1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 */ 35 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include "opt_inet.h" 40 #include "opt_inet6.h" 41 42 #include <sys/capsicum.h> 43 44 /* 45 * generally, I don't like #includes inside .h files, but it seems to 46 * be the easiest way to handle the port. 47 */ 48 #include <sys/fail.h> 49 #include <sys/hash.h> 50 #include <sys/sysctl.h> 51 #include <fs/nfs/nfsport.h> 52 #include <netinet/in_fib.h> 53 #include <netinet/if_ether.h> 54 #include <netinet6/ip6_var.h> 55 #include <net/if_types.h> 56 #include <net/route/nhop.h> 57 58 #include <fs/nfsclient/nfs_kdtrace.h> 59 60 #ifdef KDTRACE_HOOKS 61 dtrace_nfsclient_attrcache_flush_probe_func_t 62 dtrace_nfscl_attrcache_flush_done_probe; 63 uint32_t nfscl_attrcache_flush_done_id; 64 65 dtrace_nfsclient_attrcache_get_hit_probe_func_t 66 dtrace_nfscl_attrcache_get_hit_probe; 67 uint32_t nfscl_attrcache_get_hit_id; 68 69 dtrace_nfsclient_attrcache_get_miss_probe_func_t 70 dtrace_nfscl_attrcache_get_miss_probe; 71 uint32_t nfscl_attrcache_get_miss_id; 72 73 dtrace_nfsclient_attrcache_load_probe_func_t 74 dtrace_nfscl_attrcache_load_done_probe; 75 uint32_t nfscl_attrcache_load_done_id; 76 #endif /* !KDTRACE_HOOKS */ 77 78 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 79 extern struct vop_vector newnfs_vnodeops; 80 extern struct vop_vector newnfs_fifoops; 81 extern uma_zone_t newnfsnode_zone; 82 extern struct buf_ops buf_ops_newnfs; 83 extern uma_zone_t ncl_pbuf_zone; 84 extern short nfsv4_cbport; 85 extern int nfscl_enablecallb; 86 extern int nfs_numnfscbd; 87 extern int nfscl_inited; 88 struct mtx ncl_iod_mutex; 89 NFSDLOCKMUTEX; 90 extern struct mtx nfsrv_dslock_mtx; 91 92 extern void (*ncl_call_invalcaches)(struct vnode *); 93 94 SYSCTL_DECL(_vfs_nfs); 95 static int ncl_fileid_maxwarnings = 10; 96 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN, 97 &ncl_fileid_maxwarnings, 0, 98 "Limit fileid corruption warnings; 0 is off; -1 is unlimited"); 99 static volatile int ncl_fileid_nwarnings; 100 101 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *, 102 struct nfsvattr *); 103 104 /* 105 * Comparison function for vfs_hash functions. 106 */ 107 int 108 newnfs_vncmpf(struct vnode *vp, void *arg) 109 { 110 struct nfsfh *nfhp = (struct nfsfh *)arg; 111 struct nfsnode *np = VTONFS(vp); 112 113 if (np->n_fhp->nfh_len != nfhp->nfh_len || 114 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len)) 115 return (1); 116 return (0); 117 } 118 119 /* 120 * Look up a vnode/nfsnode by file handle. 121 * Callers must check for mount points!! 122 * In all cases, a pointer to a 123 * nfsnode structure is returned. 124 * This variant takes a "struct nfsfh *" as second argument and uses 125 * that structure up, either by hanging off the nfsnode or FREEing it. 126 */ 127 int 128 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp, 129 struct componentname *cnp, struct thread *td, struct nfsnode **npp, 130 void *stuff, int lkflags) 131 { 132 struct nfsnode *np, *dnp; 133 struct vnode *vp, *nvp; 134 struct nfsv4node *newd, *oldd; 135 int error; 136 u_int hash; 137 struct nfsmount *nmp; 138 139 nmp = VFSTONFS(mntp); 140 dnp = VTONFS(dvp); 141 *npp = NULL; 142 143 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT); 144 145 error = vfs_hash_get(mntp, hash, lkflags, 146 td, &nvp, newnfs_vncmpf, nfhp); 147 if (error == 0 && nvp != NULL) { 148 /* 149 * I believe there is a slight chance that vgonel() could 150 * get called on this vnode between when NFSVOPLOCK() drops 