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 #include <sys/capsicum.h> 38 #include <sys/extattr.h> 39 40 /* 41 * Functions that perform the vfs operations required by the routines in 42 * nfsd_serv.c. It is hoped that this change will make the server more 43 * portable. 44 */ 45 46 #include <fs/nfs/nfsport.h> 47 #include <security/mac/mac_framework.h> 48 #include <sys/callout.h> 49 #include <sys/filio.h> 50 #include <sys/hash.h> 51 #include <sys/osd.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 int nfsrv_sessionhashsize; 62 extern struct nfslayouthash *nfslayouthash; 63 extern int nfsrv_layouthashsize; 64 extern struct mtx nfsrv_dslock_mtx; 65 extern int nfs_pnfsiothreads; 66 extern volatile int nfsrv_devidcnt; 67 extern int nfsrv_maxpnfsmirror; 68 extern uint32_t nfs_srvmaxio; 69 extern int nfs_bufpackets; 70 extern u_long sb_max_adj; 71 72 NFSD_VNET_DECLARE(int, nfsrv_numnfsd); 73 NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst); 74 NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool); 75 NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash); 76 NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash); 77 NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash); 78 NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock); 79 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p); 80 81 NFSDLOCKMUTEX; 82 NFSSTATESPINLOCK; 83 struct mtx nfsrc_udpmtx; 84 struct mtx nfs_v4root_mutex; 85 struct mtx nfsrv_dontlistlock_mtx; 86 struct mtx nfsrv_recalllock_mtx; 87 struct nfsrvfh nfs_pubfh; 88 int nfs_pubfhset = 0; 89 int nfsd_debuglevel = 0; 90 static pid_t nfsd_master_pid = (pid_t)-1; 91 static char nfsd_master_comm[MAXCOMLEN + 1]; 92 static struct timeval nfsd_master_start; 93 static uint32_t nfsv4_sysid = 0; 94 static fhandle_t zerofh; 95 96 NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL; 97 NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl); 98 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table); 99 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table); 100 NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh); 101 NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0; 102 NFSD_VNET_DEFINE(struct callout, nfsd_callout); 103 NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt); 104 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt); 105 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt); 106 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false; 107 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false; 108 109 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, 110 struct ucred *); 111 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *); 112 113 int nfsrv_enable_crossmntpt = 1; 114 static int nfs_commit_blks; 115 static int nfs_commit_miss; 116 extern int nfsrv_issuedelegs; 117 extern int nfsrv_dolocallocks; 118 extern struct nfsdevicehead nfsrv_devidhead; 119 120 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, 121 struct iovec **); 122 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **, 123 struct mbuf **, struct iovec **); 124 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, 125 int *); 126 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, 127 NFSPROC_T *); 128 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, 129 int *, char *, fhandle_t *); 130 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, 131 NFSPROC_T *); 132 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, 133 struct thread *, int, struct mbuf **, char *, struct mbuf **, 134 struct nfsvattr *, struct acl *, off_t *, int, bool *); 135 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); 136 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, 137 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); 138 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, 139 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, 140 char *, int *); 141 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 142 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 143 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 144 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 145 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 146 struct vnode *, struct nfsmount **, int, struct acl *, int *); 147 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 148 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); 149 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 150 struct vnode *, struct nfsmount *, struct nfsvattr *); 151 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, 152 NFSPROC_T *, struct nfsmount *); 153 static int nfsrv_putfhname(fhandle_t *, char *); 154 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, 155 struct pnfsdsfile *, struct vnode **, NFSPROC_T *); 156 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, 157 struct vnode *, NFSPROC_T *); 158 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); 159 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, 160 NFSPROC_T *); 161 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); 162 163 int nfs_pnfsio(task_fn_t *, void *); 164 165 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 166 "NFS server"); 167 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, 168 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); 169 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 170 0, ""); 171 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 172 0, ""); 173 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, 174 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); 175 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, 176 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); 177 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 178 0, "Debug level for NFS server"); 179 NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid); 180 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, 181 CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid), 182 0, "Enable nfsd to accept numeric owner_names"); 183 static int nfsrv_pnfsgetdsattr = 1; 184 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, 185 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); 186 187 /* 188 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are 189 * not running. 190 * The dsN subdirectories for the increased values must have been created 191 * on all DS servers before this increase is done. 192 */ 193 u_int nfsrv_dsdirsize = 20; 194 static int 195 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) 196 { 197 int error, newdsdirsize; 198 199 newdsdirsize = nfsrv_dsdirsize; 200 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); 201 if (error != 0 || req->newptr == NULL) 202 return (error); 203 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || 204 newnfs_numnfsd != 0) 205 return (EINVAL); 206 nfsrv_dsdirsize = newdsdirsize; 207 return (0); 208 } 209 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, 210 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), 211 sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); 212 213 /* 214 * nfs_srvmaxio can only be increased and only when the nfsd threads are 215 * not running. The setting must be a power of 2, with the current limit of 216 * 1Mbyte. 217 */ 218 static int 219 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS) 220 { 221 int error; 222 u_int newsrvmaxio; 223 uint64_t tval; 224 225 newsrvmaxio = nfs_srvmaxio; 226 error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req); 227 if (error != 0 || req->newptr == NULL) 228 return (error); 229 if (newsrvmaxio == nfs_srvmaxio) 230 return (0); 231 if (newsrvmaxio < nfs_srvmaxio) { 232 printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n"); 233 return (EINVAL); 234 } 235 if (newsrvmaxio > 1048576) { 236 printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n"); 237 return (EINVAL); 238 } 239 if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) { 240 printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n"); 241 return (EINVAL); 242 } 243 244 /* 245 * Check that kern.ipc.maxsockbuf is large enough for 246 * newsrviomax, given the setting of vfs.nfs.bufpackets. 247 */ 248 if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets > 249 sb_max_adj) { 250 /* 251 * Suggest vfs.nfs.bufpackets * maximum RPC message for 252 * sb_max_adj. 253 */ 254 tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets; 255 256 /* 257 * Convert suggested sb_max_adj value to a suggested 258 * sb_max value, which is what is set via kern.ipc.maxsockbuf. 259 * Perform the inverse calculation of (from uipc_sockbuf.c): 260 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES / 261 * (MSIZE + MCLBYTES); 262 * XXX If the calculation of sb_max_adj from sb_max changes, 263 * this calculation must be changed as well. 264 */ 265 tval *= (MSIZE + MCLBYTES); /* Brackets for readability. */ 266 tval += MCLBYTES - 1; /* Round up divide. */ 267 tval /= MCLBYTES; 268 printf("nfsd: set kern.ipc.maxsockbuf to a minimum of " 269 "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval, 270 newsrvmaxio); 271 return (EINVAL); 272 } 273 274 NFSD_LOCK(); 275 if (newnfs_numnfsd != 0) { 276 NFSD_UNLOCK(); 277 printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd " 278 "threads are running\n"); 279 return (EINVAL); 280 } 281 282 283 nfs_srvmaxio = newsrvmaxio; 284 NFSD_UNLOCK(); 285 return (0); 286 } 287 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio, 288 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 289 sysctl_srvmaxio, "IU", "Maximum I/O size in bytes"); 290 291 #define MAX_REORDERED_RPC 16 292 #define NUM_HEURISTIC 1031 293 #define NHUSE_INIT 64 294 #define NHUSE_INC 16 295 #define NHUSE_MAX 2048 296 297 static struct nfsheur { 298 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 299 off_t nh_nextoff; /* next offset for sequential detection */ 300 int nh_use; /* use count for selection */ 301 int nh_seqcount; /* heuristic */ 302 } nfsheur[NUM_HEURISTIC]; 303 304 /* 305 * Heuristic to detect sequential operation. 306 */ 307 static struct nfsheur * 308 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 309 { 310 struct nfsheur *nh; 311 int hi, try; 312 313 /* Locate best candidate. */ 314 try = 32; 315 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 316 nh = &nfsheur[hi]; 317 while (try--) { 318 if (nfsheur[hi].nh_vp == vp) { 319 nh = &nfsheur[hi]; 320 break; 321 } 322 if (nfsheur[hi].nh_use > 0) 323 --nfsheur[hi].nh_use; 324 hi = (hi + 1) % NUM_HEURISTIC; 325 if (nfsheur[hi].nh_use < nh->nh_use) 326 nh = &nfsheur[hi]; 327 } 328 329 /* Initialize hint if this is a new file. */ 330 if (nh->nh_vp != vp) { 331 nh->nh_vp = vp; 332 nh->nh_nextoff = uio->uio_offset; 333 nh->nh_use = NHUSE_INIT; 334 if (uio->uio_offset == 0) 335 nh->nh_seqcount = 4; 336 else 337 nh->nh_seqcount = 1; 338 } 339 340 /* Calculate heuristic. */ 341 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 342 uio->uio_offset == nh->nh_nextoff) { 343 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 344 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 345 if (nh->nh_seqcount > IO_SEQMAX) 346 nh->nh_seqcount = IO_SEQMAX; 347 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 348 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 349 /* Probably a reordered RPC, leave seqcount alone. */ 350 } else if (nh->nh_seqcount > 1) { 351 nh->nh_seqcount /= 2; 352 } else { 353 nh->nh_seqcount = 0; 354 } 355 nh->nh_use += NHUSE_INC; 356 if (nh->nh_use > NHUSE_MAX) 357 nh->nh_use = NHUSE_MAX; 358 return (nh); 359 } 360 361 /* 362 * Get attributes into nfsvattr structure. 363 */ 364 int 365 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, 366 struct nfsrv_descript *nd, struct thread *p, int vpislocked, 367 nfsattrbit_t *attrbitp) 368 { 369 int error, gotattr, lockedit = 0; 370 struct nfsvattr na; 371 372 if (vpislocked == 0) { 373 /* 374 * When vpislocked == 0, the vnode is either exclusively 375 * locked by this thread or not locked by this thread. 376 * As such, shared lock it, if not exclusively locked. 377 */ 378 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 379 lockedit = 1; 380 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); 381 } 382 } 383 384 /* 385 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed 386 * attributes, as required. 387 * This needs to be done for regular files if: 388 * - non-NFSv4 RPCs or 389 * - when attrbitp == NULL or 390 * - an NFSv4 RPC with any of the above attributes in attrbitp. 391 * A return of 0 for nfsrv_proxyds() indicates that it has acquired 392 * these attributes. nfsrv_proxyds() will return an error if the 393 * server is not a pNFS one. 394 */ 395 gotattr = 0; 396 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || 397 (nd->nd_flag & ND_NFSV4) == 0 || 398 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || 399 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || 400 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || 401 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || 402 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { 403 error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, 404 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, 405 NULL); 406 if (error == 0) 407 gotattr = 1; 408 } 409 410 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); 411 if (lockedit != 0) 412 NFSVOPUNLOCK(vp); 413 414 /* 415 * If we got the Change, Size and Modify Time from the DS, 416 * replace them. 417 */ 418 if (gotattr != 0) { 419 nvap->na_atime = na.na_atime; 420 nvap->na_mtime = na.na_mtime; 421 nvap->na_filerev = na.na_filerev; 422 nvap->na_size = na.na_size; 423 nvap->na_bytes = na.na_bytes; 424 } 425 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, 426 error, (uintmax_t)na.na_filerev); 427 428 NFSEXITCODE(error); 429 return (error); 430 } 431 432 /* 433 * Get a file handle for a vnode. 434 */ 435 int 436 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) 437 { 438 int error; 439 440 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); 441 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 442 error = VOP_VPTOFH(vp, &fhp->fh_fid); 443 444 NFSEXITCODE(error); 445 return (error); 446 } 447 448 /* 449 * Perform access checking for vnodes obtained from file handles that would 450 * refer to files already opened by a Unix client. You cannot just use 451 * vn_writechk() and VOP_ACCESSX() for two reasons. 452 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 453 * case. 454 * 2 - The owner is to be given access irrespective of mode bits for some 455 * operations, so that processes that chmod after opening a file don't 456 * break. 457 */ 458 int 459 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, 460 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, 461 u_int32_t *supportedtypep) 462 { 463 struct vattr vattr; 464 int error = 0, getret = 0; 465 466 if (vpislocked == 0) { 467 if (NFSVOPLOCK(vp, LK_SHARED) != 0) { 468 error = EPERM; 469 goto out; 470 } 471 } 472 if (accmode & VWRITE) { 473 /* Just vn_writechk() changed to check rdonly */ 474 /* 475 * Disallow write attempts on read-only file systems; 476 * unless the file is a socket or a block or character 477 * device resident on the file system. 478 */ 479 if (NFSVNO_EXRDONLY(exp) || 480 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 481 switch (vp->v_type) { 482 case VREG: 483 case VDIR: 484 case VLNK: 485 error = EROFS; 486 default: 487 break; 488 } 489 } 490 /* 491 * If there's shared text associated with 492 * the inode, try to free it up once. If 493 * we fail, we can't allow writing. 494 */ 495 if (VOP_IS_TEXT(vp) && error == 0) 496 error = ETXTBSY; 497 } 498 if (error != 0) { 499 if (vpislocked == 0) 500 NFSVOPUNLOCK(vp); 501 goto out; 502 } 503 504 /* 505 * Should the override still be applied when ACLs are enabled? 506 */ 507 error = VOP_ACCESSX(vp, accmode, cred, p); 508 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { 509 /* 510 * Try again with VEXPLICIT_DENY, to see if the test for 511 * deletion is supported. 512 */ 513 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); 514 if (error == 0) { 515 if (vp->v_type == VDIR) { 516 accmode &= ~(VDELETE | VDELETE_CHILD); 517 accmode |= VWRITE; 518 error = VOP_ACCESSX(vp, accmode, cred, p); 519 } else if (supportedtypep != NULL) { 520 *supportedtypep &= ~NFSACCESS_DELETE; 521 } 522 } 523 } 524 525 /* 526 * Allow certain operations for the owner (reads and writes 527 * on files that are already open). 528 */ 529 if (override != NFSACCCHK_NOOVERRIDE && 530 (error == EPERM || error == EACCES)) { 531 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) 532 error = 0; 533 else if (override & NFSACCCHK_ALLOWOWNER) { 534 getret = VOP_GETATTR(vp, &vattr, cred); 535 if (getret == 0 && cred->cr_uid == vattr.va_uid) 536 error = 0; 537 } 538 } 539 if (vpislocked == 0) 540 NFSVOPUNLOCK(vp); 541 542 out: 543 NFSEXITCODE(error); 544 return (error); 545 } 546 547 /* 548 * Set attribute(s) vnop. 549 */ 550 int 551 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 552 struct thread *p, struct nfsexstuff *exp) 553 { 554 u_quad_t savsize = 0; 555 int error, savedit; 556 time_t savbtime; 557 558 /* 559 * If this is an exported file system and a pNFS service is running, 560 * don't VOP_SETATTR() of size for the MDS file system. 561 */ 562 savedit = 0; 563 error = 0; 564 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && 565 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && 566 nvap->na_vattr.va_size > 0) { 567 savsize = nvap->na_vattr.va_size; 568 nvap->na_vattr.va_size = VNOVAL; 569 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 570 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 571 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 572 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 573 nvap->na_vattr.va_mtime.tv_sec != VNOVAL) 574 savedit = 1; 575 else 576 savedit = 2; 577 } 578 if (savedit != 2) 579 error = VOP_SETATTR(vp, &nvap->na_vattr, cred); 580 if (savedit != 0) 581 nvap->na_vattr.va_size = savsize; 582 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 583 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 584 nvap->na_vattr.va_size != VNOVAL || 585 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 586 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 587 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { 588 /* Never modify birthtime on a DS file. */ 589 savbtime = nvap->na_vattr.va_birthtime.tv_sec; 590 nvap->na_vattr.va_birthtime.tv_sec = VNOVAL; 591 /* For a pNFS server, set the attributes on the DS file. */ 592 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, 593 NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); 594 nvap->na_vattr.va_birthtime.tv_sec = savbtime; 595 if (error == ENOENT) 596 error = 0; 597 } 598 NFSEXITCODE(error); 599 return (error); 600 } 601 602 /* 603 * Set up nameidata for a lookup() call and do it. 604 */ 605 int 606 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, 607 struct vnode *dp, int islocked, struct nfsexstuff *exp, 608 struct vnode **retdirp) 609 { 610 struct componentname *cnp = &ndp->ni_cnd; 611 int i; 612 struct iovec aiov; 613 struct uio auio; 614 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; 615 int error = 0; 616 char *cp; 617 618 *retdirp = NULL; 619 cnp->cn_nameptr = cnp->cn_pnbuf; 620 ndp->ni_lcf = 0; 621 /* 622 * Extract and set starting directory. 623 */ 624 if (dp->v_type != VDIR) { 625 if (islocked) 626 vput(dp); 627 else 628 vrele(dp); 629 nfsvno_relpathbuf(ndp); 630 error = ENOTDIR; 631 goto out1; 632 } 633 if (islocked) 634 NFSVOPUNLOCK(dp); 635 VREF(dp); 636 *retdirp = dp; 637 if (NFSVNO_EXRDONLY(exp)) 638 cnp->cn_flags |= RDONLY; 639 ndp->ni_segflg = UIO_SYSSPACE; 640 641 if (nd->nd_flag & ND_PUBLOOKUP) { 642 ndp->ni_loopcnt = 0; 643 if (cnp->cn_pnbuf[0] == '/') { 644 vrele(dp); 645 /* 646 * Check for degenerate pathnames here, since lookup() 647 * panics on them. 648 */ 649 for (i = 1; i < ndp->ni_pathlen; i++) 650 if (cnp->cn_pnbuf[i] != '/') 651 break; 652 if (i == ndp->ni_pathlen) { 653 error = NFSERR_ACCES; 654 goto out; 655 } 656 dp = rootvnode; 657 VREF(dp); 658 } 659 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || 660 (nd->nd_flag & ND_NFSV4) == 0) { 661 /* 662 * Only cross mount points for NFSv4 when doing a 663 * mount while traversing the file system above 664 * the mount point, unless nfsrv_enable_crossmntpt is set. 665 */ 666 cnp->cn_flags |= NOCROSSMOUNT; 667 } 668 669 /* 670 * Initialize for scan, set ni_startdir and bump ref on dp again 671 * because lookup() will dereference ni_startdir. 672 */ 673 674 ndp->ni_startdir = dp; 675 ndp->ni_rootdir = rootvnode; 676 ndp->ni_topdir = NULL; 677 678 if (!lockleaf) 679 cnp->cn_flags |= LOCKLEAF; 680 for (;;) { 681 cnp->cn_nameptr = cnp->cn_pnbuf; 682 /* 683 * Call lookup() to do the real work. If an error occurs, 684 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and 685 * we do not have to dereference anything before returning. 686 * In either case ni_startdir will be dereferenced and NULLed 687 * out. 688 */ 689 error = vfs_lookup(ndp); 690 if (error) 691 break; 692 693 /* 694 * Check for encountering a symbolic link. Trivial 695 * termination occurs if no symlink encountered. 696 */ 697 if ((cnp->cn_flags & ISSYMLINK) == 0) { 698 if (ndp->ni_vp && !lockleaf) 699 NFSVOPUNLOCK(ndp->ni_vp); 700 break; 701 } 702 703 /* 704 * Validate symlink 705 */ 706 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) 707 NFSVOPUNLOCK(ndp->ni_dvp); 708 if (!(nd->nd_flag & ND_PUBLOOKUP)) { 709 error = EINVAL; 710 goto badlink2; 711 } 712 713 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 714 error = ELOOP; 715 goto badlink2; 716 } 717 if (ndp->ni_pathlen > 1) 718 cp = uma_zalloc(namei_zone, M_WAITOK); 719 else 720 cp = cnp->cn_pnbuf; 721 aiov.iov_base = cp; 722 aiov.iov_len = MAXPATHLEN; 723 auio.uio_iov = &aiov; 724 auio.uio_iovcnt = 1; 725 auio.uio_offset = 0; 726 auio.uio_rw = UIO_READ; 727 auio.uio_segflg = UIO_SYSSPACE; 728 auio.uio_td = NULL; 729 auio.uio_resid = MAXPATHLEN; 730 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 731 if (error) { 732 badlink1: 733 if (ndp->ni_pathlen > 1) 734 uma_zfree(namei_zone, cp); 735 badlink2: 736 vrele(ndp->ni_dvp); 737 vput(ndp->ni_vp); 738 break; 739 } 740 linklen = MAXPATHLEN - auio.uio_resid; 741 if (linklen == 0) { 742 error = ENOENT; 743 goto badlink1; 744 } 745 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 746 error = ENAMETOOLONG; 747 goto badlink1; 748 } 749 750 /* 751 * Adjust or replace path 752 */ 753 if (ndp->ni_pathlen > 1) { 754 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 755 uma_zfree(namei_zone, cnp->cn_pnbuf); 756 cnp->cn_pnbuf = cp; 757 } else 758 cnp->cn_pnbuf[linklen] = '\0'; 759 ndp->ni_pathlen += linklen; 760 761 /* 762 * Cleanup refs for next loop and check if root directory 763 * should replace current directory. Normally ni_dvp 764 * becomes the new base directory and is cleaned up when 765 * we loop. Explicitly null pointers after invalidation 766 * to clarify operation. 767 */ 768 vput(ndp->ni_vp); 769 ndp->ni_vp = NULL; 770 771 if (cnp->cn_pnbuf[0] == '/') { 772 vrele(ndp->ni_dvp); 773 ndp->ni_dvp = ndp->ni_rootdir; 774 VREF(ndp->ni_dvp); 775 } 776 ndp->ni_startdir = ndp->ni_dvp; 777 ndp->ni_dvp = NULL; 778 } 779 if (!lockleaf) 780 cnp->cn_flags &= ~LOCKLEAF; 781 782 out: 783 if (error) { 784 nfsvno_relpathbuf(ndp); 785 ndp->ni_vp = NULL; 786 ndp->ni_dvp = NULL; 787 ndp->ni_startdir = NULL; 788 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { 789 ndp->ni_dvp = NULL; 790 } 791 792 out1: 793 NFSEXITCODE2(error, nd); 794 return (error); 795 } 796 797 /* 798 * Set up a pathname buffer and return a pointer to it and, optionally 799 * set a hash pointer. 800 */ 801 void 802 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) 803 { 804 struct componentname *cnp = &ndp->ni_cnd; 805 806 cnp->cn_flags |= (NOMACCHECK); 807 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 808 if (hashpp != NULL) 809 *hashpp = NULL; 810 *bufpp = cnp->cn_pnbuf; 811 } 812 813 /* 814 * Release the above path buffer, if not released by nfsvno_namei(). 815 */ 816 void 817 nfsvno_relpathbuf(struct nameidata *ndp) 818 { 819 820 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 821 ndp->ni_cnd.cn_pnbuf = NULL; 822 } 823 824 /* 825 * Readlink vnode op into an mbuf list. 826 */ 827 int 828 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz, 829 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 830 { 831 struct iovec *iv; 832 struct uio io, *uiop = &io; 833 struct mbuf *mp, *mp3; 834 int len, tlen, error = 0; 835 836 len = NFS_MAXPATHLEN; 837 if (maxextsiz > 0) 838 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 839 &mp3, &mp, &iv); 840 else 841 uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); 842 uiop->uio_iov = iv; 843 uiop->uio_offset = 0; 844 uiop->uio_resid = len; 845 uiop->uio_rw = UIO_READ; 846 uiop->uio_segflg = UIO_SYSSPACE; 847 uiop->uio_td = NULL; 848 error = VOP_READLINK(vp, uiop, cred); 849 free(iv, M_TEMP); 850 if (error) { 851 m_freem(mp3); 852 *lenp = 0; 853 goto out; 854 } 855 if (uiop->uio_resid > 0) { 856 len -= uiop->uio_resid; 857 tlen = NFSM_RNDUP(len); 858 if (tlen == 0) { 859 m_freem(mp3); 860 mp3 = mp = NULL; 861 } else if (tlen != NFS_MAXPATHLEN || tlen != len) 862 mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, 863 tlen - len); 864 } 865 *lenp = len; 866 *mpp = mp3; 867 *mpendp = mp; 868 869 out: 870 NFSEXITCODE(error); 871 return (error); 872 } 873 874 /* 875 * Create an mbuf chain and an associated iovec that can be used to Read 876 * or Getextattr of data. 877 * Upon success, return pointers to the first and last mbufs in the chain 878 * plus the malloc'd iovec and its iovlen. 