 2386 } 2387 #endif 2388 if (!nd->nd_repstat && vp->v_type != VDIR) 2389 nd->nd_repstat = NFSERR_NOTDIR; 2390 if (!nd->nd_repstat && cnt == 0) 2391 nd->nd_repstat = NFSERR_TOOSMALL; 2392 if (!nd->nd_repstat) 2393 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2394 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2395 NFSACCCHK_VPISLOCKED, NULL); 2396 if (nd->nd_repstat) { 2397 vput(vp); 2398 if (nd->nd_flag & ND_NFSV3) 2399 nfsrv_postopattr(nd, getret, &at); 2400 goto out; 2401 } 2402 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2403 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; 2404 2405 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2406 again: 2407 eofflag = 0; 2408 if (cookies) { 2409 free(cookies, M_TEMP); 2410 cookies = NULL; 2411 } 2412 2413 iv.iov_base = rbuf; 2414 iv.iov_len = siz; 2415 io.uio_iov = &iv; 2416 io.uio_iovcnt = 1; 2417 io.uio_offset = (off_t)off; 2418 io.uio_resid = siz; 2419 io.uio_segflg = UIO_SYSSPACE; 2420 io.uio_rw = UIO_READ; 2421 io.uio_td = NULL; 2422 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2423 &cookies); 2424 off = (u_int64_t)io.uio_offset; 2425 if (io.uio_resid) 2426 siz -= io.uio_resid; 2427 2428 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2429 2430 if (!cookies && !nd->nd_repstat) 2431 nd->nd_repstat = NFSERR_PERM; 2432 if (!nd->nd_repstat) 2433 nd->nd_repstat = getret; 2434 if (nd->nd_repstat) { 2435 vput(vp); 2436 if (cookies) 2437 free(cookies, M_TEMP); 2438 free(rbuf, M_TEMP); 2439 if (nd->nd_flag & ND_NFSV3) 2440 nfsrv_postopattr(nd, getret, &at); 2441 goto out; 2442 } 2443 /* 2444 * If nothing read, return eof 2445 * rpc reply 2446 */ 2447 if (siz == 0) { 2448 vput(vp); 2449 if (nd->nd_flag & ND_NFSV3) 2450 nfsrv_postopattr(nd, getret, &at); 2451 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2452 txdr_hyper(at.na_filerev, tl); 2453 tl += 2; 2454 *tl++ = newnfs_false; 2455 *tl = newnfs_true; 2456 free(cookies, M_TEMP); 2457 free(rbuf, M_TEMP); 2458 goto out; 2459 } 2460 2461 /* 2462 * Check for degenerate cases of nothing useful read. 2463 * If so go try again 2464 */ 2465 cpos = rbuf; 2466 cend = rbuf + siz; 2467 dp = (struct dirent *)cpos; 2468 cookiep = cookies; 2469 2470 /* 2471 * For some reason FreeBSD's ufs_readdir() chooses to back the 2472 * directory offset up to a block boundary, so it is necessary to 2473 * skip over the records that precede the requested offset. This 2474 * requires the assumption that file offset cookies monotonically 2475 * increase. 2476 */ 2477 while (cpos < cend && ncookies > 0 && 2478 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2479 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || 2480 ((nd->nd_flag & ND_NFSV4) && 2481 ((dp->d_namlen == 1 && dp->d_name[0] == '.') || 2482 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { 2483 cpos += dp->d_reclen; 2484 dp = (struct dirent *)cpos; 2485 cookiep++; 2486 ncookies--; 2487 } 2488 if (cpos >= cend || ncookies == 0) { 2489 siz = fullsiz; 2490 toff = off; 2491 goto again; 2492 } 2493 2494 /* 2495 * Busy the file system so that the mount point won't go away 2496 * and, as such, VFS_VGET() can be used safely. 2497 */ 2498 mp = vp->v_mount; 2499 vfs_ref(mp); 2500 NFSVOPUNLOCK(vp); 2501 nd->nd_repstat = vfs_busy(mp, 0); 2502 vfs_rel(mp); 2503 if (nd->nd_repstat != 0) { 2504 vrele(vp); 2505 free(cookies, M_TEMP); 2506 free(rbuf, M_TEMP); 2507 if (nd->nd_flag & ND_NFSV3) 2508 nfsrv_postopattr(nd, getret, &at); 2509 goto out; 2510 } 2511 2512 /* 2513 * Check to see if entries in this directory can be safely acquired 2514 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. 2515 * ZFS snapshot directories need VOP_LOOKUP(), so that any 2516 * automount of the snapshot directory that is required will 2517 * be done. 2518 * This needs to be done here for NFSv4, since NFSv4 never does 2519 * a VFS_VGET() for "." or "..". 2520 */ 2521 if (is_zfs == 1) { 2522 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); 2523 if (r == EOPNOTSUPP) { 2524 usevget = 0; 2525 cn.cn_nameiop = LOOKUP; 2526 cn.cn_lkflags = LK_SHARED | LK_RETRY; 2527 cn.cn_cred = nd->nd_cred; 2528 } else if (r == 0) 2529 vput(nvp); 2530 } 2531 2532 /* 2533 * If the reply is likely to exceed MCLBYTES and the reply will 2534 * not be saved, use ext_pgs mbufs for TLS. 2535 * It is difficult to predict how large each entry will be and 2536 * how many entries have been read, so just assume the directory 2537 * entries grow by a factor of 4 when attributes are included. 2538 * For NFSv4.0, we do not know for sure if the reply will 2539 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2540 */ 2541 if (cnt > MCLBYTES && siz > MCLBYTES / 4 && 2542 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2543 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2544 nd->nd_flag |= ND_EXTPG; 2545 2546 /* 2547 * Save this position, in case there is an error before one entry 2548 * is created. 2549 */ 2550 mb0 = nd->nd_mb; 2551 bpos0 = nd->nd_bpos; 2552 bextpg0 = nd->nd_bextpg; 2553 bextpgsiz0 = nd->nd_bextpgsiz; 2554 2555 /* 2556 * Fill in the first part of the reply. 2557 * dirlen is the reply length in bytes and cannot exceed cnt. 2558 * (Include the two booleans at the end of the reply in dirlen now, 2559 * so we recognize when we have exceeded cnt.) 2560 */ 2561 if (nd->nd_flag & ND_NFSV3) { 2562 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2563 nfsrv_postopattr(nd, getret, &at); 2564 } else { 2565 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; 2566 } 2567 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); 2568 txdr_hyper(at.na_filerev, tl); 2569 2570 /* 2571 * Save this position, in case there is an empty reply needed. 2572 */ 2573 mb1 = nd->nd_mb; 2574 bpos1 = nd->nd_bpos; 2575 bextpg1 = nd->nd_bextpg; 2576 bextpgsiz1 = nd->nd_bextpgsiz; 2577 2578 /* Loop through the records and build reply */ 2579 entrycnt = 0; 2580 while (cpos < cend && ncookies > 0 && dirlen < cnt) { 2581 nlen = dp->d_namlen; 2582 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2583 nlen <= NFS_MAXNAMLEN && 2584 ((nd->nd_flag & ND_NFSV3) || nlen > 2 || 2585 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) 2586 || (nlen == 1 && dp->d_name[0] != '.'))) { 2587 /* 2588 * Save the current position in the reply, in case 2589 * this entry exceeds cnt. 2590 */ 2591 mb1 = nd->nd_mb; 2592 bpos1 = nd->nd_bpos; 2593 bextpg1 = nd->nd_bextpg; 2594 bextpgsiz1 = nd->nd_bextpgsiz; 2595 2596 /* 2597 * For readdir_and_lookup get the vnode using 2598 * the file number. 2599 */ 2600 nvp = NULL; 2601 refp = NULL; 2602 r = 0; 2603 at_root = 0; 2604 needs_unbusy = 0; 2605 new_mp = mp; 2606 mounted_on_fileno = (uint64_t)dp->d_fileno; 2607 if ((nd->nd_flag & ND_NFSV3) || 2608 NFSNONZERO_ATTRBIT(&savbits)) { 2609 if (nd->nd_flag & ND_NFSV4) 2610 refp = nfsv4root_getreferral(NULL, 2611 vp, dp->d_fileno); 2612 if (refp == NULL) { 2613 if (usevget) 2614 r = VFS_VGET(mp, dp->d_fileno, 2615 LK_SHARED, &nvp); 2616 else 2617 r = EOPNOTSUPP; 2618 if (r == EOPNOTSUPP) { 2619 if (usevget) { 2620 usevget = 0; 2621 cn.cn_nameiop = LOOKUP; 2622 cn.cn_lkflags = 2623 LK_SHARED | 2624 LK_RETRY; 2625 cn.cn_cred = 2626 nd->nd_cred; 2627 } 2628 cn.cn_nameptr = dp->d_name; 2629 cn.cn_namelen = nlen; 2630 cn.cn_flags = ISLASTCN | 2631 NOFOLLOW | LOCKLEAF; 2632 if (nlen == 2 && 2633 dp->d_name[0] == '.' && 2634 dp->d_name[1] == '.') 2635 cn.cn_flags |= 2636 ISDOTDOT; 2637 if (NFSVOPLOCK(vp, LK_SHARED) 2638 != 0) { 2639 nd->nd_repstat = EPERM; 2640 break; 2641 } 2642 if ((vp->v_vflag & VV_ROOT) != 0 2643 && (cn.cn_flags & ISDOTDOT) 2644 != 0) { 2645 vref(vp); 2646 nvp = vp; 2647 r = 0; 2648 } else { 2649 r = VOP_LOOKUP(vp, &nvp, 2650 &cn); 2651 if (vp != nvp) 2652 NFSVOPUNLOCK(vp); 2653 } 2654 } 2655 2656 /* 2657 * For NFSv4, check to see if nvp is 2658 * a mount point and get the mount 2659 * point vnode, as required. 2660 */ 2661 if (r == 0 && 2662 nfsrv_enable_crossmntpt != 0 && 2663 (nd->nd_flag & ND_NFSV4) != 0 && 2664 nvp->v_type == VDIR && 2665 nvp->v_mountedhere != NULL) { 2666 new_mp = nvp->v_mountedhere; 2667 r = vfs_busy(new_mp, 0); 2668 vput(nvp); 2669 nvp = NULL; 2670 if (r == 0) { 2671 r = VFS_ROOT(new_mp, 2672 LK_SHARED, &nvp); 2673 needs_unbusy = 1; 2674 if (r == 0) 2675 at_root = 1; 2676 } 2677 } 2678 } 2679 2680 /* 2681 * If we failed to look up the entry, then it 2682 * has become invalid, most likely removed. 2683 */ 2684 if (r != 0) { 2685 if (needs_unbusy) 2686 vfs_unbusy(new_mp); 2687 goto invalid; 2688 } 2689 KASSERT(refp != NULL || nvp != NULL, 2690 ("%s: undetected lookup error", __func__)); 2691 2692 if (refp == NULL && 2693 ((nd->nd_flag & ND_NFSV3) || 2694 NFSNONZERO_ATTRBIT(&attrbits))) { 2695 r = nfsvno_getfh(nvp, &nfh, p); 2696 if (!r) 2697 r = nfsvno_getattr(nvp, nvap, nd, p, 2698 1, &attrbits); 2699 if (r == 0 && is_zfs == 1 && 2700 nfsrv_enable_crossmntpt != 0 && 2701 (nd->nd_flag & ND_NFSV4) != 0 && 2702 nvp->v_type == VDIR && 2703 vp->v_mount != nvp->v_mount) { 2704 /* 2705 * For a ZFS snapshot, there is a 2706 * pseudo mount that does not set 2707 * v_mountedhere, so it needs to 2708 * be detected via a different 2709 * mount structure. 2710 */ 2711 at_root = 1; 2712 if (new_mp == mp) 2713 new_mp = nvp->v_mount; 2714 } 2715 } 2716 2717 /* 2718 * If we failed to get attributes of the entry, 2719 * then just skip it for NFSv3 (the traditional 2720 * behavior in the old NFS server). 2721 * For NFSv4 the behavior is controlled by 2722 * RDATTRERROR: we either ignore the error or 2723 * fail the request. 2724 * Note that RDATTRERROR is never set for NFSv3. 2725 */ 2726 if (r != 0) { 2727 if (!NFSISSET_ATTRBIT(&attrbits, 2728 NFSATTRBIT_RDATTRERROR)) { 2729 vput(nvp); 2730 if (needs_unbusy != 0) 2731 vfs_unbusy(new_mp); 2732 if ((nd->nd_flag & ND_NFSV3)) 2733 goto invalid; 2734 nd->nd_repstat = r; 2735 break; 2736 } 2737 } 2738 } 2739 2740 /* 2741 * Build the directory record xdr 2742 */ 2743 if (nd->nd_flag & ND_NFSV3) { 2744 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2745 *tl++ = newnfs_true; 2746 *tl++ = 0; 2747 *tl = txdr_unsigned(dp->d_fileno); 2748 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2749 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2750 *tl++ = 0; 2751 *tl = txdr_unsigned(*cookiep); 2752 nfsrv_postopattr(nd, 0, nvap); 2753 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1); 2754 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); 2755 if (nvp != NULL) 2756 vput(nvp); 2757 } else { 2758 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2759 *tl++ = newnfs_true; 2760 *tl++ = 0; 2761 *tl = txdr_unsigned(*cookiep); 2762 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2763 if (nvp != NULL) { 2764 supports_nfsv4acls = 2765 nfs_supportsnfsv4acls(nvp); 2766 NFSVOPUNLOCK(nvp); 2767 } else 2768 supports_nfsv4acls = 0; 2769 if (refp != NULL) { 2770 dirlen += nfsrv_putreferralattr(nd, 2771 &savbits, refp, 0, 2772 &nd->nd_repstat); 2773 if (nd->nd_repstat) { 2774 if (nvp != NULL) 2775 vrele(nvp); 2776 if (needs_unbusy != 0) 2777 vfs_unbusy(new_mp); 2778 break; 2779 } 2780 } else if (r) { 2781 dirlen += nfsvno_fillattr(nd, new_mp, 2782 nvp, nvap, &nfh, r, &rderrbits, 2783 nd->nd_cred, p, isdgram, 0, 2784 supports_nfsv4acls, at_root, 2785 mounted_on_fileno); 2786 } else { 2787 dirlen += nfsvno_fillattr(nd, new_mp, 2788 nvp, nvap, &nfh, r, &attrbits, 2789 nd->nd_cred, p, isdgram, 0, 2790 supports_nfsv4acls, at_root, 2791 mounted_on_fileno); 2792 } 2793 if (nvp != NULL) 2794 vrele(nvp); 2795 dirlen += (3 * NFSX_UNSIGNED); 2796 } 2797 if (needs_unbusy != 0) 2798 vfs_unbusy(new_mp); 2799 if (dirlen <= cnt) 2800 entrycnt++; 2801 } 2802 invalid: 2803 cpos += dp->d_reclen; 2804 dp = (struct dirent *)cpos; 2805 cookiep++; 2806 ncookies--; 2807 } 2808 vrele(vp); 2809 vfs_unbusy(mp); 2810 2811 /* 2812 * If dirlen > cnt, we must strip off the last entry. If that 2813 * results in an empty reply, report NFSERR_TOOSMALL. 2814 */ 2815 if (dirlen > cnt || nd->nd_repstat) { 2816 if (!nd->nd_repstat && entrycnt == 0) 2817 nd->nd_repstat = NFSERR_TOOSMALL; 2818 if (nd->nd_repstat) { 2819 nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); 2820 if (nd->nd_flag & ND_NFSV3) 2821 nfsrv_postopattr(nd, getret, &at); 2822 } else 2823 nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); 2824 eofflag = 0; 2825 } else if (cpos < cend) 2826 eofflag = 0; 2827 if (!nd->nd_repstat) { 2828 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2829 *tl++ = newnfs_false; 2830 if (eofflag) 2831 *tl = newnfs_true; 2832 else 2833 *tl = newnfs_false; 2834 } 2835 free(cookies, M_TEMP); 2836 free(rbuf, M_TEMP); 2837 2838 out: 2839 NFSEXITCODE2(0, nd); 2840 return (0); 2841 nfsmout: 2842 vput(vp); 2843 NFSEXITCODE2(error, nd); 2844 return (error); 2845 } 2846 2847 /* 2848 * Get the settable attributes out of the mbuf list. 2849 * (Return 0 or EBADRPC) 2850 */ 2851 int 2852 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2853 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2854 { 2855 u_int32_t *tl; 2856 struct nfsv2_sattr *sp; 2857 int error = 0, toclient = 0; 2858 2859 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 2860 case ND_NFSV2: 2861 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2862 /* 2863 * Some old clients didn't fill in the high order 16bits. 2864 * --> check the low order 2 bytes for 0xffff 2865 */ 2866 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 2867 nvap->na_mode = nfstov_mode(sp->sa_mode); 2868 if (sp->sa_uid != newnfs_xdrneg1) 2869 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 2870 if (sp->sa_gid != newnfs_xdrneg1) 2871 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 2872 if (sp->sa_size != newnfs_xdrneg1) 2873 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 2874 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 2875 #ifdef notyet 2876 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 2877 #else 2878 nvap->na_atime.tv_sec = 2879 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 2880 nvap->na_atime.tv_nsec = 0; 2881 #endif 2882 } 2883 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 2884 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 2885 break; 2886 case ND_NFSV3: 2887 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2888 if (*tl == newnfs_true) { 2889 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2890 nvap->na_mode = nfstov_mode(*tl); 2891 } 2892 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2893 if (*tl == newnfs_true) { 2894 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2895 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 2896 } 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_gid = fxdr_unsigned(gid_t, *tl); 2901 } 2902 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2903 if (*tl == newnfs_true) { 2904 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2905 nvap->na_size = fxdr_hyper(tl); 2906 } 2907 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2908 switch (fxdr_unsigned(int, *tl)) { 2909 case NFSV3SATTRTIME_TOCLIENT: 2910 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2911 fxdr_nfsv3time(tl, &nvap->na_atime); 2912 toclient = 1; 2913 break; 2914 case NFSV3SATTRTIME_TOSERVER: 2915 vfs_timestamp(&nvap->na_atime); 2916 nvap->na_vaflags |= VA_UTIMES_NULL; 2917 break; 2918 } 2919 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2920 switch (fxdr_unsigned(int, *tl)) { 2921 case NFSV3SATTRTIME_TOCLIENT: 2922 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2923 fxdr_nfsv3time(tl, &nvap->na_mtime); 2924 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2925 break; 2926 case NFSV3SATTRTIME_TOSERVER: 2927 vfs_timestamp(&nvap->na_mtime); 2928 if (!toclient) 2929 nvap->na_vaflags |= VA_UTIMES_NULL; 2930 break; 2931 } 2932 break; 2933 case ND_NFSV4: 2934 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); 2935 } 2936 nfsmout: 2937 NFSEXITCODE2(error, nd); 2938 return (error); 2939 } 2940 2941 /* 2942 * Handle the setable attributes for V4. 