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/capsicum.h> 37 #include <sys/extattr.h> 38 39 /* 40 * Functions that perform the vfs operations required by the routines in 41 * nfsd_serv.c. It is hoped that this change will make the server more 42 * portable. 43 */ 44 45 #include <fs/nfs/nfsport.h> 46 #include <security/mac/mac_framework.h> 47 #include <sys/callout.h> 48 #include <sys/filio.h> 49 #include <sys/hash.h> 50 #include <sys/osd.h> 51 #include <sys/sysctl.h> 52 #include <nlm/nlm_prot.h> 53 #include <nlm/nlm.h> 54 #include <vm/vm_param.h> 55 #include <vm/vnode_pager.h> 56 57 FEATURE(nfsd, "NFSv4 server"); 58 59 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 60 extern int nfsrv_useacl; 61 extern int newnfs_numnfsd; 62 extern int nfsrv_sessionhashsize; 63 extern struct nfslayouthash *nfslayouthash; 64 extern int nfsrv_layouthashsize; 65 extern struct mtx nfsrv_dslock_mtx; 66 extern int nfs_pnfsiothreads; 67 extern volatile int nfsrv_devidcnt; 68 extern int nfsrv_maxpnfsmirror; 69 extern uint32_t nfs_srvmaxio; 70 extern int nfs_bufpackets; 71 extern u_long sb_max_adj; 72 extern struct nfsv4lock nfsv4rootfs_lock; 73 74 NFSD_VNET_DECLARE(int, nfsrv_numnfsd); 75 NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst); 76 NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool); 77 NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash); 78 NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash); 79 NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash); 80 NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock); 81 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p); 82 83 NFSDLOCKMUTEX; 84 NFSSTATESPINLOCK; 85 struct mtx nfsrc_udpmtx; 86 struct mtx nfs_v4root_mutex; 87 struct mtx nfsrv_dontlistlock_mtx; 88 struct mtx nfsrv_recalllock_mtx; 89 struct nfsrvfh nfs_pubfh; 90 int nfs_pubfhset = 0; 91 int nfsd_debuglevel = 0; 92 static pid_t nfsd_master_pid = (pid_t)-1; 93 static char nfsd_master_comm[MAXCOMLEN + 1]; 94 static struct timeval nfsd_master_start; 95 static uint32_t nfsv4_sysid = 0; 96 static fhandle_t zerofh; 97 98 NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL; 99 NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl); 100 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table); 101 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table); 102 NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh); 103 NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0; 104 NFSD_VNET_DEFINE(struct callout, nfsd_callout); 105 NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt); 106 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt); 107 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt); 108 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false; 109 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false; 110 111 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, 112 struct ucred *); 113 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *); 114 115 int nfsrv_enable_crossmntpt = 1; 116 static int nfs_commit_blks; 117 static int nfs_commit_miss; 118 extern int nfsrv_issuedelegs; 119 extern int nfsrv_dolocallocks; 120 extern struct nfsdevicehead nfsrv_devidhead; 121 122 /* Map d_type to vnode type. */ 123 static uint8_t dtype_to_vnode[DT_WHT + 1] = { VNON, VFIFO, VCHR, VNON, VDIR, 124 VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON }; 125 126 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **, 127 struct iovec **); 128 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **, 129 struct mbuf **, struct iovec **); 130 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **, 131 int *); 132 static void nfs_dtypetovtype(struct nfsvattr *, struct vnode *, uint8_t); 133 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *, 134 NFSPROC_T *); 135 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **, 136 int *, char *, fhandle_t *); 137 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *, 138 NFSPROC_T *); 139 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *, 140 struct thread *, int, struct mbuf **, char *, struct mbuf **, 141 struct nfsvattr *, struct acl *, off_t *, int, bool *); 142 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *); 143 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *, 144 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **); 145 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *, 146 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **, 147 char *, int *); 148 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 149 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 150 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *, 151 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *); 152 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 153 struct vnode *, struct nfsmount **, int, struct acl *, int *); 154 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 155 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *); 156 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *, 157 struct vnode *, struct nfsmount *, struct nfsvattr *); 158 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *, 159 NFSPROC_T *, struct nfsmount *); 160 static int nfsrv_putfhname(fhandle_t *, char *); 161 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *, 162 struct pnfsdsfile *, struct vnode **, NFSPROC_T *); 163 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *, 164 struct vnode *, NFSPROC_T *); 165 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *); 166 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *, 167 NFSPROC_T *); 168 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *); 169 170 int nfs_pnfsio(task_fn_t *, void *); 171 172 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 173 "NFS server"); 174 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, 175 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); 176 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 177 0, ""); 178 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 179 0, ""); 180 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, 181 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); 182 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel, 183 0, "Debug level for NFS server"); 184 NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid); 185 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, 186 CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid), 187 0, "Enable nfsd to accept numeric owner_names"); 188 static int nfsrv_pnfsgetdsattr = 1; 189 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW, 190 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC"); 191 static bool nfsrv_recalldeleg = false; 192 SYSCTL_BOOL(_vfs_nfsd, OID_AUTO, recalldeleg, CTLFLAG_RW, 193 &nfsrv_recalldeleg, 0, 194 "When set remove/rename recalls delegations for same client"); 195 196 /* 197 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are 198 * not running. 199 * The dsN subdirectories for the increased values must have been created 200 * on all DS servers before this increase is done. 201 */ 202 u_int nfsrv_dsdirsize = 20; 203 static int 204 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS) 205 { 206 int error, newdsdirsize; 207 208 newdsdirsize = nfsrv_dsdirsize; 209 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req); 210 if (error != 0 || req->newptr == NULL) 211 return (error); 212 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 || 213 newnfs_numnfsd != 0) 214 return (EINVAL); 215 nfsrv_dsdirsize = newdsdirsize; 216 return (0); 217 } 218 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, 219 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize), 220 sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers"); 221 222 /* 223 * nfs_srvmaxio can only be increased and only when the nfsd threads are 224 * not running. The setting must be a power of 2, with the current limit of 225 * 1Mbyte. 226 */ 227 static int 228 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS) 229 { 230 int error; 231 u_int newsrvmaxio; 232 uint64_t tval; 233 234 newsrvmaxio = nfs_srvmaxio; 235 error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req); 236 if (error != 0 || req->newptr == NULL) 237 return (error); 238 if (newsrvmaxio == nfs_srvmaxio) 239 return (0); 240 if (newsrvmaxio < nfs_srvmaxio) { 241 printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n"); 242 return (EINVAL); 243 } 244 if (newsrvmaxio > 1048576) { 245 printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n"); 246 return (EINVAL); 247 } 248 if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) { 249 printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n"); 250 return (EINVAL); 251 } 252 253 /* 254 * Check that kern.ipc.maxsockbuf is large enough for 255 * newsrviomax, given the setting of vfs.nfs.bufpackets. 256 */ 257 if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets > 258 sb_max_adj) { 259 /* 260 * Suggest vfs.nfs.bufpackets * maximum RPC message for 261 * sb_max_adj. 262 */ 263 tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets; 264 265 /* 266 * Convert suggested sb_max_adj value to a suggested 267 * sb_max value, which is what is set via kern.ipc.maxsockbuf. 268 * Perform the inverse calculation of (from uipc_sockbuf.c): 269 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES / 270 * (MSIZE + MCLBYTES); 271 * XXX If the calculation of sb_max_adj from sb_max changes, 272 * this calculation must be changed as well. 273 */ 274 tval *= (MSIZE + MCLBYTES); /* Brackets for readability. */ 275 tval += MCLBYTES - 1; /* Round up divide. */ 276 tval /= MCLBYTES; 277 printf("nfsd: set kern.ipc.maxsockbuf to a minimum of " 278 "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval, 279 newsrvmaxio); 280 return (EINVAL); 281 } 282 283 NFSD_LOCK(); 284 if (newnfs_numnfsd != 0) { 285 NFSD_UNLOCK(); 286 printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd " 287 "threads are running\n"); 288 return (EINVAL); 289 } 290 291 292 nfs_srvmaxio = newsrvmaxio; 293 NFSD_UNLOCK(); 294 return (0); 295 } 296 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio, 297 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 298 sysctl_srvmaxio, "IU", "Maximum I/O size in bytes"); 299 300 static int 301 sysctl_dolocallocks(SYSCTL_HANDLER_ARGS) 302 { 303 int error, igotlock, newdolocallocks; 304 305 newdolocallocks = nfsrv_dolocallocks; 306 error = sysctl_handle_int(oidp, &newdolocallocks, 0, req); 307 if (error != 0 || req->newptr == NULL) 308 return (error); 309 if (newdolocallocks == nfsrv_dolocallocks) 310 return (0); 311 if (jailed(curthread->td_ucred)) 312 return (EINVAL); 313 314 NFSLOCKV4ROOTMUTEX(); 315 do { 316 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 317 NFSV4ROOTLOCKMUTEXPTR, NULL); 318 } while (!igotlock); 319 NFSUNLOCKV4ROOTMUTEX(); 320 321 nfsrv_dolocallocks = newdolocallocks; 322 323 NFSLOCKV4ROOTMUTEX(); 324 nfsv4_unlock(&nfsv4rootfs_lock, 0); 325 NFSUNLOCKV4ROOTMUTEX(); 326 return (0); 327 } 328 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, enable_locallocks, 329 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 330 sysctl_dolocallocks, "IU", "Enable nfsd to acquire local locks on files"); 331 332 #define MAX_REORDERED_RPC 16 333 #define NUM_HEURISTIC 1031 334 #define NHUSE_INIT 64 335 #define NHUSE_INC 16 336 #define NHUSE_MAX 2048 337 338 static struct nfsheur { 339 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 340 off_t nh_nextoff; /* next offset for sequential detection */ 341 int nh_use; /* use count for selection */ 342 int nh_seqcount; /* heuristic */ 343 } nfsheur[NUM_HEURISTIC]; 344 345 /* 346 * Heuristic to detect sequential operation. 347 */ 348 static struct nfsheur * 349 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 350 { 351 struct nfsheur *nh; 352 int hi, try; 353 354 /* Locate best candidate. */ 355 try = 32; 356 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 357 nh = &nfsheur[hi]; 358 while (try--) { 359 if (nfsheur[hi].nh_vp == vp) { 360 nh = &nfsheur[hi]; 361 break; 362 } 363 if (nfsheur[hi].nh_use > 0) 364 --nfsheur[hi].nh_use; 365 hi = (hi + 1) % NUM_HEURISTIC; 366 if (nfsheur[hi].nh_use < nh->nh_use) 367 nh = &nfsheur[hi]; 368 } 369 370 /* Initialize hint if this is a new file. */ 371 if (nh->nh_vp != vp) { 372 nh->nh_vp = vp; 373 nh->nh_nextoff = uio->uio_offset; 374 nh->nh_use = NHUSE_INIT; 375 if (uio->uio_offset == 0) 376 nh->nh_seqcount = 4; 377 else 378 nh->nh_seqcount = 1; 379 } 380 381 /* Calculate heuristic. */ 382 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 383 uio->uio_offset == nh->nh_nextoff) { 384 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 385 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 386 if (nh->nh_seqcount > IO_SEQMAX) 387 nh->nh_seqcount = IO_SEQMAX; 388 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 389 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 390 /* Probably a reordered RPC, leave seqcount alone. */ 391 } else if (nh->nh_seqcount > 1) { 392 nh->nh_seqcount /= 2; 393 } else { 394 nh->nh_seqcount = 0; 395 } 396 nh->nh_use += NHUSE_INC; 397 if (nh->nh_use > NHUSE_MAX) 398 nh->nh_use = NHUSE_MAX; 399 return (nh); 400 } 401 402 /* 403 * Get attributes into nfsvattr structure. 404 */ 405 int 406 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, 407 struct nfsrv_descript *nd, struct thread *p, int vpislocked, 408 nfsattrbit_t *attrbitp) 409 { 410 int error, gotattr, lockedit = 0; 411 struct nfsvattr na; 412 413 if (vpislocked == 0) { 414 /* 415 * When vpislocked == 0, the vnode is either exclusively 416 * locked by this thread or not locked by this thread. 417 * As such, shared lock it, if not exclusively locked. 418 */ 419 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 420 lockedit = 1; 421 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); 422 } 423 } 424 425 /* 426 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed 427 * attributes, as required. 428 * This needs to be done for regular files if: 429 * - non-NFSv4 RPCs or 430 * - when attrbitp == NULL or 431 * - an NFSv4 RPC with any of the above attributes in attrbitp. 432 * A return of 0 for nfsrv_proxyds() indicates that it has acquired 433 * these attributes. nfsrv_proxyds() will return an error if the 434 * server is not a pNFS one. 435 */ 436 gotattr = 0; 437 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL || 438 (nd->nd_flag & ND_NFSV4) == 0 || 439 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) || 440 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) || 441 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) || 442 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) || 443 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) { 444 error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p, 445 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0, 446 NULL); 447 if (error == 0) 448 gotattr = 1; 449 } 450 451 nvap->na_bsdflags = 0; 452 nvap->na_flags = 0; 453 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred); 454 if (lockedit != 0) 455 NFSVOPUNLOCK(vp); 456 457 /* 458 * If we got the Change, Size and Modify Time from the DS, 459 * replace them. 460 */ 461 if (gotattr != 0) { 462 nvap->na_atime = na.na_atime; 463 nvap->na_mtime = na.na_mtime; 464 nvap->na_filerev = na.na_filerev; 465 nvap->na_size = na.na_size; 466 nvap->na_bytes = na.na_bytes; 467 } 468 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr, 469 error, (uintmax_t)na.na_filerev); 470 471 NFSEXITCODE(error); 472 return (error); 473 } 474 475 /* 476 * Get a file handle for a vnode. 477 */ 478 int 479 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) 480 { 481 int error; 482 483 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); 484 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 485 error = VOP_VPTOFH(vp, &fhp->fh_fid); 486 487 NFSEXITCODE(error); 488 return (error); 489 } 490 491 /* 492 * Perform access checking for vnodes obtained from file handles that would 493 * refer to files already opened by a Unix client. You cannot just use 494 * vn_writechk() and VOP_ACCESSX() for two reasons. 495 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 496 * case. 497 * 2 - The owner is to be given access irrespective of mode bits for some 498 * operations, so that processes that chmod after opening a file don't 499 * break. 500 */ 501 int 502 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, 503 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, 504 u_int32_t *supportedtypep) 505 { 506 struct vattr vattr; 507 int error = 0, getret = 0; 508 509 if (vpislocked == 0) { 510 if (NFSVOPLOCK(vp, LK_SHARED) != 0) { 511 error = EPERM; 512 goto out; 513 } 514 } 515 if (accmode & VWRITE) { 516 /* Just vn_writechk() changed to check rdonly */ 517 /* 518 * Disallow write attempts on read-only file systems; 519 * unless the file is a socket or a block or character 520 * device resident on the file system. 521 */ 522 if (NFSVNO_EXRDONLY(exp) || 523 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 524 switch (vp->v_type) { 525 case VREG: 526 case VDIR: 527 case VLNK: 528 error = EROFS; 529 default: 530 break; 531 } 532 } 533 /* 534 * If there's shared text associated with 535 * the inode, try to free it up once. If 536 * we fail, we can't allow writing. 537 */ 538 if (VOP_IS_TEXT(vp) && error == 0) 539 error = ETXTBSY; 540 } 541 if (error != 0) { 542 if (vpislocked == 0) 543 NFSVOPUNLOCK(vp); 544 goto out; 545 } 546 547 /* 548 * Should the override still be applied when ACLs are enabled? 549 */ 550 error = VOP_ACCESSX(vp, accmode, cred, p); 551 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { 552 /* 553 * Try again with VEXPLICIT_DENY, to see if the test for 554 * deletion is supported. 555 */ 556 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); 557 if (error == 0) { 558 if (vp->v_type == VDIR) { 559 accmode &= ~(VDELETE | VDELETE_CHILD); 560 accmode |= VWRITE; 561 error = VOP_ACCESSX(vp, accmode, cred, p); 562 } else if (supportedtypep != NULL) { 563 *supportedtypep &= ~NFSACCESS_DELETE; 564 } 565 } 566 } 567 568 /* 569 * Allow certain operations for the owner (reads and writes 570 * on files that are already open). 571 */ 572 if (override != NFSACCCHK_NOOVERRIDE && 573 (error == EPERM || error == EACCES)) { 574 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) 575 error = 0; 576 else if (override & NFSACCCHK_ALLOWOWNER) { 577 getret = VOP_GETATTR(vp, &vattr, cred); 578 if (getret == 0 && cred->cr_uid == vattr.va_uid) 579 error = 0; 580 } 581 } 582 if (vpislocked == 0) 583 NFSVOPUNLOCK(vp); 584 585 out: 586 NFSEXITCODE(error); 587 return (error); 588 } 589 590 /* 591 * Set attribute(s) vnop. 592 */ 593 int 594 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 595 struct thread *p, struct nfsexstuff *exp) 596 { 597 u_quad_t savsize = 0; 598 int error, savedit; 599 time_t savbtime; 600 601 /* 602 * If this is an exported file system and a pNFS service is running, 603 * don't VOP_SETATTR() of size for the MDS file system. 604 */ 605 savedit = 0; 606 error = 0; 607 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 && 608 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL && 609 nvap->na_vattr.va_size > 0) { 610 savsize = nvap->na_vattr.va_size; 611 nvap->na_vattr.va_size = VNOVAL; 612 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 613 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 614 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 615 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 616 nvap->na_vattr.va_mtime.tv_sec != VNOVAL) 617 savedit = 1; 618 else 619 savedit = 2; 620 } 621 if (savedit != 2) 622 error = VOP_SETATTR(vp, &nvap->na_vattr, cred); 623 if (savedit != 0) 624 nvap->na_vattr.va_size = savsize; 625 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL || 626 nvap->na_vattr.va_gid != (gid_t)VNOVAL || 627 nvap->na_vattr.va_size != VNOVAL || 628 nvap->na_vattr.va_mode != (mode_t)VNOVAL || 629 nvap->na_vattr.va_atime.tv_sec != VNOVAL || 630 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) { 631 /* Never modify birthtime on a DS file. */ 632 savbtime = nvap->na_vattr.va_birthtime.tv_sec; 633 nvap->na_vattr.va_birthtime.tv_sec = VNOVAL; 634 /* For a pNFS server, set the attributes on the DS file. */ 635 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR, 636 NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL); 637 nvap->na_vattr.va_birthtime.tv_sec = savbtime; 638 if (error == ENOENT) 639 error = 0; 640 } 641 NFSEXITCODE(error); 642 return (error); 643 } 644 645 /* 646 * Set up nameidata for a lookup() call and do it. 647 */ 648 int 649 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, 650 struct vnode *dp, int islocked, struct nfsexstuff *exp, 651 struct vnode **retdirp) 652 { 653 struct componentname *cnp = &ndp->ni_cnd; 654 int i; 655 struct iovec aiov; 656 struct uio auio; 657 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; 658 int error = 0; 659 char *cp; 660 661 *retdirp = NULL; 662 cnp->cn_nameptr = cnp->cn_pnbuf; 663 ndp->ni_lcf = 0; 664 /* 665 * Extract and set starting directory. 666 */ 667 if (dp->v_type != VDIR) { 668 if (islocked) 669 vput(dp); 670 else 671 vrele(dp); 672 nfsvno_relpathbuf(ndp); 673 error = ENOTDIR; 674 goto out1; 675 } 676 if (islocked) 677 NFSVOPUNLOCK(dp); 678 VREF(dp); 679 *retdirp = dp; 680 if (NFSVNO_EXRDONLY(exp)) 681 cnp->cn_flags |= RDONLY; 682 ndp->ni_segflg = UIO_SYSSPACE; 683 684 if (nd->nd_flag & ND_PUBLOOKUP) { 685 ndp->ni_loopcnt = 0; 686 if (cnp->cn_pnbuf[0] == '/') { 687 vrele(dp); 688 /* 689 * Check for degenerate pathnames here, since lookup() 690 * panics on them. 691 */ 692 for (i = 1; i < ndp->ni_pathlen; i++) 693 if (cnp->cn_pnbuf[i] != '/') 694 break; 695 if (i == ndp->ni_pathlen) { 696 error = NFSERR_ACCES; 697 goto out; 698 } 699 dp = rootvnode; 700 VREF(dp); 701 } 702 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || 703 (nd->nd_flag & ND_NFSV4) == 0) { 704 /* 705 * Only cross mount points for NFSv4 when doing a 706 * mount while traversing the file system above 707 * the mount point, unless nfsrv_enable_crossmntpt is set. 708 */ 709 cnp->cn_flags |= NOCROSSMOUNT; 710 } 711 712 /* 713 * Initialize for scan, set ni_startdir and bump ref on dp again 714 * because lookup() will dereference ni_startdir. 715 */ 716 717 ndp->ni_startdir = dp; 718 ndp->ni_rootdir = rootvnode; 719 ndp->ni_topdir = NULL; 720 721 if (!lockleaf) 722 cnp->cn_flags |= LOCKLEAF; 723 for (;;) { 724 cnp->cn_nameptr = cnp->cn_pnbuf; 725 /* 726 * Call lookup() to do the real work. If an error occurs, 727 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and 728 * we do not have to dereference anything before returning. 729 * In either case ni_startdir will be dereferenced and NULLed 730 * out. 731 */ 732 error = vfs_lookup(ndp); 733 if (error) 734 break; 735 736 /* 737 * Check for encountering a symbolic link. Trivial 738 * termination occurs if no symlink encountered. 739 */ 740 if ((cnp->cn_flags & ISSYMLINK) == 0) { 741 if (ndp->ni_vp && !lockleaf) 742 NFSVOPUNLOCK(ndp->ni_vp); 743 break; 744 } 745 746 /* 747 * Validate symlink 748 */ 749 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) 750 NFSVOPUNLOCK(ndp->ni_dvp); 751 if (!(nd->nd_flag & ND_PUBLOOKUP)) { 752 error = EINVAL; 753 goto badlink2; 754 } 755 756 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 757 error = ELOOP; 758 goto badlink2; 759 } 760 if (ndp->ni_pathlen > 1) 761 cp = uma_zalloc(namei_zone, M_WAITOK); 762 else 763 cp = cnp->cn_pnbuf; 764 aiov.iov_base = cp; 765 aiov.iov_len = MAXPATHLEN; 766 auio.uio_iov = &aiov; 767 auio.uio_iovcnt = 1; 768 auio.uio_offset = 0; 769 auio.uio_rw = UIO_READ; 770 auio.uio_segflg = UIO_SYSSPACE; 771 auio.uio_td = NULL; 772 auio.uio_resid = MAXPATHLEN; 773 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 774 if (error) { 775 badlink1: 776 if (ndp->ni_pathlen > 1) 777 uma_zfree(namei_zone, cp); 778 badlink2: 779 vrele(ndp->ni_dvp); 780 vput(ndp->ni_vp); 781 break; 782 } 783 linklen = MAXPATHLEN - auio.uio_resid; 784 if (linklen == 0) { 785 error = ENOENT; 786 goto badlink1; 787 } 788 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 789 error = ENAMETOOLONG; 790 goto badlink1; 791 } 792 793 /* 794 * Adjust or replace path 795 */ 796 if (ndp->ni_pathlen > 1) { 797 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 798 uma_zfree(namei_zone, cnp->cn_pnbuf); 799 cnp->cn_pnbuf = cp; 800 } else 801 cnp->cn_pnbuf[linklen] = '\0'; 802 ndp->ni_pathlen += linklen; 803 804 /* 805 * Cleanup refs for next loop and check if root directory 806 * should replace current directory. Normally ni_dvp 807 * becomes the new base directory and is cleaned up when 808 * we loop. Explicitly null pointers after invalidation 809 * to clarify operation. 810 */ 811 vput(ndp->ni_vp); 812 ndp->ni_vp = NULL; 813 814 if (cnp->cn_pnbuf[0] == '/') { 815 vrele(ndp->ni_dvp); 816 ndp->ni_dvp = ndp->ni_rootdir; 817 VREF(ndp->ni_dvp); 818 } 819 ndp->ni_startdir = ndp->ni_dvp; 820 ndp->ni_dvp = NULL; 821 } 822 if (!lockleaf) 823 cnp->cn_flags &= ~LOCKLEAF; 824 825 out: 826 if (error) { 827 nfsvno_relpathbuf(ndp); 828 ndp->ni_vp = NULL; 829 ndp->ni_dvp = NULL; 830 ndp->ni_startdir = NULL; 831 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { 832 ndp->ni_dvp = NULL; 833 } 834 835 out1: 836 NFSEXITCODE2(error, nd); 837 return (error); 838 } 839 840 /* 841 * Set up a pathname buffer and return a pointer to it and, optionally 842 * set a hash pointer. 843 */ 844 void 845 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) 846 { 847 struct componentname *cnp = &ndp->ni_cnd; 848 849 cnp->cn_flags |= (NOMACCHECK); 850 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 851 if (hashpp != NULL) 852 *hashpp = NULL; 853 *bufpp = cnp->cn_pnbuf; 854 } 855 856 /* 857 * Release the above path buffer, if not released by nfsvno_namei(). 858 */ 859 void 860 nfsvno_relpathbuf(struct nameidata *ndp) 861 { 862 863 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 864 ndp->ni_cnd.cn_pnbuf = NULL; 865 } 866 867 /* 868 * Readlink vnode op into an mbuf list. 869 */ 870 int 871 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz, 872 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 873 { 874 struct iovec *iv; 875 struct uio io, *uiop = &io; 876 struct mbuf *mp, *mp3; 877 int len, tlen, error = 0; 878 879 len = NFS_MAXPATHLEN; 880 if (maxextsiz > 0) 881 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 882 &mp3, &mp, &iv); 883 else 884 uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv); 885 uiop->uio_iov = iv; 886 uiop->uio_offset = 0; 887 uiop->uio_resid = len; 888 uiop->uio_rw = UIO_READ; 889 uiop->uio_segflg = UIO_SYSSPACE; 890 uiop->uio_td = NULL; 891 error = VOP_READLINK(vp, uiop, cred); 892 free(iv, M_TEMP); 893 if (error) { 894 m_freem(mp3); 895 *lenp = 0; 896 goto out; 897 } 898 if (uiop->uio_resid > 0) { 899 len -= uiop->uio_resid; 900 tlen = NFSM_RNDUP(len); 901 if (tlen == 0) { 902 m_freem(mp3); 903 mp3 = mp = NULL; 904 } else if (tlen != NFS_MAXPATHLEN || tlen != len) 905 mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, 906 tlen - len); 907 } 908 *lenp = len; 909 *mpp = mp3; 910 *mpendp = mp; 911 912 out: 913 NFSEXITCODE(error); 914 return (error); 915 } 916 917 /* 918 * Create an mbuf chain and an associated iovec that can be used to Read 919 * or Getextattr of data. 920 * Upon success, return pointers to the first and last mbufs in the chain 921 * plus the malloc'd iovec and its iovlen. 922 */ 923 static int 924 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp, 925 struct iovec **ivp) 926 { 927 struct mbuf *m, *m2 = NULL, *m3; 928 struct iovec *iv; 929 int i, left, siz; 930 931 left = len; 932 m3 = NULL; 933 /* 934 * Generate the mbuf list with the uio_iov ref. to it. 935 */ 936 i = 0; 937 while (left > 0) { 938 NFSMGET(m); 939 MCLGET(m, M_WAITOK); 940 m->m_len = 0; 941 siz = min(M_TRAILINGSPACE(m), left); 942 left -= siz; 943 i++; 944 if (m3) 945 m2->m_next = m; 946 else 947 m3 = m; 948 m2 = m; 949 } 950 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 951 m = m3; 952 left = len; 953 i = 0; 954 while (left > 0) { 955 if (m == NULL) 956 panic("nfsrv_createiovec iov"); 957 siz = min(M_TRAILINGSPACE(m), left); 958 if (siz > 0) { 959 iv->iov_base = mtod(m, caddr_t) + m->m_len; 960 iv->iov_len = siz; 961 m->m_len += siz; 962 left -= siz; 963 iv++; 964 i++; 965 } 966 m = m->m_next; 967 } 968 *mpp = m3; 969 *mpendp = m2; 970 return (i); 971 } 972 973 /* 974 * Create an mbuf chain and an associated iovec that can be used to Read 975 * or Getextattr of data. 976 * Upon success, return pointers to the first and last mbufs in the chain 977 * plus the malloc'd iovec and its iovlen. 