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, bool has_caseinsensitive) 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, NULL, has_caseinsensitive); 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_caseinsensitive, 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 has_caseinsensitive = false; 2953 clone_blksize = 0; 2954 if (nvp != NULL) { 2955 supports_nfsv4acls = 2956 nfs_supportsnfsv4acls(nvp); 2957 if (NFSISSET_ATTRBIT(&attrbits, 2958 NFSATTRBIT_XATTRSUPPORT)) { 2959 ret = VOP_GETEXTATTR(nvp, 2960 EXTATTR_NAMESPACE_USER, 2961 "xxx", NULL, &atsiz, 2962 nd->nd_cred, p); 2963 xattrsupp = ret != EOPNOTSUPP; 2964 } 2965 if (VOP_PATHCONF(nvp, 2966 _PC_HAS_HIDDENSYSTEM, &pathval) != 2967 0) 2968 pathval = 0; 2969 has_hiddensystem = pathval > 0; 2970 pathval = 0; 2971 if (NFSISSET_ATTRBIT(&attrbits, 2972 NFSATTRBIT_NAMEDATTR) && 2973 VOP_PATHCONF(nvp, _PC_HAS_NAMEDATTR, 2974 &pathval) != 0) 2975 pathval = 0; 2976 has_namedattr = pathval > 0; 2977 pathval = 0; 2978 if (VOP_PATHCONF(nvp, _PC_CLONE_BLKSIZE, 2979 &pathval) != 0) 2980 pathval = 0; 2981 clone_blksize = pathval; 2982 if (VOP_PATHCONF(nvp, 2983 _PC_CASE_INSENSITIVE, 2984 &pathval) != 0) 2985 pathval = 0; 2986 has_caseinsensitive = pathval > 0; 2987 NFSVOPUNLOCK(nvp); 2988 } else 2989 supports_nfsv4acls = 0; 2990 if (refp != NULL) { 2991 dirlen += nfsrv_putreferralattr(nd, 2992 &refbits, refp, 0, 2993 &nd->nd_repstat); 2994 if (nd->nd_repstat) { 2995 if (nvp != NULL) 2996 vrele(nvp); 2997 if (needs_unbusy != 0) 2998 vfs_unbusy(new_mp); 2999 break; 3000 } 3001 } else if (r) { 3002 dirlen += nfsvno_fillattr(nd, new_mp, 3003 nvp, nvap, &nfh, r, &rderrbits, 3004 nd->nd_cred, p, isdgram, 0, 3005 supports_nfsv4acls, at_root, 3006 mounted_on_fileno, xattrsupp, 3007 has_hiddensystem, has_namedattr, 3008 clone_blksize, has_caseinsensitive); 3009 } else { 3010 dirlen += nfsvno_fillattr(nd, new_mp, 3011 nvp, nvap, &nfh, r, &attrbits, 3012 nd->nd_cred, p, isdgram, 0, 3013 supports_nfsv4acls, at_root, 3014 mounted_on_fileno, xattrsupp, 3015 has_hiddensystem, has_namedattr, 3016 clone_blksize, has_caseinsensitive); 3017 } 3018 if (nvp != NULL) 3019 vrele(nvp); 3020 dirlen += (3 * NFSX_UNSIGNED); 3021 } 3022 if (needs_unbusy != 0) 3023 vfs_unbusy(new_mp); 3024 if (dirlen <= cnt) 3025 entrycnt++; 3026 } 3027 invalid: 3028 cpos += dp->d_reclen; 3029 dp = (struct dirent *)cpos; 3030 cookiep++; 3031 ncookies--; 3032 } 3033 vrele(vp); 3034 vfs_unbusy(mp); 3035 3036 /* 3037 * If dirlen > cnt, we must strip off the last entry. If that 3038 * results in an empty reply, report NFSERR_TOOSMALL. 3039 */ 3040 if (dirlen > cnt || nd->nd_repstat) { 3041 if (!nd->nd_repstat && entrycnt == 0) 3042 nd->nd_repstat = NFSERR_TOOSMALL; 3043 if (nd->nd_repstat) { 3044 nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0); 3045 if (nd->nd_flag & ND_NFSV3) 3046 nfsrv_postopattr(nd, getret, &at); 3047 } else 3048 nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1); 3049 eofflag = 0; 3050 } else if (cpos < cend) 3051 eofflag = 0; 3052 if (!nd->nd_repstat) { 3053 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3054 *tl++ = newnfs_false; 3055 if (eofflag) 3056 *tl = newnfs_true; 3057 else 3058 *tl = newnfs_false; 3059 } 3060 free(cookies, M_TEMP); 3061 free(rbuf, M_TEMP); 3062 3063 out: 3064 NFSEXITCODE2(0, nd); 3065 return (0); 3066 nfsmout: 3067 vput(vp); 3068 NFSEXITCODE2(error, nd); 3069 return (error); 3070 } 3071 3072 /* 3073 * Get the settable attributes out of the mbuf list. 3074 * (Return 0 or EBADRPC) 3075 */ 3076 int 3077 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 3078 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 3079 { 3080 u_int32_t *tl; 3081 struct nfsv2_sattr *sp; 3082 int error = 0, toclient = 0; 3083 3084 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 3085 case ND_NFSV2: 3086 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 3087 /* 3088 * Some old clients didn't fill in the high order 16bits. 3089 * --> check the low order 2 bytes for 0xffff 3090 */ 3091 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 3092 nvap->na_mode = nfstov_mode(sp->sa_mode); 3093 if (sp->sa_uid != newnfs_xdrneg1) 3094 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 3095 if (sp->sa_gid != newnfs_xdrneg1) 3096 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 3097 if (sp->sa_size != newnfs_xdrneg1) 3098 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 3099 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 3100 #ifdef notyet 3101 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 3102 #else 3103 nvap->na_atime.tv_sec = 3104 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 3105 nvap->na_atime.tv_nsec = 0; 3106 #endif 3107 } 3108 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 3109 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 3110 break; 3111 case ND_NFSV3: 3112 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3113 if (*tl == newnfs_true) { 3114 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3115 nvap->na_mode = nfstov_mode(*tl); 3116 } 3117 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3118 if (*tl == newnfs_true) { 3119 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3120 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 3121 } 3122 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3123 if (*tl == newnfs_true) { 3124 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3125 nvap->na_gid = fxdr_unsigned(gid_t, *tl); 3126 } 3127 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3128 if (*tl == newnfs_true) { 3129 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3130 nvap->na_size = fxdr_hyper(tl); 3131 } 3132 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3133 switch (fxdr_unsigned(int, *tl)) { 3134 case NFSV3SATTRTIME_TOCLIENT: 3135 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3136 fxdr_nfsv3time(tl, &nvap->na_atime); 3137 toclient = 1; 3138 break; 3139 case NFSV3SATTRTIME_TOSERVER: 3140 vfs_timestamp(&nvap->na_atime); 3141 nvap->na_vaflags |= VA_UTIMES_NULL; 3142 break; 3143 } 3144 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3145 switch (fxdr_unsigned(int, *tl)) { 3146 case NFSV3SATTRTIME_TOCLIENT: 3147 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 3148 fxdr_nfsv3time(tl, &nvap->na_mtime); 3149 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3150 break; 3151 case NFSV3SATTRTIME_TOSERVER: 3152 vfs_timestamp(&nvap->na_mtime); 3153 if (!toclient) 3154 nvap->na_vaflags |= VA_UTIMES_NULL; 3155 break; 3156 } 3157 break; 3158 case ND_NFSV4: 3159 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p); 3160 } 3161 nfsmout: 3162 NFSEXITCODE2(error, nd); 3163 return (error); 3164 } 3165 3166 /* 3167 * Handle the setable attributes for V4. 3168 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 3169 */ 3170 int 3171 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap, 3172 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 3173 { 3174 u_int32_t *tl; 3175 int attrsum = 0; 3176 int i, j; 3177 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 3178 int moderet, toclient = 0; 3179 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 3180 uid_t uid; 3181 gid_t gid; 3182 u_short mode, mask; /* Same type as va_mode. */ 3183 struct vattr va; 3184 3185 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 3186 if (error) 3187 goto nfsmout; 3188 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3189 attrsize = fxdr_unsigned(int, *tl); 3190 3191 /* 3192 * Loop around getting the setable attributes. If an unsupported 3193 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 3194 * Once nd_repstat != 0, do not set the attribute value, but keep 3195 * parsing the attribute(s). 3196 */ 3197 if (retnotsup) { 3198 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3199 bitpos = NFSATTRBIT_MAX; 3200 } else { 3201 bitpos = 0; 3202 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ARCHIVE) || 3203 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_HIDDEN) || 3204 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SYSTEM)) 3205 nvap->na_flags = 0; 3206 } 3207 moderet = 0; 3208 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 3209 if (attrsum > attrsize) { 3210 error = NFSERR_BADXDR; 3211 goto nfsmout; 3212 } 3213 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 3214 switch (bitpos) { 3215 case NFSATTRBIT_SIZE: 3216 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 3217 if (!nd->nd_repstat) { 3218 if (vp != NULL && vp->v_type != VREG) 3219 nd->nd_repstat = (vp->v_type == VDIR) ? 3220 NFSERR_ISDIR : NFSERR_INVAL; 3221 else 3222 nvap->na_size = fxdr_hyper(tl); 3223 } 3224 attrsum += NFSX_HYPER; 3225 break; 3226 case NFSATTRBIT_ACL: 3227 error = nfsrv_dissectacl(nd, aclp, true, &aceerr, 3228 &aclsize, p); 3229 if (error) 3230 goto nfsmout; 3231 if (aceerr && !nd->nd_repstat) 3232 nd->nd_repstat = aceerr; 3233 attrsum += aclsize; 3234 break; 3235 case NFSATTRBIT_ARCHIVE: 3236 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); 3237 if (nd->nd_repstat == 0) { 3238 if (*tl == newnfs_true) 3239 nvap->na_flags |= UF_ARCHIVE; 3240 } 3241 attrsum += NFSX_UNSIGNED; 3242 break; 3243 case NFSATTRBIT_HIDDEN: 3244 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); 3245 if (nd->nd_repstat == 0) { 3246 if (*tl == newnfs_true) 3247 nvap->na_flags |= UF_HIDDEN; 3248 } 3249 attrsum += NFSX_UNSIGNED; 3250 break; 3251 case NFSATTRBIT_MIMETYPE: 3252 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3253 i = fxdr_unsigned(int, *tl); 3254 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 3255 if (error) 3256 goto nfsmout; 3257 if (!nd->nd_repstat) 3258 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3259 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 3260 break; 3261 case NFSATTRBIT_MODE: 3262 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */ 3263 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3264 if (!nd->nd_repstat) 3265 nvap->na_mode = nfstov_mode(*tl); 3266 attrsum += NFSX_UNSIGNED; 3267 break; 3268 case NFSATTRBIT_OWNER: 3269 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3270 j = fxdr_unsigned(int, *tl); 3271 if (j < 0) { 3272 error = NFSERR_BADXDR; 3273 goto nfsmout; 3274 } 3275 if (j > NFSV4_SMALLSTR) 3276 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3277 else 3278 cp = namestr; 3279 error = nfsrv_mtostr(nd, cp, j); 3280 if (error) { 3281 if (j > NFSV4_SMALLSTR) 3282 free(cp, M_NFSSTRING); 3283 goto nfsmout; 3284 } 3285 if (!nd->nd_repstat) { 3286 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, 3287 &uid); 3288 if (!nd->nd_repstat) 3289 nvap->na_uid = uid; 3290 } 3291 if (j > NFSV4_SMALLSTR) 3292 free(cp, M_NFSSTRING); 3293 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3294 break; 3295 case NFSATTRBIT_OWNERGROUP: 3296 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3297 j = fxdr_unsigned(int, *tl); 3298 if (j < 0) { 3299 error = NFSERR_BADXDR; 3300 goto nfsmout; 3301 } 3302 if (j > NFSV4_SMALLSTR) 3303 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 3304 else 3305 cp = namestr; 3306 error = nfsrv_mtostr(nd, cp, j); 3307 if (error) { 3308 if (j > NFSV4_SMALLSTR) 3309 free(cp, M_NFSSTRING); 3310 goto nfsmout; 3311 } 3312 if (!nd->nd_repstat) { 3313 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, 3314 &gid); 3315 if (!nd->nd_repstat) 3316 nvap->na_gid = gid; 3317 } 3318 if (j > NFSV4_SMALLSTR) 3319 free(cp, M_NFSSTRING); 3320 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 3321 break; 3322 case NFSATTRBIT_SYSTEM: 3323 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); 3324 if (nd->nd_repstat == 0) { 3325 if (*tl == newnfs_true) 3326 nvap->na_flags |= UF_SYSTEM; 3327 } 3328 attrsum += NFSX_UNSIGNED; 3329 break; 3330 case NFSATTRBIT_TIMEACCESSSET: 3331 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3332 attrsum += NFSX_UNSIGNED; 3333 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3334 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3335 if (!nd->nd_repstat) 3336 fxdr_nfsv4time(tl, &nvap->na_atime); 3337 toclient = 1; 3338 attrsum += NFSX_V4TIME; 3339 } else if (!nd->nd_repstat) { 3340 vfs_timestamp(&nvap->na_atime); 3341 nvap->na_vaflags |= VA_UTIMES_NULL; 3342 } 3343 break; 3344 case NFSATTRBIT_TIMEBACKUP: 3345 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3346 if (!nd->nd_repstat) 3347 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3348 attrsum += NFSX_V4TIME; 3349 break; 3350 case NFSATTRBIT_TIMECREATE: 3351 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3352 if (!nd->nd_repstat) 3353 fxdr_nfsv4time(tl, &nvap->na_btime); 3354 attrsum += NFSX_V4TIME; 3355 break; 3356 case NFSATTRBIT_TIMEMODIFYSET: 3357 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 3358 attrsum += NFSX_UNSIGNED; 3359 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 3360 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 3361 if (!nd->nd_repstat) 3362 fxdr_nfsv4time(tl, &nvap->na_mtime); 3363 nvap->na_vaflags &= ~VA_UTIMES_NULL; 3364 attrsum += NFSX_V4TIME; 3365 } else if (!nd->nd_repstat) { 3366 vfs_timestamp(&nvap->na_mtime); 3367 if (!toclient) 3368 nvap->na_vaflags |= VA_UTIMES_NULL; 3369 } 3370 break; 3371 case NFSATTRBIT_MODESETMASKED: 3372 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3373 mode = fxdr_unsigned(u_short, *tl++); 3374 mask = fxdr_unsigned(u_short, *tl); 3375 /* 3376 * vp == NULL implies an Open/Create operation. 3377 * This attribute can only be used for Setattr and 3378 * only for NFSv4.1 or higher. 3379 * If moderet != 0, a mode attribute has also been 3380 * specified and this attribute cannot be done in the 3381 * same Setattr operation. 3382 */ 3383 if (!nd->nd_repstat) { 3384 if ((nd->nd_flag & ND_NFSV41) == 0) 3385 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3386 else if ((mode & ~07777) != 0 || 3387 (mask & ~07777) != 0 || vp == NULL) 3388 nd->nd_repstat = NFSERR_INVAL; 3389 else if (moderet == 0) 3390 moderet = VOP_GETATTR(vp, &va, 3391 nd->nd_cred); 3392 if (moderet == 0) 3393 nvap->na_mode = (mode & mask) | 3394 (va.va_mode & ~mask); 3395 else 3396 nd->nd_repstat = moderet; 3397 } 3398 attrsum += 2 * NFSX_UNSIGNED; 3399 break; 3400 case NFSATTRBIT_MODEUMASK: 3401 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 3402 mode = fxdr_unsigned(u_short, *tl++); 3403 mask = fxdr_unsigned(u_short, *tl); 3404 /* 3405 * If moderet != 0, mode has already been done. 3406 * If vp != NULL, this is not a file object creation. 3407 */ 3408 if (!nd->nd_repstat) { 3409 if ((nd->nd_flag & ND_NFSV42) == 0) 3410 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3411 else if ((mask & ~0777) != 0 || vp != NULL || 3412 moderet != 0) 3413 nd->nd_repstat = NFSERR_INVAL; 3414 else 3415 nvap->na_mode = (mode & ~mask); 3416 } 3417 attrsum += 2 * NFSX_UNSIGNED; 3418 break; 3419 default: 3420 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 3421 /* 3422 * set bitpos so we drop out of the loop. 3423 */ 3424 bitpos = NFSATTRBIT_MAX; 3425 break; 3426 } 3427 } 3428 3429 /* 3430 * some clients pad the attrlist, so we need to skip over the 3431 * padding. This also skips over unparsed non-supported attributes. 3432 */ 3433 if (attrsum > attrsize) { 3434 error = NFSERR_BADXDR; 3435 } else { 3436 attrsize = NFSM_RNDUP(attrsize); 3437 if (attrsum < attrsize) 3438 error = nfsm_advance(nd, attrsize - attrsum, -1); 3439 } 3440 nfsmout: 3441 NFSEXITCODE2(error, nd); 3442 return (error); 3443 } 3444 3445 /* 3446 * Check/setup export credentials. 3447 */ 3448 int 3449 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 3450 struct ucred *credanon, bool testsec) 3451 { 3452 int error; 3453 3454 /* 3455 * Check/setup credentials. 3456 */ 3457 if (nd->nd_flag & ND_GSS) 3458 exp->nes_exflag &= ~MNT_EXPORTANON; 3459 3460 /* 3461 * Check to see if the operation is allowed for this security flavor. 3462 */ 3463 error = 0; 3464 if (testsec) { 3465 error = nfsvno_testexp(nd, exp); 3466 if (error != 0) 3467 goto out; 3468 } 3469 3470 /* 3471 * Check to see if the file system is exported V4 only. 3472 */ 3473 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 3474 error = NFSERR_PROGNOTV4; 3475 goto out; 3476 } 3477 3478 /* 3479 * Now, map the user credentials. 3480 * (Note that ND_AUTHNONE will only be set for an NFSv3 3481 * Fsinfo RPC. If set for anything else, this code might need 3482 * to change.) 3483 */ 3484 if (NFSVNO_EXPORTED(exp)) { 3485 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) || 3486 NFSVNO_EXPORTANON(exp) || 3487 (nd->nd_flag & ND_AUTHNONE) != 0) { 3488 nd->nd_cred->cr_uid = credanon->cr_uid; 3489 nd->nd_cred->cr_gid = credanon->cr_gid; 3490 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 3491 credanon->cr_groups); 3492 } else if ((nd->nd_flag & ND_GSS) == 0) { 3493 /* 3494 * If using AUTH_SYS, call nfsrv_getgrpscred() to see 3495 * if there is a replacement credential with a group 3496 * list set up by "nfsuserd -manage-gids". 3497 * If there is no replacement, nfsrv_getgrpscred() 3498 * simply returns its argument. 3499 */ 3500 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred); 3501 } 3502 } 3503 3504 out: 3505 NFSEXITCODE2(error, nd); 3506 return (error); 3507 } 3508 3509 /* 3510 * Check exports. 3511 */ 3512 int 3513 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 3514 struct ucred **credp) 3515 { 3516 int error; 3517 3518 error = 0; 3519 *credp = NULL; 3520 MNT_ILOCK(mp); 3521 if (mp->mnt_exjail == NULL || 3522 mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison) 3523 error = EACCES; 3524 MNT_IUNLOCK(mp); 3525 if (error == 0) 3526 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3527 &exp->nes_numsecflavor, exp->nes_secflavors); 3528 if (error) { 3529 if (NFSD_VNET(nfs_rootfhset)) { 3530 exp->nes_exflag = 0; 3531 exp->nes_numsecflavor = 0; 3532 error = 0; 3533 } 3534 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3535 MAXSECFLAVORS) { 3536 printf("nfsvno_checkexp: numsecflavors out of range\n"); 3537 exp->nes_numsecflavor = 0; 3538 error = EACCES; 3539 } 3540 NFSEXITCODE(error); 3541 return (error); 3542 } 3543 3544 /* 3545 * Get a vnode for a file handle and export stuff. 