2943 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 2944 */ 2945 int 2946 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2947 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2948 { 2949 u_int32_t *tl; 2950 int attrsum = 0; 2951 int i, j; 2952 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 2953 int moderet, toclient = 0; 2954 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 2955 uid_t uid; 2956 gid_t gid; 2957 u_short mode, mask; /* Same type as va_mode. */ 2958 struct vattr va; 2959 2960 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 2961 if (error) 2962 goto nfsmout; 2963 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2964 attrsize = fxdr_unsigned(int, *tl); 2965 2966 /* 2967 * Loop around getting the setable attributes. If an unsupported 2968 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 2969 */ 2970 if (retnotsup) { 2971 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2972 bitpos = NFSATTRBIT_MAX; 2973 } else { 2974 bitpos = 0; 2975 } 2976 moderet = 0; 2977 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 2978 if (attrsum > attrsize) { 2979 error = NFSERR_BADXDR; 2980 goto nfsmout; 2981 } 2982 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 2983 switch (bitpos) { 2984 case NFSATTRBIT_SIZE: 2985 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 2986 if (vp != NULL && vp->v_type != VREG) { 2987 error = (vp->v_type == VDIR) ? NFSERR_ISDIR : 2988 NFSERR_INVAL; 2989 goto nfsmout; 2990 } 2991 nvap->na_size = fxdr_hyper(tl); 2992 attrsum += NFSX_HYPER; 2993 break; 2994 case NFSATTRBIT_ACL: 2995 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize, 2996 p); 2997 if (error) 2998 goto nfsmout; 2999 if (aceerr && !nd->nd_repstat) 3000 nd->nd_repstat = aceerr; 3001 attrsum += aclsize; 3002 break; 3003 case NFSATTRBIT_ARCHIVE: 3004 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3005 if (!nd->nd_repstat) 3006 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3007 attrsum += NFSX_UNSIGNED; 3008 break; 3009 case NFSATTRBIT_HIDDEN: 3010 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3011 if (!nd->nd_repstat) 3012 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3013 attrsum += NFSX_UNSIGNED; 3014 break; 3015 case NFSATTRBIT_MIMETYPE: 3016 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3017 i = fxdr_unsigned(int, *tl); 3018 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 3019 if (error) 3020 goto nfsmout; 3021 if (!nd->nd_repstat) 3022 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3023 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 3024 break; 3025 case NFSATTRBIT_MODE: 3026 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ 3027 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3028 nvap->na_mode = nfstov_mode(*tl); 3029 attrsum += NFSX_UNSIGNED; 3030 break; 3031 case NFSATTRBIT_OWNER: 3032 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3033 j = fxdr_unsigned(int, *tl); 3034 if (j < 0) { 3035 error = NFSERR_BADXDR; 3036 goto nfsmout; 3037 } 3038 if (j > NFSV4_SMALLSTR) 3039 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3040 else 3041 cp = namestr; 3042 error = nfsrv_mtostr(nd, cp, j); 3043 if (error) { 3044 if (j > NFSV4_SMALLSTR) 3045 free(cp, M_NFSSTRING); 3046 goto nfsmout; 3047 } 3048 if (!nd->nd_repstat) { 3049 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, 3050 &uid); 3051 if (!nd->nd_repstat) 3052 nvap->na_uid = uid; 3053 } 3054 if (j > NFSV4_SMALLSTR) 3055 free(cp, M_NFSSTRING); 3056 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3057 break; 3058 case NFSATTRBIT_OWNERGROUP: 3059 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3060 j = fxdr_unsigned(int, *tl); 3061 if (j < 0) { 3062 error = NFSERR_BADXDR; 3063 goto nfsmout; 3064 } 3065 if (j > NFSV4_SMALLSTR) 3066 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3067 else 3068 cp = namestr; 3069 error = nfsrv_mtostr(nd, cp, j); 3070 if (error) { 3071 if (j > NFSV4_SMALLSTR) 3072 free(cp, M_NFSSTRING); 3073 goto nfsmout; 3074 } 3075 if (!nd->nd_repstat) { 3076 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, 3077 &gid); 3078 if (!nd->nd_repstat) 3079 nvap->na_gid = gid; 3080 } 3081 if (j > NFSV4_SMALLSTR) 3082 free(cp, M_NFSSTRING); 3083 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3084 break; 3085 case NFSATTRBIT_SYSTEM: 3086 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3087 if (!nd->nd_repstat) 3088 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3089 attrsum += NFSX_UNSIGNED; 3090 break; 3091 case NFSATTRBIT_TIMEACCESSSET: 3092 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3093 attrsum += NFSX_UNSIGNED; 3094 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3095 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3096 fxdr_nfsv4time(tl, &nvap->na_atime); 3097 toclient = 1; 3098 attrsum += NFSX_V4TIME; 3099 } else { 3100 vfs_timestamp(&nvap->na_atime); 3101 nvap->na_vaflags |= VA_UTIMES_NULL; 3102 } 3103 break; 3104 case NFSATTRBIT_TIMEBACKUP: 3105 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3106 if (!nd->nd_repstat) 3107 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3108 attrsum += NFSX_V4TIME; 3109 break; 3110 case NFSATTRBIT_TIMECREATE: 3111 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3112 fxdr_nfsv4time(tl, &nvap->na_btime); 3113 attrsum += NFSX_V4TIME; 3114 break; 3115 case NFSATTRBIT_TIMEMODIFYSET: 3116 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3117 attrsum += NFSX_UNSIGNED; 3118 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3119 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3120 fxdr_nfsv4time(tl, &nvap->na_mtime); 3121 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3122 attrsum += NFSX_V4TIME; 3123 } else { 3124 vfs_timestamp(&nvap->na_mtime); 3125 if (!toclient) 3126 nvap->na_vaflags |= VA_UTIMES_NULL; 3127 } 3128 break; 3129 case NFSATTRBIT_MODESETMASKED: 3130 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3131 mode = fxdr_unsigned(u_short, *tl++); 3132 mask = fxdr_unsigned(u_short, *tl); 3133 /* 3134 * vp == NULL implies an Open/Create operation. 3135 * This attribute can only be used for Setattr and 3136 * only for NFSv4.1 or higher. 3137 * If moderet != 0, a mode attribute has also been 3138 * specified and this attribute cannot be done in the 3139 * same Setattr operation. 3140 */ 3141 if ((nd->nd_flag & ND_NFSV41) == 0) 3142 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3143 else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 || 3144 vp == NULL) 3145 nd->nd_repstat = NFSERR_INVAL; 3146 else if (moderet == 0) 3147 moderet = VOP_GETATTR(vp, &va, nd->nd_cred); 3148 if (moderet == 0) 3149 nvap->na_mode = (mode & mask) | 3150 (va.va_mode & ~mask); 3151 else 3152 nd->nd_repstat = moderet; 3153 attrsum += 2 * NFSX_UNSIGNED; 3154 break; 3155 default: 3156 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3157 /* 3158 * set bitpos so we drop out of the loop. 3159 */ 3160 bitpos = NFSATTRBIT_MAX; 3161 break; 3162 } 3163 } 3164 3165 /* 3166 * some clients pad the attrlist, so we need to skip over the 3167 * padding. 3168 */ 3169 if (attrsum > attrsize) { 3170 error = NFSERR_BADXDR; 3171 } else { 3172 attrsize = NFSM_RNDUP(attrsize); 3173 if (attrsum < attrsize) 3174 error = nfsm_advance(nd, attrsize - attrsum, -1); 3175 } 3176 nfsmout: 3177 NFSEXITCODE2(error, nd); 3178 return (error); 3179 } 3180 3181 /* 3182 * Check/setup export credentials. 3183 */ 3184 int 3185 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 3186 struct ucred *credanon, bool testsec) 3187 { 3188 int error; 3189 3190 /* 3191 * Check/setup credentials. 3192 */ 3193 if (nd->nd_flag & ND_GSS) 3194 exp->nes_exflag &= ~MNT_EXPORTANON; 3195 3196 /* 3197 * Check to see if the operation is allowed for this security flavor. 3198 */ 3199 error = 0; 3200 if (testsec) { 3201 error = nfsvno_testexp(nd, exp); 3202 if (error != 0) 3203 goto out; 3204 } 3205 3206 /* 3207 * Check to see if the file system is exported V4 only. 3208 */ 3209 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 3210 error = NFSERR_PROGNOTV4; 3211 goto out; 3212 } 3213 3214 /* 3215 * Now, map the user credentials. 3216 * (Note that ND_AUTHNONE will only be set for an NFSv3 3217 * Fsinfo RPC. If set for anything else, this code might need 3218 * to change.) 3219 */ 3220 if (NFSVNO_EXPORTED(exp)) { 3221 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || 3222 NFSVNO_EXPORTANON(exp) || 3223 (nd->nd_flag & ND_AUTHNONE) != 0) { 3224 nd->nd_cred->cr_uid = credanon->cr_uid; 3225 nd->nd_cred->cr_gid = credanon->cr_gid; 3226 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 3227 credanon->cr_groups); 3228 } else if ((nd->nd_flag & ND_GSS) == 0) { 3229 /* 3230 * If using AUTH_SYS, call nfsrv_getgrpscred() to see 3231 * if there is a replacement credential with a group 3232 * list set up by "nfsuserd -manage-gids". 3233 * If there is no replacement, nfsrv_getgrpscred() 3234 * simply returns its argument. 3235 */ 3236 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); 3237 } 3238 } 3239 3240 out: 3241 NFSEXITCODE2(error, nd); 3242 return (error); 3243 } 3244 3245 /* 3246 * Check exports. 3247 */ 3248 int 3249 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 3250 struct ucred **credp) 3251 { 3252 int error; 3253 3254 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3255 &exp->nes_numsecflavor, exp->nes_secflavors); 3256 if (error) { 3257 if (nfs_rootfhset) { 3258 exp->nes_exflag = 0; 3259 exp->nes_numsecflavor = 0; 3260 error = 0; 3261 } 3262 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3263 MAXSECFLAVORS) { 3264 printf("nfsvno_checkexp: numsecflavors out of range\n"); 3265 exp->nes_numsecflavor = 0; 3266 error = EACCES; 3267 } 3268 NFSEXITCODE(error); 3269 return (error); 3270 } 3271 3272 /* 3273 * Get a vnode for a file handle and export stuff. 3274 */ 3275 int 3276 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 3277 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 3278 struct ucred **credp) 3279 { 3280 int error; 3281 3282 *credp = NULL; 3283 exp->nes_numsecflavor = 0; 3284 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); 3285 if (error != 0) 3286 /* Make sure the server replies ESTALE to the client. */ 3287 error = ESTALE; 3288 if (nam && !error) { 3289 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3290 &exp->nes_numsecflavor, exp->nes_secflavors); 3291 if (error) { 3292 if (nfs_rootfhset) { 3293 exp->nes_exflag = 0; 3294 exp->nes_numsecflavor = 0; 3295 error = 0; 3296 } else { 3297 vput(*vpp); 3298 } 3299 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3300 MAXSECFLAVORS) { 3301 printf("nfsvno_fhtovp: numsecflavors out of range\n"); 3302 exp->nes_numsecflavor = 0; 3303 error = EACCES; 3304 vput(*vpp); 3305 } 3306 } 3307 NFSEXITCODE(error); 3308 return (error); 3309 } 3310 3311 /* 3312 * nfsd_fhtovp() - convert a fh to a vnode ptr 3313 * - look up fsid in mount list (if not found ret error) 3314 * - get vp and export rights by calling nfsvno_fhtovp() 3315 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 3316 * for AUTH_SYS 3317 * - if mpp != NULL, return the mount point so that it can 3318 * be used for vn_finished_write() by the caller 3319 */ 3320 void 3321 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 3322 struct vnode **vpp, struct nfsexstuff *exp, 3323 struct mount **mpp, int startwrite, int nextop) 3324 { 3325 struct mount *mp, *mpw; 3326 struct ucred *credanon; 3327 fhandle_t *fhp; 3328 int error; 3329 3330 if (mpp != NULL) 3331 *mpp = NULL; 3332 *vpp = NULL; 3333 fhp = (fhandle_t *)nfp->nfsrvfh_data; 3334 mp = vfs_busyfs(&fhp->fh_fsid); 3335 if (mp == NULL) { 3336 nd->nd_repstat = ESTALE; 3337 goto out; 3338 } 3339 3340 if (startwrite) { 3341 mpw = mp; 3342 error = vn_start_write(NULL, &mpw, V_WAIT); 3343 if (error != 0) { 3344 mpw = NULL; 3345 vfs_unbusy(mp); 3346 nd->nd_repstat = ESTALE; 3347 goto out; 3348 } 3349 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) 3350 lktype = LK_EXCLUSIVE; 3351 } else 3352 mpw = NULL; 3353 3354 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 3355 &credanon); 3356 vfs_unbusy(mp); 3357 3358 /* 3359 * For NFSv4 without a pseudo root fs, unexported file handles 3360 * can be returned, so that Lookup works everywhere. 3361 */ 3362 if (!nd->nd_repstat && exp->nes_exflag == 0 && 3363 !(nd->nd_flag & ND_NFSV4)) { 3364 vput(*vpp); 3365 *vpp = NULL; 3366 nd->nd_repstat = EACCES; 3367 } 3368 3369 /* 3370 * Personally, I've never seen any point in requiring a 3371 * reserved port#, since only in the rare case where the 3372 * clients are all boxes with secure system privileges, 3373 * does it provide any enhanced security, but... some people 3374 * believe it to be useful and keep putting this code back in. 3375 * (There is also some "security checker" out there that 3376 * complains if the nfs server doesn't enforce this.) 3377 * However, note the following: 3378 * RFC3530 (NFSv4) specifies that a reserved port# not be 3379 * required. 3380 * RFC2623 recommends that, if a reserved port# is checked for, 3381 * that there be a way to turn that off--> ifdef'd. 3382 */ 3383 #ifdef NFS_REQRSVPORT 3384 if (!nd->nd_repstat) { 3385 struct sockaddr_in *saddr; 3386 struct sockaddr_in6 *saddr6; 3387 3388 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 3389 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 3390 if (!(nd->nd_flag & ND_NFSV4) && 3391 ((saddr->sin_family == AF_INET && 3392 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 3393 (saddr6->sin6_family == AF_INET6 && 3394 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 3395 vput(*vpp); 3396 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3397 } 3398 } 3399 #endif /* NFS_REQRSVPORT */ 3400 3401 /* 3402 * Check/setup credentials. 3403 */ 3404 if (!nd->nd_repstat) { 3405 nd->nd_saveduid = nd->nd_cred->cr_uid; 3406 nd->nd_repstat = nfsd_excred(nd, exp, credanon, 3407 nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type)); 3408 if (nd->nd_repstat) 3409 vput(*vpp); 3410 } 3411 if (credanon != NULL) 3412 crfree(credanon); 3413 if (nd->nd_repstat) { 3414 vn_finished_write(mpw); 3415 *vpp = NULL; 3416 } else if (mpp != NULL) { 3417 *mpp = mpw; 3418 } 3419 3420 out: 3421 NFSEXITCODE2(0, nd); 3422 } 3423 3424 /* 3425 * glue for fp. 3426 */ 3427 static int 3428 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 3429 { 3430 struct filedesc *fdp; 3431 struct file *fp; 3432 int error = 0; 3433 3434 fdp = p->td_proc->p_fd; 3435 if (fd < 0 || fd >= fdp->fd_nfiles || 3436 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { 3437 error = EBADF; 3438 goto out; 3439 } 3440 *fpp = fp; 3441 3442 out: 3443 NFSEXITCODE(error); 3444 return (error); 3445 } 3446 3447 /* 3448 * Called from nfssvc() to update the exports list. Just call 3449 * vfs_export(). This has to be done, since the v4 root fake fs isn't 3450 * in the mount list. 3451 */ 3452 int 3453 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 3454 { 3455 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 3456 int error = 0; 3457 struct nameidata nd; 3458 fhandle_t fh; 3459 3460 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export); 3461 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 3462 nfs_rootfhset = 0; 3463 else if (error == 0) { 3464 if (nfsexargp->fspec == NULL) { 3465 error = EPERM; 3466 goto out; 3467 } 3468 /* 3469 * If fspec != NULL, this is the v4root path. 3470 */ 3471 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, 3472 nfsexargp->fspec, p); 3473 if ((error = namei(&nd)) != 0) 3474 goto out; 3475 error = nfsvno_getfh(nd.ni_vp, &fh, p); 3476 vrele(nd.ni_vp); 3477 if (!error) { 3478 nfs_rootfh.nfsrvfh_len = NFSX_MYFH; 3479 NFSBCOPY((caddr_t)&fh, 3480 nfs_rootfh.nfsrvfh_data, 3481 sizeof (fhandle_t)); 3482 nfs_rootfhset = 1; 3483 } 3484 } 3485 3486 out: 3487 NFSEXITCODE(error); 3488 return (error); 3489 } 3490 3491 /* 3492 * This function needs to test to see if the system is near its limit 3493 * for memory allocation via malloc() or mget() and return True iff 3494 * either of these resources are near their limit. 3495 * XXX (For now, this is just a stub.) 3496 */ 3497 int nfsrv_testmalloclimit = 0; 3498 int 3499 nfsrv_mallocmget_limit(void) 3500 { 3501 static int printmesg = 0; 3502 static int testval = 1; 3503 3504 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 3505 if ((printmesg++ % 100) == 0) 3506 printf("nfsd: malloc/mget near limit\n"); 3507 return (1); 3508 } 3509 return (0); 3510 } 3511 3512 /* 3513 * BSD specific initialization of a mount point. 3514 */ 3515 void 3516 nfsd_mntinit(void) 3517 { 3518 static int inited = 0; 3519 3520 if (inited) 3521 return; 3522 inited = 1; 3523 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 3524 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist); 3525 TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist); 3526 nfsv4root_mnt.mnt_export = NULL; 3527 TAILQ_INIT(&nfsv4root_opt); 3528 TAILQ_INIT(&nfsv4root_newopt); 3529 nfsv4root_mnt.mnt_opt = &nfsv4root_opt; 3530 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt; 3531 nfsv4root_mnt.mnt_nvnodelistsize = 0; 3532 nfsv4root_mnt.mnt_lazyvnodelistsize = 0; 3533 } 3534 3535 /* 3536 * Get a vnode for a file handle, without checking exports, etc. 3537 */ 3538 struct vnode * 3539 nfsvno_getvp(fhandle_t *fhp) 3540 { 3541 struct mount *mp; 3542 struct vnode *vp; 3543 int error; 3544 3545 mp = vfs_busyfs(&fhp->fh_fsid); 3546 if (mp == NULL) 3547 return (NULL); 3548 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 3549 vfs_unbusy(mp); 3550 if (error) 3551 return (NULL); 3552 return (vp); 3553 } 3554 3555 /* 3556 * Do a local VOP_ADVLOCK(). 3557 */ 3558 int 3559 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 3560 u_int64_t end, struct thread *td) 3561 { 3562 int error = 0; 3563 struct flock fl; 3564 u_int64_t tlen; 3565 3566 if (nfsrv_dolocallocks == 0) 3567 goto out; 3568 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); 3569 3570 fl.l_whence = SEEK_SET; 3571 fl.l_type = ftype; 3572 fl.l_start = (off_t)first; 3573 if (end == NFS64BITSSET) { 3574 fl.l_len = 0; 3575 } else { 3576 tlen = end - first; 3577 fl.l_len = (off_t)tlen; 3578 } 3579 /* 3580 * For FreeBSD8, the l_pid and l_sysid must be set to the same 3581 * values for all calls, so that all locks will be held by the 3582 * nfsd server. (The nfsd server handles conflicts between the 3583 * various clients.) 3584 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 3585 * bytes, so it can't be put in l_sysid. 3586 */ 3587 if (nfsv4_sysid == 0) 3588 nfsv4_sysid = nlm_acquire_next_sysid(); 3589 fl.l_pid = (pid_t)0; 3590 fl.l_sysid = (int)nfsv4_sysid; 3591 3592 if (ftype == F_UNLCK) 3593 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 3594 (F_POSIX | F_REMOTE)); 3595 else 3596 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 3597 (F_POSIX | F_REMOTE)); 3598 3599 out: 3600 NFSEXITCODE(error); 3601 return (error); 3602 } 3603 3604 /* 3605 * Check the nfsv4 root exports. 