879 */ 880 static int 881 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, 882 struct iovec **ivp) 883 { 884 struct mbuf *m, *m2 = NULL, *m3; 885 struct iovec *iv; 886 int i, left, siz; 887 888 left = len; 889 m3 = NULL; 890 /* 891 * Generate the mbuf list with the uio_iov ref. to it. 892 */ 893 i = 0; 894 while (left > 0) { 895 NFSMGET(m); 896 MCLGET(m, M_WAITOK); 897 m->m_len = 0; 898 siz = min(M_TRAILINGSPACE(m), left); 899 left -= siz; 900 i++; 901 if (m3) 902 m2->m_next = m; 903 else 904 m3 = m; 905 m2 = m; 906 } 907 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 908 m = m3; 909 left = len; 910 i = 0; 911 while (left > 0) { 912 if (m == NULL) 913 panic("nfsrv_createiovec iov"); 914 siz = min(M_TRAILINGSPACE(m), left); 915 if (siz > 0) { 916 iv->iov_base = mtod(m, caddr_t) + m->m_len; 917 iv->iov_len = siz; 918 m->m_len += siz; 919 left -= siz; 920 iv++; 921 i++; 922 } 923 m = m->m_next; 924 } 925 *mpp = m3; 926 *mpendp = m2; 927 return (i); 928 } 929 930 /* 931 * Create an mbuf chain and an associated iovec that can be used to Read 932 * or Getextattr of data. 933 * Upon success, return pointers to the first and last mbufs in the chain 934 * plus the malloc'd iovec and its iovlen. 935 * Same as above, but creates ext_pgs mbuf(s). 936 */ 937 static int 938 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp, 939 struct mbuf **mpendp, struct iovec **ivp) 940 { 941 struct mbuf *m, *m2 = NULL, *m3; 942 struct iovec *iv; 943 int i, left, pgno, siz; 944 945 left = len; 946 m3 = NULL; 947 /* 948 * Generate the mbuf list with the uio_iov ref. to it. 949 */ 950 i = 0; 951 while (left > 0) { 952 siz = min(left, maxextsiz); 953 m = mb_alloc_ext_plus_pages(siz, M_WAITOK); 954 left -= siz; 955 i += m->m_epg_npgs; 956 if (m3 != NULL) 957 m2->m_next = m; 958 else 959 m3 = m; 960 m2 = m; 961 } 962 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 963 m = m3; 964 left = len; 965 i = 0; 966 pgno = 0; 967 while (left > 0) { 968 if (m == NULL) 969 panic("nfsvno_createiovec_extpgs iov"); 970 siz = min(PAGE_SIZE, left); 971 if (siz > 0) { 972 iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); 973 iv->iov_len = siz; 974 m->m_len += siz; 975 if (pgno == m->m_epg_npgs - 1) 976 m->m_epg_last_len = siz; 977 left -= siz; 978 iv++; 979 i++; 980 pgno++; 981 } 982 if (pgno == m->m_epg_npgs && left > 0) { 983 m = m->m_next; 984 if (m == NULL) 985 panic("nfsvno_createiovec_extpgs iov"); 986 pgno = 0; 987 } 988 } 989 *mpp = m3; 990 *mpendp = m2; 991 return (i); 992 } 993 994 /* 995 * Read vnode op call into mbuf list. 996 */ 997 int 998 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 999 int maxextsiz, struct thread *p, struct mbuf **mpp, 1000 struct mbuf **mpendp) 1001 { 1002 struct mbuf *m; 1003 struct iovec *iv; 1004 int error = 0, len, tlen, ioflag = 0; 1005 struct mbuf *m3; 1006 struct uio io, *uiop = &io; 1007 struct nfsheur *nh; 1008 1009 /* 1010 * Attempt to read from a DS file. A return of ENOENT implies 1011 * there is no DS file to read. 1012 */ 1013 error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, 1014 NULL, mpendp, NULL, NULL, NULL, 0, NULL); 1015 if (error != ENOENT) 1016 return (error); 1017 1018 len = NFSM_RNDUP(cnt); 1019 if (maxextsiz > 0) 1020 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 1021 &m3, &m, &iv); 1022 else 1023 uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); 1024 uiop->uio_iov = iv; 1025 uiop->uio_offset = off; 1026 uiop->uio_resid = len; 1027 uiop->uio_rw = UIO_READ; 1028 uiop->uio_segflg = UIO_SYSSPACE; 1029 uiop->uio_td = NULL; 1030 nh = nfsrv_sequential_heuristic(uiop, vp); 1031 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 1032 /* XXX KDM make this more systematic? */ 1033 NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_READ] += uiop->uio_resid; 1034 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 1035 free(iv, M_TEMP); 1036 if (error) { 1037 m_freem(m3); 1038 *mpp = NULL; 1039 goto out; 1040 } 1041 nh->nh_nextoff = uiop->uio_offset; 1042 tlen = len - uiop->uio_resid; 1043 cnt = cnt < tlen ? cnt : tlen; 1044 tlen = NFSM_RNDUP(cnt); 1045 if (tlen == 0) { 1046 m_freem(m3); 1047 m3 = m = NULL; 1048 } else if (len != tlen || tlen != cnt) 1049 m = nfsrv_adj(m3, len - tlen, tlen - cnt); 1050 *mpp = m3; 1051 *mpendp = m; 1052 1053 out: 1054 NFSEXITCODE(error); 1055 return (error); 1056 } 1057 1058 /* 1059 * Create the iovec for the mbuf chain passed in as an argument. 1060 * The "cp" argument is where the data starts within the first mbuf in 1061 * the chain. It returns the iovec and the iovcnt. 1062 */ 1063 static int 1064 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, 1065 int *iovcntp) 1066 { 1067 struct mbuf *mp; 1068 struct iovec *ivp; 1069 int cnt, i, len; 1070 1071 /* 1072 * Loop through the mbuf chain, counting how many mbufs are a 1073 * part of this write operation, so the iovec size is known. 1074 */ 1075 cnt = 0; 1076 len = retlen; 1077 mp = m; 1078 i = mtod(mp, caddr_t) + mp->m_len - cp; 1079 while (len > 0) { 1080 if (i > 0) { 1081 len -= i; 1082 cnt++; 1083 } 1084 mp = mp->m_next; 1085 if (!mp) { 1086 if (len > 0) 1087 return (EBADRPC); 1088 } else 1089 i = mp->m_len; 1090 } 1091 1092 /* Now, create the iovec. */ 1093 mp = m; 1094 *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, 1095 M_WAITOK); 1096 *iovcntp = cnt; 1097 i = mtod(mp, caddr_t) + mp->m_len - cp; 1098 len = retlen; 1099 while (len > 0) { 1100 if (mp == NULL) 1101 panic("nfsrv_createiovecw"); 1102 if (i > 0) { 1103 i = min(i, len); 1104 ivp->iov_base = cp; 1105 ivp->iov_len = i; 1106 ivp++; 1107 len -= i; 1108 } 1109 mp = mp->m_next; 1110 if (mp) { 1111 i = mp->m_len; 1112 cp = mtod(mp, caddr_t); 1113 } 1114 } 1115 return (0); 1116 } 1117 1118 /* 1119 * Write vnode op from an mbuf list. 1120 */ 1121 int 1122 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, 1123 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) 1124 { 1125 struct iovec *iv; 1126 int cnt, ioflags, error; 1127 struct uio io, *uiop = &io; 1128 struct nfsheur *nh; 1129 1130 /* 1131 * Attempt to write to a DS file. A return of ENOENT implies 1132 * there is no DS file to write. 1133 */ 1134 error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, 1135 &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); 1136 if (error != ENOENT) { 1137 *stable = NFSWRITE_FILESYNC; 1138 return (error); 1139 } 1140 1141 if (*stable == NFSWRITE_UNSTABLE) 1142 ioflags = IO_NODELOCKED; 1143 else 1144 ioflags = (IO_SYNC | IO_NODELOCKED); 1145 error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); 1146 if (error != 0) 1147 return (error); 1148 uiop->uio_iov = iv; 1149 uiop->uio_iovcnt = cnt; 1150 uiop->uio_resid = retlen; 1151 uiop->uio_rw = UIO_WRITE; 1152 uiop->uio_segflg = UIO_SYSSPACE; 1153 NFSUIOPROC(uiop, p); 1154 uiop->uio_offset = off; 1155 nh = nfsrv_sequential_heuristic(uiop, vp); 1156 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 1157 /* XXX KDM make this more systematic? */ 1158 NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; 1159 error = VOP_WRITE(vp, uiop, ioflags, cred); 1160 if (error == 0) 1161 nh->nh_nextoff = uiop->uio_offset; 1162 free(iv, M_TEMP); 1163 1164 NFSEXITCODE(error); 1165 return (error); 1166 } 1167 1168 /* 1169 * Common code for creating a regular file (plus special files for V2). 1170 */ 1171 int 1172 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, 1173 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, 1174 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) 1175 { 1176 u_quad_t tempsize; 1177 int error; 1178 struct thread *p = curthread; 1179 1180 error = nd->nd_repstat; 1181 if (!error && ndp->ni_vp == NULL) { 1182 if (nvap->na_type == VREG || nvap->na_type == VSOCK) { 1183 error = VOP_CREATE(ndp->ni_dvp, 1184 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1185 /* For a pNFS server, create the data file on a DS. */ 1186 if (error == 0 && nvap->na_type == VREG) { 1187 /* 1188 * Create a data file on a DS for a pNFS server. 1189 * This function just returns if not 1190 * running a pNFS DS or the creation fails. 1191 */ 1192 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1193 nd->nd_cred, p); 1194 } 1195 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1196 NULL, false); 1197 nfsvno_relpathbuf(ndp); 1198 if (!error) { 1199 if (*exclusive_flagp) { 1200 *exclusive_flagp = 0; 1201 NFSVNO_ATTRINIT(nvap); 1202 nvap->na_atime.tv_sec = cverf[0]; 1203 nvap->na_atime.tv_nsec = cverf[1]; 1204 error = VOP_SETATTR(ndp->ni_vp, 1205 &nvap->na_vattr, nd->nd_cred); 1206 if (error != 0) { 1207 vput(ndp->ni_vp); 1208 ndp->ni_vp = NULL; 1209 error = NFSERR_NOTSUPP; 1210 } 1211 } 1212 } 1213 /* 1214 * NFS V2 Only. nfsrvd_mknod() does this for V3. 1215 * (This implies, just get out on an error.) 1216 */ 1217 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || 1218 nvap->na_type == VFIFO) { 1219 if (nvap->na_type == VCHR && rdev == 0xffffffff) 1220 nvap->na_type = VFIFO; 1221 if (nvap->na_type != VFIFO && 1222 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { 1223 nfsvno_relpathbuf(ndp); 1224 vput(ndp->ni_dvp); 1225 goto out; 1226 } 1227 nvap->na_rdev = rdev; 1228 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1229 &ndp->ni_cnd, &nvap->na_vattr); 1230 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1231 NULL, false); 1232 nfsvno_relpathbuf(ndp); 1233 if (error) 1234 goto out; 1235 } else { 1236 nfsvno_relpathbuf(ndp); 1237 vput(ndp->ni_dvp); 1238 error = ENXIO; 1239 goto out; 1240 } 1241 *vpp = ndp->ni_vp; 1242 } else { 1243 /* 1244 * Handle cases where error is already set and/or 1245 * the file exists. 1246 * 1 - clean up the lookup 1247 * 2 - iff !error and na_size set, truncate it 1248 */ 1249 nfsvno_relpathbuf(ndp); 1250 *vpp = ndp->ni_vp; 1251 if (ndp->ni_dvp == *vpp) 1252 vrele(ndp->ni_dvp); 1253 else 1254 vput(ndp->ni_dvp); 1255 if (!error && nvap->na_size != VNOVAL) { 1256 error = nfsvno_accchk(*vpp, VWRITE, 1257 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1258 NFSACCCHK_VPISLOCKED, NULL); 1259 if (!error) { 1260 tempsize = nvap->na_size; 1261 NFSVNO_ATTRINIT(nvap); 1262 nvap->na_size = tempsize; 1263 error = nfsvno_setattr(*vpp, nvap, 1264 nd->nd_cred, p, exp); 1265 } 1266 } 1267 if (error) 1268 vput(*vpp); 1269 } 1270 1271 out: 1272 NFSEXITCODE(error); 1273 return (error); 1274 } 1275 1276 /* 1277 * Do a mknod vnode op. 1278 */ 1279 int 1280 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, 1281 struct thread *p) 1282 { 1283 int error = 0; 1284 __enum_uint8(vtype) vtyp; 1285 1286 vtyp = nvap->na_type; 1287 /* 1288 * Iff doesn't exist, create it. 1289 */ 1290 if (ndp->ni_vp) { 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 nfsvno_relpathbuf(ndp); 1299 vput(ndp->ni_dvp); 1300 error = NFSERR_BADTYPE; 1301 goto out; 1302 } 1303 if (vtyp == VSOCK) { 1304 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 1305 &ndp->ni_cnd, &nvap->na_vattr); 1306 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1307 false); 1308 nfsvno_relpathbuf(ndp); 1309 } else { 1310 if (nvap->na_type != VFIFO && 1311 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { 1312 nfsvno_relpathbuf(ndp); 1313 vput(ndp->ni_dvp); 1314 goto out; 1315 } 1316 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1317 &ndp->ni_cnd, &nvap->na_vattr); 1318 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1319 false); 1320 nfsvno_relpathbuf(ndp); 1321 /* 1322 * Since VOP_MKNOD returns the ni_vp, I can't 1323 * see any reason to do the lookup. 1324 */ 1325 } 1326 1327 out: 1328 NFSEXITCODE(error); 1329 return (error); 1330 } 1331 1332 /* 1333 * Mkdir vnode op. 1334 */ 1335 int 1336 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, 1337 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 1338 { 1339 int error = 0; 1340 1341 if (ndp->ni_vp != NULL) { 1342 if (ndp->ni_dvp == ndp->ni_vp) 1343 vrele(ndp->ni_dvp); 1344 else 1345 vput(ndp->ni_dvp); 1346 vrele(ndp->ni_vp); 1347 nfsvno_relpathbuf(ndp); 1348 error = EEXIST; 1349 goto out; 1350 } 1351 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1352 &nvap->na_vattr); 1353 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); 1354 nfsvno_relpathbuf(ndp); 1355 1356 out: 1357 NFSEXITCODE(error); 1358 return (error); 1359 } 1360 1361 /* 1362 * symlink vnode op. 1363 */ 1364 int 1365 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, 1366 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, 1367 struct nfsexstuff *exp) 1368 { 1369 int error = 0; 1370 1371 if (ndp->ni_vp) { 1372 nfsvno_relpathbuf(ndp); 1373 if (ndp->ni_dvp == ndp->ni_vp) 1374 vrele(ndp->ni_dvp); 1375 else 1376 vput(ndp->ni_dvp); 1377 vrele(ndp->ni_vp); 1378 error = EEXIST; 1379 goto out; 1380 } 1381 1382 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1383 &nvap->na_vattr, pathcp); 1384 /* 1385 * Although FreeBSD still had the lookup code in 1386 * it for 7/current, there doesn't seem to be any 1387 * point, since VOP_SYMLINK() returns the ni_vp. 1388 * Just vput it for v2. 1389 */ 1390 VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0); 1391 nfsvno_relpathbuf(ndp); 1392 1393 out: 1394 NFSEXITCODE(error); 1395 return (error); 1396 } 1397 1398 /* 1399 * Parse symbolic link arguments. 1400 * This function has an ugly side effect. It will malloc() an area for 1401 * the symlink and set iov_base to point to it, only if it succeeds. 1402 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must 1403 * be FREE'd later. 1404 */ 1405 int 1406 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, 1407 struct thread *p, char **pathcpp, int *lenp) 1408 { 1409 u_int32_t *tl; 1410 char *pathcp = NULL; 1411 int error = 0, len; 1412 struct nfsv2_sattr *sp; 1413 1414 *pathcpp = NULL; 1415 *lenp = 0; 1416 if ((nd->nd_flag & ND_NFSV3) && 1417 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) 1418 goto nfsmout; 1419 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 1420 len = fxdr_unsigned(int, *tl); 1421 if (len > NFS_MAXPATHLEN || len <= 0) { 1422 error = EBADRPC; 1423 goto nfsmout; 1424 } 1425 pathcp = malloc(len + 1, M_TEMP, M_WAITOK); 1426 error = nfsrv_mtostr(nd, pathcp, len); 1427 if (error) 1428 goto nfsmout; 1429 if (nd->nd_flag & ND_NFSV2) { 1430 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1431 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 1432 } 1433 *pathcpp = pathcp; 1434 *lenp = len; 1435 NFSEXITCODE2(0, nd); 1436 return (0); 1437 nfsmout: 1438 if (pathcp) 1439 free(pathcp, M_TEMP); 1440 NFSEXITCODE2(error, nd); 1441 return (error); 1442 } 1443 1444 /* 1445 * Remove a non-directory object. 1446 */ 1447 int 1448 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1449 struct thread *p, struct nfsexstuff *exp) 1450 { 1451 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS]; 1452 int error = 0, mirrorcnt; 1453 char fname[PNFS_FILENAME_LEN + 1]; 1454 fhandle_t fh; 1455 1456 vp = ndp->ni_vp; 1457 dsdvp[0] = NULL; 1458 if (vp->v_type == VDIR) 1459 error = NFSERR_ISDIR; 1460 else if (is_v4) 1461 error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0), 1462 p); 1463 if (error == 0) 1464 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); 1465 if (!error) 1466 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); 1467 if (error == 0 && dsdvp[0] != NULL) 1468 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1469 if (ndp->ni_dvp == vp) 1470 vrele(ndp->ni_dvp); 1471 else 1472 vput(ndp->ni_dvp); 1473 vput(vp); 1474 nfsvno_relpathbuf(ndp); 1475 NFSEXITCODE(error); 1476 return (error); 1477 } 1478 1479 /* 1480 * Remove a directory. 1481 */ 1482 int 1483 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1484 struct thread *p, struct nfsexstuff *exp) 1485 { 1486 struct vnode *vp; 1487 int error = 0; 1488 1489 vp = ndp->ni_vp; 1490 if (vp->v_type != VDIR) { 1491 error = ENOTDIR; 1492 goto out; 1493 } 1494 /* 1495 * No rmdir "." please. 1496 */ 1497 if (ndp->ni_dvp == vp) { 1498 error = EINVAL; 1499 goto out; 1500 } 1501 /* 1502 * The root of a mounted filesystem cannot be deleted. 1503 */ 1504 if (vp->v_vflag & VV_ROOT) 1505 error = EBUSY; 1506 out: 1507 if (!error) 1508 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); 1509 if (ndp->ni_dvp == vp) 1510 vrele(ndp->ni_dvp); 1511 else 1512 vput(ndp->ni_dvp); 1513 vput(vp); 1514 nfsvno_relpathbuf(ndp); 1515 NFSEXITCODE(error); 1516 return (error); 1517 } 1518 1519 /* 1520 * Rename vnode op. 1521 */ 1522 int 1523 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, 1524 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) 1525 { 1526 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS]; 1527 int error = 0, mirrorcnt; 1528 char fname[PNFS_FILENAME_LEN + 1]; 1529 fhandle_t fh; 1530 1531 dsdvp[0] = NULL; 1532 fvp = fromndp->ni_vp; 1533 if (ndstat) { 1534 vrele(fromndp->ni_dvp); 1535 vrele(fvp); 1536 error = ndstat; 1537 goto out1; 1538 } 1539 tdvp = tondp->ni_dvp; 1540 tvp = tondp->ni_vp; 1541 if (tvp != NULL) { 1542 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1543 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; 1544 goto out; 1545 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1546 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; 1547 goto out; 1548 } 1549 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1550 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1551 goto out; 1552 } 1553 1554 /* 1555 * A rename to '.' or '..' results in a prematurely 1556 * unlocked vnode on FreeBSD5, so I'm just going to fail that 1557 * here. 1558 */ 1559 if ((tondp->ni_cnd.cn_namelen == 1 && 1560 tondp->ni_cnd.cn_nameptr[0] == '.') || 1561 (tondp->ni_cnd.cn_namelen == 2 && 1562 tondp->ni_cnd.cn_nameptr[0] == '.' && 1563 tondp->ni_cnd.cn_nameptr[1] == '.')) { 1564 error = EINVAL; 1565 goto out; 1566 } 1567 } 1568 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1569 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1570 goto out; 1571 } 1572 if (fvp->v_mount != tdvp->v_mount) { 1573 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1574 goto out; 1575 } 1576 if (fvp == tdvp) { 1577 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; 1578 goto out; 1579 } 1580 if (fvp == tvp) { 1581 /* 1582 * If source and destination are the same, there is nothing to 1583 * do. Set error to -1 to indicate this. 1584 */ 1585 error = -1; 1586 goto out; 1587 } 1588 if (ndflag & ND_NFSV4) { 1589 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { 1590 error = nfsrv_checkremove(fvp, 0, NULL, 1591 (nfsquad_t)((u_quad_t)0), p); 1592 NFSVOPUNLOCK(fvp); 1593 } else 1594 error = EPERM; 1595 if (tvp && !error) 1596 error = nfsrv_checkremove(tvp, 1, NULL, 1597 (nfsquad_t)((u_quad_t)0), p); 1598 } else { 1599 /* 1600 * For NFSv2 and NFSv3, try to get rid of the delegation, so 1601 * that the NFSv4 client won't be confused by the rename. 1602 * Since nfsd_recalldelegation() can only be called on an 1603 * unlocked vnode at this point and fvp is the file that will 1604 * still exist after the rename, just do fvp. 1605 */ 1606 nfsd_recalldelegation(fvp, p); 1607 } 1608 if (error == 0 && tvp != NULL) { 1609 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh); 1610 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" 1611 " dsdvp=%p\n", dsdvp[0]); 1612 } 1613 out: 1614 if (!error) { 1615 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, 1616 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, 1617 &tondp->ni_cnd); 1618 } else { 1619 if (tdvp == tvp) 1620 vrele(tdvp); 1621 else 1622 vput(tdvp); 1623 if (tvp) 1624 vput(tvp); 1625 vrele(fromndp->ni_dvp); 1626 vrele(fvp); 1627 if (error == -1) 1628 error = 0; 1629 } 1630 1631 /* 1632 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and 1633 * if the rename succeeded, the DS file for the tvp needs to be 1634 * removed. 1635 */ 1636 if (error == 0 && dsdvp[0] != NULL) { 1637 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1638 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); 1639 } 1640 1641 nfsvno_relpathbuf(tondp); 1642 out1: 1643 nfsvno_relpathbuf(fromndp); 1644 NFSEXITCODE(error); 1645 return (error); 1646 } 1647 1648 /* 1649 * Link vnode op. 1650 */ 1651 int 1652 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, 1653 struct thread *p, struct nfsexstuff *exp) 1654 { 1655 struct vnode *xp; 1656 int error = 0; 1657 1658 xp = ndp->ni_vp; 1659 if (xp != NULL) { 1660 error = EEXIST; 1661 } else { 1662 xp = ndp->ni_dvp; 1663 if (vp->v_mount != xp->v_mount) 1664 error = EXDEV; 1665 } 1666 if (!error) { 1667 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1668 if (!VN_IS_DOOMED(vp)) 1669 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); 1670 else 1671 error = EPERM; 1672 if (ndp->ni_dvp == vp) { 1673 vrele(ndp->ni_dvp); 1674 NFSVOPUNLOCK(vp); 1675 } else { 1676 vref(vp); 1677 VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true); 1678 } 1679 } else { 1680 if (ndp->ni_dvp == ndp->ni_vp) 1681 vrele(ndp->ni_dvp); 1682 else 1683 vput(ndp->ni_dvp); 1684 if (ndp->ni_vp) 1685 vrele(ndp->ni_vp); 1686 } 1687 nfsvno_relpathbuf(ndp); 1688 NFSEXITCODE(error); 1689 return (error); 1690 } 1691 1692 /* 1693 * Do the fsync() appropriate for the commit. 1694 */ 1695 int 1696 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, 1697 struct thread *td) 1698 { 1699 int error = 0; 1700 1701 /* 1702 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of 1703 * file is done. At this time VOP_FSYNC does not accept offset and 1704 * byte count parameters so call VOP_FSYNC the whole file for now. 1705 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. 1706 * File systems that do not use the buffer cache (as indicated 1707 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). 1708 */ 1709 if (cnt == 0 || cnt > MAX_COMMIT_COUNT || 1710 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { 1711 /* 1712 * Give up and do the whole thing 1713 */ 1714 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1715 VM_OBJECT_WLOCK(vp->v_object); 1716 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); 1717 VM_OBJECT_WUNLOCK(vp->v_object); 1718 } 1719 error = VOP_FSYNC(vp, MNT_WAIT, td); 1720 } else { 1721 /* 1722 * Locate and synchronously write any buffers that fall 1723 * into the requested range. Note: we are assuming that 1724 * f_iosize is a power of 2. 1725 */ 1726 int iosize = vp->v_mount->mnt_stat.f_iosize; 1727 int iomask = iosize - 1; 1728 struct bufobj *bo; 1729 daddr_t lblkno; 1730 1731 /* 1732 * Align to iosize boundary, super-align to page boundary. 1733 */ 1734 if (off & iomask) { 1735 cnt += off & iomask; 1736 off &= ~(u_quad_t)iomask; 1737 } 1738 if (off & PAGE_MASK) { 1739 cnt += off & PAGE_MASK; 1740 off &= ~(u_quad_t)PAGE_MASK; 1741 } 1742 lblkno = off / iosize; 1743 1744 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1745 VM_OBJECT_WLOCK(vp->v_object); 1746 vm_object_page_clean(vp->v_object, off, off + cnt, 1747 OBJPC_SYNC); 1748 VM_OBJECT_WUNLOCK(vp->v_object); 1749 } 1750 1751 bo = &vp->v_bufobj; 1752 BO_LOCK(bo); 1753 while (cnt > 0) { 1754 struct buf *bp; 1755 1756 /* 1757 * If we have a buffer and it is marked B_DELWRI we 1758 * have to lock and write it. Otherwise the prior 1759 * write is assumed to have already been committed. 1760 * 1761 * gbincore() can return invalid buffers now so we 1762 * have to check that bit as well (though B_DELWRI 1763 * should not be set if B_INVAL is set there could be 1764 * a race here since we haven't locked the buffer). 1765 */ 1766 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 1767 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 1768 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { 1769 BO_LOCK(bo); 1770 continue; /* retry */ 1771 } 1772 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 1773 B_DELWRI) { 1774 bremfree(bp); 1775 bp->b_flags &= ~B_ASYNC; 1776 bwrite(bp); 1777 ++nfs_commit_miss; 1778 } else 1779 BUF_UNLOCK(bp); 1780 BO_LOCK(bo); 1781 } 1782 ++nfs_commit_blks; 1783 if (cnt < iosize) 1784 break; 1785 cnt -= iosize; 1786 ++lblkno; 1787 } 1788 BO_UNLOCK(bo); 1789 } 1790 NFSEXITCODE(error); 1791 return (error); 1792 } 1793 1794 /* 1795 * Statfs vnode op. 1796 */ 1797 int 1798 nfsvno_statfs(struct vnode *vp, struct statfs *sf) 1799 { 1800 struct statfs *tsf; 1801 int error; 1802 1803 tsf = NULL; 1804 if (nfsrv_devidcnt > 0) { 1805 /* For a pNFS service, get the DS numbers. */ 1806 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); 1807 error = nfsrv_pnfsstatfs(tsf, vp->v_mount); 1808 if (error != 0) { 1809 free(tsf, M_TEMP); 1810 tsf = NULL; 1811 } 1812 } 1813 error = VFS_STATFS(vp->v_mount, sf); 1814 if (error == 0) { 1815 if (tsf != NULL) { 1816 sf->f_blocks = tsf->f_blocks; 1817 sf->f_bavail = tsf->f_bavail; 1818 sf->f_bfree = tsf->f_bfree; 1819 sf->f_bsize = tsf->f_bsize; 1820 } 1821 /* 1822 * Since NFS handles these values as unsigned on the 1823 * wire, there is no way to represent negative values, 1824 * so set them to 0. Without this, they will appear 1825 * to be very large positive values for clients like 1826 * Solaris10. 1827 */ 1828 if (sf->f_bavail < 0) 1829 sf->f_bavail = 0; 1830 if (sf->f_ffree < 0) 1831 sf->f_ffree = 0; 1832 } 1833 free(tsf, M_TEMP); 1834 NFSEXITCODE(error); 1835 return (error); 1836 } 1837 1838 /* 1839 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but 1840 * must handle nfsrv_opencheck() calls after any other access checks. 1841 */ 1842 void 1843 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, 1844 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, 1845 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, 1846 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, bool done_namei, 1847 struct nfsexstuff *exp, struct vnode **vpp) 1848 { 1849 struct vnode *vp = NULL; 1850 u_quad_t tempsize; 1851 struct nfsexstuff nes; 1852 struct thread *p = curthread; 1853 uint32_t oldrepstat; 1854 1855 if (ndp->ni_vp == NULL) { 1856 /* 1857 * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be 1858 * set true, since cleanup after nfsvno_namei() is needed. 