978 * Same as above, but creates ext_pgs mbuf(s). 979 */ 980 static int 981 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp, 982 struct mbuf **mpendp, struct iovec **ivp) 983 { 984 struct mbuf *m, *m2 = NULL, *m3; 985 struct iovec *iv; 986 int i, left, pgno, siz; 987 988 left = len; 989 m3 = NULL; 990 /* 991 * Generate the mbuf list with the uio_iov ref. to it. 992 */ 993 i = 0; 994 while (left > 0) { 995 siz = min(left, maxextsiz); 996 m = mb_alloc_ext_plus_pages(siz, M_WAITOK); 997 left -= siz; 998 i += m->m_epg_npgs; 999 if (m3 != NULL) 1000 m2->m_next = m; 1001 else 1002 m3 = m; 1003 m2 = m; 1004 } 1005 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK); 1006 m = m3; 1007 left = len; 1008 i = 0; 1009 pgno = 0; 1010 while (left > 0) { 1011 if (m == NULL) 1012 panic("nfsvno_createiovec_extpgs iov"); 1013 siz = min(PAGE_SIZE, left); 1014 if (siz > 0) { 1015 iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]); 1016 iv->iov_len = siz; 1017 m->m_len += siz; 1018 if (pgno == m->m_epg_npgs - 1) 1019 m->m_epg_last_len = siz; 1020 left -= siz; 1021 iv++; 1022 i++; 1023 pgno++; 1024 } 1025 if (pgno == m->m_epg_npgs && left > 0) { 1026 m = m->m_next; 1027 if (m == NULL) 1028 panic("nfsvno_createiovec_extpgs iov"); 1029 pgno = 0; 1030 } 1031 } 1032 *mpp = m3; 1033 *mpendp = m2; 1034 return (i); 1035 } 1036 1037 /* 1038 * Read vnode op call into mbuf list. 1039 */ 1040 int 1041 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 1042 int maxextsiz, struct thread *p, struct mbuf **mpp, 1043 struct mbuf **mpendp) 1044 { 1045 struct mbuf *m; 1046 struct iovec *iv; 1047 int error = 0, len, tlen, ioflag = 0; 1048 struct mbuf *m3; 1049 struct uio io, *uiop = &io; 1050 struct nfsheur *nh; 1051 1052 /* 1053 * Attempt to read from a DS file. A return of ENOENT implies 1054 * there is no DS file to read. 1055 */ 1056 error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp, 1057 NULL, mpendp, NULL, NULL, NULL, 0, NULL); 1058 if (error != ENOENT) 1059 return (error); 1060 1061 len = NFSM_RNDUP(cnt); 1062 if (maxextsiz > 0) 1063 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz, 1064 &m3, &m, &iv); 1065 else 1066 uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv); 1067 uiop->uio_iov = iv; 1068 uiop->uio_offset = off; 1069 uiop->uio_resid = len; 1070 uiop->uio_rw = UIO_READ; 1071 uiop->uio_segflg = UIO_SYSSPACE; 1072 uiop->uio_td = NULL; 1073 nh = nfsrv_sequential_heuristic(uiop, vp); 1074 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 1075 /* XXX KDM make this more systematic? */ 1076 NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_READ] += uiop->uio_resid; 1077 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 1078 free(iv, M_TEMP); 1079 if (error) { 1080 m_freem(m3); 1081 *mpp = NULL; 1082 goto out; 1083 } 1084 nh->nh_nextoff = uiop->uio_offset; 1085 tlen = len - uiop->uio_resid; 1086 cnt = cnt < tlen ? cnt : tlen; 1087 tlen = NFSM_RNDUP(cnt); 1088 if (tlen == 0) { 1089 m_freem(m3); 1090 m3 = m = NULL; 1091 } else if (len != tlen || tlen != cnt) 1092 m = nfsrv_adj(m3, len - tlen, tlen - cnt); 1093 *mpp = m3; 1094 *mpendp = m; 1095 1096 out: 1097 NFSEXITCODE(error); 1098 return (error); 1099 } 1100 1101 /* 1102 * Create the iovec for the mbuf chain passed in as an argument. 1103 * The "cp" argument is where the data starts within the first mbuf in 1104 * the chain. It returns the iovec and the iovcnt. 1105 */ 1106 static int 1107 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp, 1108 int *iovcntp) 1109 { 1110 struct mbuf *mp; 1111 struct iovec *ivp; 1112 int cnt, i, len; 1113 1114 /* 1115 * Loop through the mbuf chain, counting how many mbufs are a 1116 * part of this write operation, so the iovec size is known. 1117 */ 1118 cnt = 0; 1119 len = retlen; 1120 mp = m; 1121 i = mtod(mp, caddr_t) + mp->m_len - cp; 1122 while (len > 0) { 1123 if (i > 0) { 1124 len -= i; 1125 cnt++; 1126 } 1127 mp = mp->m_next; 1128 if (!mp) { 1129 if (len > 0) 1130 return (EBADRPC); 1131 } else 1132 i = mp->m_len; 1133 } 1134 1135 /* Now, create the iovec. */ 1136 mp = m; 1137 *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, 1138 M_WAITOK); 1139 *iovcntp = cnt; 1140 i = mtod(mp, caddr_t) + mp->m_len - cp; 1141 len = retlen; 1142 while (len > 0) { 1143 if (mp == NULL) 1144 panic("nfsrv_createiovecw"); 1145 if (i > 0) { 1146 i = min(i, len); 1147 ivp->iov_base = cp; 1148 ivp->iov_len = i; 1149 ivp++; 1150 len -= i; 1151 } 1152 mp = mp->m_next; 1153 if (mp) { 1154 i = mp->m_len; 1155 cp = mtod(mp, caddr_t); 1156 } 1157 } 1158 return (0); 1159 } 1160 1161 /* 1162 * Write vnode op from an mbuf list. 1163 */ 1164 int 1165 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable, 1166 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) 1167 { 1168 struct iovec *iv; 1169 int cnt, ioflags, error; 1170 struct uio io, *uiop = &io; 1171 struct nfsheur *nh; 1172 1173 /* 1174 * Attempt to write to a DS file. A return of ENOENT implies 1175 * there is no DS file to write. 1176 */ 1177 error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS, 1178 &mp, cp, NULL, NULL, NULL, NULL, 0, NULL); 1179 if (error != ENOENT) { 1180 *stable = NFSWRITE_FILESYNC; 1181 return (error); 1182 } 1183 1184 if (*stable == NFSWRITE_UNSTABLE) 1185 ioflags = IO_NODELOCKED; 1186 else 1187 ioflags = (IO_SYNC | IO_NODELOCKED); 1188 error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt); 1189 if (error != 0) 1190 return (error); 1191 uiop->uio_iov = iv; 1192 uiop->uio_iovcnt = cnt; 1193 uiop->uio_resid = retlen; 1194 uiop->uio_rw = UIO_WRITE; 1195 uiop->uio_segflg = UIO_SYSSPACE; 1196 NFSUIOPROC(uiop, p); 1197 uiop->uio_offset = off; 1198 nh = nfsrv_sequential_heuristic(uiop, vp); 1199 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 1200 /* XXX KDM make this more systematic? */ 1201 NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid; 1202 error = VOP_WRITE(vp, uiop, ioflags, cred); 1203 if (error == 0) 1204 nh->nh_nextoff = uiop->uio_offset; 1205 free(iv, M_TEMP); 1206 1207 NFSEXITCODE(error); 1208 return (error); 1209 } 1210 1211 /* 1212 * Common code for creating a regular file (plus special files for V2). 1213 */ 1214 int 1215 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, 1216 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, 1217 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp) 1218 { 1219 u_quad_t tempsize; 1220 int error; 1221 struct thread *p = curthread; 1222 1223 error = nd->nd_repstat; 1224 if (!error && ndp->ni_vp == NULL) { 1225 if (nvap->na_type == VREG || nvap->na_type == VSOCK) { 1226 error = VOP_CREATE(ndp->ni_dvp, 1227 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1228 /* For a pNFS server, create the data file on a DS. */ 1229 if (error == 0 && nvap->na_type == VREG) { 1230 /* 1231 * Create a data file on a DS for a pNFS server. 1232 * This function just returns if not 1233 * running a pNFS DS or the creation fails. 1234 */ 1235 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 1236 nd->nd_cred, p); 1237 } 1238 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1239 NULL, false); 1240 nfsvno_relpathbuf(ndp); 1241 if (!error) { 1242 if (*exclusive_flagp) { 1243 *exclusive_flagp = 0; 1244 NFSVNO_ATTRINIT(nvap); 1245 nvap->na_atime.tv_sec = cverf[0]; 1246 nvap->na_atime.tv_nsec = cverf[1]; 1247 error = VOP_SETATTR(ndp->ni_vp, 1248 &nvap->na_vattr, nd->nd_cred); 1249 if (error != 0) { 1250 vput(ndp->ni_vp); 1251 ndp->ni_vp = NULL; 1252 error = NFSERR_NOTSUPP; 1253 } 1254 } 1255 } 1256 /* 1257 * NFS V2 Only. nfsrvd_mknod() does this for V3. 1258 * (This implies, just get out on an error.) 1259 */ 1260 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || 1261 nvap->na_type == VFIFO) { 1262 if (nvap->na_type == VCHR && rdev == 0xffffffff) 1263 nvap->na_type = VFIFO; 1264 if (nvap->na_type != VFIFO && 1265 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) { 1266 nfsvno_relpathbuf(ndp); 1267 vput(ndp->ni_dvp); 1268 goto out; 1269 } 1270 nvap->na_rdev = rdev; 1271 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1272 &ndp->ni_cnd, &nvap->na_vattr); 1273 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : 1274 NULL, false); 1275 nfsvno_relpathbuf(ndp); 1276 if (error) 1277 goto out; 1278 } else { 1279 nfsvno_relpathbuf(ndp); 1280 vput(ndp->ni_dvp); 1281 error = ENXIO; 1282 goto out; 1283 } 1284 *vpp = ndp->ni_vp; 1285 } else { 1286 /* 1287 * Handle cases where error is already set and/or 1288 * the file exists. 1289 * 1 - clean up the lookup 1290 * 2 - iff !error and na_size set, truncate it 1291 */ 1292 nfsvno_relpathbuf(ndp); 1293 *vpp = ndp->ni_vp; 1294 if (ndp->ni_dvp == *vpp) 1295 vrele(ndp->ni_dvp); 1296 else 1297 vput(ndp->ni_dvp); 1298 if (!error && nvap->na_size != VNOVAL) { 1299 error = nfsvno_accchk(*vpp, VWRITE, 1300 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1301 NFSACCCHK_VPISLOCKED, NULL); 1302 if (!error) { 1303 tempsize = nvap->na_size; 1304 NFSVNO_ATTRINIT(nvap); 1305 nvap->na_size = tempsize; 1306 error = nfsvno_setattr(*vpp, nvap, 1307 nd->nd_cred, p, exp); 1308 } 1309 } 1310 if (error) 1311 vput(*vpp); 1312 } 1313 1314 out: 1315 NFSEXITCODE(error); 1316 return (error); 1317 } 1318 1319 /* 1320 * Do a mknod vnode op. 1321 */ 1322 int 1323 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, 1324 struct thread *p) 1325 { 1326 int error = 0; 1327 __enum_uint8(vtype) vtyp; 1328 1329 vtyp = nvap->na_type; 1330 /* 1331 * Iff doesn't exist, create it. 1332 */ 1333 if (ndp->ni_vp) { 1334 nfsvno_relpathbuf(ndp); 1335 vput(ndp->ni_dvp); 1336 vrele(ndp->ni_vp); 1337 error = EEXIST; 1338 goto out; 1339 } 1340 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 1341 nfsvno_relpathbuf(ndp); 1342 vput(ndp->ni_dvp); 1343 error = NFSERR_BADTYPE; 1344 goto out; 1345 } 1346 if (vtyp == VSOCK) { 1347 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 1348 &ndp->ni_cnd, &nvap->na_vattr); 1349 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1350 false); 1351 nfsvno_relpathbuf(ndp); 1352 } else { 1353 if (nvap->na_type != VFIFO && 1354 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) { 1355 nfsvno_relpathbuf(ndp); 1356 vput(ndp->ni_dvp); 1357 goto out; 1358 } 1359 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 1360 &ndp->ni_cnd, &nvap->na_vattr); 1361 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, 1362 false); 1363 nfsvno_relpathbuf(ndp); 1364 /* 1365 * Since VOP_MKNOD returns the ni_vp, I can't 1366 * see any reason to do the lookup. 1367 */ 1368 } 1369 1370 out: 1371 NFSEXITCODE(error); 1372 return (error); 1373 } 1374 1375 /* 1376 * Mkdir vnode op. 1377 */ 1378 int 1379 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, 1380 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 1381 { 1382 int error = 0; 1383 1384 if (ndp->ni_vp != NULL) { 1385 if (ndp->ni_dvp == ndp->ni_vp) 1386 vrele(ndp->ni_dvp); 1387 else 1388 vput(ndp->ni_dvp); 1389 vrele(ndp->ni_vp); 1390 nfsvno_relpathbuf(ndp); 1391 error = EEXIST; 1392 goto out; 1393 } 1394 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1395 &nvap->na_vattr); 1396 VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false); 1397 nfsvno_relpathbuf(ndp); 1398 1399 out: 1400 NFSEXITCODE(error); 1401 return (error); 1402 } 1403 1404 /* 1405 * symlink vnode op. 1406 */ 1407 int 1408 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, 1409 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, 1410 struct nfsexstuff *exp) 1411 { 1412 int error = 0; 1413 1414 if (ndp->ni_vp) { 1415 nfsvno_relpathbuf(ndp); 1416 if (ndp->ni_dvp == ndp->ni_vp) 1417 vrele(ndp->ni_dvp); 1418 else 1419 vput(ndp->ni_dvp); 1420 vrele(ndp->ni_vp); 1421 error = EEXIST; 1422 goto out; 1423 } 1424 1425 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 1426 &nvap->na_vattr, pathcp); 1427 /* 1428 * Although FreeBSD still had the lookup code in 1429 * it for 7/current, there doesn't seem to be any 1430 * point, since VOP_SYMLINK() returns the ni_vp. 1431 * Just vput it for v2. 1432 */ 1433 VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0); 1434 nfsvno_relpathbuf(ndp); 1435 1436 out: 1437 NFSEXITCODE(error); 1438 return (error); 1439 } 1440 1441 /* 1442 * Parse symbolic link arguments. 1443 * This function has an ugly side effect. It will malloc() an area for 1444 * the symlink and set iov_base to point to it, only if it succeeds. 1445 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must 1446 * be FREE'd later. 1447 */ 1448 int 1449 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, 1450 struct thread *p, char **pathcpp, int *lenp) 1451 { 1452 u_int32_t *tl; 1453 char *pathcp = NULL; 1454 int error = 0, len; 1455 struct nfsv2_sattr *sp; 1456 1457 *pathcpp = NULL; 1458 *lenp = 0; 1459 if ((nd->nd_flag & ND_NFSV3) && 1460 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p))) 1461 goto nfsmout; 1462 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 1463 len = fxdr_unsigned(int, *tl); 1464 if (len > NFS_MAXPATHLEN || len <= 0) { 1465 error = EBADRPC; 1466 goto nfsmout; 1467 } 1468 pathcp = malloc(len + 1, M_TEMP, M_WAITOK); 1469 error = nfsrv_mtostr(nd, pathcp, len); 1470 if (error) 1471 goto nfsmout; 1472 if (nd->nd_flag & ND_NFSV2) { 1473 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1474 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 1475 } 1476 *pathcpp = pathcp; 1477 *lenp = len; 1478 NFSEXITCODE2(0, nd); 1479 return (0); 1480 nfsmout: 1481 if (pathcp) 1482 free(pathcp, M_TEMP); 1483 NFSEXITCODE2(error, nd); 1484 return (error); 1485 } 1486 1487 /* 1488 * Remove a non-directory object. 1489 */ 1490 int 1491 nfsvno_removesub(struct nameidata *ndp, bool is_v4, struct nfsrv_descript *nd, 1492 struct thread *p, struct nfsexstuff *exp) 1493 { 1494 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS], *newvp; 1495 struct mount *mp; 1496 int error = 0, mirrorcnt, ret; 1497 char fname[PNFS_FILENAME_LEN + 1]; 1498 fhandle_t fh; 1499 1500 vp = ndp->ni_vp; 1501 dsdvp[0] = NULL; 1502 if (vp->v_type == VDIR) { 1503 error = NFSERR_ISDIR; 1504 } else if (is_v4) { 1505 if (nfsrv_recalldeleg || (nd->nd_flag & ND_NFSV41) == 0) 1506 error = nfsrv_checkremove(vp, 1, NULL, 1507 (nfsquad_t)((u_quad_t)0), p); 1508 else 1509 error = nfsrv_checkremove(vp, 1, NULL, nd->nd_clientid, 1510 p); 1511 } 1512 if (error == 0) 1513 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh); 1514 if (!error) 1515 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); 1516 if (error == 0 && dsdvp[0] != NULL) 1517 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1518 if (is_v4 && (nd->nd_flag & ND_NFSV41) != 0 && error == 0) 1519 error = nfsvno_getfh(vp, &fh, p); 1520 if (ndp->ni_dvp == vp) 1521 vrele(ndp->ni_dvp); 1522 else 1523 vput(ndp->ni_dvp); 1524 vput(vp); 1525 1526 /* Use ret to determine if the file still exists. */ 1527 if (is_v4 && (nd->nd_flag & ND_NFSV41) != 0 && error == 0) { 1528 mp = vfs_busyfs(&fh.fh_fsid); 1529 if (mp != NULL) { 1530 /* Find out if the file still exists. */ 1531 ret = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &newvp); 1532 if (ret == 0) 1533 vput(newvp); 1534 else 1535 ret = ESTALE; 1536 vfs_unbusy(mp); 1537 } else { 1538 ret = ESTALE; 1539 } 1540 if (ret == ESTALE) { 1541 /* Get rid of any delegation. */ 1542 nfsrv_removedeleg(&fh, nd, p); 1543 } 1544 } 1545 1546 nfsvno_relpathbuf(ndp); 1547 NFSEXITCODE(error); 1548 return (error); 1549 } 1550 1551 /* 1552 * Remove a directory. 1553 */ 1554 int 1555 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1556 struct thread *p, struct nfsexstuff *exp) 1557 { 1558 struct vnode *vp; 1559 int error = 0; 1560 1561 vp = ndp->ni_vp; 1562 if (vp->v_type != VDIR) { 1563 error = ENOTDIR; 1564 goto out; 1565 } 1566 /* 1567 * No rmdir "." please. 1568 */ 1569 if (ndp->ni_dvp == vp) { 1570 error = EINVAL; 1571 goto out; 1572 } 1573 /* 1574 * The root of a mounted filesystem cannot be deleted. 1575 */ 1576 if (vp->v_vflag & VV_ROOT) 1577 error = EBUSY; 1578 out: 1579 if (!error) 1580 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); 1581 if (ndp->ni_dvp == vp) 1582 vrele(ndp->ni_dvp); 1583 else 1584 vput(ndp->ni_dvp); 1585 vput(vp); 1586 nfsvno_relpathbuf(ndp); 1587 NFSEXITCODE(error); 1588 return (error); 1589 } 1590 1591 /* 1592 * Rename vnode op. 1593 */ 1594 int 1595 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, 1596 struct nfsrv_descript *nd, struct thread *p) 1597 { 1598 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS], *newvp; 1599 struct mount *mp; 1600 int error = 0, mirrorcnt, ret; 1601 char fname[PNFS_FILENAME_LEN + 1]; 1602 fhandle_t fh, fh2; 1603 1604 dsdvp[0] = NULL; 1605 fvp = fromndp->ni_vp; 1606 if (nd->nd_repstat != 0) { 1607 vrele(fromndp->ni_dvp); 1608 vrele(fvp); 1609 error = nd->nd_repstat; 1610 goto out1; 1611 } 1612 tdvp = tondp->ni_dvp; 1613 tvp = tondp->ni_vp; 1614 if (tvp != NULL) { 1615 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1616 error = (nd->nd_flag & ND_NFSV2) ? EISDIR : EEXIST; 1617 goto out; 1618 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1619 error = (nd->nd_flag & ND_NFSV2) ? ENOTDIR : EEXIST; 1620 goto out; 1621 } 1622 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1623 error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1624 goto out; 1625 } 1626 1627 /* 1628 * A rename to '.' or '..' results in a prematurely 1629 * unlocked vnode on FreeBSD5, so I'm just going to fail that 1630 * here. 1631 */ 1632 if ((tondp->ni_cnd.cn_namelen == 1 && 1633 tondp->ni_cnd.cn_nameptr[0] == '.') || 1634 (tondp->ni_cnd.cn_namelen == 2 && 1635 tondp->ni_cnd.cn_nameptr[0] == '.' && 1636 tondp->ni_cnd.cn_nameptr[1] == '.')) { 1637 error = EINVAL; 1638 goto out; 1639 } 1640 } 1641 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1642 error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1643 goto out; 1644 } 1645 if (fvp->v_mount != tdvp->v_mount) { 1646 error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1647 goto out; 1648 } 1649 if (fvp == tdvp) { 1650 error = (nd->nd_flag & ND_NFSV2) ? ENOTEMPTY : EINVAL; 1651 goto out; 1652 } 1653 if (fvp == tvp) { 1654 /* 1655 * If source and destination are the same, there is 1656 * nothing to do. Set error to EJUSTRETURN to indicate 1657 * this. 1658 */ 1659 error = EJUSTRETURN; 1660 goto out; 1661 } 1662 if (nd->nd_flag & ND_NFSV4) { 1663 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { 1664 if (nfsrv_recalldeleg || (nd->nd_flag & ND_NFSV41) == 0) 1665 error = nfsrv_checkremove(fvp, 0, NULL, 1666 (nfsquad_t)((u_quad_t)0), p); 1667 else 1668 error = nfsrv_checkremove(fvp, 0, NULL, 1669 nd->nd_clientid, p); 1670 NFSVOPUNLOCK(fvp); 1671 } else 1672 error = EPERM; 1673 if (tvp && !error) { 1674 if (nfsrv_recalldeleg || (nd->nd_flag & ND_NFSV41) == 0) 1675 error = nfsrv_checkremove(tvp, 1, NULL, 1676 (nfsquad_t)((u_quad_t)0), p); 1677 else 1678 error = nfsrv_checkremove(tvp, 1, NULL, 1679 nd->nd_clientid, p); 1680 } 1681 } else { 1682 /* 1683 * For NFSv2 and NFSv3, try to get rid of the delegation, so 1684 * that the NFSv4 client won't be confused by the rename. 1685 * Since nfsd_recalldelegation() can only be called on an 1686 * unlocked vnode at this point and fvp is the file that will 1687 * still exist after the rename, just do fvp. 1688 */ 1689 nfsd_recalldelegation(fvp, p); 1690 } 1691 if (error == 0 && tvp != NULL) { 1692 if ((nd->nd_flag & ND_NFSV41) != 0) 1693 error = nfsvno_getfh(tvp, &fh2, p); 1694 if (error == 0) 1695 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, 1696 &fh); 1697 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup" 1698 " dsdvp=%p\n", dsdvp[0]); 1699 } 1700 out: 1701 mp = NULL; 1702 if (error == 0) { 1703 error = VOP_GETWRITEMOUNT(tondp->ni_dvp, &mp); 1704 if (error == 0) { 1705 if (mp == NULL) { 1706 error = ENOENT; 1707 } else { 1708 error = lockmgr(&mp->mnt_renamelock, 1709 LK_EXCLUSIVE | LK_NOWAIT, NULL); 1710 if (error != 0) 1711 error = ERELOOKUP; 1712 } 1713 } 1714 } 1715 if (error == 0) { 1716 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, 1717 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, 1718 &tondp->ni_cnd); 1719 lockmgr(&mp->mnt_renamelock, LK_RELEASE, 0); 1720 vfs_rel(mp); 1721 } else { 1722 if (tdvp == tvp) 1723 vrele(tdvp); 1724 else 1725 vput(tdvp); 1726 if (tvp) 1727 vput(tvp); 1728 vrele(fromndp->ni_dvp); 1729 vrele(fvp); 1730 if (error == EJUSTRETURN) { 1731 error = 0; 1732 } else if (error == ERELOOKUP && mp != NULL) { 1733 lockmgr(&mp->mnt_renamelock, LK_EXCLUSIVE, 0); 1734 lockmgr(&mp->mnt_renamelock, LK_RELEASE, 0); 1735 vfs_rel(mp); 1736 } 1737 } 1738 1739 /* 1740 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and 1741 * if the rename succeeded, the DS file for the tvp needs to be 1742 * removed. 1743 */ 1744 if (error == 0 && dsdvp[0] != NULL) { 1745 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p); 1746 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n"); 1747 } 1748 1749 /* Use ret to determine if the file still exists. */ 1750 if ((nd->nd_flag & ND_NFSV41) != 0 && error == 0) { 1751 mp = vfs_busyfs(&fh2.fh_fsid); 1752 if (mp != NULL) { 1753 /* Find out if the file still exists. */ 1754 ret = VFS_FHTOVP(mp, &fh2.fh_fid, LK_SHARED, &newvp); 1755 if (ret == 0) 1756 vput(newvp); 1757 else 1758 ret = ESTALE; 1759 vfs_unbusy(mp); 1760 } else { 1761 ret = ESTALE; 1762 } 1763 if (ret == ESTALE) { 1764 /* Get rid of any delegation. */ 1765 nfsrv_removedeleg(&fh2, nd, p); 1766 } 1767 } 1768 1769 nfsvno_relpathbuf(tondp); 1770 out1: 1771 nfsvno_relpathbuf(fromndp); 1772 NFSEXITCODE(error); 1773 return (error); 1774 } 1775 1776 /* 1777 * Link vnode op. 1778 */ 1779 int 1780 nfsvno_link(struct nameidata *ndp, struct vnode *vp, nfsquad_t clientid, 1781 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 1782 { 1783 struct vnode *xp; 1784 int error = 0; 1785 1786 xp = ndp->ni_vp; 1787 if (xp != NULL) { 1788 error = EEXIST; 1789 } else { 1790 xp = ndp->ni_dvp; 1791 if (vp->v_mount != xp->v_mount) 1792 error = EXDEV; 1793 } 1794 if (!error) { 1795 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1796 if (!VN_IS_DOOMED(vp)) { 1797 error = nfsrv_checkremove(vp, 0, NULL, clientid, p); 1798 if (error == 0) 1799 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); 1800 } else 1801 error = EPERM; 1802 if (ndp->ni_dvp == vp) { 1803 vrele(ndp->ni_dvp); 1804 NFSVOPUNLOCK(vp); 1805 } else { 1806 vref(vp); 1807 VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true); 1808 } 1809 } else { 1810 if (ndp->ni_dvp == ndp->ni_vp) 1811 vrele(ndp->ni_dvp); 1812 else 1813 vput(ndp->ni_dvp); 1814 if (ndp->ni_vp) 1815 vrele(ndp->ni_vp); 1816 } 1817 nfsvno_relpathbuf(ndp); 1818 NFSEXITCODE(error); 1819 return (error); 1820 } 1821 1822 /* 1823 * Do the fsync() appropriate for the commit. 1824 */ 1825 int 1826 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, 1827 struct thread *td) 1828 { 1829 int error = 0; 1830 1831 /* 1832 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of 1833 * file is done. At this time VOP_FSYNC does not accept offset and 1834 * byte count parameters so call VOP_FSYNC the whole file for now. 1835 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. 1836 * File systems that do not use the buffer cache (as indicated 1837 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC(). 1838 */ 1839 if (cnt == 0 || cnt > MAX_COMMIT_COUNT || 1840 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) { 1841 /* 1842 * Give up and do the whole thing 1843 */ 1844 vnode_pager_clean_sync(vp); 1845 error = VOP_FSYNC(vp, MNT_WAIT, td); 1846 } else { 1847 /* 1848 * Locate and synchronously write any buffers that fall 1849 * into the requested range. Note: we are assuming that 1850 * f_iosize is a power of 2. 1851 */ 1852 int iosize = vp->v_mount->mnt_stat.f_iosize; 1853 int iomask = iosize - 1; 1854 struct bufobj *bo; 1855 daddr_t lblkno; 1856 1857 /* 1858 * Align to iosize boundary, super-align to page boundary. 1859 */ 1860 if (off & iomask) { 1861 cnt += off & iomask; 1862 off &= ~(u_quad_t)iomask; 1863 } 1864 if (off & PAGE_MASK) { 1865 cnt += off & PAGE_MASK; 1866 off &= ~(u_quad_t)PAGE_MASK; 1867 } 1868 lblkno = off / iosize; 1869 1870 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) { 1871 VM_OBJECT_WLOCK(vp->v_object); 1872 vm_object_page_clean(vp->v_object, off, off + cnt, 1873 OBJPC_SYNC); 1874 VM_OBJECT_WUNLOCK(vp->v_object); 1875 } 1876 1877 bo = &vp->v_bufobj; 1878 BO_LOCK(bo); 1879 while (cnt > 0) { 1880 struct buf *bp; 1881 1882 /* 1883 * If we have a buffer and it is marked B_DELWRI we 1884 * have to lock and write it. Otherwise the prior 1885 * write is assumed to have already been committed. 1886 * 1887 * gbincore() can return invalid buffers now so we 1888 * have to check that bit as well (though B_DELWRI 1889 * should not be set if B_INVAL is set there could be 1890 * a race here since we haven't locked the buffer). 1891 */ 1892 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 1893 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 1894 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { 1895 BO_LOCK(bo); 1896 continue; /* retry */ 1897 } 1898 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 1899 B_DELWRI) { 1900 bremfree(bp); 1901 bp->b_flags &= ~B_ASYNC; 1902 bwrite(bp); 1903 ++nfs_commit_miss; 1904 } else 1905 BUF_UNLOCK(bp); 1906 BO_LOCK(bo); 1907 } 1908 ++nfs_commit_blks; 1909 if (cnt < iosize) 1910 break; 1911 cnt -= iosize; 1912 ++lblkno; 1913 } 1914 BO_UNLOCK(bo); 1915 } 1916 NFSEXITCODE(error); 1917 return (error); 1918 } 1919 1920 /* 1921 * Statfs vnode op. 1922 */ 1923 int 1924 nfsvno_statfs(struct vnode *vp, struct statfs *sf) 1925 { 1926 struct statfs *tsf; 1927 int error; 1928 1929 tsf = NULL; 1930 if (nfsrv_devidcnt > 0) { 1931 /* For a pNFS service, get the DS numbers. */ 1932 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO); 1933 error = nfsrv_pnfsstatfs(tsf, vp->v_mount); 1934 if (error != 0) { 1935 free(tsf, M_TEMP); 1936 tsf = NULL; 1937 } 1938 } 1939 error = VFS_STATFS(vp->v_mount, sf); 1940 if (error == 0) { 1941 if (tsf != NULL) { 1942 sf->f_blocks = tsf->f_blocks; 1943 sf->f_bavail = tsf->f_bavail; 1944 sf->f_bfree = tsf->f_bfree; 1945 sf->f_bsize = tsf->f_bsize; 1946 } 1947 /* 1948 * Since NFS handles these values as unsigned on the 1949 * wire, there is no way to represent negative values, 1950 * so set them to 0. Without this, they will appear 1951 * to be very large positive values for clients like 1952 * Solaris10. 1953 */ 1954 if (sf->f_bavail < 0) 1955 sf->f_bavail = 0; 1956 if (sf->f_ffree < 0) 1957 sf->f_ffree = 0; 1958 } 1959 free(tsf, M_TEMP); 1960 NFSEXITCODE(error); 1961 return (error); 1962 } 1963 1964 /* 1965 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but 1966 * must handle nfsrv_opencheck() calls after any other access checks. 1967 */ 1968 void 1969 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, 1970 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, 1971 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, 1972 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, bool done_namei, 1973 struct nfsexstuff *exp, struct vnode **vpp) 1974 { 1975 struct vnode *vp = NULL; 1976 u_quad_t tempsize; 1977 struct nfsexstuff nes; 1978 struct thread *p = curthread; 1979 uint32_t oldrepstat; 1980 1981 if (ndp->ni_vp == NULL) { 1982 /* 1983 * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be 1984 * set true, since cleanup after nfsvno_namei() is needed. 1985 */ 1986 oldrepstat = nd->nd_repstat; 1987 nd->nd_repstat = nfsrv_opencheck(clientid, 1988 stateidp, stp, NULL, nd, p, nd->nd_repstat); 1989 if (nd->nd_repstat != 0 && oldrepstat == 0) 1990 done_namei = true; 1991 } 1992 if (!nd->nd_repstat) { 1993 if (ndp->ni_vp == NULL) { 1994 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, 1995 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1996 /* For a pNFS server, create the data file on a DS. */ 1997 if (nd->nd_repstat == 0) { 1998 /* 1999 * Create a data file on a DS for a pNFS server. 