3546 */ 3547 int 3548 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 3549 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 3550 struct ucred **credp) 3551 { 3552 int error; 3553 3554 *credp = NULL; 3555 exp->nes_numsecflavor = 0; 3556 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp); 3557 if (error != 0) 3558 /* Make sure the server replies ESTALE to the client. */ 3559 error = ESTALE; 3560 if (nam && !error) { 3561 MNT_ILOCK(mp); 3562 if (mp->mnt_exjail == NULL || 3563 mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison) 3564 error = EACCES; 3565 MNT_IUNLOCK(mp); 3566 if (error == 0) 3567 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 3568 &exp->nes_numsecflavor, exp->nes_secflavors); 3569 if (error) { 3570 if (NFSD_VNET(nfs_rootfhset)) { 3571 exp->nes_exflag = 0; 3572 exp->nes_numsecflavor = 0; 3573 error = 0; 3574 } else { 3575 vput(*vpp); 3576 } 3577 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor > 3578 MAXSECFLAVORS) { 3579 printf("nfsvno_fhtovp: numsecflavors out of range\n"); 3580 exp->nes_numsecflavor = 0; 3581 error = EACCES; 3582 vput(*vpp); 3583 } 3584 } 3585 NFSEXITCODE(error); 3586 return (error); 3587 } 3588 3589 /* 3590 * nfsd_fhtovp() - convert a fh to a vnode ptr 3591 * - look up fsid in mount list (if not found ret error) 3592 * - get vp and export rights by calling nfsvno_fhtovp() 3593 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 3594 * for AUTH_SYS 3595 * - if mpp != NULL, return the mount point so that it can 3596 * be used for vn_finished_write() by the caller 3597 */ 3598 void 3599 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 3600 struct vnode **vpp, struct nfsexstuff *exp, 3601 struct mount **mpp, int startwrite, int nextop) 3602 { 3603 struct mount *mp, *mpw; 3604 struct ucred *credanon; 3605 fhandle_t *fhp; 3606 int error; 3607 3608 if (mpp != NULL) 3609 *mpp = NULL; 3610 *vpp = NULL; 3611 fhp = (fhandle_t *)nfp->nfsrvfh_data; 3612 mp = vfs_busyfs(&fhp->fh_fsid); 3613 if (mp == NULL) { 3614 nd->nd_repstat = ESTALE; 3615 goto out; 3616 } 3617 3618 if (startwrite) { 3619 mpw = mp; 3620 error = vn_start_write(NULL, &mpw, V_WAIT); 3621 if (error != 0) { 3622 mpw = NULL; 3623 vfs_unbusy(mp); 3624 nd->nd_repstat = ESTALE; 3625 goto out; 3626 } 3627 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp))) 3628 lktype = LK_EXCLUSIVE; 3629 } else 3630 mpw = NULL; 3631 3632 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 3633 &credanon); 3634 vfs_unbusy(mp); 3635 3636 if (nd->nd_repstat == 0 && 3637 nfp->nfsrvfh_len >= NFSX_MYFH + NFSX_V4NAMEDDIRFH && 3638 nfp->nfsrvfh_len <= NFSX_MYFH + NFSX_V4NAMEDATTRFH) { 3639 if (nfp->nfsrvfh_len == NFSX_MYFH + NFSX_V4NAMEDDIRFH) 3640 vn_irflag_set_cond(*vpp, VIRF_NAMEDDIR); 3641 else 3642 vn_irflag_set_cond(*vpp, VIRF_NAMEDATTR); 3643 } 3644 3645 /* 3646 * For NFSv4 without a pseudo root fs, unexported file handles 3647 * can be returned, so that Lookup works everywhere. 3648 */ 3649 if (!nd->nd_repstat && exp->nes_exflag == 0 && 3650 !(nd->nd_flag & ND_NFSV4)) { 3651 vput(*vpp); 3652 *vpp = NULL; 3653 nd->nd_repstat = EACCES; 3654 } 3655 3656 /* 3657 * Personally, I've never seen any point in requiring a 3658 * reserved port#, since only in the rare case where the 3659 * clients are all boxes with secure system privileges, 3660 * does it provide any enhanced security, but... some people 3661 * believe it to be useful and keep putting this code back in. 3662 * (There is also some "security checker" out there that 3663 * complains if the nfs server doesn't enforce this.) 3664 * However, note the following: 3665 * RFC3530 (NFSv4) specifies that a reserved port# not be 3666 * required. 3667 * RFC2623 recommends that, if a reserved port# is checked for, 3668 * that there be a way to turn that off--> ifdef'd. 3669 */ 3670 #ifdef NFS_REQRSVPORT 3671 if (!nd->nd_repstat) { 3672 struct sockaddr_in *saddr; 3673 struct sockaddr_in6 *saddr6; 3674 3675 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 3676 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 3677 if (!(nd->nd_flag & ND_NFSV4) && 3678 ((saddr->sin_family == AF_INET && 3679 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 3680 (saddr6->sin6_family == AF_INET6 && 3681 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 3682 vput(*vpp); 3683 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 3684 } 3685 } 3686 #endif /* NFS_REQRSVPORT */ 3687 3688 /* 3689 * Check/setup credentials. 3690 */ 3691 if (!nd->nd_repstat) { 3692 nd->nd_saveduid = nd->nd_cred->cr_uid; 3693 nd->nd_repstat = nfsd_excred(nd, exp, credanon, 3694 nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type)); 3695 if (nd->nd_repstat) 3696 vput(*vpp); 3697 } 3698 if (credanon != NULL) 3699 crfree(credanon); 3700 if (nd->nd_repstat) { 3701 vn_finished_write(mpw); 3702 *vpp = NULL; 3703 } else if (mpp != NULL) { 3704 *mpp = mpw; 3705 } 3706 3707 out: 3708 NFSEXITCODE2(0, nd); 3709 } 3710 3711 /* 3712 * glue for fp. 3713 */ 3714 static int 3715 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 3716 { 3717 struct filedesc *fdp; 3718 struct file *fp; 3719 int error = 0; 3720 3721 fdp = p->td_proc->p_fd; 3722 if (fd < 0 || fd >= fdp->fd_nfiles || 3723 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) { 3724 error = EBADF; 3725 goto out; 3726 } 3727 *fpp = fp; 3728 3729 out: 3730 NFSEXITCODE(error); 3731 return (error); 3732 } 3733 3734 /* 3735 * Called from nfssvc() to update the exports list. Just call 3736 * vfs_export(). This has to be done, since the v4 root fake fs isn't 3737 * in the mount list. 3738 */ 3739 int 3740 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 3741 { 3742 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 3743 int error = 0; 3744 struct nameidata nd; 3745 fhandle_t fh; 3746 3747 error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export, false); 3748 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 3749 NFSD_VNET(nfs_rootfhset) = 0; 3750 else if (error == 0) { 3751 if (nfsexargp->fspec == NULL) { 3752 error = EPERM; 3753 goto out; 3754 } 3755 /* 3756 * If fspec != NULL, this is the v4root path. 3757 */ 3758 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec); 3759 if ((error = namei(&nd)) != 0) 3760 goto out; 3761 NDFREE_PNBUF(&nd); 3762 error = nfsvno_getfh(nd.ni_vp, &fh, p); 3763 vrele(nd.ni_vp); 3764 if (!error) { 3765 NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH; 3766 NFSBCOPY((caddr_t)&fh, 3767 NFSD_VNET(nfs_rootfh).nfsrvfh_data, 3768 sizeof (fhandle_t)); 3769 NFSD_VNET(nfs_rootfhset) = 1; 3770 } 3771 } 3772 3773 out: 3774 NFSEXITCODE(error); 3775 return (error); 3776 } 3777 3778 /* 3779 * This function needs to test to see if the system is near its limit 3780 * for memory allocation via malloc() or mget() and return True iff 3781 * either of these resources are near their limit. 3782 * XXX (For now, this is just a stub.) 3783 */ 3784 int nfsrv_testmalloclimit = 0; 3785 int 3786 nfsrv_mallocmget_limit(void) 3787 { 3788 static int printmesg = 0; 3789 static int testval = 1; 3790 3791 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 3792 if ((printmesg++ % 100) == 0) 3793 printf("nfsd: malloc/mget near limit\n"); 3794 return (1); 3795 } 3796 return (0); 3797 } 3798 3799 /* 3800 * BSD specific initialization of a mount point. 3801 */ 3802 void 3803 nfsd_mntinit(void) 3804 { 3805 3806 NFSD_LOCK(); 3807 if (NFSD_VNET(nfsrv_mntinited)) { 3808 NFSD_UNLOCK(); 3809 return; 3810 } 3811 NFSD_VNET(nfsrv_mntinited) = true; 3812 nfsrvd_init(0); 3813 NFSD_UNLOCK(); 3814 3815 NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP, 3816 M_WAITOK | M_ZERO); 3817 NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 3818 mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF); 3819 lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0); 3820 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist); 3821 TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist); 3822 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 3823 TAILQ_INIT(&NFSD_VNET(nfsv4root_opt)); 3824 TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt)); 3825 NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt); 3826 NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt); 3827 NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0; 3828 NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0; 3829 callout_init(&NFSD_VNET(nfsd_callout), 1); 3830 3831 nfsrvd_initcache(); 3832 nfsd_init(); 3833 } 3834 3835 static void 3836 nfsd_timer(void *arg) 3837 { 3838 struct vnet *vnetp; 3839 3840 vnetp = (struct vnet *)arg; 3841 NFSD_CURVNET_SET_QUIET(vnetp); 3842 nfsrv_servertimer(vnetp); 3843 callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer, 3844 arg, 0); 3845 NFSD_CURVNET_RESTORE(); 3846 } 3847 3848 /* 3849 * Get a vnode for a file handle, without checking exports, etc. 3850 */ 3851 struct vnode * 3852 nfsvno_getvp(fhandle_t *fhp) 3853 { 3854 struct mount *mp; 3855 struct vnode *vp; 3856 int error; 3857 3858 mp = vfs_busyfs(&fhp->fh_fsid); 3859 if (mp == NULL) 3860 return (NULL); 3861 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 3862 vfs_unbusy(mp); 3863 if (error) 3864 return (NULL); 3865 return (vp); 3866 } 3867 3868 /* 3869 * Do a local VOP_ADVLOCK(). 3870 */ 3871 int 3872 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 3873 u_int64_t end, struct thread *td) 3874 { 3875 int error = 0; 3876 struct flock fl; 3877 u_int64_t tlen; 3878 3879 if (nfsrv_dolocallocks == 0) 3880 goto out; 3881 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked"); 3882 3883 fl.l_whence = SEEK_SET; 3884 fl.l_type = ftype; 3885 fl.l_start = (off_t)first; 3886 if (end == NFS64BITSSET) { 3887 fl.l_len = 0; 3888 } else { 3889 tlen = end - first; 3890 fl.l_len = (off_t)tlen; 3891 } 3892 /* 3893 * For FreeBSD8, the l_pid and l_sysid must be set to the same 3894 * values for all calls, so that all locks will be held by the 3895 * nfsd server. (The nfsd server handles conflicts between the 3896 * various clients.) 3897 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 3898 * bytes, so it can't be put in l_sysid. 3899 */ 3900 if (nfsv4_sysid == 0) 3901 nfsv4_sysid = nlm_acquire_next_sysid(); 3902 fl.l_pid = (pid_t)0; 3903 fl.l_sysid = (int)nfsv4_sysid; 3904 3905 if (ftype == F_UNLCK) 3906 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 3907 (F_POSIX | F_REMOTE)); 3908 else 3909 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 3910 (F_POSIX | F_REMOTE)); 3911 3912 out: 3913 NFSEXITCODE(error); 3914 return (error); 3915 } 3916 3917 /* 3918 * Check the nfsv4 root exports. 3919 */ 3920 int 3921 nfsvno_v4rootexport(struct nfsrv_descript *nd) 3922 { 3923 struct ucred *credanon; 3924 int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i; 3925 uint64_t exflags; 3926 3927 error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags, 3928 &credanon, &numsecflavor, secflavors); 3929 if (error) { 3930 error = NFSERR_PROGUNAVAIL; 3931 goto out; 3932 } 3933 if (credanon != NULL) 3934 crfree(credanon); 3935 for (i = 0; i < numsecflavor; i++) { 3936 if (secflavors[i] == AUTH_SYS) 3937 nd->nd_flag |= ND_EXAUTHSYS; 3938 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3939 nd->nd_flag |= ND_EXGSS; 3940 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3941 nd->nd_flag |= ND_EXGSSINTEGRITY; 3942 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3943 nd->nd_flag |= ND_EXGSSPRIVACY; 3944 } 3945 3946 /* And set ND_EXxx flags for TLS. */ 3947 if ((exflags & MNT_EXTLS) != 0) { 3948 nd->nd_flag |= ND_EXTLS; 3949 if ((exflags & MNT_EXTLSCERT) != 0) 3950 nd->nd_flag |= ND_EXTLSCERT; 3951 if ((exflags & MNT_EXTLSCERTUSER) != 0) 3952 nd->nd_flag |= ND_EXTLSCERTUSER; 3953 } 3954 3955 out: 3956 NFSEXITCODE(error); 3957 return (error); 3958 } 3959 3960 /* 3961 * Nfs server pseudo system call for the nfsd's 3962 */ 3963 /* 3964 * MPSAFE 3965 */ 3966 static int 3967 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3968 { 3969 struct file *fp; 3970 struct nfsd_addsock_args sockarg; 3971 struct nfsd_nfsd_args nfsdarg; 3972 struct nfsd_nfsd_oargs onfsdarg; 3973 struct nfsd_pnfsd_args pnfsdarg; 3974 struct vnode *vp, *nvp, *curdvp; 3975 struct pnfsdsfile *pf; 3976 struct nfsdevice *ds, *fds; 3977 cap_rights_t rights; 3978 int buflen, error, ret; 3979 char *buf, *cp, *cp2, *cp3; 3980 char fname[PNFS_FILENAME_LEN + 1]; 3981 3982 NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td)); 3983 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3984 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3985 if (error) 3986 goto out; 3987 /* 3988 * Since we don't know what rights might be required, 3989 * pretend that we need them all. It is better to be too 3990 * careful than too reckless. 3991 */ 3992 error = fget(td, sockarg.sock, 3993 cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp); 3994 if (error != 0) 3995 goto out; 3996 if (fp->f_type != DTYPE_SOCKET) { 3997 fdrop(fp, td); 3998 error = EPERM; 3999 goto out; 4000 } 4001 error = nfsrvd_addsock(fp); 4002 fdrop(fp, td); 4003 } else if (uap->flag & NFSSVC_NFSDNFSD) { 4004 if (uap->argp == NULL) { 4005 error = EINVAL; 4006 goto out; 4007 } 4008 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) { 4009 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg)); 4010 if (error == 0) { 4011 nfsdarg.principal = onfsdarg.principal; 4012 nfsdarg.minthreads = onfsdarg.minthreads; 4013 nfsdarg.maxthreads = onfsdarg.maxthreads; 4014 nfsdarg.version = 1; 4015 nfsdarg.addr = NULL; 4016 nfsdarg.addrlen = 0; 4017 nfsdarg.dnshost = NULL; 4018 nfsdarg.dnshostlen = 0; 4019 nfsdarg.dspath = NULL; 4020 nfsdarg.dspathlen = 0; 4021 nfsdarg.mdspath = NULL; 4022 nfsdarg.mdspathlen = 0; 4023 nfsdarg.mirrorcnt = 1; 4024 } 4025 } else 4026 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg)); 4027 if (error) 4028 goto out; 4029 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 && 4030 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 && 4031 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 && 4032 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 && 4033 nfsdarg.mirrorcnt >= 1 && 4034 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS && 4035 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL && 4036 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) { 4037 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d" 4038 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen, 4039 nfsdarg.dspathlen, nfsdarg.dnshostlen, 4040 nfsdarg.mdspathlen, nfsdarg.mirrorcnt); 4041 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK); 4042 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen); 4043 if (error != 0) { 4044 free(cp, M_TEMP); 4045 goto out; 4046 } 4047 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */ 4048 nfsdarg.addr = cp; 4049 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK); 4050 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen); 4051 if (error != 0) { 4052 free(nfsdarg.addr, M_TEMP); 4053 free(cp, M_TEMP); 4054 goto out; 4055 } 4056 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */ 4057 nfsdarg.dnshost = cp; 4058 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK); 4059 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen); 4060 if (error != 0) { 4061 free(nfsdarg.addr, M_TEMP); 4062 free(nfsdarg.dnshost, M_TEMP); 4063 free(cp, M_TEMP); 4064 goto out; 4065 } 4066 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */ 4067 nfsdarg.dspath = cp; 4068 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK); 4069 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen); 4070 if (error != 0) { 4071 free(nfsdarg.addr, M_TEMP); 4072 free(nfsdarg.dnshost, M_TEMP); 4073 free(nfsdarg.dspath, M_TEMP); 4074 free(cp, M_TEMP); 4075 goto out; 4076 } 4077 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */ 4078 nfsdarg.mdspath = cp; 4079 } else { 4080 nfsdarg.addr = NULL; 4081 nfsdarg.addrlen = 0; 4082 nfsdarg.dnshost = NULL; 4083 nfsdarg.dnshostlen = 0; 4084 nfsdarg.dspath = NULL; 4085 nfsdarg.dspathlen = 0; 4086 nfsdarg.mdspath = NULL; 4087 nfsdarg.mdspathlen = 0; 4088 nfsdarg.mirrorcnt = 1; 4089 } 4090 nfsd_timer(NFSD_TD_TO_VNET(td)); 4091 error = nfsrvd_nfsd(td, &nfsdarg); 4092 callout_drain(&NFSD_VNET(nfsd_callout)); 4093 free(nfsdarg.addr, M_TEMP); 4094 free(nfsdarg.dnshost, M_TEMP); 4095 free(nfsdarg.dspath, M_TEMP); 4096 free(nfsdarg.mdspath, M_TEMP); 4097 } else if (uap->flag & NFSSVC_PNFSDS) { 4098 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg)); 4099 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER || 4100 pnfsdarg.op == PNFSDOP_FORCEDELDS)) { 4101 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4102 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1, 4103 NULL); 4104 if (error == 0) 4105 error = nfsrv_deldsserver(pnfsdarg.op, cp, td); 4106 free(cp, M_TEMP); 4107 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) { 4108 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4109 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS; 4110 buf = malloc(buflen, M_TEMP, M_WAITOK); 4111 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1, 4112 NULL); 4113 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error); 4114 if (error == 0 && pnfsdarg.dspath != NULL) { 4115 