1859 */ 1860 oldrepstat = nd->nd_repstat; 1861 nd->nd_repstat = nfsrv_opencheck(clientid, 1862 stateidp, stp, NULL, nd, p, nd->nd_repstat); 1863 if (nd->nd_repstat != 0 && oldrepstat == 0) 1864 done_namei = true; 1865 } 1866 if (!nd->nd_repstat) { 1867 if (ndp->ni_vp == NULL) { 1868 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, 1869 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1870 /* For a pNFS server, create the data file on a DS. */ 1871 if (nd->nd_repstat == 0) { 1872 /* 1873 * Create a data file on a DS for a pNFS server. 1874 * This function just returns if not 1875 * running a pNFS DS or the creation fails. 1876 */ 1877 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1878 cred, p); 1879 } 1880 VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ? 1881 &ndp->ni_vp : NULL, false); 1882 nfsvno_relpathbuf(ndp); 1883 if (!nd->nd_repstat) { 1884 if (*exclusive_flagp) { 1885 *exclusive_flagp = 0; 1886 NFSVNO_ATTRINIT(nvap); 1887 nvap->na_atime.tv_sec = cverf[0]; 1888 nvap->na_atime.tv_nsec = cverf[1]; 1889 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, 1890 &nvap->na_vattr, cred); 1891 if (nd->nd_repstat != 0) { 1892 vput(ndp->ni_vp); 1893 ndp->ni_vp = NULL; 1894 nd->nd_repstat = NFSERR_NOTSUPP; 1895 } else 1896 NFSSETBIT_ATTRBIT(attrbitp, 1897 NFSATTRBIT_TIMEACCESS); 1898 } else { 1899 nfsrv_fixattr(nd, ndp->ni_vp, nvap, 1900 aclp, p, attrbitp, exp); 1901 } 1902 } 1903 vp = ndp->ni_vp; 1904 } else { 1905 nfsvno_relpathbuf(ndp); 1906 vp = ndp->ni_vp; 1907 if (create == NFSV4OPEN_CREATE) { 1908 if (ndp->ni_dvp == vp) 1909 vrele(ndp->ni_dvp); 1910 else 1911 vput(ndp->ni_dvp); 1912 } 1913 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { 1914 if (ndp->ni_cnd.cn_flags & RDONLY) 1915 NFSVNO_SETEXRDONLY(&nes); 1916 else 1917 NFSVNO_EXINIT(&nes); 1918 nd->nd_repstat = nfsvno_accchk(vp, 1919 VWRITE, cred, &nes, p, 1920 NFSACCCHK_NOOVERRIDE, 1921 NFSACCCHK_VPISLOCKED, NULL); 1922 nd->nd_repstat = nfsrv_opencheck(clientid, 1923 stateidp, stp, vp, nd, p, nd->nd_repstat); 1924 if (!nd->nd_repstat) { 1925 tempsize = nvap->na_size; 1926 NFSVNO_ATTRINIT(nvap); 1927 nvap->na_size = tempsize; 1928 nd->nd_repstat = nfsvno_setattr(vp, 1929 nvap, cred, p, exp); 1930 } 1931 } else if (vp->v_type == VREG) { 1932 nd->nd_repstat = nfsrv_opencheck(clientid, 1933 stateidp, stp, vp, nd, p, nd->nd_repstat); 1934 } 1935 } 1936 } else if (done_namei) { 1937 KASSERT(create == NFSV4OPEN_CREATE, 1938 ("nfsvno_open: not create")); 1939 /* 1940 * done_namei is set when nfsvno_namei() has completed 1941 * successfully, but a subsequent error was set in 1942 * nd_repstat. As such, cleanup of the nfsvno_namei() 1943 * results is required. 1944 */ 1945 nfsvno_relpathbuf(ndp); 1946 if (ndp->ni_dvp == ndp->ni_vp) 1947 vrele(ndp->ni_dvp); 1948 else 1949 vput(ndp->ni_dvp); 1950 if (ndp->ni_vp) 1951 vput(ndp->ni_vp); 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 uint64_t *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 txdr_hyper(dp->d_fileno, tl); 2253 } else { 2254 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2255 *tl++ = newnfs_true; 2256 *tl = txdr_unsigned(dp->d_fileno); 2257 } 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 txdr_hyper(*cookiep, tl); 2262 } else { 2263 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 2264 *tl = txdr_unsigned(*cookiep); 2265 } 2266 } 2267 cpos += dp->d_reclen; 2268 dp = (struct dirent *)cpos; 2269 cookiep++; 2270 ncookies--; 2271 } 2272 if (cpos < cend) 2273 eofflag = 0; 2274 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2275 *tl++ = newnfs_false; 2276 if (eofflag) 2277 *tl = newnfs_true; 2278 else 2279 *tl = newnfs_false; 2280 free(rbuf, M_TEMP); 2281 free(cookies, M_TEMP); 2282 2283 out: 2284 NFSEXITCODE2(0, nd); 2285 return (0); 2286 nfsmout: 2287 vput(vp); 2288 NFSEXITCODE2(error, nd); 2289 return (error); 2290 } 2291 2292 /* 2293 * Readdirplus for V3 and Readdir for V4. 2294 */ 2295 int 2296 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, 2297 struct vnode *vp, struct nfsexstuff *exp) 2298 { 2299 struct dirent *dp; 2300 u_int32_t *tl; 2301 int dirlen; 2302 char *cpos, *cend, *rbuf; 2303 struct vnode *nvp; 2304 fhandle_t nfh; 2305 struct nfsvattr nva, at, *nvap = &nva; 2306 struct mbuf *mb0, *mb1; 2307 struct nfsreferral *refp; 2308 int nlen, r, error = 0, getret = 1, usevget = 1; 2309 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; 2310 caddr_t bpos0, bpos1; 2311 u_int64_t off, toff, verf __unused; 2312 uint64_t *cookies = NULL, *cookiep; 2313 nfsattrbit_t attrbits, rderrbits, savbits; 2314 struct uio io; 2315 struct iovec iv; 2316 struct componentname cn; 2317 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; 2318 struct mount *mp, *new_mp; 2319 uint64_t mounted_on_fileno; 2320 struct thread *p = curthread; 2321 int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1; 2322 2323 if (nd->nd_repstat) { 2324 nfsrv_postopattr(nd, getret, &at); 2325 goto out; 2326 } 2327 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2328 off = fxdr_hyper(tl); 2329 toff = off; 2330 tl += 2; 2331 verf = fxdr_hyper(tl); 2332 tl += 2; 2333 siz = fxdr_unsigned(int, *tl++); 2334 cnt = fxdr_unsigned(int, *tl); 2335 2336 /* 2337 * Use the server's maximum data transfer size as the upper bound 2338 * on reply datalen. 2339 */ 2340 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2341 cnt = NFS_SRVMAXDATA(nd); 2342 2343 /* 2344 * siz is a "hint" of how much directory information (name, fileid, 2345 * cookie) should be in the reply. At least one client "hints" 0, 2346 * so I set it to cnt for that case. I also round it up to the 2347 * next multiple of DIRBLKSIZ. 2348 * Since the size of a Readdirplus directory entry reply will always 2349 * be greater than a directory entry returned by VOP_READDIR(), it 2350 * does not make sense to read more than NFS_SRVMAXDATA() via 2351 * VOP_READDIR(). 2352 */ 2353 if (siz <= 0) 2354 siz = cnt; 2355 else if (siz > NFS_SRVMAXDATA(nd)) 2356 siz = NFS_SRVMAXDATA(nd); 2357 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2358 2359 if (nd->nd_flag & ND_NFSV4) { 2360 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 2361 if (error) 2362 goto nfsmout; 2363 NFSSET_ATTRBIT(&savbits, &attrbits); 2364 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); 2365 NFSZERO_ATTRBIT(&rderrbits); 2366 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); 2367 } else { 2368 NFSZERO_ATTRBIT(&attrbits); 2369 } 2370 fullsiz = siz; 2371 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2372 #if 0 2373 if (!nd->nd_repstat) { 2374 if (off && verf != at.na_filerev) { 2375 /* 2376 * va_filerev is not sufficient as a cookie verifier, 2377 * since it is not supposed to change when entries are 2378 * removed/added unless that offset cookies returned to 2379 * the client are no longer valid. 2380 */ 2381 if (nd->nd_flag & ND_NFSV4) { 2382 nd->nd_repstat = NFSERR_NOTSAME; 2383 } else { 2384 nd->nd_repstat = NFSERR_BAD_COOKIE; 2385 } 2386 } 2387 } 2388 #endif 2389 if (!nd->nd_repstat && vp->v_type != VDIR) 2390 nd->nd_repstat = NFSERR_NOTDIR; 2391 if (!nd->nd_repstat && cnt == 0) 2392 nd->nd_repstat = NFSERR_TOOSMALL; 2393 if (!nd->nd_repstat) 2394 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2395 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2396 NFSACCCHK_VPISLOCKED, NULL); 2397 if (nd->nd_repstat) { 2398 vput(vp); 2399 if (nd->nd_flag & ND_NFSV3) 2400 nfsrv_postopattr(nd, getret, &at); 2401 goto out; 2402 } 2403 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2404 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; 2405 2406 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2407 again: 2408 eofflag = 0; 2409 if (cookies) { 2410 free(cookies, M_TEMP); 2411 cookies = NULL; 2412 } 2413 2414 iv.iov_base = rbuf; 2415 iv.iov_len = siz; 2416 io.uio_iov = &iv; 2417 io.uio_iovcnt = 1; 2418 io.uio_offset = (off_t)off; 2419 io.uio_resid = siz; 2420 io.uio_segflg = UIO_SYSSPACE; 2421 io.uio_rw = UIO_READ; 2422 io.uio_td = NULL; 2423 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2424 &cookies); 2425 off = (u_int64_t)io.uio_offset; 2426 if (io.uio_resid) 2427 siz -= io.uio_resid; 2428 2429 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2430 2431 if (!cookies && !nd->nd_repstat) 2432 nd->nd_repstat = NFSERR_PERM; 2433 if (!nd->nd_repstat) 2434 nd->nd_repstat = getret; 2435 if (nd->nd_repstat) { 2436 vput(vp); 2437 if (cookies) 2438 free(cookies, M_TEMP); 2439 free(rbuf, M_TEMP); 2440 if (nd->nd_flag & ND_NFSV3) 2441 nfsrv_postopattr(nd, getret, &at); 2442 goto out; 2443 } 2444 /* 2445 * If nothing read, return eof 2446 * rpc reply 2447 */ 2448 if (siz == 0) { 2449 vput(vp); 2450 if (nd->nd_flag & ND_NFSV3) 2451 nfsrv_postopattr(nd, getret, &at); 2452 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2453 txdr_hyper(at.na_filerev, tl); 2454 tl += 2; 2455 *tl++ = newnfs_false; 2456 *tl = newnfs_true; 2457 free(cookies, M_TEMP); 2458 free(rbuf, M_TEMP); 2459 goto out; 2460 } 2461 2462 /* 2463 * Check for degenerate cases of nothing useful read. 2464 * If so go try again 2465 */ 2466 cpos = rbuf; 2467 cend = rbuf + siz; 2468 dp = (struct dirent *)cpos; 2469 cookiep = cookies; 2470 2471 /* 2472 * For some reason FreeBSD's ufs_readdir() chooses to back the 2473 * directory offset up to a block boundary, so it is necessary to 2474 * skip over the records that precede the requested offset. This 2475 * requires the assumption that file offset cookies monotonically 2476 * increase. 2477 */ 2478 while (cpos < cend && ncookies > 0 && 2479 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2480 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || 2481 ((nd->nd_flag & ND_NFSV4) && 2482 ((dp->d_namlen == 1 && dp->d_name[0] == '.') || 2483 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { 2484 cpos += dp->d_reclen; 2485 dp = (struct dirent *)cpos; 2486 cookiep++; 2487 ncookies--; 2488 } 2489 if (cpos >= cend || ncookies == 0) { 2490 siz = fullsiz; 2491 toff = off; 2492 goto again; 2493 } 2494 2495 /* 2496 * Busy the file system so that the mount point won't go away 2497 * and, as such, VFS_VGET() can be used safely. 2498 */ 2499 mp = vp->v_mount; 2500 vfs_ref(mp); 2501 NFSVOPUNLOCK(vp); 2502 nd->nd_repstat = vfs_busy(mp, 0); 2503 vfs_rel(mp); 2504 if (nd->nd_repstat != 0) { 2505 vrele(vp); 2506 free(cookies, M_TEMP); 2507 free(rbuf, M_TEMP); 2508 if (nd->nd_flag & ND_NFSV3) 2509 nfsrv_postopattr(nd, getret, &at); 2510 goto out; 2511 } 2512 2513 /* 2514 * Check to see if entries in this directory can be safely acquired 2515 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. 2516 * ZFS snapshot directories need VOP_LOOKUP(), so that any 2517 * automount of the snapshot directory that is required will 2518 * be done. 2519 * This needs to be done here for NFSv4, since NFSv4 never does 2520 * a VFS_VGET() for "." or "..". 2521 */ 2522 if (is_zfs == 1) { 2523 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); 2524 if (r == EOPNOTSUPP) { 2525 usevget = 0; 2526 cn.cn_nameiop = LOOKUP; 2527 cn.cn_lkflags = LK_SHARED | LK_RETRY; 2528 cn.cn_cred = nd->nd_cred; 2529 } else if (r == 0) 2530 vput(nvp); 2531 } 2532 2533 /* 2534 * If the reply is likely to exceed MCLBYTES and the reply will 2535 * not be saved, use ext_pgs mbufs for TLS. 2536 * It is difficult to predict how large each entry will be and 2537 * how many entries have been read, so just assume the directory 2538 * entries grow by a factor of 4 when attributes are included. 2539 * For NFSv4.0, we do not know for sure if the reply will 2540 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2541 */ 2542 if (cnt > MCLBYTES && siz > MCLBYTES / 4 && 2543 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2544 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2545 nd->nd_flag |= ND_EXTPG; 2546 2547 /* 2548 * Save this position, in case there is an error before one entry 2549 * is created. 2550 */ 2551 mb0 = nd->nd_mb; 2552 bpos0 = nd->nd_bpos; 2553 bextpg0 = nd->nd_bextpg; 2554 bextpgsiz0 = nd->nd_bextpgsiz; 2555 2556 /* 2557 * Fill in the first part of the reply. 2558 * dirlen is the reply length in bytes and cannot exceed cnt. 2559 * (Include the two booleans at the end of the reply in dirlen now, 2560 * so we recognize when we have exceeded cnt.) 2561 */ 2562 if (nd->nd_flag & ND_NFSV3) { 2563 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2564 nfsrv_postopattr(nd, getret, &at); 2565 } else { 2566 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; 2567 } 2568 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); 2569 txdr_hyper(at.na_filerev, tl); 2570 2571 /* 2572 * Save this position, in case there is an empty reply needed. 2573 */ 2574 mb1 = nd->nd_mb; 2575 bpos1 = nd->nd_bpos; 2576 bextpg1 = nd->nd_bextpg; 2577 bextpgsiz1 = nd->nd_bextpgsiz; 2578 2579 /* Loop through the records and build reply */ 2580 entrycnt = 0; 2581 while (cpos < cend && ncookies > 0 && dirlen < cnt) { 2582 nlen = dp->d_namlen; 2583 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2584 nlen <= NFS_MAXNAMLEN && 2585 ((nd->nd_flag & ND_NFSV3) || nlen > 2 || 2586 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) 2587 || (nlen == 1 && dp->d_name[0] != '.'))) { 2588 /* 2589 * Save the current position in the reply, in case 2590 * this entry exceeds cnt. 2591 */ 2592 mb1 = nd->nd_mb; 2593 bpos1 = nd->nd_bpos; 2594 bextpg1 = nd->nd_bextpg; 2595 bextpgsiz1 = nd->nd_bextpgsiz; 2596 2597 /* 2598 * For readdir_and_lookup get the vnode using 2599 * the file number. 2600 */ 2601 nvp = NULL; 2602 refp = NULL; 2603 r = 0; 2604 at_root = 0; 2605 needs_unbusy = 0; 2606 new_mp = mp; 2607 mounted_on_fileno = (uint64_t)dp->d_fileno; 2608 if ((nd->nd_flag & ND_NFSV3) || 2609 NFSNONZERO_ATTRBIT(&savbits)) { 2610 if (nd->nd_flag & ND_NFSV4) 2611 refp = nfsv4root_getreferral(NULL, 2612 vp, dp->d_fileno); 2613 if (refp == NULL) { 2614 if (usevget) 2615 r = VFS_VGET(mp, dp->d_fileno, 2616 LK_SHARED, &nvp); 2617 else 2618 r = EOPNOTSUPP; 2619 if (r == EOPNOTSUPP) { 2620 if (usevget) { 2621 usevget = 0; 2622 cn.cn_nameiop = LOOKUP; 2623 cn.cn_lkflags = 2624 LK_SHARED | 2625 LK_RETRY; 2626 cn.cn_cred = 2627 nd->nd_cred; 2628 } 2629 cn.cn_nameptr = dp->d_name; 2630 cn.cn_namelen = nlen; 2631 cn.cn_flags = ISLASTCN | 2632 NOFOLLOW | LOCKLEAF; 2633 if (nlen == 2 && 2634 dp->d_name[0] == '.' && 2635 dp->d_name[1] == '.') 2636 cn.cn_flags |= 2637 ISDOTDOT; 2638 if (NFSVOPLOCK(vp, LK_SHARED) 2639 != 0) { 2640 nd->nd_repstat = EPERM; 2641 break; 2642 } 2643 if ((vp->v_vflag & VV_ROOT) != 0 2644 && (cn.cn_flags & ISDOTDOT) 2645 != 0) { 2646 vref(vp); 2647 nvp = vp; 2648 r = 0; 2649 } else { 2650 r = VOP_LOOKUP(vp, &nvp, 2651 &cn); 2652 if (vp != nvp) 2653 NFSVOPUNLOCK(vp); 2654 } 2655 } 2656 2657 /* 2658 * For NFSv4, check to see if nvp is 2659 * a mount point and get the mount 2660 * point vnode, as required. 2661 */ 2662 if (r == 0 && 2663 nfsrv_enable_crossmntpt != 0 && 2664 (nd->nd_flag & ND_NFSV4) != 0 && 2665 nvp->v_type == VDIR && 2666 nvp->v_mountedhere != NULL) { 2667 new_mp = nvp->v_mountedhere; 2668 r = vfs_busy(new_mp, 0); 2669 vput(nvp); 2670 nvp = NULL; 2671 if (r == 0) { 2672 r = VFS_ROOT(new_mp, 2673 LK_SHARED, &nvp); 2674 needs_unbusy = 1; 2675 if (r == 0) 2676 at_root = 1; 2677 } 2678 } 2679 } 2680 2681 /* 2682 * If we failed to look up the entry, then it 2683 * has become invalid, most likely removed. 2684 */ 2685 if (r != 0) { 2686 if (needs_unbusy) 2687 vfs_unbusy(new_mp); 2688 goto invalid; 2689 } 2690 KASSERT(refp != NULL || nvp != NULL, 2691 ("%s: undetected lookup error", __func__)); 2692 2693 if (refp == NULL && 2694 ((nd->nd_flag & ND_NFSV3) || 2695 NFSNONZERO_ATTRBIT(&attrbits))) { 2696 r = nfsvno_getfh(nvp, &nfh, p); 2697 if (!r) 2698 r = nfsvno_getattr(nvp, nvap, nd, p, 2699 1, &attrbits); 2700 if (r == 0 && is_zfs == 1 && 2701 nfsrv_enable_crossmntpt != 0 && 2702 (nd->nd_flag & ND_NFSV4) != 0 && 2703 nvp->v_type == VDIR && 2704 vp->v_mount != nvp->v_mount) { 2705 /* 2706 * For a ZFS snapshot, there is a 2707 * pseudo mount that does not set 2708 * v_mountedhere, so it needs to 2709 * be detected via a different 2710 * mount structure. 2711 */ 2712 at_root = 1; 2713 if (new_mp == mp) 2714 new_mp = nvp->v_mount; 2715 } 2716 } 2717 2718 /* 2719 * If we failed to get attributes of the entry, 2720 * then just skip it for NFSv3 (the traditional 2721 * behavior in the old NFS server). 2722 * For NFSv4 the behavior is controlled by 2723 * RDATTRERROR: we either ignore the error or 2724 * fail the request. 2725 * The exception is EOPNOTSUPP, which can be 2726 * returned by nfsvno_getfh() for certain 2727 * file systems, such as devfs. This indicates 2728 * that the file system cannot be exported, 2729 * so just skip over the entry. 2730 * Note that RDATTRERROR is never set for NFSv3. 2731 */ 2732 if (r != 0) { 2733 if (!NFSISSET_ATTRBIT(&attrbits, 2734 NFSATTRBIT_RDATTRERROR) || 2735 r == EOPNOTSUPP) { 2736 vput(nvp); 2737 if (needs_unbusy != 0) 2738 vfs_unbusy(new_mp); 2739 if ((nd->nd_flag & ND_NFSV3) || 2740 r == EOPNOTSUPP) 2741 goto invalid; 2742 nd->nd_repstat = r; 2743 break; 2744 } 2745 } 2746 } 2747 2748 /* 2749 * Build the directory record xdr 2750 */ 2751 if (nd->nd_flag & ND_NFSV3) { 2752 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2753 *tl++ = newnfs_true; 2754 txdr_hyper(dp->d_fileno, tl); 2755 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2756 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2757 txdr_hyper(*cookiep, tl); 2758 nfsrv_postopattr(nd, 0, nvap); 2759 dirlen += nfsm_fhtom(NULL, nd, (u_int8_t *)&nfh, 2760 0, 1); 2761 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); 2762 if (nvp != NULL) 2763 vput(nvp); 2764 } else { 2765 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2766 *tl++ = newnfs_true; 2767 txdr_hyper(*cookiep, tl); 2768 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2769 if (nvp != NULL) { 2770 supports_nfsv4acls = 2771 nfs_supportsnfsv4acls(nvp); 2772 NFSVOPUNLOCK(nvp); 2773 } else 2774 supports_nfsv4acls = 0; 2775 if (refp != NULL) { 2776 dirlen += nfsrv_putreferralattr(nd, 2777 &savbits, refp, 0, 2778 &nd->nd_repstat); 2779 if (nd->nd_repstat) { 2780 if (nvp != NULL) 2781 vrele(nvp); 2782 if (needs_unbusy != 0) 2783 vfs_unbusy(new_mp); 2784 break; 2785 } 2786 } else if (r) { 2787 dirlen += nfsvno_fillattr(nd, new_mp, 2788 nvp, nvap, &nfh, r, &rderrbits, 2789 nd->nd_cred, p, isdgram, 0, 2790 supports_nfsv4acls, at_root, 2791 mounted_on_fileno); 2792 } else { 2793 dirlen += nfsvno_fillattr(nd, new_mp, 2794 nvp, nvap, &nfh, r, &attrbits, 2795 nd->nd_cred, p, isdgram, 0, 2796 supports_nfsv4acls, at_root, 2797 mounted_on_fileno); 2798 } 2799 if (nvp != NULL) 2800 vrele(nvp); 2801 dirlen += (3 * NFSX_UNSIGNED); 2802 } 2803 if (needs_unbusy != 0) 2804 vfs_unbusy(new_mp); 2805 if (dirlen <= cnt) 2806 entrycnt++; 2807 } 2808 invalid: 2809 cpos += dp->d_reclen; 2810 dp = (struct dirent *)cpos; 2811 cookiep++; 2812 ncookies--; 2813 } 2814 vrele(vp); 2815 vfs_unbusy(mp); 2816 2817 /* 2818 * If dirlen > cnt, we must strip off the last entry. If that 2819 * results in an empty reply, report NFSERR_TOOSMALL. 2820 */ 2821 if (dirlen > cnt || nd->nd_repstat) { 2822 if (!nd->nd_repstat && entrycnt == 0) 2823 nd->nd_repstat = NFSERR_TOOSMALL; 2824 if (nd->nd_repstat) { 2825 nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); 2826 if (nd->nd_flag & ND_NFSV3) 2827 nfsrv_postopattr(nd, getret, &at); 2828 } else 2829 nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); 2830 eofflag = 0; 2831 } else if (cpos < cend) 2832 eofflag = 0; 2833 if (!nd->nd_repstat) { 2834 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2835 *tl++ = newnfs_false; 2836 if (eofflag) 2837 *tl = newnfs_true; 2838 else 2839 *tl = newnfs_false; 2840 } 2841 free(cookies, M_TEMP); 2842 free(rbuf, M_TEMP); 2843 2844 out: 2845 NFSEXITCODE2(0, nd); 2846 return (0); 2847 nfsmout: 2848 vput(vp); 2849 NFSEXITCODE2(error, nd); 2850 return (error); 2851 } 2852 2853 /* 2854 * Get the settable attributes out of the mbuf list. 2855 * (Return 0 or EBADRPC) 2856 */ 2857 int 2858 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2859 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2860 { 2861 u_int32_t *tl; 2862 struct nfsv2_sattr *sp; 2863 int error = 0, toclient = 0; 2864 2865 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 2866 case ND_NFSV2: 2867 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2868 /* 2869 * Some old clients didn't fill in the high order 16bits. 2870 * --> check the low order 2 bytes for 0xffff 2871 */ 2872 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 2873 nvap->na_mode = nfstov_mode(sp->sa_mode); 2874 if (sp->sa_uid != newnfs_xdrneg1) 2875 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 2876 if (sp->sa_gid != newnfs_xdrneg1) 2877 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 2878 if (sp->sa_size != newnfs_xdrneg1) 2879 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 2880 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 2881 #ifdef notyet 2882 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 2883 #else 2884 nvap->na_atime.tv_sec = 2885 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 2886 nvap->na_atime.tv_nsec = 0; 2887 #endif 2888 } 2889 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 2890 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 2891 break; 2892 case ND_NFSV3: 2893 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2894 if (*tl == newnfs_true) { 2895 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2896 nvap->na_mode = nfstov_mode(*tl); 2897 } 2898 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2899 if (*tl == newnfs_true) { 2900 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2901 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 2902 } 2903 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2904 if (*tl == newnfs_true) { 2905 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2906 nvap->na_gid = fxdr_unsigned(gid_t, *tl); 2907 } 2908 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2909 if (*tl == newnfs_true) { 2910 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2911 nvap->na_size = fxdr_hyper(tl); 2912 } 2913 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2914 switch (fxdr_unsigned(int, *tl)) { 2915 case NFSV3SATTRTIME_TOCLIENT: 2916 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2917 fxdr_nfsv3time(tl, &nvap->na_atime); 2918 toclient = 1; 2919 break; 2920 case NFSV3SATTRTIME_TOSERVER: 2921 vfs_timestamp(&nvap->na_atime); 2922 nvap->na_vaflags |= VA_UTIMES_NULL; 2923 break; 2924 } 2925 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2926 switch (fxdr_unsigned(int, *tl)) { 2927 case NFSV3SATTRTIME_TOCLIENT: 2928 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2929 fxdr_nfsv3time(tl, &nvap->na_mtime); 2930 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2931 break; 2932 case NFSV3SATTRTIME_TOSERVER: 2933 vfs_timestamp(&nvap->na_mtime); 2934 if (!toclient) 2935 nvap->na_vaflags |= VA_UTIMES_NULL; 2936 break; 2937 } 2938 break; 2939 case ND_NFSV4: 2940 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); 2941 } 2942 nfsmout: 2943 NFSEXITCODE2(error, nd); 2944 return (error); 2945 } 2946 2947 /* 2948 * Handle the setable attributes for V4. 