151 * the VI_LOCK() and vget() acquires it again, so that it 152 * hasn't yet had v_usecount incremented. If this were to 153 * happen, the VIRF_DOOMED flag would be set, so check for 154 * that here. Since we now have the v_usecount incremented, 155 * we should be ok until we vrele() it, if the VIRF_DOOMED 156 * flag isn't set now. 157 */ 158 VI_LOCK(nvp); 159 if (VN_IS_DOOMED(nvp)) { 160 VI_UNLOCK(nvp); 161 vrele(nvp); 162 error = ENOENT; 163 } else { 164 VI_UNLOCK(nvp); 165 } 166 } 167 if (error) { 168 free(nfhp, M_NFSFH); 169 return (error); 170 } 171 if (nvp != NULL) { 172 np = VTONFS(nvp); 173 /* 174 * For NFSv4, check to see if it is the same name and 175 * replace the name, if it is different. 176 */ 177 oldd = newd = NULL; 178 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL && 179 nvp->v_type == VREG && 180 (np->n_v4->n4_namelen != cnp->cn_namelen || 181 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 182 cnp->cn_namelen) || 183 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 184 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 185 dnp->n_fhp->nfh_len))) { 186 newd = malloc( 187 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + 188 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); 189 NFSLOCKNODE(np); 190 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG 191 && (np->n_v4->n4_namelen != cnp->cn_namelen || 192 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 193 cnp->cn_namelen) || 194 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 195 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 196 dnp->n_fhp->nfh_len))) { 197 oldd = np->n_v4; 198 np->n_v4 = newd; 199 newd = NULL; 200 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 201 np->n_v4->n4_namelen = cnp->cn_namelen; 202 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 203 dnp->n_fhp->nfh_len); 204 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 205 cnp->cn_namelen); 206 } 207 NFSUNLOCKNODE(np); 208 } 209 if (newd != NULL) 210 free(newd, M_NFSV4NODE); 211 if (oldd != NULL) 212 free(oldd, M_NFSV4NODE); 213 *npp = np; 214 free(nfhp, M_NFSFH); 215 return (0); 216 } 217 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO); 218 219 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp); 220 if (error) { 221 uma_zfree(newnfsnode_zone, np); 222 free(nfhp, M_NFSFH); 223 return (error); 224 } 225 vp = nvp; 226 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0")); 227 vp->v_bufobj.bo_ops = &buf_ops_newnfs; 228 vp->v_data = np; 229 np->n_vnode = vp; 230 /* 231 * Initialize the mutex even if the vnode is going to be a loser. 232 * This simplifies the logic in reclaim, which can then unconditionally 233 * destroy the mutex (in the case of the loser, or if hash_insert 234 * happened to return an error no special casing is needed). 235 */ 236 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK); 237 lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE | 238 LK_CANRECURSE); 239 240 /* 241 * Are we getting the root? If so, make sure the vnode flags 242 * are correct 243 */ 244 if ((nfhp->nfh_len == nmp->nm_fhsize) && 245 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) { 246 if (vp->v_type == VNON) 247 vp->v_type = VDIR; 248 vp->v_vflag |= VV_ROOT; 249 } 250 251 vp->v_vflag |= VV_VMSIZEVNLOCK; 252 253 np->n_fhp = nfhp; 254 /* 255 * For NFSv4, we have to attach the directory file handle and 256 * file name, so that Open Ops can be done later. 