3606 */ 3607 int 3608 nfsvno_v4rootexport(struct nfsrv_descript *nd) 3609 { 3610 struct ucred *credanon; 3611 int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; 3612 uint64_t exflags; 3613 3614 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags, 3615 &credanon, &numsecflavor, secflavors); 3616 if (error) { 3617 error = NFSERR_PROGUNAVAIL; 3618 goto out; 3619 } 3620 if (credanon != NULL) 3621 crfree(credanon); 3622 for (i = 0; i < numsecflavor; i++) { 3623 if (secflavors[i] == AUTH_SYS) 3624 nd->nd_flag |= ND_EXAUTHSYS; 3625 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3626 nd->nd_flag |= ND_EXGSS; 3627 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3628 nd->nd_flag |= ND_EXGSSINTEGRITY; 3629 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3630 nd->nd_flag |= ND_EXGSSPRIVACY; 3631 } 3632 3633 /* And set ND_EXxx flags for TLS. */ 3634 if ((exflags & MNT_EXTLS) != 0) { 3635 nd->nd_flag |= ND_EXTLS; 3636 if ((exflags & MNT_EXTLSCERT) != 0) 3637 nd->nd_flag |= ND_EXTLSCERT; 3638 if ((exflags & MNT_EXTLSCERTUSER) != 0) 3639 nd->nd_flag |= ND_EXTLSCERTUSER; 3640 } 3641 3642 out: 3643 NFSEXITCODE(error); 3644 return (error); 3645 } 3646 3647 /* 3648 * Nfs server pseudo system call for the nfsd's 3649 */ 3650 /* 3651 * MPSAFE 3652 */ 3653 static int 3654 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3655 { 3656 struct file *fp; 3657 struct nfsd_addsock_args sockarg; 3658 struct nfsd_nfsd_args nfsdarg; 3659 struct nfsd_nfsd_oargs onfsdarg; 3660 struct nfsd_pnfsd_args pnfsdarg; 3661 struct vnode *vp, *nvp, *curdvp; 3662 struct pnfsdsfile *pf; 3663 struct nfsdevice *ds, *fds; 3664 cap_rights_t rights; 3665 int buflen, error, ret; 3666 char *buf, *cp, *cp2, *cp3; 3667 char fname[PNFS_FILENAME_LEN + 1]; 3668 3669 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3670 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3671 if (error) 3672 goto out; 3673 /* 3674 * Since we don't know what rights might be required, 3675 * pretend that we need them all. It is better to be too 3676 * careful than too reckless. 3677 */ 3678 error = fget(td, sockarg.sock, 3679 cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); 3680 if (error != 0) 3681 goto out; 3682 if (fp->f_type != DTYPE_SOCKET) { 3683 fdrop(fp, td); 3684 error = EPERM; 3685 goto out; 3686 } 3687 error = nfsrvd_addsock(fp); 3688 fdrop(fp, td); 3689 } else if (uap->flag & NFSSVC_NFSDNFSD) { 3690 if (uap->argp == NULL) { 3691 error = EINVAL; 3692 goto out; 3693 } 3694 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { 3695 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); 3696 if (error == 0) { 3697 nfsdarg.principal = onfsdarg.principal; 3698 nfsdarg.minthreads = onfsdarg.minthreads; 3699 nfsdarg.maxthreads = onfsdarg.maxthreads; 3700 nfsdarg.version = 1; 3701 nfsdarg.addr = NULL; 3702 nfsdarg.addrlen = 0; 3703 nfsdarg.dnshost = NULL; 3704 nfsdarg.dnshostlen = 0; 3705 nfsdarg.dspath = NULL; 3706 nfsdarg.dspathlen = 0; 3707 nfsdarg.mdspath = NULL; 3708 nfsdarg.mdspathlen = 0; 3709 nfsdarg.mirrorcnt = 1; 3710 } 3711 } else 3712 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); 3713 if (error) 3714 goto out; 3715 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && 3716 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && 3717 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && 3718 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && 3719 nfsdarg.mirrorcnt >= 1 && 3720 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && 3721 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && 3722 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { 3723 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" 3724 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, 3725 nfsdarg.dspathlen, nfsdarg.dnshostlen, 3726 nfsdarg.mdspathlen, nfsdarg.mirrorcnt); 3727 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); 3728 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); 3729 if (error != 0) { 3730 free(cp, M_TEMP); 3731 goto out; 3732 } 3733 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ 3734 nfsdarg.addr = cp; 3735 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); 3736 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); 3737 if (error != 0) { 3738 free(nfsdarg.addr, M_TEMP); 3739 free(cp, M_TEMP); 3740 goto out; 3741 } 3742 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ 3743 nfsdarg.dnshost = cp; 3744 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); 3745 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); 3746 if (error != 0) { 3747 free(nfsdarg.addr, M_TEMP); 3748 free(nfsdarg.dnshost, M_TEMP); 3749 free(cp, M_TEMP); 3750 goto out; 3751 } 3752 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ 3753 nfsdarg.dspath = cp; 3754 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); 3755 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); 3756 if (error != 0) { 3757 free(nfsdarg.addr, M_TEMP); 3758 free(nfsdarg.dnshost, M_TEMP); 3759 free(nfsdarg.dspath, M_TEMP); 3760 free(cp, M_TEMP); 3761 goto out; 3762 } 3763 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ 3764 nfsdarg.mdspath = cp; 3765 } else { 3766 nfsdarg.addr = NULL; 3767 nfsdarg.addrlen = 0; 3768 nfsdarg.dnshost = NULL; 3769 nfsdarg.dnshostlen = 0; 3770 nfsdarg.dspath = NULL; 3771 nfsdarg.dspathlen = 0; 3772 nfsdarg.mdspath = NULL; 3773 nfsdarg.mdspathlen = 0; 3774 nfsdarg.mirrorcnt = 1; 3775 } 3776 error = nfsrvd_nfsd(td, &nfsdarg); 3777 free(nfsdarg.addr, M_TEMP); 3778 free(nfsdarg.dnshost, M_TEMP); 3779 free(nfsdarg.dspath, M_TEMP); 3780 free(nfsdarg.mdspath, M_TEMP); 3781 } else if (uap->flag & NFSSVC_PNFSDS) { 3782 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); 3783 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || 3784 pnfsdarg.op == PNFSDOP_FORCEDELDS)) { 3785 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3786 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, 3787 NULL); 3788 if (error == 0) 3789 error = nfsrv_deldsserver(pnfsdarg.op, cp, td); 3790 free(cp, M_TEMP); 3791 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { 3792 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3793 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; 3794 buf = malloc(buflen, M_TEMP, M_WAITOK); 3795 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, 3796 NULL); 3797 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); 3798 if (error == 0 && pnfsdarg.dspath != NULL) { 3799 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3800 error = copyinstr(pnfsdarg.dspath, cp2, 3801 PATH_MAX + 1, NULL); 3802 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", 3803 error); 3804 } else 3805 cp2 = NULL; 3806 if (error == 0 && pnfsdarg.curdspath != NULL) { 3807 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3808 error = copyinstr(pnfsdarg.curdspath, cp3, 3809 PATH_MAX + 1, NULL); 3810 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", 3811 error); 3812 } else 3813 cp3 = NULL; 3814 curdvp = NULL; 3815 fds = NULL; 3816 if (error == 0) 3817 error = nfsrv_mdscopymr(cp, cp2, cp3, buf, 3818 &buflen, fname, td, &vp, &nvp, &pf, &ds, 3819 &fds); 3820 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); 3821 if (error == 0) { 3822 if (pf->dsf_dir >= nfsrv_dsdirsize) { 3823 printf("copymr: dsdir out of range\n"); 3824 pf->dsf_dir = 0; 3825 } 3826 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); 3827 error = nfsrv_copymr(vp, nvp, 3828 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, 3829 (struct pnfsdsfile *)buf, 3830 buflen / sizeof(*pf), td->td_ucred, td); 3831 vput(vp); 3832 vput(nvp); 3833 if (fds != NULL && error == 0) { 3834 curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; 3835 ret = vn_lock(curdvp, LK_EXCLUSIVE); 3836 if (ret == 0) { 3837 nfsrv_dsremove(curdvp, fname, 3838 td->td_ucred, td); 3839 NFSVOPUNLOCK(curdvp); 3840 } 3841 } 3842 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); 3843 } 3844 free(cp, M_TEMP); 3845 free(cp2, M_TEMP); 3846 free(cp3, M_TEMP); 3847 free(buf, M_TEMP); 3848 } 3849 } else { 3850 error = nfssvc_srvcall(td, uap, td->td_ucred); 3851 } 3852 3853 out: 3854 NFSEXITCODE(error); 3855 return (error); 3856 } 3857 3858 static int 3859 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 3860 { 3861 struct nfsex_args export; 3862 struct nfsex_oldargs oexp; 3863 struct file *fp = NULL; 3864 int stablefd, i, len; 3865 struct nfsd_clid adminrevoke; 3866 struct nfsd_dumplist dumplist; 3867 struct nfsd_dumpclients *dumpclients; 3868 struct nfsd_dumplocklist dumplocklist; 3869 struct nfsd_dumplocks *dumplocks; 3870 struct nameidata nd; 3871 vnode_t vp; 3872 int error = EINVAL, igotlock; 3873 struct proc *procp; 3874 gid_t *grps; 3875 static int suspend_nfsd = 0; 3876 3877 if (uap->flag & NFSSVC_PUBLICFH) { 3878 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 3879 sizeof (fhandle_t)); 3880 error = copyin(uap->argp, 3881 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 3882 if (!error) 3883 nfs_pubfhset = 1; 3884 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 3885 (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { 3886 error = copyin(uap->argp,(caddr_t)&export, 3887 sizeof (struct nfsex_args)); 3888 if (!error) { 3889 grps = NULL; 3890 if (export.export.ex_ngroups > NGROUPS_MAX || 3891 export.export.ex_ngroups < 0) 3892 error = EINVAL; 3893 else if (export.export.ex_ngroups > 0) { 3894 grps = malloc(export.export.ex_ngroups * 3895 sizeof(gid_t), M_TEMP, M_WAITOK); 3896 error = copyin(export.export.ex_groups, grps, 3897 export.export.ex_ngroups * sizeof(gid_t)); 3898 export.export.ex_groups = grps; 3899 } else 3900 export.export.ex_groups = NULL; 3901 if (!error) 3902 error = nfsrv_v4rootexport(&export, cred, p); 3903 free(grps, M_TEMP); 3904 } 3905 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 3906 NFSSVC_V4ROOTEXPORT) { 3907 error = copyin(uap->argp,(caddr_t)&oexp, 3908 sizeof (struct nfsex_oldargs)); 3909 if (!error) { 3910 memset(&export.export, 0, sizeof(export.export)); 3911 export.export.ex_flags = (uint64_t)oexp.export.ex_flags; 3912 export.export.ex_root = oexp.export.ex_root; 3913 export.export.ex_uid = oexp.export.ex_anon.cr_uid; 3914 export.export.ex_ngroups = 3915 oexp.export.ex_anon.cr_ngroups; 3916 export.export.ex_groups = NULL; 3917 if (export.export.ex_ngroups > XU_NGROUPS || 3918 export.export.ex_ngroups < 0) 3919 error = EINVAL; 3920 else if (export.export.ex_ngroups > 0) { 3921 export.export.ex_groups = malloc( 3922 export.export.ex_ngroups * sizeof(gid_t), 3923 M_TEMP, M_WAITOK); 3924 for (i = 0; i < export.export.ex_ngroups; i++) 3925 export.export.ex_groups[i] = 3926 oexp.export.ex_anon.cr_groups[i]; 3927 } 3928 export.export.ex_addr = oexp.export.ex_addr; 3929 export.export.ex_addrlen = oexp.export.ex_addrlen; 3930 export.export.ex_mask = oexp.export.ex_mask; 3931 export.export.ex_masklen = oexp.export.ex_masklen; 3932 export.export.ex_indexfile = oexp.export.ex_indexfile; 3933 export.export.ex_numsecflavors = 3934 oexp.export.ex_numsecflavors; 3935 if (export.export.ex_numsecflavors >= MAXSECFLAVORS || 3936 export.export.ex_numsecflavors < 0) 3937 error = EINVAL; 3938 else { 3939 for (i = 0; i < export.export.ex_numsecflavors; 3940 i++) 3941 export.export.ex_secflavors[i] = 3942 oexp.export.ex_secflavors[i]; 3943 } 3944 export.fspec = oexp.fspec; 3945 if (error == 0) 3946 error = nfsrv_v4rootexport(&export, cred, p); 3947 free(export.export.ex_groups, M_TEMP); 3948 } 3949 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 3950 nfs_pubfhset = 0; 3951 error = 0; 3952 } else if (uap->flag & NFSSVC_STABLERESTART) { 3953 error = copyin(uap->argp, (caddr_t)&stablefd, 3954 sizeof (int)); 3955 if (!error) 3956 error = fp_getfvp(p, stablefd, &fp, &vp); 3957 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 3958 error = EBADF; 3959 if (!error && newnfs_numnfsd != 0) 3960 error = EPERM; 3961 if (!error) { 3962 nfsrv_stablefirst.nsf_fp = fp; 3963 nfsrv_setupstable(p); 3964 } 3965 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 3966 error = copyin(uap->argp, (caddr_t)&adminrevoke, 3967 sizeof (struct nfsd_clid)); 3968 if (!error) 3969 error = nfsrv_adminrevoke(&adminrevoke, p); 3970 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 3971 error = copyin(uap->argp, (caddr_t)&dumplist, 3972 sizeof (struct nfsd_dumplist)); 3973 if (!error && (dumplist.ndl_size < 1 || 3974 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 3975 error = EPERM; 3976 if (!error) { 3977 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 3978 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 3979 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 3980 error = copyout(dumpclients, dumplist.ndl_list, len); 3981 free(dumpclients, M_TEMP); 3982 } 3983 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 3984 error = copyin(uap->argp, (caddr_t)&dumplocklist, 3985 sizeof (struct nfsd_dumplocklist)); 3986 if (!error && (dumplocklist.ndllck_size < 1 || 3987 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 3988 error = EPERM; 3989 if (!error) 3990 error = nfsrv_lookupfilename(&nd, 3991 dumplocklist.ndllck_fname, p); 3992 if (!error) { 3993 len = sizeof (struct nfsd_dumplocks) * 3994 dumplocklist.ndllck_size; 3995 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 3996 nfsrv_dumplocks(nd.ni_vp, dumplocks, 3997 dumplocklist.ndllck_size, p); 3998 vput(nd.ni_vp); 3999 error = copyout(dumplocks, dumplocklist.ndllck_list, 4000 len); 4001 free(dumplocks, M_TEMP); 4002 } 4003 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 4004 procp = p->td_proc; 4005 PROC_LOCK(procp); 4006 nfsd_master_pid = procp->p_pid; 4007 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 4008 nfsd_master_start = procp->p_stats->p_start; 4009 nfsd_master_proc = procp; 4010 PROC_UNLOCK(procp); 4011 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { 4012 NFSLOCKV4ROOTMUTEX(); 4013 if (suspend_nfsd == 0) { 4014 /* Lock out all nfsd threads */ 4015 do { 4016 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1, 4017 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 4018 } while (igotlock == 0 && suspend_nfsd == 0); 4019 suspend_nfsd = 1; 4020 } 4021 NFSUNLOCKV4ROOTMUTEX(); 4022 error = 0; 4023 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { 4024 NFSLOCKV4ROOTMUTEX(); 4025 if (suspend_nfsd != 0) { 4026 nfsv4_unlock(&nfsd_suspend_lock, 0); 4027 suspend_nfsd = 0; 4028 } 4029 NFSUNLOCKV4ROOTMUTEX(); 4030 error = 0; 4031 } 4032 4033 NFSEXITCODE(error); 4034 return (error); 4035 } 4036 4037 /* 4038 * Check exports. 4039 * Returns 0 if ok, 1 otherwise. 4040 */ 4041 int 4042 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 4043 { 4044 int i; 4045 4046 /* 4047 * Allow NFSv3 Fsinfo per RFC2623. 4048 */ 4049 if (((nd->nd_flag & ND_NFSV4) != 0 || 4050 nd->nd_procnum != NFSPROC_FSINFO) && 4051 ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) || 4052 (NFSVNO_EXTLSCERT(exp) && 4053 (nd->nd_flag & ND_TLSCERT) == 0) || 4054 (NFSVNO_EXTLSCERTUSER(exp) && 4055 (nd->nd_flag & ND_TLSCERTUSER) == 0))) { 4056 if ((nd->nd_flag & ND_NFSV4) != 0) 4057 return (NFSERR_WRONGSEC); 4058 else if ((nd->nd_flag & ND_TLS) == 0) 4059 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS); 4060 else 4061 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST); 4062 } 4063 4064 /* 4065 * This seems odd, but allow the case where the security flavor 4066 * list is empty. This happens when NFSv4 is traversing non-exported 4067 * file systems. Exported file systems should always have a non-empty 4068 * security flavor list. 4069 */ 4070 if (exp->nes_numsecflavor == 0) 4071 return (0); 4072 4073 for (i = 0; i < exp->nes_numsecflavor; i++) { 4074 /* 4075 * The tests for privacy and integrity must be first, 4076 * since ND_GSS is set for everything but AUTH_SYS. 4077 */ 4078 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 4079 (nd->nd_flag & ND_GSSPRIVACY)) 4080 return (0); 4081 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 4082 (nd->nd_flag & ND_GSSINTEGRITY)) 4083 return (0); 4084 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 4085 (nd->nd_flag & ND_GSS)) 4086 return (0); 4087 if (exp->nes_secflavors[i] == AUTH_SYS && 4088 (nd->nd_flag & ND_GSS) == 0) 4089 return (0); 4090 } 4091 if ((nd->nd_flag & ND_NFSV4) != 0) 4092 return (NFSERR_WRONGSEC); 4093 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4094 } 4095 4096 /* 4097 * Calculate a hash value for the fid in a file handle. 4098 */ 4099 uint32_t 4100 nfsrv_hashfh(fhandle_t *fhp) 4101 { 4102 uint32_t hashval; 4103 4104 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 4105 return (hashval); 4106 } 4107 4108 /* 4109 * Calculate a hash value for the sessionid. 4110 */ 4111 uint32_t 4112 nfsrv_hashsessionid(uint8_t *sessionid) 4113 { 4114 uint32_t hashval; 4115 4116 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); 4117 return (hashval); 4118 } 4119 4120 /* 4121 * Signal the userland master nfsd to backup the stable restart file. 4122 */ 4123 void 4124 nfsrv_backupstable(void) 4125 { 4126 struct proc *procp; 4127 4128 if (nfsd_master_proc != NULL) { 4129 procp = pfind(nfsd_master_pid); 4130 /* Try to make sure it is the correct process. */ 4131 if (procp == nfsd_master_proc && 4132 procp->p_stats->p_start.tv_sec == 4133 nfsd_master_start.tv_sec && 4134 procp->p_stats->p_start.tv_usec == 4135 nfsd_master_start.tv_usec && 4136 strcmp(procp->p_comm, nfsd_master_comm) == 0) 4137 kern_psignal(procp, SIGUSR2); 4138 else 4139 nfsd_master_proc = NULL; 4140 4141 if (procp != NULL) 4142 PROC_UNLOCK(procp); 4143 } 4144 } 4145 4146 /* 4147 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. 4148 * The arguments are in a structure, so that they can be passed through 4149 * taskqueue for a kernel process to execute this function. 