2000 * This function just returns if not 2001 * running a pNFS DS or the creation fails. 2002 */ 2003 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr, 2004 cred, p); 2005 } 2006 VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ? 2007 &ndp->ni_vp : NULL, false); 2008 nfsvno_relpathbuf(ndp); 2009 if (!nd->nd_repstat) { 2010 if (*exclusive_flagp) { 2011 *exclusive_flagp = 0; 2012 NFSVNO_ATTRINIT(nvap); 2013 nvap->na_atime.tv_sec = cverf[0]; 2014 nvap->na_atime.tv_nsec = cverf[1]; 2015 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, 2016 &nvap->na_vattr, cred); 2017 if (nd->nd_repstat != 0) { 2018 vput(ndp->ni_vp); 2019 ndp->ni_vp = NULL; 2020 nd->nd_repstat = NFSERR_NOTSUPP; 2021 } else 2022 NFSSETBIT_ATTRBIT(attrbitp, 2023 NFSATTRBIT_TIMEACCESS); 2024 } else { 2025 nfsrv_fixattr(nd, ndp->ni_vp, nvap, 2026 aclp, p, attrbitp, exp); 2027 } 2028 } 2029 vp = ndp->ni_vp; 2030 } else { 2031 nfsvno_relpathbuf(ndp); 2032 vp = ndp->ni_vp; 2033 if (create == NFSV4OPEN_CREATE) { 2034 if (ndp->ni_dvp == vp) 2035 vrele(ndp->ni_dvp); 2036 else 2037 vput(ndp->ni_dvp); 2038 } 2039 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { 2040 if (ndp->ni_cnd.cn_flags & RDONLY) 2041 NFSVNO_SETEXRDONLY(&nes); 2042 else 2043 NFSVNO_EXINIT(&nes); 2044 nd->nd_repstat = nfsvno_accchk(vp, 2045 VWRITE, cred, &nes, p, 2046 NFSACCCHK_NOOVERRIDE, 2047 NFSACCCHK_VPISLOCKED, NULL); 2048 nd->nd_repstat = nfsrv_opencheck(clientid, 2049 stateidp, stp, vp, nd, p, nd->nd_repstat); 2050 if (!nd->nd_repstat) { 2051 tempsize = nvap->na_size; 2052 NFSVNO_ATTRINIT(nvap); 2053 nvap->na_size = tempsize; 2054 nd->nd_repstat = nfsvno_setattr(vp, 2055 nvap, cred, p, exp); 2056 } 2057 } else if (vp->v_type == VREG) { 2058 nd->nd_repstat = nfsrv_opencheck(clientid, 2059 stateidp, stp, vp, nd, p, nd->nd_repstat); 2060 } 2061 } 2062 } else if (done_namei) { 2063 KASSERT(create == NFSV4OPEN_CREATE, 2064 ("nfsvno_open: not create")); 2065 /* 2066 * done_namei is set when nfsvno_namei() has completed 2067 * successfully, but a subsequent error was set in 2068 * nd_repstat. As such, cleanup of the nfsvno_namei() 2069 * results is required. 2070 */ 2071 nfsvno_relpathbuf(ndp); 2072 if (ndp->ni_dvp == ndp->ni_vp) 2073 vrele(ndp->ni_dvp); 2074 else 2075 vput(ndp->ni_dvp); 2076 if (ndp->ni_vp) 2077 vput(ndp->ni_vp); 2078 } 2079 *vpp = vp; 2080 2081 NFSEXITCODE2(0, nd); 2082 } 2083 2084 /* 2085 * Updates the file rev and sets the mtime and ctime 2086 * to the current clock time, returning the va_filerev and va_Xtime 2087 * values. 2088 * Return ESTALE to indicate the vnode is VIRF_DOOMED. 2089 */ 2090 int 2091 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, 2092 struct nfsrv_descript *nd, struct thread *p) 2093 { 2094 struct vattr va; 2095 2096 VATTR_NULL(&va); 2097 vfs_timestamp(&va.va_mtime); 2098 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 2099 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY); 2100 if (VN_IS_DOOMED(vp)) 2101 return (ESTALE); 2102 } 2103 (void) VOP_SETATTR(vp, &va, nd->nd_cred); 2104 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL); 2105 return (0); 2106 } 2107 2108 /* 2109 * Glue routine to nfsv4_fillattr(). 2110 */ 2111 int 2112 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, 2113 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, 2114 struct ucred *cred, struct thread *p, int isdgram, int reterr, 2115 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno, 2116 bool xattrsupp, bool has_hiddensystem, bool has_namedattr, 2117 uint32_t clone_blksize) 2118 { 2119 struct statfs *sf; 2120 int error; 2121 2122 sf = NULL; 2123 if (nfsrv_devidcnt > 0 && 2124 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) || 2125 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) || 2126 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) { 2127 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO); 2128 error = nfsrv_pnfsstatfs(sf, mp); 2129 if (error != 0) { 2130 free(sf, M_TEMP); 2131 sf = NULL; 2132 } 2133 } 2134 2135 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, 2136 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, 2137 mounted_on_fileno, sf, xattrsupp, has_hiddensystem, has_namedattr, 2138 clone_blksize); 2139 free(sf, M_TEMP); 2140 NFSEXITCODE2(0, nd); 2141 return (error); 2142 } 2143 2144 /* 2145 * Convert a dirent d_type to a vnode type. 2146 */ 2147 static void nfs_dtypetovtype(struct nfsvattr *nvap, struct vnode *vp, 2148 uint8_t dtype) 2149 { 2150 2151 if ((vn_irflag_read(vp) & VIRF_NAMEDDIR) != 0) { 2152 nvap->na_type = VREG; 2153 nvap->na_bsdflags |= SFBSD_NAMEDATTR; 2154 } else if (dtype <= DT_WHT) { 2155 nvap->na_type = dtype_to_vnode[dtype]; 2156 } else { 2157 nvap->na_type = VNON; 2158 } 2159 } 2160 2161 /* Since the Readdir vnode ops vary, put the entire functions in here. */ 2162 /* 2163 * nfs readdir service 2164 * - mallocs what it thinks is enough to read 2165 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR 2166 * - calls VOP_READDIR() 2167 * - loops around building the reply 2168 * if the output generated exceeds count break out of loop 2169 * The NFSM_CLGET macro is used here so that the reply will be packed 2170 * tightly in mbuf clusters. 2171 * - it trims out records with d_fileno == 0 2172 * this doesn't matter for Unix clients, but they might confuse clients 2173 * for other os'. 2174 * - it trims out records with d_type == DT_WHT 2175 * these cannot be seen through NFS (unless we extend the protocol) 2176 * The alternate call nfsrvd_readdirplus() does lookups as well. 2177 * PS: The NFS protocol spec. does not clarify what the "count" byte 2178 * argument is a count of.. just name strings and file id's or the 2179 * entire reply rpc or ... 2180 * I tried just file name and id sizes and it confused the Sun client, 2181 * so I am using the full rpc size now. The "paranoia.." comment refers 2182 * to including the status longwords that are not a part of the dir. 2183 * "entry" structures, but are in the rpc. 2184 */ 2185 int 2186 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, 2187 struct vnode *vp, struct nfsexstuff *exp) 2188 { 2189 struct dirent *dp; 2190 u_int32_t *tl; 2191 int dirlen; 2192 char *cpos, *cend, *rbuf; 2193 struct nfsvattr at; 2194 int nlen, error = 0, getret = 1; 2195 int siz, cnt, fullsiz, eofflag, ncookies; 2196 u_int64_t off, toff, verf __unused; 2197 uint64_t *cookies = NULL, *cookiep; 2198 struct uio io; 2199 struct iovec iv; 2200 int is_ufs; 2201 struct thread *p = curthread; 2202 2203 if (nd->nd_repstat) { 2204 nfsrv_postopattr(nd, getret, &at); 2205 goto out; 2206 } 2207 if (nd->nd_flag & ND_NFSV2) { 2208 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2209 off = fxdr_unsigned(u_quad_t, *tl++); 2210 } else { 2211 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 2212 off = fxdr_hyper(tl); 2213 tl += 2; 2214 verf = fxdr_hyper(tl); 2215 tl += 2; 2216 } 2217 toff = off; 2218 cnt = fxdr_unsigned(int, *tl); 2219 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2220 cnt = NFS_SRVMAXDATA(nd); 2221 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2222 fullsiz = siz; 2223 if (nd->nd_flag & ND_NFSV3) { 2224 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, 2225 NULL); 2226 #if 0 2227 /* 2228 * va_filerev is not sufficient as a cookie verifier, 2229 * since it is not supposed to change when entries are 2230 * removed/added unless that offset cookies returned to 2231 * the client are no longer valid. 2232 */ 2233 if (!nd->nd_repstat && toff && verf != at.na_filerev) 2234 nd->nd_repstat = NFSERR_BAD_COOKIE; 2235 #endif 2236 } 2237 if (!nd->nd_repstat && vp->v_type != VDIR) 2238 nd->nd_repstat = NFSERR_NOTDIR; 2239 if (nd->nd_repstat == 0 && cnt == 0) { 2240 if (nd->nd_flag & ND_NFSV2) 2241 /* NFSv2 does not have NFSERR_TOOSMALL */ 2242 nd->nd_repstat = EPERM; 2243 else 2244 nd->nd_repstat = NFSERR_TOOSMALL; 2245 } 2246 if (!nd->nd_repstat) 2247 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2248 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2249 NFSACCCHK_VPISLOCKED, NULL); 2250 if (nd->nd_repstat) { 2251 vput(vp); 2252 if (nd->nd_flag & ND_NFSV3) 2253 nfsrv_postopattr(nd, getret, &at); 2254 goto out; 2255 } 2256 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2257 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2258 again: 2259 eofflag = 0; 2260 if (cookies) { 2261 free(cookies, M_TEMP); 2262 cookies = NULL; 2263 } 2264 2265 iv.iov_base = rbuf; 2266 iv.iov_len = siz; 2267 io.uio_iov = &iv; 2268 io.uio_iovcnt = 1; 2269 io.uio_offset = (off_t)off; 2270 io.uio_resid = siz; 2271 io.uio_segflg = UIO_SYSSPACE; 2272 io.uio_rw = UIO_READ; 2273 io.uio_td = NULL; 2274 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2275 &cookies); 2276 off = (u_int64_t)io.uio_offset; 2277 if (io.uio_resid) 2278 siz -= io.uio_resid; 2279 2280 if (!cookies && !nd->nd_repstat) 2281 nd->nd_repstat = NFSERR_PERM; 2282 if (nd->nd_flag & ND_NFSV3) { 2283 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2284 if (!nd->nd_repstat) 2285 nd->nd_repstat = getret; 2286 } 2287 2288 /* 2289 * Handles the failed cases. nd->nd_repstat == 0 past here. 2290 */ 2291 if (nd->nd_repstat) { 2292 vput(vp); 2293 free(rbuf, M_TEMP); 2294 if (cookies) 2295 free(cookies, M_TEMP); 2296 if (nd->nd_flag & ND_NFSV3) 2297 nfsrv_postopattr(nd, getret, &at); 2298 goto out; 2299 } 2300 /* 2301 * If nothing read, return eof 2302 * rpc reply 2303 */ 2304 if (siz == 0) { 2305 vput(vp); 2306 if (nd->nd_flag & ND_NFSV2) { 2307 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2308 } else { 2309 nfsrv_postopattr(nd, getret, &at); 2310 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2311 txdr_hyper(at.na_filerev, tl); 2312 tl += 2; 2313 } 2314 *tl++ = newnfs_false; 2315 *tl = newnfs_true; 2316 free(rbuf, M_TEMP); 2317 free(cookies, M_TEMP); 2318 goto out; 2319 } 2320 2321 /* 2322 * Check for degenerate cases of nothing useful read. 2323 * If so go try again 2324 */ 2325 cpos = rbuf; 2326 cend = rbuf + siz; 2327 dp = (struct dirent *)cpos; 2328 cookiep = cookies; 2329 2330 /* 2331 * For some reason FreeBSD's ufs_readdir() chooses to back the 2332 * directory offset up to a block boundary, so it is necessary to 2333 * skip over the records that precede the requested offset. This 2334 * requires the assumption that file offset cookies monotonically 2335 * increase. 2336 */ 2337 while (cpos < cend && ncookies > 0 && 2338 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2339 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) { 2340 cpos += dp->d_reclen; 2341 dp = (struct dirent *)cpos; 2342 cookiep++; 2343 ncookies--; 2344 } 2345 if (cpos >= cend || ncookies == 0) { 2346 siz = fullsiz; 2347 toff = off; 2348 goto again; 2349 } 2350 vput(vp); 2351 2352 /* 2353 * If cnt > MCLBYTES and the reply will not be saved, use 2354 * ext_pgs mbufs for TLS. 2355 * For NFSv4.0, we do not know for sure if the reply will 2356 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2357 */ 2358 if (cnt > MCLBYTES && siz > MCLBYTES && 2359 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2360 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2361 nd->nd_flag |= ND_EXTPG; 2362 2363 /* 2364 * dirlen is the size of the reply, including all XDR and must 2365 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate 2366 * if the XDR should be included in "count", but to be safe, we do. 2367 * (Include the two booleans at the end of the reply in dirlen now.) 2368 */ 2369 if (nd->nd_flag & ND_NFSV3) { 2370 nfsrv_postopattr(nd, getret, &at); 2371 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2372 txdr_hyper(at.na_filerev, tl); 2373 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2374 } else { 2375 dirlen = 2 * NFSX_UNSIGNED; 2376 } 2377 2378 /* Loop through the records and build reply */ 2379 while (cpos < cend && ncookies > 0) { 2380 nlen = dp->d_namlen; 2381 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2382 nlen <= NFS_MAXNAMLEN) { 2383 if (nd->nd_flag & ND_NFSV3) 2384 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2385 else 2386 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 2387 if (dirlen > cnt) { 2388 eofflag = 0; 2389 break; 2390 } 2391 2392 /* 2393 * Build the directory record xdr from 2394 * the dirent entry. 2395 */ 2396 if (nd->nd_flag & ND_NFSV3) { 2397 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2398 *tl++ = newnfs_true; 2399 txdr_hyper(dp->d_fileno, tl); 2400 } else { 2401 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2402 *tl++ = newnfs_true; 2403 *tl = txdr_unsigned(dp->d_fileno); 2404 } 2405 (void) nfsm_strtom(nd, dp->d_name, nlen); 2406 if (nd->nd_flag & ND_NFSV3) { 2407 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2408 txdr_hyper(*cookiep, tl); 2409 } else { 2410 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 2411 *tl = txdr_unsigned(*cookiep); 2412 } 2413 } 2414 cpos += dp->d_reclen; 2415 dp = (struct dirent *)cpos; 2416 cookiep++; 2417 ncookies--; 2418 } 2419 if (cpos < cend) 2420 eofflag = 0; 2421 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2422 *tl++ = newnfs_false; 2423 if (eofflag) 2424 *tl = newnfs_true; 2425 else 2426 *tl = newnfs_false; 2427 free(rbuf, M_TEMP); 2428 free(cookies, M_TEMP); 2429 2430 out: 2431 NFSEXITCODE2(0, nd); 2432 return (0); 2433 nfsmout: 2434 vput(vp); 2435 NFSEXITCODE2(error, nd); 2436 return (error); 2437 } 2438 2439 /* 2440 * Readdirplus for V3 and Readdir for V4. 2441 */ 2442 int 2443 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, 2444 struct vnode *vp, struct nfsexstuff *exp) 2445 { 2446 struct dirent *dp; 2447 uint32_t clone_blksize, *tl; 2448 int dirlen; 2449 char *cpos, *cend, *rbuf; 2450 struct vnode *nvp; 2451 fhandle_t nfh; 2452 struct nfsvattr nva, at, *nvap = &nva; 2453 struct mbuf *mb0, *mb1; 2454 struct nfsreferral *refp; 2455 int nlen, r, error = 0, getret = 1, ret, usevget = 1; 2456 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; 2457 caddr_t bpos0, bpos1; 2458 u_int64_t off, toff, verf __unused; 2459 uint64_t *cookies = NULL, *cookiep; 2460 nfsattrbit_t attrbits, rderrbits, savbits, refbits; 2461 struct uio io; 2462 struct iovec iv; 2463 struct componentname cn; 2464 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls; 2465 struct mount *mp, *new_mp; 2466 uint64_t mounted_on_fileno; 2467 struct thread *p = curthread; 2468 int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1; 2469 size_t atsiz; 2470 long pathval; 2471 bool has_hiddensystem, has_namedattr, xattrsupp; 2472 2473 if (nd->nd_repstat) { 2474 nfsrv_postopattr(nd, getret, &at); 2475 goto out; 2476 } 2477 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 2478 off = fxdr_hyper(tl); 2479 toff = off; 2480 tl += 2; 2481 verf = fxdr_hyper(tl); 2482 tl += 2; 2483 siz = fxdr_unsigned(int, *tl++); 2484 cnt = fxdr_unsigned(int, *tl); 2485 2486 /* 2487 * Use the server's maximum data transfer size as the upper bound 2488 * on reply datalen. 2489 */ 2490 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 2491 cnt = NFS_SRVMAXDATA(nd); 2492 2493 /* 2494 * siz is a "hint" of how much directory information (name, fileid, 2495 * cookie) should be in the reply. At least one client "hints" 0, 2496 * so I set it to cnt for that case. I also round it up to the 2497 * next multiple of DIRBLKSIZ. 2498 * Since the size of a Readdirplus directory entry reply will always 2499 * be greater than a directory entry returned by VOP_READDIR(), it 2500 * does not make sense to read more than NFS_SRVMAXDATA() via 2501 * VOP_READDIR(). 2502 */ 2503 if (siz <= 0) 2504 siz = cnt; 2505 else if (siz > NFS_SRVMAXDATA(nd)) 2506 siz = NFS_SRVMAXDATA(nd); 2507 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 2508 2509 if (nd->nd_flag & ND_NFSV4) { 2510 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 2511 if (error) 2512 goto nfsmout; 2513 NFSSET_ATTRBIT(&savbits, &attrbits); 2514 NFSSET_ATTRBIT(&refbits, &attrbits); 2515 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd); 2516 NFSZERO_ATTRBIT(&rderrbits); 2517 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); 2518 /* 2519 * If these 4 bits are the only attributes requested by the 2520 * client, they can be satisfied without acquiring the vnode 2521 * for the file object unless it is a directory. 2522 * This will be indicated by savbits being all 0s. 2523 */ 2524 NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_TYPE); 2525 NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_FILEID); 2526 NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_MOUNTEDONFILEID); 2527 NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_RDATTRERROR); 2528 } else { 2529 NFSZERO_ATTRBIT(&attrbits); 2530 } 2531 fullsiz = siz; 2532 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2533 #if 0 2534 if (!nd->nd_repstat) { 2535 if (off && verf != at.na_filerev) { 2536 /* 2537 * va_filerev is not sufficient as a cookie verifier, 2538 * since it is not supposed to change when entries are 2539 * removed/added unless that offset cookies returned to 2540 * the client are no longer valid. 2541 */ 2542 if (nd->nd_flag & ND_NFSV4) { 2543 nd->nd_repstat = NFSERR_NOTSAME; 2544 } else { 2545 nd->nd_repstat = NFSERR_BAD_COOKIE; 2546 } 2547 } 2548 } 2549 #endif 2550 if (!nd->nd_repstat && vp->v_type != VDIR) 2551 nd->nd_repstat = NFSERR_NOTDIR; 2552 if (!nd->nd_repstat && cnt == 0) 2553 nd->nd_repstat = NFSERR_TOOSMALL; 2554 if (!nd->nd_repstat) 2555 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 2556 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 2557 NFSACCCHK_VPISLOCKED, NULL); 2558 if (nd->nd_repstat) { 2559 vput(vp); 2560 if (nd->nd_flag & ND_NFSV3) 2561 nfsrv_postopattr(nd, getret, &at); 2562 goto out; 2563 } 2564 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0; 2565 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0; 2566 2567 rbuf = malloc(siz, M_TEMP, M_WAITOK); 2568 again: 2569 eofflag = 0; 2570 if (cookies) { 2571 free(cookies, M_TEMP); 2572 cookies = NULL; 2573 } 2574 2575 iv.iov_base = rbuf; 2576 iv.iov_len = siz; 2577 io.uio_iov = &iv; 2578 io.uio_iovcnt = 1; 2579 io.uio_offset = (off_t)off; 2580 io.uio_resid = siz; 2581 io.uio_segflg = UIO_SYSSPACE; 2582 io.uio_rw = UIO_READ; 2583 io.uio_td = NULL; 2584 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 2585 &cookies); 2586 off = (u_int64_t)io.uio_offset; 2587 if (io.uio_resid) 2588 siz -= io.uio_resid; 2589 2590 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL); 2591 2592 if (!cookies && !nd->nd_repstat) 2593 nd->nd_repstat = NFSERR_PERM; 2594 if (!nd->nd_repstat) 2595 nd->nd_repstat = getret; 2596 if (nd->nd_repstat) { 2597 vput(vp); 2598 if (cookies) 2599 free(cookies, M_TEMP); 2600 free(rbuf, M_TEMP); 2601 if (nd->nd_flag & ND_NFSV3) 2602 nfsrv_postopattr(nd, getret, &at); 2603 goto out; 2604 } 2605 /* 2606 * If nothing read, return eof 2607 * rpc reply 2608 */ 2609 if (siz == 0) { 2610 ateof: 2611 vput(vp); 2612 if (nd->nd_flag & ND_NFSV3) 2613 nfsrv_postopattr(nd, getret, &at); 2614 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 2615 txdr_hyper(at.na_filerev, tl); 2616 tl += 2; 2617 *tl++ = newnfs_false; 2618 *tl = newnfs_true; 2619 free(cookies, M_TEMP); 2620 free(rbuf, M_TEMP); 2621 goto out; 2622 } 2623 2624 /* 2625 * Check for degenerate cases of nothing useful read. 2626 * If so go try again 2627 */ 2628 cpos = rbuf; 2629 cend = rbuf + siz; 2630 dp = (struct dirent *)cpos; 2631 cookiep = cookies; 2632 2633 /* 2634 * For some reason FreeBSD's ufs_readdir() chooses to back the 2635 * directory offset up to a block boundary, so it is necessary to 2636 * skip over the records that precede the requested offset. This 2637 * requires the assumption that file offset cookies monotonically 2638 * increase. 2639 */ 2640 while (cpos < cend && ncookies > 0 && 2641 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 2642 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) || 2643 ((nd->nd_flag & ND_NFSV4) && 2644 ((dp->d_namlen == 1 && dp->d_name[0] == '.') || 2645 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { 2646 cpos += dp->d_reclen; 2647 dp = (struct dirent *)cpos; 2648 cookiep++; 2649 ncookies--; 2650 } 2651 if (cpos >= cend || ncookies == 0) { 2652 if (eofflag != 0) 2653 goto ateof; 2654 siz = fullsiz; 2655 toff = off; 2656 goto again; 2657 } 2658 2659 /* 2660 * Busy the file system so that the mount point won't go away 2661 * and, as such, VFS_VGET() can be used safely. 2662 */ 2663 mp = vp->v_mount; 2664 vfs_ref(mp); 2665 NFSVOPUNLOCK(vp); 2666 nd->nd_repstat = vfs_busy(mp, 0); 2667 vfs_rel(mp); 2668 if (nd->nd_repstat != 0) { 2669 vrele(vp); 2670 free(cookies, M_TEMP); 2671 free(rbuf, M_TEMP); 2672 if (nd->nd_flag & ND_NFSV3) 2673 nfsrv_postopattr(nd, getret, &at); 2674 goto out; 2675 } 2676 2677 /* 2678 * Check to see if entries in this directory can be safely acquired 2679 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required. 2680 * ZFS snapshot directories need VOP_LOOKUP(), so that any 2681 * automount of the snapshot directory that is required will 2682 * be done. 2683 * This needs to be done here for NFSv4, since NFSv4 never does 2684 * a VFS_VGET() for "." or "..". 2685 */ 2686 if (is_zfs == 1) { 2687 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp); 2688 if (r == EOPNOTSUPP) { 2689 usevget = 0; 2690 cn.cn_nameiop = LOOKUP; 2691 cn.cn_lkflags = LK_SHARED | LK_RETRY; 2692 cn.cn_cred = nd->nd_cred; 2693 } else if (r == 0) 2694 vput(nvp); 2695 } 2696 2697 /* 2698 * If the reply is likely to exceed MCLBYTES and the reply will 2699 * not be saved, use ext_pgs mbufs for TLS. 2700 * It is difficult to predict how large each entry will be and 2701 * how many entries have been read, so just assume the directory 2702 * entries grow by a factor of 4 when attributes are included. 2703 * For NFSv4.0, we do not know for sure if the reply will 2704 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 2705 */ 2706 if (cnt > MCLBYTES && siz > MCLBYTES / 4 && 2707 (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS && 2708 (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4) 2709 nd->nd_flag |= ND_EXTPG; 2710 2711 /* 2712 * Save this position, in case there is an error before one entry 2713 * is created. 2714 */ 2715 mb0 = nd->nd_mb; 2716 bpos0 = nd->nd_bpos; 2717 bextpg0 = nd->nd_bextpg; 2718 bextpgsiz0 = nd->nd_bextpgsiz; 2719 2720 /* 2721 * Fill in the first part of the reply. 2722 * dirlen is the reply length in bytes and cannot exceed cnt. 2723 * (Include the two booleans at the end of the reply in dirlen now, 2724 * so we recognize when we have exceeded cnt.) 2725 */ 2726 if (nd->nd_flag & ND_NFSV3) { 2727 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 2728 nfsrv_postopattr(nd, getret, &at); 2729 } else { 2730 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; 2731 } 2732 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); 2733 txdr_hyper(at.na_filerev, tl); 2734 2735 /* 2736 * Save this position, in case there is an empty reply needed. 2737 */ 2738 mb1 = nd->nd_mb; 2739 bpos1 = nd->nd_bpos; 2740 bextpg1 = nd->nd_bextpg; 2741 bextpgsiz1 = nd->nd_bextpgsiz; 2742 2743 /* Loop through the records and build reply */ 2744 entrycnt = 0; 2745 while (cpos < cend && ncookies > 0 && dirlen < cnt) { 2746 nlen = dp->d_namlen; 2747 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2748 nlen <= NFS_MAXNAMLEN && 2749 ((nd->nd_flag & ND_NFSV3) || nlen > 2 || 2750 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) 2751 || (nlen == 1 && dp->d_name[0] != '.'))) { 2752 /* 2753 * Save the current position in the reply, in case 2754 * this entry exceeds cnt. 2755 */ 2756 mb1 = nd->nd_mb; 2757 bpos1 = nd->nd_bpos; 2758 bextpg1 = nd->nd_bextpg; 2759 bextpgsiz1 = nd->nd_bextpgsiz; 2760 2761 /* 2762 * For readdir_and_lookup get the vnode using 2763 * the file number. 2764 */ 2765 nvp = NULL; 2766 refp = NULL; 2767 r = 0; 2768 at_root = 0; 2769 needs_unbusy = 0; 2770 new_mp = mp; 2771 mounted_on_fileno = (uint64_t)dp->d_fileno; 2772 if ((nd->nd_flag & ND_NFSV3) || 2773 NFSNONZERO_ATTRBIT(&savbits) || 2774 dp->d_type == DT_UNKNOWN || 2775 (dp->d_type == DT_DIR && 2776 nfsrv_enable_crossmntpt != 0)) { 2777 if (nd->nd_flag & ND_NFSV4) 2778 refp = nfsv4root_getreferral(NULL, 2779 vp, dp->d_fileno); 2780 if (refp == NULL) { 2781 if (usevget) 2782 r = VFS_VGET(mp, dp->d_fileno, 2783 LK_SHARED, &nvp); 2784 else 2785 r = EOPNOTSUPP; 2786 if (r == 0 && (vn_irflag_read(vp) & 2787 VIRF_NAMEDDIR) != 0) 2788 vn_irflag_set_cond(nvp, 2789 VIRF_NAMEDATTR); 2790 if (r == EOPNOTSUPP) { 2791 if (usevget) { 2792 usevget = 0; 2793 cn.cn_nameiop = LOOKUP; 2794 cn.cn_lkflags = 2795 LK_SHARED | 2796 LK_RETRY; 2797 cn.cn_cred = 2798 nd->nd_cred; 2799 } 2800 cn.cn_nameptr = dp->d_name; 2801 cn.cn_namelen = nlen; 2802 cn.cn_flags = ISLASTCN | 2803 NOFOLLOW | LOCKLEAF; 2804 if ((vn_irflag_read(vp) & 2805 VIRF_NAMEDDIR) != 0) 2806 cn.cn_flags |= 2807 OPENNAMED; 2808 if (nlen == 2 && 2809 dp->d_name[0] == '.' && 2810 dp->d_name[1] == '.') 2811 cn.cn_flags |= 2812 ISDOTDOT; 2813 if (NFSVOPLOCK(vp, LK_SHARED) 2814 != 0) { 2815 nd->nd_repstat = EPERM; 2816 break; 2817 } 2818 if ((vp->v_vflag & VV_ROOT) != 0 2819 && (cn.cn_flags & ISDOTDOT) 2820 != 0) { 2821 vref(vp); 2822 nvp = vp; 2823 r = 0; 2824 } else { 2825 r = VOP_LOOKUP(vp, &nvp, 2826 &cn); 2827 if (vp != nvp) 2828 NFSVOPUNLOCK(vp); 2829 } 2830 } 2831 2832 /* 2833 * For NFSv4, check to see if nvp is 2834 * a mount point and get the mount 2835 * point vnode, as required. 2836 */ 2837 if (r == 0 && 2838 nfsrv_enable_crossmntpt != 0 && 2839 (nd->nd_flag & ND_NFSV4) != 0 && 2840 nvp->v_type == VDIR && 2841 nvp->v_mountedhere != NULL) { 2842 new_mp = nvp->v_mountedhere; 2843 r = vfs_busy(new_mp, 0); 2844 vput(nvp); 2845 nvp = NULL; 2846 if (r == 0) { 2847 r = VFS_ROOT(new_mp, 2848 LK_SHARED, &nvp); 2849 needs_unbusy = 1; 2850 if (r == 0) 2851 at_root = 1; 2852 } 2853 } 2854 } 2855 2856 /* 2857 * If we failed to look up the entry, then it 2858 * has become invalid, most likely removed. 2859 */ 2860 if (r != 0) { 2861 if (needs_unbusy) 2862 vfs_unbusy(new_mp); 2863 goto invalid; 2864 } 2865 KASSERT(refp != NULL || nvp != NULL, 2866 ("%s: undetected lookup error", __func__)); 2867 2868 if (refp == NULL && 2869 ((nd->nd_flag & ND_NFSV3) || 2870 NFSNONZERO_ATTRBIT(&attrbits))) { 2871 r = nfsvno_getfh(nvp, &nfh, p); 2872 if (!r) 2873 r = nfsvno_getattr(nvp, nvap, nd, p, 2874 1, &attrbits); 2875 if (r == 0 && is_zfs == 1 && 2876 nfsrv_enable_crossmntpt != 0 && 2877 (nd->nd_flag & ND_NFSV4) != 0 && 2878 nvp->v_type == VDIR && 2879 vp->v_mount != nvp->v_mount) { 2880 /* 2881 * For a ZFS snapshot, there is a 2882 * pseudo mount that does not set 2883 * v_mountedhere, so it needs to 2884 * be detected via a different 2885 * mount structure. 2886 */ 2887 at_root = 1; 2888 if (new_mp == mp) 2889 new_mp = nvp->v_mount; 2890 } 2891 } 2892 2893 /* 2894 * If we failed to get attributes of the entry, 2895 * then just skip it for NFSv3 (the traditional 2896 * behavior in the old NFS server). 2897 * For NFSv4 the behavior is controlled by 2898 * RDATTRERROR: we either ignore the error or 2899 * fail the request. 2900 * The exception is EOPNOTSUPP, which can be 2901 * returned by nfsvno_getfh() for certain 2902 * file systems, such as devfs. This indicates 2903 * that the file system cannot be exported, 2904 * so just skip over the entry. 2905 * Note that RDATTRERROR is never set for NFSv3. 