2949 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 2950 */ 2951 int 2952 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 2953 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2954 { 2955 u_int32_t *tl; 2956 int attrsum = 0; 2957 int i, j; 2958 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 2959 int moderet, toclient = 0; 2960 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 2961 uid_t uid; 2962 gid_t gid; 2963 u_short mode, mask; /* Same type as va_mode. */ 2964 struct vattr va; 2965 2966 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 2967 if (error) 2968 goto nfsmout; 2969 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2970 attrsize = fxdr_unsigned(int, *tl); 2971 2972 /* 2973 * Loop around getting the setable attributes. If an unsupported 2974 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 2975 */ 2976 if (retnotsup) { 2977 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2978 bitpos = NFSATTRBIT_MAX; 2979 } else { 2980 bitpos = 0; 2981 } 2982 moderet = 0; 2983 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 2984 if (attrsum > attrsize) { 2985 error = NFSERR_BADXDR; 2986 goto nfsmout; 2987 } 2988 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 2989 switch (bitpos) { 2990 case NFSATTRBIT_SIZE: 2991 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 2992 if (vp != NULL && vp->v_type != VREG) { 2993 error = (vp->v_type == VDIR) ? NFSERR_ISDIR : 2994 NFSERR_INVAL; 2995 goto nfsmout; 2996 } 2997 nvap->na_size = fxdr_hyper(tl); 2998 attrsum += NFSX_HYPER; 2999 break; 3000 case NFSATTRBIT_ACL: 3001 error = nfsrv_dissectacl(nd, aclp, true, &aceerr, 3002 &aclsize, p); 3003 if (error) 3004 goto nfsmout; 3005 if (aceerr && !nd->nd_repstat) 3006 nd->nd_repstat = aceerr; 3007 attrsum += aclsize; 3008 break; 3009 case NFSATTRBIT_ARCHIVE: 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_HIDDEN: 3016 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3017 if (!nd->nd_repstat) 3018 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3019 attrsum += NFSX_UNSIGNED; 3020 break; 3021 case NFSATTRBIT_MIMETYPE: 3022 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3023 i = fxdr_unsigned(int, *tl); 3024 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 3025 if (error) 3026 goto nfsmout; 3027 if (!nd->nd_repstat) 3028 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3029 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 3030 break; 3031 case NFSATTRBIT_MODE: 3032 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ 3033 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3034 nvap->na_mode = nfstov_mode(*tl); 3035 attrsum += NFSX_UNSIGNED; 3036 break; 3037 case NFSATTRBIT_OWNER: 3038 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3039 j = fxdr_unsigned(int, *tl); 3040 if (j < 0) { 3041 error = NFSERR_BADXDR; 3042 goto nfsmout; 3043 } 3044 if (j > NFSV4_SMALLSTR) 3045 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3046 else 3047 cp = namestr; 3048 error = nfsrv_mtostr(nd, cp, j); 3049 if (error) { 3050 if (j > NFSV4_SMALLSTR) 3051 free(cp, M_NFSSTRING); 3052 goto nfsmout; 3053 } 3054 if (!nd->nd_repstat) { 3055 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, 3056 &uid); 3057 if (!nd->nd_repstat) 3058 nvap->na_uid = uid; 3059 } 3060 if (j > NFSV4_SMALLSTR) 3061 free(cp, M_NFSSTRING); 3062 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3063 break; 3064 case NFSATTRBIT_OWNERGROUP: 3065 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3066 j = fxdr_unsigned(int, *tl); 3067 if (j < 0) { 3068 error = NFSERR_BADXDR; 3069 goto nfsmout; 3070 } 3071 if (j > NFSV4_SMALLSTR) 3072 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3073 else 3074 cp = namestr; 3075 error = nfsrv_mtostr(nd, cp, j); 3076 if (error) { 3077 if (j > NFSV4_SMALLSTR) 3078 free(cp, M_NFSSTRING); 3079 goto nfsmout; 3080 } 3081 if (!nd->nd_repstat) { 3082 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, 3083 &gid); 3084 if (!nd->nd_repstat) 3085 nvap->na_gid = gid; 3086 } 3087 if (j > NFSV4_SMALLSTR) 3088 free(cp, M_NFSSTRING); 3089 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3090 break; 3091 case NFSATTRBIT_SYSTEM: 3092 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3093 if (!nd->nd_repstat) 3094 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3095 attrsum += NFSX_UNSIGNED; 3096 break; 3097 case NFSATTRBIT_TIMEACCESSSET: 3098 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3099 attrsum += NFSX_UNSIGNED; 3100 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3101 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3102 fxdr_nfsv4time(tl, &nvap->na_atime); 3103 toclient = 1; 3104 attrsum += NFSX_V4TIME; 3105 } else { 3106 vfs_timestamp(&nvap->na_atime); 3107 nvap->na_vaflags |= VA_UTIMES_NULL; 3108 } 3109 break; 3110 case NFSATTRBIT_TIMEBACKUP: 3111 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3112 if (!nd->nd_repstat) 3113 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3114 attrsum += NFSX_V4TIME; 3115 break; 3116 case NFSATTRBIT_TIMECREATE: 3117 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3118 fxdr_nfsv4time(tl, &nvap->na_btime); 3119 attrsum += NFSX_V4TIME; 3120 break; 3121 case NFSATTRBIT_TIMEMODIFYSET: 3122 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3123 attrsum += NFSX_UNSIGNED; 3124 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3125 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3126 fxdr_nfsv4time(tl, &nvap->na_mtime); 3127 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3128 attrsum += NFSX_V4TIME; 3129 } else { 3130 vfs_timestamp(&nvap->na_mtime); 3131 if (!toclient) 3132 nvap->na_vaflags |= VA_UTIMES_NULL; 3133 } 3134 break; 3135 case NFSATTRBIT_MODESETMASKED: 3136 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3137 mode = fxdr_unsigned(u_short, *tl++); 3138 mask = fxdr_unsigned(u_short, *tl); 3139 /* 3140 * vp == NULL implies an Open/Create operation. 3141 * This attribute can only be used for Setattr and 3142 * only for NFSv4.1 or higher. 3143 * If moderet != 0, a mode attribute has also been 3144 * specified and this attribute cannot be done in the 3145 * same Setattr operation. 3146 */ 3147 if ((nd->nd_flag & ND_NFSV41) == 0) 3148 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3149 else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 || 3150 vp == NULL) 3151 nd->nd_repstat = NFSERR_INVAL; 3152 else if (moderet == 0) 3153 moderet = VOP_GETATTR(vp, &va, nd->nd_cred); 3154 if (moderet == 0) 3155 nvap->na_mode = (mode & mask) | 3156 (va.va_mode & ~mask); 3157 else 3158 nd->nd_repstat = moderet; 3159 attrsum += 2 * NFSX_UNSIGNED; 3160 break; 3161 default: 3162 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3163 /* 3164 * set bitpos so we drop out of the loop. 3165 */ 3166 bitpos = NFSATTRBIT_MAX; 3167 break; 3168 } 3169 } 3170 3171 /* 3172 * some clients pad the attrlist, so we need to skip over the 3173 * padding. 3174 */ 3175 if (attrsum > attrsize) { 3176 error = NFSERR_BADXDR; 3177 } else { 3178 attrsize = NFSM_RNDUP(attrsize); 3179 if (attrsum < attrsize) 3180 error = nfsm_advance(nd, attrsize - attrsum, -1); 3181 } 3182 nfsmout: 3183 NFSEXITCODE2(error, nd); 3184 return (error); 3185 } 3186 3187 /* 3188 * Check/setup export credentials. 3189 */ 3190 int 3191 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 3192 struct ucred *credanon, bool testsec) 3193 { 3194 int error; 3195 3196 /* 3197 * Check/setup credentials. 3198 */ 3199 if (nd->nd_flag & ND_GSS) 3200 exp->nes_exflag &= ~MNT_EXPORTANON; 3201 3202 /* 3203 * Check to see if the operation is allowed for this security flavor. 3204 */ 3205 error = 0; 3206 if (testsec) { 3207 error = nfsvno_testexp(nd, exp); 3208 if (error != 0) 3209 goto out; 3210 } 3211 3212 /* 3213 * Check to see if the file system is exported V4 only. 3214 */ 3215 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 3216 error = NFSERR_PROGNOTV4; 3217 goto out; 3218 } 3219 3220 /* 3221 * Now, map the user credentials. 3222 * (Note that ND_AUTHNONE will only be set for an NFSv3 3223 * Fsinfo RPC. If set for anything else, this code might need 3224 * to change.) 3225 */ 3226 if (NFSVNO_EXPORTED(exp)) { 3227 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || 3228 NFSVNO_EXPORTANON(exp) || 3229 (nd->nd_flag & ND_AUTHNONE) != 0) { 3230 nd->nd_cred->cr_uid = credanon->cr_uid; 3231 nd->nd_cred->cr_gid = credanon->cr_gid; 3232 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 3233 credanon->cr_groups); 3234 } else if ((nd->nd_flag & ND_GSS) == 0) { 3235 /* 3236 * If using AUTH_SYS, call nfsrv_getgrpscred() to see 3237 * if there is a replacement credential with a group 3238 * list set up by "nfsuserd -manage-gids". 3239 * If there is no replacement, nfsrv_getgrpscred() 3240 * simply returns its argument. 3241 */ 3242 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); 3243 } 3244 } 3245 3246 out: 3247 NFSEXITCODE2(error, nd); 3248 return (error); 3249 } 3250 3251 /* 3252 * Check exports. 3253 */ 3254 int 3255 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 3256 struct ucred **credp) 3257 { 3258 int error; 3259 3260 error = 0; 3261 *credp = NULL; 3262 MNT_ILOCK(mp); 3263 if (mp->mnt_exjail == NULL || 3264 mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison) 3265 error = EACCES; 3266 MNT_IUNLOCK(mp); 3267 if (error == 0) 3268 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3269 &exp->nes_numsecflavor, exp->nes_secflavors); 3270 if (error) { 3271 if (NFSD_VNET(nfs_rootfhset)) { 3272 exp->nes_exflag = 0; 3273 exp->nes_numsecflavor = 0; 3274 error = 0; 3275 } 3276 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3277 MAXSECFLAVORS) { 3278 printf("nfsvno_checkexp: numsecflavors out of range\n"); 3279 exp->nes_numsecflavor = 0; 3280 error = EACCES; 3281 } 3282 NFSEXITCODE(error); 3283 return (error); 3284 } 3285 3286 /* 3287 * Get a vnode for a file handle and export stuff. 3288 */ 3289 int 3290 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 3291 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 3292 struct ucred **credp) 3293 { 3294 int error; 3295 3296 *credp = NULL; 3297 exp->nes_numsecflavor = 0; 3298 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); 3299 if (error != 0) 3300 /* Make sure the server replies ESTALE to the client. */ 3301 error = ESTALE; 3302 if (nam && !error) { 3303 MNT_ILOCK(mp); 3304 if (mp->mnt_exjail == NULL || 3305 mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison) 3306 error = EACCES; 3307 MNT_IUNLOCK(mp); 3308 if (error == 0) 3309 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3310 &exp->nes_numsecflavor, exp->nes_secflavors); 3311 if (error) { 3312 if (NFSD_VNET(nfs_rootfhset)) { 3313 exp->nes_exflag = 0; 3314 exp->nes_numsecflavor = 0; 3315 error = 0; 3316 } else { 3317 vput(*vpp); 3318 } 3319 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3320 MAXSECFLAVORS) { 3321 printf("nfsvno_fhtovp: numsecflavors out of range\n"); 3322 exp->nes_numsecflavor = 0; 3323 error = EACCES; 3324 vput(*vpp); 3325 } 3326 } 3327 NFSEXITCODE(error); 3328 return (error); 3329 } 3330 3331 /* 3332 * nfsd_fhtovp() - convert a fh to a vnode ptr 3333 * - look up fsid in mount list (if not found ret error) 3334 * - get vp and export rights by calling nfsvno_fhtovp() 3335 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 3336 * for AUTH_SYS 3337 * - if mpp != NULL, return the mount point so that it can 3338 * be used for vn_finished_write() by the caller 3339 */ 3340 void 3341 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 3342 struct vnode **vpp, struct nfsexstuff *exp, 3343 struct mount **mpp, int startwrite, int nextop) 3344 { 3345 struct mount *mp, *mpw; 3346 struct ucred *credanon; 3347 fhandle_t *fhp; 3348 int error; 3349 3350 if (mpp != NULL) 3351 *mpp = NULL; 3352 *vpp = NULL; 3353 fhp = (fhandle_t *)nfp->nfsrvfh_data; 3354 mp = vfs_busyfs(&fhp->fh_fsid); 3355 if (mp == NULL) { 3356 nd->nd_repstat = ESTALE; 3357 goto out; 3358 } 3359 3360 if (startwrite) { 3361 mpw = mp; 3362 error = vn_start_write(NULL, &mpw, V_WAIT); 3363 if (error != 0) { 3364 mpw = NULL; 3365 vfs_unbusy(mp); 3366 nd->nd_repstat = ESTALE; 3367 goto out; 3368 } 3369 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) 3370 lktype = LK_EXCLUSIVE; 3371 } else 3372 mpw = NULL; 3373 3374 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 3375 &credanon); 3376 vfs_unbusy(mp); 3377 3378 /* 3379 * For NFSv4 without a pseudo root fs, unexported file handles 3380 * can be returned, so that Lookup works everywhere. 3381 */ 3382 if (!nd->nd_repstat && exp->nes_exflag == 0 && 3383 !(nd->nd_flag & ND_NFSV4)) { 3384 vput(*vpp); 3385 *vpp = NULL; 3386 nd->nd_repstat = EACCES; 3387 } 3388 3389 /* 3390 * Personally, I've never seen any point in requiring a 3391 * reserved port#, since only in the rare case where the 3392 * clients are all boxes with secure system privileges, 3393 * does it provide any enhanced security, but... some people 3394 * believe it to be useful and keep putting this code back in. 3395 * (There is also some "security checker" out there that 3396 * complains if the nfs server doesn't enforce this.) 3397 * However, note the following: 3398 * RFC3530 (NFSv4) specifies that a reserved port# not be 3399 * required. 3400 * RFC2623 recommends that, if a reserved port# is checked for, 3401 * that there be a way to turn that off--> ifdef'd. 3402 */ 3403 #ifdef NFS_REQRSVPORT 3404 if (!nd->nd_repstat) { 3405 struct sockaddr_in *saddr; 3406 struct sockaddr_in6 *saddr6; 3407 3408 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 3409 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 3410 if (!(nd->nd_flag & ND_NFSV4) && 3411 ((saddr->sin_family == AF_INET && 3412 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 3413 (saddr6->sin6_family == AF_INET6 && 3414 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 3415 vput(*vpp); 3416 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3417 } 3418 } 3419 #endif /* NFS_REQRSVPORT */ 3420 3421 /* 3422 * Check/setup credentials. 3423 */ 3424 if (!nd->nd_repstat) { 3425 nd->nd_saveduid = nd->nd_cred->cr_uid; 3426 nd->nd_repstat = nfsd_excred(nd, exp, credanon, 3427 nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type)); 3428 if (nd->nd_repstat) 3429 vput(*vpp); 3430 } 3431 if (credanon != NULL) 3432 crfree(credanon); 3433 if (nd->nd_repstat) { 3434 vn_finished_write(mpw); 3435 *vpp = NULL; 3436 } else if (mpp != NULL) { 3437 *mpp = mpw; 3438 } 3439 3440 out: 3441 NFSEXITCODE2(0, nd); 3442 } 3443 3444 /* 3445 * glue for fp. 3446 */ 3447 static int 3448 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 3449 { 3450 struct filedesc *fdp; 3451 struct file *fp; 3452 int error = 0; 3453 3454 fdp = p->td_proc->p_fd; 3455 if (fd < 0 || fd >= fdp->fd_nfiles || 3456 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { 3457 error = EBADF; 3458 goto out; 3459 } 3460 *fpp = fp; 3461 3462 out: 3463 NFSEXITCODE(error); 3464 return (error); 3465 } 3466 3467 /* 3468 * Called from nfssvc() to update the exports list. Just call 3469 * vfs_export(). This has to be done, since the v4 root fake fs isn't 3470 * in the mount list. 3471 */ 3472 int 3473 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 3474 { 3475 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 3476 int error = 0; 3477 struct nameidata nd; 3478 fhandle_t fh; 3479 3480 error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export, false); 3481 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 3482 NFSD_VNET(nfs_rootfhset) = 0; 3483 else if (error == 0) { 3484 if (nfsexargp->fspec == NULL) { 3485 error = EPERM; 3486 goto out; 3487 } 3488 /* 3489 * If fspec != NULL, this is the v4root path. 3490 */ 3491 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec); 3492 if ((error = namei(&nd)) != 0) 3493 goto out; 3494 error = nfsvno_getfh(nd.ni_vp, &fh, p); 3495 vrele(nd.ni_vp); 3496 if (!error) { 3497 NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH; 3498 NFSBCOPY((caddr_t)&fh, 3499 NFSD_VNET(nfs_rootfh).nfsrvfh_data, 3500 sizeof (fhandle_t)); 3501 NFSD_VNET(nfs_rootfhset) = 1; 3502 } 3503 } 3504 3505 out: 3506 NFSEXITCODE(error); 3507 return (error); 3508 } 3509 3510 /* 3511 * This function needs to test to see if the system is near its limit 3512 * for memory allocation via malloc() or mget() and return True iff 3513 * either of these resources are near their limit. 3514 * XXX (For now, this is just a stub.) 3515 */ 3516 int nfsrv_testmalloclimit = 0; 3517 int 3518 nfsrv_mallocmget_limit(void) 3519 { 3520 static int printmesg = 0; 3521 static int testval = 1; 3522 3523 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 3524 if ((printmesg++ % 100) == 0) 3525 printf("nfsd: malloc/mget near limit\n"); 3526 return (1); 3527 } 3528 return (0); 3529 } 3530 3531 /* 3532 * BSD specific initialization of a mount point. 3533 */ 3534 void 3535 nfsd_mntinit(void) 3536 { 3537 3538 NFSD_LOCK(); 3539 if (NFSD_VNET(nfsrv_mntinited)) { 3540 NFSD_UNLOCK(); 3541 return; 3542 } 3543 NFSD_VNET(nfsrv_mntinited) = true; 3544 nfsrvd_init(0); 3545 NFSD_UNLOCK(); 3546 3547 NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP, 3548 M_WAITOK | M_ZERO); 3549 NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 3550 mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF); 3551 lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0); 3552 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist); 3553 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist); 3554 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 3555 TAILQ_INIT(&NFSD_VNET(nfsv4root_opt)); 3556 TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt)); 3557 NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt); 3558 NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt); 3559 NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0; 3560 NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0; 3561 callout_init(&NFSD_VNET(nfsd_callout), 1); 3562 3563 nfsrvd_initcache(); 3564 nfsd_init(); 3565 } 3566 3567 static void 3568 nfsd_timer(void *arg) 3569 { 3570 struct vnet *vnetp; 3571 3572 vnetp = (struct vnet *)arg; 3573 NFSD_CURVNET_SET_QUIET(vnetp); 3574 nfsrv_servertimer(vnetp); 3575 callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer, 3576 arg, 0); 3577 NFSD_CURVNET_RESTORE(); 3578 } 3579 3580 /* 3581 * Get a vnode for a file handle, without checking exports, etc. 3582 */ 3583 struct vnode * 3584 nfsvno_getvp(fhandle_t *fhp) 3585 { 3586 struct mount *mp; 3587 struct vnode *vp; 3588 int error; 3589 3590 mp = vfs_busyfs(&fhp->fh_fsid); 3591 if (mp == NULL) 3592 return (NULL); 3593 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 3594 vfs_unbusy(mp); 3595 if (error) 3596 return (NULL); 3597 return (vp); 3598 } 3599 3600 /* 3601 * Do a local VOP_ADVLOCK(). 3602 */ 3603 int 3604 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 3605 u_int64_t end, struct thread *td) 3606 { 3607 int error = 0; 3608 struct flock fl; 3609 u_int64_t tlen; 3610 3611 if (nfsrv_dolocallocks == 0) 3612 goto out; 3613 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); 3614 3615 fl.l_whence = SEEK_SET; 3616 fl.l_type = ftype; 3617 fl.l_start = (off_t)first; 3618 if (end == NFS64BITSSET) { 3619 fl.l_len = 0; 3620 } else { 3621 tlen = end - first; 3622 fl.l_len = (off_t)tlen; 3623 } 3624 /* 3625 * For FreeBSD8, the l_pid and l_sysid must be set to the same 3626 * values for all calls, so that all locks will be held by the 3627 * nfsd server. (The nfsd server handles conflicts between the 3628 * various clients.) 3629 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 3630 * bytes, so it can't be put in l_sysid. 3631 */ 3632 if (nfsv4_sysid == 0) 3633 nfsv4_sysid = nlm_acquire_next_sysid(); 3634 fl.l_pid = (pid_t)0; 3635 fl.l_sysid = (int)nfsv4_sysid; 3636 3637 if (ftype == F_UNLCK) 3638 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 3639 (F_POSIX | F_REMOTE)); 3640 else 3641 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 3642 (F_POSIX | F_REMOTE)); 3643 3644 out: 3645 NFSEXITCODE(error); 3646 return (error); 3647 } 3648 3649 /* 3650 * Check the nfsv4 root exports. 3651 */ 3652 int 3653 nfsvno_v4rootexport(struct nfsrv_descript *nd) 3654 { 3655 struct ucred *credanon; 3656 int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; 3657 uint64_t exflags; 3658 3659 error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags, 3660 &credanon, &numsecflavor, secflavors); 3661 if (error) { 3662 error = NFSERR_PROGUNAVAIL; 3663 goto out; 3664 } 3665 if (credanon != NULL) 3666 crfree(credanon); 3667 for (i = 0; i < numsecflavor; i++) { 3668 if (secflavors[i] == AUTH_SYS) 3669 nd->nd_flag |= ND_EXAUTHSYS; 3670 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3671 nd->nd_flag |= ND_EXGSS; 3672 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3673 nd->nd_flag |= ND_EXGSSINTEGRITY; 3674 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3675 nd->nd_flag |= ND_EXGSSPRIVACY; 3676 } 3677 3678 /* And set ND_EXxx flags for TLS. */ 3679 if ((exflags & MNT_EXTLS) != 0) { 3680 nd->nd_flag |= ND_EXTLS; 3681 if ((exflags & MNT_EXTLSCERT) != 0) 3682 nd->nd_flag |= ND_EXTLSCERT; 3683 if ((exflags & MNT_EXTLSCERTUSER) != 0) 3684 nd->nd_flag |= ND_EXTLSCERTUSER; 3685 } 3686 3687 out: 3688 NFSEXITCODE(error); 3689 return (error); 3690 } 3691 3692 /* 3693 * Nfs server pseudo system call for the nfsd's 3694 */ 3695 /* 3696 * MPSAFE 3697 */ 3698 static int 3699 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3700 { 3701 struct file *fp; 3702 struct nfsd_addsock_args sockarg; 3703 struct nfsd_nfsd_args nfsdarg; 3704 struct nfsd_nfsd_oargs onfsdarg; 3705 struct nfsd_pnfsd_args pnfsdarg; 3706 struct vnode *vp, *nvp, *curdvp; 3707 struct pnfsdsfile *pf; 3708 struct nfsdevice *ds, *fds; 3709 cap_rights_t rights; 3710 int buflen, error, ret; 3711 char *buf, *cp, *cp2, *cp3; 3712 char fname[PNFS_FILENAME_LEN + 1]; 3713 3714 NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td)); 3715 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3716 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3717 if (error) 3718 goto out; 3719 /* 3720 * Since we don't know what rights might be required, 3721 * pretend that we need them all. It is better to be too 3722 * careful than too reckless. 3723 */ 3724 error = fget(td, sockarg.sock, 3725 cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); 3726 if (error != 0) 3727 goto out; 3728 if (fp->f_type != DTYPE_SOCKET) { 3729 fdrop(fp, td); 3730 error = EPERM; 3731 goto out; 3732 } 3733 error = nfsrvd_addsock(fp); 3734 fdrop(fp, td); 3735 } else if (uap->flag & NFSSVC_NFSDNFSD) { 3736 if (uap->argp == NULL) { 3737 error = EINVAL; 3738 goto out; 3739 } 3740 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { 3741 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); 3742 if (error == 0) { 3743 nfsdarg.principal = onfsdarg.principal; 3744 nfsdarg.minthreads = onfsdarg.minthreads; 3745 nfsdarg.maxthreads = onfsdarg.maxthreads; 3746 nfsdarg.version = 1; 3747 nfsdarg.addr = NULL; 3748 nfsdarg.addrlen = 0; 3749 nfsdarg.dnshost = NULL; 3750 nfsdarg.dnshostlen = 0; 3751 nfsdarg.dspath = NULL; 3752 nfsdarg.dspathlen = 0; 3753 nfsdarg.mdspath = NULL; 3754 nfsdarg.mdspathlen = 0; 3755 nfsdarg.mirrorcnt = 1; 3756 } 3757 } else 3758 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); 3759 if (error) 3760 goto out; 3761 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && 3762 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && 3763 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && 3764 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && 3765 nfsdarg.mirrorcnt >= 1 && 3766 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && 3767 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && 3768 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { 3769 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" 3770 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, 3771 nfsdarg.dspathlen, nfsdarg.dnshostlen, 3772 nfsdarg.mdspathlen, nfsdarg.mirrorcnt); 3773 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); 3774 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); 3775 if (error != 0) { 3776 free(cp, M_TEMP); 3777 goto out; 3778 } 3779 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ 3780 nfsdarg.addr = cp; 3781 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); 3782 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); 3783 if (error != 0) { 3784 free(nfsdarg.addr, M_TEMP); 3785 free(cp, M_TEMP); 3786 goto out; 3787 } 3788 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ 3789 nfsdarg.dnshost = cp; 3790 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); 3791 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); 3792 if (error != 0) { 3793 free(nfsdarg.addr, M_TEMP); 3794 free(nfsdarg.dnshost, M_TEMP); 3795 free(cp, M_TEMP); 3796 goto out; 3797 } 3798 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ 3799 nfsdarg.dspath = cp; 3800 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); 3801 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); 3802 if (error != 0) { 3803 free(nfsdarg.addr, M_TEMP); 3804 free(nfsdarg.dnshost, M_TEMP); 3805 free(nfsdarg.dspath, M_TEMP); 3806 free(cp, M_TEMP); 3807 goto out; 3808 } 3809 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ 3810 nfsdarg.mdspath = cp; 3811 } else { 3812 nfsdarg.addr = NULL; 3813 nfsdarg.addrlen = 0; 3814 nfsdarg.dnshost = NULL; 3815 nfsdarg.dnshostlen = 0; 3816 nfsdarg.dspath = NULL; 3817 nfsdarg.dspathlen = 0; 3818 nfsdarg.mdspath = NULL; 3819 nfsdarg.mdspathlen = 0; 3820 nfsdarg.mirrorcnt = 1; 3821 } 3822 nfsd_timer(NFSD_TD_TO_VNET(td)); 3823 error = nfsrvd_nfsd(td, &nfsdarg); 3824 callout_drain(&NFSD_VNET(nfsd_callout)); 3825 free(nfsdarg.addr, M_TEMP); 3826 free(nfsdarg.dnshost, M_TEMP); 3827 free(nfsdarg.dspath, M_TEMP); 3828 free(nfsdarg.mdspath, M_TEMP); 3829 } else if (uap->flag & NFSSVC_PNFSDS) { 3830 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); 3831 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || 3832 pnfsdarg.op == PNFSDOP_FORCEDELDS)) { 3833 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3834 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, 3835 NULL); 3836 if (error == 0) 3837 error = nfsrv_deldsserver(pnfsdarg.op, cp, td); 3838 free(cp, M_TEMP); 3839 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { 3840 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3841 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; 3842 buf = malloc(buflen, M_TEMP, M_WAITOK); 3843 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, 3844 NULL); 3845 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); 3846 if (error == 0 && pnfsdarg.dspath != NULL) { 3847 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3848 error = copyinstr(pnfsdarg.dspath, cp2, 3849 PATH_MAX + 1, NULL); 3850 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", 3851 error); 3852 } else 3853 cp2 = NULL; 3854 if (error == 0 && pnfsdarg.curdspath != NULL) { 3855 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 3856 error = copyinstr(pnfsdarg.curdspath, cp3, 3857 PATH_MAX + 1, NULL); 3858 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", 3859 error); 3860 } else 3861 cp3 = NULL; 3862 curdvp = NULL; 3863 fds = NULL; 3864 if (error == 0) 3865 error = nfsrv_mdscopymr(cp, cp2, cp3, buf, 3866 &buflen, fname, td, &vp, &nvp, &pf, &ds, 3867 &fds); 3868 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); 3869 if (error == 0) { 3870 if (pf->dsf_dir >= nfsrv_dsdirsize) { 3871 printf("copymr: dsdir out of range\n"); 3872 pf->dsf_dir = 0; 3873 } 3874 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); 3875 error = nfsrv_copymr(vp, nvp, 3876 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, 3877 (struct pnfsdsfile *)buf, 3878 buflen / sizeof(*pf), td->td_ucred, td); 3879 vput(vp); 3880 vput(nvp); 3881 if (fds != NULL && error == 0) { 3882 curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; 3883 ret = vn_lock(curdvp, LK_EXCLUSIVE); 3884 if (ret == 0) { 3885 nfsrv_dsremove(curdvp, fname, 3886 td->td_ucred, td); 3887 NFSVOPUNLOCK(curdvp); 3888 } 3889 } 3890 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); 3891 } 3892 free(cp, M_TEMP); 3893 free(cp2, M_TEMP); 3894 free(cp3, M_TEMP); 3895 free(buf, M_TEMP); 3896 } 3897 } else { 3898 error = nfssvc_srvcall(td, uap, td->td_ucred); 3899 } 3900 3901 out: 3902 NFSD_CURVNET_RESTORE(); 3903 NFSEXITCODE(error); 3904 return (error); 3905 } 3906 3907 static int 3908 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 3909 { 3910 struct nfsex_args export; 