4150 */ 4151 struct nfsrvdscreate { 4152 int done; 4153 int inprog; 4154 struct task tsk; 4155 struct ucred *tcred; 4156 struct vnode *dvp; 4157 NFSPROC_T *p; 4158 struct pnfsdsfile *pf; 4159 int err; 4160 fhandle_t fh; 4161 struct vattr va; 4162 struct vattr createva; 4163 }; 4164 4165 int 4166 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, 4167 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, 4168 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) 4169 { 4170 struct vnode *nvp; 4171 struct nameidata named; 4172 struct vattr va; 4173 char *bufp; 4174 u_long *hashp; 4175 struct nfsnode *np; 4176 struct nfsmount *nmp; 4177 int error; 4178 4179 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, 4180 LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE); 4181 nfsvno_setpathbuf(&named, &bufp, &hashp); 4182 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 4183 named.ni_cnd.cn_nameptr = bufp; 4184 if (fnamep != NULL) { 4185 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); 4186 named.ni_cnd.cn_namelen = strlen(bufp); 4187 } else 4188 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); 4189 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); 4190 4191 /* Create the date file in the DS mount. */ 4192 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4193 if (error == 0) { 4194 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); 4195 vref(dvp); 4196 VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false); 4197 if (error == 0) { 4198 /* Set the ownership of the file. */ 4199 error = VOP_SETATTR(nvp, nvap, tcred); 4200 NFSD_DEBUG(4, "nfsrv_dscreate:" 4201 " setattr-uid=%d\n", error); 4202 if (error != 0) 4203 vput(nvp); 4204 } 4205 if (error != 0) 4206 printf("pNFS: pnfscreate failed=%d\n", error); 4207 } else 4208 printf("pNFS: pnfscreate vnlock=%d\n", error); 4209 if (error == 0) { 4210 np = VTONFS(nvp); 4211 nmp = VFSTONFS(nvp->v_mount); 4212 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") 4213 != 0 || nmp->nm_nam->sa_len > sizeof( 4214 struct sockaddr_in6) || 4215 np->n_fhp->nfh_len != NFSX_MYFH) { 4216 printf("Bad DS file: fstype=%s salen=%d" 4217 " fhlen=%d\n", 4218 nvp->v_mount->mnt_vfc->vfc_name, 4219 nmp->nm_nam->sa_len, np->n_fhp->nfh_len); 4220 error = ENOENT; 4221 } 4222 4223 /* Set extattrs for the DS on the MDS file. */ 4224 if (error == 0) { 4225 if (dsa != NULL) { 4226 error = VOP_GETATTR(nvp, &va, tcred); 4227 if (error == 0) { 4228 dsa->dsa_filerev = va.va_filerev; 4229 dsa->dsa_size = va.va_size; 4230 dsa->dsa_atime = va.va_atime; 4231 dsa->dsa_mtime = va.va_mtime; 4232 dsa->dsa_bytes = va.va_bytes; 4233 } 4234 } 4235 if (error == 0) { 4236 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, 4237 NFSX_MYFH); 4238 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, 4239 nmp->nm_nam->sa_len); 4240 NFSBCOPY(named.ni_cnd.cn_nameptr, 4241 pf->dsf_filename, 4242 sizeof(pf->dsf_filename)); 4243 } 4244 } else 4245 printf("pNFS: pnfscreate can't get DS" 4246 " attr=%d\n", error); 4247 if (nvpp != NULL && error == 0) 4248 *nvpp = nvp; 4249 else 4250 vput(nvp); 4251 } 4252 nfsvno_relpathbuf(&named); 4253 return (error); 4254 } 4255 4256 /* 4257 * Start up the thread that will execute nfsrv_dscreate(). 4258 */ 4259 static void 4260 start_dscreate(void *arg, int pending) 4261 { 4262 struct nfsrvdscreate *dsc; 4263 4264 dsc = (struct nfsrvdscreate *)arg; 4265 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, 4266 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); 4267 dsc->done = 1; 4268 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); 4269 } 4270 4271 /* 4272 * Create a pNFS data file on the Data Server(s). 4273 */ 4274 static void 4275 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, 4276 NFSPROC_T *p) 4277 { 4278 struct nfsrvdscreate *dsc, *tdsc = NULL; 4279 struct nfsdevice *ds, *tds, *fds; 4280 struct mount *mp; 4281 struct pnfsdsfile *pf, *tpf; 4282 struct pnfsdsattr dsattr; 4283 struct vattr va; 4284 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4285 struct nfsmount *nmp; 4286 fhandle_t fh; 4287 uid_t vauid; 4288 gid_t vagid; 4289 u_short vamode; 4290 struct ucred *tcred; 4291 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; 4292 int failpos, timo; 4293 4294 /* Get a DS server directory in a round-robin order. */ 4295 mirrorcnt = 1; 4296 mp = vp->v_mount; 4297 ds = fds = NULL; 4298 NFSDDSLOCK(); 4299 /* 4300 * Search for the first entry that handles this MDS fs, but use the 4301 * first entry for all MDS fs's otherwise. 4302 */ 4303 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { 4304 if (tds->nfsdev_nmp != NULL) { 4305 if (tds->nfsdev_mdsisset == 0 && ds == NULL) 4306 ds = tds; 4307 else if (tds->nfsdev_mdsisset != 0 && fsidcmp( 4308 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { 4309 ds = fds = tds; 4310 break; 4311 } 4312 } 4313 } 4314 if (ds == NULL) { 4315 NFSDDSUNLOCK(); 4316 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); 4317 return; 4318 } 4319 i = dsdir[0] = ds->nfsdev_nextdir; 4320 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; 4321 dvp[0] = ds->nfsdev_dsdir[i]; 4322 tds = TAILQ_NEXT(ds, nfsdev_list); 4323 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { 4324 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { 4325 if (tds->nfsdev_nmp != NULL && 4326 ((tds->nfsdev_mdsisset == 0 && fds == NULL) || 4327 (tds->nfsdev_mdsisset != 0 && fds != NULL && 4328 fsidcmp(&mp->mnt_stat.f_fsid, 4329 &tds->nfsdev_mdsfsid) == 0))) { 4330 dsdir[mirrorcnt] = i; 4331 dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; 4332 mirrorcnt++; 4333 if (mirrorcnt >= nfsrv_maxpnfsmirror) 4334 break; 4335 } 4336 } 4337 } 4338 /* Put at end of list to implement round-robin usage. */ 4339 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 4340 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 4341 NFSDDSUNLOCK(); 4342 dsc = NULL; 4343 if (mirrorcnt > 1) 4344 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, 4345 M_WAITOK | M_ZERO); 4346 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | 4347 M_ZERO); 4348 4349 error = nfsvno_getfh(vp, &fh, p); 4350 if (error == 0) 4351 error = VOP_GETATTR(vp, &va, cred); 4352 if (error == 0) { 4353 /* Set the attributes for "vp" to Setattr the DS vp. */ 4354 vauid = va.va_uid; 4355 vagid = va.va_gid; 4356 vamode = va.va_mode; 4357 VATTR_NULL(&va); 4358 va.va_uid = vauid; 4359 va.va_gid = vagid; 4360 va.va_mode = vamode; 4361 va.va_size = 0; 4362 } else 4363 printf("pNFS: pnfscreate getfh+attr=%d\n", error); 4364 4365 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, 4366 cred->cr_gid); 4367 /* Make data file name based on FH. */ 4368 tcred = newnfs_getcred(); 4369 4370 /* 4371 * Create the file on each DS mirror, using kernel process(es) for the 4372 * additional mirrors. 4373 */ 4374 failpos = -1; 4375 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { 4376 tpf->dsf_dir = dsdir[i]; 4377 tdsc->tcred = tcred; 4378 tdsc->p = p; 4379 tdsc->pf = tpf; 4380 tdsc->createva = *vap; 4381 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); 4382 tdsc->va = va; 4383 tdsc->dvp = dvp[i]; 4384 tdsc->done = 0; 4385 tdsc->inprog = 0; 4386 tdsc->err = 0; 4387 ret = EIO; 4388 if (nfs_pnfsiothreads != 0) { 4389 ret = nfs_pnfsio(start_dscreate, tdsc); 4390 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); 4391 } 4392 if (ret != 0) { 4393 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, 4394 NULL, tcred, p, NULL); 4395 if (ret != 0) { 4396 KASSERT(error == 0, ("nfsrv_dscreate err=%d", 4397 error)); 4398 if (failpos == -1 && nfsds_failerr(ret)) 4399 failpos = i; 4400 else 4401 error = ret; 4402 } 4403 } 4404 } 4405 if (error == 0) { 4406 tpf->dsf_dir = dsdir[mirrorcnt - 1]; 4407 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, 4408 &dsattr, NULL, tcred, p, NULL); 4409 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { 4410 failpos = mirrorcnt - 1; 4411 error = 0; 4412 } 4413 } 4414 timo = hz / 50; /* Wait for 20msec. */ 4415 if (timo < 1) 4416 timo = 1; 4417 /* Wait for kernel task(s) to complete. */ 4418 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { 4419 while (tdsc->inprog != 0 && tdsc->done == 0) 4420 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); 4421 if (tdsc->err != 0) { 4422 if (failpos == -1 && nfsds_failerr(tdsc->err)) 4423 failpos = i; 4424 else if (error == 0) 4425 error = tdsc->err; 4426 } 4427 } 4428 4429 /* 4430 * If failpos has been set, that mirror has failed, so it needs 4431 * to be disabled. 4432 */ 4433 if (failpos >= 0) { 4434 nmp = VFSTONFS(dvp[failpos]->v_mount); 4435 NFSLOCKMNT(nmp); 4436 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4437 NFSMNTP_CANCELRPCS)) == 0) { 4438 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4439 NFSUNLOCKMNT(nmp); 4440 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4441 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, 4442 ds); 4443 if (ds != NULL) 4444 nfsrv_killrpcs(nmp); 4445 NFSLOCKMNT(nmp); 4446 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4447 wakeup(nmp); 4448 } 4449 NFSUNLOCKMNT(nmp); 4450 } 4451 4452 NFSFREECRED(tcred); 4453 if (error == 0) { 4454 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); 4455 4456 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", 4457 mirrorcnt, nfsrv_maxpnfsmirror); 4458 /* 4459 * For all mirrors that couldn't be created, fill in the 4460 * *pf structure, but with an IP address == 0.0.0.0. 4461 */ 4462 tpf = pf + mirrorcnt; 4463 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { 4464 *tpf = *pf; 4465 tpf->dsf_sin.sin_family = AF_INET; 4466 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); 4467 tpf->dsf_sin.sin_addr.s_addr = 0; 4468 tpf->dsf_sin.sin_port = 0; 4469 } 4470 4471 error = vn_extattr_set(vp, IO_NODELOCKED, 4472 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 4473 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); 4474 if (error == 0) 4475 error = vn_extattr_set(vp, IO_NODELOCKED, 4476 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", 4477 sizeof(dsattr), (char *)&dsattr, p); 4478 if (error != 0) 4479 printf("pNFS: pnfscreate setextattr=%d\n", 4480 error); 4481 } else 4482 printf("pNFS: pnfscreate=%d\n", error); 4483 free(pf, M_TEMP); 4484 free(dsc, M_TEMP); 4485 } 4486 4487 /* 4488 * Get the information needed to remove the pNFS Data Server file from the 4489 * Metadata file. Upon success, ddvp is set non-NULL to the locked 4490 * DS directory vnode. The caller must unlock *ddvp when done with it. 4491 */ 4492 static void 4493 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, 4494 int *mirrorcntp, char *fname, fhandle_t *fhp) 4495 { 4496 struct vattr va; 4497 struct ucred *tcred; 4498 char *buf; 4499 int buflen, error; 4500 4501 dvpp[0] = NULL; 4502 /* If not an exported regular file or not a pNFS server, just return. */ 4503 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4504 nfsrv_devidcnt == 0) 4505 return; 4506 4507 /* Check to see if this is the last hard link. */ 4508 tcred = newnfs_getcred(); 4509 error = VOP_GETATTR(vp, &va, tcred); 4510 NFSFREECRED(tcred); 4511 if (error != 0) { 4512 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); 4513 return; 4514 } 4515 if (va.va_nlink > 1) 4516 return; 4517 4518 error = nfsvno_getfh(vp, fhp, p); 4519 if (error != 0) { 4520 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); 4521 return; 4522 } 4523 4524 buflen = 1024; 4525 buf = malloc(buflen, M_TEMP, M_WAITOK); 4526 /* Get the directory vnode for the DS mount and the file handle. */ 4527 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, 4528 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); 4529 free(buf, M_TEMP); 4530 if (error != 0) 4531 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); 4532 } 4533 4534 /* 4535 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. 4536 * The arguments are in a structure, so that they can be passed through 4537 * taskqueue for a kernel process to execute this function. 4538 */ 4539 struct nfsrvdsremove { 4540 int done; 4541 int inprog; 4542 struct task tsk; 4543 struct ucred *tcred; 4544 struct vnode *dvp; 4545 NFSPROC_T *p; 4546 int err; 4547 char fname[PNFS_FILENAME_LEN + 1]; 4548 }; 4549 4550 static int 4551 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, 4552 NFSPROC_T *p) 4553 { 4554 struct nameidata named; 4555 struct vnode *nvp; 4556 char *bufp; 4557 u_long *hashp; 4558 int error; 4559 4560 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4561 if (error != 0) 4562 return (error); 4563 named.ni_cnd.cn_nameiop = DELETE; 4564 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 4565 named.ni_cnd.cn_cred = tcred; 4566 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; 4567 nfsvno_setpathbuf(&named, &bufp, &hashp); 4568 named.ni_cnd.cn_nameptr = bufp; 4569 named.ni_cnd.cn_namelen = strlen(fname); 4570 strlcpy(bufp, fname, NAME_MAX); 4571 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); 4572 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 4573 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); 4574 if (error == 0) { 4575 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); 4576 vput(nvp); 4577 } 4578 NFSVOPUNLOCK(dvp); 4579 nfsvno_relpathbuf(&named); 4580 if (error != 0) 4581 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); 4582 return (error); 4583 } 4584 4585 /* 4586 * Start up the thread that will execute nfsrv_dsremove(). 4587 */ 4588 static void 4589 start_dsremove(void *arg, int pending) 4590 { 4591 struct nfsrvdsremove *dsrm; 4592 4593 dsrm = (struct nfsrvdsremove *)arg; 4594 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, 4595 dsrm->p); 4596 dsrm->done = 1; 4597 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); 4598 } 4599 4600 /* 4601 * Remove a pNFS data file from a Data Server. 4602 * nfsrv_pnfsremovesetup() must have been called before the MDS file was 4603 * removed to set up the dvp and fill in the FH. 4604 */ 4605 static void 4606 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, 4607 NFSPROC_T *p) 4608 { 4609 struct ucred *tcred; 4610 struct nfsrvdsremove *dsrm, *tdsrm; 4611 struct nfsdevice *ds; 4612 struct nfsmount *nmp; 4613 int failpos, i, ret, timo; 4614 4615 tcred = newnfs_getcred(); 4616 dsrm = NULL; 4617 if (mirrorcnt > 1) 4618 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); 4619 /* 4620 * Remove the file on each DS mirror, using kernel process(es) for the 4621 * additional mirrors. 4622 */ 4623 failpos = -1; 4624 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4625 tdsrm->tcred = tcred; 4626 tdsrm->p = p; 4627 tdsrm->dvp = dvp[i]; 4628 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); 4629 tdsrm->inprog = 0; 4630 tdsrm->done = 0; 4631 tdsrm->err = 0; 4632 ret = EIO; 4633 if (nfs_pnfsiothreads != 0) { 4634 ret = nfs_pnfsio(start_dsremove, tdsrm); 4635 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); 4636 } 4637 if (ret != 0) { 4638 ret = nfsrv_dsremove(dvp[i], fname, tcred, p); 4639 if (failpos == -1 && nfsds_failerr(ret)) 4640 failpos = i; 4641 } 4642 } 4643 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); 4644 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) 4645 failpos = mirrorcnt - 1; 4646 timo = hz / 50; /* Wait for 20msec. */ 4647 if (timo < 1) 4648 timo = 1; 4649 /* Wait for kernel task(s) to complete. */ 4650 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4651 while (tdsrm->inprog != 0 && tdsrm->done == 0) 4652 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); 4653 if (failpos == -1 && nfsds_failerr(tdsrm->err)) 4654 failpos = i; 4655 } 4656 4657 /* 4658 * If failpos has been set, that mirror has failed, so it needs 4659 * to be disabled. 4660 */ 4661 if (failpos >= 0) { 4662 nmp = VFSTONFS(dvp[failpos]->v_mount); 4663 NFSLOCKMNT(nmp); 4664 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4665 NFSMNTP_CANCELRPCS)) == 0) { 4666 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4667 NFSUNLOCKMNT(nmp); 4668 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4669 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, 4670 ds); 4671 if (ds != NULL) 4672 nfsrv_killrpcs(nmp); 4673 NFSLOCKMNT(nmp); 4674 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4675 wakeup(nmp); 4676 } 4677 NFSUNLOCKMNT(nmp); 4678 } 4679 4680 /* Get rid all layouts for the file. */ 4681 nfsrv_freefilelayouts(fhp); 4682 4683 NFSFREECRED(tcred); 4684 free(dsrm, M_TEMP); 4685 } 4686 4687 /* 4688 * Generate a file name based on the file handle and put it in *bufp. 4689 * Return the number of bytes generated. 4690 */ 4691 static int 4692 nfsrv_putfhname(fhandle_t *fhp, char *bufp) 4693 { 4694 int i; 4695 uint8_t *cp; 4696 const uint8_t *hexdigits = "0123456789abcdef"; 4697 4698 cp = (uint8_t *)fhp; 4699 for (i = 0; i < sizeof(*fhp); i++) { 4700 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; 4701 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; 4702 } 4703 bufp[2 * i] = '\0'; 4704 return (2 * i); 4705 } 4706 4707 /* 4708 * Update the Metadata file's attributes from the DS file when a Read/Write 4709 * layout is returned. 4710 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN 4711 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. 4712 */ 4713 int 4714 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 4715 { 4716 struct ucred *tcred; 4717 int error; 4718 4719 /* Do this as root so that it won't fail with EACCES. */ 4720 tcred = newnfs_getcred(); 4721 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, 4722 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); 4723 NFSFREECRED(tcred); 4724 return (error); 4725 } 4726 4727 /* 4728 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. 4729 */ 4730 static int 4731 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, 4732 NFSPROC_T *p) 4733 { 4734 int error; 4735 4736 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, 4737 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); 4738 return (error); 4739 } 4740 4741 static int 4742 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 4743 struct thread *p, int ioproc, struct mbuf **mpp, char *cp, 4744 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, 4745 off_t *offp, int content, bool *eofp) 4746 { 4747 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; 4748 fhandle_t fh[NFSDEV_MAXMIRRORS]; 4749 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4750 struct nfsdevice *ds; 4751 struct pnfsdsattr dsattr; 4752 struct opnfsdsattr odsattr; 4753 char *buf; 4754 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; 4755 4756 NFSD_DEBUG(4, "in nfsrv_proxyds\n"); 4757 /* 4758 * If not a regular file, not exported or not a pNFS server, 4759 * just return ENOENT. 4760 */ 4761 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4762 nfsrv_devidcnt == 0) 4763 return (ENOENT); 4764 4765 buflen = 1024; 4766 buf = malloc(buflen, M_TEMP, M_WAITOK); 4767 error = 0; 4768 4769 /* 4770 * For Getattr, get the Change attribute (va_filerev) and size (va_size) 4771 * from the MetaData file's extended attribute. 