257 */ 258 if (nmp->nm_flag & NFSMNT_NFSV4) { 259 np->n_v4 = malloc(sizeof (struct nfsv4node) 260 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE, 261 M_WAITOK); 262 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 263 np->n_v4->n4_namelen = cnp->cn_namelen; 264 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 265 dnp->n_fhp->nfh_len); 266 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 267 cnp->cn_namelen); 268 } else { 269 np->n_v4 = NULL; 270 } 271 272 /* 273 * NFS supports recursive and shared locking. 274 */ 275 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL); 276 VN_LOCK_AREC(vp); 277 VN_LOCK_ASHARE(vp); 278 error = insmntque(vp, mntp); 279 if (error != 0) { 280 *npp = NULL; 281 mtx_destroy(&np->n_mtx); 282 lockdestroy(&np->n_excl); 283 free(nfhp, M_NFSFH); 284 if (np->n_v4 != NULL) 285 free(np->n_v4, M_NFSV4NODE); 286 uma_zfree(newnfsnode_zone, np); 287 return (error); 288 } 289 error = vfs_hash_insert(vp, hash, lkflags, 290 td, &nvp, newnfs_vncmpf, nfhp); 291 if (error) 292 return (error); 293 if (nvp != NULL) { 294 *npp = VTONFS(nvp); 295 /* vfs_hash_insert() vput()'s the losing vnode */ 296 return (0); 297 } 298 *npp = np; 299 300 return (0); 301 } 302 303 /* 304 * Another variant of nfs_nget(). This one is only used by reopen. It 305 * takes almost the same args as nfs_nget(), but only succeeds if an entry 306 * exists in the cache. (Since files should already be "open" with a 307 * vnode ref cnt on the node when reopen calls this, it should always 308 * succeed.) 309 * Also, don't get a vnode lock, since it may already be locked by some 310 * other process that is handling it. This is ok, since all other threads 311 * on the client are blocked by the nfsc_lock being exclusively held by the 312 * caller of this function. 313 */ 314 int 315 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize, 316 struct thread *td, struct nfsnode **npp) 317 { 318 struct vnode *nvp; 319 u_int hash; 320 struct nfsfh *nfhp; 321 int error; 322 323 *npp = NULL; 324 /* For forced dismounts, just return error. */ 325 if (NFSCL_FORCEDISM(mntp)) 326 return (EINTR); 327 nfhp = malloc(sizeof (struct nfsfh) + fhsize, 328 M_NFSFH, M_WAITOK); 329 bcopy(fhp, &nfhp->nfh_fh[0], fhsize); 330 nfhp->nfh_len = fhsize; 331 332 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT); 333 334 /* 335 * First, try to get the vnode locked, but don't block for the lock. 336 */ 337 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp, 338 newnfs_vncmpf, nfhp); 339 if (error == 0 && nvp != NULL) { 340 NFSVOPUNLOCK(nvp); 341 } else if (error == EBUSY) { 342 /* 343 * It is safe so long as a vflush() with 344 * FORCECLOSE has not been done. Since the Renew thread is 345 * stopped and the MNTK_UNMOUNTF flag is set before doing 346 * a vflush() with FORCECLOSE, we should be ok here. 347 */ 348 if (NFSCL_FORCEDISM(mntp)) 349 error = EINTR; 350 else { 351 vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp); 352 if (nvp == NULL) { 353 error = ENOENT; 354 } else if (VN_IS_DOOMED(nvp)) { 355 error = ENOENT; 356 vrele(nvp); 357 } else { 358 error = 0; 359 } 360 } 361 } 362 free(nfhp, M_NFSFH); 363 if (error) 364 return (error); 365 if (nvp != NULL) { 366 *npp = VTONFS(nvp); 367 return (0); 368 } 369 return (EINVAL); 370 } 371 372 static void 373 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap, 374 struct nfsvattr *newnap) 375 { 376 int off; 377 378 if (ncl_fileid_maxwarnings >= 0 && 379 ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) 380 return; 381 off = 0; 382 if (ncl_fileid_maxwarnings >= 0) { 383 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) 384 off = 1; 385 } 386 387 printf("newnfs: server '%s' error: fileid changed. " 388 "fsid %jx:%jx: expected fileid %#jx, got %#jx. " 389 "(BROKEN NFS SERVER OR MIDDLEWARE)\n", 390 nmp->nm_com.nmcom_hostname, 391 (uintmax_t)nmp->nm_fsid[0], 392 (uintmax_t)nmp->nm_fsid[1], 393 (uintmax_t)oldnap->na_fileid, 394 (uintmax_t)newnap->na_fileid); 395 396 if (off) 397 printf("newnfs: Logged %d times about fileid corruption; " 398 "going quiet to avoid spamming logs excessively. (Limit " 399 "is: %d).