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4116 error = copyinstr(pnfsdarg.dspath, cp2, 4117 PATH_MAX + 1, NULL); 4118 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n", 4119 error); 4120 } else 4121 cp2 = NULL; 4122 if (error == 0 && pnfsdarg.curdspath != NULL) { 4123 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK); 4124 error = copyinstr(pnfsdarg.curdspath, cp3, 4125 PATH_MAX + 1, NULL); 4126 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n", 4127 error); 4128 } else 4129 cp3 = NULL; 4130 curdvp = NULL; 4131 fds = NULL; 4132 if (error == 0) 4133 error = nfsrv_mdscopymr(cp, cp2, cp3, buf, 4134 &buflen, fname, td, &vp, &nvp, &pf, &ds, 4135 &fds); 4136 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error); 4137 if (error == 0) { 4138 if (pf->dsf_dir >= nfsrv_dsdirsize) { 4139 printf("copymr: dsdir out of range\n"); 4140 pf->dsf_dir = 0; 4141 } 4142 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen); 4143 error = nfsrv_copymr(vp, nvp, 4144 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf, 4145 (struct pnfsdsfile *)buf, 4146 buflen / sizeof(*pf), td->td_ucred, td); 4147 vput(vp); 4148 vput(nvp); 4149 if (fds != NULL && error == 0) { 4150 curdvp = fds->nfsdev_dsdir[pf->dsf_dir]; 4151 ret = vn_lock(curdvp, LK_EXCLUSIVE); 4152 if (ret == 0) { 4153 nfsrv_dsremove(curdvp, fname, 4154 td->td_ucred, td); 4155 NFSVOPUNLOCK(curdvp); 4156 } 4157 } 4158 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error); 4159 } 4160 free(cp, M_TEMP); 4161 free(cp2, M_TEMP); 4162 free(cp3, M_TEMP); 4163 free(buf, M_TEMP); 4164 } 4165 } else { 4166 error = nfssvc_srvcall(td, uap, td->td_ucred); 4167 } 4168 4169 out: 4170 NFSD_CURVNET_RESTORE(); 4171 NFSEXITCODE(error); 4172 return (error); 4173 } 4174 4175 static int 4176 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 4177 { 4178 struct nfsex_args export; 4179 struct nfsex_oldargs oexp; 4180 struct file *fp = NULL; 4181 int stablefd, i, len; 4182 struct nfsd_clid adminrevoke; 4183 struct nfsd_dumplist dumplist; 4184 struct nfsd_dumpclients *dumpclients; 4185 struct nfsd_dumplocklist dumplocklist; 4186 struct nfsd_dumplocks *dumplocks; 4187 struct nameidata nd; 4188 vnode_t vp; 4189 int error = EINVAL, igotlock; 4190 struct proc *procp; 4191 gid_t *grps; 4192 4193 if (uap->flag & NFSSVC_PUBLICFH) { 4194 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 4195 sizeof (fhandle_t)); 4196 error = copyin(uap->argp, 4197 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 4198 if (!error) 4199 nfs_pubfhset = 1; 4200 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 4201 (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) { 4202 error = copyin(uap->argp,(caddr_t)&export, 4203 sizeof (struct nfsex_args)); 4204 if (!error) { 4205 grps = NULL; 4206 if (export.export.ex_ngroups > NGROUPS_MAX || 4207 export.export.ex_ngroups < 0) 4208 error = EINVAL; 4209 else if (export.export.ex_ngroups > 0) { 4210 grps = malloc(export.export.ex_ngroups * 4211 sizeof(gid_t), M_TEMP, M_WAITOK); 4212 error = copyin(export.export.ex_groups, grps, 4213 export.export.ex_ngroups * sizeof(gid_t)); 4214 export.export.ex_groups = grps; 4215 } else 4216 export.export.ex_groups = NULL; 4217 if (!error) 4218 error = nfsrv_v4rootexport(&export, cred, p); 4219 free(grps, M_TEMP); 4220 } 4221 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) == 4222 NFSSVC_V4ROOTEXPORT) { 4223 error = copyin(uap->argp,(caddr_t)&oexp, 4224 sizeof (struct nfsex_oldargs)); 4225 if (!error) { 4226 memset(&export.export, 0, sizeof(export.export)); 4227 export.export.ex_flags = (uint64_t)oexp.export.ex_flags; 4228 export.export.ex_root = oexp.export.ex_root; 4229 export.export.ex_uid = oexp.export.ex_anon.cr_uid; 4230 export.export.ex_ngroups = 4231 oexp.export.ex_anon.cr_ngroups; 4232 export.export.ex_groups = NULL; 4233 if (export.export.ex_ngroups > XU_NGROUPS || 4234 export.export.ex_ngroups < 0) 4235 error = EINVAL; 4236 else if (export.export.ex_ngroups > 0) { 4237 export.export.ex_groups = malloc( 4238 export.export.ex_ngroups * sizeof(gid_t), 4239 M_TEMP, M_WAITOK); 4240 for (i = 0; i < export.export.ex_ngroups; i++) 4241 export.export.ex_groups[i] = 4242 oexp.export.ex_anon.cr_groups[i]; 4243 } 4244 export.export.ex_addr = oexp.export.ex_addr; 4245 export.export.ex_addrlen = oexp.export.ex_addrlen; 4246 export.export.ex_mask = oexp.export.ex_mask; 4247 export.export.ex_masklen = oexp.export.ex_masklen; 4248 export.export.ex_indexfile = oexp.export.ex_indexfile; 4249 export.export.ex_numsecflavors = 4250 oexp.export.ex_numsecflavors; 4251 if (export.export.ex_numsecflavors >= MAXSECFLAVORS || 4252 export.export.ex_numsecflavors < 0) 4253 error = EINVAL; 4254 else { 4255 for (i = 0; i < export.export.ex_numsecflavors; 4256 i++) 4257 export.export.ex_secflavors[i] = 4258 oexp.export.ex_secflavors[i]; 4259 } 4260 export.fspec = oexp.fspec; 4261 if (error == 0) 4262 error = nfsrv_v4rootexport(&export, cred, p); 4263 free(export.export.ex_groups, M_TEMP); 4264 } 4265 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 4266 nfs_pubfhset = 0; 4267 error = 0; 4268 } else if (uap->flag & NFSSVC_STABLERESTART) { 4269 error = copyin(uap->argp, (caddr_t)&stablefd, 4270 sizeof (int)); 4271 if (!error) 4272 error = fp_getfvp(p, stablefd, &fp, &vp); 4273 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 4274 error = EBADF; 4275 if (!error && NFSD_VNET(nfsrv_numnfsd) != 0) 4276 error = ENXIO; 4277 if (!error) { 4278 NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp; 4279 nfsrv_setupstable(p); 4280 } 4281 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 4282 error = copyin(uap->argp, (caddr_t)&adminrevoke, 4283 sizeof (struct nfsd_clid)); 4284 if (!error) 4285 error = nfsrv_adminrevoke(&adminrevoke, p); 4286 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 4287 error = copyin(uap->argp, (caddr_t)&dumplist, 4288 sizeof (struct nfsd_dumplist)); 4289 if (!error && (dumplist.ndl_size < 1 || 4290 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 4291 error = EPERM; 4292 if (!error) { 4293 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 4294 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4295 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 4296 error = copyout(dumpclients, dumplist.ndl_list, len); 4297 free(dumpclients, M_TEMP); 4298 } 4299 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 4300 error = copyin(uap->argp, (caddr_t)&dumplocklist, 4301 sizeof (struct nfsd_dumplocklist)); 4302 if (!error && (dumplocklist.ndllck_size < 1 || 4303 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 4304 error = EPERM; 4305 if (!error) 4306 error = nfsrv_lookupfilename(&nd, 4307 dumplocklist.ndllck_fname, p); 4308 if (!error) { 4309 len = sizeof (struct nfsd_dumplocks) * 4310 dumplocklist.ndllck_size; 4311 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 4312 nfsrv_dumplocks(nd.ni_vp, dumplocks, 4313 dumplocklist.ndllck_size, p); 4314 vput(nd.ni_vp); 4315 error = copyout(dumplocks, dumplocklist.ndllck_list, 4316 len); 4317 free(dumplocks, M_TEMP); 4318 } 4319 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 4320 procp = p->td_proc; 4321 PROC_LOCK(procp); 4322 nfsd_master_pid = procp->p_pid; 4323 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 4324 nfsd_master_start = procp->p_stats->p_start; 4325 NFSD_VNET(nfsd_master_proc) = procp; 4326 PROC_UNLOCK(procp); 4327 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) { 4328 NFSLOCKV4ROOTMUTEX(); 4329 if (!NFSD_VNET(nfsrv_suspend_nfsd)) { 4330 /* Lock out all nfsd threads */ 4331 do { 4332 igotlock = nfsv4_lock( 4333 &NFSD_VNET(nfsd_suspend_lock), 1, NULL, 4334 NFSV4ROOTLOCKMUTEXPTR, NULL); 4335 } while (igotlock == 0 && 4336 !NFSD_VNET(nfsrv_suspend_nfsd)); 4337 NFSD_VNET(nfsrv_suspend_nfsd) = true; 4338 } 4339 NFSUNLOCKV4ROOTMUTEX(); 4340 error = 0; 4341 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) { 4342 NFSLOCKV4ROOTMUTEX(); 4343 if (NFSD_VNET(nfsrv_suspend_nfsd)) { 4344 nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0); 4345 NFSD_VNET(nfsrv_suspend_nfsd) = false; 4346 } 4347 NFSUNLOCKV4ROOTMUTEX(); 4348 error = 0; 4349 } 4350 4351 NFSEXITCODE(error); 4352 return (error); 4353 } 4354 4355 /* 4356 * Check exports. 4357 * Returns 0 if ok, 1 otherwise. 4358 */ 4359 int 4360 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 4361 { 4362 int i; 4363 4364 if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) || 4365 (NFSVNO_EXTLSCERT(exp) && 4366 (nd->nd_flag & ND_TLSCERT) == 0) || 4367 (NFSVNO_EXTLSCERTUSER(exp) && 4368 (nd->nd_flag & ND_TLSCERTUSER) == 0)) { 4369 if ((nd->nd_flag & ND_NFSV4) != 0) 4370 return (NFSERR_WRONGSEC); 4371 #ifdef notnow 4372 /* There is currently no auth_stat for this. */ 4373 else if ((nd->nd_flag & ND_TLS) == 0) 4374 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS); 4375 else 4376 return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST); 4377 #endif 4378 else 4379 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4380 } 4381 4382 /* 4383 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use 4384 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. 4385 */ 4386 if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO) 4387 return (0); 4388 4389 /* 4390 * This seems odd, but allow the case where the security flavor 4391 * list is empty. This happens when NFSv4 is traversing non-exported 4392 * file systems. Exported file systems should always have a non-empty 4393 * security flavor list. 4394 */ 4395 if (exp->nes_numsecflavor == 0) 4396 return (0); 4397 4398 for (i = 0; i < exp->nes_numsecflavor; i++) { 4399 /* 4400 * The tests for privacy and integrity must be first, 4401 * since ND_GSS is set for everything but AUTH_SYS. 4402 */ 4403 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 4404 (nd->nd_flag & ND_GSSPRIVACY)) 4405 return (0); 4406 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 4407 (nd->nd_flag & ND_GSSINTEGRITY)) 4408 return (0); 4409 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 4410 (nd->nd_flag & ND_GSS)) 4411 return (0); 4412 if (exp->nes_secflavors[i] == AUTH_SYS && 4413 (nd->nd_flag & ND_GSS) == 0) 4414 return (0); 4415 } 4416 if ((nd->nd_flag & ND_NFSV4) != 0) 4417 return (NFSERR_WRONGSEC); 4418 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 4419 } 4420 4421 /* 4422 * Calculate a hash value for the fid in a file handle. 4423 */ 4424 uint32_t 4425 nfsrv_hashfh(fhandle_t *fhp) 4426 { 4427 uint32_t hashval; 4428 4429 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 4430 return (hashval); 4431 } 4432 4433 /* 4434 * Calculate a hash value for the sessionid. 4435 */ 4436 uint32_t 4437 nfsrv_hashsessionid(uint8_t *sessionid) 4438 { 4439 uint32_t hashval; 4440 4441 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0); 4442 return (hashval); 4443 } 4444 4445 /* 4446 * Signal the userland master nfsd to backup the stable restart file. 4447 */ 4448 void 4449 nfsrv_backupstable(void) 4450 { 4451 struct proc *procp; 4452 4453 if (NFSD_VNET(nfsd_master_proc) != NULL) { 4454 procp = pfind(nfsd_master_pid); 4455 /* Try to make sure it is the correct process. */ 4456 if (procp == NFSD_VNET(nfsd_master_proc) && 4457 procp->p_stats->p_start.tv_sec == 4458 nfsd_master_start.tv_sec && 4459 procp->p_stats->p_start.tv_usec == 4460 nfsd_master_start.tv_usec && 4461 strcmp(procp->p_comm, nfsd_master_comm) == 0) 4462 kern_psignal(procp, SIGUSR2); 4463 else 4464 NFSD_VNET(nfsd_master_proc) = NULL; 4465 4466 if (procp != NULL) 4467 PROC_UNLOCK(procp); 4468 } 4469 } 4470 4471 /* 4472 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror. 4473 * The arguments are in a structure, so that they can be passed through 4474 * taskqueue for a kernel process to execute this function. 4475 */ 4476 struct nfsrvdscreate { 4477 int done; 4478 int inprog; 4479 struct task tsk; 4480 struct ucred *tcred; 4481 struct vnode *dvp; 4482 NFSPROC_T *p; 4483 struct pnfsdsfile *pf; 4484 int err; 4485 fhandle_t fh; 4486 struct vattr va; 4487 struct vattr createva; 4488 }; 4489 4490 int 4491 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap, 4492 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa, 4493 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp) 4494 { 4495 struct vnode *nvp; 4496 struct nameidata named; 4497 struct vattr va; 4498 char *bufp; 4499 u_long *hashp; 4500 struct nfsnode *np; 4501 struct nfsmount *nmp; 4502 int error; 4503 4504 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE, 4505 LOCKPARENT | LOCKLEAF | NOCACHE); 4506 nfsvno_setpathbuf(&named, &bufp, &hashp); 4507 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE; 4508 named.ni_cnd.cn_nameptr = bufp; 4509 if (fnamep != NULL) { 4510 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1); 4511 named.ni_cnd.cn_namelen = strlen(bufp); 4512 } else 4513 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp); 4514 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp); 4515 4516 /* Create the date file in the DS mount. */ 4517 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4518 if (error == 0) { 4519 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap); 4520 vref(dvp); 4521 VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false); 4522 if (error == 0) { 4523 /* Set the ownership of the file. */ 4524 error = VOP_SETATTR(nvp, nvap, tcred); 4525 NFSD_DEBUG(4, "nfsrv_dscreate:" 4526 " setattr-uid=%d\n", error); 4527 if (error != 0) 4528 vput(nvp); 4529 } 4530 if (error != 0) 4531 printf("pNFS: pnfscreate failed=%d\n", error); 4532 } else 4533 printf("pNFS: pnfscreate vnlock=%d\n", error); 4534 if (error == 0) { 4535 np = VTONFS(nvp); 4536 nmp = VFSTONFS(nvp->v_mount); 4537 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs") 4538 != 0 || nmp->nm_nam->sa_len > sizeof( 4539 struct sockaddr_in6) || 4540 np->n_fhp->nfh_len != NFSX_MYFH) { 4541 printf("Bad DS file: fstype=%s salen=%d" 4542 " fhlen=%d\n", 4543 nvp->v_mount->mnt_vfc->vfc_name, 4544 nmp->nm_nam->sa_len, np->n_fhp->nfh_len); 4545 error = ENOENT; 4546 } 4547 4548 /* Set extattrs for the DS on the MDS file. */ 4549 if (error == 0) { 4550 if (dsa != NULL) { 4551 error = VOP_GETATTR(nvp, &va, tcred); 4552 if (error == 0) { 4553 dsa->dsa_filerev = va.va_filerev; 4554 dsa->dsa_size = va.va_size; 4555 dsa->dsa_atime = va.va_atime; 4556 dsa->dsa_mtime = va.va_mtime; 4557 dsa->dsa_bytes = va.va_bytes; 4558 } 4559 } 4560 if (error == 0) { 4561 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, 4562 NFSX_MYFH); 4563 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin, 4564 nmp->nm_nam->sa_len); 4565 NFSBCOPY(named.ni_cnd.cn_nameptr, 4566 pf->dsf_filename, 4567 sizeof(pf->dsf_filename)); 4568 } 4569 } else 4570 printf("pNFS: pnfscreate can't get DS" 4571 " attr=%d\n", error); 4572 if (nvpp != NULL && error == 0) 4573 *nvpp = nvp; 4574 else 4575 vput(nvp); 4576 } 4577 nfsvno_relpathbuf(&named); 4578 return (error); 4579 } 4580 4581 /* 4582 * Start up the thread that will execute nfsrv_dscreate(). 4583 */ 4584 static void 4585 start_dscreate(void *arg, int pending) 4586 { 4587 struct nfsrvdscreate *dsc; 4588 4589 dsc = (struct nfsrvdscreate *)arg; 4590 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh, 4591 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL); 4592 dsc->done = 1; 4593 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err); 4594 } 4595 4596 /* 4597 * Create a pNFS data file on the Data Server(s). 4598 */ 4599 static void 4600 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred, 4601 NFSPROC_T *p) 4602 { 4603 struct nfsrvdscreate *dsc, *tdsc = NULL; 4604 struct nfsdevice *ds, *tds, *fds; 4605 struct mount *mp; 4606 struct pnfsdsfile *pf, *tpf; 4607 struct pnfsdsattr dsattr; 4608 struct vattr va; 4609 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 4610 struct nfsmount *nmp; 4611 fhandle_t fh; 4612 uid_t vauid; 4613 gid_t vagid; 4614 u_short vamode; 4615 struct ucred *tcred; 4616 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret; 4617 int failpos, timo; 4618 4619 /* Get a DS server directory in a round-robin order. */ 4620 mirrorcnt = 1; 4621 mp = vp->v_mount; 4622 ds = fds = NULL; 4623 NFSDDSLOCK(); 4624 /* 4625 * Search for the first entry that handles this MDS fs, but use the 4626 * first entry for all MDS fs's otherwise. 4627 */ 4628 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) { 4629 if (tds->nfsdev_nmp != NULL) { 4630 if (tds->nfsdev_mdsisset == 0 && ds == NULL) 4631 ds = tds; 4632 else if (tds->nfsdev_mdsisset != 0 && fsidcmp( 4633 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) { 4634 ds = fds = tds; 4635 break; 4636 } 4637 } 4638 } 4639 if (ds == NULL) { 4640 NFSDDSUNLOCK(); 4641 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n"); 4642 return; 4643 } 4644 i = dsdir[0] = ds->nfsdev_nextdir; 4645 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize; 4646 dvp[0] = ds->nfsdev_dsdir[i]; 4647 tds = TAILQ_NEXT(ds, nfsdev_list); 4648 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) { 4649 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) { 4650 if (tds->nfsdev_nmp != NULL && 4651 ((tds->nfsdev_mdsisset == 0 && fds == NULL) || 4652 (tds->nfsdev_mdsisset != 0 && fds != NULL && 4653 fsidcmp(&mp->mnt_stat.f_fsid, 4654 &tds->nfsdev_mdsfsid) == 0))) { 4655 dsdir[mirrorcnt] = i; 4656 dvp[mirrorcnt] = tds->nfsdev_dsdir[i]; 4657 mirrorcnt++; 4658 if (mirrorcnt >= nfsrv_maxpnfsmirror) 4659 break; 4660 } 4661 } 4662 } 4663 /* Put at end of list to implement round-robin usage. */ 4664 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 4665 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 4666 NFSDDSUNLOCK(); 4667 dsc = NULL; 4668 if (mirrorcnt > 1) 4669 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP, 4670 M_WAITOK | M_ZERO); 4671 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK | 4672 M_ZERO); 4673 4674 error = nfsvno_getfh(vp, &fh, p); 4675 if (error == 0) 4676 error = VOP_GETATTR(vp, &va, cred); 4677 if (error == 0) { 4678 /* Set the attributes for "vp" to Setattr the DS vp. */ 4679 vauid = va.va_uid; 4680 vagid = va.va_gid; 4681 vamode = va.va_mode; 4682 VATTR_NULL(&va); 4683 va.va_uid = vauid; 4684 va.va_gid = vagid; 4685 va.va_mode = vamode; 4686 va.va_size = 0; 4687 } else 4688 printf("pNFS: pnfscreate getfh+attr=%d\n", error); 4689 4690 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid, 4691 cred->cr_gid); 4692 /* Make data file name based on FH. */ 4693 tcred = newnfs_getcred(); 4694 4695 /* 4696 * Create the file on each DS mirror, using kernel process(es) for the 4697 * additional mirrors. 