4772 */ 4773 if (ioproc == NFSPROC_GETATTR) { 4774 error = vn_extattr_get(vp, IO_NODELOCKED, 4775 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, 4776 p); 4777 if (error == 0) { 4778 if (buflen == sizeof(odsattr)) { 4779 NFSBCOPY(buf, &odsattr, buflen); 4780 nap->na_filerev = odsattr.dsa_filerev; 4781 nap->na_size = odsattr.dsa_size; 4782 nap->na_atime = odsattr.dsa_atime; 4783 nap->na_mtime = odsattr.dsa_mtime; 4784 /* 4785 * Fake na_bytes by rounding up na_size. 4786 * Since we don't know the block size, just 4787 * use BLKDEV_IOSIZE. 4788 */ 4789 nap->na_bytes = (odsattr.dsa_size + 4790 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); 4791 } else if (buflen == sizeof(dsattr)) { 4792 NFSBCOPY(buf, &dsattr, buflen); 4793 nap->na_filerev = dsattr.dsa_filerev; 4794 nap->na_size = dsattr.dsa_size; 4795 nap->na_atime = dsattr.dsa_atime; 4796 nap->na_mtime = dsattr.dsa_mtime; 4797 nap->na_bytes = dsattr.dsa_bytes; 4798 } else 4799 error = ENXIO; 4800 } 4801 if (error == 0) { 4802 /* 4803 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() 4804 * returns 0, just return now. nfsrv_checkdsattr() 4805 * returns 0 if there is no Read/Write layout 4806 * plus either an Open/Write_access or Write 4807 * delegation issued to a client for the file. 4808 */ 4809 if (nfsrv_pnfsgetdsattr == 0 || 4810 nfsrv_checkdsattr(vp, p) == 0) { 4811 free(buf, M_TEMP); 4812 return (error); 4813 } 4814 } 4815 4816 /* 4817 * Clear ENOATTR so the code below will attempt to do a 4818 * nfsrv_getattrdsrpc() to get the attributes and (re)create 4819 * the extended attribute. 4820 */ 4821 if (error == ENOATTR) 4822 error = 0; 4823 } 4824 4825 origmircnt = -1; 4826 trycnt = 0; 4827 tryagain: 4828 if (error == 0) { 4829 buflen = 1024; 4830 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == 4831 LK_EXCLUSIVE) 4832 printf("nfsrv_proxyds: Readds vp exclusively locked\n"); 4833 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, 4834 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, 4835 NULL, NULL); 4836 if (error == 0) { 4837 for (i = 0; i < mirrorcnt; i++) 4838 nmp[i] = VFSTONFS(dvp[i]->v_mount); 4839 } else 4840 printf("pNFS: proxy getextattr sockaddr=%d\n", error); 4841 } else 4842 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); 4843 if (error == 0) { 4844 failpos = -1; 4845 if (origmircnt == -1) 4846 origmircnt = mirrorcnt; 4847 /* 4848 * If failpos is set to a mirror#, then that mirror has 4849 * failed and will be disabled. For Read, Getattr and Seek, the 4850 * function only tries one mirror, so if that mirror has 4851 * failed, it will need to be retried. As such, increment 4852 * tryitagain for these cases. 4853 * For Write, Setattr and Setacl, the function tries all 4854 * mirrors and will not return an error for the case where 4855 * one mirror has failed. For these cases, the functioning 4856 * mirror(s) will have been modified, so a retry isn't 4857 * necessary. These functions will set failpos for the 4858 * failed mirror#. 4859 */ 4860 if (ioproc == NFSPROC_READDS) { 4861 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], 4862 mpp, mpp2); 4863 if (nfsds_failerr(error) && mirrorcnt > 1) { 4864 /* 4865 * Setting failpos will cause the mirror 4866 * to be disabled and then a retry of this 4867 * read is required. 4868 */ 4869 failpos = 0; 4870 error = 0; 4871 trycnt++; 4872 } 4873 } else if (ioproc == NFSPROC_WRITEDS) 4874 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, 4875 &nmp[0], mirrorcnt, mpp, cp, &failpos); 4876 else if (ioproc == NFSPROC_SETATTR) 4877 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], 4878 mirrorcnt, nap, &failpos); 4879 else if (ioproc == NFSPROC_SETACL) 4880 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], 4881 mirrorcnt, aclp, &failpos); 4882 else if (ioproc == NFSPROC_SEEKDS) { 4883 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, 4884 p, nmp[0]); 4885 if (nfsds_failerr(error) && mirrorcnt > 1) { 4886 /* 4887 * Setting failpos will cause the mirror 4888 * to be disabled and then a retry of this 4889 * read is required. 4890 */ 4891 failpos = 0; 4892 error = 0; 4893 trycnt++; 4894 } 4895 } else if (ioproc == NFSPROC_ALLOCATE) 4896 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, 4897 &nmp[0], mirrorcnt, &failpos); 4898 else if (ioproc == NFSPROC_DEALLOCATE) 4899 error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p, 4900 vp, &nmp[0], mirrorcnt, &failpos); 4901 else { 4902 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, 4903 vp, nmp[mirrorcnt - 1], nap); 4904 if (nfsds_failerr(error) && mirrorcnt > 1) { 4905 /* 4906 * Setting failpos will cause the mirror 4907 * to be disabled and then a retry of this 4908 * getattr is required. 4909 */ 4910 failpos = mirrorcnt - 1; 4911 error = 0; 4912 trycnt++; 4913 } 4914 } 4915 ds = NULL; 4916 if (failpos >= 0) { 4917 failnmp = nmp[failpos]; 4918 NFSLOCKMNT(failnmp); 4919 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | 4920 NFSMNTP_CANCELRPCS)) == 0) { 4921 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4922 NFSUNLOCKMNT(failnmp); 4923 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, 4924 failnmp, p); 4925 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", 4926 failpos, ds); 4927 if (ds != NULL) 4928 nfsrv_killrpcs(failnmp); 4929 NFSLOCKMNT(failnmp); 4930 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4931 wakeup(failnmp); 4932 } 4933 NFSUNLOCKMNT(failnmp); 4934 } 4935 for (i = 0; i < mirrorcnt; i++) 4936 NFSVOPUNLOCK(dvp[i]); 4937 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, 4938 trycnt); 4939 /* Try the Read/Getattr again if a mirror was deleted. */ 4940 if (ds != NULL && trycnt > 0 && trycnt < origmircnt) 4941 goto tryagain; 4942 } else { 4943 /* Return ENOENT for any Extended Attribute error. */ 4944 error = ENOENT; 4945 } 4946 free(buf, M_TEMP); 4947 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); 4948 return (error); 4949 } 4950 4951 /* 4952 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended 4953 * attribute. 4954 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs 4955 * to be checked. If it points to a NULL nmp, then it returns 4956 * a suitable destination. 4957 * curnmp - If non-NULL, it is the source mount for the copy. 4958 */ 4959 int 4960 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, 4961 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, 4962 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, 4963 struct nfsmount *curnmp, int *ippos, int *dsdirp) 4964 { 4965 struct vnode *dvp, *nvp = NULL, **tdvpp; 4966 struct mount *mp; 4967 struct nfsmount *nmp, *newnmp; 4968 struct sockaddr *sad; 4969 struct sockaddr_in *sin; 4970 struct nfsdevice *ds, *tds, *fndds; 4971 struct pnfsdsfile *pf; 4972 uint32_t dsdir; 4973 int error, fhiszero, fnd, gotone, i, mirrorcnt; 4974 4975 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); 4976 *mirrorcntp = 1; 4977 tdvpp = dvpp; 4978 if (nvpp != NULL) 4979 *nvpp = NULL; 4980 if (dvpp != NULL) 4981 *dvpp = NULL; 4982 if (ippos != NULL) 4983 *ippos = -1; 4984 if (newnmpp != NULL) 4985 newnmp = *newnmpp; 4986 else 4987 newnmp = NULL; 4988 mp = vp->v_mount; 4989 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 4990 "pnfsd.dsfile", buflenp, buf, p); 4991 mirrorcnt = *buflenp / sizeof(*pf); 4992 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || 4993 *buflenp != sizeof(*pf) * mirrorcnt)) 4994 error = ENOATTR; 4995 4996 pf = (struct pnfsdsfile *)buf; 4997 /* If curnmp != NULL, check for a match in the mirror list. */ 4998 if (curnmp != NULL && error == 0) { 4999 fnd = 0; 5000 for (i = 0; i < mirrorcnt; i++, pf++) { 5001 sad = (struct sockaddr *)&pf->dsf_sin; 5002 if (nfsaddr2_match(sad, curnmp->nm_nam)) { 5003 if (ippos != NULL) 5004 *ippos = i; 5005 fnd = 1; 5006 break; 5007 } 5008 } 5009 if (fnd == 0) 5010 error = ENXIO; 5011 } 5012 5013 gotone = 0; 5014 pf = (struct pnfsdsfile *)buf; 5015 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, 5016 error); 5017 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { 5018 fhiszero = 0; 5019 sad = (struct sockaddr *)&pf->dsf_sin; 5020 sin = &pf->dsf_sin; 5021 dsdir = pf->dsf_dir; 5022 if (dsdir >= nfsrv_dsdirsize) { 5023 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); 5024 error = ENOATTR; 5025 } else if (nvpp != NULL && newnmp != NULL && 5026 nfsaddr2_match(sad, newnmp->nm_nam)) 5027 error = EEXIST; 5028 if (error == 0) { 5029 if (ippos != NULL && curnmp == NULL && 5030 sad->sa_family == AF_INET && 5031 sin->sin_addr.s_addr == 0) 5032 *ippos = i; 5033 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) 5034 fhiszero = 1; 5035 /* Use the socket address to find the mount point. */ 5036 fndds = NULL; 5037 NFSDDSLOCK(); 5038 /* Find a match for the IP address. */ 5039 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5040 if (ds->nfsdev_nmp != NULL) { 5041 dvp = ds->nfsdev_dvp; 5042 nmp = VFSTONFS(dvp->v_mount); 5043 if (nmp != ds->nfsdev_nmp) 5044 printf("different2 nmp %p %p\n", 5045 nmp, ds->nfsdev_nmp); 5046 if (nfsaddr2_match(sad, nmp->nm_nam)) { 5047 fndds = ds; 5048 break; 5049 } 5050 } 5051 } 5052 if (fndds != NULL && newnmpp != NULL && 5053 newnmp == NULL) { 5054 /* Search for a place to make a mirror copy. */ 5055 TAILQ_FOREACH(tds, &nfsrv_devidhead, 5056 nfsdev_list) { 5057 if (tds->nfsdev_nmp != NULL && 5058 fndds != tds && 5059 ((tds->nfsdev_mdsisset == 0 && 5060 fndds->nfsdev_mdsisset == 0) || 5061 (tds->nfsdev_mdsisset != 0 && 5062 fndds->nfsdev_mdsisset != 0 && 5063 fsidcmp(&tds->nfsdev_mdsfsid, 5064 &mp->mnt_stat.f_fsid) == 0))) { 5065 *newnmpp = tds->nfsdev_nmp; 5066 break; 5067 } 5068 } 5069 if (tds != NULL) { 5070 /* 5071 * Move this entry to the end of the 5072 * list, so it won't be selected as 5073 * easily the next time. 5074 */ 5075 TAILQ_REMOVE(&nfsrv_devidhead, tds, 5076 nfsdev_list); 5077 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, 5078 nfsdev_list); 5079 } 5080 } 5081 NFSDDSUNLOCK(); 5082 if (fndds != NULL) { 5083 dvp = fndds->nfsdev_dsdir[dsdir]; 5084 if (lktype != 0 || fhiszero != 0 || 5085 (nvpp != NULL && *nvpp == NULL)) { 5086 if (fhiszero != 0) 5087 error = vn_lock(dvp, 5088 LK_EXCLUSIVE); 5089 else if (lktype != 0) 5090 error = vn_lock(dvp, lktype); 5091 else 5092 error = vn_lock(dvp, LK_SHARED); 5093 /* 5094 * If the file handle is all 0's, try to 5095 * do a Lookup against the DS to acquire 5096 * it. 5097 * If dvpp == NULL or the Lookup fails, 5098 * unlock dvp after the call. 5099 */ 5100 if (error == 0 && (fhiszero != 0 || 5101 (nvpp != NULL && *nvpp == NULL))) { 5102 error = nfsrv_pnfslookupds(vp, 5103 dvp, pf, &nvp, p); 5104 if (error == 0) { 5105 if (fhiszero != 0) 5106 nfsrv_pnfssetfh( 5107 vp, pf, 5108 devid, 5109 fnamep, 5110 nvp, p); 5111 if (nvpp != NULL && 5112 *nvpp == NULL) { 5113 *nvpp = nvp; 5114 *dsdirp = dsdir; 5115 } else 5116 vput(nvp); 5117 } 5118 if (error != 0 || lktype == 0) 5119 NFSVOPUNLOCK(dvp); 5120 } 5121 } 5122 if (error == 0) { 5123 gotone++; 5124 NFSD_DEBUG(4, "gotone=%d\n", gotone); 5125 if (devid != NULL) { 5126 NFSBCOPY(fndds->nfsdev_deviceid, 5127 devid, NFSX_V4DEVICEID); 5128 devid += NFSX_V4DEVICEID; 5129 } 5130 if (dvpp != NULL) 5131 *tdvpp++ = dvp; 5132 if (fhp != NULL) 5133 NFSBCOPY(&pf->dsf_fh, fhp++, 5134 NFSX_MYFH); 5135 if (fnamep != NULL && gotone == 1) 5136 strlcpy(fnamep, 5137 pf->dsf_filename, 5138 sizeof(pf->dsf_filename)); 5139 } else 5140 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " 5141 "err=%d\n", error); 5142 } 5143 } 5144 } 5145 if (error == 0 && gotone == 0) 5146 error = ENOENT; 5147 5148 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, 5149 error); 5150 if (error == 0) 5151 *mirrorcntp = gotone; 5152 else { 5153 if (gotone > 0 && dvpp != NULL) { 5154 /* 5155 * If the error didn't occur on the first one and 5156 * dvpp != NULL, the one(s) prior to the failure will 5157 * have locked dvp's that need to be unlocked. 5158 */ 5159 for (i = 0; i < gotone; i++) { 5160 NFSVOPUNLOCK(*dvpp); 5161 *dvpp++ = NULL; 5162 } 5163 } 5164 /* 5165 * If it found the vnode to be copied from before a failure, 5166 * it needs to be vput()'d. 5167 */ 5168 if (nvpp != NULL && *nvpp != NULL) { 5169 vput(*nvpp); 5170 *nvpp = NULL; 5171 } 5172 } 5173 return (error); 5174 } 5175 5176 /* 5177 * Set the extended attribute for the Change attribute. 5178 */ 5179 static int 5180 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 5181 { 5182 struct pnfsdsattr dsattr; 5183 int error; 5184 5185 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); 5186 dsattr.dsa_filerev = nap->na_filerev; 5187 dsattr.dsa_size = nap->na_size; 5188 dsattr.dsa_atime = nap->na_atime; 5189 dsattr.dsa_mtime = nap->na_mtime; 5190 dsattr.dsa_bytes = nap->na_bytes; 5191 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5192 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); 5193 if (error != 0) 5194 printf("pNFS: setextattr=%d\n", error); 5195 return (error); 5196 } 5197 5198 static int 5199 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5200 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) 5201 { 5202 uint32_t *tl; 5203 struct nfsrv_descript *nd; 5204 nfsv4stateid_t st; 5205 struct mbuf *m, *m2; 5206 int error = 0, retlen, tlen, trimlen; 5207 5208 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); 5209 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5210 *mpp = NULL; 5211 /* 5212 * Use a stateid where other is an alternating 01010 pattern and 5213 * seqid is 0xffffffff. This value is not defined as special by 5214 * the RFC and is used by the FreeBSD NFS server to indicate an 5215 * MDS->DS proxy operation. 5216 */ 5217 st.other[0] = 0x55555555; 5218 st.other[1] = 0x55555555; 5219 st.other[2] = 0x55555555; 5220 st.seqid = 0xffffffff; 5221 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5222 NULL, NULL, 0, 0); 5223 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5224 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); 5225 txdr_hyper(off, tl); 5226 *(tl + 2) = txdr_unsigned(len); 5227 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5228 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5229 if (error != 0) { 5230 free(nd, M_TEMP); 5231 return (error); 5232 } 5233 if (nd->nd_repstat == 0) { 5234 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 5235 NFSM_STRSIZ(retlen, len); 5236 if (retlen > 0) { 5237 /* Trim off the pre-data XDR from the mbuf chain. */ 5238 m = nd->nd_mrep; 5239 while (m != NULL && m != nd->nd_md) { 5240 if (m->m_next == nd->nd_md) { 5241 m->m_next = NULL; 5242 m_freem(nd->nd_mrep); 5243 nd->nd_mrep = m = nd->nd_md; 5244 } else 5245 m = m->m_next; 5246 } 5247 if (m == NULL) { 5248 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5249 error = ENOENT; 5250 goto nfsmout; 5251 } 5252 5253 /* 5254 * Now, adjust first mbuf so that any XDR before the 5255 * read data is skipped over. 5256 */ 5257 trimlen = nd->nd_dpos - mtod(m, char *); 5258 if (trimlen > 0) { 5259 m->m_len -= trimlen; 5260 NFSM_DATAP(m, trimlen); 5261 } 5262 5263 /* 5264 * Truncate the mbuf chain at retlen bytes of data, 5265 * plus XDR padding that brings the length up to a 5266 * multiple of 4. 5267 */ 5268 tlen = NFSM_RNDUP(retlen); 5269 do { 5270 if (m->m_len >= tlen) { 5271 m->m_len = tlen; 5272 tlen = 0; 5273 m2 = m->m_next; 5274 m->m_next = NULL; 5275 m_freem(m2); 5276 break; 5277 } 5278 tlen -= m->m_len; 5279 m = m->m_next; 5280 } while (m != NULL); 5281 if (tlen > 0) { 5282 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5283 error = ENOENT; 5284 goto nfsmout; 5285 } 5286 *mpp = nd->nd_mrep; 5287 *mpendp = m; 5288 nd->nd_mrep = NULL; 5289 } 5290 } else 5291 error = nd->nd_repstat; 5292 nfsmout: 5293 /* If nd->nd_mrep is already NULL, this is a no-op. */ 5294 m_freem(nd->nd_mrep); 5295 free(nd, M_TEMP); 5296 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); 5297 return (error); 5298 } 5299 5300 /* 5301 * Do a write RPC on a DS data file, using this structure for the arguments, 5302 * so that this function can be executed by a separate kernel process. 5303 */ 5304 struct nfsrvwritedsdorpc { 5305 int done; 5306 int inprog; 5307 struct task tsk; 5308 fhandle_t fh; 5309 off_t off; 5310 int len; 5311 struct nfsmount *nmp; 5312 struct ucred *cred; 5313 NFSPROC_T *p; 5314 struct mbuf *m; 5315 int err; 5316 }; 5317 5318 static int 5319 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, 5320 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) 5321 { 5322 uint32_t *tl; 5323 struct nfsrv_descript *nd; 5324 nfsattrbit_t attrbits; 5325 nfsv4stateid_t st; 5326 int commit, error, retlen; 5327 5328 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5329 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, 5330 sizeof(fhandle_t), NULL, NULL, 0, 0); 5331 5332 /* 5333 * Use a stateid where other is an alternating 01010 pattern and 5334 * seqid is 0xffffffff. This value is not defined as special by 5335 * the RFC and is used by the FreeBSD NFS server to indicate an 5336 * MDS->DS proxy operation. 5337 */ 5338 st.other[0] = 0x55555555; 5339 st.other[1] = 0x55555555; 5340 st.other[2] = 0x55555555; 5341 st.seqid = 0xffffffff; 5342 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5343 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); 5344 txdr_hyper(off, tl); 5345 tl += 2; 5346 /* 5347 * Do all writes FileSync, since the server doesn't hold onto dirty 5348 * buffers. Since clients should be accessing the DS servers directly 5349 * using the pNFS layouts, this just needs to work correctly as a 5350 * fallback. 