\n", ncl_fileid_nwarnings, 400 ncl_fileid_maxwarnings); 401 } 402 403 void 404 ncl_copy_vattr(struct vattr *dst, struct vattr *src) 405 { 406 dst->va_type = src->va_type; 407 dst->va_mode = src->va_mode; 408 dst->va_nlink = src->va_nlink; 409 dst->va_uid = src->va_uid; 410 dst->va_gid = src->va_gid; 411 dst->va_fsid = src->va_fsid; 412 dst->va_fileid = src->va_fileid; 413 dst->va_size = src->va_size; 414 dst->va_blocksize = src->va_blocksize; 415 dst->va_atime = src->va_atime; 416 dst->va_mtime = src->va_mtime; 417 dst->va_ctime = src->va_ctime; 418 dst->va_birthtime = src->va_birthtime; 419 dst->va_gen = src->va_gen; 420 dst->va_flags = src->va_flags; 421 dst->va_rdev = src->va_rdev; 422 dst->va_bytes = src->va_bytes; 423 dst->va_filerev = src->va_filerev; 424 } 425 426 /* 427 * Load the attribute cache (that lives in the nfsnode entry) with 428 * the attributes of the second argument and 429 * Iff vaper not NULL 430 * copy the attributes to *vaper 431 * Similar to nfs_loadattrcache(), except the attributes are passed in 432 * instead of being parsed out of the mbuf list. 433 */ 434 int 435 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper, 436 void *stuff, int writeattr, int dontshrink) 437 { 438 struct vnode *vp = *vpp; 439 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper; 440 struct nfsnode *np; 441 struct nfsmount *nmp; 442 struct timespec mtime_save; 443 int error, force_fid_err; 444 445 error = 0; 446 447 /* 448 * If v_type == VNON it is a new node, so fill in the v_type, 449 * n_mtime fields. Check to see if it represents a special 450 * device, and if so, check for a possible alias. Once the 451 * correct vnode has been obtained, fill in the rest of the 452 * information. 453 */ 454 np = VTONFS(vp); 455 NFSLOCKNODE(np); 456 if (vp->v_type != nvap->va_type) { 457 vp->v_type = nvap->va_type; 458 if (vp->v_type == VFIFO) 459 vp->v_op = &newnfs_fifoops; 460 np->n_mtime = nvap->va_mtime; 461 } 462 nmp = VFSTONFS(vp->v_mount); 463 vap = &np->n_vattr.na_vattr; 464 mtime_save = vap->va_mtime; 465 if (writeattr) { 466 np->n_vattr.na_filerev = nap->na_filerev; 467 np->n_vattr.na_size = nap->na_size; 468 np->n_vattr.na_mtime = nap->na_mtime; 469 np->n_vattr.na_ctime = nap->na_ctime; 470 np->n_vattr.na_btime = nap->na_btime; 471 np->n_vattr.na_fsid = nap->na_fsid; 472 np->n_vattr.na_mode = nap->na_mode; 473 } else { 474 force_fid_err = 0; 475 KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning, 476 force_fid_err); 477 /* 478 * BROKEN NFS SERVER OR MIDDLEWARE 479 * 480 * Certain NFS servers (certain old proprietary filers ca. 481 * 2006) or broken middleboxes (e.g. WAN accelerator products) 482 * will respond to GETATTR requests with results for a 483 * different fileid. 484 * 485 * The WAN accelerator we've observed not only serves stale 486 * cache results for a given file, it also occasionally serves 487 * results for wholly different files. This causes surprising 488 * problems; for example the cached size attribute of a file 489 * may truncate down and then back up, resulting in zero 490 * regions in file contents read by applications. We observed 491 * this reliably with Clang and .c files during parallel build. 