3911 struct nfsex_oldargs oexp; 3912 struct file *fp = NULL; 3913 int stablefd, i, len; 3914 struct nfsd_clid adminrevoke; 3915 struct nfsd_dumplist dumplist; 3916 struct nfsd_dumpclients *dumpclients; 3917 struct nfsd_dumplocklist dumplocklist; 3918 struct nfsd_dumplocks *dumplocks; 3919 struct nameidata nd; 3920 vnode_t vp; 3921 int error = EINVAL, igotlock; 3922 struct proc *procp; 3923 gid_t *grps; 3924 3925 if (uap->flag & NFSSVC_PUBLICFH) { 3926 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 3927 sizeof (fhandle_t)); 3928 error = copyin(uap->argp, 3929 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 3930 if (!error) 3931 nfs_pubfhset = 1; 3932 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 3933 (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { 3934 error = copyin(uap->argp,(caddr_t)&export, 3935 sizeof (struct nfsex_args)); 3936 if (!error) { 3937 grps = NULL; 3938 if (export.export.ex_ngroups > NGROUPS_MAX || 3939 export.export.ex_ngroups < 0) 3940 error = EINVAL; 3941 else if (export.export.ex_ngroups > 0) { 3942 grps = malloc(export.export.ex_ngroups * 3943 sizeof(gid_t), M_TEMP, M_WAITOK); 3944 error = copyin(export.export.ex_groups, grps, 3945 export.export.ex_ngroups * sizeof(gid_t)); 3946 export.export.ex_groups = grps; 3947 } else 3948 export.export.ex_groups = NULL; 3949 if (!error) 3950 error = nfsrv_v4rootexport(&export, cred, p); 3951 free(grps, M_TEMP); 3952 } 3953 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 3954 NFSSVC_V4ROOTEXPORT) { 3955 error = copyin(uap->argp,(caddr_t)&oexp, 3956 sizeof (struct nfsex_oldargs)); 3957 if (!error) { 3958 memset(&export.export, 0, sizeof(export.export)); 3959 export.export.ex_flags = (uint64_t)oexp.export.ex_flags; 3960 export.export.ex_root = oexp.export.ex_root; 3961 export.export.ex_uid = oexp.export.ex_anon.cr_uid; 3962 export.export.ex_ngroups = 3963 oexp.export.ex_anon.cr_ngroups; 3964 export.export.ex_groups = NULL; 3965 if (export.export.ex_ngroups > XU_NGROUPS || 3966 export.export.ex_ngroups < 0) 3967 error = EINVAL; 3968 else if (export.export.ex_ngroups > 0) { 3969 export.export.ex_groups = malloc( 3970 export.export.ex_ngroups * sizeof(gid_t), 3971 M_TEMP, M_WAITOK); 3972 for (i = 0; i < export.export.ex_ngroups; i++) 3973 export.export.ex_groups[i] = 3974 oexp.export.ex_anon.cr_groups[i]; 3975 } 3976 export.export.ex_addr = oexp.export.ex_addr; 3977 export.export.ex_addrlen = oexp.export.ex_addrlen; 3978 export.export.ex_mask = oexp.export.ex_mask; 3979 export.export.ex_masklen = oexp.export.ex_masklen; 3980 export.export.ex_indexfile = oexp.export.ex_indexfile; 3981 export.export.ex_numsecflavors = 3982 oexp.export.ex_numsecflavors; 3983 if (export.export.ex_numsecflavors >= MAXSECFLAVORS || 3984 export.export.ex_numsecflavors < 0) 3985 error = EINVAL; 3986 else { 3987 for (i = 0; i < export.export.ex_numsecflavors; 3988 i++) 3989 export.export.ex_secflavors[i] = 3990 oexp.export.ex_secflavors[i]; 3991 } 3992 export.fspec = oexp.fspec; 3993 if (error == 0) 3994 error = nfsrv_v4rootexport(&export, cred, p); 3995 free(export.export.ex_groups, M_TEMP); 3996 } 3997 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 3998 nfs_pubfhset = 0; 3999 error = 0; 4000 } else if (uap->flag & NFSSVC_STABLERESTART) { 4001 error = copyin(uap->argp, (caddr_t)&stablefd, 4002 sizeof (int)); 4003 if (!error) 4004 error = fp_getfvp(p, stablefd, &fp, &vp); 4005 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 4006 error = EBADF; 4007 if (!error && NFSD_VNET(nfsrv_numnfsd) != 0) 4008 error = ENXIO; 4009 if (!error) { 4010 NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp; 4011 nfsrv_setupstable(p); 4012 } 4013 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 4014 error = copyin(uap->argp, (caddr_t)&adminrevoke, 4015 sizeof (struct nfsd_clid)); 4016 if (!error) 4017 error = nfsrv_adminrevoke(&adminrevoke, p); 4018 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 4019 error = copyin(uap->argp, (caddr_t)&dumplist, 4020 sizeof (struct nfsd_dumplist)); 4021 if (!error && (dumplist.ndl_size < 1 || 4022 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 4023 error = EPERM; 4024 if (!error) { 4025 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 4026 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4027 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 4028 error = copyout(dumpclients, dumplist.ndl_list, len); 4029 free(dumpclients, M_TEMP); 4030 } 4031 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 4032 error = copyin(uap->argp, (caddr_t)&dumplocklist, 4033 sizeof (struct nfsd_dumplocklist)); 4034 if (!error && (dumplocklist.ndllck_size < 1 || 4035 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 4036 error = EPERM; 4037 if (!error) 4038 error = nfsrv_lookupfilename(&nd, 4039 dumplocklist.ndllck_fname, p); 4040 if (!error) { 4041 len = sizeof (struct nfsd_dumplocks) * 4042 dumplocklist.ndllck_size; 4043 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4044 nfsrv_dumplocks(nd.ni_vp, dumplocks, 4045 dumplocklist.ndllck_size, p); 4046 vput(nd.ni_vp); 4047 error = copyout(dumplocks, dumplocklist.ndllck_list, 4048 len); 4049 free(dumplocks, M_TEMP); 4050 } 4051 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 4052 procp = p->td_proc; 4053 PROC_LOCK(procp); 4054 nfsd_master_pid = procp->p_pid; 4055 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 4056 nfsd_master_start = procp->p_stats->p_start; 4057 NFSD_VNET(nfsd_master_proc) = procp; 4058 PROC_UNLOCK(procp); 4059 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { 4060 NFSLOCKV4ROOTMUTEX(); 4061 if (!NFSD_VNET(nfsrv_suspend_nfsd)) { 4062 /* Lock out all nfsd threads */ 4063 do { 4064 igotlock = nfsv4_lock( 4065 &NFSD_VNET(nfsd_suspend_lock), 1, NULL, 4066 NFSV4ROOTLOCKMUTEXPTR, NULL); 4067 } while (igotlock == 0 && 4068 !NFSD_VNET(nfsrv_suspend_nfsd)); 4069 NFSD_VNET(nfsrv_suspend_nfsd) = true; 4070 } 4071 NFSUNLOCKV4ROOTMUTEX(); 4072 error = 0; 4073 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { 4074 NFSLOCKV4ROOTMUTEX(); 4075 if (NFSD_VNET(nfsrv_suspend_nfsd)) { 4076 nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0); 4077 NFSD_VNET(nfsrv_suspend_nfsd) = false; 4078 } 4079 NFSUNLOCKV4ROOTMUTEX(); 4080 error = 0; 4081 } 4082 4083 NFSEXITCODE(error); 4084 return (error); 4085 } 4086 4087 /* 4088 * Check exports. 4089 * Returns 0 if ok, 1 otherwise. 4090 */ 4091 int 4092 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 4093 { 4094 int i; 4095 4096 if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) || 4097 (NFSVNO_EXTLSCERT(exp) && 4098 (nd->nd_flag & ND_TLSCERT) == 0) || 4099 (NFSVNO_EXTLSCERTUSER(exp) && 4100 (nd->nd_flag & ND_TLSCERTUSER) == 0)) { 4101 if ((nd->nd_flag & ND_NFSV4) != 0) 4102 return (NFSERR_WRONGSEC); 4103 #ifdef notnow 4104 /* There is currently no auth_stat for this. */ 4105 else if ((nd->nd_flag & ND_TLS) == 0) 4106 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS); 4107 else 4108 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST); 4109 #endif 4110 else 4111 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4112 } 4113 4114 /* 4115 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use 4116 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. 4117 */ 4118 if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO) 4119 return (0); 4120 4121 /* 4122 * This seems odd, but allow the case where the security flavor 4123 * list is empty. This happens when NFSv4 is traversing non-exported 4124 * file systems. Exported file systems should always have a non-empty 4125 * security flavor list. 4126 */ 4127 if (exp->nes_numsecflavor == 0) 4128 return (0); 4129 4130 for (i = 0; i < exp->nes_numsecflavor; i++) { 4131 /* 4132 * The tests for privacy and integrity must be first, 4133 * since ND_GSS is set for everything but AUTH_SYS. 4134 */ 4135 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 4136 (nd->nd_flag & ND_GSSPRIVACY)) 4137 return (0); 4138 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 4139 (nd->nd_flag & ND_GSSINTEGRITY)) 4140 return (0); 4141 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 4142 (nd->nd_flag & ND_GSS)) 4143 return (0); 4144 if (exp->nes_secflavors[i] == AUTH_SYS && 4145 (nd->nd_flag & ND_GSS) == 0) 4146 return (0); 4147 } 4148 if ((nd->nd_flag & ND_NFSV4) != 0) 4149 return (NFSERR_WRONGSEC); 4150 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4151 } 4152 4153 /* 4154 * Calculate a hash value for the fid in a file handle. 4155 */ 4156 uint32_t 4157 nfsrv_hashfh(fhandle_t *fhp) 4158 { 4159 uint32_t hashval; 4160 4161 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 4162 return (hashval); 4163 } 4164 4165 /* 4166 * Calculate a hash value for the sessionid. 4167 */ 4168 uint32_t 4169 nfsrv_hashsessionid(uint8_t *sessionid) 4170 { 4171 uint32_t hashval; 4172 4173 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); 4174 return (hashval); 4175 } 4176 4177 /* 4178 * Signal the userland master nfsd to backup the stable restart file. 4179 */ 4180 void 4181 nfsrv_backupstable(void) 4182 { 4183 struct proc *procp; 4184 4185 if (NFSD_VNET(nfsd_master_proc) != NULL) { 4186 procp = pfind(nfsd_master_pid); 4187 /* Try to make sure it is the correct process. */ 4188 if (procp == NFSD_VNET(nfsd_master_proc) && 4189 procp->p_stats->p_start.tv_sec == 4190 nfsd_master_start.tv_sec && 4191 procp->p_stats->p_start.tv_usec == 4192 nfsd_master_start.tv_usec && 4193 strcmp(procp->p_comm, nfsd_master_comm) == 0) 4194 kern_psignal(procp, SIGUSR2); 4195 else 4196 NFSD_VNET(nfsd_master_proc) = NULL; 4197 4198 if (procp != NULL) 4199 PROC_UNLOCK(procp); 4200 } 4201 } 4202 4203 /* 4204 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. 4205 * The arguments are in a structure, so that they can be passed through 4206 * taskqueue for a kernel process to execute this function. 4207 */ 4208 struct nfsrvdscreate { 4209 int done; 4210 int inprog; 4211 struct task tsk; 4212 struct ucred *tcred; 4213 struct vnode *dvp; 4214 NFSPROC_T *p; 4215 struct pnfsdsfile *pf; 4216 int err; 4217 fhandle_t fh; 4218 struct vattr va; 4219 struct vattr createva; 4220 }; 4221 4222 int 4223 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, 4224 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, 4225 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) 4226 { 4227 struct vnode *nvp; 4228 struct nameidata named; 4229 struct vattr va; 4230 char *bufp; 4231 u_long *hashp; 4232 struct nfsnode *np; 4233 struct nfsmount *nmp; 4234 int error; 4235 4236 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, 4237 LOCKPARENT | LOCKLEAF | NOCACHE); 4238 nfsvno_setpathbuf(&named, &bufp, &hashp); 4239 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 4240 named.ni_cnd.cn_nameptr = bufp; 4241 if (fnamep != NULL) { 4242 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); 4243 named.ni_cnd.cn_namelen = strlen(bufp); 4244 } else 4245 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); 4246 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); 4247 4248 /* Create the date file in the DS mount. */ 4249 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4250 if (error == 0) { 4251 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); 4252 vref(dvp); 4253 VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false); 4254 if (error == 0) { 4255 /* Set the ownership of the file. */ 4256 error = VOP_SETATTR(nvp, nvap, tcred); 4257 NFSD_DEBUG(4, "nfsrv_dscreate:" 4258 " setattr-uid=%d\n", error); 4259 if (error != 0) 4260 vput(nvp); 4261 } 4262 if (error != 0) 4263 printf("pNFS: pnfscreate failed=%d\n", error); 4264 } else 4265 printf("pNFS: pnfscreate vnlock=%d\n", error); 4266 if (error == 0) { 4267 np = VTONFS(nvp); 4268 nmp = VFSTONFS(nvp->v_mount); 4269 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") 4270 != 0 || nmp->nm_nam->sa_len > sizeof( 4271 struct sockaddr_in6) || 4272 np->n_fhp->nfh_len != NFSX_MYFH) { 4273 printf("Bad DS file: fstype=%s salen=%d" 4274 " fhlen=%d\n", 4275 nvp->v_mount->mnt_vfc->vfc_name, 4276 nmp->nm_nam->sa_len, np->n_fhp->nfh_len); 4277 error = ENOENT; 4278 } 4279 4280 /* Set extattrs for the DS on the MDS file. */ 4281 if (error == 0) { 4282 if (dsa != NULL) { 4283 error = VOP_GETATTR(nvp, &va, tcred); 4284 if (error == 0) { 4285 dsa->dsa_filerev = va.va_filerev; 4286 dsa->dsa_size = va.va_size; 4287 dsa->dsa_atime = va.va_atime; 4288 dsa->dsa_mtime = va.va_mtime; 4289 dsa->dsa_bytes = va.va_bytes; 4290 } 4291 } 4292 if (error == 0) { 4293 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, 4294 NFSX_MYFH); 4295 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, 4296 nmp->nm_nam->sa_len); 4297 NFSBCOPY(named.ni_cnd.cn_nameptr, 4298 pf->dsf_filename, 4299 sizeof(pf->dsf_filename)); 4300 } 4301 } else 4302 printf("pNFS: pnfscreate can't get DS" 4303 " attr=%d\n", error); 4304 if (nvpp != NULL && error == 0) 4305 *nvpp = nvp; 4306 else 4307 vput(nvp); 4308 } 4309 nfsvno_relpathbuf(&named); 4310 return (error); 4311 } 4312 4313 /* 4314 * Start up the thread that will execute nfsrv_dscreate(). 4315 */ 4316 static void 4317 start_dscreate(void *arg, int pending) 4318 { 4319 struct nfsrvdscreate *dsc; 4320 4321 dsc = (struct nfsrvdscreate *)arg; 4322 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, 4323 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); 4324 dsc->done = 1; 4325 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); 4326 } 4327 4328 /* 4329 * Create a pNFS data file on the Data Server(s). 4330 */ 4331 static void 4332 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, 4333 NFSPROC_T *p) 4334 { 4335 struct nfsrvdscreate *dsc, *tdsc = NULL; 4336 struct nfsdevice *ds, *tds, *fds; 4337 struct mount *mp; 4338 struct pnfsdsfile *pf, *tpf; 4339 struct pnfsdsattr dsattr; 4340 struct vattr va; 4341 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4342 struct nfsmount *nmp; 4343 fhandle_t fh; 4344 uid_t vauid; 4345 gid_t vagid; 4346 u_short vamode; 4347 struct ucred *tcred; 4348 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; 4349 int failpos, timo; 4350 4351 /* Get a DS server directory in a round-robin order. */ 4352 mirrorcnt = 1; 4353 mp = vp->v_mount; 4354 ds = fds = NULL; 4355 NFSDDSLOCK(); 4356 /* 4357 * Search for the first entry that handles this MDS fs, but use the 4358 * first entry for all MDS fs's otherwise. 4359 */ 4360 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { 4361 if (tds->nfsdev_nmp != NULL) { 4362 if (tds->nfsdev_mdsisset == 0 && ds == NULL) 4363 ds = tds; 4364 else if (tds->nfsdev_mdsisset != 0 && fsidcmp( 4365 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { 4366 ds = fds = tds; 4367 break; 4368 } 4369 } 4370 } 4371 if (ds == NULL) { 4372 NFSDDSUNLOCK(); 4373 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); 4374 return; 4375 } 4376 i = dsdir[0] = ds->nfsdev_nextdir; 4377 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; 4378 dvp[0] = ds->nfsdev_dsdir[i]; 4379 tds = TAILQ_NEXT(ds, nfsdev_list); 4380 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { 4381 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { 4382 if (tds->nfsdev_nmp != NULL && 4383 ((tds->nfsdev_mdsisset == 0 && fds == NULL) || 4384 (tds->nfsdev_mdsisset != 0 && fds != NULL && 4385 fsidcmp(&mp->mnt_stat.f_fsid, 4386 &tds->nfsdev_mdsfsid) == 0))) { 4387 dsdir[mirrorcnt] = i; 4388 dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; 4389 mirrorcnt++; 4390 if (mirrorcnt >= nfsrv_maxpnfsmirror) 4391 break; 4392 } 4393 } 4394 } 4395 /* Put at end of list to implement round-robin usage. */ 4396 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 4397 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 4398 NFSDDSUNLOCK(); 4399 dsc = NULL; 4400 if (mirrorcnt > 1) 4401 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, 4402 M_WAITOK | M_ZERO); 4403 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | 4404 M_ZERO); 4405 4406 error = nfsvno_getfh(vp, &fh, p); 4407 if (error == 0) 4408 error = VOP_GETATTR(vp, &va, cred); 4409 if (error == 0) { 4410 /* Set the attributes for "vp" to Setattr the DS vp. */ 4411 vauid = va.va_uid; 4412 vagid = va.va_gid; 4413 vamode = va.va_mode; 4414 VATTR_NULL(&va); 4415 va.va_uid = vauid; 4416 va.va_gid = vagid; 4417 va.va_mode = vamode; 4418 va.va_size = 0; 4419 } else 4420 printf("pNFS: pnfscreate getfh+attr=%d\n", error); 4421 4422 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, 4423 cred->cr_gid); 4424 /* Make data file name based on FH. */ 4425 tcred = newnfs_getcred(); 4426 4427 /* 4428 * Create the file on each DS mirror, using kernel process(es) for the 4429 * additional mirrors. 4430 */ 4431 failpos = -1; 4432 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { 4433 tpf->dsf_dir = dsdir[i]; 4434 tdsc->tcred = tcred; 4435 tdsc->p = p; 4436 tdsc->pf = tpf; 4437 tdsc->createva = *vap; 4438 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); 4439 tdsc->va = va; 4440 tdsc->dvp = dvp[i]; 4441 tdsc->done = 0; 4442 tdsc->inprog = 0; 4443 tdsc->err = 0; 4444 ret = EIO; 4445 if (nfs_pnfsiothreads != 0) { 4446 ret = nfs_pnfsio(start_dscreate, tdsc); 4447 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); 4448 } 4449 if (ret != 0) { 4450 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, 4451 NULL, tcred, p, NULL); 4452 if (ret != 0) { 4453 KASSERT(error == 0, ("nfsrv_dscreate err=%d", 4454 error)); 4455 if (failpos == -1 && nfsds_failerr(ret)) 4456 failpos = i; 4457 else 4458 error = ret; 4459 } 4460 } 4461 } 4462 if (error == 0) { 4463 tpf->dsf_dir = dsdir[mirrorcnt - 1]; 4464 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, 4465 &dsattr, NULL, tcred, p, NULL); 4466 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { 4467 failpos = mirrorcnt - 1; 4468 error = 0; 4469 } 4470 } 4471 timo = hz / 50; /* Wait for 20msec. */ 4472 if (timo < 1) 4473 timo = 1; 4474 /* Wait for kernel task(s) to complete. */ 4475 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { 4476 while (tdsc->inprog != 0 && tdsc->done == 0) 4477 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); 4478 if (tdsc->err != 0) { 4479 if (failpos == -1 && nfsds_failerr(tdsc->err)) 4480 failpos = i; 4481 else if (error == 0) 4482 error = tdsc->err; 4483 } 4484 } 4485 4486 /* 4487 * If failpos has been set, that mirror has failed, so it needs 4488 * to be disabled. 4489 */ 4490 if (failpos >= 0) { 4491 nmp = VFSTONFS(dvp[failpos]->v_mount); 4492 NFSLOCKMNT(nmp); 4493 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4494 NFSMNTP_CANCELRPCS)) == 0) { 4495 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4496 NFSUNLOCKMNT(nmp); 4497 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4498 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, 4499 ds); 4500 if (ds != NULL) 4501 nfsrv_killrpcs(nmp); 4502 NFSLOCKMNT(nmp); 4503 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4504 wakeup(nmp); 4505 } 4506 NFSUNLOCKMNT(nmp); 4507 } 4508 4509 NFSFREECRED(tcred); 4510 if (error == 0) { 4511 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); 4512 4513 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", 4514 mirrorcnt, nfsrv_maxpnfsmirror); 4515 /* 4516 * For all mirrors that couldn't be created, fill in the 4517 * *pf structure, but with an IP address == 0.0.0.0. 4518 */ 4519 tpf = pf + mirrorcnt; 4520 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { 4521 *tpf = *pf; 4522 tpf->dsf_sin.sin_family = AF_INET; 4523 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); 4524 tpf->dsf_sin.sin_addr.s_addr = 0; 4525 tpf->dsf_sin.sin_port = 0; 4526 } 4527 4528 error = vn_extattr_set(vp, IO_NODELOCKED, 4529 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 4530 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); 4531 if (error == 0) 4532 error = vn_extattr_set(vp, IO_NODELOCKED, 4533 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", 4534 sizeof(dsattr), (char *)&dsattr, p); 4535 if (error != 0) 4536 printf("pNFS: pnfscreate setextattr=%d\n", 4537 error); 4538 } else 4539 printf("pNFS: pnfscreate=%d\n", error); 4540 free(pf, M_TEMP); 4541 free(dsc, M_TEMP); 4542 } 4543 4544 /* 4545 * Get the information needed to remove the pNFS Data Server file from the 4546 * Metadata file. Upon success, ddvp is set non-NULL to the locked 4547 * DS directory vnode. The caller must unlock *ddvp when done with it. 4548 */ 4549 static void 4550 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, 4551 int *mirrorcntp, char *fname, fhandle_t *fhp) 4552 { 4553 struct vattr va; 4554 struct ucred *tcred; 4555 char *buf; 4556 int buflen, error; 4557 4558 dvpp[0] = NULL; 4559 /* If not an exported regular file or not a pNFS server, just return. */ 4560 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4561 nfsrv_devidcnt == 0) 4562 return; 4563 4564 /* Check to see if this is the last hard link. */ 4565 tcred = newnfs_getcred(); 4566 error = VOP_GETATTR(vp, &va, tcred); 4567 NFSFREECRED(tcred); 4568 if (error != 0) { 4569 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); 4570 return; 4571 } 4572 if (va.va_nlink > 1) 4573 return; 4574 4575 error = nfsvno_getfh(vp, fhp, p); 4576 if (error != 0) { 4577 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); 4578 return; 4579 } 4580 4581 buflen = 1024; 4582 buf = malloc(buflen, M_TEMP, M_WAITOK); 4583 /* Get the directory vnode for the DS mount and the file handle. */ 4584 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, 4585 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); 4586 free(buf, M_TEMP); 4587 if (error != 0) 4588 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); 4589 } 4590 4591 /* 4592 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. 4593 * The arguments are in a structure, so that they can be passed through 4594 * taskqueue for a kernel process to execute this function. 4595 */ 4596 struct nfsrvdsremove { 4597 int done; 4598 int inprog; 4599 struct task tsk; 4600 struct ucred *tcred; 4601 struct vnode *dvp; 4602 NFSPROC_T *p; 4603 int err; 4604 char fname[PNFS_FILENAME_LEN + 1]; 4605 }; 4606 4607 static int 4608 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, 4609 NFSPROC_T *p) 4610 { 4611 struct nameidata named; 4612 struct vnode *nvp; 4613 char *bufp; 4614 u_long *hashp; 4615 int error; 4616 4617 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4618 if (error != 0) 4619 return (error); 4620 named.ni_cnd.cn_nameiop = DELETE; 4621 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 4622 named.ni_cnd.cn_cred = tcred; 4623 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 4624 nfsvno_setpathbuf(&named, &bufp, &hashp); 4625 named.ni_cnd.cn_nameptr = bufp; 4626 named.ni_cnd.cn_namelen = strlen(fname); 4627 strlcpy(bufp, fname, NAME_MAX); 4628 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); 4629 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 4630 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); 4631 if (error == 0) { 4632 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); 4633 vput(nvp); 4634 } 4635 NFSVOPUNLOCK(dvp); 4636 nfsvno_relpathbuf(&named); 4637 if (error != 0) 4638 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); 4639 return (error); 4640 } 4641 4642 /* 4643 * Start up the thread that will execute nfsrv_dsremove(). 4644 */ 4645 static void 4646 start_dsremove(void *arg, int pending) 4647 { 4648 struct nfsrvdsremove *dsrm; 4649 4650 dsrm = (struct nfsrvdsremove *)arg; 4651 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, 4652 dsrm->p); 4653 dsrm->done = 1; 4654 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); 4655 } 4656 4657 /* 4658 * Remove a pNFS data file from a Data Server. 4659 * nfsrv_pnfsremovesetup() must have been called before the MDS file was 4660 * removed to set up the dvp and fill in the FH. 4661 */ 4662 static void 4663 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, 4664 NFSPROC_T *p) 4665 { 4666 struct ucred *tcred; 4667 struct nfsrvdsremove *dsrm, *tdsrm; 4668 struct nfsdevice *ds; 4669 struct nfsmount *nmp; 4670 int failpos, i, ret, timo; 4671 4672 tcred = newnfs_getcred(); 4673 dsrm = NULL; 4674 if (mirrorcnt > 1) 4675 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); 4676 /* 4677 * Remove the file on each DS mirror, using kernel process(es) for the 4678 * additional mirrors. 4679 */ 4680 failpos = -1; 4681 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4682 tdsrm->tcred = tcred; 4683 tdsrm->p = p; 4684 tdsrm->dvp = dvp[i]; 4685 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); 4686 tdsrm->inprog = 0; 4687 tdsrm->done = 0; 4688 tdsrm->err = 0; 4689 ret = EIO; 4690 if (nfs_pnfsiothreads != 0) { 4691 ret = nfs_pnfsio(start_dsremove, tdsrm); 4692 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); 4693 } 4694 if (ret != 0) { 4695 ret = nfsrv_dsremove(dvp[i], fname, tcred, p); 4696 if (failpos == -1 && nfsds_failerr(ret)) 4697 failpos = i; 4698 } 4699 } 4700 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); 4701 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) 4702 failpos = mirrorcnt - 1; 4703 timo = hz / 50; /* Wait for 20msec. */ 4704 if (timo < 1) 4705 timo = 1; 4706 /* Wait for kernel task(s) to complete. */ 4707 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4708 while (tdsrm->inprog != 0 && tdsrm->done == 0) 4709 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); 4710 if (failpos == -1 && nfsds_failerr(tdsrm->err)) 4711 failpos = i; 4712 } 4713 4714 /* 4715 * If failpos has been set, that mirror has failed, so it needs 4716 * to be disabled. 