5351 */ 5352 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); 5353 *tl = txdr_unsigned(len); 5354 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); 5355 5356 /* Put data in mbuf chain. */ 5357 nd->nd_mb->m_next = m; 5358 5359 /* Set nd_mb and nd_bpos to end of data. */ 5360 while (m->m_next != NULL) 5361 m = m->m_next; 5362 nd->nd_mb = m; 5363 nfsm_set(nd, m->m_len); 5364 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); 5365 5366 /* Do a Getattr for the attributes that change upon writing. */ 5367 NFSZERO_ATTRBIT(&attrbits); 5368 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5369 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5370 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5371 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5372 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5373 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5374 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5375 (void) nfsrv_putattrbit(nd, &attrbits); 5376 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5377 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5378 if (error != 0) { 5379 free(nd, M_TEMP); 5380 return (error); 5381 } 5382 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); 5383 /* Get rid of weak cache consistency data for now. */ 5384 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5385 (ND_NFSV4 | ND_V4WCCATTR)) { 5386 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5387 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5388 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); 5389 if (error != 0) 5390 goto nfsmout; 5391 /* 5392 * Get rid of Op# and status for next op. 5393 */ 5394 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5395 if (*++tl != 0) 5396 nd->nd_flag |= ND_NOMOREDATA; 5397 } 5398 if (nd->nd_repstat == 0) { 5399 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); 5400 retlen = fxdr_unsigned(int, *tl++); 5401 commit = fxdr_unsigned(int, *tl); 5402 if (commit != NFSWRITE_FILESYNC) 5403 error = NFSERR_IO; 5404 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", 5405 retlen, commit, error); 5406 } else 5407 error = nd->nd_repstat; 5408 /* We have no use for the Write Verifier since we use FileSync. */ 5409 5410 /* 5411 * Get the Change, Size, Access Time and Modify Time attributes and set 5412 * on the Metadata file, so its attributes will be what the file's 5413 * would be if it had been written. 5414 */ 5415 if (error == 0) { 5416 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5417 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5418 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5419 } 5420 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); 5421 nfsmout: 5422 m_freem(nd->nd_mrep); 5423 free(nd, M_TEMP); 5424 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); 5425 return (error); 5426 } 5427 5428 /* 5429 * Start up the thread that will execute nfsrv_writedsdorpc(). 5430 */ 5431 static void 5432 start_writedsdorpc(void *arg, int pending) 5433 { 5434 struct nfsrvwritedsdorpc *drpc; 5435 5436 drpc = (struct nfsrvwritedsdorpc *)arg; 5437 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5438 drpc->len, NULL, drpc->m, drpc->cred, drpc->p); 5439 drpc->done = 1; 5440 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); 5441 } 5442 5443 static int 5444 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5445 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5446 struct mbuf **mpp, char *cp, int *failposp) 5447 { 5448 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; 5449 struct nfsvattr na; 5450 struct mbuf *m; 5451 int error, i, offs, ret, timo; 5452 5453 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); 5454 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); 5455 drpc = NULL; 5456 if (mirrorcnt > 1) 5457 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5458 M_WAITOK); 5459 5460 /* Calculate offset in mbuf chain that data starts. */ 5461 offs = cp - mtod(*mpp, char *); 5462 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); 5463 5464 /* 5465 * Do the write RPC for every DS, using a separate kernel process 5466 * for every DS except the last one. 5467 */ 5468 error = 0; 5469 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5470 tdrpc->done = 0; 5471 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5472 tdrpc->off = off; 5473 tdrpc->len = len; 5474 tdrpc->nmp = *nmpp; 5475 tdrpc->cred = cred; 5476 tdrpc->p = p; 5477 tdrpc->inprog = 0; 5478 tdrpc->err = 0; 5479 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5480 ret = EIO; 5481 if (nfs_pnfsiothreads != 0) { 5482 ret = nfs_pnfsio(start_writedsdorpc, tdrpc); 5483 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", 5484 ret); 5485 } 5486 if (ret != 0) { 5487 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, 5488 tdrpc->m, cred, p); 5489 if (nfsds_failerr(ret) && *failposp == -1) 5490 *failposp = i; 5491 else if (error == 0 && ret != 0) 5492 error = ret; 5493 } 5494 nmpp++; 5495 fhp++; 5496 } 5497 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5498 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); 5499 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5500 *failposp = mirrorcnt - 1; 5501 else if (error == 0 && ret != 0) 5502 error = ret; 5503 if (error == 0) 5504 error = nfsrv_setextattr(vp, &na, p); 5505 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); 5506 tdrpc = drpc; 5507 timo = hz / 50; /* Wait for 20msec. */ 5508 if (timo < 1) 5509 timo = 1; 5510 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5511 /* Wait for RPCs on separate threads to complete. */ 5512 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5513 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); 5514 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5515 *failposp = i; 5516 else if (error == 0 && tdrpc->err != 0) 5517 error = tdrpc->err; 5518 } 5519 free(drpc, M_TEMP); 5520 return (error); 5521 } 5522 5523 /* 5524 * Do a allocate RPC on a DS data file, using this structure for the arguments, 5525 * so that this function can be executed by a separate kernel process. 5526 */ 5527 struct nfsrvallocatedsdorpc { 5528 int done; 5529 int inprog; 5530 struct task tsk; 5531 fhandle_t fh; 5532 off_t off; 5533 off_t len; 5534 struct nfsmount *nmp; 5535 struct ucred *cred; 5536 NFSPROC_T *p; 5537 int err; 5538 }; 5539 5540 static int 5541 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5542 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5543 { 5544 uint32_t *tl; 5545 struct nfsrv_descript *nd; 5546 nfsattrbit_t attrbits; 5547 nfsv4stateid_t st; 5548 int error; 5549 5550 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5551 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, 5552 sizeof(fhandle_t), NULL, NULL, 0, 0); 5553 5554 /* 5555 * Use a stateid where other is an alternating 01010 pattern and 5556 * seqid is 0xffffffff. This value is not defined as special by 5557 * the RFC and is used by the FreeBSD NFS server to indicate an 5558 * MDS->DS proxy operation. 5559 */ 5560 st.other[0] = 0x55555555; 5561 st.other[1] = 0x55555555; 5562 st.other[2] = 0x55555555; 5563 st.seqid = 0xffffffff; 5564 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5565 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5566 txdr_hyper(off, tl); tl += 2; 5567 txdr_hyper(len, tl); tl += 2; 5568 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); 5569 5570 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5571 NFSGETATTR_ATTRBIT(&attrbits); 5572 nfsrv_putattrbit(nd, &attrbits); 5573 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5574 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5575 if (error != 0) { 5576 free(nd, M_TEMP); 5577 return (error); 5578 } 5579 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", 5580 nd->nd_repstat); 5581 if (nd->nd_repstat == 0) { 5582 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5583 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5584 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5585 } else 5586 error = nd->nd_repstat; 5587 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); 5588 nfsmout: 5589 m_freem(nd->nd_mrep); 5590 free(nd, M_TEMP); 5591 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); 5592 return (error); 5593 } 5594 5595 /* 5596 * Start up the thread that will execute nfsrv_allocatedsdorpc(). 5597 */ 5598 static void 5599 start_allocatedsdorpc(void *arg, int pending) 5600 { 5601 struct nfsrvallocatedsdorpc *drpc; 5602 5603 drpc = (struct nfsrvallocatedsdorpc *)arg; 5604 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5605 drpc->len, NULL, drpc->cred, drpc->p); 5606 drpc->done = 1; 5607 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); 5608 } 5609 5610 static int 5611 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5612 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5613 int *failposp) 5614 { 5615 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; 5616 struct nfsvattr na; 5617 int error, i, ret, timo; 5618 5619 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); 5620 drpc = NULL; 5621 if (mirrorcnt > 1) 5622 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5623 M_WAITOK); 5624 5625 /* 5626 * Do the allocate RPC for every DS, using a separate kernel process 5627 * for every DS except the last one. 5628 */ 5629 error = 0; 5630 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5631 tdrpc->done = 0; 5632 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5633 tdrpc->off = off; 5634 tdrpc->len = len; 5635 tdrpc->nmp = *nmpp; 5636 tdrpc->cred = cred; 5637 tdrpc->p = p; 5638 tdrpc->inprog = 0; 5639 tdrpc->err = 0; 5640 ret = EIO; 5641 if (nfs_pnfsiothreads != 0) { 5642 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); 5643 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", 5644 ret); 5645 } 5646 if (ret != 0) { 5647 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, 5648 cred, p); 5649 if (nfsds_failerr(ret) && *failposp == -1) 5650 *failposp = i; 5651 else if (error == 0 && ret != 0) 5652 error = ret; 5653 } 5654 nmpp++; 5655 fhp++; 5656 } 5657 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5658 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5659 *failposp = mirrorcnt - 1; 5660 else if (error == 0 && ret != 0) 5661 error = ret; 5662 if (error == 0) 5663 error = nfsrv_setextattr(vp, &na, p); 5664 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); 5665 tdrpc = drpc; 5666 timo = hz / 50; /* Wait for 20msec. */ 5667 if (timo < 1) 5668 timo = 1; 5669 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5670 /* Wait for RPCs on separate threads to complete. */ 5671 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5672 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5673 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5674 *failposp = i; 5675 else if (error == 0 && tdrpc->err != 0) 5676 error = tdrpc->err; 5677 } 5678 free(drpc, M_TEMP); 5679 return (error); 5680 } 5681 5682 /* 5683 * Do a deallocate RPC on a DS data file, using this structure for the 5684 * arguments, so that this function can be executed by a separate kernel 5685 * process. 5686 */ 5687 struct nfsrvdeallocatedsdorpc { 5688 int done; 5689 int inprog; 5690 struct task tsk; 5691 fhandle_t fh; 5692 off_t off; 5693 off_t len; 5694 struct nfsmount *nmp; 5695 struct ucred *cred; 5696 NFSPROC_T *p; 5697 int err; 5698 }; 5699 5700 static int 5701 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5702 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5703 { 5704 uint32_t *tl; 5705 struct nfsrv_descript *nd; 5706 nfsattrbit_t attrbits; 5707 nfsv4stateid_t st; 5708 int error; 5709 5710 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5711 nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp, 5712 sizeof(fhandle_t), NULL, NULL, 0, 0); 5713 5714 /* 5715 * Use a stateid where other is an alternating 01010 pattern and 5716 * seqid is 0xffffffff. This value is not defined as special by 5717 * the RFC and is used by the FreeBSD NFS server to indicate an 5718 * MDS->DS proxy operation. 5719 */ 5720 st.other[0] = 0x55555555; 5721 st.other[1] = 0x55555555; 5722 st.other[2] = 0x55555555; 5723 st.seqid = 0xffffffff; 5724 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5725 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5726 txdr_hyper(off, tl); tl += 2; 5727 txdr_hyper(len, tl); tl += 2; 5728 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len); 5729 5730 /* Do a Getattr for the attributes that change upon writing. */ 5731 NFSZERO_ATTRBIT(&attrbits); 5732 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5733 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5734 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5735 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5736 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5737 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5738 nfsrv_putattrbit(nd, &attrbits); 5739 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5740 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5741 if (error != 0) { 5742 free(nd, M_TEMP); 5743 return (error); 5744 } 5745 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n", 5746 nd->nd_repstat); 5747 /* Get rid of weak cache consistency data for now. */ 5748 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5749 (ND_NFSV4 | ND_V4WCCATTR)) { 5750 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5751 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5752 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error); 5753 if (error != 0) 5754 goto nfsmout; 5755 /* 5756 * Get rid of Op# and status for next op. 5757 */ 5758 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5759 if (*++tl != 0) 5760 nd->nd_flag |= ND_NOMOREDATA; 5761 } 5762 if (nd->nd_repstat == 0) { 5763 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5764 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5765 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5766 } else 5767 error = nd->nd_repstat; 5768 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error); 5769 nfsmout: 5770 m_freem(nd->nd_mrep); 5771 free(nd, M_TEMP); 5772 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error); 5773 return (error); 5774 } 5775 5776 /* 5777 * Start up the thread that will execute nfsrv_deallocatedsdorpc(). 5778 */ 5779 static void 5780 start_deallocatedsdorpc(void *arg, int pending) 5781 { 5782 struct nfsrvdeallocatedsdorpc *drpc; 5783 5784 drpc = (struct nfsrvdeallocatedsdorpc *)arg; 5785 drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5786 drpc->len, NULL, drpc->cred, drpc->p); 5787 drpc->done = 1; 5788 NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err); 5789 } 5790 5791 static int 5792 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5793 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5794 int *failposp) 5795 { 5796 struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL; 5797 struct nfsvattr na; 5798 int error, i, ret, timo; 5799 5800 NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n"); 5801 drpc = NULL; 5802 if (mirrorcnt > 1) 5803 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5804 M_WAITOK); 5805 5806 /* 5807 * Do the deallocate RPC for every DS, using a separate kernel process 5808 * for every DS except the last one. 5809 */ 5810 error = 0; 5811 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5812 tdrpc->done = 0; 5813 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5814 tdrpc->off = off; 5815 tdrpc->len = len; 5816 tdrpc->nmp = *nmpp; 5817 tdrpc->cred = cred; 5818 tdrpc->p = p; 5819 tdrpc->inprog = 0; 5820 tdrpc->err = 0; 5821 ret = EIO; 5822 if (nfs_pnfsiothreads != 0) { 5823 ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc); 5824 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n", 5825 ret); 5826 } 5827 if (ret != 0) { 5828 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, 5829 NULL, cred, p); 5830 if (nfsds_failerr(ret) && *failposp == -1) 5831 *failposp = i; 5832 else if (error == 0 && ret != 0) 5833 error = ret; 5834 } 5835 nmpp++; 5836 fhp++; 5837 } 5838 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5839 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5840 *failposp = mirrorcnt - 1; 5841 else if (error == 0 && ret != 0) 5842 error = ret; 5843 if (error == 0) 5844 error = nfsrv_setextattr(vp, &na, p); 5845 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error); 5846 tdrpc = drpc; 5847 timo = hz / 50; /* Wait for 20msec. */ 5848 if (timo < 1) 5849 timo = 1; 5850 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5851 /* Wait for RPCs on separate threads to complete. */ 5852 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5853 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5854 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5855 *failposp = i; 5856 else if (error == 0 && tdrpc->err != 0) 5857 error = tdrpc->err; 5858 } 5859 free(drpc, M_TEMP); 5860 return (error); 5861 } 5862 5863 static int 5864 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5865 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, 5866 struct nfsvattr *dsnap) 5867 { 5868 uint32_t *tl; 5869 struct nfsrv_descript *nd; 5870 nfsv4stateid_t st; 5871 nfsattrbit_t attrbits; 5872 int error; 5873 5874 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); 5875 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5876 /* 5877 * Use a stateid where other is an alternating 01010 pattern and 5878 * seqid is 0xffffffff. This value is not defined as special by 5879 * the RFC and is used by the FreeBSD NFS server to indicate an 5880 * MDS->DS proxy operation. 5881 */ 5882 st.other[0] = 0x55555555; 5883 st.other[1] = 0x55555555; 5884 st.other[2] = 0x55555555; 5885 st.seqid = 0xffffffff; 5886 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5887 NULL, NULL, 0, 0); 5888 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5889 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); 5890 5891 /* Do a Getattr for the attributes that change due to writing. */ 5892 NFSZERO_ATTRBIT(&attrbits); 5893 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5894 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5895 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5896 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5897 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5898 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5899 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5900 (void) nfsrv_putattrbit(nd, &attrbits); 5901 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5902 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5903 if (error != 0) { 5904 free(nd, M_TEMP); 5905 return (error); 5906 } 5907 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", 5908 nd->nd_repstat); 5909 /* Get rid of weak cache consistency data for now. */ 5910 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5911 (ND_NFSV4 | ND_V4WCCATTR)) { 5912 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5913 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5914 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); 5915 if (error != 0) 5916 goto nfsmout; 5917 /* 5918 * Get rid of Op# and status for next op. 5919 */ 5920 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5921 if (*++tl != 0) 5922 nd->nd_flag |= ND_NOMOREDATA; 5923 } 5924 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 5925 if (error != 0) 5926 goto nfsmout; 5927 if (nd->nd_repstat != 0) 5928 error = nd->nd_repstat; 5929 /* 5930 * Get the Change, Size, Access Time and Modify Time attributes and set 5931 * on the Metadata file, so its attributes will be what the file's 5932 * would be if it had been written. 