492 * A pcap revealed packet fragmentation and GETATTR RPC 493 * responses with wholly wrong fileids. 494 */ 495 if ((np->n_vattr.na_fileid != 0 && 496 np->n_vattr.na_fileid != nap->na_fileid) || 497 force_fid_err) { 498 nfscl_warn_fileid(nmp, &np->n_vattr, nap); 499 error = EIDRM; 500 goto out; 501 } 502 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr, 503 sizeof (struct nfsvattr)); 504 } 505 506 /* 507 * For NFSv4, if the node's fsid is not equal to the mount point's 508 * fsid, return the low order 32bits of the node's fsid. This 509 * allows getcwd(3) to work. There is a chance that the fsid might 510 * be the same as a local fs, but since this is in an NFS mount 511 * point, I don't think that will cause any problems? 512 */ 513 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) && 514 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] || 515 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) { 516 /* 517 * va_fsid needs to be set to some value derived from 518 * np->n_vattr.na_filesid that is not equal 519 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes 520 * from the value used for the top level server volume 521 * in the mounted subtree. 522 */ 523 vn_fsid(vp, vap); 524 if ((uint32_t)vap->va_fsid == np->n_vattr.na_filesid[0]) 525 vap->va_fsid = hash32_buf( 526 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0); 527 } else 528 vn_fsid(vp, vap); 529 np->n_attrstamp = time_second; 530 if (vap->va_size != np->n_size) { 531 if (vap->va_type == VREG) { 532 if (dontshrink && vap->va_size < np->n_size) { 533 /* 534 * We've been told not to shrink the file; 535 * zero np->n_attrstamp to indicate that 536 * the attributes are stale. 537 */ 538 vap->va_size = np->n_size; 539 np->n_attrstamp = 0; 540 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 541 } else if (np->n_flag & NMODIFIED) { 542 /* 543 * We've modified the file: Use the larger 544 * of our size, and the server's size. 545 */ 546 if (vap->va_size < np->n_size) { 547 vap->va_size = np->n_size; 548 } else { 549 np->n_size = vap->va_size; 550 np->n_flag |= NSIZECHANGED; 551 } 552 } else { 553 np->n_size = vap->va_size; 554 np->n_flag |= NSIZECHANGED; 555 } 556 } else { 557 np->n_size = vap->va_size; 558 } 559 } 560 /* 561 * The following checks are added to prevent a race between (say) 562 * a READDIR+ and a WRITE. 563 * READDIR+, WRITE requests sent out. 564 * READDIR+ resp, WRITE resp received on client. 565 * However, the WRITE resp was handled before the READDIR+ resp 566 * causing the post op attrs from the write to be loaded first 567 * and the attrs from the READDIR+ to be loaded later. If this 568 * happens, we have stale attrs loaded into the attrcache. 569 * We detect this by for the mtime moving back. We invalidate the 570 * attrcache when this happens. 571 */ 572 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) { 573 /* Size changed or mtime went backwards */ 574 np->n_attrstamp = 0; 575 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 576 } 577 if (vaper != NULL) { 578 ncl_copy_vattr(vaper, vap); 579 if (np->n_flag & NCHG) { 580 if (np->n_flag & NACC) 581 vaper->va_atime = np->n_atim; 582 if (np->n_flag & NUPD) 583 vaper->va_mtime = np->n_mtim; 584 } 585 } 586 587 out: 588 #ifdef KDTRACE_HOOKS 589 if (np->n_attrstamp != 0) 590 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error); 591 #endif 592 (void)ncl_pager_setsize(vp, NULL); 593 return (error); 594 } 595 596 /* 597 * Call vnode_pager_setsize() if the size of the node changed, as 598 * recorded in nfsnode vs. v_object, or delay the call if notifying 599 * the pager is not possible at the moment. 