4698 */ 4699 failpos = -1; 4700 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) { 4701 tpf->dsf_dir = dsdir[i]; 4702 tdsc->tcred = tcred; 4703 tdsc->p = p; 4704 tdsc->pf = tpf; 4705 tdsc->createva = *vap; 4706 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh)); 4707 tdsc->va = va; 4708 tdsc->dvp = dvp[i]; 4709 tdsc->done = 0; 4710 tdsc->inprog = 0; 4711 tdsc->err = 0; 4712 ret = EIO; 4713 if (nfs_pnfsiothreads != 0) { 4714 ret = nfs_pnfsio(start_dscreate, tdsc); 4715 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret); 4716 } 4717 if (ret != 0) { 4718 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL, 4719 NULL, tcred, p, NULL); 4720 if (ret != 0) { 4721 KASSERT(error == 0, ("nfsrv_dscreate err=%d", 4722 error)); 4723 if (failpos == -1 && nfsds_failerr(ret)) 4724 failpos = i; 4725 else 4726 error = ret; 4727 } 4728 } 4729 } 4730 if (error == 0) { 4731 tpf->dsf_dir = dsdir[mirrorcnt - 1]; 4732 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf, 4733 &dsattr, NULL, tcred, p, NULL); 4734 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) { 4735 failpos = mirrorcnt - 1; 4736 error = 0; 4737 } 4738 } 4739 timo = hz / 50; /* Wait for 20msec. */ 4740 if (timo < 1) 4741 timo = 1; 4742 /* Wait for kernel task(s) to complete. */ 4743 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) { 4744 while (tdsc->inprog != 0 && tdsc->done == 0) 4745 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo); 4746 if (tdsc->err != 0) { 4747 if (failpos == -1 && nfsds_failerr(tdsc->err)) 4748 failpos = i; 4749 else if (error == 0) 4750 error = tdsc->err; 4751 } 4752 } 4753 4754 /* 4755 * If failpos has been set, that mirror has failed, so it needs 4756 * to be disabled. 4757 */ 4758 if (failpos >= 0) { 4759 nmp = VFSTONFS(dvp[failpos]->v_mount); 4760 NFSLOCKMNT(nmp); 4761 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4762 NFSMNTP_CANCELRPCS)) == 0) { 4763 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4764 NFSUNLOCKMNT(nmp); 4765 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4766 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos, 4767 ds); 4768 if (ds != NULL) 4769 nfsrv_killrpcs(nmp); 4770 NFSLOCKMNT(nmp); 4771 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 4772 wakeup(nmp); 4773 } 4774 NFSUNLOCKMNT(nmp); 4775 } 4776 4777 NFSFREECRED(tcred); 4778 if (error == 0) { 4779 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp"); 4780 4781 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n", 4782 mirrorcnt, nfsrv_maxpnfsmirror); 4783 /* 4784 * For all mirrors that couldn't be created, fill in the 4785 * *pf structure, but with an IP address == 0.0.0.0. 4786 */ 4787 tpf = pf + mirrorcnt; 4788 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) { 4789 *tpf = *pf; 4790 tpf->dsf_sin.sin_family = AF_INET; 4791 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in); 4792 tpf->dsf_sin.sin_addr.s_addr = 0; 4793 tpf->dsf_sin.sin_port = 0; 4794 } 4795 4796 error = vn_extattr_set(vp, IO_NODELOCKED, 4797 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 4798 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p); 4799 if (error == 0) 4800 error = vn_extattr_set(vp, IO_NODELOCKED, 4801 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", 4802 sizeof(dsattr), (char *)&dsattr, p); 4803 if (error != 0) 4804 printf("pNFS: pnfscreate setextattr=%d\n", 4805 error); 4806 } else 4807 printf("pNFS: pnfscreate=%d\n", error); 4808 free(pf, M_TEMP); 4809 free(dsc, M_TEMP); 4810 } 4811 4812 /* 4813 * Get the information needed to remove the pNFS Data Server file from the 4814 * Metadata file. Upon success, ddvp is set non-NULL to the locked 4815 * DS directory vnode. The caller must unlock *ddvp when done with it. 4816 */ 4817 static void 4818 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp, 4819 int *mirrorcntp, char *fname, fhandle_t *fhp) 4820 { 4821 struct vattr va; 4822 struct ucred *tcred; 4823 char *buf; 4824 int buflen, error; 4825 4826 dvpp[0] = NULL; 4827 /* If not an exported regular file or not a pNFS server, just return. */ 4828 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 4829 nfsrv_devidcnt == 0) 4830 return; 4831 4832 /* Check to see if this is the last hard link. */ 4833 tcred = newnfs_getcred(); 4834 error = VOP_GETATTR(vp, &va, tcred); 4835 NFSFREECRED(tcred); 4836 if (error != 0) { 4837 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error); 4838 return; 4839 } 4840 if (va.va_nlink > 1) 4841 return; 4842 4843 error = nfsvno_getfh(vp, fhp, p); 4844 if (error != 0) { 4845 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error); 4846 return; 4847 } 4848 4849 buflen = 1024; 4850 buf = malloc(buflen, M_TEMP, M_WAITOK); 4851 /* Get the directory vnode for the DS mount and the file handle. */ 4852 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp, 4853 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL); 4854 free(buf, M_TEMP); 4855 if (error != 0) 4856 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error); 4857 } 4858 4859 /* 4860 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror. 4861 * The arguments are in a structure, so that they can be passed through 4862 * taskqueue for a kernel process to execute this function. 4863 */ 4864 struct nfsrvdsremove { 4865 int done; 4866 int inprog; 4867 struct task tsk; 4868 struct ucred *tcred; 4869 struct vnode *dvp; 4870 NFSPROC_T *p; 4871 int err; 4872 char fname[PNFS_FILENAME_LEN + 1]; 4873 }; 4874 4875 static int 4876 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred, 4877 NFSPROC_T *p) 4878 { 4879 struct nameidata named; 4880 struct vnode *nvp; 4881 char *bufp; 4882 u_long *hashp; 4883 int error; 4884 4885 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE); 4886 if (error != 0) 4887 return (error); 4888 named.ni_cnd.cn_nameiop = DELETE; 4889 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 4890 named.ni_cnd.cn_cred = tcred; 4891 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 4892 nfsvno_setpathbuf(&named, &bufp, &hashp); 4893 named.ni_cnd.cn_nameptr = bufp; 4894 named.ni_cnd.cn_namelen = strlen(fname); 4895 strlcpy(bufp, fname, NAME_MAX); 4896 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp); 4897 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 4898 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error); 4899 if (error == 0) { 4900 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd); 4901 vput(nvp); 4902 } 4903 NFSVOPUNLOCK(dvp); 4904 nfsvno_relpathbuf(&named); 4905 if (error != 0) 4906 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error); 4907 return (error); 4908 } 4909 4910 /* 4911 * Start up the thread that will execute nfsrv_dsremove(). 4912 */ 4913 static void 4914 start_dsremove(void *arg, int pending) 4915 { 4916 struct nfsrvdsremove *dsrm; 4917 4918 dsrm = (struct nfsrvdsremove *)arg; 4919 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred, 4920 dsrm->p); 4921 dsrm->done = 1; 4922 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err); 4923 } 4924 4925 /* 4926 * Remove a pNFS data file from a Data Server. 4927 * nfsrv_pnfsremovesetup() must have been called before the MDS file was 4928 * removed to set up the dvp and fill in the FH. 4929 */ 4930 static void 4931 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp, 4932 NFSPROC_T *p) 4933 { 4934 struct ucred *tcred; 4935 struct nfsrvdsremove *dsrm, *tdsrm; 4936 struct nfsdevice *ds; 4937 struct nfsmount *nmp; 4938 int failpos, i, ret, timo; 4939 4940 tcred = newnfs_getcred(); 4941 dsrm = NULL; 4942 if (mirrorcnt > 1) 4943 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK); 4944 /* 4945 * Remove the file on each DS mirror, using kernel process(es) for the 4946 * additional mirrors. 4947 */ 4948 failpos = -1; 4949 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4950 tdsrm->tcred = tcred; 4951 tdsrm->p = p; 4952 tdsrm->dvp = dvp[i]; 4953 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1); 4954 tdsrm->inprog = 0; 4955 tdsrm->done = 0; 4956 tdsrm->err = 0; 4957 ret = EIO; 4958 if (nfs_pnfsiothreads != 0) { 4959 ret = nfs_pnfsio(start_dsremove, tdsrm); 4960 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret); 4961 } 4962 if (ret != 0) { 4963 ret = nfsrv_dsremove(dvp[i], fname, tcred, p); 4964 if (failpos == -1 && nfsds_failerr(ret)) 4965 failpos = i; 4966 } 4967 } 4968 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p); 4969 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret)) 4970 failpos = mirrorcnt - 1; 4971 timo = hz / 50; /* Wait for 20msec. */ 4972 if (timo < 1) 4973 timo = 1; 4974 /* Wait for kernel task(s) to complete. */ 4975 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) { 4976 while (tdsrm->inprog != 0 && tdsrm->done == 0) 4977 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo); 4978 if (failpos == -1 && nfsds_failerr(tdsrm->err)) 4979 failpos = i; 4980 } 4981 4982 /* 4983 * If failpos has been set, that mirror has failed, so it needs 4984 * to be disabled. 4985 */ 4986 if (failpos >= 0) { 4987 nmp = VFSTONFS(dvp[failpos]->v_mount); 4988 NFSLOCKMNT(nmp); 4989 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 4990 NFSMNTP_CANCELRPCS)) == 0) { 4991 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 4992 NFSUNLOCKMNT(nmp); 4993 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p); 4994 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos, 4995 ds); 4996 if (ds != NULL) 4997 nfsrv_killrpcs(nmp); 4998 NFSLOCKMNT(nmp); 4999 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 5000 wakeup(nmp); 5001 } 5002 NFSUNLOCKMNT(nmp); 5003 } 5004 5005 /* Get rid all layouts for the file. */ 5006 nfsrv_freefilelayouts(fhp); 5007 5008 NFSFREECRED(tcred); 5009 free(dsrm, M_TEMP); 5010 } 5011 5012 /* 5013 * Generate a file name based on the file handle and put it in *bufp. 5014 * Return the number of bytes generated. 5015 */ 5016 static int 5017 nfsrv_putfhname(fhandle_t *fhp, char *bufp) 5018 { 5019 int i; 5020 uint8_t *cp; 5021 const uint8_t *hexdigits = "0123456789abcdef"; 5022 5023 cp = (uint8_t *)fhp; 5024 for (i = 0; i < sizeof(*fhp); i++) { 5025 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf]; 5026 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf]; 5027 } 5028 bufp[2 * i] = '\0'; 5029 return (2 * i); 5030 } 5031 5032 /* 5033 * Update the Metadata file's attributes from the DS file when a Read/Write 5034 * layout is returned. 5035 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN 5036 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file. 5037 */ 5038 int 5039 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 5040 { 5041 struct ucred *tcred; 5042 int error; 5043 5044 /* Do this as root so that it won't fail with EACCES. */ 5045 tcred = newnfs_getcred(); 5046 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN, 5047 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL); 5048 NFSFREECRED(tcred); 5049 return (error); 5050 } 5051 5052 /* 5053 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file. 5054 */ 5055 static int 5056 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred, 5057 NFSPROC_T *p) 5058 { 5059 int error; 5060 5061 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL, 5062 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL); 5063 return (error); 5064 } 5065 5066 static int 5067 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 5068 struct thread *p, int ioproc, struct mbuf **mpp, char *cp, 5069 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp, 5070 off_t *offp, int content, bool *eofp) 5071 { 5072 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp; 5073 fhandle_t fh[NFSDEV_MAXMIRRORS]; 5074 struct vnode *dvp[NFSDEV_MAXMIRRORS]; 5075 struct nfsdevice *ds; 5076 struct pnfsdsattr dsattr; 5077 struct opnfsdsattr odsattr; 5078 char *buf; 5079 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt; 5080 5081 NFSD_DEBUG(4, "in nfsrv_proxyds\n"); 5082 /* 5083 * If not a regular file, not exported or not a pNFS server, 5084 * just return ENOENT. 5085 */ 5086 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 || 5087 nfsrv_devidcnt == 0) 5088 return (ENOENT); 5089 5090 buflen = 1024; 5091 buf = malloc(buflen, M_TEMP, M_WAITOK); 5092 error = 0; 5093 5094 /* 5095 * For Getattr, get the Change attribute (va_filerev) and size (va_size) 5096 * from the MetaData file's extended attribute. 5097 */ 5098 if (ioproc == NFSPROC_GETATTR) { 5099 error = vn_extattr_get(vp, IO_NODELOCKED, 5100 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf, 5101 p); 5102 if (error == 0) { 5103 if (buflen == sizeof(odsattr)) { 5104 NFSBCOPY(buf, &odsattr, buflen); 5105 nap->na_filerev = odsattr.dsa_filerev; 5106 nap->na_size = odsattr.dsa_size; 5107 nap->na_atime = odsattr.dsa_atime; 5108 nap->na_mtime = odsattr.dsa_mtime; 5109 /* 5110 * Fake na_bytes by rounding up na_size. 5111 * Since we don't know the block size, just 5112 * use BLKDEV_IOSIZE. 5113 */ 5114 nap->na_bytes = (odsattr.dsa_size + 5115 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1); 5116 } else if (buflen == sizeof(dsattr)) { 5117 NFSBCOPY(buf, &dsattr, buflen); 5118 nap->na_filerev = dsattr.dsa_filerev; 5119 nap->na_size = dsattr.dsa_size; 5120 nap->na_atime = dsattr.dsa_atime; 5121 nap->na_mtime = dsattr.dsa_mtime; 5122 nap->na_bytes = dsattr.dsa_bytes; 5123 } else 5124 error = ENXIO; 5125 } 5126 if (error == 0) { 5127 /* 5128 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr() 5129 * returns 0, just return now. nfsrv_checkdsattr() 5130 * returns 0 if there is no Read/Write layout 5131 * plus either an Open/Write_access or Write 5132 * delegation issued to a client for the file. 5133 */ 5134 if (nfsrv_pnfsgetdsattr == 0 || 5135 nfsrv_checkdsattr(vp, p) == 0) { 5136 free(buf, M_TEMP); 5137 return (error); 5138 } 5139 } 5140 5141 /* 5142 * Clear ENOATTR so the code below will attempt to do a 5143 * nfsrv_getattrdsrpc() to get the attributes and (re)create 5144 * the extended attribute. 5145 */ 5146 if (error == ENOATTR) 5147 error = 0; 5148 } 5149 5150 origmircnt = -1; 5151 trycnt = 0; 5152 tryagain: 5153 if (error == 0) { 5154 buflen = 1024; 5155 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) == 5156 LK_EXCLUSIVE) 5157 printf("nfsrv_proxyds: Readds vp exclusively locked\n"); 5158 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen, 5159 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL, 5160 NULL, NULL); 5161 if (error == 0) { 5162 for (i = 0; i < mirrorcnt; i++) 5163 nmp[i] = VFSTONFS(dvp[i]->v_mount); 5164 } else 5165 printf("pNFS: proxy getextattr sockaddr=%d\n", error); 5166 } else 5167 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error); 5168 if (error == 0) { 5169 failpos = -1; 5170 if (origmircnt == -1) 5171 origmircnt = mirrorcnt; 5172 /* 5173 * If failpos is set to a mirror#, then that mirror has 5174 * failed and will be disabled. For Read, Getattr and Seek, the 5175 * function only tries one mirror, so if that mirror has 5176 * failed, it will need to be retried. As such, increment 5177 * tryitagain for these cases. 5178 * For Write, Setattr and Setacl, the function tries all 5179 * mirrors and will not return an error for the case where 5180 * one mirror has failed. For these cases, the functioning 5181 * mirror(s) will have been modified, so a retry isn't 5182 * necessary. These functions will set failpos for the 5183 * failed mirror#. 5184 */ 5185 if (ioproc == NFSPROC_READDS) { 5186 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0], 5187 mpp, mpp2); 5188 if (nfsds_failerr(error) && mirrorcnt > 1) { 5189 /* 5190 * Setting failpos will cause the mirror 5191 * to be disabled and then a retry of this 5192 * read is required. 5193 */ 5194 failpos = 0; 5195 error = 0; 5196 trycnt++; 5197 } 5198 } else if (ioproc == NFSPROC_WRITEDS) 5199 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp, 5200 &nmp[0], mirrorcnt, mpp, cp, &failpos); 5201 else if (ioproc == NFSPROC_SETATTR) 5202 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0], 5203 mirrorcnt, nap, &failpos); 5204 else if (ioproc == NFSPROC_SETACL) 5205 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0], 5206 mirrorcnt, aclp, &failpos); 5207 else if (ioproc == NFSPROC_SEEKDS) { 5208 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred, 5209 p, nmp[0]); 5210 if (nfsds_failerr(error) && mirrorcnt > 1) { 5211 /* 5212 * Setting failpos will cause the mirror 5213 * to be disabled and then a retry of this 5214 * read is required. 5215 */ 5216 failpos = 0; 5217 error = 0; 5218 trycnt++; 5219 } 5220 } else if (ioproc == NFSPROC_ALLOCATE) 5221 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp, 5222 &nmp[0], mirrorcnt, &failpos); 5223 else if (ioproc == NFSPROC_DEALLOCATE) 5224 error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p, 5225 vp, &nmp[0], mirrorcnt, &failpos); 5226 else { 5227 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p, 5228 vp, nmp[mirrorcnt - 1], nap); 5229 if (nfsds_failerr(error) && mirrorcnt > 1) { 5230 /* 5231 * Setting failpos will cause the mirror 5232 * to be disabled and then a retry of this 5233 * getattr is required. 