4717 */ 4718 if (failpos >= 0) { 4719 nmp = VFSTONFS(dvp[failpos]->v_mount); 4720 NFSLOCKMNT(nmp); 4721 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4722 NFSMNTP_CANCELRPCS)) == 0) { 4723 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4724 NFSUNLOCKMNT(nmp); 4725 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4726 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, 4727 ds); 4728 if (ds != NULL) 4729 nfsrv_killrpcs(nmp); 4730 NFSLOCKMNT(nmp); 4731 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4732 wakeup(nmp); 4733 } 4734 NFSUNLOCKMNT(nmp); 4735 } 4736 4737 /* Get rid all layouts for the file. */ 4738 nfsrv_freefilelayouts(fhp); 4739 4740 NFSFREECRED(tcred); 4741 free(dsrm, M_TEMP); 4742 } 4743 4744 /* 4745 * Generate a file name based on the file handle and put it in *bufp. 4746 * Return the number of bytes generated. 4747 */ 4748 static int 4749 nfsrv_putfhname(fhandle_t *fhp, char *bufp) 4750 { 4751 int i; 4752 uint8_t *cp; 4753 const uint8_t *hexdigits = "0123456789abcdef"; 4754 4755 cp = (uint8_t *)fhp; 4756 for (i = 0; i < sizeof(*fhp); i++) { 4757 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; 4758 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; 4759 } 4760 bufp[2 * i] = '\0'; 4761 return (2 * i); 4762 } 4763 4764 /* 4765 * Update the Metadata file's attributes from the DS file when a Read/Write 4766 * layout is returned. 4767 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN 4768 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. 4769 */ 4770 int 4771 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 4772 { 4773 struct ucred *tcred; 4774 int error; 4775 4776 /* Do this as root so that it won't fail with EACCES. */ 4777 tcred = newnfs_getcred(); 4778 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, 4779 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); 4780 NFSFREECRED(tcred); 4781 return (error); 4782 } 4783 4784 /* 4785 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. 4786 */ 4787 static int 4788 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, 4789 NFSPROC_T *p) 4790 { 4791 int error; 4792 4793 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, 4794 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); 4795 return (error); 4796 } 4797 4798 static int 4799 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 4800 struct thread *p, int ioproc, struct mbuf **mpp, char *cp, 4801 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, 4802 off_t *offp, int content, bool *eofp) 4803 { 4804 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; 4805 fhandle_t fh[NFSDEV_MAXMIRRORS]; 4806 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4807 struct nfsdevice *ds; 4808 struct pnfsdsattr dsattr; 4809 struct opnfsdsattr odsattr; 4810 char *buf; 4811 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; 4812 4813 NFSD_DEBUG(4, "in nfsrv_proxyds\n"); 4814 /* 4815 * If not a regular file, not exported or not a pNFS server, 4816 * just return ENOENT. 4817 */ 4818 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4819 nfsrv_devidcnt == 0) 4820 return (ENOENT); 4821 4822 buflen = 1024; 4823 buf = malloc(buflen, M_TEMP, M_WAITOK); 4824 error = 0; 4825 4826 /* 4827 * For Getattr, get the Change attribute (va_filerev) and size (va_size) 4828 * from the MetaData file's extended attribute. 4829 */ 4830 if (ioproc == NFSPROC_GETATTR) { 4831 error = vn_extattr_get(vp, IO_NODELOCKED, 4832 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, 4833 p); 4834 if (error == 0) { 4835 if (buflen == sizeof(odsattr)) { 4836 NFSBCOPY(buf, &odsattr, buflen); 4837 nap->na_filerev = odsattr.dsa_filerev; 4838 nap->na_size = odsattr.dsa_size; 4839 nap->na_atime = odsattr.dsa_atime; 4840 nap->na_mtime = odsattr.dsa_mtime; 4841 /* 4842 * Fake na_bytes by rounding up na_size. 4843 * Since we don't know the block size, just 4844 * use BLKDEV_IOSIZE. 4845 */ 4846 nap->na_bytes = (odsattr.dsa_size + 4847 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); 4848 } else if (buflen == sizeof(dsattr)) { 4849 NFSBCOPY(buf, &dsattr, buflen); 4850 nap->na_filerev = dsattr.dsa_filerev; 4851 nap->na_size = dsattr.dsa_size; 4852 nap->na_atime = dsattr.dsa_atime; 4853 nap->na_mtime = dsattr.dsa_mtime; 4854 nap->na_bytes = dsattr.dsa_bytes; 4855 } else 4856 error = ENXIO; 4857 } 4858 if (error == 0) { 4859 /* 4860 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() 4861 * returns 0, just return now. nfsrv_checkdsattr() 4862 * returns 0 if there is no Read/Write layout 4863 * plus either an Open/Write_access or Write 4864 * delegation issued to a client for the file. 4865 */ 4866 if (nfsrv_pnfsgetdsattr == 0 || 4867 nfsrv_checkdsattr(vp, p) == 0) { 4868 free(buf, M_TEMP); 4869 return (error); 4870 } 4871 } 4872 4873 /* 4874 * Clear ENOATTR so the code below will attempt to do a 4875 * nfsrv_getattrdsrpc() to get the attributes and (re)create 4876 * the extended attribute. 4877 */ 4878 if (error == ENOATTR) 4879 error = 0; 4880 } 4881 4882 origmircnt = -1; 4883 trycnt = 0; 4884 tryagain: 4885 if (error == 0) { 4886 buflen = 1024; 4887 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == 4888 LK_EXCLUSIVE) 4889 printf("nfsrv_proxyds: Readds vp exclusively locked\n"); 4890 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, 4891 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, 4892 NULL, NULL); 4893 if (error == 0) { 4894 for (i = 0; i < mirrorcnt; i++) 4895 nmp[i] = VFSTONFS(dvp[i]->v_mount); 4896 } else 4897 printf("pNFS: proxy getextattr sockaddr=%d\n", error); 4898 } else 4899 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); 4900 if (error == 0) { 4901 failpos = -1; 4902 if (origmircnt == -1) 4903 origmircnt = mirrorcnt; 4904 /* 4905 * If failpos is set to a mirror#, then that mirror has 4906 * failed and will be disabled. For Read, Getattr and Seek, the 4907 * function only tries one mirror, so if that mirror has 4908 * failed, it will need to be retried. As such, increment 4909 * tryitagain for these cases. 4910 * For Write, Setattr and Setacl, the function tries all 4911 * mirrors and will not return an error for the case where 4912 * one mirror has failed. For these cases, the functioning 4913 * mirror(s) will have been modified, so a retry isn't 4914 * necessary. These functions will set failpos for the 4915 * failed mirror#. 4916 */ 4917 if (ioproc == NFSPROC_READDS) { 4918 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], 4919 mpp, mpp2); 4920 if (nfsds_failerr(error) && mirrorcnt > 1) { 4921 /* 4922 * Setting failpos will cause the mirror 4923 * to be disabled and then a retry of this 4924 * read is required. 4925 */ 4926 failpos = 0; 4927 error = 0; 4928 trycnt++; 4929 } 4930 } else if (ioproc == NFSPROC_WRITEDS) 4931 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, 4932 &nmp[0], mirrorcnt, mpp, cp, &failpos); 4933 else if (ioproc == NFSPROC_SETATTR) 4934 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], 4935 mirrorcnt, nap, &failpos); 4936 else if (ioproc == NFSPROC_SETACL) 4937 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], 4938 mirrorcnt, aclp, &failpos); 4939 else if (ioproc == NFSPROC_SEEKDS) { 4940 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, 4941 p, nmp[0]); 4942 if (nfsds_failerr(error) && mirrorcnt > 1) { 4943 /* 4944 * Setting failpos will cause the mirror 4945 * to be disabled and then a retry of this 4946 * read is required. 4947 */ 4948 failpos = 0; 4949 error = 0; 4950 trycnt++; 4951 } 4952 } else if (ioproc == NFSPROC_ALLOCATE) 4953 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, 4954 &nmp[0], mirrorcnt, &failpos); 4955 else if (ioproc == NFSPROC_DEALLOCATE) 4956 error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p, 4957 vp, &nmp[0], mirrorcnt, &failpos); 4958 else { 4959 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, 4960 vp, nmp[mirrorcnt - 1], nap); 4961 if (nfsds_failerr(error) && mirrorcnt > 1) { 4962 /* 4963 * Setting failpos will cause the mirror 4964 * to be disabled and then a retry of this 4965 * getattr is required. 4966 */ 4967 failpos = mirrorcnt - 1; 4968 error = 0; 4969 trycnt++; 4970 } 4971 } 4972 ds = NULL; 4973 if (failpos >= 0) { 4974 failnmp = nmp[failpos]; 4975 NFSLOCKMNT(failnmp); 4976 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | 4977 NFSMNTP_CANCELRPCS)) == 0) { 4978 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4979 NFSUNLOCKMNT(failnmp); 4980 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, 4981 failnmp, p); 4982 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", 4983 failpos, ds); 4984 if (ds != NULL) 4985 nfsrv_killrpcs(failnmp); 4986 NFSLOCKMNT(failnmp); 4987 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4988 wakeup(failnmp); 4989 } 4990 NFSUNLOCKMNT(failnmp); 4991 } 4992 for (i = 0; i < mirrorcnt; i++) 4993 NFSVOPUNLOCK(dvp[i]); 4994 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, 4995 trycnt); 4996 /* Try the Read/Getattr again if a mirror was deleted. */ 4997 if (ds != NULL && trycnt > 0 && trycnt < origmircnt) 4998 goto tryagain; 4999 } else { 5000 /* Return ENOENT for any Extended Attribute error. */ 5001 error = ENOENT; 5002 } 5003 free(buf, M_TEMP); 5004 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); 5005 return (error); 5006 } 5007 5008 /* 5009 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended 5010 * attribute. 5011 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs 5012 * to be checked. If it points to a NULL nmp, then it returns 5013 * a suitable destination. 5014 * curnmp - If non-NULL, it is the source mount for the copy. 5015 */ 5016 int 5017 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, 5018 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, 5019 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, 5020 struct nfsmount *curnmp, int *ippos, int *dsdirp) 5021 { 5022 struct vnode *dvp, *nvp = NULL, **tdvpp; 5023 struct mount *mp; 5024 struct nfsmount *nmp, *newnmp; 5025 struct sockaddr *sad; 5026 struct sockaddr_in *sin; 5027 struct nfsdevice *ds, *tds, *fndds; 5028 struct pnfsdsfile *pf; 5029 uint32_t dsdir; 5030 int error, fhiszero, fnd, gotone, i, mirrorcnt; 5031 5032 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); 5033 *mirrorcntp = 1; 5034 tdvpp = dvpp; 5035 if (nvpp != NULL) 5036 *nvpp = NULL; 5037 if (dvpp != NULL) 5038 *dvpp = NULL; 5039 if (ippos != NULL) 5040 *ippos = -1; 5041 if (newnmpp != NULL) 5042 newnmp = *newnmpp; 5043 else 5044 newnmp = NULL; 5045 mp = vp->v_mount; 5046 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5047 "pnfsd.dsfile", buflenp, buf, p); 5048 mirrorcnt = *buflenp / sizeof(*pf); 5049 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || 5050 *buflenp != sizeof(*pf) * mirrorcnt)) 5051 error = ENOATTR; 5052 5053 pf = (struct pnfsdsfile *)buf; 5054 /* If curnmp != NULL, check for a match in the mirror list. */ 5055 if (curnmp != NULL && error == 0) { 5056 fnd = 0; 5057 for (i = 0; i < mirrorcnt; i++, pf++) { 5058 sad = (struct sockaddr *)&pf->dsf_sin; 5059 if (nfsaddr2_match(sad, curnmp->nm_nam)) { 5060 if (ippos != NULL) 5061 *ippos = i; 5062 fnd = 1; 5063 break; 5064 } 5065 } 5066 if (fnd == 0) 5067 error = ENXIO; 5068 } 5069 5070 gotone = 0; 5071 pf = (struct pnfsdsfile *)buf; 5072 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, 5073 error); 5074 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { 5075 fhiszero = 0; 5076 sad = (struct sockaddr *)&pf->dsf_sin; 5077 sin = &pf->dsf_sin; 5078 dsdir = pf->dsf_dir; 5079 if (dsdir >= nfsrv_dsdirsize) { 5080 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); 5081 error = ENOATTR; 5082 } else if (nvpp != NULL && newnmp != NULL && 5083 nfsaddr2_match(sad, newnmp->nm_nam)) 5084 error = EEXIST; 5085 if (error == 0) { 5086 if (ippos != NULL && curnmp == NULL && 5087 sad->sa_family == AF_INET && 5088 sin->sin_addr.s_addr == 0) 5089 *ippos = i; 5090 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) 5091 fhiszero = 1; 5092 /* Use the socket address to find the mount point. */ 5093 fndds = NULL; 5094 NFSDDSLOCK(); 5095 /* Find a match for the IP address. */ 5096 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5097 if (ds->nfsdev_nmp != NULL) { 5098 dvp = ds->nfsdev_dvp; 5099 nmp = VFSTONFS(dvp->v_mount); 5100 if (nmp != ds->nfsdev_nmp) 5101 printf("different2 nmp %p %p\n", 5102 nmp, ds->nfsdev_nmp); 5103 if (nfsaddr2_match(sad, nmp->nm_nam)) { 5104 fndds = ds; 5105 break; 5106 } 5107 } 5108 } 5109 if (fndds != NULL && newnmpp != NULL && 5110 newnmp == NULL) { 5111 /* Search for a place to make a mirror copy. */ 5112 TAILQ_FOREACH(tds, &nfsrv_devidhead, 5113 nfsdev_list) { 5114 if (tds->nfsdev_nmp != NULL && 5115 fndds != tds && 5116 ((tds->nfsdev_mdsisset == 0 && 5117 fndds->nfsdev_mdsisset == 0) || 5118 (tds->nfsdev_mdsisset != 0 && 5119 fndds->nfsdev_mdsisset != 0 && 5120 fsidcmp(&tds->nfsdev_mdsfsid, 5121 &mp->mnt_stat.f_fsid) == 0))) { 5122 *newnmpp = tds->nfsdev_nmp; 5123 break; 5124 } 5125 } 5126 if (tds != NULL) { 5127 /* 5128 * Move this entry to the end of the 5129 * list, so it won't be selected as 5130 * easily the next time. 5131 */ 5132 TAILQ_REMOVE(&nfsrv_devidhead, tds, 5133 nfsdev_list); 5134 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, 5135 nfsdev_list); 5136 } 5137 } 5138 NFSDDSUNLOCK(); 5139 if (fndds != NULL) { 5140 dvp = fndds->nfsdev_dsdir[dsdir]; 5141 if (lktype != 0 || fhiszero != 0 || 5142 (nvpp != NULL && *nvpp == NULL)) { 5143 if (fhiszero != 0) 5144 error = vn_lock(dvp, 5145 LK_EXCLUSIVE); 5146 else if (lktype != 0) 5147 error = vn_lock(dvp, lktype); 5148 else 5149 error = vn_lock(dvp, LK_SHARED); 5150 /* 5151 * If the file handle is all 0's, try to 5152 * do a Lookup against the DS to acquire 5153 * it. 5154 * If dvpp == NULL or the Lookup fails, 5155 * unlock dvp after the call. 5156 */ 5157 if (error == 0 && (fhiszero != 0 || 5158 (nvpp != NULL && *nvpp == NULL))) { 5159 error = nfsrv_pnfslookupds(vp, 5160 dvp, pf, &nvp, p); 5161 if (error == 0) { 5162 if (fhiszero != 0) 5163 nfsrv_pnfssetfh( 5164 vp, pf, 5165 devid, 5166 fnamep, 5167 nvp, p); 5168 if (nvpp != NULL && 5169 *nvpp == NULL) { 5170 *nvpp = nvp; 5171 *dsdirp = dsdir; 5172 } else 5173 vput(nvp); 5174 } 5175 if (error != 0 || lktype == 0) 5176 NFSVOPUNLOCK(dvp); 5177 } 5178 } 5179 if (error == 0) { 5180 gotone++; 5181 NFSD_DEBUG(4, "gotone=%d\n", gotone); 5182 if (devid != NULL) { 5183 NFSBCOPY(fndds->nfsdev_deviceid, 5184 devid, NFSX_V4DEVICEID); 5185 devid += NFSX_V4DEVICEID; 5186 } 5187 if (dvpp != NULL) 5188 *tdvpp++ = dvp; 5189 if (fhp != NULL) 5190 NFSBCOPY(&pf->dsf_fh, fhp++, 5191 NFSX_MYFH); 5192 if (fnamep != NULL && gotone == 1) 5193 strlcpy(fnamep, 5194 pf->dsf_filename, 5195 sizeof(pf->dsf_filename)); 5196 } else 5197 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " 5198 "err=%d\n", error); 5199 } 5200 } 5201 } 5202 if (error == 0 && gotone == 0) 5203 error = ENOENT; 5204 5205 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, 5206 error); 5207 if (error == 0) 5208 *mirrorcntp = gotone; 5209 else { 5210 if (gotone > 0 && dvpp != NULL) { 5211 /* 5212 * If the error didn't occur on the first one and 5213 * dvpp != NULL, the one(s) prior to the failure will 5214 * have locked dvp's that need to be unlocked. 5215 */ 5216 for (i = 0; i < gotone; i++) { 5217 NFSVOPUNLOCK(*dvpp); 5218 *dvpp++ = NULL; 5219 } 5220 } 5221 /* 5222 * If it found the vnode to be copied from before a failure, 5223 * it needs to be vput()'d. 5224 */ 5225 if (nvpp != NULL && *nvpp != NULL) { 5226 vput(*nvpp); 5227 *nvpp = NULL; 5228 } 5229 } 5230 return (error); 5231 } 5232 5233 /* 5234 * Set the extended attribute for the Change attribute. 5235 */ 5236 static int 5237 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 5238 { 5239 struct pnfsdsattr dsattr; 5240 int error; 5241 5242 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); 5243 dsattr.dsa_filerev = nap->na_filerev; 5244 dsattr.dsa_size = nap->na_size; 5245 dsattr.dsa_atime = nap->na_atime; 5246 dsattr.dsa_mtime = nap->na_mtime; 5247 dsattr.dsa_bytes = nap->na_bytes; 5248 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5249 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); 5250 if (error != 0) 5251 printf("pNFS: setextattr=%d\n", error); 5252 return (error); 5253 } 5254 5255 static int 5256 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5257 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) 5258 { 5259 uint32_t *tl; 5260 struct nfsrv_descript *nd; 5261 nfsv4stateid_t st; 5262 struct mbuf *m, *m2; 5263 int error = 0, retlen, tlen, trimlen; 5264 5265 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); 5266 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5267 *mpp = NULL; 5268 /* 5269 * Use a stateid where other is an alternating 01010 pattern and 5270 * seqid is 0xffffffff. This value is not defined as special by 5271 * the RFC and is used by the FreeBSD NFS server to indicate an 5272 * MDS->DS proxy operation. 5273 */ 5274 st.other[0] = 0x55555555; 5275 st.other[1] = 0x55555555; 5276 st.other[2] = 0x55555555; 5277 st.seqid = 0xffffffff; 5278 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5279 NULL, NULL, 0, 0, cred); 5280 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5281 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); 5282 txdr_hyper(off, tl); 5283 *(tl + 2) = txdr_unsigned(len); 5284 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5285 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5286 if (error != 0) { 5287 free(nd, M_TEMP); 5288 return (error); 5289 } 5290 if (nd->nd_repstat == 0) { 5291 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 5292 NFSM_STRSIZ(retlen, len); 5293 if (retlen > 0) { 5294 /* Trim off the pre-data XDR from the mbuf chain. */ 5295 m = nd->nd_mrep; 5296 while (m != NULL && m != nd->nd_md) { 5297 if (m->m_next == nd->nd_md) { 5298 m->m_next = NULL; 5299 m_freem(nd->nd_mrep); 5300 nd->nd_mrep = m = nd->nd_md; 5301 } else 5302 m = m->m_next; 5303 } 5304 if (m == NULL) { 5305 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5306 error = ENOENT; 5307 goto nfsmout; 5308 } 5309 5310 /* 5311 * Now, adjust first mbuf so that any XDR before the 5312 * read data is skipped over. 5313 */ 5314 trimlen = nd->nd_dpos - mtod(m, char *); 5315 if (trimlen > 0) { 5316 m->m_len -= trimlen; 5317 NFSM_DATAP(m, trimlen); 5318 } 5319 5320 /* 5321 * Truncate the mbuf chain at retlen bytes of data, 5322 * plus XDR padding that brings the length up to a 5323 * multiple of 4. 5324 */ 5325 tlen = NFSM_RNDUP(retlen); 5326 do { 5327 if (m->m_len >= tlen) { 5328 m->m_len = tlen; 5329 tlen = 0; 5330 m2 = m->m_next; 5331 m->m_next = NULL; 5332 m_freem(m2); 5333 break; 5334 } 5335 tlen -= m->m_len; 5336 m = m->m_next; 5337 } while (m != NULL); 5338 if (tlen > 0) { 5339 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5340 error = ENOENT; 5341 goto nfsmout; 5342 } 5343 *mpp = nd->nd_mrep; 5344 *mpendp = m; 5345 nd->nd_mrep = NULL; 5346 } 5347 } else 5348 error = nd->nd_repstat; 5349 nfsmout: 5350 /* If nd->nd_mrep is already NULL, this is a no-op. */ 5351 m_freem(nd->nd_mrep); 5352 free(nd, M_TEMP); 5353 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); 5354 return (error); 5355 } 5356 5357 /* 5358 * Do a write RPC on a DS data file, using this structure for the arguments, 5359 * so that this function can be executed by a separate kernel process. 5360 */ 5361 struct nfsrvwritedsdorpc { 5362 int done; 5363 int inprog; 5364 struct task tsk; 5365 fhandle_t fh; 5366 off_t off; 5367 int len; 5368 struct nfsmount *nmp; 5369 struct ucred *cred; 5370 NFSPROC_T *p; 5371 struct mbuf *m; 5372 int err; 5373 }; 5374 5375 static int 5376 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, 5377 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) 5378 { 5379 uint32_t *tl; 5380 struct nfsrv_descript *nd; 5381 nfsattrbit_t attrbits; 5382 nfsv4stateid_t st; 5383 int commit, error, retlen; 5384 5385 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5386 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, 5387 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5388 5389 /* 5390 * Use a stateid where other is an alternating 01010 pattern and 5391 * seqid is 0xffffffff. This value is not defined as special by 5392 * the RFC and is used by the FreeBSD NFS server to indicate an 5393 * MDS->DS proxy operation. 5394 */ 5395 st.other[0] = 0x55555555; 5396 st.other[1] = 0x55555555; 5397 st.other[2] = 0x55555555; 5398 st.seqid = 0xffffffff; 5399 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5400 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); 5401 txdr_hyper(off, tl); 5402 tl += 2; 5403 /* 5404 * Do all writes FileSync, since the server doesn't hold onto dirty 5405 * buffers. Since clients should be accessing the DS servers directly 5406 * using the pNFS layouts, this just needs to work correctly as a 5407 * fallback. 5408 */ 5409 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); 5410 *tl = txdr_unsigned(len); 5411 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); 5412 5413 /* Put data in mbuf chain. */ 5414 nd->nd_mb->m_next = m; 5415 5416 /* Set nd_mb and nd_bpos to end of data. */ 5417 while (m->m_next != NULL) 5418 m = m->m_next; 5419 nd->nd_mb = m; 5420 nfsm_set(nd, m->m_len); 5421 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); 5422 5423 /* Do a Getattr for the attributes that change upon writing. */ 5424 NFSZERO_ATTRBIT(&attrbits); 5425 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5426 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5427 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5428 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5429 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5430 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5431 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5432 (void) nfsrv_putattrbit(nd, &attrbits); 5433 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5434 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5435 if (error != 0) { 5436 free(nd, M_TEMP); 5437 return (error); 5438 } 5439 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); 5440 /* Get rid of weak cache consistency data for now. */ 5441 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5442 (ND_NFSV4 | ND_V4WCCATTR)) { 5443 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5444 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5445 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); 5446 if (error != 0) 5447 goto nfsmout; 5448 /* 5449 * Get rid of Op# and status for next op. 5450 */ 5451 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5452 if (*++tl != 0) 5453 nd->nd_flag |= ND_NOMOREDATA; 5454 } 5455 if (nd->nd_repstat == 0) { 5456 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); 5457 retlen = fxdr_unsigned(int, *tl++); 5458 commit = fxdr_unsigned(int, *tl); 5459 if (commit != NFSWRITE_FILESYNC) 5460 error = NFSERR_IO; 5461 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", 5462 retlen, commit, error); 5463 } else 5464 error = nd->nd_repstat; 5465 /* We have no use for the Write Verifier since we use FileSync. */ 5466 5467 /* 5468 * Get the Change, Size, Access Time and Modify Time attributes and set 5469 * on the Metadata file, so its attributes will be what the file's 5470 * would be if it had been written. 5471 */ 5472 if (error == 0) { 5473 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5474 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5475 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5476 } 5477 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); 5478 nfsmout: 5479 m_freem(nd->nd_mrep); 5480 free(nd, M_TEMP); 5481 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); 5482 return (error); 5483 } 5484 5485 /* 5486 * Start up the thread that will execute nfsrv_writedsdorpc(). 5487 */ 5488 static void 5489 start_writedsdorpc(void *arg, int pending) 5490 { 5491 struct nfsrvwritedsdorpc *drpc; 5492 5493 drpc = (struct nfsrvwritedsdorpc *)arg; 5494 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5495 drpc->len, NULL, drpc->m, drpc->cred, drpc->p); 5496 drpc->done = 1; 5497 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); 5498 } 5499 5500 static int 5501 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5502 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5503 struct mbuf **mpp, char *cp, int *failposp) 5504 { 5505 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; 5506 struct nfsvattr na; 5507 struct mbuf *m; 5508 int error, i, offs, ret, timo; 5509 5510 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); 5511 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); 5512 drpc = NULL; 5513 if (mirrorcnt > 1) 5514 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5515 M_WAITOK); 5516 5517 /* Calculate offset in mbuf chain that data starts. */ 5518 offs = cp - mtod(*mpp, char *); 5519 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); 5520 5521 /* 5522 * Do the write RPC for every DS, using a separate kernel process 5523 * for every DS except the last one. 5524 */ 5525 error = 0; 5526 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5527 tdrpc->done = 0; 5528 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5529 tdrpc->off = off; 5530 tdrpc->len = len; 5531 tdrpc->nmp = *nmpp; 5532 tdrpc->cred = cred; 5533 tdrpc->p = p; 5534 tdrpc->inprog = 0; 5535 tdrpc->err = 0; 5536 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5537 ret = EIO; 5538 if (nfs_pnfsiothreads != 0) { 5539 ret = nfs_pnfsio(start_writedsdorpc, tdrpc); 5540 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", 5541 ret); 5542 } 5543 if (ret != 0) { 5544 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, 5545 tdrpc->m, cred, p); 5546 if (nfsds_failerr(ret) && *failposp == -1) 5547 *failposp = i; 5548 else if (error == 0 && ret != 0) 5549 error = ret; 5550 } 5551 nmpp++; 5552 fhp++; 5553 } 5554 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5555 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); 5556 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5557 *failposp = mirrorcnt - 1; 5558 else if (error == 0 && ret != 0) 5559 error = ret; 5560 if (error == 0) 5561 error = nfsrv_setextattr(vp, &na, p); 5562 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); 5563 tdrpc = drpc; 5564 timo = hz / 50; /* Wait for 20msec. */ 5565 if (timo < 1) 5566 timo = 1; 5567 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5568 /* Wait for RPCs on separate threads to complete. */ 5569 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5570 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); 5571 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5572 *failposp = i; 5573 else if (error == 0 && tdrpc->err != 0) 5574 error = tdrpc->err; 5575 } 5576 free(drpc, M_TEMP); 5577 return (error); 5578 } 5579 5580 /* 5581 * Do a allocate RPC on a DS data file, using this structure for the arguments, 5582 * so that this function can be executed by a separate kernel process. 