600 * 601 * If nsizep is non-NULL, the call is delayed and the new node size is 602 * provided. Caller should itself call vnode_pager_setsize() if 603 * function returned true. If nsizep is NULL, function tries to call 604 * vnode_pager_setsize() itself if needed and possible, and the nfs 605 * node is unlocked unconditionally, the return value is not useful. 606 */ 607 bool 608 ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep) 609 { 610 struct nfsnode *np; 611 vm_object_t object; 612 struct vattr *vap; 613 u_quad_t nsize; 614 bool setnsize; 615 616 np = VTONFS(vp); 617 NFSASSERTNODE(np); 618 619 vap = &np->n_vattr.na_vattr; 620 nsize = vap->va_size; 621 object = vp->v_object; 622 setnsize = false; 623 624 if (object != NULL && nsize != object->un_pager.vnp.vnp_size) { 625 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && 626 (curthread->td_pflags2 & TDP2_SBPAGES) == 0) 627 setnsize = true; 628 else 629 np->n_flag |= NVNSETSZSKIP; 630 } 631 if (nsizep == NULL) { 632 NFSUNLOCKNODE(np); 633 if (setnsize) 634 vnode_pager_setsize(vp, nsize); 635 setnsize = false; 636 } else { 637 *nsizep = nsize; 638 } 639 return (setnsize); 640 } 641 642 /* 643 * Fill in the client id name. For these bytes: 644 * 1 - they must be unique 645 * 2 - they should be persistent across client reboots 646 * 1 is more critical than 2 647 * Use the mount point's unique id plus either the uuid or, if that 648 * isn't set, random junk. 649 */ 650 void 651 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen) 652 { 653 int uuidlen; 654 655 /* 656 * First, put in the 64bit mount point identifier. 657 */ 658 if (idlen >= sizeof (u_int64_t)) { 659 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t)); 660 cp += sizeof (u_int64_t); 661 idlen -= sizeof (u_int64_t); 662 } 663 664 /* 665 * If uuid is non-zero length, use it. 666 */ 667 uuidlen = strlen(uuid); 668 if (uuidlen > 0 && idlen >= uuidlen) { 669 NFSBCOPY(uuid, cp, uuidlen); 670 cp += uuidlen; 671 idlen -= uuidlen; 672 } 673 674 /* 675 * This only normally happens if the uuid isn't set. 676 */ 677 while (idlen > 0) { 678 *cp++ = (u_int8_t)(arc4random() % 256); 679 idlen--; 680 } 681 } 682 683 /* 684 * Fill in a lock owner name. For now, pid + the process's creation time. 685 */ 686 void 687 nfscl_filllockowner(void *id, u_int8_t *cp, int flags) 688 { 689 union { 690 u_int32_t lval; 691 u_int8_t cval[4]; 692 } tl; 693 struct proc *p; 694 695 if (id == NULL) { 696 /* Return the single open_owner of all 0 bytes. */ 697 bzero(cp, NFSV4CL_LOCKNAMELEN); 698 return; 699 } 700 if ((flags & F_POSIX) != 0) { 701 p = (struct proc *)id; 702 tl.lval = p->p_pid; 703 *cp++ = tl.cval[0]; 704 *cp++ = tl.cval[1]; 705 *cp++ = tl.cval[2]; 706 *cp++ = tl.cval[3]; 707 tl.lval = p->p_stats->p_start.tv_sec; 708 *cp++ = tl.cval[0]; 709 *cp++ = tl.cval[1]; 710 *cp++ = tl.cval[2]; 711 *cp++ = tl.cval[3]; 712 tl.lval = p->p_stats->p_start.tv_usec; 713 *cp++ = tl.cval[0]; 714 *cp++ = tl.cval[1]; 715 *cp++ = tl.cval[2]; 716 *cp = tl.cval[3]; 717 } else if ((flags & F_FLOCK) != 0) { 718 bcopy(&id, cp, sizeof(id)); 719 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id)); 720 } else { 721 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n"); 722 bzero(cp, NFSV4CL_LOCKNAMELEN); 723 } 724 } 725 726 /* 727 * Find the parent process for the thread passed in as an argument. 