2906 */ 2907 if (r != 0) { 2908 if (!NFSISSET_ATTRBIT(&attrbits, 2909 NFSATTRBIT_RDATTRERROR) || 2910 r == EOPNOTSUPP) { 2911 vput(nvp); 2912 if (needs_unbusy != 0) 2913 vfs_unbusy(new_mp); 2914 if ((nd->nd_flag & ND_NFSV3) || 2915 r == EOPNOTSUPP) 2916 goto invalid; 2917 nd->nd_repstat = r; 2918 break; 2919 } 2920 } 2921 } else if (NFSNONZERO_ATTRBIT(&attrbits)) { 2922 /* Only need Type and/or Fileid. */ 2923 VATTR_NULL(&nvap->na_vattr); 2924 nvap->na_fileid = dp->d_fileno; 2925 nfs_dtypetovtype(nvap, vp, dp->d_type); 2926 } 2927 2928 /* 2929 * Build the directory record xdr 2930 */ 2931 if (nd->nd_flag & ND_NFSV3) { 2932 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2933 *tl++ = newnfs_true; 2934 txdr_hyper(dp->d_fileno, tl); 2935 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2936 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2937 txdr_hyper(*cookiep, tl); 2938 nfsrv_postopattr(nd, 0, nvap); 2939 dirlen += nfsm_fhtom(NULL, nd, (u_int8_t *)&nfh, 2940 0, 1); 2941 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); 2942 if (nvp != NULL) 2943 vput(nvp); 2944 } else { 2945 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2946 *tl++ = newnfs_true; 2947 txdr_hyper(*cookiep, tl); 2948 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2949 xattrsupp = false; 2950 has_hiddensystem = false; 2951 has_namedattr = false; 2952 clone_blksize = 0; 2953 if (nvp != NULL) { 2954 supports_nfsv4acls = 2955 nfs_supportsnfsv4acls(nvp); 2956 if (NFSISSET_ATTRBIT(&attrbits, 2957 NFSATTRBIT_XATTRSUPPORT)) { 2958 ret = VOP_GETEXTATTR(nvp, 2959 EXTATTR_NAMESPACE_USER, 2960 "xxx", NULL, &atsiz, 2961 nd->nd_cred, p); 2962 xattrsupp = ret != EOPNOTSUPP; 2963 } 2964 if (VOP_PATHCONF(nvp, 2965 _PC_HAS_HIDDENSYSTEM, &pathval) != 2966 0) 2967 pathval = 0; 2968 has_hiddensystem = pathval > 0; 2969 pathval = 0; 2970 if (NFSISSET_ATTRBIT(&attrbits, 2971 NFSATTRBIT_NAMEDATTR) && 2972 VOP_PATHCONF(nvp, _PC_HAS_NAMEDATTR, 2973 &pathval) != 0) 2974 pathval = 0; 2975 has_namedattr = pathval > 0; 2976 pathval = 0; 2977 if (VOP_PATHCONF(nvp, _PC_CLONE_BLKSIZE, 2978 &pathval) != 0) 2979 pathval = 0; 2980 clone_blksize = pathval; 2981 NFSVOPUNLOCK(nvp); 2982 } else 2983 supports_nfsv4acls = 0; 2984 if (refp != NULL) { 2985 dirlen += nfsrv_putreferralattr(nd, 2986 &refbits, refp, 0, 2987 &nd->nd_repstat); 2988 if (nd->nd_repstat) { 2989 if (nvp != NULL) 2990 vrele(nvp); 2991 if (needs_unbusy != 0) 2992 vfs_unbusy(new_mp); 2993 break; 2994 } 2995 } else if (r) { 2996 dirlen += nfsvno_fillattr(nd, new_mp, 2997 nvp, nvap, &nfh, r, &rderrbits, 2998 nd->nd_cred, p, isdgram, 0, 2999 supports_nfsv4acls, at_root, 3000 mounted_on_fileno, xattrsupp, 3001 has_hiddensystem, has_namedattr, 3002 clone_blksize); 3003 } else { 3004 dirlen += nfsvno_fillattr(nd, new_mp, 3005 nvp, nvap, &nfh, r, &attrbits, 3006 nd->nd_cred, p, isdgram, 0, 3007 supports_nfsv4acls, at_root, 3008 mounted_on_fileno, xattrsupp, 3009 has_hiddensystem, has_namedattr, 3010 clone_blksize); 3011 } 3012 if (nvp != NULL) 3013 vrele(nvp); 3014 dirlen += (3 * NFSX_UNSIGNED); 3015 } 3016 if (needs_unbusy != 0) 3017 vfs_unbusy(new_mp); 3018 if (dirlen <= cnt) 3019 entrycnt++; 3020 } 3021 invalid: 3022 cpos += dp->d_reclen; 3023 dp = (struct dirent *)cpos; 3024 cookiep++; 3025 ncookies--; 3026 } 3027 vrele(vp); 3028 vfs_unbusy(mp); 3029 3030 /* 3031 * If dirlen > cnt, we must strip off the last entry. If that 3032 * results in an empty reply, report NFSERR_TOOSMALL. 3033 */ 3034 if (dirlen > cnt || nd->nd_repstat) { 3035 if (!nd->nd_repstat && entrycnt == 0) 3036 nd->nd_repstat = NFSERR_TOOSMALL; 3037 if (nd->nd_repstat) { 3038 nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); 3039 if (nd->nd_flag & ND_NFSV3) 3040 nfsrv_postopattr(nd, getret, &at); 3041 } else 3042 nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); 3043 eofflag = 0; 3044 } else if (cpos < cend) 3045 eofflag = 0; 3046 if (!nd->nd_repstat) { 3047 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3048 *tl++ = newnfs_false; 3049 if (eofflag) 3050 *tl = newnfs_true; 3051 else 3052 *tl = newnfs_false; 3053 } 3054 free(cookies, M_TEMP); 3055 free(rbuf, M_TEMP); 3056 3057 out: 3058 NFSEXITCODE2(0, nd); 3059 return (0); 3060 nfsmout: 3061 vput(vp); 3062 NFSEXITCODE2(error, nd); 3063 return (error); 3064 } 3065 3066 /* 3067 * Get the settable attributes out of the mbuf list. 3068 * (Return 0 or EBADRPC) 3069 */ 3070 int 3071 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 3072 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 3073 { 3074 u_int32_t *tl; 3075 struct nfsv2_sattr *sp; 3076 int error = 0, toclient = 0; 3077 3078 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 3079 case ND_NFSV2: 3080 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 3081 /* 3082 * Some old clients didn't fill in the high order 16bits. 3083 * --> check the low order 2 bytes for 0xffff 3084 */ 3085 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 3086 nvap->na_mode = nfstov_mode(sp->sa_mode); 3087 if (sp->sa_uid != newnfs_xdrneg1) 3088 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 3089 if (sp->sa_gid != newnfs_xdrneg1) 3090 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 3091 if (sp->sa_size != newnfs_xdrneg1) 3092 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 3093 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 3094 #ifdef notyet 3095 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 3096 #else 3097 nvap->na_atime.tv_sec = 3098 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 3099 nvap->na_atime.tv_nsec = 0; 3100 #endif 3101 } 3102 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 3103 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 3104 break; 3105 case ND_NFSV3: 3106 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3107 if (*tl == newnfs_true) { 3108 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3109 nvap->na_mode = nfstov_mode(*tl); 3110 } 3111 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3112 if (*tl == newnfs_true) { 3113 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3114 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 3115 } 3116 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3117 if (*tl == newnfs_true) { 3118 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3119 nvap->na_gid = fxdr_unsigned(gid_t, *tl); 3120 } 3121 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3122 if (*tl == newnfs_true) { 3123 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3124 nvap->na_size = fxdr_hyper(tl); 3125 } 3126 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3127 switch (fxdr_unsigned(int, *tl)) { 3128 case NFSV3SATTRTIME_TOCLIENT: 3129 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3130 fxdr_nfsv3time(tl, &nvap->na_atime); 3131 toclient = 1; 3132 break; 3133 case NFSV3SATTRTIME_TOSERVER: 3134 vfs_timestamp(&nvap->na_atime); 3135 nvap->na_vaflags |= VA_UTIMES_NULL; 3136 break; 3137 } 3138 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3139 switch (fxdr_unsigned(int, *tl)) { 3140 case NFSV3SATTRTIME_TOCLIENT: 3141 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3142 fxdr_nfsv3time(tl, &nvap->na_mtime); 3143 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3144 break; 3145 case NFSV3SATTRTIME_TOSERVER: 3146 vfs_timestamp(&nvap->na_mtime); 3147 if (!toclient) 3148 nvap->na_vaflags |= VA_UTIMES_NULL; 3149 break; 3150 } 3151 break; 3152 case ND_NFSV4: 3153 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); 3154 } 3155 nfsmout: 3156 NFSEXITCODE2(error, nd); 3157 return (error); 3158 } 3159 3160 /* 3161 * Handle the setable attributes for V4. 3162 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 3163 */ 3164 int 3165 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 3166 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 3167 { 3168 u_int32_t *tl; 3169 int attrsum = 0; 3170 int i, j; 3171 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 3172 int moderet, toclient = 0; 3173 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 3174 uid_t uid; 3175 gid_t gid; 3176 u_short mode, mask; /* Same type as va_mode. */ 3177 struct vattr va; 3178 3179 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 3180 if (error) 3181 goto nfsmout; 3182 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3183 attrsize = fxdr_unsigned(int, *tl); 3184 3185 /* 3186 * Loop around getting the setable attributes. If an unsupported 3187 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 3188 * Once nd_repstat != 0, do not set the attribute value, but keep 3189 * parsing the attribute(s). 3190 */ 3191 if (retnotsup) { 3192 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3193 bitpos = NFSATTRBIT_MAX; 3194 } else { 3195 bitpos = 0; 3196 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_HIDDEN) || 3197 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SYSTEM)) 3198 nvap->na_flags = 0; 3199 } 3200 moderet = 0; 3201 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 3202 if (attrsum > attrsize) { 3203 error = NFSERR_BADXDR; 3204 goto nfsmout; 3205 } 3206 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 3207 switch (bitpos) { 3208 case NFSATTRBIT_SIZE: 3209 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 3210 if (!nd->nd_repstat) { 3211 if (vp != NULL && vp->v_type != VREG) 3212 nd->nd_repstat = (vp->v_type == VDIR) ? 3213 NFSERR_ISDIR : NFSERR_INVAL; 3214 else 3215 nvap->na_size = fxdr_hyper(tl); 3216 } 3217 attrsum += NFSX_HYPER; 3218 break; 3219 case NFSATTRBIT_ACL: 3220 error = nfsrv_dissectacl(nd, aclp, true, &aceerr, 3221 &aclsize, p); 3222 if (error) 3223 goto nfsmout; 3224 if (aceerr && !nd->nd_repstat) 3225 nd->nd_repstat = aceerr; 3226 attrsum += aclsize; 3227 break; 3228 case NFSATTRBIT_ARCHIVE: 3229 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3230 if (!nd->nd_repstat) 3231 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3232 attrsum += NFSX_UNSIGNED; 3233 break; 3234 case NFSATTRBIT_HIDDEN: 3235 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); 3236 if (nd->nd_repstat == 0) { 3237 if (*tl == newnfs_true) 3238 nvap->na_flags |= UF_HIDDEN; 3239 } 3240 attrsum += NFSX_UNSIGNED; 3241 break; 3242 case NFSATTRBIT_MIMETYPE: 3243 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3244 i = fxdr_unsigned(int, *tl); 3245 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 3246 if (error) 3247 goto nfsmout; 3248 if (!nd->nd_repstat) 3249 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3250 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 3251 break; 3252 case NFSATTRBIT_MODE: 3253 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ 3254 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3255 if (!nd->nd_repstat) 3256 nvap->na_mode = nfstov_mode(*tl); 3257 attrsum += NFSX_UNSIGNED; 3258 break; 3259 case NFSATTRBIT_OWNER: 3260 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3261 j = fxdr_unsigned(int, *tl); 3262 if (j < 0) { 3263 error = NFSERR_BADXDR; 3264 goto nfsmout; 3265 } 3266 if (j > NFSV4_SMALLSTR) 3267 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3268 else 3269 cp = namestr; 3270 error = nfsrv_mtostr(nd, cp, j); 3271 if (error) { 3272 if (j > NFSV4_SMALLSTR) 3273 free(cp, M_NFSSTRING); 3274 goto nfsmout; 3275 } 3276 if (!nd->nd_repstat) { 3277 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, 3278 &uid); 3279 if (!nd->nd_repstat) 3280 nvap->na_uid = uid; 3281 } 3282 if (j > NFSV4_SMALLSTR) 3283 free(cp, M_NFSSTRING); 3284 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3285 break; 3286 case NFSATTRBIT_OWNERGROUP: 3287 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3288 j = fxdr_unsigned(int, *tl); 3289 if (j < 0) { 3290 error = NFSERR_BADXDR; 3291 goto nfsmout; 3292 } 3293 if (j > NFSV4_SMALLSTR) 3294 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3295 else 3296 cp = namestr; 3297 error = nfsrv_mtostr(nd, cp, j); 3298 if (error) { 3299 if (j > NFSV4_SMALLSTR) 3300 free(cp, M_NFSSTRING); 3301 goto nfsmout; 3302 } 3303 if (!nd->nd_repstat) { 3304 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, 3305 &gid); 3306 if (!nd->nd_repstat) 3307 nvap->na_gid = gid; 3308 } 3309 if (j > NFSV4_SMALLSTR) 3310 free(cp, M_NFSSTRING); 3311 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3312 break; 3313 case NFSATTRBIT_SYSTEM: 3314 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); 3315 if (nd->nd_repstat == 0) { 3316 if (*tl == newnfs_true) 3317 nvap->na_flags |= UF_SYSTEM; 3318 } 3319 attrsum += NFSX_UNSIGNED; 3320 break; 3321 case NFSATTRBIT_TIMEACCESSSET: 3322 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3323 attrsum += NFSX_UNSIGNED; 3324 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3325 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3326 if (!nd->nd_repstat) 3327 fxdr_nfsv4time(tl, &nvap->na_atime); 3328 toclient = 1; 3329 attrsum += NFSX_V4TIME; 3330 } else if (!nd->nd_repstat) { 3331 vfs_timestamp(&nvap->na_atime); 3332 nvap->na_vaflags |= VA_UTIMES_NULL; 3333 } 3334 break; 3335 case NFSATTRBIT_TIMEBACKUP: 3336 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3337 if (!nd->nd_repstat) 3338 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3339 attrsum += NFSX_V4TIME; 3340 break; 3341 case NFSATTRBIT_TIMECREATE: 3342 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3343 if (!nd->nd_repstat) 3344 fxdr_nfsv4time(tl, &nvap->na_btime); 3345 attrsum += NFSX_V4TIME; 3346 break; 3347 case NFSATTRBIT_TIMEMODIFYSET: 3348 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3349 attrsum += NFSX_UNSIGNED; 3350 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3351 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3352 if (!nd->nd_repstat) 3353 fxdr_nfsv4time(tl, &nvap->na_mtime); 3354 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3355 attrsum += NFSX_V4TIME; 3356 } else if (!nd->nd_repstat) { 3357 vfs_timestamp(&nvap->na_mtime); 3358 if (!toclient) 3359 nvap->na_vaflags |= VA_UTIMES_NULL; 3360 } 3361 break; 3362 case NFSATTRBIT_MODESETMASKED: 3363 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3364 mode = fxdr_unsigned(u_short, *tl++); 3365 mask = fxdr_unsigned(u_short, *tl); 3366 /* 3367 * vp == NULL implies an Open/Create operation. 3368 * This attribute can only be used for Setattr and 3369 * only for NFSv4.1 or higher. 3370 * If moderet != 0, a mode attribute has also been 3371 * specified and this attribute cannot be done in the 3372 * same Setattr operation. 3373 */ 3374 if (!nd->nd_repstat) { 3375 if ((nd->nd_flag & ND_NFSV41) == 0) 3376 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3377 else if ((mode & ~07777) != 0 || 3378 (mask & ~07777) != 0 || vp == NULL) 3379 nd->nd_repstat = NFSERR_INVAL; 3380 else if (moderet == 0) 3381 moderet = VOP_GETATTR(vp, &va, 3382 nd->nd_cred); 3383 if (moderet == 0) 3384 nvap->na_mode = (mode & mask) | 3385 (va.va_mode & ~mask); 3386 else 3387 nd->nd_repstat = moderet; 3388 } 3389 attrsum += 2 * NFSX_UNSIGNED; 3390 break; 3391 case NFSATTRBIT_MODEUMASK: 3392 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3393 mode = fxdr_unsigned(u_short, *tl++); 3394 mask = fxdr_unsigned(u_short, *tl); 3395 /* 3396 * If moderet != 0, mode has already been done. 3397 * If vp != NULL, this is not a file object creation. 3398 */ 3399 if (!nd->nd_repstat) { 3400 if ((nd->nd_flag & ND_NFSV42) == 0) 3401 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3402 else if ((mask & ~0777) != 0 || vp != NULL || 3403 moderet != 0) 3404 nd->nd_repstat = NFSERR_INVAL; 3405 else 3406 nvap->na_mode = (mode & ~mask); 3407 } 3408 attrsum += 2 * NFSX_UNSIGNED; 3409 break; 3410 default: 3411 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3412 /* 3413 * set bitpos so we drop out of the loop. 3414 */ 3415 bitpos = NFSATTRBIT_MAX; 3416 break; 3417 } 3418 } 3419 3420 /* 3421 * some clients pad the attrlist, so we need to skip over the 3422 * padding. This also skips over unparsed non-supported attributes. 3423 */ 3424 if (attrsum > attrsize) { 3425 error = NFSERR_BADXDR; 3426 } else { 3427 attrsize = NFSM_RNDUP(attrsize); 3428 if (attrsum < attrsize) 3429 error = nfsm_advance(nd, attrsize - attrsum, -1); 3430 } 3431 nfsmout: 3432 NFSEXITCODE2(error, nd); 3433 return (error); 3434 } 3435 3436 /* 3437 * Check/setup export credentials. 3438 */ 3439 int 3440 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 3441 struct ucred *credanon, bool testsec) 3442 { 3443 int error; 3444 3445 /* 3446 * Check/setup credentials. 3447 */ 3448 if (nd->nd_flag & ND_GSS) 3449 exp->nes_exflag &= ~MNT_EXPORTANON; 3450 3451 /* 3452 * Check to see if the operation is allowed for this security flavor. 3453 */ 3454 error = 0; 3455 if (testsec) { 3456 error = nfsvno_testexp(nd, exp); 3457 if (error != 0) 3458 goto out; 3459 } 3460 3461 /* 3462 * Check to see if the file system is exported V4 only. 3463 */ 3464 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 3465 error = NFSERR_PROGNOTV4; 3466 goto out; 3467 } 3468 3469 /* 3470 * Now, map the user credentials. 3471 * (Note that ND_AUTHNONE will only be set for an NFSv3 3472 * Fsinfo RPC. If set for anything else, this code might need 3473 * to change.) 3474 */ 3475 if (NFSVNO_EXPORTED(exp)) { 3476 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || 3477 NFSVNO_EXPORTANON(exp) || 3478 (nd->nd_flag & ND_AUTHNONE) != 0) { 3479 nd->nd_cred->cr_uid = credanon->cr_uid; 3480 nd->nd_cred->cr_gid = credanon->cr_gid; 3481 /* 3482 * 'credanon' is already a 'struct ucred' that was built 3483 * internally with calls to crsetgroups_and_egid(), so 3484 * we don't need a fallback here. 3485 */ 3486 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 3487 credanon->cr_groups); 3488 } else if ((nd->nd_flag & ND_GSS) == 0) { 3489 /* 3490 * If using AUTH_SYS, call nfsrv_getgrpscred() to see 3491 * if there is a replacement credential with a group 3492 * list set up by "nfsuserd -manage-gids". 3493 * If there is no replacement, nfsrv_getgrpscred() 3494 * simply returns its argument. 3495 */ 3496 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); 3497 } 3498 } 3499 3500 out: 3501 NFSEXITCODE2(error, nd); 3502 return (error); 3503 } 3504 3505 /* 3506 * Check exports. 3507 */ 3508 int 3509 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 3510 struct ucred **credp) 3511 { 3512 int error; 3513 3514 error = 0; 3515 *credp = NULL; 3516 MNT_ILOCK(mp); 3517 if (mp->mnt_exjail == NULL || 3518 mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison) 3519 error = EACCES; 3520 MNT_IUNLOCK(mp); 3521 if (error == 0) 3522 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3523 &exp->nes_numsecflavor, exp->nes_secflavors); 3524 if (error) { 3525 if (NFSD_VNET(nfs_rootfhset)) { 3526 exp->nes_exflag = 0; 3527 exp->nes_numsecflavor = 0; 3528 error = 0; 3529 } 3530 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3531 MAXSECFLAVORS) { 3532 printf("nfsvno_checkexp: numsecflavors out of range\n"); 3533 exp->nes_numsecflavor = 0; 3534 error = EACCES; 3535 } 3536 NFSEXITCODE(error); 3537 return (error); 3538 } 3539 3540 /* 3541 * Get a vnode for a file handle and export stuff. 3542 */ 3543 int 3544 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 3545 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 3546 struct ucred **credp) 3547 { 3548 int error; 3549 3550 *credp = NULL; 3551 exp->nes_numsecflavor = 0; 3552 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); 3553 if (error != 0) 3554 /* Make sure the server replies ESTALE to the client. */ 3555 error = ESTALE; 3556 if (nam && !error) { 3557 MNT_ILOCK(mp); 3558 if (mp->mnt_exjail == NULL || 3559 mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison) 3560 error = EACCES; 3561 MNT_IUNLOCK(mp); 3562 if (error == 0) 3563 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3564 &exp->nes_numsecflavor, exp->nes_secflavors); 3565 if (error) { 3566 if (NFSD_VNET(nfs_rootfhset)) { 3567 exp->nes_exflag = 0; 3568 exp->nes_numsecflavor = 0; 3569 error = 0; 3570 } else { 3571 vput(*vpp); 3572 } 3573 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3574 MAXSECFLAVORS) { 3575 printf("nfsvno_fhtovp: numsecflavors out of range\n"); 3576 exp->nes_numsecflavor = 0; 3577 error = EACCES; 3578 vput(*vpp); 3579 } 3580 } 3581 NFSEXITCODE(error); 3582 return (error); 3583 } 3584 3585 /* 3586 * nfsd_fhtovp() - convert a fh to a vnode ptr 3587 * - look up fsid in mount list (if not found ret error) 3588 * - get vp and export rights by calling nfsvno_fhtovp() 3589 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 3590 * for AUTH_SYS 3591 * - if mpp != NULL, return the mount point so that it can 3592 * be used for vn_finished_write() by the caller 3593 */ 3594 void 3595 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 3596 struct vnode **vpp, struct nfsexstuff *exp, 3597 struct mount **mpp, int startwrite, int nextop) 3598 { 3599 struct mount *mp, *mpw; 3600 struct ucred *credanon; 3601 fhandle_t *fhp; 3602 int error; 3603 3604 if (mpp != NULL) 3605 *mpp = NULL; 3606 *vpp = NULL; 3607 fhp = (fhandle_t *)nfp->nfsrvfh_data; 3608 mp = vfs_busyfs(&fhp->fh_fsid); 3609 if (mp == NULL) { 3610 nd->nd_repstat = ESTALE; 3611 goto out; 3612 } 3613 3614 if (startwrite) { 3615 mpw = mp; 3616 error = vn_start_write(NULL, &mpw, V_WAIT); 3617 if (error != 0) { 3618 mpw = NULL; 3619 vfs_unbusy(mp); 3620 nd->nd_repstat = ESTALE; 3621 goto out; 3622 } 3623 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) 3624 lktype = LK_EXCLUSIVE; 3625 } else 3626 mpw = NULL; 3627 3628 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 3629 &credanon); 3630 vfs_unbusy(mp); 3631 3632 if (nd->nd_repstat == 0 && 3633 nfp->nfsrvfh_len >= NFSX_MYFH + NFSX_V4NAMEDDIRFH && 3634 nfp->nfsrvfh_len <= NFSX_MYFH + NFSX_V4NAMEDATTRFH) { 3635 if (nfp->nfsrvfh_len == NFSX_MYFH + NFSX_V4NAMEDDIRFH) 3636 vn_irflag_set_cond(*vpp, VIRF_NAMEDDIR); 3637 else 3638 vn_irflag_set_cond(*vpp, VIRF_NAMEDATTR); 3639 } 3640 3641 /* 3642 * For NFSv4 without a pseudo root fs, unexported file handles 3643 * can be returned, so that Lookup works everywhere. 3644 */ 3645 if (!nd->nd_repstat && exp->nes_exflag == 0 && 3646 !(nd->nd_flag & ND_NFSV4)) { 3647 vput(*vpp); 3648 *vpp = NULL; 3649 nd->nd_repstat = EACCES; 3650 } 3651 3652 /* 3653 * Personally, I've never seen any point in requiring a 3654 * reserved port#, since only in the rare case where the 3655 * clients are all boxes with secure system privileges, 3656 * does it provide any enhanced security, but... some people 3657 * believe it to be useful and keep putting this code back in. 3658 * (There is also some "security checker" out there that 3659 * complains if the nfs server doesn't enforce this.) 3660 * However, note the following: 3661 * RFC3530 (NFSv4) specifies that a reserved port# not be 3662 * required. 3663 * RFC2623 recommends that, if a reserved port# is checked for, 3664 * that there be a way to turn that off--> ifdef'd. 3665 */ 3666 #ifdef NFS_REQRSVPORT 3667 if (!nd->nd_repstat) { 3668 struct sockaddr_in *saddr; 3669 struct sockaddr_in6 *saddr6; 3670 3671 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 3672 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 3673 if (!(nd->nd_flag & ND_NFSV4) && 3674 ((saddr->sin_family == AF_INET && 3675 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 3676 (saddr6->sin6_family == AF_INET6 && 3677 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 3678 vput(*vpp); 3679 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3680 } 3681 } 3682 #endif /* NFS_REQRSVPORT */ 3683 3684 /* 3685 * Check/setup credentials. 3686 */ 3687 if (!nd->nd_repstat) { 3688 nd->nd_saveduid = nd->nd_cred->cr_uid; 3689 nd->nd_repstat = nfsd_excred(nd, exp, credanon, 3690 nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type)); 3691 if (nd->nd_repstat) 3692 vput(*vpp); 3693 } 3694 if (credanon != NULL) 3695 crfree(credanon); 3696 if (nd->nd_repstat) { 3697 vn_finished_write(mpw); 3698 *vpp = NULL; 3699 } else if (mpp != NULL) { 3700 *mpp = mpw; 3701 } 3702 3703 out: 3704 NFSEXITCODE2(0, nd); 3705 } 3706 3707 /* 3708 * glue for fp. 3709 */ 3710 static int 3711 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 3712 { 3713 struct filedesc *fdp; 3714 struct file *fp; 3715 int error = 0; 3716 3717 fdp = p->td_proc->p_fd; 3718 if (fd < 0 || fd >= fdp->fd_nfiles || 3719 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { 3720 error = EBADF; 3721 goto out; 3722 } 3723 *fpp = fp; 3724 3725 out: 3726 NFSEXITCODE(error); 3727 return (error); 3728 } 3729 3730 /* 3731 * Called from nfssvc() to update the exports list. Just call 3732 * vfs_export(). This has to be done, since the v4 root fake fs isn't 3733 * in the mount list. 3734 */ 3735 int 3736 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 3737 { 3738 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 3739 int error = 0; 3740 struct nameidata nd; 3741 fhandle_t fh; 3742 3743 error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export, false); 3744 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 3745 NFSD_VNET(nfs_rootfhset) = 0; 3746 else if (error == 0) { 3747 if (nfsexargp->fspec == NULL) { 3748 error = EPERM; 3749 goto out; 3750 } 3751 /* 3752 * If fspec != NULL, this is the v4root path. 3753 */ 3754 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec); 3755 if ((error = namei(&nd)) != 0) 3756 goto out; 3757 NDFREE_PNBUF(&nd); 3758 error = nfsvno_getfh(nd.ni_vp, &fh, p); 3759 vrele(nd.ni_vp); 3760 if (!error) { 3761 NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH; 3762 NFSBCOPY((caddr_t)&fh, 3763 NFSD_VNET(nfs_rootfh).nfsrvfh_data, 3764 sizeof (fhandle_t)); 3765 NFSD_VNET(nfs_rootfhset) = 1; 3766 } 3767 } 3768 3769 out: 3770 NFSEXITCODE(error); 3771 return (error); 3772 } 3773 3774 /* 3775 * This function needs to test to see if the system is near its limit 3776 * for memory allocation via malloc() or mget() and return True iff 3777 * either of these resources are near their limit. 3778 * XXX (For now, this is just a stub.) 3779 */ 3780 int nfsrv_testmalloclimit = 0; 3781 int 3782 nfsrv_mallocmget_limit(void) 3783 { 3784 static int printmesg = 0; 3785 static int testval = 1; 3786 3787 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 3788 if ((printmesg++ % 100) == 0) 3789 printf("nfsd: malloc/mget near limit\n"); 3790 return (1); 3791 } 3792 return (0); 3793 } 3794 3795 /* 3796 * BSD specific initialization of a mount point. 3797 */ 3798 void 3799 nfsd_mntinit(void) 3800 { 3801 3802 NFSD_LOCK(); 3803 if (NFSD_VNET(nfsrv_mntinited)) { 3804 NFSD_UNLOCK(); 3805 return; 3806 } 3807 NFSD_VNET(nfsrv_mntinited) = true; 3808 nfsrvd_init(0); 3809 NFSD_UNLOCK(); 3810 3811 NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP, 3812 M_WAITOK | M_ZERO); 3813 NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 3814 mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF); 3815 lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0); 3816 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist); 3817 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist); 3818 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 3819 TAILQ_INIT(&NFSD_VNET(nfsv4root_opt)); 3820 TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt)); 3821 NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt); 3822 NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt); 3823 NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0; 3824 NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0; 3825 callout_init(&NFSD_VNET(nfsd_callout), 1); 3826 3827 nfsrvd_initcache(); 3828 nfsd_init(); 3829 } 3830 3831 static void 3832 nfsd_timer(void *arg) 3833 { 3834 struct vnet *vnetp; 3835 3836 vnetp = (struct vnet *)arg; 3837 NFSD_CURVNET_SET_QUIET(vnetp); 3838 nfsrv_servertimer(vnetp); 3839 callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer, 3840 arg, 0); 3841 NFSD_CURVNET_RESTORE(); 3842 } 3843 3844 /* 3845 * Get a vnode for a file handle, without checking exports, etc. 3846 */ 3847 struct vnode * 3848 nfsvno_getvp(fhandle_t *fhp) 3849 { 3850 struct mount *mp; 3851 struct vnode *vp; 3852 int error; 3853 3854 mp = vfs_busyfs(&fhp->fh_fsid); 3855 if (mp == NULL) 3856 return (NULL); 3857 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 3858 vfs_unbusy(mp); 3859 if (error) 3860 return (NULL); 3861 return (vp); 3862 } 3863 3864 /* 3865 * Do a local VOP_ADVLOCK(). 3866 */ 3867 int 3868 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 3869 u_int64_t end, struct thread *td) 3870 { 3871 int error = 0; 3872 struct flock fl; 3873 u_int64_t tlen; 3874 3875 if (nfsrv_dolocallocks == 0) 3876 goto out; 3877 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); 3878 3879 fl.l_whence = SEEK_SET; 3880 fl.l_type = ftype; 3881 fl.l_start = (off_t)first; 3882 if (end == NFS64BITSSET) { 3883 fl.l_len = 0; 3884 } else { 3885 tlen = end - first; 3886 fl.l_len = (off_t)tlen; 3887 } 3888 /* 3889 * For FreeBSD8, the l_pid and l_sysid must be set to the same 3890 * values for all calls, so that all locks will be held by the 3891 * nfsd server. (The nfsd server handles conflicts between the 3892 * various clients.) 