5583 */ 5584 struct nfsrvallocatedsdorpc { 5585 int done; 5586 int inprog; 5587 struct task tsk; 5588 fhandle_t fh; 5589 off_t off; 5590 off_t len; 5591 struct nfsmount *nmp; 5592 struct ucred *cred; 5593 NFSPROC_T *p; 5594 int err; 5595 }; 5596 5597 static int 5598 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5599 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5600 { 5601 uint32_t *tl; 5602 struct nfsrv_descript *nd; 5603 nfsattrbit_t attrbits; 5604 nfsv4stateid_t st; 5605 int error; 5606 5607 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5608 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, 5609 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5610 5611 /* 5612 * Use a stateid where other is an alternating 01010 pattern and 5613 * seqid is 0xffffffff. This value is not defined as special by 5614 * the RFC and is used by the FreeBSD NFS server to indicate an 5615 * MDS->DS proxy operation. 5616 */ 5617 st.other[0] = 0x55555555; 5618 st.other[1] = 0x55555555; 5619 st.other[2] = 0x55555555; 5620 st.seqid = 0xffffffff; 5621 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5622 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5623 txdr_hyper(off, tl); tl += 2; 5624 txdr_hyper(len, tl); tl += 2; 5625 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); 5626 5627 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5628 NFSGETATTR_ATTRBIT(&attrbits); 5629 nfsrv_putattrbit(nd, &attrbits); 5630 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5631 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5632 if (error != 0) { 5633 free(nd, M_TEMP); 5634 return (error); 5635 } 5636 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", 5637 nd->nd_repstat); 5638 if (nd->nd_repstat == 0) { 5639 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5640 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5641 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5642 } else 5643 error = nd->nd_repstat; 5644 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); 5645 nfsmout: 5646 m_freem(nd->nd_mrep); 5647 free(nd, M_TEMP); 5648 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); 5649 return (error); 5650 } 5651 5652 /* 5653 * Start up the thread that will execute nfsrv_allocatedsdorpc(). 5654 */ 5655 static void 5656 start_allocatedsdorpc(void *arg, int pending) 5657 { 5658 struct nfsrvallocatedsdorpc *drpc; 5659 5660 drpc = (struct nfsrvallocatedsdorpc *)arg; 5661 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5662 drpc->len, NULL, drpc->cred, drpc->p); 5663 drpc->done = 1; 5664 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); 5665 } 5666 5667 static int 5668 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5669 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5670 int *failposp) 5671 { 5672 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; 5673 struct nfsvattr na; 5674 int error, i, ret, timo; 5675 5676 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); 5677 drpc = NULL; 5678 if (mirrorcnt > 1) 5679 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5680 M_WAITOK); 5681 5682 /* 5683 * Do the allocate RPC for every DS, using a separate kernel process 5684 * for every DS except the last one. 5685 */ 5686 error = 0; 5687 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5688 tdrpc->done = 0; 5689 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5690 tdrpc->off = off; 5691 tdrpc->len = len; 5692 tdrpc->nmp = *nmpp; 5693 tdrpc->cred = cred; 5694 tdrpc->p = p; 5695 tdrpc->inprog = 0; 5696 tdrpc->err = 0; 5697 ret = EIO; 5698 if (nfs_pnfsiothreads != 0) { 5699 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); 5700 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", 5701 ret); 5702 } 5703 if (ret != 0) { 5704 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, 5705 cred, p); 5706 if (nfsds_failerr(ret) && *failposp == -1) 5707 *failposp = i; 5708 else if (error == 0 && ret != 0) 5709 error = ret; 5710 } 5711 nmpp++; 5712 fhp++; 5713 } 5714 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5715 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5716 *failposp = mirrorcnt - 1; 5717 else if (error == 0 && ret != 0) 5718 error = ret; 5719 if (error == 0) 5720 error = nfsrv_setextattr(vp, &na, p); 5721 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); 5722 tdrpc = drpc; 5723 timo = hz / 50; /* Wait for 20msec. */ 5724 if (timo < 1) 5725 timo = 1; 5726 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5727 /* Wait for RPCs on separate threads to complete. */ 5728 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5729 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5730 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5731 *failposp = i; 5732 else if (error == 0 && tdrpc->err != 0) 5733 error = tdrpc->err; 5734 } 5735 free(drpc, M_TEMP); 5736 return (error); 5737 } 5738 5739 /* 5740 * Do a deallocate RPC on a DS data file, using this structure for the 5741 * arguments, so that this function can be executed by a separate kernel 5742 * process. 5743 */ 5744 struct nfsrvdeallocatedsdorpc { 5745 int done; 5746 int inprog; 5747 struct task tsk; 5748 fhandle_t fh; 5749 off_t off; 5750 off_t len; 5751 struct nfsmount *nmp; 5752 struct ucred *cred; 5753 NFSPROC_T *p; 5754 int err; 5755 }; 5756 5757 static int 5758 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5759 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5760 { 5761 uint32_t *tl; 5762 struct nfsrv_descript *nd; 5763 nfsattrbit_t attrbits; 5764 nfsv4stateid_t st; 5765 int error; 5766 5767 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5768 nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp, 5769 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5770 5771 /* 5772 * Use a stateid where other is an alternating 01010 pattern and 5773 * seqid is 0xffffffff. This value is not defined as special by 5774 * the RFC and is used by the FreeBSD NFS server to indicate an 5775 * MDS->DS proxy operation. 5776 */ 5777 st.other[0] = 0x55555555; 5778 st.other[1] = 0x55555555; 5779 st.other[2] = 0x55555555; 5780 st.seqid = 0xffffffff; 5781 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5782 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5783 txdr_hyper(off, tl); tl += 2; 5784 txdr_hyper(len, tl); tl += 2; 5785 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len); 5786 5787 /* Do a Getattr for the attributes that change upon writing. */ 5788 NFSZERO_ATTRBIT(&attrbits); 5789 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5790 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5791 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5792 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5793 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5794 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5795 nfsrv_putattrbit(nd, &attrbits); 5796 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5797 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5798 if (error != 0) { 5799 free(nd, M_TEMP); 5800 return (error); 5801 } 5802 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n", 5803 nd->nd_repstat); 5804 /* Get rid of weak cache consistency data for now. */ 5805 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5806 (ND_NFSV4 | ND_V4WCCATTR)) { 5807 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5808 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5809 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error); 5810 if (error != 0) 5811 goto nfsmout; 5812 /* 5813 * Get rid of Op# and status for next op. 5814 */ 5815 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5816 if (*++tl != 0) 5817 nd->nd_flag |= ND_NOMOREDATA; 5818 } 5819 if (nd->nd_repstat == 0) { 5820 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5821 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5822 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5823 } else 5824 error = nd->nd_repstat; 5825 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error); 5826 nfsmout: 5827 m_freem(nd->nd_mrep); 5828 free(nd, M_TEMP); 5829 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error); 5830 return (error); 5831 } 5832 5833 /* 5834 * Start up the thread that will execute nfsrv_deallocatedsdorpc(). 5835 */ 5836 static void 5837 start_deallocatedsdorpc(void *arg, int pending) 5838 { 5839 struct nfsrvdeallocatedsdorpc *drpc; 5840 5841 drpc = (struct nfsrvdeallocatedsdorpc *)arg; 5842 drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5843 drpc->len, NULL, drpc->cred, drpc->p); 5844 drpc->done = 1; 5845 NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err); 5846 } 5847 5848 static int 5849 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5850 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5851 int *failposp) 5852 { 5853 struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL; 5854 struct nfsvattr na; 5855 int error, i, ret, timo; 5856 5857 NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n"); 5858 drpc = NULL; 5859 if (mirrorcnt > 1) 5860 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5861 M_WAITOK); 5862 5863 /* 5864 * Do the deallocate RPC for every DS, using a separate kernel process 5865 * for every DS except the last one. 5866 */ 5867 error = 0; 5868 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5869 tdrpc->done = 0; 5870 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5871 tdrpc->off = off; 5872 tdrpc->len = len; 5873 tdrpc->nmp = *nmpp; 5874 tdrpc->cred = cred; 5875 tdrpc->p = p; 5876 tdrpc->inprog = 0; 5877 tdrpc->err = 0; 5878 ret = EIO; 5879 if (nfs_pnfsiothreads != 0) { 5880 ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc); 5881 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n", 5882 ret); 5883 } 5884 if (ret != 0) { 5885 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, 5886 NULL, cred, p); 5887 if (nfsds_failerr(ret) && *failposp == -1) 5888 *failposp = i; 5889 else if (error == 0 && ret != 0) 5890 error = ret; 5891 } 5892 nmpp++; 5893 fhp++; 5894 } 5895 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5896 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5897 *failposp = mirrorcnt - 1; 5898 else if (error == 0 && ret != 0) 5899 error = ret; 5900 if (error == 0) 5901 error = nfsrv_setextattr(vp, &na, p); 5902 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error); 5903 tdrpc = drpc; 5904 timo = hz / 50; /* Wait for 20msec. */ 5905 if (timo < 1) 5906 timo = 1; 5907 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5908 /* Wait for RPCs on separate threads to complete. */ 5909 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5910 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5911 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5912 *failposp = i; 5913 else if (error == 0 && tdrpc->err != 0) 5914 error = tdrpc->err; 5915 } 5916 free(drpc, M_TEMP); 5917 return (error); 5918 } 5919 5920 static int 5921 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 5922 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, 5923 struct nfsvattr *dsnap) 5924 { 5925 uint32_t *tl; 5926 struct nfsrv_descript *nd; 5927 nfsv4stateid_t st; 5928 nfsattrbit_t attrbits; 5929 int error; 5930 5931 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); 5932 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5933 /* 5934 * Use a stateid where other is an alternating 01010 pattern and 5935 * seqid is 0xffffffff. This value is not defined as special by 5936 * the RFC and is used by the FreeBSD NFS server to indicate an 5937 * MDS->DS proxy operation. 5938 */ 5939 st.other[0] = 0x55555555; 5940 st.other[1] = 0x55555555; 5941 st.other[2] = 0x55555555; 5942 st.seqid = 0xffffffff; 5943 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5944 NULL, NULL, 0, 0, cred); 5945 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5946 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); 5947 5948 /* Do a Getattr for the attributes that change due to writing. */ 5949 NFSZERO_ATTRBIT(&attrbits); 5950 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5951 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5952 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5953 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5954 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5955 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5956 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5957 (void) nfsrv_putattrbit(nd, &attrbits); 5958 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5959 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5960 if (error != 0) { 5961 free(nd, M_TEMP); 5962 return (error); 5963 } 5964 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", 5965 nd->nd_repstat); 5966 /* Get rid of weak cache consistency data for now. */ 5967 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5968 (ND_NFSV4 | ND_V4WCCATTR)) { 5969 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5970 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5971 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); 5972 if (error != 0) 5973 goto nfsmout; 5974 /* 5975 * Get rid of Op# and status for next op. 5976 */ 5977 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5978 if (*++tl != 0) 5979 nd->nd_flag |= ND_NOMOREDATA; 5980 } 5981 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 5982 if (error != 0) 5983 goto nfsmout; 5984 if (nd->nd_repstat != 0) 5985 error = nd->nd_repstat; 5986 /* 5987 * Get the Change, Size, Access Time and Modify Time attributes and set 5988 * on the Metadata file, so its attributes will be what the file's 5989 * would be if it had been written. 5990 */ 5991 if (error == 0) { 5992 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5993 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 5994 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); 5995 } 5996 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); 5997 nfsmout: 5998 m_freem(nd->nd_mrep); 5999 free(nd, M_TEMP); 6000 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); 6001 return (error); 6002 } 6003 6004 struct nfsrvsetattrdsdorpc { 6005 int done; 6006 int inprog; 6007 struct task tsk; 6008 fhandle_t fh; 6009 struct nfsmount *nmp; 6010 struct vnode *vp; 6011 struct ucred *cred; 6012 NFSPROC_T *p; 6013 struct nfsvattr na; 6014 struct nfsvattr dsna; 6015 int err; 6016 }; 6017 6018 /* 6019 * Start up the thread that will execute nfsrv_setattrdsdorpc(). 6020 */ 6021 static void 6022 start_setattrdsdorpc(void *arg, int pending) 6023 { 6024 struct nfsrvsetattrdsdorpc *drpc; 6025 6026 drpc = (struct nfsrvsetattrdsdorpc *)arg; 6027 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, 6028 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); 6029 drpc->done = 1; 6030 } 6031 6032 static int 6033 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6034 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 6035 struct nfsvattr *nap, int *failposp) 6036 { 6037 struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; 6038 struct nfsvattr na; 6039 int error, i, ret, timo; 6040 6041 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); 6042 drpc = NULL; 6043 if (mirrorcnt > 1) 6044 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6045 M_WAITOK); 6046 6047 /* 6048 * Do the setattr RPC for every DS, using a separate kernel process 6049 * for every DS except the last one. 6050 */ 6051 error = 0; 6052 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6053 tdrpc->done = 0; 6054 tdrpc->inprog = 0; 6055 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6056 tdrpc->nmp = *nmpp; 6057 tdrpc->vp = vp; 6058 tdrpc->cred = cred; 6059 tdrpc->p = p; 6060 tdrpc->na = *nap; 6061 tdrpc->err = 0; 6062 ret = EIO; 6063 if (nfs_pnfsiothreads != 0) { 6064 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); 6065 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", 6066 ret); 6067 } 6068 if (ret != 0) { 6069 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, 6070 &na); 6071 if (nfsds_failerr(ret) && *failposp == -1) 6072 *failposp = i; 6073 else if (error == 0 && ret != 0) 6074 error = ret; 6075 } 6076 nmpp++; 6077 fhp++; 6078 } 6079 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); 6080 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6081 *failposp = mirrorcnt - 1; 6082 else if (error == 0 && ret != 0) 6083 error = ret; 6084 if (error == 0) 6085 error = nfsrv_setextattr(vp, &na, p); 6086 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); 6087 tdrpc = drpc; 6088 timo = hz / 50; /* Wait for 20msec. */ 6089 if (timo < 1) 6090 timo = 1; 6091 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6092 /* Wait for RPCs on separate threads to complete. */ 6093 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6094 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); 6095 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6096 *failposp = i; 6097 else if (error == 0 && tdrpc->err != 0) 6098 error = tdrpc->err; 6099 } 6100 free(drpc, M_TEMP); 6101 return (error); 6102 } 6103 6104 /* 6105 * Do a Setattr of an NFSv4 ACL on the DS file. 6106 */ 6107 static int 6108 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6109 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) 6110 { 6111 struct nfsrv_descript *nd; 6112 nfsv4stateid_t st; 6113 nfsattrbit_t attrbits; 6114 int error; 6115 6116 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); 6117 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6118 /* 6119 * Use a stateid where other is an alternating 01010 pattern and 6120 * seqid is 0xffffffff. This value is not defined as special by 6121 * the RFC and is used by the FreeBSD NFS server to indicate an 6122 * MDS->DS proxy operation. 6123 */ 6124 st.other[0] = 0x55555555; 6125 st.other[1] = 0x55555555; 6126 st.other[2] = 0x55555555; 6127 st.seqid = 0xffffffff; 6128 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), 6129 NULL, NULL, 0, 0, cred); 6130 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6131 NFSZERO_ATTRBIT(&attrbits); 6132 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); 6133 /* 6134 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), 6135 * so passing in the metadata "vp" will be ok, since it is of 6136 * the same type (VREG). 6137 */ 6138 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, 6139 NULL, 0, 0, 0, 0, 0, NULL); 6140 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6141 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6142 if (error != 0) { 6143 free(nd, M_TEMP); 6144 return (error); 6145 } 6146 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", 6147 nd->nd_repstat); 6148 error = nd->nd_repstat; 6149 m_freem(nd->nd_mrep); 6150 free(nd, M_TEMP); 6151 return (error); 6152 } 6153 6154 struct nfsrvsetacldsdorpc { 6155 int done; 6156 int inprog; 6157 struct task tsk; 6158 fhandle_t fh; 6159 struct nfsmount *nmp; 6160 struct vnode *vp; 6161 struct ucred *cred; 6162 NFSPROC_T *p; 6163 struct acl *aclp; 6164 int err; 6165 }; 6166 6167 /* 6168 * Start up the thread that will execute nfsrv_setacldsdorpc(). 6169 */ 6170 static void 6171 start_setacldsdorpc(void *arg, int pending) 6172 { 6173 struct nfsrvsetacldsdorpc *drpc; 6174 6175 drpc = (struct nfsrvsetacldsdorpc *)arg; 6176 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, 6177 drpc->vp, drpc->nmp, drpc->aclp); 6178 drpc->done = 1; 6179 } 6180 6181 static int 6182 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6183 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, 6184 int *failposp) 6185 { 6186 struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; 6187 int error, i, ret, timo; 6188 6189 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); 6190 drpc = NULL; 6191 if (mirrorcnt > 1) 6192 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6193 M_WAITOK); 6194 6195 /* 6196 * Do the setattr RPC for every DS, using a separate kernel process 6197 * for every DS except the last one. 6198 */ 6199 error = 0; 6200 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6201 tdrpc->done = 0; 6202 tdrpc->inprog = 0; 6203 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6204 tdrpc->nmp = *nmpp; 6205 tdrpc->vp = vp; 6206 tdrpc->cred = cred; 6207 tdrpc->p = p; 6208 tdrpc->aclp = aclp; 6209 tdrpc->err = 0; 6210 ret = EIO; 6211 if (nfs_pnfsiothreads != 0) { 6212 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); 6213 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", 6214 ret); 6215 } 6216 if (ret != 0) { 6217 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, 6218 aclp); 6219 if (nfsds_failerr(ret) && *failposp == -1) 6220 *failposp = i; 6221 else if (error == 0 && ret != 0) 6222 error = ret; 6223 } 6224 nmpp++; 6225 fhp++; 6226 } 6227 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); 6228 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6229 *failposp = mirrorcnt - 1; 6230 else if (error == 0 && ret != 0) 6231 error = ret; 6232 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); 6233 tdrpc = drpc; 6234 timo = hz / 50; /* Wait for 20msec. */ 6235 if (timo < 1) 6236 timo = 1; 6237 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6238 /* Wait for RPCs on separate threads to complete. */ 6239 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6240 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); 6241 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6242 *failposp = i; 6243 else if (error == 0 && tdrpc->err != 0) 6244 error = tdrpc->err; 6245 } 6246 free(drpc, M_TEMP); 6247 return (error); 6248 } 6249 6250 /* 6251 * Getattr call to the DS for the attributes that change due to writing. 6252 */ 6253 static int 6254 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6255 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) 6256 { 6257 struct nfsrv_descript *nd; 6258 int error; 6259 nfsattrbit_t attrbits; 6260 6261 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); 6262 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6263 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, 6264 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6265 NFSZERO_ATTRBIT(&attrbits); 6266 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6267 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6268 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6269 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6270 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6271 (void) nfsrv_putattrbit(nd, &attrbits); 6272 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6273 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6274 if (error != 0) { 6275 free(nd, M_TEMP); 6276 return (error); 6277 } 6278 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", 6279 nd->nd_repstat); 6280 if (nd->nd_repstat == 0) { 6281 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, 6282 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, 6283 NULL, NULL); 6284 /* 6285 * We can only save the updated values in the extended 6286 * attribute if the vp is exclusively locked. 6287 * This should happen when any of the following operations 6288 * occur on the vnode: 6289 * Close, Delegreturn, LayoutCommit, LayoutReturn 6290 * As such, the updated extended attribute should get saved 6291 * before nfsrv_checkdsattr() returns 0 and allows the cached 6292 * attributes to be returned without calling this function. 6293 */ 6294 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { 6295 error = nfsrv_setextattr(vp, nap, p); 6296 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", 6297 error); 6298 } 6299 } else 6300 error = nd->nd_repstat; 6301 m_freem(nd->nd_mrep); 6302 free(nd, M_TEMP); 6303 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); 6304 return (error); 6305 } 6306 6307 /* 6308 * Seek call to a DS. 6309 */ 6310 static int 6311 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, 6312 struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) 6313 { 6314 uint32_t *tl; 6315 struct nfsrv_descript *nd; 6316 nfsv4stateid_t st; 6317 int error; 6318 6319 NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); 6320 /* 6321 * Use a stateid where other is an alternating 01010 pattern and 6322 * seqid is 0xffffffff. This value is not defined as special by 6323 * the RFC and is used by the FreeBSD NFS server to indicate an 6324 * MDS->DS proxy operation. 6325 */ 6326 st.other[0] = 0x55555555; 6327 st.other[1] = 0x55555555; 6328 st.other[2] = 0x55555555; 6329 st.seqid = 0xffffffff; 6330 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6331 nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, 6332 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6333 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6334 NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); 6335 txdr_hyper(*offp, tl); tl += 2; 6336 *tl = txdr_unsigned(content); 6337 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6338 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6339 if (error != 0) { 6340 free(nd, M_TEMP); 6341 return (error); 6342 } 6343 NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); 6344 if (nd->nd_repstat == 0) { 6345 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); 6346 if (*tl++ == newnfs_true) 6347 *eofp = true; 6348 else 6349 *eofp = false; 6350 *offp = fxdr_hyper(tl); 6351 } else 6352 error = nd->nd_repstat; 6353 nfsmout: 6354 m_freem(nd->nd_mrep); 6355 free(nd, M_TEMP); 6356 NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); 6357 return (error); 6358 } 6359 6360 /* 6361 * Get the device id and file handle for a DS file. 6362 */ 6363 int 6364 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, 6365 fhandle_t *fhp, char *devid) 6366 { 6367 int buflen, error; 6368 char *buf; 6369 6370 buflen = 1024; 6371 buf = malloc(buflen, M_TEMP, M_WAITOK); 6372 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, 6373 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); 6374 free(buf, M_TEMP); 6375 return (error); 6376 } 6377 6378 /* 6379 * Do a Lookup against the DS for the filename. 