5234 */ 5235 failpos = mirrorcnt - 1; 5236 error = 0; 5237 trycnt++; 5238 } 5239 } 5240 ds = NULL; 5241 if (failpos >= 0) { 5242 failnmp = nmp[failpos]; 5243 NFSLOCKMNT(failnmp); 5244 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM | 5245 NFSMNTP_CANCELRPCS)) == 0) { 5246 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS; 5247 NFSUNLOCKMNT(failnmp); 5248 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, 5249 failnmp, p); 5250 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n", 5251 failpos, ds); 5252 if (ds != NULL) 5253 nfsrv_killrpcs(failnmp); 5254 NFSLOCKMNT(failnmp); 5255 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 5256 wakeup(failnmp); 5257 } 5258 NFSUNLOCKMNT(failnmp); 5259 } 5260 for (i = 0; i < mirrorcnt; i++) 5261 NFSVOPUNLOCK(dvp[i]); 5262 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error, 5263 trycnt); 5264 /* Try the Read/Getattr again if a mirror was deleted. */ 5265 if (ds != NULL && trycnt > 0 && trycnt < origmircnt) 5266 goto tryagain; 5267 } else { 5268 /* Return ENOENT for any Extended Attribute error. */ 5269 error = ENOENT; 5270 } 5271 free(buf, M_TEMP); 5272 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error); 5273 return (error); 5274 } 5275 5276 /* 5277 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended 5278 * attribute. 5279 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs 5280 * to be checked. If it points to a NULL nmp, then it returns 5281 * a suitable destination. 5282 * curnmp - If non-NULL, it is the source mount for the copy. 5283 */ 5284 int 5285 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp, 5286 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp, 5287 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp, 5288 struct nfsmount *curnmp, int *ippos, int *dsdirp) 5289 { 5290 struct vnode *dvp, *nvp = NULL, **tdvpp; 5291 struct mount *mp; 5292 struct nfsmount *nmp, *newnmp; 5293 struct sockaddr *sad; 5294 struct sockaddr_in *sin; 5295 struct nfsdevice *ds, *tds, *fndds; 5296 struct pnfsdsfile *pf; 5297 uint32_t dsdir; 5298 int error, fhiszero, fnd, gotone, i, mirrorcnt; 5299 5300 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp"); 5301 *mirrorcntp = 1; 5302 tdvpp = dvpp; 5303 if (nvpp != NULL) 5304 *nvpp = NULL; 5305 if (dvpp != NULL) 5306 *dvpp = NULL; 5307 if (ippos != NULL) 5308 *ippos = -1; 5309 if (newnmpp != NULL) 5310 newnmp = *newnmpp; 5311 else 5312 newnmp = NULL; 5313 mp = vp->v_mount; 5314 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5315 "pnfsd.dsfile", buflenp, buf, p); 5316 mirrorcnt = *buflenp / sizeof(*pf); 5317 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS || 5318 *buflenp != sizeof(*pf) * mirrorcnt)) 5319 error = ENOATTR; 5320 5321 pf = (struct pnfsdsfile *)buf; 5322 /* If curnmp != NULL, check for a match in the mirror list. */ 5323 if (curnmp != NULL && error == 0) { 5324 fnd = 0; 5325 for (i = 0; i < mirrorcnt; i++, pf++) { 5326 sad = (struct sockaddr *)&pf->dsf_sin; 5327 if (nfsaddr2_match(sad, curnmp->nm_nam)) { 5328 if (ippos != NULL) 5329 *ippos = i; 5330 fnd = 1; 5331 break; 5332 } 5333 } 5334 if (fnd == 0) 5335 error = ENXIO; 5336 } 5337 5338 gotone = 0; 5339 pf = (struct pnfsdsfile *)buf; 5340 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt, 5341 error); 5342 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) { 5343 fhiszero = 0; 5344 sad = (struct sockaddr *)&pf->dsf_sin; 5345 sin = &pf->dsf_sin; 5346 dsdir = pf->dsf_dir; 5347 if (dsdir >= nfsrv_dsdirsize) { 5348 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir); 5349 error = ENOATTR; 5350 } else if (nvpp != NULL && newnmp != NULL && 5351 nfsaddr2_match(sad, newnmp->nm_nam)) 5352 error = EEXIST; 5353 if (error == 0) { 5354 if (ippos != NULL && curnmp == NULL && 5355 sad->sa_family == AF_INET && 5356 sin->sin_addr.s_addr == 0) 5357 *ippos = i; 5358 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0) 5359 fhiszero = 1; 5360 /* Use the socket address to find the mount point. */ 5361 fndds = NULL; 5362 NFSDDSLOCK(); 5363 /* Find a match for the IP address. */ 5364 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 5365 if (ds->nfsdev_nmp != NULL) { 5366 dvp = ds->nfsdev_dvp; 5367 nmp = VFSTONFS(dvp->v_mount); 5368 if (nmp != ds->nfsdev_nmp) 5369 printf("different2 nmp %p %p\n", 5370 nmp, ds->nfsdev_nmp); 5371 if (nfsaddr2_match(sad, nmp->nm_nam)) { 5372 fndds = ds; 5373 break; 5374 } 5375 } 5376 } 5377 if (fndds != NULL && newnmpp != NULL && 5378 newnmp == NULL) { 5379 /* Search for a place to make a mirror copy. */ 5380 TAILQ_FOREACH(tds, &nfsrv_devidhead, 5381 nfsdev_list) { 5382 if (tds->nfsdev_nmp != NULL && 5383 fndds != tds && 5384 ((tds->nfsdev_mdsisset == 0 && 5385 fndds->nfsdev_mdsisset == 0) || 5386 (tds->nfsdev_mdsisset != 0 && 5387 fndds->nfsdev_mdsisset != 0 && 5388 fsidcmp(&tds->nfsdev_mdsfsid, 5389 &mp->mnt_stat.f_fsid) == 0))) { 5390 *newnmpp = tds->nfsdev_nmp; 5391 break; 5392 } 5393 } 5394 if (tds != NULL) { 5395 /* 5396 * Move this entry to the end of the 5397 * list, so it won't be selected as 5398 * easily the next time. 5399 */ 5400 TAILQ_REMOVE(&nfsrv_devidhead, tds, 5401 nfsdev_list); 5402 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds, 5403 nfsdev_list); 5404 } 5405 } 5406 NFSDDSUNLOCK(); 5407 if (fndds != NULL) { 5408 dvp = fndds->nfsdev_dsdir[dsdir]; 5409 if (lktype != 0 || fhiszero != 0 || 5410 (nvpp != NULL && *nvpp == NULL)) { 5411 if (fhiszero != 0) 5412 error = vn_lock(dvp, 5413 LK_EXCLUSIVE); 5414 else if (lktype != 0) 5415 error = vn_lock(dvp, lktype); 5416 else 5417 error = vn_lock(dvp, LK_SHARED); 5418 /* 5419 * If the file handle is all 0's, try to 5420 * do a Lookup against the DS to acquire 5421 * it. 5422 * If dvpp == NULL or the Lookup fails, 5423 * unlock dvp after the call. 5424 */ 5425 if (error == 0 && (fhiszero != 0 || 5426 (nvpp != NULL && *nvpp == NULL))) { 5427 error = nfsrv_pnfslookupds(vp, 5428 dvp, pf, &nvp, p); 5429 if (error == 0) { 5430 if (fhiszero != 0) 5431 nfsrv_pnfssetfh( 5432 vp, pf, 5433 devid, 5434 fnamep, 5435 nvp, p); 5436 if (nvpp != NULL && 5437 *nvpp == NULL) { 5438 *nvpp = nvp; 5439 *dsdirp = dsdir; 5440 } else 5441 vput(nvp); 5442 } 5443 if (error != 0 || lktype == 0) 5444 NFSVOPUNLOCK(dvp); 5445 } 5446 } 5447 if (error == 0) { 5448 gotone++; 5449 NFSD_DEBUG(4, "gotone=%d\n", gotone); 5450 if (devid != NULL) { 5451 NFSBCOPY(fndds->nfsdev_deviceid, 5452 devid, NFSX_V4DEVICEID); 5453 devid += NFSX_V4DEVICEID; 5454 } 5455 if (dvpp != NULL) 5456 *tdvpp++ = dvp; 5457 if (fhp != NULL) 5458 NFSBCOPY(&pf->dsf_fh, fhp++, 5459 NFSX_MYFH); 5460 if (fnamep != NULL && gotone == 1) 5461 strlcpy(fnamep, 5462 pf->dsf_filename, 5463 sizeof(pf->dsf_filename)); 5464 } else 5465 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt " 5466 "err=%d\n", error); 5467 } 5468 } 5469 } 5470 if (error == 0 && gotone == 0) 5471 error = ENOENT; 5472 5473 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone, 5474 error); 5475 if (error == 0) 5476 *mirrorcntp = gotone; 5477 else { 5478 if (gotone > 0 && dvpp != NULL) { 5479 /* 5480 * If the error didn't occur on the first one and 5481 * dvpp != NULL, the one(s) prior to the failure will 5482 * have locked dvp's that need to be unlocked. 5483 */ 5484 for (i = 0; i < gotone; i++) { 5485 NFSVOPUNLOCK(*dvpp); 5486 *dvpp++ = NULL; 5487 } 5488 } 5489 /* 5490 * If it found the vnode to be copied from before a failure, 5491 * it needs to be vput()'d. 5492 */ 5493 if (nvpp != NULL && *nvpp != NULL) { 5494 vput(*nvpp); 5495 *nvpp = NULL; 5496 } 5497 } 5498 return (error); 5499 } 5500 5501 /* 5502 * Set the extended attribute for the Change attribute. 5503 */ 5504 static int 5505 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p) 5506 { 5507 struct pnfsdsattr dsattr; 5508 int error; 5509 5510 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp"); 5511 dsattr.dsa_filerev = nap->na_filerev; 5512 dsattr.dsa_size = nap->na_size; 5513 dsattr.dsa_atime = nap->na_atime; 5514 dsattr.dsa_mtime = nap->na_mtime; 5515 dsattr.dsa_bytes = nap->na_bytes; 5516 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM, 5517 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p); 5518 if (error != 0) 5519 printf("pNFS: setextattr=%d\n", error); 5520 return (error); 5521 } 5522 5523 static int 5524 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5525 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp) 5526 { 5527 uint32_t *tl; 5528 struct nfsrv_descript *nd; 5529 nfsv4stateid_t st; 5530 struct mbuf *m, *m2; 5531 int error = 0, retlen, tlen, trimlen; 5532 5533 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n"); 5534 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5535 *mpp = NULL; 5536 /* 5537 * Use a stateid where other is an alternating 01010 pattern and 5538 * seqid is 0xffffffff. This value is not defined as special by 5539 * the RFC and is used by the FreeBSD NFS server to indicate an 5540 * MDS->DS proxy operation. 5541 */ 5542 st.other[0] = 0x55555555; 5543 st.other[1] = 0x55555555; 5544 st.other[2] = 0x55555555; 5545 st.seqid = 0xffffffff; 5546 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp), 5547 NULL, NULL, 0, 0, cred); 5548 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5549 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); 5550 txdr_hyper(off, tl); 5551 *(tl + 2) = txdr_unsigned(len); 5552 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 5553 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5554 if (error != 0) { 5555 free(nd, M_TEMP); 5556 return (error); 5557 } 5558 if (nd->nd_repstat == 0) { 5559 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 5560 NFSM_STRSIZ(retlen, len); 5561 if (retlen > 0) { 5562 /* Trim off the pre-data XDR from the mbuf chain. */ 5563 m = nd->nd_mrep; 5564 while (m != NULL && m != nd->nd_md) { 5565 if (m->m_next == nd->nd_md) { 5566 m->m_next = NULL; 5567 m_freem(nd->nd_mrep); 5568 nd->nd_mrep = m = nd->nd_md; 5569 } else 5570 m = m->m_next; 5571 } 5572 if (m == NULL) { 5573 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5574 error = ENOENT; 5575 goto nfsmout; 5576 } 5577 5578 /* 5579 * Now, adjust first mbuf so that any XDR before the 5580 * read data is skipped over. 5581 */ 5582 trimlen = nd->nd_dpos - mtod(m, char *); 5583 if (trimlen > 0) { 5584 m->m_len -= trimlen; 5585 NFSM_DATAP(m, trimlen); 5586 } 5587 5588 /* 5589 * Truncate the mbuf chain at retlen bytes of data, 5590 * plus XDR padding that brings the length up to a 5591 * multiple of 4. 5592 */ 5593 tlen = NFSM_RNDUP(retlen); 5594 do { 5595 if (m->m_len >= tlen) { 5596 m->m_len = tlen; 5597 tlen = 0; 5598 m2 = m->m_next; 5599 m->m_next = NULL; 5600 m_freem(m2); 5601 break; 5602 } 5603 tlen -= m->m_len; 5604 m = m->m_next; 5605 } while (m != NULL); 5606 if (tlen > 0) { 5607 printf("nfsrv_readdsrpc: busted mbuf list\n"); 5608 error = ENOENT; 5609 goto nfsmout; 5610 } 5611 *mpp = nd->nd_mrep; 5612 *mpendp = m; 5613 nd->nd_mrep = NULL; 5614 } 5615 } else 5616 error = nd->nd_repstat; 5617 nfsmout: 5618 /* If nd->nd_mrep is already NULL, this is a no-op. */ 5619 m_freem(nd->nd_mrep); 5620 free(nd, M_TEMP); 5621 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error); 5622 return (error); 5623 } 5624 5625 /* 5626 * Do a write RPC on a DS data file, using this structure for the arguments, 5627 * so that this function can be executed by a separate kernel process. 5628 */ 5629 struct nfsrvwritedsdorpc { 5630 int done; 5631 int inprog; 5632 struct task tsk; 5633 fhandle_t fh; 5634 off_t off; 5635 int len; 5636 struct nfsmount *nmp; 5637 struct ucred *cred; 5638 NFSPROC_T *p; 5639 struct mbuf *m; 5640 int err; 5641 }; 5642 5643 static int 5644 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len, 5645 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p) 5646 { 5647 uint32_t *tl; 5648 struct nfsrv_descript *nd; 5649 nfsattrbit_t attrbits; 5650 nfsv4stateid_t st; 5651 int commit, error, retlen; 5652 5653 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5654 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp, 5655 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5656 5657 /* 5658 * Use a stateid where other is an alternating 01010 pattern and 5659 * seqid is 0xffffffff. This value is not defined as special by 5660 * the RFC and is used by the FreeBSD NFS server to indicate an 5661 * MDS->DS proxy operation. 5662 */ 5663 st.other[0] = 0x55555555; 5664 st.other[1] = 0x55555555; 5665 st.other[2] = 0x55555555; 5666 st.seqid = 0xffffffff; 5667 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5668 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); 5669 txdr_hyper(off, tl); 5670 tl += 2; 5671 /* 5672 * Do all writes FileSync, since the server doesn't hold onto dirty 5673 * buffers. Since clients should be accessing the DS servers directly 5674 * using the pNFS layouts, this just needs to work correctly as a 5675 * fallback. 5676 */ 5677 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC); 5678 *tl = txdr_unsigned(len); 5679 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len); 5680 5681 /* Put data in mbuf chain. */ 5682 nd->nd_mb->m_next = m; 5683 5684 /* Set nd_mb and nd_bpos to end of data. */ 5685 while (m->m_next != NULL) 5686 m = m->m_next; 5687 nd->nd_mb = m; 5688 nfsm_set(nd, m->m_len); 5689 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len); 5690 5691 /* Do a Getattr for the attributes that change upon writing. */ 5692 NFSZERO_ATTRBIT(&attrbits); 5693 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 5694 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 5695 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 5696 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 5697 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 5698 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 5699 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5700 (void) nfsrv_putattrbit(nd, &attrbits); 5701 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5702 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5703 if (error != 0) { 5704 free(nd, M_TEMP); 5705 return (error); 5706 } 5707 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat); 5708 /* Get rid of weak cache consistency data for now. */ 5709 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 5710 (ND_NFSV4 | ND_V4WCCATTR)) { 5711 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5712 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 5713 NULL); 5714 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error); 5715 if (error != 0) 5716 goto nfsmout; 5717 /* 5718 * Get rid of Op# and status for next op. 5719 */ 5720 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5721 if (*++tl != 0) 5722 nd->nd_flag |= ND_NOMOREDATA; 5723 } 5724 if (nd->nd_repstat == 0) { 5725 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); 5726 retlen = fxdr_unsigned(int, *tl++); 5727 commit = fxdr_unsigned(int, *tl); 5728 if (commit != NFSWRITE_FILESYNC) 5729 error = NFSERR_IO; 5730 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n", 5731 retlen, commit, error); 5732 } else 5733 error = nd->nd_repstat; 5734 /* We have no use for the Write Verifier since we use FileSync. */ 5735 5736 /* 5737 * Get the Change, Size, Access Time and Modify Time attributes and set 5738 * on the Metadata file, so its attributes will be what the file's 5739 * would be if it had been written. 5740 */ 5741 if (error == 0) { 5742 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5743 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5744 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 5745 NULL); 5746 } 5747 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error); 5748 nfsmout: 5749 m_freem(nd->nd_mrep); 5750 free(nd, M_TEMP); 5751 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error); 5752 return (error); 5753 } 5754 5755 /* 5756 * Start up the thread that will execute nfsrv_writedsdorpc(). 5757 */ 5758 static void 5759 start_writedsdorpc(void *arg, int pending) 5760 { 5761 struct nfsrvwritedsdorpc *drpc; 5762 5763 drpc = (struct nfsrvwritedsdorpc *)arg; 5764 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5765 drpc->len, NULL, drpc->m, drpc->cred, drpc->p); 5766 drpc->done = 1; 5767 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err); 5768 } 5769 5770 static int 5771 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred, 5772 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5773 struct mbuf **mpp, char *cp, int *failposp) 5774 { 5775 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL; 5776 struct nfsvattr na; 5777 struct mbuf *m; 5778 int error, i, offs, ret, timo; 5779 5780 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n"); 5781 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain")); 5782 drpc = NULL; 5783 if (mirrorcnt > 1) 5784 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5785 M_WAITOK); 5786 5787 /* Calculate offset in mbuf chain that data starts. */ 5788 offs = cp - mtod(*mpp, char *); 5789 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len); 5790 5791 /* 5792 * Do the write RPC for every DS, using a separate kernel process 5793 * for every DS except the last one. 5794 */ 5795 error = 0; 5796 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5797 tdrpc->done = 0; 5798 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5799 tdrpc->off = off; 5800 tdrpc->len = len; 5801 tdrpc->nmp = *nmpp; 5802 tdrpc->cred = cred; 5803 tdrpc->p = p; 5804 tdrpc->inprog = 0; 5805 tdrpc->err = 0; 5806 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5807 ret = EIO; 5808 if (nfs_pnfsiothreads != 0) { 5809 ret = nfs_pnfsio(start_writedsdorpc, tdrpc); 5810 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n", 5811 ret); 5812 } 5813 if (ret != 0) { 5814 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL, 5815 tdrpc->m, cred, p); 5816 if (nfsds_failerr(ret) && *failposp == -1) 5817 *failposp = i; 5818 else if (error == 0 && ret != 0) 5819 error = ret; 5820 } 5821 nmpp++; 5822 fhp++; 5823 } 5824 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK); 5825 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p); 5826 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5827 *failposp = mirrorcnt - 1; 5828 else if (error == 0 && ret != 0) 5829 error = ret; 5830 if (error == 0) 5831 error = nfsrv_setextattr(vp, &na, p); 5832 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error); 5833 tdrpc = drpc; 5834 timo = hz / 50; /* Wait for 20msec. */ 5835 if (timo < 1) 5836 timo = 1; 5837 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5838 /* Wait for RPCs on separate threads to complete. */ 5839 while (tdrpc->inprog != 0 && tdrpc->done == 0) 5840 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo); 5841 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 5842 *failposp = i; 5843 else if (error == 0 && tdrpc->err != 0) 5844 error = tdrpc->err; 5845 } 5846 free(drpc, M_TEMP); 5847 return (error); 5848 } 5849 5850 /* 5851 * Do a allocate RPC on a DS data file, using this structure for the arguments, 5852 * so that this function can be executed by a separate kernel process. 