3893 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 3894 * bytes, so it can't be put in l_sysid. 3895 */ 3896 if (nfsv4_sysid == 0) 3897 nfsv4_sysid = nlm_acquire_next_sysid(); 3898 fl.l_pid = (pid_t)0; 3899 fl.l_sysid = (int)nfsv4_sysid; 3900 3901 if (ftype == F_UNLCK) 3902 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 3903 (F_POSIX | F_REMOTE)); 3904 else 3905 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 3906 (F_POSIX | F_REMOTE)); 3907 3908 out: 3909 NFSEXITCODE(error); 3910 return (error); 3911 } 3912 3913 /* 3914 * Check the nfsv4 root exports. 3915 */ 3916 int 3917 nfsvno_v4rootexport(struct nfsrv_descript *nd) 3918 { 3919 struct ucred *credanon; 3920 int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; 3921 uint64_t exflags; 3922 3923 error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags, 3924 &credanon, &numsecflavor, secflavors); 3925 if (error) { 3926 error = NFSERR_PROGUNAVAIL; 3927 goto out; 3928 } 3929 if (credanon != NULL) 3930 crfree(credanon); 3931 for (i = 0; i < numsecflavor; i++) { 3932 if (secflavors[i] == AUTH_SYS) 3933 nd->nd_flag |= ND_EXAUTHSYS; 3934 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3935 nd->nd_flag |= ND_EXGSS; 3936 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3937 nd->nd_flag |= ND_EXGSSINTEGRITY; 3938 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3939 nd->nd_flag |= ND_EXGSSPRIVACY; 3940 } 3941 3942 /* And set ND_EXxx flags for TLS. */ 3943 if ((exflags & MNT_EXTLS) != 0) { 3944 nd->nd_flag |= ND_EXTLS; 3945 if ((exflags & MNT_EXTLSCERT) != 0) 3946 nd->nd_flag |= ND_EXTLSCERT; 3947 if ((exflags & MNT_EXTLSCERTUSER) != 0) 3948 nd->nd_flag |= ND_EXTLSCERTUSER; 3949 } 3950 3951 out: 3952 NFSEXITCODE(error); 3953 return (error); 3954 } 3955 3956 /* 3957 * Nfs server pseudo system call for the nfsd's 3958 */ 3959 /* 3960 * MPSAFE 3961 */ 3962 static int 3963 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3964 { 3965 struct file *fp; 3966 struct nfsd_addsock_args sockarg; 3967 struct nfsd_nfsd_args nfsdarg; 3968 struct nfsd_nfsd_oargs onfsdarg; 3969 struct nfsd_pnfsd_args pnfsdarg; 3970 struct vnode *vp, *nvp, *curdvp; 3971 struct pnfsdsfile *pf; 3972 struct nfsdevice *ds, *fds; 3973 cap_rights_t rights; 3974 int buflen, error, ret; 3975 char *buf, *cp, *cp2, *cp3; 3976 char fname[PNFS_FILENAME_LEN + 1]; 3977 3978 NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td)); 3979 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3980 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3981 if (error) 3982 goto out; 3983 /* 3984 * Since we don't know what rights might be required, 3985 * pretend that we need them all. It is better to be too 3986 * careful than too reckless. 3987 */ 3988 error = fget(td, sockarg.sock, 3989 cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); 3990 if (error != 0) 3991 goto out; 3992 if (fp->f_type != DTYPE_SOCKET) { 3993 fdrop(fp, td); 3994 error = EPERM; 3995 goto out; 3996 } 3997 error = nfsrvd_addsock(fp); 3998 fdrop(fp, td); 3999 } else if (uap->flag & NFSSVC_NFSDNFSD) { 4000 if (uap->argp == NULL) { 4001 error = EINVAL; 4002 goto out; 4003 } 4004 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { 4005 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); 4006 if (error == 0) { 4007 nfsdarg.principal = onfsdarg.principal; 4008 nfsdarg.minthreads = onfsdarg.minthreads; 4009 nfsdarg.maxthreads = onfsdarg.maxthreads; 4010 nfsdarg.version = 1; 4011 nfsdarg.addr = NULL; 4012 nfsdarg.addrlen = 0; 4013 nfsdarg.dnshost = NULL; 4014 nfsdarg.dnshostlen = 0; 4015 nfsdarg.dspath = NULL; 4016 nfsdarg.dspathlen = 0; 4017 nfsdarg.mdspath = NULL; 4018 nfsdarg.mdspathlen = 0; 4019 nfsdarg.mirrorcnt = 1; 4020 } 4021 } else 4022 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); 4023 if (error) 4024 goto out; 4025 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && 4026 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && 4027 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && 4028 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && 4029 nfsdarg.mirrorcnt >= 1 && 4030 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && 4031 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && 4032 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { 4033 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" 4034 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, 4035 nfsdarg.dspathlen, nfsdarg.dnshostlen, 4036 nfsdarg.mdspathlen, nfsdarg.mirrorcnt); 4037 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); 4038 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); 4039 if (error != 0) { 4040 free(cp, M_TEMP); 4041 goto out; 4042 } 4043 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ 4044 nfsdarg.addr = cp; 4045 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); 4046 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); 4047 if (error != 0) { 4048 free(nfsdarg.addr, M_TEMP); 4049 free(cp, M_TEMP); 4050 goto out; 4051 } 4052 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ 4053 nfsdarg.dnshost = cp; 4054 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); 4055 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); 4056 if (error != 0) { 4057 free(nfsdarg.addr, M_TEMP); 4058 free(nfsdarg.dnshost, M_TEMP); 4059 free(cp, M_TEMP); 4060 goto out; 4061 } 4062 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ 4063 nfsdarg.dspath = cp; 4064 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); 4065 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); 4066 if (error != 0) { 4067 free(nfsdarg.addr, M_TEMP); 4068 free(nfsdarg.dnshost, M_TEMP); 4069 free(nfsdarg.dspath, M_TEMP); 4070 free(cp, M_TEMP); 4071 goto out; 4072 } 4073 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ 4074 nfsdarg.mdspath = cp; 4075 } else { 4076 nfsdarg.addr = NULL; 4077 nfsdarg.addrlen = 0; 4078 nfsdarg.dnshost = NULL; 4079 nfsdarg.dnshostlen = 0; 4080 nfsdarg.dspath = NULL; 4081 nfsdarg.dspathlen = 0; 4082 nfsdarg.mdspath = NULL; 4083 nfsdarg.mdspathlen = 0; 4084 nfsdarg.mirrorcnt = 1; 4085 } 4086 nfsd_timer(NFSD_TD_TO_VNET(td)); 4087 error = nfsrvd_nfsd(td, &nfsdarg); 4088 callout_drain(&NFSD_VNET(nfsd_callout)); 4089 free(nfsdarg.addr, M_TEMP); 4090 free(nfsdarg.dnshost, M_TEMP); 4091 free(nfsdarg.dspath, M_TEMP); 4092 free(nfsdarg.mdspath, M_TEMP); 4093 } else if (uap->flag & NFSSVC_PNFSDS) { 4094 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); 4095 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || 4096 pnfsdarg.op == PNFSDOP_FORCEDELDS)) { 4097 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4098 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, 4099 NULL); 4100 if (error == 0) 4101 error = nfsrv_deldsserver(pnfsdarg.op, cp, td); 4102 free(cp, M_TEMP); 4103 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { 4104 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4105 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; 4106 buf = malloc(buflen, M_TEMP, M_WAITOK); 4107 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, 4108 NULL); 4109 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); 4110 if (error == 0 && pnfsdarg.dspath != NULL) { 4111 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4112 error = copyinstr(pnfsdarg.dspath, cp2, 4113 PATH_MAX + 1, NULL); 4114 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", 4115 error); 4116 } else 4117 cp2 = NULL; 4118 if (error == 0 && pnfsdarg.curdspath != NULL) { 4119 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4120 error = copyinstr(pnfsdarg.curdspath, cp3, 4121 PATH_MAX + 1, NULL); 4122 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", 4123 error); 4124 } else 4125 cp3 = NULL; 4126 curdvp = NULL; 4127 fds = NULL; 4128 if (error == 0) 4129 error = nfsrv_mdscopymr(cp, cp2, cp3, buf, 4130 &buflen, fname, td, &vp, &nvp, &pf, &ds, 4131 &fds); 4132 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); 4133 if (error == 0) { 4134 if (pf->dsf_dir >= nfsrv_dsdirsize) { 4135 printf("copymr: dsdir out of range\n"); 4136 pf->dsf_dir = 0; 4137 } 4138 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); 4139 error = nfsrv_copymr(vp, nvp, 4140 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, 4141 (struct pnfsdsfile *)buf, 4142 buflen / sizeof(*pf), td->td_ucred, td); 4143 vput(vp); 4144 vput(nvp); 4145 if (fds != NULL && error == 0) { 4146 curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; 4147 ret = vn_lock(curdvp, LK_EXCLUSIVE); 4148 if (ret == 0) { 4149 nfsrv_dsremove(curdvp, fname, 4150 td->td_ucred, td); 4151 NFSVOPUNLOCK(curdvp); 4152 } 4153 } 4154 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); 4155 } 4156 free(cp, M_TEMP); 4157 free(cp2, M_TEMP); 4158 free(cp3, M_TEMP); 4159 free(buf, M_TEMP); 4160 } 4161 } else { 4162 error = nfssvc_srvcall(td, uap, td->td_ucred); 4163 } 4164 4165 out: 4166 NFSD_CURVNET_RESTORE(); 4167 NFSEXITCODE(error); 4168 return (error); 4169 } 4170 4171 static int 4172 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 4173 { 4174 struct nfsex_args export; 4175 struct nfsex_oldargs oexp; 4176 struct file *fp = NULL; 4177 int stablefd, i, len; 4178 struct nfsd_clid adminrevoke; 4179 struct nfsd_dumplist dumplist; 4180 struct nfsd_dumpclients *dumpclients; 4181 struct nfsd_dumplocklist dumplocklist; 4182 struct nfsd_dumplocks *dumplocks; 4183 struct nameidata nd; 4184 vnode_t vp; 4185 int error = EINVAL, igotlock; 4186 struct proc *procp; 4187 gid_t *grps; 4188 4189 if (uap->flag & NFSSVC_PUBLICFH) { 4190 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 4191 sizeof (fhandle_t)); 4192 error = copyin(uap->argp, 4193 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 4194 if (!error) 4195 nfs_pubfhset = 1; 4196 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 4197 (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { 4198 error = copyin(uap->argp,(caddr_t)&export, 4199 sizeof (struct nfsex_args)); 4200 if (!error) { 4201 grps = NULL; 4202 if (export.export.ex_ngroups > NGROUPS_MAX || 4203 export.export.ex_ngroups < 0) 4204 error = EINVAL; 4205 else if (export.export.ex_ngroups > 0) { 4206 grps = malloc(export.export.ex_ngroups * 4207 sizeof(gid_t), M_TEMP, M_WAITOK); 4208 error = copyin(export.export.ex_groups, grps, 4209 export.export.ex_ngroups * sizeof(gid_t)); 4210 export.export.ex_groups = grps; 4211 } else 4212 export.export.ex_groups = NULL; 4213 if (!error) 4214 error = nfsrv_v4rootexport(&export, cred, p); 4215 free(grps, M_TEMP); 4216 } 4217 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 4218 NFSSVC_V4ROOTEXPORT) { 4219 error = copyin(uap->argp,(caddr_t)&oexp, 4220 sizeof (struct nfsex_oldargs)); 4221 if (!error) { 4222 memset(&export.export, 0, sizeof(export.export)); 4223 export.export.ex_flags = (uint64_t)oexp.export.ex_flags; 4224 export.export.ex_root = oexp.export.ex_root; 4225 export.export.ex_uid = oexp.export.ex_anon.cr_uid; 4226 export.export.ex_ngroups = 4227 oexp.export.ex_anon.cr_ngroups; 4228 export.export.ex_groups = NULL; 4229 if (export.export.ex_ngroups > XU_NGROUPS || 4230 export.export.ex_ngroups < 0) 4231 error = EINVAL; 4232 else if (export.export.ex_ngroups > 0) { 4233 export.export.ex_groups = malloc( 4234 export.export.ex_ngroups * sizeof(gid_t), 4235 M_TEMP, M_WAITOK); 4236 for (i = 0; i < export.export.ex_ngroups; i++) 4237 export.export.ex_groups[i] = 4238 oexp.export.ex_anon.cr_groups[i]; 4239 } 4240 export.export.ex_addr = oexp.export.ex_addr; 4241 export.export.ex_addrlen = oexp.export.ex_addrlen; 4242 export.export.ex_mask = oexp.export.ex_mask; 4243 export.export.ex_masklen = oexp.export.ex_masklen; 4244 export.export.ex_indexfile = oexp.export.ex_indexfile; 4245 export.export.ex_numsecflavors = 4246 oexp.export.ex_numsecflavors; 4247 if (export.export.ex_numsecflavors >= MAXSECFLAVORS || 4248 export.export.ex_numsecflavors < 0) 4249 error = EINVAL; 4250 else { 4251 for (i = 0; i < export.export.ex_numsecflavors; 4252 i++) 4253 export.export.ex_secflavors[i] = 4254 oexp.export.ex_secflavors[i]; 4255 } 4256 export.fspec = oexp.fspec; 4257 if (error == 0) 4258 error = nfsrv_v4rootexport(&export, cred, p); 4259 free(export.export.ex_groups, M_TEMP); 4260 } 4261 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 4262 nfs_pubfhset = 0; 4263 error = 0; 4264 } else if (uap->flag & NFSSVC_STABLERESTART) { 4265 error = copyin(uap->argp, (caddr_t)&stablefd, 4266 sizeof (int)); 4267 if (!error) 4268 error = fp_getfvp(p, stablefd, &fp, &vp); 4269 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 4270 error = EBADF; 4271 if (!error && NFSD_VNET(nfsrv_numnfsd) != 0) 4272 error = ENXIO; 4273 if (!error) { 4274 NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp; 4275 nfsrv_setupstable(p); 4276 } 4277 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 4278 error = copyin(uap->argp, (caddr_t)&adminrevoke, 4279 sizeof (struct nfsd_clid)); 4280 if (!error) 4281 error = nfsrv_adminrevoke(&adminrevoke, p); 4282 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 4283 error = copyin(uap->argp, (caddr_t)&dumplist, 4284 sizeof (struct nfsd_dumplist)); 4285 if (!error && (dumplist.ndl_size < 1 || 4286 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 4287 error = EPERM; 4288 if (!error) { 4289 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 4290 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4291 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 4292 error = copyout(dumpclients, dumplist.ndl_list, len); 4293 free(dumpclients, M_TEMP); 4294 } 4295 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 4296 error = copyin(uap->argp, (caddr_t)&dumplocklist, 4297 sizeof (struct nfsd_dumplocklist)); 4298 if (!error && (dumplocklist.ndllck_size < 1 || 4299 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 4300 error = EPERM; 4301 if (!error) 4302 error = nfsrv_lookupfilename(&nd, 4303 dumplocklist.ndllck_fname, p); 4304 if (!error) { 4305 len = sizeof (struct nfsd_dumplocks) * 4306 dumplocklist.ndllck_size; 4307 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4308 nfsrv_dumplocks(nd.ni_vp, dumplocks, 4309 dumplocklist.ndllck_size, p); 4310 vput(nd.ni_vp); 4311 error = copyout(dumplocks, dumplocklist.ndllck_list, 4312 len); 4313 free(dumplocks, M_TEMP); 4314 } 4315 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 4316 procp = p->td_proc; 4317 PROC_LOCK(procp); 4318 nfsd_master_pid = procp->p_pid; 4319 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 4320 nfsd_master_start = procp->p_stats->p_start; 4321 NFSD_VNET(nfsd_master_proc) = procp; 4322 PROC_UNLOCK(procp); 4323 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { 4324 NFSLOCKV4ROOTMUTEX(); 4325 if (!NFSD_VNET(nfsrv_suspend_nfsd)) { 4326 /* Lock out all nfsd threads */ 4327 do { 4328 igotlock = nfsv4_lock( 4329 &NFSD_VNET(nfsd_suspend_lock), 1, NULL, 4330 NFSV4ROOTLOCKMUTEXPTR, NULL); 4331 } while (igotlock == 0 && 4332 !NFSD_VNET(nfsrv_suspend_nfsd)); 4333 NFSD_VNET(nfsrv_suspend_nfsd) = true; 4334 } 4335 NFSUNLOCKV4ROOTMUTEX(); 4336 error = 0; 4337 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { 4338 NFSLOCKV4ROOTMUTEX(); 4339 if (NFSD_VNET(nfsrv_suspend_nfsd)) { 4340 nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0); 4341 NFSD_VNET(nfsrv_suspend_nfsd) = false; 4342 } 4343 NFSUNLOCKV4ROOTMUTEX(); 4344 error = 0; 4345 } 4346 4347 NFSEXITCODE(error); 4348 return (error); 4349 } 4350 4351 /* 4352 * Check exports. 4353 * Returns 0 if ok, 1 otherwise. 4354 */ 4355 int 4356 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 4357 { 4358 int i; 4359 4360 if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) || 4361 (NFSVNO_EXTLSCERT(exp) && 4362 (nd->nd_flag & ND_TLSCERT) == 0) || 4363 (NFSVNO_EXTLSCERTUSER(exp) && 4364 (nd->nd_flag & ND_TLSCERTUSER) == 0)) { 4365 if ((nd->nd_flag & ND_NFSV4) != 0) 4366 return (NFSERR_WRONGSEC); 4367 #ifdef notnow 4368 /* There is currently no auth_stat for this. */ 4369 else if ((nd->nd_flag & ND_TLS) == 0) 4370 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS); 4371 else 4372 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST); 4373 #endif 4374 else 4375 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4376 } 4377 4378 /* 4379 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use 4380 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. 4381 */ 4382 if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO) 4383 return (0); 4384 4385 /* 4386 * This seems odd, but allow the case where the security flavor 4387 * list is empty. This happens when NFSv4 is traversing non-exported 4388 * file systems. Exported file systems should always have a non-empty 4389 * security flavor list. 4390 */ 4391 if (exp->nes_numsecflavor == 0) 4392 return (0); 4393 4394 for (i = 0; i < exp->nes_numsecflavor; i++) { 4395 /* 4396 * The tests for privacy and integrity must be first, 4397 * since ND_GSS is set for everything but AUTH_SYS. 4398 */ 4399 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 4400 (nd->nd_flag & ND_GSSPRIVACY)) 4401 return (0); 4402 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 4403 (nd->nd_flag & ND_GSSINTEGRITY)) 4404 return (0); 4405 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 4406 (nd->nd_flag & ND_GSS)) 4407 return (0); 4408 if (exp->nes_secflavors[i] == AUTH_SYS && 4409 (nd->nd_flag & ND_GSS) == 0) 4410 return (0); 4411 } 4412 if ((nd->nd_flag & ND_NFSV4) != 0) 4413 return (NFSERR_WRONGSEC); 4414 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4415 } 4416 4417 /* 4418 * Calculate a hash value for the fid in a file handle. 4419 */ 4420 uint32_t 4421 nfsrv_hashfh(fhandle_t *fhp) 4422 { 4423 uint32_t hashval; 4424 4425 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 4426 return (hashval); 4427 } 4428 4429 /* 4430 * Calculate a hash value for the sessionid. 4431 */ 4432 uint32_t 4433 nfsrv_hashsessionid(uint8_t *sessionid) 4434 { 4435 uint32_t hashval; 4436 4437 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); 4438 return (hashval); 4439 } 4440 4441 /* 4442 * Signal the userland master nfsd to backup the stable restart file. 4443 */ 4444 void 4445 nfsrv_backupstable(void) 4446 { 4447 struct proc *procp; 4448 4449 if (NFSD_VNET(nfsd_master_proc) != NULL) { 4450 procp = pfind(nfsd_master_pid); 4451 /* Try to make sure it is the correct process. */ 4452 if (procp == NFSD_VNET(nfsd_master_proc) && 4453 procp->p_stats->p_start.tv_sec == 4454 nfsd_master_start.tv_sec && 4455 procp->p_stats->p_start.tv_usec == 4456 nfsd_master_start.tv_usec && 4457 strcmp(procp->p_comm, nfsd_master_comm) == 0) 4458 kern_psignal(procp, SIGUSR2); 4459 else 4460 NFSD_VNET(nfsd_master_proc) = NULL; 4461 4462 if (procp != NULL) 4463 PROC_UNLOCK(procp); 4464 } 4465 } 4466 4467 /* 4468 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. 4469 * The arguments are in a structure, so that they can be passed through 4470 * taskqueue for a kernel process to execute this function. 4471 */ 4472 struct nfsrvdscreate { 4473 int done; 4474 int inprog; 4475 struct task tsk; 4476 struct ucred *tcred; 4477 struct vnode *dvp; 4478 NFSPROC_T *p; 4479 struct pnfsdsfile *pf; 4480 int err; 4481 fhandle_t fh; 4482 struct vattr va; 4483 struct vattr createva; 4484 }; 4485 4486 int 4487 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, 4488 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, 4489 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) 4490 { 4491 struct vnode *nvp; 4492 struct nameidata named; 4493 struct vattr va; 4494 char *bufp; 4495 u_long *hashp; 4496 struct nfsnode *np; 4497 struct nfsmount *nmp; 4498 int error; 4499 4500 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, 4501 LOCKPARENT | LOCKLEAF | NOCACHE); 4502 nfsvno_setpathbuf(&named, &bufp, &hashp); 4503 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 4504 named.ni_cnd.cn_nameptr = bufp; 4505 if (fnamep != NULL) { 4506 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); 4507 named.ni_cnd.cn_namelen = strlen(bufp); 4508 } else 4509 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); 4510 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); 4511 4512 /* Create the date file in the DS mount. */ 4513 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4514 if (error == 0) { 4515 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); 4516 vref(dvp); 4517 VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false); 4518 if (error == 0) { 4519 /* Set the ownership of the file. */ 4520 error = VOP_SETATTR(nvp, nvap, tcred); 4521 NFSD_DEBUG(4, "nfsrv_dscreate:" 4522 " setattr-uid=%d\n", error); 4523 if (error != 0) 4524 vput(nvp); 4525 } 4526 if (error != 0) 4527 printf("pNFS: pnfscreate failed=%d\n", error); 4528 } else 4529 printf("pNFS: pnfscreate vnlock=%d\n", error); 4530 if (error == 0) { 4531 np = VTONFS(nvp); 4532 nmp = VFSTONFS(nvp->v_mount); 4533 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") 4534 != 0 || nmp->nm_nam->sa_len > sizeof( 4535 struct sockaddr_in6) || 4536 np->n_fhp->nfh_len != NFSX_MYFH) { 4537 printf("Bad DS file: fstype=%s salen=%d" 4538 " fhlen=%d\n", 4539 nvp->v_mount->mnt_vfc->vfc_name, 4540 nmp->nm_nam->sa_len, np->n_fhp->nfh_len); 4541 error = ENOENT; 4542 } 4543 4544 /* Set extattrs for the DS on the MDS file. */ 4545 if (error == 0) { 4546 if (dsa != NULL) { 4547 error = VOP_GETATTR(nvp, &va, tcred); 4548 if (error == 0) { 4549 dsa->dsa_filerev = va.va_filerev; 4550 dsa->dsa_size = va.va_size; 4551 dsa->dsa_atime = va.va_atime; 4552 dsa->dsa_mtime = va.va_mtime; 4553 dsa->dsa_bytes = va.va_bytes; 4554 } 4555 } 4556 if (error == 0) { 4557 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, 4558 NFSX_MYFH); 4559 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, 4560 nmp->nm_nam->sa_len); 4561 NFSBCOPY(named.ni_cnd.cn_nameptr, 4562 pf->dsf_filename, 4563 sizeof(pf->dsf_filename)); 4564 } 4565 } else 4566 printf("pNFS: pnfscreate can't get DS" 4567 " attr=%d\n", error); 4568 if (nvpp != NULL && error == 0) 4569 *nvpp = nvp; 4570 else 4571 vput(nvp); 4572 } 4573 nfsvno_relpathbuf(&named); 4574 return (error); 4575 } 4576 4577 /* 4578 * Start up the thread that will execute nfsrv_dscreate(). 4579 */ 4580 static void 4581 start_dscreate(void *arg, int pending) 4582 { 4583 struct nfsrvdscreate *dsc; 4584 4585 dsc = (struct nfsrvdscreate *)arg; 4586 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, 4587 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); 4588 dsc->done = 1; 4589 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); 4590 } 4591 4592 /* 4593 * Create a pNFS data file on the Data Server(s). 4594 */ 4595 static void 4596 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, 4597 NFSPROC_T *p) 4598 { 4599 struct nfsrvdscreate *dsc, *tdsc = NULL; 4600 struct nfsdevice *ds, *tds, *fds; 4601 struct mount *mp; 4602 struct pnfsdsfile *pf, *tpf; 4603 struct pnfsdsattr dsattr; 4604 struct vattr va; 4605 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4606 struct nfsmount *nmp; 4607 fhandle_t fh; 4608 uid_t vauid; 4609 gid_t vagid; 4610 u_short vamode; 4611 struct ucred *tcred; 4612 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; 4613 int failpos, timo; 4614 4615 /* Get a DS server directory in a round-robin order. */ 4616 mirrorcnt = 1; 4617 mp = vp->v_mount; 4618 ds = fds = NULL; 4619 NFSDDSLOCK(); 4620 /* 4621 * Search for the first entry that handles this MDS fs, but use the 4622 * first entry for all MDS fs's otherwise. 4623 */ 4624 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { 4625 if (tds->nfsdev_nmp != NULL) { 4626 if (tds->nfsdev_mdsisset == 0 && ds == NULL) 4627 ds = tds; 4628 else if (tds->nfsdev_mdsisset != 0 && fsidcmp( 4629 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { 4630 ds = fds = tds; 4631 break; 4632 } 4633 } 4634 } 4635 if (ds == NULL) { 4636 NFSDDSUNLOCK(); 4637 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); 4638 return; 4639 } 4640 i = dsdir[0] = ds->nfsdev_nextdir; 4641 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; 4642 dvp[0] = ds->nfsdev_dsdir[i]; 4643 tds = TAILQ_NEXT(ds, nfsdev_list); 4644 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { 4645 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { 4646 if (tds->nfsdev_nmp != NULL && 4647 ((tds->nfsdev_mdsisset == 0 && fds == NULL) || 4648 (tds->nfsdev_mdsisset != 0 && fds != NULL && 4649 fsidcmp(&mp->mnt_stat.f_fsid, 4650 &tds->nfsdev_mdsfsid) == 0))) { 4651 dsdir[mirrorcnt] = i; 4652 dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; 4653 mirrorcnt++; 4654 if (mirrorcnt >= nfsrv_maxpnfsmirror) 4655 break; 4656 } 4657 } 4658 } 4659 /* Put at end of list to implement round-robin usage. */ 4660 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 4661 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 4662 NFSDDSUNLOCK(); 4663 dsc = NULL; 4664 if (mirrorcnt > 1) 4665 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, 4666 M_WAITOK | M_ZERO); 4667 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | 4668 M_ZERO); 4669 4670 error = nfsvno_getfh(vp, &fh, p); 4671 if (error == 0) 4672 error = VOP_GETATTR(vp, &va, cred); 4673 if (error == 0) { 4674 /* Set the attributes for "vp" to Setattr the DS vp. */ 4675 vauid = va.va_uid; 4676 vagid = va.va_gid; 4677 vamode = va.va_mode; 4678 VATTR_NULL(&va); 4679 va.va_uid = vauid; 4680 va.va_gid = vagid; 4681 va.va_mode = vamode; 4682 va.va_size = 0; 4683 } else 4684 printf("pNFS: pnfscreate getfh+attr=%d\n", error); 4685 4686 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, 4687 cred->cr_gid); 4688 /* Make data file name based on FH. */ 4689 tcred = newnfs_getcred(); 4690 4691 /* 4692 * Create the file on each DS mirror, using kernel process(es) for the 4693 * additional mirrors. 4694 */ 4695 failpos = -1; 4696 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { 4697 tpf->dsf_dir = dsdir[i]; 4698 tdsc->tcred = tcred; 4699 tdsc->p = p; 4700 tdsc->pf = tpf; 4701 tdsc->createva = *vap; 4702 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); 4703 tdsc->va = va; 4704 tdsc->dvp = dvp[i]; 4705 tdsc->done = 0; 4706 tdsc->inprog = 0; 4707 tdsc->err = 0; 4708 ret = EIO; 4709 if (nfs_pnfsiothreads != 0) { 4710 ret = nfs_pnfsio(start_dscreate, tdsc); 4711 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); 4712 } 4713 if (ret != 0) { 4714 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, 4715 NULL, tcred, p, NULL); 4716 if (ret != 0) { 4717 KASSERT(error == 0, ("nfsrv_dscreate err=%d", 4718 error)); 4719 if (failpos == -1 && nfsds_failerr(ret)) 4720 failpos = i; 4721 else 4722 error = ret; 4723 } 4724 } 4725 } 4726 if (error == 0) { 4727 tpf->dsf_dir = dsdir[mirrorcnt - 1]; 4728 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, 4729 &dsattr, NULL, tcred, p, NULL); 4730 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { 4731 failpos = mirrorcnt - 1; 4732 error = 0; 4733 } 4734 } 4735 timo = hz / 50; /* Wait for 20msec. */ 4736 if (timo < 1) 4737 timo = 1; 4738 /* Wait for kernel task(s) to complete. */ 4739 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { 4740 while (tdsc->inprog != 0 && tdsc->done == 0) 4741 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); 4742 if (tdsc->err != 0) { 4743 if (failpos == -1 && nfsds_failerr(tdsc->err)) 4744 failpos = i; 4745 else if (error == 0) 4746 error = tdsc->err; 4747 } 4748 } 4749 4750 /* 4751 * If failpos has been set, that mirror has failed, so it needs 4752 * to be disabled. 4753 */ 4754 if (failpos >= 0) { 4755 nmp = VFSTONFS(dvp[failpos]->v_mount); 4756 NFSLOCKMNT(nmp); 4757 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4758 NFSMNTP_CANCELRPCS)) == 0) { 4759 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4760 NFSUNLOCKMNT(nmp); 4761 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4762 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, 4763 ds); 4764 if (ds != NULL) 4765 nfsrv_killrpcs(nmp); 4766 NFSLOCKMNT(nmp); 4767 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4768 wakeup(nmp); 4769 } 4770 NFSUNLOCKMNT(nmp); 4771 } 4772 4773 NFSFREECRED(tcred); 4774 if (error == 0) { 4775 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); 4776 4777 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", 4778 mirrorcnt, nfsrv_maxpnfsmirror); 4779 /* 4780 * For all mirrors that couldn't be created, fill in the 4781 * *pf structure, but with an IP address == 0.0.0.0. 