5933 */ 5934 if (error == 0) { 5935 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5936 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5937 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5938 } 5939 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); 5940 nfsmout: 5941 m_freem(nd->nd_mrep); 5942 free(nd, M_TEMP); 5943 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); 5944 return (error); 5945 } 5946 5947 struct nfsrvsetattrdsdorpc { 5948 int done; 5949 int inprog; 5950 struct task tsk; 5951 fhandle_t fh; 5952 struct nfsmount *nmp; 5953 struct vnode *vp; 5954 struct ucred *cred; 5955 NFSPROC_T *p; 5956 struct nfsvattr na; 5957 struct nfsvattr dsna; 5958 int err; 5959 }; 5960 5961 /* 5962 * Start up the thread that will execute nfsrv_setattrdsdorpc(). 5963 */ 5964 static void 5965 start_setattrdsdorpc(void *arg, int pending) 5966 { 5967 struct nfsrvsetattrdsdorpc *drpc; 5968 5969 drpc = (struct nfsrvsetattrdsdorpc *)arg; 5970 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, 5971 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); 5972 drpc->done = 1; 5973 } 5974 5975 static int 5976 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5977 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5978 struct nfsvattr *nap, int *failposp) 5979 { 5980 struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; 5981 struct nfsvattr na; 5982 int error, i, ret, timo; 5983 5984 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); 5985 drpc = NULL; 5986 if (mirrorcnt > 1) 5987 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5988 M_WAITOK); 5989 5990 /* 5991 * Do the setattr RPC for every DS, using a separate kernel process 5992 * for every DS except the last one. 5993 */ 5994 error = 0; 5995 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5996 tdrpc->done = 0; 5997 tdrpc->inprog = 0; 5998 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5999 tdrpc->nmp = *nmpp; 6000 tdrpc->vp = vp; 6001 tdrpc->cred = cred; 6002 tdrpc->p = p; 6003 tdrpc->na = *nap; 6004 tdrpc->err = 0; 6005 ret = EIO; 6006 if (nfs_pnfsiothreads != 0) { 6007 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); 6008 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", 6009 ret); 6010 } 6011 if (ret != 0) { 6012 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, 6013 &na); 6014 if (nfsds_failerr(ret) && *failposp == -1) 6015 *failposp = i; 6016 else if (error == 0 && ret != 0) 6017 error = ret; 6018 } 6019 nmpp++; 6020 fhp++; 6021 } 6022 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); 6023 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6024 *failposp = mirrorcnt - 1; 6025 else if (error == 0 && ret != 0) 6026 error = ret; 6027 if (error == 0) 6028 error = nfsrv_setextattr(vp, &na, p); 6029 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); 6030 tdrpc = drpc; 6031 timo = hz / 50; /* Wait for 20msec. */ 6032 if (timo < 1) 6033 timo = 1; 6034 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6035 /* Wait for RPCs on separate threads to complete. */ 6036 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6037 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); 6038 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6039 *failposp = i; 6040 else if (error == 0 && tdrpc->err != 0) 6041 error = tdrpc->err; 6042 } 6043 free(drpc, M_TEMP); 6044 return (error); 6045 } 6046 6047 /* 6048 * Do a Setattr of an NFSv4 ACL on the DS file. 6049 */ 6050 static int 6051 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6052 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) 6053 { 6054 struct nfsrv_descript *nd; 6055 nfsv4stateid_t st; 6056 nfsattrbit_t attrbits; 6057 int error; 6058 6059 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); 6060 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6061 /* 6062 * Use a stateid where other is an alternating 01010 pattern and 6063 * seqid is 0xffffffff. This value is not defined as special by 6064 * the RFC and is used by the FreeBSD NFS server to indicate an 6065 * MDS->DS proxy operation. 6066 */ 6067 st.other[0] = 0x55555555; 6068 st.other[1] = 0x55555555; 6069 st.other[2] = 0x55555555; 6070 st.seqid = 0xffffffff; 6071 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), 6072 NULL, NULL, 0, 0); 6073 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6074 NFSZERO_ATTRBIT(&attrbits); 6075 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); 6076 /* 6077 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), 6078 * so passing in the metadata "vp" will be ok, since it is of 6079 * the same type (VREG). 6080 */ 6081 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, 6082 NULL, 0, 0, 0, 0, 0, NULL); 6083 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6084 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6085 if (error != 0) { 6086 free(nd, M_TEMP); 6087 return (error); 6088 } 6089 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", 6090 nd->nd_repstat); 6091 error = nd->nd_repstat; 6092 m_freem(nd->nd_mrep); 6093 free(nd, M_TEMP); 6094 return (error); 6095 } 6096 6097 struct nfsrvsetacldsdorpc { 6098 int done; 6099 int inprog; 6100 struct task tsk; 6101 fhandle_t fh; 6102 struct nfsmount *nmp; 6103 struct vnode *vp; 6104 struct ucred *cred; 6105 NFSPROC_T *p; 6106 struct acl *aclp; 6107 int err; 6108 }; 6109 6110 /* 6111 * Start up the thread that will execute nfsrv_setacldsdorpc(). 6112 */ 6113 static void 6114 start_setacldsdorpc(void *arg, int pending) 6115 { 6116 struct nfsrvsetacldsdorpc *drpc; 6117 6118 drpc = (struct nfsrvsetacldsdorpc *)arg; 6119 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, 6120 drpc->vp, drpc->nmp, drpc->aclp); 6121 drpc->done = 1; 6122 } 6123 6124 static int 6125 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6126 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, 6127 int *failposp) 6128 { 6129 struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; 6130 int error, i, ret, timo; 6131 6132 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); 6133 drpc = NULL; 6134 if (mirrorcnt > 1) 6135 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6136 M_WAITOK); 6137 6138 /* 6139 * Do the setattr RPC for every DS, using a separate kernel process 6140 * for every DS except the last one. 6141 */ 6142 error = 0; 6143 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6144 tdrpc->done = 0; 6145 tdrpc->inprog = 0; 6146 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6147 tdrpc->nmp = *nmpp; 6148 tdrpc->vp = vp; 6149 tdrpc->cred = cred; 6150 tdrpc->p = p; 6151 tdrpc->aclp = aclp; 6152 tdrpc->err = 0; 6153 ret = EIO; 6154 if (nfs_pnfsiothreads != 0) { 6155 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); 6156 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", 6157 ret); 6158 } 6159 if (ret != 0) { 6160 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, 6161 aclp); 6162 if (nfsds_failerr(ret) && *failposp == -1) 6163 *failposp = i; 6164 else if (error == 0 && ret != 0) 6165 error = ret; 6166 } 6167 nmpp++; 6168 fhp++; 6169 } 6170 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); 6171 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6172 *failposp = mirrorcnt - 1; 6173 else if (error == 0 && ret != 0) 6174 error = ret; 6175 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); 6176 tdrpc = drpc; 6177 timo = hz / 50; /* Wait for 20msec. */ 6178 if (timo < 1) 6179 timo = 1; 6180 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6181 /* Wait for RPCs on separate threads to complete. */ 6182 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6183 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); 6184 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6185 *failposp = i; 6186 else if (error == 0 && tdrpc->err != 0) 6187 error = tdrpc->err; 6188 } 6189 free(drpc, M_TEMP); 6190 return (error); 6191 } 6192 6193 /* 6194 * Getattr call to the DS for the attributes that change due to writing. 6195 */ 6196 static int 6197 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6198 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) 6199 { 6200 struct nfsrv_descript *nd; 6201 int error; 6202 nfsattrbit_t attrbits; 6203 6204 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); 6205 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6206 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, 6207 sizeof(fhandle_t), NULL, NULL, 0, 0); 6208 NFSZERO_ATTRBIT(&attrbits); 6209 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6210 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6211 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6212 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6213 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6214 (void) nfsrv_putattrbit(nd, &attrbits); 6215 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6216 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6217 if (error != 0) { 6218 free(nd, M_TEMP); 6219 return (error); 6220 } 6221 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", 6222 nd->nd_repstat); 6223 if (nd->nd_repstat == 0) { 6224 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, 6225 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, 6226 NULL, NULL); 6227 /* 6228 * We can only save the updated values in the extended 6229 * attribute if the vp is exclusively locked. 6230 * This should happen when any of the following operations 6231 * occur on the vnode: 6232 * Close, Delegreturn, LayoutCommit, LayoutReturn 6233 * As such, the updated extended attribute should get saved 6234 * before nfsrv_checkdsattr() returns 0 and allows the cached 6235 * attributes to be returned without calling this function. 6236 */ 6237 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { 6238 error = nfsrv_setextattr(vp, nap, p); 6239 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", 6240 error); 6241 } 6242 } else 6243 error = nd->nd_repstat; 6244 m_freem(nd->nd_mrep); 6245 free(nd, M_TEMP); 6246 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); 6247 return (error); 6248 } 6249 6250 /* 6251 * Seek call to a DS. 6252 */ 6253 static int 6254 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, 6255 struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) 6256 { 6257 uint32_t *tl; 6258 struct nfsrv_descript *nd; 6259 nfsv4stateid_t st; 6260 int error; 6261 6262 NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); 6263 /* 6264 * Use a stateid where other is an alternating 01010 pattern and 6265 * seqid is 0xffffffff. This value is not defined as special by 6266 * the RFC and is used by the FreeBSD NFS server to indicate an 6267 * MDS->DS proxy operation. 6268 */ 6269 st.other[0] = 0x55555555; 6270 st.other[1] = 0x55555555; 6271 st.other[2] = 0x55555555; 6272 st.seqid = 0xffffffff; 6273 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6274 nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, 6275 sizeof(fhandle_t), NULL, NULL, 0, 0); 6276 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6277 NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); 6278 txdr_hyper(*offp, tl); tl += 2; 6279 *tl = txdr_unsigned(content); 6280 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6281 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6282 if (error != 0) { 6283 free(nd, M_TEMP); 6284 return (error); 6285 } 6286 NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); 6287 if (nd->nd_repstat == 0) { 6288 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); 6289 if (*tl++ == newnfs_true) 6290 *eofp = true; 6291 else 6292 *eofp = false; 6293 *offp = fxdr_hyper(tl); 6294 } else 6295 error = nd->nd_repstat; 6296 nfsmout: 6297 m_freem(nd->nd_mrep); 6298 free(nd, M_TEMP); 6299 NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); 6300 return (error); 6301 } 6302 6303 /* 6304 * Get the device id and file handle for a DS file. 6305 */ 6306 int 6307 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, 6308 fhandle_t *fhp, char *devid) 6309 { 6310 int buflen, error; 6311 char *buf; 6312 6313 buflen = 1024; 6314 buf = malloc(buflen, M_TEMP, M_WAITOK); 6315 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, 6316 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); 6317 free(buf, M_TEMP); 6318 return (error); 6319 } 6320 6321 /* 6322 * Do a Lookup against the DS for the filename. 6323 */ 6324 static int 6325 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, 6326 struct vnode **nvpp, NFSPROC_T *p) 6327 { 6328 struct nameidata named; 6329 struct ucred *tcred; 6330 char *bufp; 6331 u_long *hashp; 6332 struct vnode *nvp; 6333 int error; 6334 6335 tcred = newnfs_getcred(); 6336 named.ni_cnd.cn_nameiop = LOOKUP; 6337 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; 6338 named.ni_cnd.cn_cred = tcred; 6339 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME; 6340 nfsvno_setpathbuf(&named, &bufp, &hashp); 6341 named.ni_cnd.cn_nameptr = bufp; 6342 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); 6343 strlcpy(bufp, pf->dsf_filename, NAME_MAX); 6344 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); 6345 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 6346 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); 6347 NFSFREECRED(tcred); 6348 nfsvno_relpathbuf(&named); 6349 if (error == 0) 6350 *nvpp = nvp; 6351 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); 6352 return (error); 6353 } 6354 6355 /* 6356 * Set the file handle to the correct one. 6357 */ 6358 static void 6359 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, 6360 char *fnamep, struct vnode *nvp, NFSPROC_T *p) 6361 { 6362 struct nfsnode *np; 6363 int ret = 0; 6364 6365 np = VTONFS(nvp); 6366 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); 6367 /* 6368 * We can only do a vn_set_extattr() if the vnode is exclusively 6369 * locked and vn_start_write() has been done. If devid != NULL or 6370 * fnamep != NULL or the vnode is shared locked, vn_start_write() 6371 * may not have been done. 6372 * If not done now, it will be done on a future call. 6373 */ 6374 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == 6375 LK_EXCLUSIVE) 6376 ret = vn_extattr_set(vp, IO_NODELOCKED, 6377 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), 6378 (char *)pf, p); 6379 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); 6380 } 6381 6382 /* 6383 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point 6384 * when the DS has failed. 6385 */ 6386 void 6387 nfsrv_killrpcs(struct nfsmount *nmp) 6388 { 6389 6390 /* 6391 * Call newnfs_nmcancelreqs() to cause 6392 * any RPCs in progress on the mount point to 6393 * fail. 6394 * This will cause any process waiting for an 6395 * RPC to complete while holding a vnode lock 6396 * on the mounted-on vnode (such as "df" or 6397 * a non-forced "umount") to fail. 6398 * This will unlock the mounted-on vnode so 6399 * a forced dismount can succeed. 6400 * The NFSMNTP_CANCELRPCS flag should be set when this function is 6401 * called. 6402 */ 6403 newnfs_nmcancelreqs(nmp); 6404 } 6405 6406 /* 6407 * Sum up the statfs info for each of the DSs, so that the client will 6408 * receive the total for all DSs. 6409 */ 6410 static int 6411 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) 6412 { 6413 struct statfs *tsf; 6414 struct nfsdevice *ds; 6415 struct vnode **dvpp, **tdvpp, *dvp; 6416 uint64_t tot; 6417 int cnt, error = 0, i; 6418 6419 if (nfsrv_devidcnt <= 0) 6420 return (ENXIO); 6421 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 6422 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 6423 6424 /* Get an array of the dvps for the DSs. */ 6425 tdvpp = dvpp; 6426 i = 0; 6427 NFSDDSLOCK(); 6428 /* First, search for matches for same file system. */ 6429 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6430 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && 6431 fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { 6432 if (++i > nfsrv_devidcnt) 6433 break; 6434 *tdvpp++ = ds->nfsdev_dvp; 6435 } 6436 } 6437 /* 6438 * If no matches for same file system, total all servers not assigned 6439 * to a file system. 6440 */ 6441 if (i == 0) { 6442 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6443 if (ds->nfsdev_nmp != NULL && 6444 ds->nfsdev_mdsisset == 0) { 6445 if (++i > nfsrv_devidcnt) 6446 break; 6447 *tdvpp++ = ds->nfsdev_dvp; 6448 } 6449 } 6450 } 6451 NFSDDSUNLOCK(); 6452 cnt = i; 6453 6454 /* Do a VFS_STATFS() for each of the DSs and sum them up. */ 6455 tdvpp = dvpp; 6456 for (i = 0; i < cnt && error == 0; i++) { 6457 dvp = *tdvpp++; 6458 error = VFS_STATFS(dvp->v_mount, tsf); 6459 if (error == 0) { 6460 if (sf->f_bsize == 0) { 6461 if (tsf->f_bsize > 0) 6462 sf->f_bsize = tsf->f_bsize; 6463 else 6464 sf->f_bsize = 8192; 6465 } 6466 if (tsf->f_blocks > 0) { 6467 if (sf->f_bsize != tsf->f_bsize) { 6468 tot = tsf->f_blocks * tsf->f_bsize; 6469 sf->f_blocks += (tot / sf->f_bsize); 6470 } else 6471 sf->f_blocks += tsf->f_blocks; 6472 } 6473 if (tsf->f_bfree > 0) { 6474 if (sf->f_bsize != tsf->f_bsize) { 6475 tot = tsf->f_bfree * tsf->f_bsize; 6476 sf->f_bfree += (tot / sf->f_bsize); 6477 } else 6478 sf->f_bfree += tsf->f_bfree; 6479 } 6480 if (tsf->f_bavail > 0) { 6481 if (sf->f_bsize != tsf->f_bsize) { 6482 tot = tsf->f_bavail * tsf->f_bsize; 6483 sf->f_bavail += (tot / sf->f_bsize); 6484 } else 6485 sf->f_bavail += tsf->f_bavail; 6486 } 6487 } 6488 } 6489 free(tsf, M_TEMP); 6490 free(dvpp, M_TEMP); 6491 return (error); 6492 } 6493 6494 /* 6495 * Set an NFSv4 acl. 6496 */ 6497 int 6498 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) 6499 { 6500 int error; 6501 6502 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { 6503 error = NFSERR_ATTRNOTSUPP; 6504 goto out; 6505 } 6506 /* 6507 * With NFSv4 ACLs, chmod(2) may need to add additional entries. 