5853 */ 5854 struct nfsrvallocatedsdorpc { 5855 int done; 5856 int inprog; 5857 struct task tsk; 5858 fhandle_t fh; 5859 off_t off; 5860 off_t len; 5861 struct nfsmount *nmp; 5862 struct ucred *cred; 5863 NFSPROC_T *p; 5864 int err; 5865 }; 5866 5867 static int 5868 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 5869 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 5870 { 5871 uint32_t *tl; 5872 struct nfsrv_descript *nd; 5873 nfsattrbit_t attrbits; 5874 nfsv4stateid_t st; 5875 int error; 5876 5877 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 5878 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp, 5879 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 5880 5881 /* 5882 * Use a stateid where other is an alternating 01010 pattern and 5883 * seqid is 0xffffffff. This value is not defined as special by 5884 * the RFC and is used by the FreeBSD NFS server to indicate an 5885 * MDS->DS proxy operation. 5886 */ 5887 st.other[0] = 0x55555555; 5888 st.other[1] = 0x55555555; 5889 st.other[2] = 0x55555555; 5890 st.seqid = 0xffffffff; 5891 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 5892 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 5893 txdr_hyper(off, tl); tl += 2; 5894 txdr_hyper(len, tl); tl += 2; 5895 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len); 5896 5897 *tl = txdr_unsigned(NFSV4OP_GETATTR); 5898 NFSGETATTR_ATTRBIT(&attrbits); 5899 nfsrv_putattrbit(nd, &attrbits); 5900 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 5901 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 5902 if (error != 0) { 5903 free(nd, M_TEMP); 5904 return (error); 5905 } 5906 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n", 5907 nd->nd_repstat); 5908 if (nd->nd_repstat == 0) { 5909 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 5910 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 5911 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 5912 NULL); 5913 } else 5914 error = nd->nd_repstat; 5915 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error); 5916 nfsmout: 5917 m_freem(nd->nd_mrep); 5918 free(nd, M_TEMP); 5919 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error); 5920 return (error); 5921 } 5922 5923 /* 5924 * Start up the thread that will execute nfsrv_allocatedsdorpc(). 5925 */ 5926 static void 5927 start_allocatedsdorpc(void *arg, int pending) 5928 { 5929 struct nfsrvallocatedsdorpc *drpc; 5930 5931 drpc = (struct nfsrvallocatedsdorpc *)arg; 5932 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 5933 drpc->len, NULL, drpc->cred, drpc->p); 5934 drpc->done = 1; 5935 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err); 5936 } 5937 5938 static int 5939 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 5940 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 5941 int *failposp) 5942 { 5943 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL; 5944 struct nfsvattr na; 5945 int error, i, ret, timo; 5946 5947 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n"); 5948 drpc = NULL; 5949 if (mirrorcnt > 1) 5950 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 5951 M_WAITOK); 5952 5953 /* 5954 * Do the allocate RPC for every DS, using a separate kernel process 5955 * for every DS except the last one. 5956 */ 5957 error = 0; 5958 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5959 tdrpc->done = 0; 5960 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 5961 tdrpc->off = off; 5962 tdrpc->len = len; 5963 tdrpc->nmp = *nmpp; 5964 tdrpc->cred = cred; 5965 tdrpc->p = p; 5966 tdrpc->inprog = 0; 5967 tdrpc->err = 0; 5968 ret = EIO; 5969 if (nfs_pnfsiothreads != 0) { 5970 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc); 5971 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n", 5972 ret); 5973 } 5974 if (ret != 0) { 5975 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL, 5976 cred, p); 5977 if (nfsds_failerr(ret) && *failposp == -1) 5978 *failposp = i; 5979 else if (error == 0 && ret != 0) 5980 error = ret; 5981 } 5982 nmpp++; 5983 fhp++; 5984 } 5985 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 5986 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 5987 *failposp = mirrorcnt - 1; 5988 else if (error == 0 && ret != 0) 5989 error = ret; 5990 if (error == 0) 5991 error = nfsrv_setextattr(vp, &na, p); 5992 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error); 5993 tdrpc = drpc; 5994 timo = hz / 50; /* Wait for 20msec. */ 5995 if (timo < 1) 5996 timo = 1; 5997 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 5998 /* Wait for RPCs on separate threads to complete. */ 5999 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6000 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 6001 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6002 *failposp = i; 6003 else if (error == 0 && tdrpc->err != 0) 6004 error = tdrpc->err; 6005 } 6006 free(drpc, M_TEMP); 6007 return (error); 6008 } 6009 6010 /* 6011 * Do a deallocate RPC on a DS data file, using this structure for the 6012 * arguments, so that this function can be executed by a separate kernel 6013 * process. 6014 */ 6015 struct nfsrvdeallocatedsdorpc { 6016 int done; 6017 int inprog; 6018 struct task tsk; 6019 fhandle_t fh; 6020 off_t off; 6021 off_t len; 6022 struct nfsmount *nmp; 6023 struct ucred *cred; 6024 NFSPROC_T *p; 6025 int err; 6026 }; 6027 6028 static int 6029 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, 6030 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p) 6031 { 6032 uint32_t *tl; 6033 struct nfsrv_descript *nd; 6034 nfsattrbit_t attrbits; 6035 nfsv4stateid_t st; 6036 int error; 6037 6038 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6039 nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp, 6040 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6041 6042 /* 6043 * Use a stateid where other is an alternating 01010 pattern and 6044 * seqid is 0xffffffff. This value is not defined as special by 6045 * the RFC and is used by the FreeBSD NFS server to indicate an 6046 * MDS->DS proxy operation. 6047 */ 6048 st.other[0] = 0x55555555; 6049 st.other[1] = 0x55555555; 6050 st.other[2] = 0x55555555; 6051 st.seqid = 0xffffffff; 6052 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6053 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED); 6054 txdr_hyper(off, tl); tl += 2; 6055 txdr_hyper(len, tl); tl += 2; 6056 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len); 6057 6058 /* Do a Getattr for the attributes that change upon writing. */ 6059 NFSZERO_ATTRBIT(&attrbits); 6060 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6061 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6062 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6063 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6064 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6065 *tl = txdr_unsigned(NFSV4OP_GETATTR); 6066 nfsrv_putattrbit(nd, &attrbits); 6067 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, 6068 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6069 if (error != 0) { 6070 free(nd, M_TEMP); 6071 return (error); 6072 } 6073 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n", 6074 nd->nd_repstat); 6075 /* Get rid of weak cache consistency data for now. */ 6076 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 6077 (ND_NFSV4 | ND_V4WCCATTR)) { 6078 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 6079 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 6080 NULL); 6081 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error); 6082 if (error != 0) 6083 goto nfsmout; 6084 /* 6085 * Get rid of Op# and status for next op. 6086 */ 6087 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6088 if (*++tl != 0) 6089 nd->nd_flag |= ND_NOMOREDATA; 6090 } 6091 if (nd->nd_repstat == 0) { 6092 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6093 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, 6094 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL, 6095 NULL); 6096 } else 6097 error = nd->nd_repstat; 6098 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error); 6099 nfsmout: 6100 m_freem(nd->nd_mrep); 6101 free(nd, M_TEMP); 6102 NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error); 6103 return (error); 6104 } 6105 6106 /* 6107 * Start up the thread that will execute nfsrv_deallocatedsdorpc(). 6108 */ 6109 static void 6110 start_deallocatedsdorpc(void *arg, int pending) 6111 { 6112 struct nfsrvdeallocatedsdorpc *drpc; 6113 6114 drpc = (struct nfsrvdeallocatedsdorpc *)arg; 6115 drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off, 6116 drpc->len, NULL, drpc->cred, drpc->p); 6117 drpc->done = 1; 6118 NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err); 6119 } 6120 6121 static int 6122 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred, 6123 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 6124 int *failposp) 6125 { 6126 struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL; 6127 struct nfsvattr na; 6128 int error, i, ret, timo; 6129 6130 NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n"); 6131 drpc = NULL; 6132 if (mirrorcnt > 1) 6133 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6134 M_WAITOK); 6135 6136 /* 6137 * Do the deallocate RPC for every DS, using a separate kernel process 6138 * for every DS except the last one. 6139 */ 6140 error = 0; 6141 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6142 tdrpc->done = 0; 6143 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6144 tdrpc->off = off; 6145 tdrpc->len = len; 6146 tdrpc->nmp = *nmpp; 6147 tdrpc->cred = cred; 6148 tdrpc->p = p; 6149 tdrpc->inprog = 0; 6150 tdrpc->err = 0; 6151 ret = EIO; 6152 if (nfs_pnfsiothreads != 0) { 6153 ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc); 6154 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n", 6155 ret); 6156 } 6157 if (ret != 0) { 6158 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, 6159 NULL, cred, p); 6160 if (nfsds_failerr(ret) && *failposp == -1) 6161 *failposp = i; 6162 else if (error == 0 && ret != 0) 6163 error = ret; 6164 } 6165 nmpp++; 6166 fhp++; 6167 } 6168 ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p); 6169 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6170 *failposp = mirrorcnt - 1; 6171 else if (error == 0 && ret != 0) 6172 error = ret; 6173 if (error == 0) 6174 error = nfsrv_setextattr(vp, &na, p); 6175 NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error); 6176 tdrpc = drpc; 6177 timo = hz / 50; /* Wait for 20msec. */ 6178 if (timo < 1) 6179 timo = 1; 6180 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6181 /* Wait for RPCs on separate threads to complete. */ 6182 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6183 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo); 6184 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6185 *failposp = i; 6186 else if (error == 0 && tdrpc->err != 0) 6187 error = tdrpc->err; 6188 } 6189 free(drpc, M_TEMP); 6190 return (error); 6191 } 6192 6193 static int 6194 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6195 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap, 6196 struct nfsvattr *dsnap) 6197 { 6198 uint32_t *tl; 6199 struct nfsrv_descript *nd; 6200 nfsv4stateid_t st; 6201 nfsattrbit_t attrbits; 6202 int error; 6203 6204 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n"); 6205 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6206 /* 6207 * Use a stateid where other is an alternating 01010 pattern and 6208 * seqid is 0xffffffff. This value is not defined as special by 6209 * the RFC and is used by the FreeBSD NFS server to indicate an 6210 * MDS->DS proxy operation. 6211 */ 6212 st.other[0] = 0x55555555; 6213 st.other[1] = 0x55555555; 6214 st.other[2] = 0x55555555; 6215 st.seqid = 0xffffffff; 6216 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp), 6217 NULL, NULL, 0, 0, cred); 6218 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6219 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0); 6220 6221 /* Do a Getattr for the attributes that change due to writing. */ 6222 NFSZERO_ATTRBIT(&attrbits); 6223 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6224 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6225 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6226 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6227 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6228 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 6229 *tl = txdr_unsigned(NFSV4OP_GETATTR); 6230 (void) nfsrv_putattrbit(nd, &attrbits); 6231 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6232 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6233 if (error != 0) { 6234 free(nd, M_TEMP); 6235 return (error); 6236 } 6237 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n", 6238 nd->nd_repstat); 6239 /* Get rid of weak cache consistency data for now. */ 6240 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == 6241 (ND_NFSV4 | ND_V4WCCATTR)) { 6242 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 6243 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 6244 NULL, NULL); 6245 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error); 6246 if (error != 0) 6247 goto nfsmout; 6248 /* 6249 * Get rid of Op# and status for next op. 6250 */ 6251 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6252 if (*++tl != 0) 6253 nd->nd_flag |= ND_NOMOREDATA; 6254 } 6255 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 6256 if (error != 0) 6257 goto nfsmout; 6258 if (nd->nd_repstat != 0) 6259 error = nd->nd_repstat; 6260 /* 6261 * Get the Change, Size, Access Time and Modify Time attributes and set 6262 * on the Metadata file, so its attributes will be what the file's 6263 * would be if it had been written. 6264 */ 6265 if (error == 0) { 6266 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); 6267 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL, 6268 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 6269 NULL, NULL); 6270 } 6271 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error); 6272 nfsmout: 6273 m_freem(nd->nd_mrep); 6274 free(nd, M_TEMP); 6275 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error); 6276 return (error); 6277 } 6278 6279 struct nfsrvsetattrdsdorpc { 6280 int done; 6281 int inprog; 6282 struct task tsk; 6283 fhandle_t fh; 6284 struct nfsmount *nmp; 6285 struct vnode *vp; 6286 struct ucred *cred; 6287 NFSPROC_T *p; 6288 struct nfsvattr na; 6289 struct nfsvattr dsna; 6290 int err; 6291 }; 6292 6293 /* 6294 * Start up the thread that will execute nfsrv_setattrdsdorpc(). 