4782 */ 4783 tpf = pf + mirrorcnt; 4784 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { 4785 *tpf = *pf; 4786 tpf->dsf_sin.sin_family = AF_INET; 4787 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); 4788 tpf->dsf_sin.sin_addr.s_addr = 0; 4789 tpf->dsf_sin.sin_port = 0; 4790 } 4791 4792 error = vn_extattr_set(vp, IO_NODELOCKED, 4793 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 4794 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); 4795 if (error == 0) 4796 error = vn_extattr_set(vp, IO_NODELOCKED, 4797 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", 4798 sizeof(dsattr), (char *)&dsattr, p); 4799 if (error != 0) 4800 printf("pNFS: pnfscreate setextattr=%d\n", 4801 error); 4802 } else 4803 printf("pNFS: pnfscreate=%d\n", error); 4804 free(pf, M_TEMP); 4805 free(dsc, M_TEMP); 4806 } 4807 4808 /* 4809 * Get the information needed to remove the pNFS Data Server file from the 4810 * Metadata file. Upon success, ddvp is set non-NULL to the locked 4811 * DS directory vnode. The caller must unlock *ddvp when done with it. 4812 */ 4813 static void 4814 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, 4815 int *mirrorcntp, char *fname, fhandle_t *fhp) 4816 { 4817 struct vattr va; 4818 struct ucred *tcred; 4819 char *buf; 4820 int buflen, error; 4821 4822 dvpp[0] = NULL; 4823 /* If not an exported regular file or not a pNFS server, just return. */ 4824 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4825 nfsrv_devidcnt == 0) 4826 return; 4827 4828 /* Check to see if this is the last hard link. */ 4829 tcred = newnfs_getcred(); 4830 error = VOP_GETATTR(vp, &va, tcred); 4831 NFSFREECRED(tcred); 4832 if (error != 0) { 4833 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); 4834 return; 4835 } 4836 if (va.va_nlink > 1) 4837 return; 4838 4839 error = nfsvno_getfh(vp, fhp, p); 4840 if (error != 0) { 4841 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); 4842 return; 4843 } 4844 4845 buflen = 1024; 4846 buf = malloc(buflen, M_TEMP, M_WAITOK); 4847 /* Get the directory vnode for the DS mount and the file handle. */ 4848 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, 4849 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); 4850 free(buf, M_TEMP); 4851 if (error != 0) 4852 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); 4853 } 4854 4855 /* 4856 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. 4857 * The arguments are in a structure, so that they can be passed through 4858 * taskqueue for a kernel process to execute this function. 4859 */ 4860 struct nfsrvdsremove { 4861 int done; 4862 int inprog; 4863 struct task tsk; 4864 struct ucred *tcred; 4865 struct vnode *dvp; 4866 NFSPROC_T *p; 4867 int err; 4868 char fname[PNFS_FILENAME_LEN + 1]; 4869 }; 4870 4871 static int 4872 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, 4873 NFSPROC_T *p) 4874 { 4875 struct nameidata named; 4876 struct vnode *nvp; 4877 char *bufp; 4878 u_long *hashp; 4879 int error; 4880 4881 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4882 if (error != 0) 4883 return (error); 4884 named.ni_cnd.cn_nameiop = DELETE; 4885 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 4886 named.ni_cnd.cn_cred = tcred; 4887 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 4888 nfsvno_setpathbuf(&named, &bufp, &hashp); 4889 named.ni_cnd.cn_nameptr = bufp; 4890 named.ni_cnd.cn_namelen = strlen(fname); 4891 strlcpy(bufp, fname, NAME_MAX); 4892 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); 4893 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 4894 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); 4895 if (error == 0) { 4896 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); 4897 vput(nvp); 4898 } 4899 NFSVOPUNLOCK(dvp); 4900 nfsvno_relpathbuf(&named); 4901 if (error != 0) 4902 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); 4903 return (error); 4904 } 4905 4906 /* 4907 * Start up the thread that will execute nfsrv_dsremove(). 4908 */ 4909 static void 4910 start_dsremove(void *arg, int pending) 4911 { 4912 struct nfsrvdsremove *dsrm; 4913 4914 dsrm = (struct nfsrvdsremove *)arg; 4915 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, 4916 dsrm->p); 4917 dsrm->done = 1; 4918 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); 4919 } 4920 4921 /* 4922 * Remove a pNFS data file from a Data Server. 4923 * nfsrv_pnfsremovesetup() must have been called before the MDS file was 4924 * removed to set up the dvp and fill in the FH. 4925 */ 4926 static void 4927 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, 4928 NFSPROC_T *p) 4929 { 4930 struct ucred *tcred; 4931 struct nfsrvdsremove *dsrm, *tdsrm; 4932 struct nfsdevice *ds; 4933 struct nfsmount *nmp; 4934 int failpos, i, ret, timo; 4935 4936 tcred = newnfs_getcred(); 4937 dsrm = NULL; 4938 if (mirrorcnt > 1) 4939 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); 4940 /* 4941 * Remove the file on each DS mirror, using kernel process(es) for the 4942 * additional mirrors. 4943 */ 4944 failpos = -1; 4945 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4946 tdsrm->tcred = tcred; 4947 tdsrm->p = p; 4948 tdsrm->dvp = dvp[i]; 4949 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); 4950 tdsrm->inprog = 0; 4951 tdsrm->done = 0; 4952 tdsrm->err = 0; 4953 ret = EIO; 4954 if (nfs_pnfsiothreads != 0) { 4955 ret = nfs_pnfsio(start_dsremove, tdsrm); 4956 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); 4957 } 4958 if (ret != 0) { 4959 ret = nfsrv_dsremove(dvp[i], fname, tcred, p); 4960 if (failpos == -1 && nfsds_failerr(ret)) 4961 failpos = i; 4962 } 4963 } 4964 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); 4965 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) 4966 failpos = mirrorcnt - 1; 4967 timo = hz / 50; /* Wait for 20msec. */ 4968 if (timo < 1) 4969 timo = 1; 4970 /* Wait for kernel task(s) to complete. */ 4971 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4972 while (tdsrm->inprog != 0 && tdsrm->done == 0) 4973 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); 4974 if (failpos == -1 && nfsds_failerr(tdsrm->err)) 4975 failpos = i; 4976 } 4977 4978 /* 4979 * If failpos has been set, that mirror has failed, so it needs 4980 * to be disabled. 4981 */ 4982 if (failpos >= 0) { 4983 nmp = VFSTONFS(dvp[failpos]->v_mount); 4984 NFSLOCKMNT(nmp); 4985 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4986 NFSMNTP_CANCELRPCS)) == 0) { 4987 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4988 NFSUNLOCKMNT(nmp); 4989 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4990 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, 4991 ds); 4992 if (ds != NULL) 4993 nfsrv_killrpcs(nmp); 4994 NFSLOCKMNT(nmp); 4995 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4996 wakeup(nmp); 4997 } 4998 NFSUNLOCKMNT(nmp); 4999 } 5000 5001 /* Get rid all layouts for the file. */ 5002 nfsrv_freefilelayouts(fhp); 5003 5004 NFSFREECRED(tcred); 5005 free(dsrm, M_TEMP); 5006 } 5007 5008 /* 5009 * Generate a file name based on the file handle and put it in *bufp. 5010 * Return the number of bytes generated. 5011 */ 5012 static int 5013 nfsrv_putfhname(fhandle_t *fhp, char *bufp) 5014 { 5015 int i; 5016 uint8_t *cp; 5017 const uint8_t *hexdigits = "0123456789abcdef"; 5018 5019 cp = (uint8_t *)fhp; 5020 for (i = 0; i < sizeof(*fhp); i++) { 5021 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; 5022 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; 5023 } 5024 bufp[2 * i] = '\0'; 5025 return (2 * i); 5026 } 5027 5028 /* 5029 * Update the Metadata file's attributes from the DS file when a Read/Write 5030 * layout is returned. 5031 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN 5032 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. 5033 */ 5034 int 5035 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 5036 { 5037 struct ucred *tcred; 5038 int error; 5039 5040 /* Do this as root so that it won't fail with EACCES. */ 5041 tcred = newnfs_getcred(); 5042 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, 5043 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); 5044 NFSFREECRED(tcred); 5045 return (error); 5046 } 5047 5048 /* 5049 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. 5050 */ 5051 static int 5052 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, 5053 NFSPROC_T *p) 5054 { 5055 int error; 5056 5057 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, 5058 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); 5059 return (error); 5060 } 5061 5062 static int 5063 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 5064 struct thread *p, int ioproc, struct mbuf **mpp, char *cp, 5065 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, 5066 off_t *offp, int content, bool *eofp) 5067 { 5068 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; 5069 fhandle_t fh[NFSDEV_MAXMIRRORS]; 5070 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 5071 struct nfsdevice *ds; 5072 struct pnfsdsattr dsattr; 5073 struct opnfsdsattr odsattr; 5074 char *buf; 5075 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; 5076 5077 NFSD_DEBUG(4, "in nfsrv_proxyds\n"); 5078 /* 5079 * If not a regular file, not exported or not a pNFS server, 5080 * just return ENOENT. 5081 */ 5082 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 5083 nfsrv_devidcnt == 0) 5084 return (ENOENT); 5085 5086 buflen = 1024; 5087 buf = malloc(buflen, M_TEMP, M_WAITOK); 5088 error = 0; 5089 5090 /* 5091 * For Getattr, get the Change attribute (va_filerev) and size (va_size) 5092 * from the MetaData file's extended attribute. 5093 */ 5094 if (ioproc == NFSPROC_GETATTR) { 5095 error = vn_extattr_get(vp, IO_NODELOCKED, 5096 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, 5097 p); 5098 if (error == 0) { 5099 if (buflen == sizeof(odsattr)) { 5100 NFSBCOPY(buf, &odsattr, buflen); 5101 nap->na_filerev = odsattr.dsa_filerev; 5102 nap->na_size = odsattr.dsa_size; 5103 nap->na_atime = odsattr.dsa_atime; 5104 nap->na_mtime = odsattr.dsa_mtime; 5105 /* 5106 * Fake na_bytes by rounding up na_size. 5107 * Since we don't know the block size, just 5108 * use BLKDEV_IOSIZE. 5109 */ 5110 nap->na_bytes = (odsattr.dsa_size + 5111 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); 5112 } else if (buflen == sizeof(dsattr)) { 5113 NFSBCOPY(buf, &dsattr, buflen); 5114 nap->na_filerev = dsattr.dsa_filerev; 5115 nap->na_size = dsattr.dsa_size; 5116 nap->na_atime = dsattr.dsa_atime; 5117 nap->na_mtime = dsattr.dsa_mtime; 5118 nap->na_bytes = dsattr.dsa_bytes; 5119 } else 5120 error = ENXIO; 5121 } 5122 if (error == 0) { 5123 /* 5124 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() 5125 * returns 0, just return now. nfsrv_checkdsattr() 5126 * returns 0 if there is no Read/Write layout 5127 * plus either an Open/Write_access or Write 5128 * delegation issued to a client for the file. 5129 */ 5130 if (nfsrv_pnfsgetdsattr == 0 || 5131 nfsrv_checkdsattr(vp, p) == 0) { 5132 free(buf, M_TEMP); 5133 return (error); 5134 } 5135 } 5136 5137 /* 5138 * Clear ENOATTR so the code below will attempt to do a 5139 * nfsrv_getattrdsrpc() to get the attributes and (re)create 5140 * the extended attribute. 5141 */ 5142 if (error == ENOATTR) 5143 error = 0; 5144 } 5145 5146 origmircnt = -1; 5147 trycnt = 0; 5148 tryagain: 5149 if (error == 0) { 5150 buflen = 1024; 5151 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == 5152 LK_EXCLUSIVE) 5153 printf("nfsrv_proxyds: Readds vp exclusively locked\n"); 5154 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, 5155 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, 5156 NULL, NULL); 5157 if (error == 0) { 5158 for (i = 0; i < mirrorcnt; i++) 5159 nmp[i] = VFSTONFS(dvp[i]->v_mount); 5160 } else 5161 printf("pNFS: proxy getextattr sockaddr=%d\n", error); 5162 } else 5163 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); 5164 if (error == 0) { 5165 failpos = -1; 5166 if (origmircnt == -1) 5167 origmircnt = mirrorcnt; 5168 /* 5169 * If failpos is set to a mirror#, then that mirror has 5170 * failed and will be disabled. For Read, Getattr and Seek, the 5171 * function only tries one mirror, so if that mirror has 5172 * failed, it will need to be retried. As such, increment 5173 * tryitagain for these cases. 5174 * For Write, Setattr and Setacl, the function tries all 5175 * mirrors and will not return an error for the case where 5176 * one mirror has failed. For these cases, the functioning 5177 * mirror(s) will have been modified, so a retry isn't 5178 * necessary. These functions will set failpos for the 5179 * failed mirror#. 5180 */ 5181 if (ioproc == NFSPROC_READDS) { 5182 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], 5183 mpp, mpp2); 5184 if (nfsds_failerr(error) && mirrorcnt > 1) { 5185 /* 5186 * Setting failpos will cause the mirror 5187 * to be disabled and then a retry of this 5188 * read is required. 5189 */ 5190 failpos = 0; 5191 error = 0; 5192 trycnt++; 5193 } 5194 } else if (ioproc == NFSPROC_WRITEDS) 5195 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, 5196 &nmp[0], mirrorcnt, mpp, cp, &failpos); 5197 else if (ioproc == NFSPROC_SETATTR) 5198 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], 5199 mirrorcnt, nap, &failpos); 5200 else if (ioproc == NFSPROC_SETACL) 5201 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], 5202 mirrorcnt, aclp, &failpos); 5203 else if (ioproc == NFSPROC_SEEKDS) { 5204 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, 5205 p, nmp[0]); 5206 if (nfsds_failerr(error) && mirrorcnt > 1) { 5207 /* 5208 * Setting failpos will cause the mirror 5209 * to be disabled and then a retry of this 5210 * read is required. 5211 */ 5212 failpos = 0; 5213 error = 0; 5214 trycnt++; 5215 } 5216 } else if (ioproc == NFSPROC_ALLOCATE) 5217 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, 5218 &nmp[0], mirrorcnt, &failpos); 5219 else if (ioproc == NFSPROC_DEALLOCATE) 5220 error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p, 5221 vp, &nmp[0], mirrorcnt, &failpos); 5222 else { 5223 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, 5224 vp, nmp[mirrorcnt - 1], nap); 5225 if (nfsds_failerr(error) && mirrorcnt > 1) { 5226 /* 5227 * Setting failpos will cause the mirror 5228 * to be disabled and then a retry of this 5229 * getattr is required. 5230 */ 5231 failpos = mirrorcnt - 1; 5232 error = 0; 5233 trycnt++; 5234 } 5235 } 5236 ds = NULL; 5237 if (failpos >= 0) { 5238 failnmp = nmp[failpos]; 5239 NFSLOCKMNT(failnmp); 5240 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | 5241 NFSMNTP_CANCELRPCS)) == 0) { 5242 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; 5243 NFSUNLOCKMNT(failnmp); 5244 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, 5245 failnmp, p); 5246 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", 5247 failpos, ds); 5248 if (ds != NULL) 5249 nfsrv_killrpcs(failnmp); 5250 NFSLOCKMNT(failnmp); 5251 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 5252 wakeup(failnmp); 5253 } 5254 NFSUNLOCKMNT(failnmp); 5255 } 5256 for (i = 0; i < mirrorcnt; i++) 5257 NFSVOPUNLOCK(dvp[i]); 5258 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, 5259 trycnt); 5260 /* Try the Read/Getattr again if a mirror was deleted. */ 5261 if (ds != NULL && trycnt > 0 && trycnt < origmircnt) 5262 goto tryagain; 5263 } else { 5264 /* Return ENOENT for any Extended Attribute error. */ 5265 error = ENOENT; 5266 } 5267 free(buf, M_TEMP); 5268 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); 5269 return (error); 5270 } 5271 5272 /* 5273 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended 5274 * attribute. 5275 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs 5276 * to be checked. If it points to a NULL nmp, then it returns 5277 * a suitable destination. 5278 * curnmp - If non-NULL, it is the source mount for the copy. 5279 */ 5280 int 5281 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, 5282 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, 5283 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, 5284 struct nfsmount *curnmp, int *ippos, int *dsdirp) 5285 { 5286 struct vnode *dvp, *nvp = NULL, **tdvpp; 5287 struct mount *mp; 5288 struct nfsmount *nmp, *newnmp; 5289 struct sockaddr *sad; 5290 struct sockaddr_in *sin; 5291 struct nfsdevice *ds, *tds, *fndds; 5292 struct pnfsdsfile *pf; 5293 uint32_t dsdir; 5294 int error, fhiszero, fnd, gotone, i, mirrorcnt; 5295 5296 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); 5297 *mirrorcntp = 1; 5298 tdvpp = dvpp; 5299 if (nvpp != NULL) 5300 *nvpp = NULL; 5301 if (dvpp != NULL) 5302 *dvpp = NULL; 5303 if (ippos != NULL) 5304 *ippos = -1; 5305 if (newnmpp != NULL) 5306 newnmp = *newnmpp; 5307 else 5308 newnmp = NULL; 5309 mp = vp->v_mount; 5310 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5311 "pnfsd.dsfile", buflenp, buf, p); 5312 mirrorcnt = *buflenp / sizeof(*pf); 5313 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || 5314 *buflenp != sizeof(*pf) * mirrorcnt)) 5315 error = ENOATTR; 5316 5317 pf = (struct pnfsdsfile *)buf; 5318 /* If curnmp != NULL, check for a match in the mirror list. */ 5319 if (curnmp != NULL && error == 0) { 5320 fnd = 0; 5321 for (i = 0; i < mirrorcnt; i++, pf++) { 5322 sad = (struct sockaddr *)&pf->dsf_sin; 5323 if (nfsaddr2_match(sad, curnmp->nm_nam)) { 5324 if (ippos != NULL) 5325 *ippos = i; 5326 fnd = 1; 5327 break; 5328 } 5329 } 5330 if (fnd == 0) 5331 error = ENXIO; 5332 } 5333 5334 gotone = 0; 5335 pf = (struct pnfsdsfile *)buf; 5336 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, 5337 error); 5338 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { 5339 fhiszero = 0; 5340 sad = (struct sockaddr *)&pf->dsf_sin; 5341 sin = &pf->dsf_sin; 5342 dsdir = pf->dsf_dir; 5343 if (dsdir >= nfsrv_dsdirsize) { 5344 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); 5345 error = ENOATTR; 5346 } else if (nvpp != NULL && newnmp != NULL && 5347 nfsaddr2_match(sad, newnmp->nm_nam)) 5348 error = EEXIST; 5349 if (error == 0) { 5350 if (ippos != NULL && curnmp == NULL && 5351 sad->sa_family == AF_INET && 5352 sin->sin_addr.s_addr == 0) 5353 *ippos = i; 5354 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) 5355 fhiszero = 1; 5356 /* Use the socket address to find the mount point. */ 5357 fndds = NULL; 5358 NFSDDSLOCK(); 5359 /* Find a match for the IP address. */ 5360 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5361 if (ds->nfsdev_nmp != NULL) { 5362 dvp = ds->nfsdev_dvp; 5363 nmp = VFSTONFS(dvp->v_mount); 5364 if (nmp != ds->nfsdev_nmp) 5365 printf("different2 nmp %p %p\n", 5366 nmp, ds->nfsdev_nmp); 5367 if (nfsaddr2_match(sad, nmp->nm_nam)) { 5368 fndds = ds; 5369 break; 5370 } 5371 } 5372 } 5373 if (fndds != NULL && newnmpp != NULL && 5374 newnmp == NULL) { 5375 /* Search for a place to make a mirror copy. */ 5376 TAILQ_FOREACH(tds, &nfsrv_devidhead, 5377 nfsdev_list) { 5378 if (tds->nfsdev_nmp != NULL && 5379 fndds != tds && 5380 ((tds->nfsdev_mdsisset == 0 && 5381 fndds->nfsdev_mdsisset == 0) || 5382 (tds->nfsdev_mdsisset != 0 && 5383 fndds->nfsdev_mdsisset != 0 && 5384 fsidcmp(&tds->nfsdev_mdsfsid, 5385 &mp->mnt_stat.f_fsid) == 0))) { 5386 *newnmpp = tds->nfsdev_nmp; 5387 break; 5388 } 5389 } 5390 if (tds != NULL) { 5391 /* 5392 * Move this entry to the end of the 5393 * list, so it won't be selected as 5394 * easily the next time. 5395 */ 5396 TAILQ_REMOVE(&nfsrv_devidhead, tds, 5397 nfsdev_list); 5398 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, 5399 nfsdev_list); 5400 } 5401 } 5402 NFSDDSUNLOCK(); 5403 if (fndds != NULL) { 5404 dvp = fndds->nfsdev_dsdir[dsdir]; 5405 if (lktype != 0 || fhiszero != 0 || 5406 (nvpp != NULL && *nvpp == NULL)) { 5407 if (fhiszero != 0) 5408 error = vn_lock(dvp, 5409 LK_EXCLUSIVE); 5410 else if (lktype != 0) 5411 error = vn_lock(dvp, lktype); 5412 else 5413 error = vn_lock(dvp, LK_SHARED); 5414 /* 5415 * If the file handle is all 0's, try to 5416 * do a Lookup against the DS to acquire 5417 * it. 5418 * If dvpp == NULL or the Lookup fails, 5419 * unlock dvp after the call. 5420 */ 5421 if (error == 0 && (fhiszero != 0 || 5422 (nvpp != NULL && *nvpp == NULL))) { 5423 error = nfsrv_pnfslookupds(vp, 5424 dvp, pf, &nvp, p); 5425 if (error == 0) { 5426 if (fhiszero != 0) 5427 nfsrv_pnfssetfh( 5428 vp, pf, 5429 devid, 5430 fnamep, 5431 nvp, p); 5432 if (nvpp != NULL && 5433 *nvpp == NULL) { 5434 *nvpp = nvp; 5435 *dsdirp = dsdir; 5436 } else 5437 vput(nvp); 5438 } 5439 if (error != 0 || lktype == 0) 5440 NFSVOPUNLOCK(dvp); 5441 } 5442 } 5443 if (error == 0) { 5444 gotone++; 5445 NFSD_DEBUG(4, "gotone=%d\n", gotone); 5446 if (devid != NULL) { 5447 NFSBCOPY(fndds->nfsdev_deviceid, 5448 devid, NFSX_V4DEVICEID); 5449 devid += NFSX_V4DEVICEID; 5450 } 5451 if (dvpp != NULL) 5452 *tdvpp++ = dvp; 5453 if (fhp != NULL) 5454 NFSBCOPY(&pf->dsf_fh, fhp++, 5455 NFSX_MYFH); 5456 if (fnamep != NULL && gotone == 1) 5457 strlcpy(fnamep, 5458 pf->dsf_filename, 5459 sizeof(pf->dsf_filename)); 5460 } else 5461 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " 5462 "err=%d\n", error); 5463 } 5464 } 5465 } 5466 if (error == 0 && gotone == 0) 5467 error = ENOENT; 5468 5469 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, 5470 error); 5471 if (error == 0) 5472 *mirrorcntp = gotone; 5473 else { 5474 if (gotone > 0 && dvpp != NULL) { 5475 /* 5476 * If the error didn't occur on the first one and 5477 * dvpp != NULL, the one(s) prior to the failure will 5478 * have locked dvp's that need to be unlocked. 5479 */ 5480 for (i = 0; i < gotone; i++) { 5481 NFSVOPUNLOCK(*dvpp); 5482 *dvpp++ = NULL; 5483 } 5484 } 5485 /* 5486 * If it found the vnode to be copied from before a failure, 5487 * it needs to be vput()'d. 5488 */ 5489 if (nvpp != NULL && *nvpp != NULL) { 5490 vput(*nvpp); 5491 *nvpp = NULL; 5492 } 5493 } 5494 return (error); 5495 } 5496 5497 /* 5498 * Set the extended attribute for the Change attribute. 5499 */ 5500 static int 5501 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 5502 { 5503 struct pnfsdsattr dsattr; 5504 int error; 5505 5506 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); 5507 dsattr.dsa_filerev = nap->na_filerev; 5508 dsattr.dsa_size = nap->na_size; 5509 dsattr.dsa_atime = nap->na_atime; 5510 dsattr.dsa_mtime = nap->na_mtime; 5511 dsattr.dsa_bytes = nap->na_bytes; 5512 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5513 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); 5514 if (error != 0) 5515 printf("pNFS: setextattr=%d\n", error); 5516 return (error); 5517 } 5518 5519 static int 5520 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5521 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) 5522 { 5523 uint32_t *tl; 5524 struct nfsrv_descript *nd; 5525 nfsv4stateid_t st; 5526 struct mbuf *m, *m2; 5527 int error = 0, retlen, tlen, trimlen; 5528 5529 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); 5530 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5531 *mpp = NULL; 5532 /* 5533 * Use a stateid where other is an alternating 01010 pattern and 5534 * seqid is 0xffffffff. This value is not defined as special by 5535 * the RFC and is used by the FreeBSD NFS server to indicate an 5536 * MDS->DS proxy operation. 5537 */ 5538 st.other[0] = 0x55555555; 5539 st.other[1] = 0x55555555; 5540 st.other[2] = 0x55555555; 5541 st.seqid = 0xffffffff; 5542 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5543 NULL, NULL, 0, 0, cred); 5544 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5545 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); 5546 txdr_hyper(off, tl); 5547 *(tl + 2) = txdr_unsigned(len); 5548 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5549 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5550 if (error != 0) { 5551 free(nd, M_TEMP); 5552 return (error); 5553 } 5554 if (nd->nd_repstat == 0) { 5555 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 5556 NFSM_STRSIZ(retlen, len); 5557 if (retlen > 0) { 5558 /* Trim off the pre-data XDR from the mbuf chain. */ 5559 m = nd->nd_mrep; 5560 while (m != NULL && m != nd->nd_md) { 5561 if (m->m_next == nd->nd_md) { 5562 m->m_next = NULL; 5563 m_freem(nd->nd_mrep); 5564 nd->nd_mrep = m = nd->nd_md; 5565 } else 5566 m = m->m_next; 5567 } 5568 if (m == NULL) { 5569 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5570 error = ENOENT; 5571 goto nfsmout; 5572 } 5573 5574 /* 5575 * Now, adjust first mbuf so that any XDR before the 5576 * read data is skipped over. 5577 */ 5578 trimlen = nd->nd_dpos - mtod(m, char *); 5579 if (trimlen > 0) { 5580 m->m_len -= trimlen; 5581 NFSM_DATAP(m, trimlen); 5582 } 5583 5584 /* 5585 * Truncate the mbuf chain at retlen bytes of data, 5586 * plus XDR padding that brings the length up to a 5587 * multiple of 4. 5588 */ 5589 tlen = NFSM_RNDUP(retlen); 5590 do { 5591 if (m->m_len >= tlen) { 5592 m->m_len = tlen; 5593 tlen = 0; 5594 m2 = m->m_next; 5595 m->m_next = NULL; 5596 m_freem(m2); 5597 break; 5598 } 5599 tlen -= m->m_len; 5600 m = m->m_next; 5601 } while (m != NULL); 5602 if (tlen > 0) { 5603 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5604 error = ENOENT; 5605 goto nfsmout; 5606 } 5607 *mpp = nd->nd_mrep; 5608 *mpendp = m; 5609 nd->nd_mrep = NULL; 5610 } 5611 } else 5612 error = nd->nd_repstat; 5613 nfsmout: 5614 /* If nd->nd_mrep is already NULL, this is a no-op. */ 5615 m_freem(nd->nd_mrep); 5616 free(nd, M_TEMP); 5617 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); 5618 return (error); 5619 } 5620 5621 /* 5622 * Do a write RPC on a DS data file, using this structure for the arguments, 5623 * so that this function can be executed by a separate kernel process. 5624 */ 5625 struct nfsrvwritedsdorpc { 5626 int done; 5627 int inprog; 5628 struct task tsk; 5629 fhandle_t fh; 5630 off_t off; 5631 int len; 5632 struct nfsmount *nmp; 5633 struct ucred *cred; 5634 NFSPROC_T *p; 5635 struct mbuf *m; 5636 int err; 5637 }; 5638 5639 static int 5640 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, 5641 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) 5642 { 5643 uint32_t *tl; 5644 struct nfsrv_descript *nd; 5645 nfsattrbit_t attrbits; 5646 nfsv4stateid_t st; 5647 int commit, error, retlen; 5648 5649 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5650 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, 5651 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5652 5653 /* 5654 * Use a stateid where other is an alternating 01010 pattern and 5655 * seqid is 0xffffffff. This value is not defined as special by 5656 * the RFC and is used by the FreeBSD NFS server to indicate an 5657 * MDS->DS proxy operation. 5658 */ 5659 st.other[0] = 0x55555555; 5660 st.other[1] = 0x55555555; 5661 st.other[2] = 0x55555555; 5662 st.seqid = 0xffffffff; 5663 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5664 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); 5665 txdr_hyper(off, tl); 5666 tl += 2; 5667 /* 5668 * Do all writes FileSync, since the server doesn't hold onto dirty 5669 * buffers. Since clients should be accessing the DS servers directly 5670 * using the pNFS layouts, this just needs to work correctly as a 5671 * fallback. 