728 * If none exists, return NULL, otherwise return a thread for the parent. 729 * (Can be any of the threads, since it is only used for td->td_proc.) 730 */ 731 NFSPROC_T * 732 nfscl_getparent(struct thread *td) 733 { 734 struct proc *p; 735 struct thread *ptd; 736 737 if (td == NULL) 738 return (NULL); 739 p = td->td_proc; 740 if (p->p_pid == 0) 741 return (NULL); 742 p = p->p_pptr; 743 if (p == NULL) 744 return (NULL); 745 ptd = TAILQ_FIRST(&p->p_threads); 746 return (ptd); 747 } 748 749 /* 750 * Start up the renew kernel thread. 751 */ 752 static void 753 start_nfscl(void *arg) 754 { 755 struct nfsclclient *clp; 756 struct thread *td; 757 758 clp = (struct nfsclclient *)arg; 759 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads); 760 nfscl_renewthread(clp, td); 761 kproc_exit(0); 762 } 763 764 void 765 nfscl_start_renewthread(struct nfsclclient *clp) 766 { 767 768 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0, 769 "nfscl"); 770 } 771 772 /* 773 * Handle wcc_data. 774 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr 775 * as the first Op after PutFH. 776 * (For NFSv4, the postop attributes are after the Op, so they can't be 777 * parsed here. A separate call to nfscl_postop_attr() is required.) 778 */ 779 int 780 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp, 781 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff) 782 { 783 u_int32_t *tl; 784 struct nfsnode *np = VTONFS(vp); 785 struct nfsvattr nfsva; 786 int error = 0; 787 788 if (wccflagp != NULL) 789 *wccflagp = 0; 790 if (nd->nd_flag & ND_NFSV3) { 791 *flagp = 0; 792 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 793 if (*tl == newnfs_true) { 794 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 795 if (wccflagp != NULL) { 796 NFSLOCKNODE(np); 797 *wccflagp = (np->n_mtime.tv_sec == 798 fxdr_unsigned(u_int32_t, *(tl + 2)) && 799 np->n_mtime.tv_nsec == 800 fxdr_unsigned(u_int32_t, *(tl + 3))); 801 NFSUNLOCKNODE(np); 802 } 803 } 804 error = nfscl_postop_attr(nd, nap, flagp, stuff); 805 if (wccflagp != NULL && *flagp == 0) 806 *wccflagp = 0; 807 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) 808 == (ND_NFSV4 | ND_V4WCCATTR)) { 809 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, 810 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, 811 NULL, NULL, NULL, NULL, NULL); 812 if (error) 813 return (error); 814 /* 815 * Get rid of Op# and status for next op. 816 */ 817 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 818 if (*++tl) 819 nd->nd_flag |= ND_NOMOREDATA; 820 if (wccflagp != NULL && 821 nfsva.na_vattr.va_mtime.tv_sec != 0) { 822 NFSLOCKNODE(np); 823 *wccflagp = (np->n_mtime.tv_sec == 824 nfsva.na_vattr.va_mtime.tv_sec && 825 np->n_mtime.tv_nsec == 826 nfsva.na_vattr.va_mtime.tv_sec); 827 NFSUNLOCKNODE(np); 828 } 829 } 830 nfsmout: 831 return (error); 832 } 833 834 /* 835 * Get postop attributes. 836 */ 837 int 838 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp, 839 void *stuff) 840 { 841 u_int32_t *tl; 842 int error = 0; 843 844 *retp = 0; 845 if (nd->nd_flag & ND_NOMOREDATA) 846 return (error); 847 if (nd->nd_flag & ND_NFSV3) { 848 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 849 *retp = fxdr_unsigned(int, *tl); 850 } else if (nd->nd_flag & ND_NFSV4) { 851 /* 852 * For NFSv4, the postop attr are at the end, so no point 853 * in looking if nd_repstat != 0. 