6380 */ 6381 static int 6382 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, 6383 struct vnode **nvpp, NFSPROC_T *p) 6384 { 6385 struct nameidata named; 6386 struct ucred *tcred; 6387 char *bufp; 6388 u_long *hashp; 6389 struct vnode *nvp; 6390 int error; 6391 6392 tcred = newnfs_getcred(); 6393 named.ni_cnd.cn_nameiop = LOOKUP; 6394 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; 6395 named.ni_cnd.cn_cred = tcred; 6396 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 6397 nfsvno_setpathbuf(&named, &bufp, &hashp); 6398 named.ni_cnd.cn_nameptr = bufp; 6399 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); 6400 strlcpy(bufp, pf->dsf_filename, NAME_MAX); 6401 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); 6402 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 6403 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); 6404 NFSFREECRED(tcred); 6405 nfsvno_relpathbuf(&named); 6406 if (error == 0) 6407 *nvpp = nvp; 6408 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); 6409 return (error); 6410 } 6411 6412 /* 6413 * Set the file handle to the correct one. 6414 */ 6415 static void 6416 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, 6417 char *fnamep, struct vnode *nvp, NFSPROC_T *p) 6418 { 6419 struct nfsnode *np; 6420 int ret = 0; 6421 6422 np = VTONFS(nvp); 6423 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); 6424 /* 6425 * We can only do a vn_set_extattr() if the vnode is exclusively 6426 * locked and vn_start_write() has been done. If devid != NULL or 6427 * fnamep != NULL or the vnode is shared locked, vn_start_write() 6428 * may not have been done. 6429 * If not done now, it will be done on a future call. 6430 */ 6431 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == 6432 LK_EXCLUSIVE) 6433 ret = vn_extattr_set(vp, IO_NODELOCKED, 6434 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), 6435 (char *)pf, p); 6436 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); 6437 } 6438 6439 /* 6440 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point 6441 * when the DS has failed. 6442 */ 6443 void 6444 nfsrv_killrpcs(struct nfsmount *nmp) 6445 { 6446 6447 /* 6448 * Call newnfs_nmcancelreqs() to cause 6449 * any RPCs in progress on the mount point to 6450 * fail. 6451 * This will cause any process waiting for an 6452 * RPC to complete while holding a vnode lock 6453 * on the mounted-on vnode (such as "df" or 6454 * a non-forced "umount") to fail. 6455 * This will unlock the mounted-on vnode so 6456 * a forced dismount can succeed. 6457 * The NFSMNTP_CANCELRPCS flag should be set when this function is 6458 * called. 6459 */ 6460 newnfs_nmcancelreqs(nmp); 6461 } 6462 6463 /* 6464 * Sum up the statfs info for each of the DSs, so that the client will 6465 * receive the total for all DSs. 6466 */ 6467 static int 6468 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) 6469 { 6470 struct statfs *tsf; 6471 struct nfsdevice *ds; 6472 struct vnode **dvpp, **tdvpp, *dvp; 6473 uint64_t tot; 6474 int cnt, error = 0, i; 6475 6476 if (nfsrv_devidcnt <= 0) 6477 return (ENXIO); 6478 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 6479 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 6480 6481 /* Get an array of the dvps for the DSs. */ 6482 tdvpp = dvpp; 6483 i = 0; 6484 NFSDDSLOCK(); 6485 /* First, search for matches for same file system. */ 6486 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6487 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && 6488 fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { 6489 if (++i > nfsrv_devidcnt) 6490 break; 6491 *tdvpp++ = ds->nfsdev_dvp; 6492 } 6493 } 6494 /* 6495 * If no matches for same file system, total all servers not assigned 6496 * to a file system. 6497 */ 6498 if (i == 0) { 6499 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6500 if (ds->nfsdev_nmp != NULL && 6501 ds->nfsdev_mdsisset == 0) { 6502 if (++i > nfsrv_devidcnt) 6503 break; 6504 *tdvpp++ = ds->nfsdev_dvp; 6505 } 6506 } 6507 } 6508 NFSDDSUNLOCK(); 6509 cnt = i; 6510 6511 /* Do a VFS_STATFS() for each of the DSs and sum them up. */ 6512 tdvpp = dvpp; 6513 for (i = 0; i < cnt && error == 0; i++) { 6514 dvp = *tdvpp++; 6515 error = VFS_STATFS(dvp->v_mount, tsf); 6516 if (error == 0) { 6517 if (sf->f_bsize == 0) { 6518 if (tsf->f_bsize > 0) 6519 sf->f_bsize = tsf->f_bsize; 6520 else 6521 sf->f_bsize = 8192; 6522 } 6523 if (tsf->f_blocks > 0) { 6524 if (sf->f_bsize != tsf->f_bsize) { 6525 tot = tsf->f_blocks * tsf->f_bsize; 6526 sf->f_blocks += (tot / sf->f_bsize); 6527 } else 6528 sf->f_blocks += tsf->f_blocks; 6529 } 6530 if (tsf->f_bfree > 0) { 6531 if (sf->f_bsize != tsf->f_bsize) { 6532 tot = tsf->f_bfree * tsf->f_bsize; 6533 sf->f_bfree += (tot / sf->f_bsize); 6534 } else 6535 sf->f_bfree += tsf->f_bfree; 6536 } 6537 if (tsf->f_bavail > 0) { 6538 if (sf->f_bsize != tsf->f_bsize) { 6539 tot = tsf->f_bavail * tsf->f_bsize; 6540 sf->f_bavail += (tot / sf->f_bsize); 6541 } else 6542 sf->f_bavail += tsf->f_bavail; 6543 } 6544 } 6545 } 6546 free(tsf, M_TEMP); 6547 free(dvpp, M_TEMP); 6548 return (error); 6549 } 6550 6551 /* 6552 * Set an NFSv4 acl. 6553 */ 6554 int 6555 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) 6556 { 6557 int error; 6558 6559 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { 6560 error = NFSERR_ATTRNOTSUPP; 6561 goto out; 6562 } 6563 /* 6564 * With NFSv4 ACLs, chmod(2) may need to add additional entries. 6565 * Make sure it has enough room for that - splitting every entry 6566 * into two and appending "canonical six" entries at the end. 6567 * Cribbed out of kern/vfs_acl.c - Rick M. 6568 */ 6569 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { 6570 error = NFSERR_ATTRNOTSUPP; 6571 goto out; 6572 } 6573 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 6574 if (error == 0) { 6575 error = nfsrv_dssetacl(vp, aclp, cred, p); 6576 if (error == ENOENT) 6577 error = 0; 6578 } 6579 6580 out: 6581 NFSEXITCODE(error); 6582 return (error); 6583 } 6584 6585 /* 6586 * Seek vnode op call (actually it is a VOP_IOCTL()). 6587 * This function is called with the vnode locked, but unlocks and vrele()s 6588 * the vp before returning. 6589 */ 6590 int 6591 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, 6592 off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) 6593 { 6594 struct nfsvattr at; 6595 int error, ret; 6596 6597 ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); 6598 /* 6599 * Attempt to seek on a DS file. A return of ENOENT implies 6600 * there is no DS file to seek on. 6601 */ 6602 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, 6603 NULL, NULL, NULL, NULL, offp, content, eofp); 6604 if (error != ENOENT) { 6605 vput(vp); 6606 return (error); 6607 } 6608 6609 /* 6610 * Do the VOP_IOCTL() call. For the case where *offp == file_size, 6611 * VOP_IOCTL() will return ENXIO. However, the correct reply for 6612 * NFSv4.2 is *eofp == true and error == 0 for this case. 6613 */ 6614 NFSVOPUNLOCK(vp); 6615 error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); 6616 *eofp = false; 6617 if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { 6618 /* Handle the cases where we might be at EOF. */ 6619 ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); 6620 if (ret == 0 && *offp == at.na_size) { 6621 *eofp = true; 6622 error = 0; 6623 } 6624 if (ret != 0 && error == 0) 6625 error = ret; 6626 } 6627 vrele(vp); 6628 NFSEXITCODE(error); 6629 return (error); 6630 } 6631 6632 /* 6633 * Allocate vnode op call. 6634 */ 6635 int 6636 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6637 NFSPROC_T *p) 6638 { 6639 int error; 6640 off_t olen; 6641 6642 ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); 6643 /* 6644 * Attempt to allocate on a DS file. A return of ENOENT implies 6645 * there is no DS file to allocate on. 6646 */ 6647 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, 6648 NULL, NULL, NULL, NULL, &len, 0, NULL); 6649 if (error != ENOENT) 6650 return (error); 6651 6652 /* 6653 * Do the actual VOP_ALLOCATE(), looping so long as 6654 * progress is being made, to achieve completion. 6655 */ 6656 do { 6657 olen = len; 6658 error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred); 6659 if (error == 0 && len > 0 && olen > len) 6660 maybe_yield(); 6661 } while (error == 0 && len > 0 && olen > len); 6662 if (error == 0 && len > 0) 6663 error = NFSERR_IO; 6664 NFSEXITCODE(error); 6665 return (error); 6666 } 6667 6668 /* 6669 * Deallocate vnode op call. 6670 */ 6671 int 6672 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6673 NFSPROC_T *p) 6674 { 6675 int error; 6676 off_t olen; 6677 6678 ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp"); 6679 /* 6680 * Attempt to deallocate on a DS file. A return of ENOENT implies 6681 * there is no DS file to deallocate on. 6682 */ 6683 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL, 6684 NULL, NULL, NULL, NULL, &len, 0, NULL); 6685 if (error != ENOENT) 6686 return (error); 6687 6688 /* 6689 * Do the actual VOP_DEALLOCATE(), looping so long as 6690 * progress is being made, to achieve completion. 6691 */ 6692 do { 6693 olen = len; 6694 error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred); 6695 if (error == 0 && len > 0 && olen > len) 6696 maybe_yield(); 6697 } while (error == 0 && len > 0 && olen > len); 6698 if (error == 0 && len > 0) 6699 error = NFSERR_IO; 6700 NFSEXITCODE(error); 6701 return (error); 6702 } 6703 6704 /* 6705 * Get Extended Atribute vnode op into an mbuf list. 6706 */ 6707 int 6708 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, 6709 struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p, 6710 struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 6711 { 6712 struct iovec *iv; 6713 struct uio io, *uiop = &io; 6714 struct mbuf *m, *m2; 6715 int alen, error, len, tlen; 6716 size_t siz; 6717 6718 /* First, find out the size of the extended attribute. */ 6719 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6720 &siz, cred, p); 6721 if (error != 0) 6722 return (NFSERR_NOXATTR); 6723 if (siz > maxresp - NFS_MAXXDR) 6724 return (NFSERR_XATTR2BIG); 6725 len = siz; 6726 tlen = NFSM_RNDUP(len); 6727 if (tlen > 0) { 6728 /* 6729 * If cnt > MCLBYTES and the reply will not be saved, use 6730 * ext_pgs mbufs for TLS. 6731 * For NFSv4.0, we do not know for sure if the reply will 6732 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 6733 * Always use ext_pgs mbufs if ND_EXTPG is set. 6734 */ 6735 if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES && 6736 (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS && 6737 (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)) 6738 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen, 6739 maxextsiz, &m, &m2, &iv); 6740 else 6741 uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, 6742 &iv); 6743 uiop->uio_iov = iv; 6744 } else { 6745 uiop->uio_iovcnt = 0; 6746 uiop->uio_iov = iv = NULL; 6747 m = m2 = NULL; 6748 } 6749 uiop->uio_offset = 0; 6750 uiop->uio_resid = tlen; 6751 uiop->uio_rw = UIO_READ; 6752 uiop->uio_segflg = UIO_SYSSPACE; 6753 uiop->uio_td = p; 6754 #ifdef MAC 6755 error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6756 name); 6757 if (error != 0) 6758 goto out; 6759 #endif 6760 6761 if (tlen > 0) 6762 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 6763 NULL, cred, p); 6764 if (error != 0) 6765 goto out; 6766 if (uiop->uio_resid > 0) { 6767 alen = tlen; 6768 len = tlen - uiop->uio_resid; 6769 tlen = NFSM_RNDUP(len); 6770 if (alen != tlen) 6771 printf("nfsvno_getxattr: weird size read\n"); 6772 if (tlen == 0) { 6773 m_freem(m); 6774 m = m2 = NULL; 6775 } else if (alen != tlen || tlen != len) 6776 m2 = nfsrv_adj(m, alen - tlen, tlen - len); 6777 } 6778 *lenp = len; 6779 *mpp = m; 6780 *mpendp = m2; 6781 6782 out: 6783 if (error != 0) { 6784 if (m != NULL) 6785 m_freem(m); 6786 *lenp = 0; 6787 } 6788 free(iv, M_TEMP); 6789 NFSEXITCODE(error); 6790 return (error); 6791 } 6792 6793 /* 6794 * Set Extended attribute vnode op from an mbuf list. 6795 */ 6796 int 6797 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, 6798 char *cp, struct ucred *cred, struct thread *p) 6799 { 6800 struct iovec *iv; 6801 struct uio uio, *uiop = &uio; 6802 int cnt, error; 6803 6804 error = 0; 6805 #ifdef MAC 6806 error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6807 name); 6808 #endif 6809 if (error != 0) 6810 goto out; 6811 6812 uiop->uio_rw = UIO_WRITE; 6813 uiop->uio_segflg = UIO_SYSSPACE; 6814 uiop->uio_td = p; 6815 uiop->uio_offset = 0; 6816 uiop->uio_resid = len; 6817 if (len > 0) { 6818 error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); 6819 uiop->uio_iov = iv; 6820 uiop->uio_iovcnt = cnt; 6821 } else { 6822 uiop->uio_iov = iv = NULL; 6823 uiop->uio_iovcnt = 0; 6824 } 6825 if (error == 0) { 6826 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 6827 cred, p); 6828 if (error == 0) { 6829 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 6830 nfsvno_updateds(vp, cred, p); 6831 error = VOP_FSYNC(vp, MNT_WAIT, p); 6832 } 6833 free(iv, M_TEMP); 6834 } 6835 6836 out: 6837 NFSEXITCODE(error); 6838 return (error); 6839 } 6840 6841 /* 6842 * For a pNFS server, the DS file's ctime and 6843 * va_filerev (TimeMetadata and Change) needs to 6844 * be updated. This is a hack, but works by 6845 * flipping the S_ISGID bit in va_mode and then 6846 * flipping it back. 6847 * It does result in two MDS->DS RPCs, but creating 6848 * a custom RPC just to do this seems overkill, since 6849 * Setxattr/Rmxattr will not be done that frequently. 6850 * If it fails part way through, that is not too 6851 * serious, since the DS file is never executed. 6852 */ 6853 static void 6854 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p) 6855 { 6856 struct nfsvattr nva; 6857 int ret; 6858 u_short tmode; 6859 6860 ret = VOP_GETATTR(vp, &nva.na_vattr, cred); 6861 if (ret == 0) { 6862 tmode = nva.na_mode; 6863 NFSVNO_ATTRINIT(&nva); 6864 tmode ^= S_ISGID; 6865 NFSVNO_SETATTRVAL(&nva, mode, tmode); 6866 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 6867 NFSPROC_SETATTR, NULL, NULL, NULL, &nva, 6868 NULL, NULL, 0, NULL); 6869 if (ret == 0) { 6870 tmode ^= S_ISGID; 6871 NFSVNO_SETATTRVAL(&nva, mode, tmode); 6872 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 6873 NFSPROC_SETATTR, NULL, NULL, NULL, 6874 &nva, NULL, NULL, 0, NULL); 6875 } 6876 } 6877 } 6878 6879 /* 6880 * Remove Extended attribute vnode op. 6881 */ 6882 int 6883 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, 6884 struct ucred *cred, struct thread *p) 6885 { 6886 int error; 6887 6888 /* 6889 * Get rid of any delegations. I am not sure why this is required, 6890 * but RFC-8276 says so. 6891 */ 6892 error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); 6893 if (error != 0) 6894 goto out; 6895 #ifdef MAC 6896 error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, 6897 name); 6898 if (error != 0) 6899 goto out; 6900 #endif 6901 6902 error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); 6903 if (error == EOPNOTSUPP) 6904 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6905 cred, p); 6906 if (error == 0) { 6907 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 6908 nfsvno_updateds(vp, cred, p); 6909 error = VOP_FSYNC(vp, MNT_WAIT, p); 6910 } 6911 out: 6912 NFSEXITCODE(error); 6913 return (error); 6914 } 6915 6916 /* 6917 * List Extended Atribute vnode op into an mbuf list. 6918 */ 6919 int 6920 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, 6921 struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) 6922 { 6923 struct iovec iv; 6924 struct uio io; 6925 int error; 6926 size_t siz; 6927 6928 *bufp = NULL; 6929 /* First, find out the size of the extended attribute. */ 6930 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, 6931 p); 6932 if (error != 0) 6933 return (NFSERR_NOXATTR); 6934 if (siz <= cookie) { 6935 *lenp = 0; 6936 *eofp = true; 6937 goto out; 6938 } 6939 if (siz > cookie + *lenp) { 6940 siz = cookie + *lenp; 6941 *eofp = false; 6942 } else 6943 *eofp = true; 6944 /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ 6945 if (siz > 10 * 1024 * 1024) { 6946 error = NFSERR_XATTR2BIG; 6947 goto out; 6948 } 6949 *bufp = malloc(siz, M_TEMP, M_WAITOK); 6950 iv.iov_base = *bufp; 6951 iv.iov_len = siz; 6952 io.uio_iovcnt = 1; 6953 io.uio_iov = &iv; 6954 io.uio_offset = 0; 6955 io.uio_resid = siz; 6956 io.uio_rw = UIO_READ; 6957 io.uio_segflg = UIO_SYSSPACE; 6958 io.uio_td = p; 6959 #ifdef MAC 6960 error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); 6961 if (error != 0) 6962 goto out; 6963 #endif 6964 6965 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, 6966 p); 6967 if (error != 0) 6968 goto out; 6969 if (io.uio_resid > 0) 6970 siz -= io.uio_resid; 6971 *lenp = siz; 6972 6973 out: 6974 if (error != 0) { 6975 free(*bufp, M_TEMP); 6976 *bufp = NULL; 6977 } 6978 NFSEXITCODE(error); 6979 return (error); 6980 } 6981 6982 /* 6983 * Trim trailing data off the mbuf list being built. 6984 */ 6985 void 6986 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos, 6987 int bextpg, int bextpgsiz) 6988 { 6989 vm_page_t pg; 6990 int fullpgsiz, i; 6991 6992 if (mb->m_next != NULL) { 6993 m_freem(mb->m_next); 6994 mb->m_next = NULL; 6995 } 6996 if ((mb->m_flags & M_EXTPG) != 0) { 6997 KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs, 6998 ("nfsm_trimtrailing: bextpg out of range")); 6999 KASSERT(bpos == (char *)(void *) 7000 PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz, 7001 ("nfsm_trimtrailing: bextpgsiz bad!")); 7002 7003 /* First, get rid of any pages after this position. */ 7004 for (i = mb->m_epg_npgs - 1; i > bextpg; i--) { 7005 pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]); 7006 vm_page_unwire_noq(pg); 7007 vm_page_free(pg); 7008 } 7009 mb->m_epg_npgs = bextpg + 1; 7010 if (bextpg == 0) 7011 fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off; 7012 else 7013 fullpgsiz = PAGE_SIZE; 7014 mb->m_epg_last_len = fullpgsiz - bextpgsiz; 7015 mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off); 7016 for (i = 1; i < mb->m_epg_npgs; i++) 7017 mb->m_len += m_epg_pagelen(mb, i, 0); 7018 nd->nd_bextpgsiz = bextpgsiz; 7019 nd->nd_bextpg = bextpg; 7020 } else 7021 mb->m_len = bpos - mtod(mb, char *); 7022 nd->nd_mb = mb; 7023 nd->nd_bpos = bpos; 7024 } 7025 7026 7027 /* 7028 * Check to see if a put file handle operation should test for 7029 * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR. 7030 * When Open is the next operation, NFSERR_WRONGSEC cannot be 7031 * replied for the Open cases that use a component. This can 7032 * be identified by the fact that the file handle's type is VDIR. 7033 */ 7034 bool 7035 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, __enum_uint8(vtype) vtyp) 7036 { 7037 7038 if ((nd->nd_flag & ND_NFSV4) == 0) 7039 return (true); 7040 7041 if ((nd->nd_flag & ND_LASTOP) != 0) 7042 return (false); 7043 7044 if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH || 7045 nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH || 7046 nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP || 7047 nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME) 7048 return (false); 7049 if (nextop == NFSV4OP_OPEN && vtyp == VDIR) 7050 return (false); 7051 return (true); 7052 } 7053 7054 /* 7055 * Check DSs marked no space. 7056 */ 7057 void 7058 nfsrv_checknospc(void) 7059 { 7060 struct statfs *tsf; 7061 struct nfsdevice *ds; 7062 struct vnode **dvpp, **tdvpp, *dvp; 7063 char *devid, *tdevid; 7064 int cnt, error = 0, i; 7065 7066 if (nfsrv_devidcnt <= 0) 7067 return; 7068 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 7069 devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK); 7070 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 7071 7072 /* Get an array of the dvps for the DSs. */ 7073 tdvpp = dvpp; 7074 tdevid = devid; 7075 i = 0; 7076 NFSDDSLOCK(); 7077 /* First, search for matches for same file system. */ 7078 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7079 if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) { 7080 if (++i > nfsrv_devidcnt) 7081 break; 7082 *tdvpp++ = ds->nfsdev_dvp; 7083 NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID); 7084 tdevid += NFSX_V4DEVICEID; 7085 } 7086 } 7087 NFSDDSUNLOCK(); 7088 7089 /* Do a VFS_STATFS() for each of the DSs and clear no space. */ 7090 cnt = i; 7091 tdvpp = dvpp; 7092 tdevid = devid; 7093 for (i = 0; i < cnt && error == 0; i++) { 7094 dvp = *tdvpp++; 7095 error = VFS_STATFS(dvp->v_mount, tsf); 7096 if (error == 0 && tsf->f_bavail > 0) { 7097 NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n"); 7098 nfsrv_marknospc(tdevid, false); 7099 } 7100 tdevid += NFSX_V4DEVICEID; 7101 } 7102 free(tsf, M_TEMP); 7103 free(dvpp, M_TEMP); 7104 free(devid, M_TEMP); 7105 } 7106 7107 /* 7108 * Initialize everything that needs to be initialized for a vnet. 7109 */ 7110 static void 7111 nfsrv_vnetinit(const void *unused __unused) 7112 { 7113 7114 nfsd_mntinit(); 7115 } 7116 VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY, 7117 nfsrv_vnetinit, NULL); 7118 7119 /* 7120 * Clean up everything that is in a vnet and needs to be 7121 * done when the jail is destroyed or the module unloaded. 7122 */ 7123 static void 7124 nfsrv_cleanup(const void *unused __unused) 7125 { 7126 int i; 7127 7128 NFSD_LOCK(); 7129 if (!NFSD_VNET(nfsrv_mntinited)) { 7130 NFSD_UNLOCK(); 7131 return; 7132 } 7133 NFSD_VNET(nfsrv_mntinited) = false; 7134 NFSD_UNLOCK(); 7135 7136 /* Clean out all NFSv4 state. */ 7137 nfsrv_throwawayallstate(curthread); 7138 7139 /* Clean the NFS server reply cache */ 7140 nfsrvd_cleancache(); 7141 7142 /* Clean out v4root exports. */ 7143 if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) { 7144 vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export); 7145 free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT); 7146 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 7147 } 7148 7149 /* Free up the krpc server pool. */ 7150 if (NFSD_VNET(nfsrvd_pool) != NULL) 7151 svcpool_destroy(NFSD_VNET(nfsrvd_pool)); 7152 7153 /* and get rid of the locks */ 7154 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 7155 mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx); 7156 mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx); 7157 } 7158 mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx); 7159 for (i = 0; i < nfsrv_sessionhashsize; i++) 7160 mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx); 7161 lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock); 7162 free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE); 7163 free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE); 7164 free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE); 7165 free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT); 7166 free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE); 7167 free(NFSD_VNET(nfssessionhash), M_NFSDSESSION); 7168 free(NFSD_VNET(nfsv4root_mnt), M_TEMP); 7169 NFSD_VNET(nfsv4root_mnt) = NULL; 7170 } 7171 VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY, 7172 nfsrv_cleanup, NULL); 7173 7174 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); 7175 7176 /* 7177 * Called once to initialize data structures... 7178 */ 7179 static int 7180 nfsd_modevent(module_t mod, int type, void *data) 7181 { 7182 int error = 0, i; 7183 static int loaded = 0; 7184 7185 switch (type) { 7186 case MOD_LOAD: 7187 if (loaded) 7188 goto out; 7189 newnfs_portinit(); 7190 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); 7191 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); 7192 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); 7193 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); 7194 #ifdef VV_DISABLEDELEG 7195 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; 7196 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; 7197 #endif 7198 nfsd_call_nfsd = nfssvc_nfsd; 7199 loaded = 1; 7200 break; 7201 7202 case MOD_UNLOAD: 7203 if (newnfs_numnfsd != 0) { 7204 error = EBUSY; 7205 break; 7206 } 7207 7208 #ifdef VV_DISABLEDELEG 7209 vn_deleg_ops.vndeleg_recall = NULL; 7210 vn_deleg_ops.vndeleg_disable = NULL; 7211 #endif 7212 nfsd_call_nfsd = NULL; 7213 mtx_destroy(&nfsrc_udpmtx); 7214 mtx_destroy(&nfs_v4root_mutex); 7215 mtx_destroy(&nfsrv_dontlistlock_mtx); 7216 mtx_destroy(&nfsrv_recalllock_mtx); 7217 if (nfslayouthash != NULL) { 7218 for (i = 0; i < nfsrv_layouthashsize; i++) 7219 mtx_destroy(&nfslayouthash[i].mtx); 7220 free(nfslayouthash, M_NFSDSESSION); 7221 } 7222 loaded = 0; 7223 break; 7224 default: 7225 error = EOPNOTSUPP; 7226 break; 7227 } 7228 7229 out: 7230 NFSEXITCODE(error); 7231 return (error); 7232 } 7233 static moduledata_t nfsd_mod = { 7234 "nfsd", 7235 nfsd_modevent, 7236 NULL, 7237 }; 7238 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); 7239 7240 /* So that loader and kldload(2) can find us, wherever we are.. */ 7241 MODULE_VERSION(nfsd, 1); 7242 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); 7243 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); 7244 MODULE_DEPEND(nfsd, krpc, 1, 1, 1); 7245 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); 7246