6295 */ 6296 static void 6297 start_setattrdsdorpc(void *arg, int pending) 6298 { 6299 struct nfsrvsetattrdsdorpc *drpc; 6300 6301 drpc = (struct nfsrvsetattrdsdorpc *)arg; 6302 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p, 6303 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna); 6304 drpc->done = 1; 6305 } 6306 6307 static int 6308 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6309 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, 6310 struct nfsvattr *nap, int *failposp) 6311 { 6312 struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL; 6313 struct nfsvattr na; 6314 int error, i, ret, timo; 6315 6316 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n"); 6317 drpc = NULL; 6318 if (mirrorcnt > 1) 6319 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6320 M_WAITOK); 6321 6322 /* 6323 * Do the setattr RPC for every DS, using a separate kernel process 6324 * for every DS except the last one. 6325 */ 6326 error = 0; 6327 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6328 tdrpc->done = 0; 6329 tdrpc->inprog = 0; 6330 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6331 tdrpc->nmp = *nmpp; 6332 tdrpc->vp = vp; 6333 tdrpc->cred = cred; 6334 tdrpc->p = p; 6335 tdrpc->na = *nap; 6336 tdrpc->err = 0; 6337 ret = EIO; 6338 if (nfs_pnfsiothreads != 0) { 6339 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc); 6340 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n", 6341 ret); 6342 } 6343 if (ret != 0) { 6344 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, 6345 &na); 6346 if (nfsds_failerr(ret) && *failposp == -1) 6347 *failposp = i; 6348 else if (error == 0 && ret != 0) 6349 error = ret; 6350 } 6351 nmpp++; 6352 fhp++; 6353 } 6354 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na); 6355 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6356 *failposp = mirrorcnt - 1; 6357 else if (error == 0 && ret != 0) 6358 error = ret; 6359 if (error == 0) 6360 error = nfsrv_setextattr(vp, &na, p); 6361 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error); 6362 tdrpc = drpc; 6363 timo = hz / 50; /* Wait for 20msec. */ 6364 if (timo < 1) 6365 timo = 1; 6366 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6367 /* Wait for RPCs on separate threads to complete. */ 6368 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6369 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo); 6370 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6371 *failposp = i; 6372 else if (error == 0 && tdrpc->err != 0) 6373 error = tdrpc->err; 6374 } 6375 free(drpc, M_TEMP); 6376 return (error); 6377 } 6378 6379 /* 6380 * Do a Setattr of an NFSv4 ACL on the DS file. 6381 */ 6382 static int 6383 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6384 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp) 6385 { 6386 struct nfsrv_descript *nd; 6387 nfsv4stateid_t st; 6388 nfsattrbit_t attrbits; 6389 int error; 6390 6391 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n"); 6392 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6393 /* 6394 * Use a stateid where other is an alternating 01010 pattern and 6395 * seqid is 0xffffffff. This value is not defined as special by 6396 * the RFC and is used by the FreeBSD NFS server to indicate an 6397 * MDS->DS proxy operation. 6398 */ 6399 st.other[0] = 0x55555555; 6400 st.other[1] = 0x55555555; 6401 st.other[2] = 0x55555555; 6402 st.seqid = 0xffffffff; 6403 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp), 6404 NULL, NULL, 0, 0, cred); 6405 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6406 NFSZERO_ATTRBIT(&attrbits); 6407 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); 6408 /* 6409 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(), 6410 * so passing in the metadata "vp" will be ok, since it is of 6411 * the same type (VREG). 6412 */ 6413 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL, 6414 NULL, 0, 0, 0, 0, 0, NULL, false, false, false, 0, NULL, false); 6415 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6416 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6417 if (error != 0) { 6418 free(nd, M_TEMP); 6419 return (error); 6420 } 6421 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n", 6422 nd->nd_repstat); 6423 error = nd->nd_repstat; 6424 m_freem(nd->nd_mrep); 6425 free(nd, M_TEMP); 6426 return (error); 6427 } 6428 6429 struct nfsrvsetacldsdorpc { 6430 int done; 6431 int inprog; 6432 struct task tsk; 6433 fhandle_t fh; 6434 struct nfsmount *nmp; 6435 struct vnode *vp; 6436 struct ucred *cred; 6437 NFSPROC_T *p; 6438 struct acl *aclp; 6439 int err; 6440 }; 6441 6442 /* 6443 * Start up the thread that will execute nfsrv_setacldsdorpc(). 6444 */ 6445 static void 6446 start_setacldsdorpc(void *arg, int pending) 6447 { 6448 struct nfsrvsetacldsdorpc *drpc; 6449 6450 drpc = (struct nfsrvsetacldsdorpc *)arg; 6451 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p, 6452 drpc->vp, drpc->nmp, drpc->aclp); 6453 drpc->done = 1; 6454 } 6455 6456 static int 6457 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6458 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp, 6459 int *failposp) 6460 { 6461 struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL; 6462 int error, i, ret, timo; 6463 6464 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n"); 6465 drpc = NULL; 6466 if (mirrorcnt > 1) 6467 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP, 6468 M_WAITOK); 6469 6470 /* 6471 * Do the setattr RPC for every DS, using a separate kernel process 6472 * for every DS except the last one. 6473 */ 6474 error = 0; 6475 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6476 tdrpc->done = 0; 6477 tdrpc->inprog = 0; 6478 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp)); 6479 tdrpc->nmp = *nmpp; 6480 tdrpc->vp = vp; 6481 tdrpc->cred = cred; 6482 tdrpc->p = p; 6483 tdrpc->aclp = aclp; 6484 tdrpc->err = 0; 6485 ret = EIO; 6486 if (nfs_pnfsiothreads != 0) { 6487 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc); 6488 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n", 6489 ret); 6490 } 6491 if (ret != 0) { 6492 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, 6493 aclp); 6494 if (nfsds_failerr(ret) && *failposp == -1) 6495 *failposp = i; 6496 else if (error == 0 && ret != 0) 6497 error = ret; 6498 } 6499 nmpp++; 6500 fhp++; 6501 } 6502 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp); 6503 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1) 6504 *failposp = mirrorcnt - 1; 6505 else if (error == 0 && ret != 0) 6506 error = ret; 6507 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error); 6508 tdrpc = drpc; 6509 timo = hz / 50; /* Wait for 20msec. */ 6510 if (timo < 1) 6511 timo = 1; 6512 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) { 6513 /* Wait for RPCs on separate threads to complete. */ 6514 while (tdrpc->inprog != 0 && tdrpc->done == 0) 6515 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo); 6516 if (nfsds_failerr(tdrpc->err) && *failposp == -1) 6517 *failposp = i; 6518 else if (error == 0 && tdrpc->err != 0) 6519 error = tdrpc->err; 6520 } 6521 free(drpc, M_TEMP); 6522 return (error); 6523 } 6524 6525 /* 6526 * Getattr call to the DS for the attributes that change due to writing. 6527 */ 6528 static int 6529 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p, 6530 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap) 6531 { 6532 struct nfsrv_descript *nd; 6533 int error; 6534 nfsattrbit_t attrbits; 6535 6536 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n"); 6537 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6538 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp, 6539 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6540 NFSZERO_ATTRBIT(&attrbits); 6541 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); 6542 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); 6543 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS); 6544 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); 6545 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED); 6546 (void) nfsrv_putattrbit(nd, &attrbits); 6547 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6548 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6549 if (error != 0) { 6550 free(nd, M_TEMP); 6551 return (error); 6552 } 6553 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n", 6554 nd->nd_repstat); 6555 if (nd->nd_repstat == 0) { 6556 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, 6557 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, 6558 NULL, NULL, NULL, NULL); 6559 /* 6560 * We can only save the updated values in the extended 6561 * attribute if the vp is exclusively locked. 6562 * This should happen when any of the following operations 6563 * occur on the vnode: 6564 * Close, Delegreturn, LayoutCommit, LayoutReturn 6565 * As such, the updated extended attribute should get saved 6566 * before nfsrv_checkdsattr() returns 0 and allows the cached 6567 * attributes to be returned without calling this function. 6568 */ 6569 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) { 6570 error = nfsrv_setextattr(vp, nap, p); 6571 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n", 6572 error); 6573 } 6574 } else 6575 error = nd->nd_repstat; 6576 m_freem(nd->nd_mrep); 6577 free(nd, M_TEMP); 6578 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error); 6579 return (error); 6580 } 6581 6582 /* 6583 * Seek call to a DS. 6584 */ 6585 static int 6586 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp, 6587 struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp) 6588 { 6589 uint32_t *tl; 6590 struct nfsrv_descript *nd; 6591 nfsv4stateid_t st; 6592 int error; 6593 6594 NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n"); 6595 /* 6596 * Use a stateid where other is an alternating 01010 pattern and 6597 * seqid is 0xffffffff. This value is not defined as special by 6598 * the RFC and is used by the FreeBSD NFS server to indicate an 6599 * MDS->DS proxy operation. 6600 */ 6601 st.other[0] = 0x55555555; 6602 st.other[1] = 0x55555555; 6603 st.other[2] = 0x55555555; 6604 st.seqid = 0xffffffff; 6605 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 6606 nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp, 6607 sizeof(fhandle_t), NULL, NULL, 0, 0, cred); 6608 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID); 6609 NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); 6610 txdr_hyper(*offp, tl); tl += 2; 6611 *tl = txdr_unsigned(content); 6612 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, 6613 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); 6614 if (error != 0) { 6615 free(nd, M_TEMP); 6616 return (error); 6617 } 6618 NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat); 6619 if (nd->nd_repstat == 0) { 6620 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER); 6621 if (*tl++ == newnfs_true) 6622 *eofp = true; 6623 else 6624 *eofp = false; 6625 *offp = fxdr_hyper(tl); 6626 } else 6627 error = nd->nd_repstat; 6628 nfsmout: 6629 m_freem(nd->nd_mrep); 6630 free(nd, M_TEMP); 6631 NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error); 6632 return (error); 6633 } 6634 6635 /* 6636 * Get the device id and file handle for a DS file. 6637 */ 6638 int 6639 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp, 6640 fhandle_t *fhp, char *devid) 6641 { 6642 int buflen, error; 6643 char *buf; 6644 6645 buflen = 1024; 6646 buf = malloc(buflen, M_TEMP, M_WAITOK); 6647 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL, 6648 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL); 6649 free(buf, M_TEMP); 6650 return (error); 6651 } 6652 6653 /* 6654 * Do a Lookup against the DS for the filename. 6655 */ 6656 static int 6657 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf, 6658 struct vnode **nvpp, NFSPROC_T *p) 6659 { 6660 struct nameidata named; 6661 struct ucred *tcred; 6662 char *bufp; 6663 u_long *hashp; 6664 struct vnode *nvp; 6665 int error; 6666 6667 tcred = newnfs_getcred(); 6668 named.ni_cnd.cn_nameiop = LOOKUP; 6669 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY; 6670 named.ni_cnd.cn_cred = tcred; 6671 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF; 6672 nfsvno_setpathbuf(&named, &bufp, &hashp); 6673 named.ni_cnd.cn_nameptr = bufp; 6674 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename); 6675 strlcpy(bufp, pf->dsf_filename, NAME_MAX); 6676 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp); 6677 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd); 6678 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error); 6679 NFSFREECRED(tcred); 6680 nfsvno_relpathbuf(&named); 6681 if (error == 0) 6682 *nvpp = nvp; 6683 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error); 6684 return (error); 6685 } 6686 6687 /* 6688 * Set the file handle to the correct one. 6689 */ 6690 static void 6691 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid, 6692 char *fnamep, struct vnode *nvp, NFSPROC_T *p) 6693 { 6694 struct nfsnode *np; 6695 int ret = 0; 6696 6697 np = VTONFS(nvp); 6698 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH); 6699 /* 6700 * We can only do a vn_set_extattr() if the vnode is exclusively 6701 * locked and vn_start_write() has been done. If devid != NULL or 6702 * fnamep != NULL or the vnode is shared locked, vn_start_write() 6703 * may not have been done. 6704 * If not done now, it will be done on a future call. 6705 */ 6706 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) == 6707 LK_EXCLUSIVE) 6708 ret = vn_extattr_set(vp, IO_NODELOCKED, 6709 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf), 6710 (char *)pf, p); 6711 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret); 6712 } 6713 6714 /* 6715 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point 6716 * when the DS has failed. 6717 */ 6718 void 6719 nfsrv_killrpcs(struct nfsmount *nmp) 6720 { 6721 6722 /* 6723 * Call newnfs_nmcancelreqs() to cause 6724 * any RPCs in progress on the mount point to 6725 * fail. 6726 * This will cause any process waiting for an 6727 * RPC to complete while holding a vnode lock 6728 * on the mounted-on vnode (such as "df" or 6729 * a non-forced "umount") to fail. 6730 * This will unlock the mounted-on vnode so 6731 * a forced dismount can succeed. 6732 * The NFSMNTP_CANCELRPCS flag should be set when this function is 6733 * called. 6734 */ 6735 newnfs_nmcancelreqs(nmp); 6736 } 6737 6738 /* 6739 * Sum up the statfs info for each of the DSs, so that the client will 6740 * receive the total for all DSs. 6741 */ 6742 static int 6743 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp) 6744 { 6745 struct statfs *tsf; 6746 struct nfsdevice *ds; 6747 struct vnode **dvpp, **tdvpp, *dvp; 6748 uint64_t tot; 6749 int cnt, error = 0, i; 6750 6751 if (nfsrv_devidcnt <= 0) 6752 return (ENXIO); 6753 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 6754 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 6755 6756 /* Get an array of the dvps for the DSs. */ 6757 tdvpp = dvpp; 6758 i = 0; 6759 NFSDDSLOCK(); 6760 /* First, search for matches for same file system. */ 6761 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6762 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 && 6763 fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) { 6764 if (++i > nfsrv_devidcnt) 6765 break; 6766 *tdvpp++ = ds->nfsdev_dvp; 6767 } 6768 } 6769 /* 6770 * If no matches for same file system, total all servers not assigned 6771 * to a file system. 6772 */ 6773 if (i == 0) { 6774 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 6775 if (ds->nfsdev_nmp != NULL && 6776 ds->nfsdev_mdsisset == 0) { 6777 if (++i > nfsrv_devidcnt) 6778 break; 6779 *tdvpp++ = ds->nfsdev_dvp; 6780 } 6781 } 6782 } 6783 NFSDDSUNLOCK(); 6784 cnt = i; 6785 6786 /* Do a VFS_STATFS() for each of the DSs and sum them up. */ 6787 tdvpp = dvpp; 6788 for (i = 0; i < cnt && error == 0; i++) { 6789 dvp = *tdvpp++; 6790 error = VFS_STATFS(dvp->v_mount, tsf); 6791 if (error == 0) { 6792 if (sf->f_bsize == 0) { 6793 if (tsf->f_bsize > 0) 6794 sf->f_bsize = tsf->f_bsize; 6795 else 6796 sf->f_bsize = 8192; 6797 } 6798 if (tsf->f_blocks > 0) { 6799 if (sf->f_bsize != tsf->f_bsize) { 6800 tot = tsf->f_blocks * tsf->f_bsize; 6801 sf->f_blocks += (tot / sf->f_bsize); 6802 } else 6803 sf->f_blocks += tsf->f_blocks; 6804 } 6805 if (tsf->f_bfree > 0) { 6806 if (sf->f_bsize != tsf->f_bsize) { 6807 tot = tsf->f_bfree * tsf->f_bsize; 6808 sf->f_bfree += (tot / sf->f_bsize); 6809 } else 6810 sf->f_bfree += tsf->f_bfree; 6811 } 6812 if (tsf->f_bavail > 0) { 6813 if (sf->f_bsize != tsf->f_bsize) { 6814 tot = tsf->f_bavail * tsf->f_bsize; 6815 sf->f_bavail += (tot / sf->f_bsize); 6816 } else 6817 sf->f_bavail += tsf->f_bavail; 6818 } 6819 } 6820 } 6821 free(tsf, M_TEMP); 6822 free(dvpp, M_TEMP); 6823 return (error); 6824 } 6825 6826 /* 6827 * Set an NFSv4 acl. 6828 */ 6829 int 6830 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p) 6831 { 6832 int error; 6833 6834 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { 6835 error = NFSERR_ATTRNOTSUPP; 6836 goto out; 6837 } 6838 /* 6839 * With NFSv4 ACLs, chmod(2) may need to add additional entries. 6840 * Make sure it has enough room for that - splitting every entry 6841 * into two and appending "canonical six" entries at the end. 6842 * Cribbed out of kern/vfs_acl.c - Rick M. 6843 */ 6844 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) { 6845 error = NFSERR_ATTRNOTSUPP; 6846 goto out; 6847 } 6848 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 6849 if (error == 0) { 6850 error = nfsrv_dssetacl(vp, aclp, cred, p); 6851 if (error == ENOENT) 6852 error = 0; 6853 } 6854 6855 out: 6856 NFSEXITCODE(error); 6857 return (error); 6858 } 6859 6860 /* 6861 * Seek vnode op call (actually it is a VOP_IOCTL()). 6862 * This function is called with the vnode locked, but unlocks and vrele()s 6863 * the vp before returning. 6864 */ 6865 int 6866 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd, 6867 off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p) 6868 { 6869 struct nfsvattr at; 6870 int error, ret; 6871 6872 ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp"); 6873 /* 6874 * Attempt to seek on a DS file. A return of ENOENT implies 6875 * there is no DS file to seek on. 6876 */ 6877 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL, 6878 NULL, NULL, NULL, NULL, offp, content, eofp); 6879 if (error != ENOENT) { 6880 vput(vp); 6881 return (error); 6882 } 6883 6884 /* 6885 * Do the VOP_IOCTL() call. For the case where *offp == file_size, 6886 * VOP_IOCTL() will return ENXIO. However, the correct reply for 6887 * NFSv4.2 is *eofp == true and error == 0 for this case. 6888 */ 6889 NFSVOPUNLOCK(vp); 6890 error = VOP_IOCTL(vp, cmd, offp, 0, cred, p); 6891 *eofp = false; 6892 if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) { 6893 /* Handle the cases where we might be at EOF. */ 6894 ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL); 6895 if (ret == 0 && *offp == at.na_size) { 6896 *eofp = true; 6897 error = 0; 6898 } 6899 if (ret != 0 && error == 0) 6900 error = ret; 6901 } 6902 vrele(vp); 6903 NFSEXITCODE(error); 6904 return (error); 6905 } 6906 6907 /* 6908 * Allocate vnode op call. 6909 */ 6910 int 6911 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6912 NFSPROC_T *p) 6913 { 6914 int error; 6915 off_t olen; 6916 6917 ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp"); 6918 /* 6919 * Attempt to allocate on a DS file. A return of ENOENT implies 6920 * there is no DS file to allocate on. 6921 */ 6922 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL, 6923 NULL, NULL, NULL, NULL, &len, 0, NULL); 6924 if (error != ENOENT) 6925 return (error); 6926 6927 /* 6928 * Do the actual VOP_ALLOCATE(), looping so long as 6929 * progress is being made, to achieve completion. 