6508 * Make sure it has enough room for that - splitting every entry 6509 * into two and appending "canonical six" entries at the end. 6510 * Cribbed out of kern/vfs_acl.c - Rick M. 6511 */ 6512 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { 6513 error = NFSERR_ATTRNOTSUPP; 6514 goto out; 6515 } 6516 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 6517 if (error == 0) { 6518 error = nfsrv_dssetacl(vp, aclp, cred, p); 6519 if (error == ENOENT) 6520 error = 0; 6521 } 6522 6523 out: 6524 NFSEXITCODE(error); 6525 return (error); 6526 } 6527 6528 /* 6529 * Seek vnode op call (actually it is a VOP_IOCTL()). 6530 * This function is called with the vnode locked, but unlocks and vrele()s 6531 * the vp before returning. 6532 */ 6533 int 6534 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, 6535 off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) 6536 { 6537 struct nfsvattr at; 6538 int error, ret; 6539 6540 ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); 6541 /* 6542 * Attempt to seek on a DS file. A return of ENOENT implies 6543 * there is no DS file to seek on. 6544 */ 6545 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, 6546 NULL, NULL, NULL, NULL, offp, content, eofp); 6547 if (error != ENOENT) { 6548 vput(vp); 6549 return (error); 6550 } 6551 6552 /* 6553 * Do the VOP_IOCTL() call. For the case where *offp == file_size, 6554 * VOP_IOCTL() will return ENXIO. However, the correct reply for 6555 * NFSv4.2 is *eofp == true and error == 0 for this case. 6556 */ 6557 NFSVOPUNLOCK(vp); 6558 error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); 6559 *eofp = false; 6560 if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { 6561 /* Handle the cases where we might be at EOF. */ 6562 ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); 6563 if (ret == 0 && *offp == at.na_size) { 6564 *eofp = true; 6565 error = 0; 6566 } 6567 if (ret != 0 && error == 0) 6568 error = ret; 6569 } 6570 vrele(vp); 6571 NFSEXITCODE(error); 6572 return (error); 6573 } 6574 6575 /* 6576 * Allocate vnode op call. 6577 */ 6578 int 6579 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6580 NFSPROC_T *p) 6581 { 6582 int error; 6583 off_t olen; 6584 6585 ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); 6586 /* 6587 * Attempt to allocate on a DS file. A return of ENOENT implies 6588 * there is no DS file to allocate on. 6589 */ 6590 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, 6591 NULL, NULL, NULL, NULL, &len, 0, NULL); 6592 if (error != ENOENT) 6593 return (error); 6594 6595 /* 6596 * Do the actual VOP_ALLOCATE(), looping so long as 6597 * progress is being made, to achieve completion. 6598 */ 6599 do { 6600 olen = len; 6601 error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred); 6602 if (error == 0 && len > 0 && olen > len) 6603 maybe_yield(); 6604 } while (error == 0 && len > 0 && olen > len); 6605 if (error == 0 && len > 0) 6606 error = NFSERR_IO; 6607 NFSEXITCODE(error); 6608 return (error); 6609 } 6610 6611 /* 6612 * Deallocate vnode op call. 6613 */ 6614 int 6615 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6616 NFSPROC_T *p) 6617 { 6618 int error; 6619 off_t olen; 6620 6621 ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp"); 6622 /* 6623 * Attempt to deallocate on a DS file. A return of ENOENT implies 6624 * there is no DS file to deallocate on. 6625 */ 6626 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL, 6627 NULL, NULL, NULL, NULL, &len, 0, NULL); 6628 if (error != ENOENT) 6629 return (error); 6630 6631 /* 6632 * Do the actual VOP_DEALLOCATE(), looping so long as 6633 * progress is being made, to achieve completion. 6634 */ 6635 do { 6636 olen = len; 6637 error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred); 6638 if (error == 0 && len > 0 && olen > len) 6639 maybe_yield(); 6640 } while (error == 0 && len > 0 && olen > len); 6641 if (error == 0 && len > 0) 6642 error = NFSERR_IO; 6643 NFSEXITCODE(error); 6644 return (error); 6645 } 6646 6647 /* 6648 * Get Extended Atribute vnode op into an mbuf list. 6649 */ 6650 int 6651 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, 6652 struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p, 6653 struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 6654 { 6655 struct iovec *iv; 6656 struct uio io, *uiop = &io; 6657 struct mbuf *m, *m2; 6658 int alen, error, len, tlen; 6659 size_t siz; 6660 6661 /* First, find out the size of the extended attribute. */ 6662 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6663 &siz, cred, p); 6664 if (error != 0) 6665 return (NFSERR_NOXATTR); 6666 if (siz > maxresp - NFS_MAXXDR) 6667 return (NFSERR_XATTR2BIG); 6668 len = siz; 6669 tlen = NFSM_RNDUP(len); 6670 if (tlen > 0) { 6671 /* 6672 * If cnt > MCLBYTES and the reply will not be saved, use 6673 * ext_pgs mbufs for TLS. 6674 * For NFSv4.0, we do not know for sure if the reply will 6675 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 6676 * Always use ext_pgs mbufs if ND_EXTPG is set. 6677 */ 6678 if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES && 6679 (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS && 6680 (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)) 6681 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen, 6682 maxextsiz, &m, &m2, &iv); 6683 else 6684 uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, 6685 &iv); 6686 uiop->uio_iov = iv; 6687 } else { 6688 uiop->uio_iovcnt = 0; 6689 uiop->uio_iov = iv = NULL; 6690 m = m2 = NULL; 6691 } 6692 uiop->uio_offset = 0; 6693 uiop->uio_resid = tlen; 6694 uiop->uio_rw = UIO_READ; 6695 uiop->uio_segflg = UIO_SYSSPACE; 6696 uiop->uio_td = p; 6697 #ifdef MAC 6698 error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6699 name); 6700 if (error != 0) 6701 goto out; 6702 #endif 6703 6704 if (tlen > 0) 6705 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 6706 NULL, cred, p); 6707 if (error != 0) 6708 goto out; 6709 if (uiop->uio_resid > 0) { 6710 alen = tlen; 6711 len = tlen - uiop->uio_resid; 6712 tlen = NFSM_RNDUP(len); 6713 if (alen != tlen) 6714 printf("nfsvno_getxattr: weird size read\n"); 6715 if (tlen == 0) { 6716 m_freem(m); 6717 m = m2 = NULL; 6718 } else if (alen != tlen || tlen != len) 6719 m2 = nfsrv_adj(m, alen - tlen, tlen - len); 6720 } 6721 *lenp = len; 6722 *mpp = m; 6723 *mpendp = m2; 6724 6725 out: 6726 if (error != 0) { 6727 if (m != NULL) 6728 m_freem(m); 6729 *lenp = 0; 6730 } 6731 free(iv, M_TEMP); 6732 NFSEXITCODE(error); 6733 return (error); 6734 } 6735 6736 /* 6737 * Set Extended attribute vnode op from an mbuf list. 6738 */ 6739 int 6740 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, 6741 char *cp, struct ucred *cred, struct thread *p) 6742 { 6743 struct iovec *iv; 6744 struct uio uio, *uiop = &uio; 6745 int cnt, error; 6746 6747 error = 0; 6748 #ifdef MAC 6749 error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6750 name); 6751 #endif 6752 if (error != 0) 6753 goto out; 6754 6755 uiop->uio_rw = UIO_WRITE; 6756 uiop->uio_segflg = UIO_SYSSPACE; 6757 uiop->uio_td = p; 6758 uiop->uio_offset = 0; 6759 uiop->uio_resid = len; 6760 if (len > 0) { 6761 error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); 6762 uiop->uio_iov = iv; 6763 uiop->uio_iovcnt = cnt; 6764 } else { 6765 uiop->uio_iov = iv = NULL; 6766 uiop->uio_iovcnt = 0; 6767 } 6768 if (error == 0) { 6769 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 6770 cred, p); 6771 free(iv, M_TEMP); 6772 } 6773 6774 out: 6775 NFSEXITCODE(error); 6776 return (error); 6777 } 6778 6779 /* 6780 * Remove Extended attribute vnode op. 6781 */ 6782 int 6783 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, 6784 struct ucred *cred, struct thread *p) 6785 { 6786 int error; 6787 6788 /* 6789 * Get rid of any delegations. I am not sure why this is required, 6790 * but RFC-8276 says so. 6791 */ 6792 error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); 6793 if (error != 0) 6794 goto out; 6795 #ifdef MAC 6796 error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6797 name); 6798 if (error != 0) 6799 goto out; 6800 #endif 6801 6802 error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); 6803 if (error == EOPNOTSUPP) 6804 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6805 cred, p); 6806 out: 6807 NFSEXITCODE(error); 6808 return (error); 6809 } 6810 6811 /* 6812 * List Extended Atribute vnode op into an mbuf list. 6813 */ 6814 int 6815 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, 6816 struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) 6817 { 6818 struct iovec iv; 6819 struct uio io; 6820 int error; 6821 size_t siz; 6822 6823 *bufp = NULL; 6824 /* First, find out the size of the extended attribute. */ 6825 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, 6826 p); 6827 if (error != 0) 6828 return (NFSERR_NOXATTR); 6829 if (siz <= cookie) { 6830 *lenp = 0; 6831 *eofp = true; 6832 goto out; 6833 } 6834 if (siz > cookie + *lenp) { 6835 siz = cookie + *lenp; 6836 *eofp = false; 6837 } else 6838 *eofp = true; 6839 /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ 6840 if (siz > 10 * 1024 * 1024) { 6841 error = NFSERR_XATTR2BIG; 6842 goto out; 6843 } 6844 *bufp = malloc(siz, M_TEMP, M_WAITOK); 6845 iv.iov_base = *bufp; 6846 iv.iov_len = siz; 6847 io.uio_iovcnt = 1; 6848 io.uio_iov = &iv; 6849 io.uio_offset = 0; 6850 io.uio_resid = siz; 6851 io.uio_rw = UIO_READ; 6852 io.uio_segflg = UIO_SYSSPACE; 6853 io.uio_td = p; 6854 #ifdef MAC 6855 error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); 6856 if (error != 0) 6857 goto out; 6858 #endif 6859 6860 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, 6861 p); 6862 if (error != 0) 6863 goto out; 6864 if (io.uio_resid > 0) 6865 siz -= io.uio_resid; 6866 *lenp = siz; 6867 6868 out: 6869 if (error != 0) { 6870 free(*bufp, M_TEMP); 6871 *bufp = NULL; 6872 } 6873 NFSEXITCODE(error); 6874 return (error); 6875 } 6876 6877 /* 6878 * Trim trailing data off the mbuf list being built. 6879 */ 6880 void 6881 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos, 6882 int bextpg, int bextpgsiz) 6883 { 6884 vm_page_t pg; 6885 int fullpgsiz, i; 6886 6887 if (mb->m_next != NULL) { 6888 m_freem(mb->m_next); 6889 mb->m_next = NULL; 6890 } 6891 if ((mb->m_flags & M_EXTPG) != 0) { 6892 KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs, 6893 ("nfsm_trimtrailing: bextpg out of range")); 6894 KASSERT(bpos == (char *)(void *) 6895 PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz, 6896 ("nfsm_trimtrailing: bextpgsiz bad!")); 6897 6898 /* First, get rid of any pages after this position. */ 6899 for (i = mb->m_epg_npgs - 1; i > bextpg; i--) { 6900 pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]); 6901 vm_page_unwire_noq(pg); 6902 vm_page_free(pg); 6903 } 6904 mb->m_epg_npgs = bextpg + 1; 6905 if (bextpg == 0) 6906 fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off; 6907 else 6908 fullpgsiz = PAGE_SIZE; 6909 mb->m_epg_last_len = fullpgsiz - bextpgsiz; 6910 mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off); 6911 for (i = 1; i < mb->m_epg_npgs; i++) 6912 mb->m_len += m_epg_pagelen(mb, i, 0); 6913 nd->nd_bextpgsiz = bextpgsiz; 6914 nd->nd_bextpg = bextpg; 6915 } else 6916 mb->m_len = bpos - mtod(mb, char *); 6917 nd->nd_mb = mb; 6918 nd->nd_bpos = bpos; 6919 } 6920 6921 6922 /* 6923 * Check to see if a put file handle operation should test for 6924 * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR. 6925 * When Open is the next operation, NFSERR_WRONGSEC cannot be 6926 * replied for the Open cases that use a component. Thia can 6927 * be identified by the fact that the file handle's type is VDIR. 6928 */ 6929 bool 6930 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp) 6931 { 6932 6933 if ((nd->nd_flag & ND_NFSV4) == 0) { 6934 if (nd->nd_procnum == NFSPROC_FSINFO) 6935 return (false); 6936 return (true); 6937 } 6938 6939 if ((nd->nd_flag & ND_LASTOP) != 0) 6940 return (false); 6941 6942 if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH || 6943 nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH || 6944 nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP || 6945 nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME) 6946 return (false); 6947 if (nextop == NFSV4OP_OPEN && vtyp == VDIR) 6948 return (false); 6949 return (true); 6950 } 6951 6952 /* 6953 * Check DSs marked no space. 6954 */ 6955 void 6956 nfsrv_checknospc(void) 6957 { 6958 struct statfs *tsf; 6959 struct nfsdevice *ds; 6960 struct vnode **dvpp, **tdvpp, *dvp; 6961 char *devid, *tdevid; 6962 int cnt, error = 0, i; 6963 6964 if (nfsrv_devidcnt <= 0) 6965 return; 6966 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 6967 devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK); 6968 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 6969 6970 /* Get an array of the dvps for the DSs. */ 6971 tdvpp = dvpp; 6972 tdevid = devid; 6973 i = 0; 6974 NFSDDSLOCK(); 6975 /* First, search for matches for same file system. */ 6976 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6977 if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) { 6978 if (++i > nfsrv_devidcnt) 6979 break; 6980 *tdvpp++ = ds->nfsdev_dvp; 6981 NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID); 6982 tdevid += NFSX_V4DEVICEID; 6983 } 6984 } 6985 NFSDDSUNLOCK(); 6986 6987 /* Do a VFS_STATFS() for each of the DSs and clear no space. */ 6988 cnt = i; 6989 tdvpp = dvpp; 6990 tdevid = devid; 6991 for (i = 0; i < cnt && error == 0; i++) { 6992 dvp = *tdvpp++; 6993 error = VFS_STATFS(dvp->v_mount, tsf); 6994 if (error == 0 && tsf->f_bavail > 0) { 6995 NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n"); 6996 nfsrv_marknospc(tdevid, false); 6997 } 6998 tdevid += NFSX_V4DEVICEID; 6999 } 7000 free(tsf, M_TEMP); 7001 free(dvpp, M_TEMP); 7002 free(devid, M_TEMP); 7003 } 7004 7005 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); 7006 7007 /* 7008 * Called once to initialize data structures... 7009 */ 7010 static int 7011 nfsd_modevent(module_t mod, int type, void *data) 7012 { 7013 int error = 0, i; 7014 static int loaded = 0; 7015 7016 switch (type) { 7017 case MOD_LOAD: 7018 if (loaded) 7019 goto out; 7020 newnfs_portinit(); 7021 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 7022 mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL, 7023 MTX_DEF); 7024 mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL, 7025 MTX_DEF); 7026 } 7027 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); 7028 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); 7029 mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF); 7030 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); 7031 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); 7032 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0); 7033 nfsrvd_initcache(); 7034 nfsd_init(); 7035 NFSD_LOCK(); 7036 nfsrvd_init(0); 7037 NFSD_UNLOCK(); 7038 nfsd_mntinit(); 7039 #ifdef VV_DISABLEDELEG 7040 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; 7041 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; 7042 #endif 7043 nfsd_call_servertimer = nfsrv_servertimer; 7044 nfsd_call_nfsd = nfssvc_nfsd; 7045 loaded = 1; 7046 break; 7047 7048 case MOD_UNLOAD: 7049 if (newnfs_numnfsd != 0) { 7050 error = EBUSY; 7051 break; 7052 } 7053 7054 #ifdef VV_DISABLEDELEG 7055 vn_deleg_ops.vndeleg_recall = NULL; 7056 vn_deleg_ops.vndeleg_disable = NULL; 7057 #endif 7058 nfsd_call_servertimer = NULL; 7059 nfsd_call_nfsd = NULL; 7060 7061 /* Clean out all NFSv4 state. */ 7062 nfsrv_throwawayallstate(curthread); 7063 7064 /* Clean the NFS server reply cache */ 7065 nfsrvd_cleancache(); 7066 7067 /* Free up the krpc server pool. */ 7068 if (nfsrvd_pool != NULL) 7069 svcpool_destroy(nfsrvd_pool); 7070 7071 /* and get rid of the locks */ 7072 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 7073 mtx_destroy(&nfsrchash_table[i].mtx); 7074 mtx_destroy(&nfsrcahash_table[i].mtx); 7075 } 7076 mtx_destroy(&nfsrc_udpmtx); 7077 mtx_destroy(&nfs_v4root_mutex); 7078 mtx_destroy(&nfsv4root_mnt.mnt_mtx); 7079 mtx_destroy(&nfsrv_dontlistlock_mtx); 7080 mtx_destroy(&nfsrv_recalllock_mtx); 7081 for (i = 0; i < nfsrv_sessionhashsize; i++) 7082 mtx_destroy(&nfssessionhash[i].mtx); 7083 if (nfslayouthash != NULL) { 7084 for (i = 0; i < nfsrv_layouthashsize; i++) 7085 mtx_destroy(&nfslayouthash[i].mtx); 7086 free(nfslayouthash, M_NFSDSESSION); 7087 } 7088 lockdestroy(&nfsv4root_mnt.mnt_explock); 7089 free(nfsclienthash, M_NFSDCLIENT); 7090 free(nfslockhash, M_NFSDLOCKFILE); 7091 free(nfssessionhash, M_NFSDSESSION); 7092 loaded = 0; 7093 break; 7094 default: 7095 error = EOPNOTSUPP; 7096 break; 7097 } 7098 7099 out: 7100 NFSEXITCODE(error); 7101 return (error); 7102 } 7103 static moduledata_t nfsd_mod = { 7104 "nfsd", 7105 nfsd_modevent, 7106 NULL, 7107 }; 7108 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); 7109 7110 /* So that loader and kldload(2) can find us, wherever we are.. */ 7111 MODULE_VERSION(nfsd, 1); 7112 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); 7113 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); 7114 MODULE_DEPEND(nfsd, krpc, 1, 1, 1); 7115 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); 7116