5672 */ 5673 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); 5674 *tl = txdr_unsigned(len); 5675 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); 5676 5677 /* Put data in mbuf chain. */ 5678 nd->nd_mb->m_next = m; 5679 5680 /* Set nd_mb and nd_bpos to end of data. */ 5681 while (m->m_next != NULL) 5682 m = m->m_next; 5683 nd->nd_mb = m; 5684 nfsm_set(nd, m->m_len); 5685 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); 5686 5687 /* Do a Getattr for the attributes that change upon writing. */ 5688 NFSZERO_ATTRBIT(&attrbits); 5689 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5690 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5691 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5692 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5693 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5694 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5695 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5696 (void) nfsrv_putattrbit(nd, &attrbits); 5697 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5698 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5699 if (error != 0) { 5700 free(nd, M_TEMP); 5701 return (error); 5702 } 5703 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); 5704 /* Get rid of weak cache consistency data for now. */ 5705 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5706 (ND_NFSV4 | ND_V4WCCATTR)) { 5707 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5708 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 5709 NULL); 5710 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); 5711 if (error != 0) 5712 goto nfsmout; 5713 /* 5714 * Get rid of Op# and status for next op. 5715 */ 5716 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5717 if (*++tl != 0) 5718 nd->nd_flag |= ND_NOMOREDATA; 5719 } 5720 if (nd->nd_repstat == 0) { 5721 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); 5722 retlen = fxdr_unsigned(int, *tl++); 5723 commit = fxdr_unsigned(int, *tl); 5724 if (commit != NFSWRITE_FILESYNC) 5725 error = NFSERR_IO; 5726 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", 5727 retlen, commit, error); 5728 } else 5729 error = nd->nd_repstat; 5730 /* We have no use for the Write Verifier since we use FileSync. */ 5731 5732 /* 5733 * Get the Change, Size, Access Time and Modify Time attributes and set 5734 * on the Metadata file, so its attributes will be what the file's 5735 * would be if it had been written. 5736 */ 5737 if (error == 0) { 5738 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5739 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5740 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 5741 NULL); 5742 } 5743 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); 5744 nfsmout: 5745 m_freem(nd->nd_mrep); 5746 free(nd, M_TEMP); 5747 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); 5748 return (error); 5749 } 5750 5751 /* 5752 * Start up the thread that will execute nfsrv_writedsdorpc(). 5753 */ 5754 static void 5755 start_writedsdorpc(void *arg, int pending) 5756 { 5757 struct nfsrvwritedsdorpc *drpc; 5758 5759 drpc = (struct nfsrvwritedsdorpc *)arg; 5760 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5761 drpc->len, NULL, drpc->m, drpc->cred, drpc->p); 5762 drpc->done = 1; 5763 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); 5764 } 5765 5766 static int 5767 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5768 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5769 struct mbuf **mpp, char *cp, int *failposp) 5770 { 5771 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; 5772 struct nfsvattr na; 5773 struct mbuf *m; 5774 int error, i, offs, ret, timo; 5775 5776 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); 5777 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); 5778 drpc = NULL; 5779 if (mirrorcnt > 1) 5780 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5781 M_WAITOK); 5782 5783 /* Calculate offset in mbuf chain that data starts. */ 5784 offs = cp - mtod(*mpp, char *); 5785 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); 5786 5787 /* 5788 * Do the write RPC for every DS, using a separate kernel process 5789 * for every DS except the last one. 5790 */ 5791 error = 0; 5792 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5793 tdrpc->done = 0; 5794 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5795 tdrpc->off = off; 5796 tdrpc->len = len; 5797 tdrpc->nmp = *nmpp; 5798 tdrpc->cred = cred; 5799 tdrpc->p = p; 5800 tdrpc->inprog = 0; 5801 tdrpc->err = 0; 5802 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5803 ret = EIO; 5804 if (nfs_pnfsiothreads != 0) { 5805 ret = nfs_pnfsio(start_writedsdorpc, tdrpc); 5806 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", 5807 ret); 5808 } 5809 if (ret != 0) { 5810 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, 5811 tdrpc->m, cred, p); 5812 if (nfsds_failerr(ret) && *failposp == -1) 5813 *failposp = i; 5814 else if (error == 0 && ret != 0) 5815 error = ret; 5816 } 5817 nmpp++; 5818 fhp++; 5819 } 5820 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5821 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); 5822 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5823 *failposp = mirrorcnt - 1; 5824 else if (error == 0 && ret != 0) 5825 error = ret; 5826 if (error == 0) 5827 error = nfsrv_setextattr(vp, &na, p); 5828 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); 5829 tdrpc = drpc; 5830 timo = hz / 50; /* Wait for 20msec. */ 5831 if (timo < 1) 5832 timo = 1; 5833 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5834 /* Wait for RPCs on separate threads to complete. */ 5835 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5836 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); 5837 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5838 *failposp = i; 5839 else if (error == 0 && tdrpc->err != 0) 5840 error = tdrpc->err; 5841 } 5842 free(drpc, M_TEMP); 5843 return (error); 5844 } 5845 5846 /* 5847 * Do a allocate RPC on a DS data file, using this structure for the arguments, 5848 * so that this function can be executed by a separate kernel process. 5849 */ 5850 struct nfsrvallocatedsdorpc { 5851 int done; 5852 int inprog; 5853 struct task tsk; 5854 fhandle_t fh; 5855 off_t off; 5856 off_t len; 5857 struct nfsmount *nmp; 5858 struct ucred *cred; 5859 NFSPROC_T *p; 5860 int err; 5861 }; 5862 5863 static int 5864 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5865 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5866 { 5867 uint32_t *tl; 5868 struct nfsrv_descript *nd; 5869 nfsattrbit_t attrbits; 5870 nfsv4stateid_t st; 5871 int error; 5872 5873 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5874 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, 5875 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5876 5877 /* 5878 * Use a stateid where other is an alternating 01010 pattern and 5879 * seqid is 0xffffffff. This value is not defined as special by 5880 * the RFC and is used by the FreeBSD NFS server to indicate an 5881 * MDS->DS proxy operation. 5882 */ 5883 st.other[0] = 0x55555555; 5884 st.other[1] = 0x55555555; 5885 st.other[2] = 0x55555555; 5886 st.seqid = 0xffffffff; 5887 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5888 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5889 txdr_hyper(off, tl); tl += 2; 5890 txdr_hyper(len, tl); tl += 2; 5891 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); 5892 5893 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5894 NFSGETATTR_ATTRBIT(&attrbits); 5895 nfsrv_putattrbit(nd, &attrbits); 5896 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5897 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5898 if (error != 0) { 5899 free(nd, M_TEMP); 5900 return (error); 5901 } 5902 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", 5903 nd->nd_repstat); 5904 if (nd->nd_repstat == 0) { 5905 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5906 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5907 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 5908 NULL); 5909 } else 5910 error = nd->nd_repstat; 5911 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); 5912 nfsmout: 5913 m_freem(nd->nd_mrep); 5914 free(nd, M_TEMP); 5915 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); 5916 return (error); 5917 } 5918 5919 /* 5920 * Start up the thread that will execute nfsrv_allocatedsdorpc(). 5921 */ 5922 static void 5923 start_allocatedsdorpc(void *arg, int pending) 5924 { 5925 struct nfsrvallocatedsdorpc *drpc; 5926 5927 drpc = (struct nfsrvallocatedsdorpc *)arg; 5928 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5929 drpc->len, NULL, drpc->cred, drpc->p); 5930 drpc->done = 1; 5931 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); 5932 } 5933 5934 static int 5935 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5936 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5937 int *failposp) 5938 { 5939 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; 5940 struct nfsvattr na; 5941 int error, i, ret, timo; 5942 5943 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); 5944 drpc = NULL; 5945 if (mirrorcnt > 1) 5946 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5947 M_WAITOK); 5948 5949 /* 5950 * Do the allocate RPC for every DS, using a separate kernel process 5951 * for every DS except the last one. 5952 */ 5953 error = 0; 5954 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5955 tdrpc->done = 0; 5956 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5957 tdrpc->off = off; 5958 tdrpc->len = len; 5959 tdrpc->nmp = *nmpp; 5960 tdrpc->cred = cred; 5961 tdrpc->p = p; 5962 tdrpc->inprog = 0; 5963 tdrpc->err = 0; 5964 ret = EIO; 5965 if (nfs_pnfsiothreads != 0) { 5966 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); 5967 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", 5968 ret); 5969 } 5970 if (ret != 0) { 5971 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, 5972 cred, p); 5973 if (nfsds_failerr(ret) && *failposp == -1) 5974 *failposp = i; 5975 else if (error == 0 && ret != 0) 5976 error = ret; 5977 } 5978 nmpp++; 5979 fhp++; 5980 } 5981 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5982 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5983 *failposp = mirrorcnt - 1; 5984 else if (error == 0 && ret != 0) 5985 error = ret; 5986 if (error == 0) 5987 error = nfsrv_setextattr(vp, &na, p); 5988 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); 5989 tdrpc = drpc; 5990 timo = hz / 50; /* Wait for 20msec. */ 5991 if (timo < 1) 5992 timo = 1; 5993 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5994 /* Wait for RPCs on separate threads to complete. */ 5995 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5996 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 5997 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5998 *failposp = i; 5999 else if (error == 0 && tdrpc->err != 0) 6000 error = tdrpc->err; 6001 } 6002 free(drpc, M_TEMP); 6003 return (error); 6004 } 6005 6006 /* 6007 * Do a deallocate RPC on a DS data file, using this structure for the 6008 * arguments, so that this function can be executed by a separate kernel 6009 * process. 6010 */ 6011 struct nfsrvdeallocatedsdorpc { 6012 int done; 6013 int inprog; 6014 struct task tsk; 6015 fhandle_t fh; 6016 off_t off; 6017 off_t len; 6018 struct nfsmount *nmp; 6019 struct ucred *cred; 6020 NFSPROC_T *p; 6021 int err; 6022 }; 6023 6024 static int 6025 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 6026 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 6027 { 6028 uint32_t *tl; 6029 struct nfsrv_descript *nd; 6030 nfsattrbit_t attrbits; 6031 nfsv4stateid_t st; 6032 int error; 6033 6034 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6035 nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp, 6036 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6037 6038 /* 6039 * Use a stateid where other is an alternating 01010 pattern and 6040 * seqid is 0xffffffff. This value is not defined as special by 6041 * the RFC and is used by the FreeBSD NFS server to indicate an 6042 * MDS->DS proxy operation. 6043 */ 6044 st.other[0] = 0x55555555; 6045 st.other[1] = 0x55555555; 6046 st.other[2] = 0x55555555; 6047 st.seqid = 0xffffffff; 6048 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6049 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 6050 txdr_hyper(off, tl); tl += 2; 6051 txdr_hyper(len, tl); tl += 2; 6052 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len); 6053 6054 /* Do a Getattr for the attributes that change upon writing. */ 6055 NFSZERO_ATTRBIT(&attrbits); 6056 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6057 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6058 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6059 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6060 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6061 *tl = txdr_unsigned(NFSV4OP_GETATTR); 6062 nfsrv_putattrbit(nd, &attrbits); 6063 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 6064 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6065 if (error != 0) { 6066 free(nd, M_TEMP); 6067 return (error); 6068 } 6069 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n", 6070 nd->nd_repstat); 6071 /* Get rid of weak cache consistency data for now. */ 6072 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 6073 (ND_NFSV4 | ND_V4WCCATTR)) { 6074 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 6075 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 6076 NULL); 6077 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error); 6078 if (error != 0) 6079 goto nfsmout; 6080 /* 6081 * Get rid of Op# and status for next op. 6082 */ 6083 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6084 if (*++tl != 0) 6085 nd->nd_flag |= ND_NOMOREDATA; 6086 } 6087 if (nd->nd_repstat == 0) { 6088 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6089 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 6090 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 6091 NULL); 6092 } else 6093 error = nd->nd_repstat; 6094 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error); 6095 nfsmout: 6096 m_freem(nd->nd_mrep); 6097 free(nd, M_TEMP); 6098 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error); 6099 return (error); 6100 } 6101 6102 /* 6103 * Start up the thread that will execute nfsrv_deallocatedsdorpc(). 6104 */ 6105 static void 6106 start_deallocatedsdorpc(void *arg, int pending) 6107 { 6108 struct nfsrvdeallocatedsdorpc *drpc; 6109 6110 drpc = (struct nfsrvdeallocatedsdorpc *)arg; 6111 drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 6112 drpc->len, NULL, drpc->cred, drpc->p); 6113 drpc->done = 1; 6114 NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err); 6115 } 6116 6117 static int 6118 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 6119 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 6120 int *failposp) 6121 { 6122 struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL; 6123 struct nfsvattr na; 6124 int error, i, ret, timo; 6125 6126 NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n"); 6127 drpc = NULL; 6128 if (mirrorcnt > 1) 6129 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6130 M_WAITOK); 6131 6132 /* 6133 * Do the deallocate RPC for every DS, using a separate kernel process 6134 * for every DS except the last one. 6135 */ 6136 error = 0; 6137 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6138 tdrpc->done = 0; 6139 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6140 tdrpc->off = off; 6141 tdrpc->len = len; 6142 tdrpc->nmp = *nmpp; 6143 tdrpc->cred = cred; 6144 tdrpc->p = p; 6145 tdrpc->inprog = 0; 6146 tdrpc->err = 0; 6147 ret = EIO; 6148 if (nfs_pnfsiothreads != 0) { 6149 ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc); 6150 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n", 6151 ret); 6152 } 6153 if (ret != 0) { 6154 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, 6155 NULL, cred, p); 6156 if (nfsds_failerr(ret) && *failposp == -1) 6157 *failposp = i; 6158 else if (error == 0 && ret != 0) 6159 error = ret; 6160 } 6161 nmpp++; 6162 fhp++; 6163 } 6164 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 6165 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6166 *failposp = mirrorcnt - 1; 6167 else if (error == 0 && ret != 0) 6168 error = ret; 6169 if (error == 0) 6170 error = nfsrv_setextattr(vp, &na, p); 6171 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error); 6172 tdrpc = drpc; 6173 timo = hz / 50; /* Wait for 20msec. */ 6174 if (timo < 1) 6175 timo = 1; 6176 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6177 /* Wait for RPCs on separate threads to complete. */ 6178 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6179 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 6180 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6181 *failposp = i; 6182 else if (error == 0 && tdrpc->err != 0) 6183 error = tdrpc->err; 6184 } 6185 free(drpc, M_TEMP); 6186 return (error); 6187 } 6188 6189 static int 6190 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6191 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, 6192 struct nfsvattr *dsnap) 6193 { 6194 uint32_t *tl; 6195 struct nfsrv_descript *nd; 6196 nfsv4stateid_t st; 6197 nfsattrbit_t attrbits; 6198 int error; 6199 6200 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); 6201 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6202 /* 6203 * Use a stateid where other is an alternating 01010 pattern and 6204 * seqid is 0xffffffff. This value is not defined as special by 6205 * the RFC and is used by the FreeBSD NFS server to indicate an 6206 * MDS->DS proxy operation. 6207 */ 6208 st.other[0] = 0x55555555; 6209 st.other[1] = 0x55555555; 6210 st.other[2] = 0x55555555; 6211 st.seqid = 0xffffffff; 6212 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), 6213 NULL, NULL, 0, 0, cred); 6214 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6215 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); 6216 6217 /* Do a Getattr for the attributes that change due to writing. */ 6218 NFSZERO_ATTRBIT(&attrbits); 6219 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6220 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6221 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6222 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6223 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6224 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 6225 *tl = txdr_unsigned(NFSV4OP_GETATTR); 6226 (void) nfsrv_putattrbit(nd, &attrbits); 6227 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6228 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6229 if (error != 0) { 6230 free(nd, M_TEMP); 6231 return (error); 6232 } 6233 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", 6234 nd->nd_repstat); 6235 /* Get rid of weak cache consistency data for now. */ 6236 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 6237 (ND_NFSV4 | ND_V4WCCATTR)) { 6238 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 6239 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 6240 NULL, NULL); 6241 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); 6242 if (error != 0) 6243 goto nfsmout; 6244 /* 6245 * Get rid of Op# and status for next op. 6246 */ 6247 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6248 if (*++tl != 0) 6249 nd->nd_flag |= ND_NOMOREDATA; 6250 } 6251 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 6252 if (error != 0) 6253 goto nfsmout; 6254 if (nd->nd_repstat != 0) 6255 error = nd->nd_repstat; 6256 /* 6257 * Get the Change, Size, Access Time and Modify Time attributes and set 6258 * on the Metadata file, so its attributes will be what the file's 6259 * would be if it had been written. 6260 */ 6261 if (error == 0) { 6262 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6263 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 6264 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 6265 NULL, NULL); 6266 } 6267 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); 6268 nfsmout: 6269 m_freem(nd->nd_mrep); 6270 free(nd, M_TEMP); 6271 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); 6272 return (error); 6273 } 6274 6275 struct nfsrvsetattrdsdorpc { 6276 int done; 6277 int inprog; 6278 struct task tsk; 6279 fhandle_t fh; 6280 struct nfsmount *nmp; 6281 struct vnode *vp; 6282 struct ucred *cred; 6283 NFSPROC_T *p; 6284 struct nfsvattr na; 6285 struct nfsvattr dsna; 6286 int err; 6287 }; 6288 6289 /* 6290 * Start up the thread that will execute nfsrv_setattrdsdorpc(). 6291 */ 6292 static void 6293 start_setattrdsdorpc(void *arg, int pending) 6294 { 6295 struct nfsrvsetattrdsdorpc *drpc; 6296 6297 drpc = (struct nfsrvsetattrdsdorpc *)arg; 6298 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, 6299 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); 6300 drpc->done = 1; 6301 } 6302 6303 static int 6304 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6305 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 6306 struct nfsvattr *nap, int *failposp) 6307 { 6308 struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; 6309 struct nfsvattr na; 6310 int error, i, ret, timo; 6311 6312 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); 6313 drpc = NULL; 6314 if (mirrorcnt > 1) 6315 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6316 M_WAITOK); 6317 6318 /* 6319 * Do the setattr RPC for every DS, using a separate kernel process 6320 * for every DS except the last one. 6321 */ 6322 error = 0; 6323 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6324 tdrpc->done = 0; 6325 tdrpc->inprog = 0; 6326 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6327 tdrpc->nmp = *nmpp; 6328 tdrpc->vp = vp; 6329 tdrpc->cred = cred; 6330 tdrpc->p = p; 6331 tdrpc->na = *nap; 6332 tdrpc->err = 0; 6333 ret = EIO; 6334 if (nfs_pnfsiothreads != 0) { 6335 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); 6336 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", 6337 ret); 6338 } 6339 if (ret != 0) { 6340 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, 6341 &na); 6342 if (nfsds_failerr(ret) && *failposp == -1) 6343 *failposp = i; 6344 else if (error == 0 && ret != 0) 6345 error = ret; 6346 } 6347 nmpp++; 6348 fhp++; 6349 } 6350 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); 6351 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6352 *failposp = mirrorcnt - 1; 6353 else if (error == 0 && ret != 0) 6354 error = ret; 6355 if (error == 0) 6356 error = nfsrv_setextattr(vp, &na, p); 6357 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); 6358 tdrpc = drpc; 6359 timo = hz / 50; /* Wait for 20msec. */ 6360 if (timo < 1) 6361 timo = 1; 6362 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6363 /* Wait for RPCs on separate threads to complete. */ 6364 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6365 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); 6366 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6367 *failposp = i; 6368 else if (error == 0 && tdrpc->err != 0) 6369 error = tdrpc->err; 6370 } 6371 free(drpc, M_TEMP); 6372 return (error); 6373 } 6374 6375 /* 6376 * Do a Setattr of an NFSv4 ACL on the DS file. 6377 */ 6378 static int 6379 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6380 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) 6381 { 6382 struct nfsrv_descript *nd; 6383 nfsv4stateid_t st; 6384 nfsattrbit_t attrbits; 6385 int error; 6386 6387 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); 6388 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6389 /* 6390 * Use a stateid where other is an alternating 01010 pattern and 6391 * seqid is 0xffffffff. This value is not defined as special by 6392 * the RFC and is used by the FreeBSD NFS server to indicate an 6393 * MDS->DS proxy operation. 6394 */ 6395 st.other[0] = 0x55555555; 6396 st.other[1] = 0x55555555; 6397 st.other[2] = 0x55555555; 6398 st.seqid = 0xffffffff; 6399 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), 6400 NULL, NULL, 0, 0, cred); 6401 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6402 NFSZERO_ATTRBIT(&attrbits); 6403 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); 6404 /* 6405 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), 6406 * so passing in the metadata "vp" will be ok, since it is of 6407 * the same type (VREG). 6408 */ 6409 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, 6410 NULL, 0, 0, 0, 0, 0, NULL, false, false, false, 0); 6411 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6412 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6413 if (error != 0) { 6414 free(nd, M_TEMP); 6415 return (error); 6416 } 6417 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", 6418 nd->nd_repstat); 6419 error = nd->nd_repstat; 6420 m_freem(nd->nd_mrep); 6421 free(nd, M_TEMP); 6422 return (error); 6423 } 6424 6425 struct nfsrvsetacldsdorpc { 6426 int done; 6427 int inprog; 6428 struct task tsk; 6429 fhandle_t fh; 6430 struct nfsmount *nmp; 6431 struct vnode *vp; 6432 struct ucred *cred; 6433 NFSPROC_T *p; 6434 struct acl *aclp; 6435 int err; 6436 }; 6437 6438 /* 6439 * Start up the thread that will execute nfsrv_setacldsdorpc(). 6440 */ 6441 static void 6442 start_setacldsdorpc(void *arg, int pending) 6443 { 6444 struct nfsrvsetacldsdorpc *drpc; 6445 6446 drpc = (struct nfsrvsetacldsdorpc *)arg; 6447 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, 6448 drpc->vp, drpc->nmp, drpc->aclp); 6449 drpc->done = 1; 6450 } 6451 6452 static int 6453 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6454 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, 6455 int *failposp) 6456 { 6457 struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; 6458 int error, i, ret, timo; 6459 6460 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); 6461 drpc = NULL; 6462 if (mirrorcnt > 1) 6463 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6464 M_WAITOK); 6465 6466 /* 6467 * Do the setattr RPC for every DS, using a separate kernel process 6468 * for every DS except the last one. 6469 */ 6470 error = 0; 6471 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6472 tdrpc->done = 0; 6473 tdrpc->inprog = 0; 6474 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6475 tdrpc->nmp = *nmpp; 6476 tdrpc->vp = vp; 6477 tdrpc->cred = cred; 6478 tdrpc->p = p; 6479 tdrpc->aclp = aclp; 6480 tdrpc->err = 0; 6481 ret = EIO; 6482 if (nfs_pnfsiothreads != 0) { 6483 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); 6484 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", 6485 ret); 6486 } 6487 if (ret != 0) { 6488 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, 6489 aclp); 6490 if (nfsds_failerr(ret) && *failposp == -1) 6491 *failposp = i; 6492 else if (error == 0 && ret != 0) 6493 error = ret; 6494 } 6495 nmpp++; 6496 fhp++; 6497 } 6498 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); 6499 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6500 *failposp = mirrorcnt - 1; 6501 else if (error == 0 && ret != 0) 6502 error = ret; 6503 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); 6504 tdrpc = drpc; 6505 timo = hz / 50; /* Wait for 20msec. */ 6506 if (timo < 1) 6507 timo = 1; 6508 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6509 /* Wait for RPCs on separate threads to complete. */ 6510 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6511 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); 6512 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6513 *failposp = i; 6514 else if (error == 0 && tdrpc->err != 0) 6515 error = tdrpc->err; 6516 } 6517 free(drpc, M_TEMP); 6518 return (error); 6519 } 6520 6521 /* 6522 * Getattr call to the DS for the attributes that change due to writing. 6523 */ 6524 static int 6525 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6526 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) 6527 { 6528 struct nfsrv_descript *nd; 6529 int error; 6530 nfsattrbit_t attrbits; 6531 6532 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); 6533 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6534 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, 6535 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6536 NFSZERO_ATTRBIT(&attrbits); 6537 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6538 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6539 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6540 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6541 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6542 (void) nfsrv_putattrbit(nd, &attrbits); 6543 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6544 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6545 if (error != 0) { 6546 free(nd, M_TEMP); 6547 return (error); 6548 } 6549 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", 6550 nd->nd_repstat); 6551 if (nd->nd_repstat == 0) { 6552 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, 6553 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, 6554 NULL, NULL, NULL, NULL); 6555 /* 6556 * We can only save the updated values in the extended 6557 * attribute if the vp is exclusively locked. 