6930 */ 6931 do { 6932 olen = len; 6933 error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred); 6934 if (error == 0 && len > 0 && olen > len) 6935 maybe_yield(); 6936 } while (error == 0 && len > 0 && olen > len); 6937 if (error == 0 && len > 0) 6938 error = NFSERR_IO; 6939 NFSEXITCODE(error); 6940 return (error); 6941 } 6942 6943 /* 6944 * Deallocate vnode op call. 6945 */ 6946 int 6947 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred, 6948 NFSPROC_T *p) 6949 { 6950 int error; 6951 off_t olen; 6952 6953 ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp"); 6954 /* 6955 * Attempt to deallocate on a DS file. A return of ENOENT implies 6956 * there is no DS file to deallocate on. 6957 */ 6958 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL, 6959 NULL, NULL, NULL, NULL, &len, 0, NULL); 6960 if (error != ENOENT) 6961 return (error); 6962 6963 /* 6964 * Do the actual VOP_DEALLOCATE(), looping so long as 6965 * progress is being made, to achieve completion. 6966 */ 6967 do { 6968 olen = len; 6969 error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred); 6970 if (error == 0 && len > 0 && olen > len) 6971 maybe_yield(); 6972 } while (error == 0 && len > 0 && olen > len); 6973 if (error == 0 && len > 0) 6974 error = NFSERR_IO; 6975 NFSEXITCODE(error); 6976 return (error); 6977 } 6978 6979 /* 6980 * Get Extended Atribute vnode op into an mbuf list. 6981 */ 6982 int 6983 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp, 6984 struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p, 6985 struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 6986 { 6987 struct iovec *iv; 6988 struct uio io, *uiop = &io; 6989 struct mbuf *m, *m2; 6990 int alen, error, len, tlen; 6991 size_t siz; 6992 6993 /* First, find out the size of the extended attribute. */ 6994 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 6995 &siz, cred, p); 6996 if (error != 0) 6997 return (NFSERR_NOXATTR); 6998 if (siz > maxresp - NFS_MAXXDR) 6999 return (NFSERR_XATTR2BIG); 7000 len = siz; 7001 tlen = NFSM_RNDUP(len); 7002 if (tlen > 0) { 7003 /* 7004 * If cnt > MCLBYTES and the reply will not be saved, use 7005 * ext_pgs mbufs for TLS. 7006 * For NFSv4.0, we do not know for sure if the reply will 7007 * be saved, so do not use ext_pgs mbufs for NFSv4.0. 7008 * Always use ext_pgs mbufs if ND_EXTPG is set. 7009 */ 7010 if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES && 7011 (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS && 7012 (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)) 7013 uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen, 7014 maxextsiz, &m, &m2, &iv); 7015 else 7016 uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, 7017 &iv); 7018 uiop->uio_iov = iv; 7019 } else { 7020 uiop->uio_iovcnt = 0; 7021 uiop->uio_iov = iv = NULL; 7022 m = m2 = NULL; 7023 } 7024 uiop->uio_offset = 0; 7025 uiop->uio_resid = tlen; 7026 uiop->uio_rw = UIO_READ; 7027 uiop->uio_segflg = UIO_SYSSPACE; 7028 uiop->uio_td = p; 7029 #ifdef MAC 7030 error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER, 7031 name); 7032 if (error != 0) 7033 goto out; 7034 #endif 7035 7036 if (tlen > 0) 7037 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 7038 NULL, cred, p); 7039 if (error != 0) 7040 goto out; 7041 if (uiop->uio_resid > 0) { 7042 alen = tlen; 7043 len = tlen - uiop->uio_resid; 7044 tlen = NFSM_RNDUP(len); 7045 if (alen != tlen) 7046 printf("nfsvno_getxattr: weird size read\n"); 7047 if (tlen == 0) { 7048 m_freem(m); 7049 m = m2 = NULL; 7050 } else if (alen != tlen || tlen != len) 7051 m2 = nfsrv_adj(m, alen - tlen, tlen - len); 7052 } 7053 *lenp = len; 7054 *mpp = m; 7055 *mpendp = m2; 7056 7057 out: 7058 if (error != 0) { 7059 if (m != NULL) 7060 m_freem(m); 7061 *lenp = 0; 7062 } 7063 free(iv, M_TEMP); 7064 NFSEXITCODE(error); 7065 return (error); 7066 } 7067 7068 /* 7069 * Set Extended attribute vnode op from an mbuf list. 7070 */ 7071 int 7072 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m, 7073 char *cp, struct ucred *cred, struct thread *p) 7074 { 7075 struct iovec *iv; 7076 struct uio uio, *uiop = &uio; 7077 int cnt, error; 7078 7079 error = 0; 7080 #ifdef MAC 7081 error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER, 7082 name); 7083 #endif 7084 if (error != 0) 7085 goto out; 7086 7087 uiop->uio_rw = UIO_WRITE; 7088 uiop->uio_segflg = UIO_SYSSPACE; 7089 uiop->uio_td = p; 7090 uiop->uio_offset = 0; 7091 uiop->uio_resid = len; 7092 if (len > 0) { 7093 error = nfsrv_createiovecw(len, m, cp, &iv, &cnt); 7094 uiop->uio_iov = iv; 7095 uiop->uio_iovcnt = cnt; 7096 } else { 7097 uiop->uio_iov = iv = NULL; 7098 uiop->uio_iovcnt = 0; 7099 } 7100 if (error == 0) { 7101 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop, 7102 cred, p); 7103 if (error == 0) { 7104 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 7105 nfsvno_updateds(vp, cred, p); 7106 error = VOP_FSYNC(vp, MNT_WAIT, p); 7107 } 7108 free(iv, M_TEMP); 7109 } 7110 7111 out: 7112 NFSEXITCODE(error); 7113 return (error); 7114 } 7115 7116 /* 7117 * For a pNFS server, the DS file's ctime and 7118 * va_filerev (TimeMetadata and Change) needs to 7119 * be updated. This is a hack, but works by 7120 * flipping the S_ISGID bit in va_mode and then 7121 * flipping it back. 7122 * It does result in two MDS->DS RPCs, but creating 7123 * a custom RPC just to do this seems overkill, since 7124 * Setxattr/Rmxattr will not be done that frequently. 7125 * If it fails part way through, that is not too 7126 * serious, since the DS file is never executed. 7127 */ 7128 static void 7129 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p) 7130 { 7131 struct nfsvattr nva; 7132 int ret; 7133 u_short tmode; 7134 7135 ret = VOP_GETATTR(vp, &nva.na_vattr, cred); 7136 if (ret == 0) { 7137 tmode = nva.na_mode; 7138 NFSVNO_ATTRINIT(&nva); 7139 tmode ^= S_ISGID; 7140 NFSVNO_SETATTRVAL(&nva, mode, tmode); 7141 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 7142 NFSPROC_SETATTR, NULL, NULL, NULL, &nva, 7143 NULL, NULL, 0, NULL); 7144 if (ret == 0) { 7145 tmode ^= S_ISGID; 7146 NFSVNO_SETATTRVAL(&nva, mode, tmode); 7147 ret = nfsrv_proxyds(vp, 0, 0, cred, p, 7148 NFSPROC_SETATTR, NULL, NULL, NULL, 7149 &nva, NULL, NULL, 0, NULL); 7150 } 7151 } 7152 } 7153 7154 /* 7155 * Remove Extended attribute vnode op. 7156 */ 7157 int 7158 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name, 7159 struct ucred *cred, struct thread *p) 7160 { 7161 int error; 7162 7163 /* 7164 * Get rid of any delegations. I am not sure why this is required, 7165 * but RFC-8276 says so. 7166 */ 7167 error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p); 7168 if (error != 0) 7169 goto out; 7170 #ifdef MAC 7171 error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER, 7172 name); 7173 if (error != 0) 7174 goto out; 7175 #endif 7176 7177 error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p); 7178 if (error == EOPNOTSUPP) 7179 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL, 7180 cred, p); 7181 if (error == 0) { 7182 if (vp->v_type == VREG && nfsrv_devidcnt != 0) 7183 nfsvno_updateds(vp, cred, p); 7184 error = VOP_FSYNC(vp, MNT_WAIT, p); 7185 } 7186 out: 7187 NFSEXITCODE(error); 7188 return (error); 7189 } 7190 7191 /* 7192 * List Extended Atribute vnode op into an mbuf list. 7193 */ 7194 int 7195 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred, 7196 struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp) 7197 { 7198 struct iovec iv; 7199 struct uio io; 7200 int error; 7201 size_t siz; 7202 7203 *bufp = NULL; 7204 /* First, find out the size of the extended attribute. */ 7205 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred, 7206 p); 7207 if (error != 0) 7208 return (NFSERR_NOXATTR); 7209 if (siz <= cookie) { 7210 *lenp = 0; 7211 *eofp = true; 7212 goto out; 7213 } 7214 if (siz > cookie + *lenp) { 7215 siz = cookie + *lenp; 7216 *eofp = false; 7217 } else 7218 *eofp = true; 7219 /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */ 7220 if (siz > 10 * 1024 * 1024) { 7221 error = NFSERR_XATTR2BIG; 7222 goto out; 7223 } 7224 *bufp = malloc(siz, M_TEMP, M_WAITOK); 7225 iv.iov_base = *bufp; 7226 iv.iov_len = siz; 7227 io.uio_iovcnt = 1; 7228 io.uio_iov = &iv; 7229 io.uio_offset = 0; 7230 io.uio_resid = siz; 7231 io.uio_rw = UIO_READ; 7232 io.uio_segflg = UIO_SYSSPACE; 7233 io.uio_td = p; 7234 #ifdef MAC 7235 error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER); 7236 if (error != 0) 7237 goto out; 7238 #endif 7239 7240 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred, 7241 p); 7242 if (error != 0) 7243 goto out; 7244 if (io.uio_resid > 0) 7245 siz -= io.uio_resid; 7246 *lenp = siz; 7247 7248 out: 7249 if (error != 0) { 7250 free(*bufp, M_TEMP); 7251 *bufp = NULL; 7252 } 7253 NFSEXITCODE(error); 7254 return (error); 7255 } 7256 7257 /* 7258 * Trim trailing data off the mbuf list being built. 7259 */ 7260 void 7261 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos, 7262 int bextpg, int bextpgsiz) 7263 { 7264 vm_page_t pg; 7265 int fullpgsiz, i; 7266 7267 if (mb->m_next != NULL) { 7268 m_freem(mb->m_next); 7269 mb->m_next = NULL; 7270 } 7271 if ((mb->m_flags & M_EXTPG) != 0) { 7272 KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs, 7273 ("nfsm_trimtrailing: bextpg out of range")); 7274 KASSERT(bpos == (char *)(void *) 7275 PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz, 7276 ("nfsm_trimtrailing: bextpgsiz bad!")); 7277 7278 /* First, get rid of any pages after this position. */ 7279 for (i = mb->m_epg_npgs - 1; i > bextpg; i--) { 7280 pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]); 7281 vm_page_unwire_noq(pg); 7282 vm_page_free(pg); 7283 } 7284 mb->m_epg_npgs = bextpg + 1; 7285 if (bextpg == 0) 7286 fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off; 7287 else 7288 fullpgsiz = PAGE_SIZE; 7289 mb->m_epg_last_len = fullpgsiz - bextpgsiz; 7290 mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off); 7291 for (i = 1; i < mb->m_epg_npgs; i++) 7292 mb->m_len += m_epg_pagelen(mb, i, 0); 7293 nd->nd_bextpgsiz = bextpgsiz; 7294 nd->nd_bextpg = bextpg; 7295 } else 7296 mb->m_len = bpos - mtod(mb, char *); 7297 nd->nd_mb = mb; 7298 nd->nd_bpos = bpos; 7299 } 7300 7301 7302 /* 7303 * Check to see if a put file handle operation should test for 7304 * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR. 7305 * When Open is the next operation, NFSERR_WRONGSEC cannot be 7306 * replied for the Open cases that use a component. This can 7307 * be identified by the fact that the file handle's type is VDIR. 7308 */ 7309 bool 7310 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, __enum_uint8(vtype) vtyp) 7311 { 7312 7313 if ((nd->nd_flag & ND_NFSV4) == 0) 7314 return (true); 7315 7316 if ((nd->nd_flag & ND_LASTOP) != 0) 7317 return (false); 7318 7319 if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH || 7320 nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH || 7321 nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP || 7322 nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME) 7323 return (false); 7324 if (nextop == NFSV4OP_OPEN && vtyp == VDIR) 7325 return (false); 7326 return (true); 7327 } 7328 7329 /* 7330 * Check DSs marked no space. 7331 */ 7332 void 7333 nfsrv_checknospc(void) 7334 { 7335 struct statfs *tsf; 7336 struct nfsdevice *ds; 7337 struct vnode **dvpp, **tdvpp, *dvp; 7338 char *devid, *tdevid; 7339 int cnt, error = 0, i; 7340 7341 if (nfsrv_devidcnt <= 0) 7342 return; 7343 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK); 7344 devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK); 7345 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK); 7346 7347 /* Get an array of the dvps for the DSs. */ 7348 tdvpp = dvpp; 7349 tdevid = devid; 7350 i = 0; 7351 NFSDDSLOCK(); 7352 /* First, search for matches for same file system. */ 7353 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7354 if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) { 7355 if (++i > nfsrv_devidcnt) 7356 break; 7357 *tdvpp++ = ds->nfsdev_dvp; 7358 NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID); 7359 tdevid += NFSX_V4DEVICEID; 7360 } 7361 } 7362 NFSDDSUNLOCK(); 7363 7364 /* Do a VFS_STATFS() for each of the DSs and clear no space. */ 7365 cnt = i; 7366 tdvpp = dvpp; 7367 tdevid = devid; 7368 for (i = 0; i < cnt && error == 0; i++) { 7369 dvp = *tdvpp++; 7370 error = VFS_STATFS(dvp->v_mount, tsf); 7371 if (error == 0 && tsf->f_bavail > 0) { 7372 NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n"); 7373 nfsrv_marknospc(tdevid, false); 7374 } 7375 tdevid += NFSX_V4DEVICEID; 7376 } 7377 free(tsf, M_TEMP); 7378 free(dvpp, M_TEMP); 7379 free(devid, M_TEMP); 7380 } 7381 7382 /* 7383 * Initialize everything that needs to be initialized for a vnet. 7384 */ 7385 static void 7386 nfsrv_vnetinit(const void *unused __unused) 7387 { 7388 7389 nfsd_mntinit(); 7390 } 7391 VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY, 7392 nfsrv_vnetinit, NULL); 7393 7394 /* 7395 * Clean up everything that is in a vnet and needs to be 7396 * done when the jail is destroyed or the module unloaded. 7397 */ 7398 static void 7399 nfsrv_cleanup(const void *unused __unused) 7400 { 7401 int i; 7402 7403 NFSD_LOCK(); 7404 if (!NFSD_VNET(nfsrv_mntinited)) { 7405 NFSD_UNLOCK(); 7406 return; 7407 } 7408 NFSD_VNET(nfsrv_mntinited) = false; 7409 NFSD_UNLOCK(); 7410 7411 /* Clean out all NFSv4 state. */ 7412 nfsrv_throwawayallstate(curthread); 7413 7414 /* Clean the NFS server reply cache */ 7415 nfsrvd_cleancache(); 7416 7417 /* Clean out v4root exports. */ 7418 if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) { 7419 vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export); 7420 free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT); 7421 NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL; 7422 } 7423 7424 /* Free up the krpc server pool. */ 7425 if (NFSD_VNET(nfsrvd_pool) != NULL) 7426 svcpool_destroy(NFSD_VNET(nfsrvd_pool)); 7427 7428 /* and get rid of the locks */ 7429 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) { 7430 mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx); 7431 mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx); 7432 } 7433 mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx); 7434 for (i = 0; i < nfsrv_sessionhashsize; i++) 7435 mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx); 7436 lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock); 7437 free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE); 7438 free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE); 7439 free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE); 7440 free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT); 7441 free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE); 7442 free(NFSD_VNET(nfssessionhash), M_NFSDSESSION); 7443 free(NFSD_VNET(nfsv4root_mnt), M_TEMP); 7444 NFSD_VNET(nfsv4root_mnt) = NULL; 7445 } 7446 VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY, 7447 nfsrv_cleanup, NULL); 7448 7449 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); 7450 7451 /* 7452 * Called once to initialize data structures... 7453 */ 7454 static int 7455 nfsd_modevent(module_t mod, int type, void *data) 7456 { 7457 int error = 0, i; 7458 static int loaded = 0; 7459 7460 switch (type) { 7461 case MOD_LOAD: 7462 if (loaded) 7463 goto out; 7464 newnfs_portinit(); 7465 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF); 7466 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF); 7467 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF); 7468 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF); 7469 #ifdef VV_DISABLEDELEG 7470 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; 7471 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; 7472 #endif 7473 nfsd_call_nfsd = nfssvc_nfsd; 7474 loaded = 1; 7475 break; 7476 7477 case MOD_UNLOAD: 7478 if (newnfs_numnfsd != 0) { 7479 error = EBUSY; 7480 break; 7481 } 7482 7483 #ifdef VV_DISABLEDELEG 7484 vn_deleg_ops.vndeleg_recall = NULL; 7485 vn_deleg_ops.vndeleg_disable = NULL; 7486 #endif 7487 nfsd_call_nfsd = NULL; 7488 mtx_destroy(&nfsrc_udpmtx); 7489 mtx_destroy(&nfs_v4root_mutex); 7490 mtx_destroy(&nfsrv_dontlistlock_mtx); 7491 mtx_destroy(&nfsrv_recalllock_mtx); 7492 if (nfslayouthash != NULL) { 7493 for (i = 0; i < nfsrv_layouthashsize; i++) 7494 mtx_destroy(&nfslayouthash[i].mtx); 7495 free(nfslayouthash, M_NFSDSESSION); 7496 } 7497 loaded = 0; 7498 break; 7499 default: 7500 error = EOPNOTSUPP; 7501 break; 7502 } 7503 7504 out: 7505 NFSEXITCODE(error); 7506 return (error); 7507 } 7508 static moduledata_t nfsd_mod = { 7509 "nfsd", 7510 nfsd_modevent, 7511 NULL, 7512 }; 7513 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); 7514 7515 /* So that loader and kldload(2) can find us, wherever we are.. */ 7516 MODULE_VERSION(nfsd, 1); 7517 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); 7518 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); 7519 MODULE_DEPEND(nfsd, krpc, 1, 1, 1); 7520 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); 7521