6558 * This should happen when any of the following operations 6559 * occur on the vnode: 6560 * Close, Delegreturn, LayoutCommit, LayoutReturn 6561 * As such, the updated extended attribute should get saved 6562 * before nfsrv_checkdsattr() returns 0 and allows the cached 6563 * attributes to be returned without calling this function. 6564 */ 6565 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { 6566 error = nfsrv_setextattr(vp, nap, p); 6567 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", 6568 error); 6569 } 6570 } else 6571 error = nd->nd_repstat; 6572 m_freem(nd->nd_mrep); 6573 free(nd, M_TEMP); 6574 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); 6575 return (error); 6576 } 6577 6578 /* 6579 * Seek call to a DS. 6580 */ 6581 static int 6582 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, 6583 struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) 6584 { 6585 uint32_t *tl; 6586 struct nfsrv_descript *nd; 6587 nfsv4stateid_t st; 6588 int error; 6589 6590 NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); 6591 /* 6592 * Use a stateid where other is an alternating 01010 pattern and 6593 * seqid is 0xffffffff. This value is not defined as special by 6594 * the RFC and is used by the FreeBSD NFS server to indicate an 6595 * MDS->DS proxy operation. 6596 */ 6597 st.other[0] = 0x55555555; 6598 st.other[1] = 0x55555555; 6599 st.other[2] = 0x55555555; 6600 st.seqid = 0xffffffff; 6601 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6602 nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, 6603 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6604 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6605 NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); 6606 txdr_hyper(*offp, tl); tl += 2; 6607 *tl = txdr_unsigned(content); 6608 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6609 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6610 if (error != 0) { 6611 free(nd, M_TEMP); 6612 return (error); 6613 } 6614 NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); 6615 if (nd->nd_repstat == 0) { 6616 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); 6617 if (*tl++ == newnfs_true) 6618 *eofp = true; 6619 else 6620 *eofp = false; 6621 *offp = fxdr_hyper(tl); 6622 } else 6623 error = nd->nd_repstat; 6624 nfsmout: 6625 m_freem(nd->nd_mrep); 6626 free(nd, M_TEMP); 6627 NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); 6628 return (error); 6629 } 6630 6631 /* 6632 * Get the device id and file handle for a DS file. 6633 */ 6634 int 6635 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, 6636 fhandle_t *fhp, char *devid) 6637 { 6638 int buflen, error; 6639 char *buf; 6640 6641 buflen = 1024; 6642 buf = malloc(buflen, M_TEMP, M_WAITOK); 6643 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, 6644 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); 6645 free(buf, M_TEMP); 6646 return (error); 6647 } 6648 6649 /* 6650 * Do a Lookup against the DS for the filename. 6651 */ 6652 static int 6653 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, 6654 struct vnode **nvpp, NFSPROC_T *p) 6655 { 6656 struct nameidata named; 6657 struct ucred *tcred; 6658 char *bufp; 6659 u_long *hashp; 6660 struct vnode *nvp; 6661 int error; 6662 6663 tcred = newnfs_getcred(); 6664 named.ni_cnd.cn_nameiop = LOOKUP; 6665 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; 6666 named.ni_cnd.cn_cred = tcred; 6667 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 6668 nfsvno_setpathbuf(&named, &bufp, &hashp); 6669 named.ni_cnd.cn_nameptr = bufp; 6670 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); 6671 strlcpy(bufp, pf->dsf_filename, NAME_MAX); 6672 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); 6673 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 6674 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); 6675 NFSFREECRED(tcred); 6676 nfsvno_relpathbuf(&named); 6677 if (error == 0) 6678 *nvpp = nvp; 6679 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); 6680 return (error); 6681 } 6682 6683 /* 6684 * Set the file handle to the correct one. 6685 */ 6686 static void 6687 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, 6688 char *fnamep, struct vnode *nvp, NFSPROC_T *p) 6689 { 6690 struct nfsnode *np; 6691 int ret = 0; 6692 6693 np = VTONFS(nvp); 6694 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); 6695 /* 6696 * We can only do a vn_set_extattr() if the vnode is exclusively 6697 * locked and vn_start_write() has been done. If devid != NULL or 6698 * fnamep != NULL or the vnode is shared locked, vn_start_write() 6699 * may not have been done. 6700 * If not done now, it will be done on a future call. 6701 */ 6702 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == 6703 LK_EXCLUSIVE) 6704 ret = vn_extattr_set(vp, IO_NODELOCKED, 6705 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), 6706 (char *)pf, p); 6707 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); 6708 } 6709 6710 /* 6711 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point 6712 * when the DS has failed. 6713 */ 6714 void 6715 nfsrv_killrpcs(struct nfsmount *nmp) 6716 { 6717 6718 /* 6719 * Call newnfs_nmcancelreqs() to cause 6720 * any RPCs in progress on the mount point to 6721 * fail. 6722 * This will cause any process waiting for an 6723 * RPC to complete while holding a vnode lock 6724 * on the mounted-on vnode (such as "df" or 6725 * a non-forced "umount") to fail. 6726 * This will unlock the mounted-on vnode so 6727 * a forced dismount can succeed. 6728 * The NFSMNTP_CANCELRPCS flag should be set when this function is 6729 * called. 6730 */ 6731 newnfs_nmcancelreqs(nmp); 6732 } 6733 6734 /* 6735 * Sum up the statfs info for each of the DSs, so that the client will 6736 * receive the total for all DSs. 6737 */ 6738 static int 6739 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) 6740 { 6741 struct statfs *tsf; 6742 struct nfsdevice *ds; 6743 struct vnode **dvpp, **tdvpp, *dvp; 6744 uint64_t tot; 6745 int cnt, error = 0, i; 6746 6747 if (nfsrv_devidcnt <= 0) 6748 return (ENXIO); 6749 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 6750 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 6751 6752 /* Get an array of the dvps for the DSs. */ 6753 tdvpp = dvpp; 6754 i = 0; 6755 NFSDDSLOCK(); 6756 /* First, search for matches for same file system. */ 6757 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6758 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && 6759 fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { 6760 if (++i > nfsrv_devidcnt) 6761 break; 6762 *tdvpp++ = ds->nfsdev_dvp; 6763 } 6764 } 6765 /* 6766 * If no matches for same file system, total all servers not assigned 6767 * to a file system. 6768 */ 6769 if (i == 0) { 6770 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6771 if (ds->nfsdev_nmp != NULL && 6772 ds->nfsdev_mdsisset == 0) { 6773 if (++i > nfsrv_devidcnt) 6774 break; 6775 *tdvpp++ = ds->nfsdev_dvp; 6776 } 6777 } 6778 } 6779 NFSDDSUNLOCK(); 6780 cnt = i; 6781 6782 /* Do a VFS_STATFS() for each of the DSs and sum them up. */ 6783 tdvpp = dvpp; 6784 for (i = 0; i < cnt && error == 0; i++) { 6785 dvp = *tdvpp++; 6786 error = VFS_STATFS(dvp->v_mount, tsf); 6787 if (error == 0) { 6788 if (sf->f_bsize == 0) { 6789 if (tsf->f_bsize > 0) 6790 sf->f_bsize = tsf->f_bsize; 6791 else 6792 sf->f_bsize = 8192; 6793 } 6794 if (tsf->f_blocks > 0) { 6795 if (sf->f_bsize != tsf->f_bsize) { 6796 tot = tsf->f_blocks * tsf->f_bsize; 6797 sf->f_blocks += (tot / sf->f_bsize); 6798 } else 6799 sf->f_blocks += tsf->f_blocks; 6800 } 6801 if (tsf->f_bfree > 0) { 6802 if (sf->f_bsize != tsf->f_bsize) { 6803 tot = tsf->f_bfree * tsf->f_bsize; 6804 sf->f_bfree += (tot / sf->f_bsize); 6805 } else 6806 sf->f_bfree += tsf->f_bfree; 6807 } 6808 if (tsf->f_bavail > 0) { 6809 if (sf->f_bsize != tsf->f_bsize) { 6810 tot = tsf->f_bavail * tsf->f_bsize; 6811 sf->f_bavail += (tot / sf->f_bsize); 6812 } else 6813 sf->f_bavail += tsf->f_bavail; 6814 } 6815 } 6816 } 6817 free(tsf, M_TEMP); 6818 free(dvpp, M_TEMP); 6819 return (error); 6820 } 6821 6822 /* 6823 * Set an NFSv4 acl. 6824 */ 6825 int 6826 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) 6827 { 6828 int error; 6829 6830 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { 6831 error = NFSERR_ATTRNOTSUPP; 6832 goto out; 6833 } 6834 /* 6835 * With NFSv4 ACLs, chmod(2) may need to add additional entries. 6836 * Make sure it has enough room for that - splitting every entry 6837 * into two and appending "canonical six" entries at the end. 6838 * Cribbed out of kern/vfs_acl.c - Rick M. 6839 */ 6840 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { 6841 error = NFSERR_ATTRNOTSUPP; 6842 goto out; 6843 } 6844 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 6845 if (error == 0) { 6846 error = nfsrv_dssetacl(vp, aclp, cred, p); 6847 if (error == ENOENT) 6848 error = 0; 6849 } 6850 6851 out: 6852 NFSEXITCODE(error); 6853 return (error); 6854 } 6855 6856 /* 6857 * Seek vnode op call (actually it is a VOP_IOCTL()). 6858 * This function is called with the vnode locked, but unlocks and vrele()s 6859 * the vp before returning. 6860 */ 6861 int 6862 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, 6863 off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) 6864 { 6865 struct nfsvattr at; 6866 int error, ret; 6867 6868 ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); 6869 /* 6870 * Attempt to seek on a DS file. A return of ENOENT implies 6871 * there is no DS file to seek on. 6872 */ 6873 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, 6874 NULL, NULL, NULL, NULL, offp, content, eofp); 6875 if (error != ENOENT) { 6876 vput(vp); 6877 return (error); 6878 } 6879 6880 /* 6881 * Do the VOP_IOCTL() call. For the case where *offp == file_size, 6882 * VOP_IOCTL() will return ENXIO. However, the correct reply for 6883 * NFSv4.2 is *eofp == true and error == 0 for this case. 6884 */ 6885 NFSVOPUNLOCK(vp); 6886 error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); 6887 *eofp = false; 6888 if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { 6889 /* Handle the cases where we might be at EOF. */ 6890 ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); 6891 if (ret == 0 && *offp == at.na_size) { 6892 *eofp = true; 6893 error = 0; 6894 } 6895 if (ret != 0 && error == 0) 6896 error = ret; 6897 } 6898 vrele(vp); 6899 NFSEXITCODE(error); 6900 return (error); 6901 } 6902 6903 /* 6904 * Allocate vnode op call. 6905 */ 6906 int 6907 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6908 NFSPROC_T *p) 6909 { 6910 int error; 6911 off_t olen; 6912 6913 ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); 6914 /* 6915 * Attempt to allocate on a DS file. A return of ENOENT implies 6916 * there is no DS file to allocate on. 6917 */ 6918 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, 6919 NULL, NULL, NULL, NULL, &len, 0, NULL); 6920 if (error != ENOENT) 6921 return (error); 6922 6923 /* 6924 * Do the actual VOP_ALLOCATE(), looping so long as 6925 * progress is being made, to achieve completion. 6926 */ 6927 do { 6928 olen = len; 6929 error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred); 6930 if (error == 0 && len > 0 && olen > len) 6931 maybe_yield(); 6932 } while (error == 0 && len > 0 && olen > len); 6933 if (error == 0 && len > 0) 6934 error = NFSERR_IO; 6935 NFSEXITCODE(error); 6936 return (error); 6937 } 6938 6939 /* 6940 * Deallocate vnode op call. 6941 */ 6942 int 6943 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6944 NFSPROC_T *p) 6945 { 6946 int error; 6947 off_t olen; 6948 6949 ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp"); 6950 /* 6951 * Attempt to deallocate on a DS file. A return of ENOENT implies 6952 * there is no DS file to deallocate on. 6953 */ 6954 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL, 6955 NULL, NULL, NULL, NULL, &len, 0, NULL); 6956 if (error != ENOENT) 6957 return (error); 6958 6959 /* 6960 * Do the actual VOP_DEALLOCATE(), looping so long as 6961 * progress is being made, to achieve completion. 6962 */ 6963 do { 6964 olen = len; 6965 error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred); 6966 if (error == 0 && len > 0 && olen > len) 6967 maybe_yield(); 6968 } while (error == 0 && len > 0 && olen > len); 6969 if (error == 0 && len > 0) 6970 error = NFSERR_IO; 6971 NFSEXITCODE(error); 6972 return (error); 6973 } 6974 6975 /* 6976 * Get Extended Atribute vnode op into an mbuf list. 6977 */ 6978 int 6979 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, 6980 struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p, 6981 struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 6982 { 6983 struct iovec *iv; 6984 struct uio io, *uiop = &io; 6985 struct mbuf *m, *m2; 6986 int alen, error, len, tlen; 6987 size_t siz; 6988 6989 /* First, find out the size of the extended attribute. */ 6990 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6991 &siz, cred, p); 6992 if (error != 0) 6993 return (NFSERR_NOXATTR); 6994 if (siz > maxresp - NFS_MAXXDR) 6995 return (NFSERR_XATTR2BIG); 6996 len = siz; 6997 tlen = NFSM_RNDUP(len); 6998 if (tlen > 0) { 6999 /* 7000 * If cnt > MCLBYTES and the reply will not be saved, use 7001 * ext_pgs mbufs for TLS. 7002 * For NFSv4.0, we do not know for sure if the reply will 7003 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 7004 * Always use ext_pgs mbufs if ND_EXTPG is set. 7005 */ 7006 if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES && 7007 (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS && 7008 (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)) 7009 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen, 7010 maxextsiz, &m, &m2, &iv); 7011 else 7012 uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, 7013 &iv); 7014 uiop->uio_iov = iv; 7015 } else { 7016 uiop->uio_iovcnt = 0; 7017 uiop->uio_iov = iv = NULL; 7018 m = m2 = NULL; 7019 } 7020 uiop->uio_offset = 0; 7021 uiop->uio_resid = tlen; 7022 uiop->uio_rw = UIO_READ; 7023 uiop->uio_segflg = UIO_SYSSPACE; 7024 uiop->uio_td = p; 7025 #ifdef MAC 7026 error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, 7027 name); 7028 if (error != 0) 7029 goto out; 7030 #endif 7031 7032 if (tlen > 0) 7033 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 7034 NULL, cred, p); 7035 if (error != 0) 7036 goto out; 7037 if (uiop->uio_resid > 0) { 7038 alen = tlen; 7039 len = tlen - uiop->uio_resid; 7040 tlen = NFSM_RNDUP(len); 7041 if (alen != tlen) 7042 printf("nfsvno_getxattr: weird size read\n"); 7043 if (tlen == 0) { 7044 m_freem(m); 7045 m = m2 = NULL; 7046 } else if (alen != tlen || tlen != len) 7047 m2 = nfsrv_adj(m, alen - tlen, tlen - len); 7048 } 7049 *lenp = len; 7050 *mpp = m; 7051 *mpendp = m2; 7052 7053 out: 7054 if (error != 0) { 7055 if (m != NULL) 7056 m_freem(m); 7057 *lenp = 0; 7058 } 7059 free(iv, M_TEMP); 7060 NFSEXITCODE(error); 7061 return (error); 7062 } 7063 7064 /* 7065 * Set Extended attribute vnode op from an mbuf list. 7066 */ 7067 int 7068 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, 7069 char *cp, struct ucred *cred, struct thread *p) 7070 { 7071 struct iovec *iv; 7072 struct uio uio, *uiop = &uio; 7073 int cnt, error; 7074 7075 error = 0; 7076 #ifdef MAC 7077 error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, 7078 name); 7079 #endif 7080 if (error != 0) 7081 goto out; 7082 7083 uiop->uio_rw = UIO_WRITE; 7084 uiop->uio_segflg = UIO_SYSSPACE; 7085 uiop->uio_td = p; 7086 uiop->uio_offset = 0; 7087 uiop->uio_resid = len; 7088 if (len > 0) { 7089 error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); 7090 uiop->uio_iov = iv; 7091 uiop->uio_iovcnt = cnt; 7092 } else { 7093 uiop->uio_iov = iv = NULL; 7094 uiop->uio_iovcnt = 0; 7095 } 7096 if (error == 0) { 7097 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 7098 cred, p); 7099 if (error == 0) { 7100 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 7101 nfsvno_updateds(vp, cred, p); 7102 error = VOP_FSYNC(vp, MNT_WAIT, p); 7103 } 7104 free(iv, M_TEMP); 7105 } 7106 7107 out: 7108 NFSEXITCODE(error); 7109 return (error); 7110 } 7111 7112 /* 7113 * For a pNFS server, the DS file's ctime and 7114 * va_filerev (TimeMetadata and Change) needs to 7115 * be updated. This is a hack, but works by 7116 * flipping the S_ISGID bit in va_mode and then 7117 * flipping it back. 7118 * It does result in two MDS->DS RPCs, but creating 7119 * a custom RPC just to do this seems overkill, since 7120 * Setxattr/Rmxattr will not be done that frequently. 7121 * If it fails part way through, that is not too 7122 * serious, since the DS file is never executed. 7123 */ 7124 static void 7125 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p) 7126 { 7127 struct nfsvattr nva; 7128 int ret; 7129 u_short tmode; 7130 7131 ret = VOP_GETATTR(vp, &nva.na_vattr, cred); 7132 if (ret == 0) { 7133 tmode = nva.na_mode; 7134 NFSVNO_ATTRINIT(&nva); 7135 tmode ^= S_ISGID; 7136 NFSVNO_SETATTRVAL(&nva, mode, tmode); 7137 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 7138 NFSPROC_SETATTR, NULL, NULL, NULL, &nva, 7139 NULL, NULL, 0, NULL); 7140 if (ret == 0) { 7141 tmode ^= S_ISGID; 7142 NFSVNO_SETATTRVAL(&nva, mode, tmode); 7143 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 7144 NFSPROC_SETATTR, NULL, NULL, NULL, 7145 &nva, NULL, NULL, 0, NULL); 7146 } 7147 } 7148 } 7149 7150 /* 7151 * Remove Extended attribute vnode op. 7152 */ 7153 int 7154 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, 7155 struct ucred *cred, struct thread *p) 7156 { 7157 int error; 7158 7159 /* 7160 * Get rid of any delegations. I am not sure why this is required, 7161 * but RFC-8276 says so. 7162 */ 7163 error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); 7164 if (error != 0) 7165 goto out; 7166 #ifdef MAC 7167 error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, 7168 name); 7169 if (error != 0) 7170 goto out; 7171 #endif 7172 7173 error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); 7174 if (error == EOPNOTSUPP) 7175 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 7176 cred, p); 7177 if (error == 0) { 7178 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 7179 nfsvno_updateds(vp, cred, p); 7180 error = VOP_FSYNC(vp, MNT_WAIT, p); 7181 } 7182 out: 7183 NFSEXITCODE(error); 7184 return (error); 7185 } 7186 7187 /* 7188 * List Extended Atribute vnode op into an mbuf list. 7189 */ 7190 int 7191 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, 7192 struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) 7193 { 7194 struct iovec iv; 7195 struct uio io; 7196 int error; 7197 size_t siz; 7198 7199 *bufp = NULL; 7200 /* First, find out the size of the extended attribute. */ 7201 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, 7202 p); 7203 if (error != 0) 7204 return (NFSERR_NOXATTR); 7205 if (siz <= cookie) { 7206 *lenp = 0; 7207 *eofp = true; 7208 goto out; 7209 } 7210 if (siz > cookie + *lenp) { 7211 siz = cookie + *lenp; 7212 *eofp = false; 7213 } else 7214 *eofp = true; 7215 /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ 7216 if (siz > 10 * 1024 * 1024) { 7217 error = NFSERR_XATTR2BIG; 7218 goto out; 7219 } 7220 *bufp = malloc(siz, M_TEMP, M_WAITOK); 7221 iv.iov_base = *bufp; 7222 iv.iov_len = siz; 7223 io.uio_iovcnt = 1; 7224 io.uio_iov = &iv; 7225 io.uio_offset = 0; 7226 io.uio_resid = siz; 7227 io.uio_rw = UIO_READ; 7228 io.uio_segflg = UIO_SYSSPACE; 7229 io.uio_td = p; 7230 #ifdef MAC 7231 error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); 7232 if (error != 0) 7233 goto out; 7234 #endif 7235 7236 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, 7237 p); 7238 if (error != 0) 7239 goto out; 7240 if (io.uio_resid > 0) 7241 siz -= io.uio_resid; 7242 *lenp = siz; 7243 7244 out: 7245 if (error != 0) { 7246 free(*bufp, M_TEMP); 7247 *bufp = NULL; 7248 } 7249 NFSEXITCODE(error); 7250 return (error); 7251 } 7252 7253 /* 7254 * Trim trailing data off the mbuf list being built. 7255 */ 7256 void 7257 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos, 7258 int bextpg, int bextpgsiz) 7259 { 7260 vm_page_t pg; 7261 int fullpgsiz, i; 7262 7263 if (mb->m_next != NULL) { 7264 m_freem(mb->m_next); 7265 mb->m_next = NULL; 7266 } 7267 if ((mb->m_flags & M_EXTPG) != 0) { 7268 KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs, 7269 ("nfsm_trimtrailing: bextpg out of range")); 7270 KASSERT(bpos == (char *)(void *) 7271 PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz, 7272 ("nfsm_trimtrailing: bextpgsiz bad!")); 7273 7274 /* First, get rid of any pages after this position. */ 7275 for (i = mb->m_epg_npgs - 1; i > bextpg; i--) { 7276 pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]); 7277 vm_page_unwire_noq(pg); 7278 vm_page_free(pg); 7279 } 7280 mb->m_epg_npgs = bextpg + 1; 7281 if (bextpg == 0) 7282 fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off; 7283 else 7284 fullpgsiz = PAGE_SIZE; 7285 mb->m_epg_last_len = fullpgsiz - bextpgsiz; 7286 mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off); 7287 for (i = 1; i < mb->m_epg_npgs; i++) 7288 mb->m_len += m_epg_pagelen(mb, i, 0); 7289 nd->nd_bextpgsiz = bextpgsiz; 7290 nd->nd_bextpg = bextpg; 7291 } else 7292 mb->m_len = bpos - mtod(mb, char *); 7293 nd->nd_mb = mb; 7294 nd->nd_bpos = bpos; 7295 } 7296 7297 7298 /* 7299 * Check to see if a put file handle operation should test for 7300 * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR. 7301 * When Open is the next operation, NFSERR_WRONGSEC cannot be 7302 * replied for the Open cases that use a component. This can 7303 * be identified by the fact that the file handle's type is VDIR. 7304 */ 7305 bool 7306 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, __enum_uint8(vtype) vtyp) 7307 { 7308 7309 if ((nd->nd_flag & ND_NFSV4) == 0) 7310 return (true); 7311 7312 if ((nd->nd_flag & ND_LASTOP) != 0) 7313 return (false); 7314 7315 if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH || 7316 nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH || 7317 nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP || 7318 nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME) 7319 return (false); 7320 if (nextop == NFSV4OP_OPEN && vtyp == VDIR) 7321 return (false); 7322 return (true); 7323 } 7324 7325 /* 7326 * Check DSs marked no space. 7327 */ 7328 void 7329 nfsrv_checknospc(void) 7330 { 7331 struct statfs *tsf; 7332 struct nfsdevice *ds; 7333 struct vnode **dvpp, **tdvpp, *dvp; 7334 char *devid, *tdevid; 7335 int cnt, error = 0, i; 7336 7337 if (nfsrv_devidcnt <= 0) 7338 return; 7339 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 7340 devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK); 7341 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 7342 7343 /* Get an array of the dvps for the DSs. */ 7344 tdvpp = dvpp; 7345 tdevid = devid; 7346 i = 0; 7347 NFSDDSLOCK(); 7348 /* First, search for matches for same file system. */ 7349 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7350 if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) { 7351 if (++i > nfsrv_devidcnt) 7352 break; 7353 *tdvpp++ = ds->nfsdev_dvp; 7354 NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID); 7355 tdevid += NFSX_V4DEVICEID; 7356 } 7357 } 7358 NFSDDSUNLOCK(); 7359 7360 /* Do a VFS_STATFS() for each of the DSs and clear no space. */ 7361 cnt = i; 7362 tdvpp = dvpp; 7363 tdevid = devid; 7364 for (i = 0; i < cnt && error == 0; i++) { 7365 dvp = *tdvpp++; 7366 error = VFS_STATFS(dvp->v_mount, tsf); 7367 if (error == 0 && tsf->f_bavail > 0) { 7368 NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n"); 7369 nfsrv_marknospc(tdevid, false); 7370 } 7371 tdevid += NFSX_V4DEVICEID; 7372 } 7373 free(tsf, M_TEMP); 7374 free(dvpp, M_TEMP); 7375 free(devid, M_TEMP); 7376 } 7377 7378 /* 7379 * Initialize everything that needs to be initialized for a vnet. 7380 */ 7381 static void 7382 nfsrv_vnetinit(const void *unused __unused) 7383 { 7384 7385 nfsd_mntinit(); 7386 } 7387 VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY, 7388 nfsrv_vnetinit, NULL); 7389 7390 /* 7391 * Clean up everything that is in a vnet and needs to be 7392 * done when the jail is destroyed or the module unloaded. 7393 */ 7394 static void 7395 nfsrv_cleanup(const void *unused __unused) 7396 { 7397 int i; 7398 7399 NFSD_LOCK(); 7400 if (!NFSD_VNET(nfsrv_mntinited)) { 7401 NFSD_UNLOCK(); 7402 return; 7403 } 7404 NFSD_VNET(nfsrv_mntinited) = false; 7405 NFSD_UNLOCK(); 7406 7407 /* Clean out all NFSv4 state. */ 7408 nfsrv_throwawayallstate(curthread); 7409 7410 /* Clean the NFS server reply cache */ 7411 nfsrvd_cleancache(); 7412 7413 /* Clean out v4root exports. */ 7414 if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) { 7415 vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export); 7416 free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT); 7417 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 7418 } 7419 7420 /* Free up the krpc server pool. */ 7421 if (NFSD_VNET(nfsrvd_pool) != NULL) 7422 svcpool_destroy(NFSD_VNET(nfsrvd_pool)); 7423 7424 /* and get rid of the locks */ 7425 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 7426 mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx); 7427 mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx); 7428 } 7429 mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx); 7430 for (i = 0; i < nfsrv_sessionhashsize; i++) 7431 mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx); 7432 lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock); 7433 free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE); 7434 free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE); 7435 free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE); 7436 free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT); 7437 free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE); 7438 free(NFSD_VNET(nfssessionhash), M_NFSDSESSION); 7439 free(NFSD_VNET(nfsv4root_mnt), M_TEMP); 7440 NFSD_VNET(nfsv4root_mnt) = NULL; 7441 } 7442 VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY, 7443 nfsrv_cleanup, NULL); 7444 7445 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); 7446 7447 /* 7448 * Called once to initialize data structures... 7449 */ 7450 static int 7451 nfsd_modevent(module_t mod, int type, void *data) 7452 { 7453 int error = 0, i; 7454 static int loaded = 0; 7455 7456 switch (type) { 7457 case MOD_LOAD: 7458 if (loaded) 7459 goto out; 7460 newnfs_portinit(); 7461 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); 7462 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); 7463 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); 7464 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); 7465 #ifdef VV_DISABLEDELEG 7466 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; 7467 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; 7468 #endif 7469 nfsd_call_nfsd = nfssvc_nfsd; 7470 loaded = 1; 7471 break; 7472 7473 case MOD_UNLOAD: 7474 if (newnfs_numnfsd != 0) { 7475 error = EBUSY; 7476 break; 7477 } 7478 7479 #ifdef VV_DISABLEDELEG 7480 vn_deleg_ops.vndeleg_recall = NULL; 7481 vn_deleg_ops.vndeleg_disable = NULL; 7482 #endif 7483 nfsd_call_nfsd = NULL; 7484 mtx_destroy(&nfsrc_udpmtx); 7485 mtx_destroy(&nfs_v4root_mutex); 7486 mtx_destroy(&nfsrv_dontlistlock_mtx); 7487 mtx_destroy(&nfsrv_recalllock_mtx); 7488 if (nfslayouthash != NULL) { 7489 for (i = 0; i < nfsrv_layouthashsize; i++) 7490 mtx_destroy(&nfslayouthash[i].mtx); 7491 free(nfslayouthash, M_NFSDSESSION); 7492 } 7493 loaded = 0; 7494 break; 7495 default: 7496 error = EOPNOTSUPP; 7497 break; 7498 } 7499 7500 out: 7501 NFSEXITCODE(error); 7502 return (error); 7503 } 7504 static moduledata_t nfsd_mod = { 7505 "nfsd", 7506 nfsd_modevent, 7507 NULL, 7508 }; 7509 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); 7510 7511 /* So that loader and kldload(2) can find us, wherever we are.. */ 7512 MODULE_VERSION(nfsd, 1); 7513 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); 7514 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); 7515 MODULE_DEPEND(nfsd, krpc, 1, 1, 1); 7516 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); 7517