854 */ 855 if (!nd->nd_repstat) { 856 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 857 if (*(tl + 1)) 858 /* should never happen since nd_repstat != 0 */ 859 nd->nd_flag |= ND_NOMOREDATA; 860 else 861 *retp = 1; 862 } 863 } else if (!nd->nd_repstat) { 864 /* For NFSv2, the attributes are here iff nd_repstat == 0 */ 865 *retp = 1; 866 } 867 if (*retp) { 868 error = nfsm_loadattr(nd, nap); 869 if (error) 870 *retp = 0; 871 } 872 nfsmout: 873 return (error); 874 } 875 876 /* 877 * nfscl_request() - mostly a wrapper for newnfs_request(). 878 */ 879 int 880 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p, 881 struct ucred *cred, void *stuff) 882 { 883 int ret, vers; 884 struct nfsmount *nmp; 885 886 nmp = VFSTONFS(vp->v_mount); 887 if (nd->nd_flag & ND_NFSV4) 888 vers = NFS_VER4; 889 else if (nd->nd_flag & ND_NFSV3) 890 vers = NFS_VER3; 891 else 892 vers = NFS_VER2; 893 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred, 894 NFS_PROG, vers, NULL, 1, NULL, NULL); 895 return (ret); 896 } 897 898 /* 899 * fill in this bsden's variant of statfs using nfsstatfs. 900 */ 901 void 902 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs) 903 { 904 struct statfs *sbp = (struct statfs *)statfs; 905 906 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) { 907 sbp->f_bsize = NFS_FABLKSIZE; 908 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE; 909 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE; 910 /* 911 * Although sf_abytes is uint64_t and f_bavail is int64_t, 912 * the value after dividing by NFS_FABLKSIZE is small 913 * enough that it will fit in 63bits, so it is ok to 914 * assign it to f_bavail without fear that it will become 915 * negative. 916 */ 917 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE; 918 sbp->f_files = sfp->sf_tfiles; 919 /* Since f_ffree is int64_t, clip it to 63bits. */ 920 if (sfp->sf_ffiles > INT64_MAX) 921 sbp->f_ffree = INT64_MAX; 922 else 923 sbp->f_ffree = sfp->sf_ffiles; 924 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) { 925 /* 926 * The type casts to (int32_t) ensure that this code is 927 * compatible with the old NFS client, in that it will 928 * propagate bit31 to the high order bits. This may or may 929 * not be correct for NFSv2, but since it is a legacy 930 * environment, I'd rather retain backwards compatibility. 931 */ 932 sbp->f_bsize = (int32_t)sfp->sf_bsize; 933 sbp->f_blocks = (int32_t)sfp->sf_blocks; 934 sbp->f_bfree = (int32_t)sfp->sf_bfree; 935 sbp->f_bavail = (int32_t)sfp->sf_bavail; 936 sbp->f_files = 0; 937 sbp->f_ffree = 0; 938 } 939 } 940 941 /* 942 * Use the fsinfo stuff to update the mount point. 943 */ 944 void 945 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp) 946 { 947 948 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) && 949 fsp->fs_wtpref >= NFS_FABLKSIZE) 950 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) & 951 ~(NFS_FABLKSIZE - 1); 952 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) { 953 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1); 954 if (nmp->nm_wsize == 0) 955 nmp->nm_wsize = fsp->fs_wtmax; 956 } 957 if (nmp->nm_wsize < NFS_FABLKSIZE) 958 nmp->nm_wsize = NFS_FABLKSIZE; 959 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) && 960 fsp->fs_rtpref >= NFS_FABLKSIZE) 961 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) & 962 ~(NFS_FABLKSIZE - 1); 963 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) { 964 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1); 965 if (nmp->nm_rsize == 0) 966 nmp->nm_rsize = fsp->fs_rtmax; 967 } 968 if (nmp->nm_rsize < NFS_FABLKSIZE) 969 nmp->nm_rsize = NFS_FABLKSIZE; 970 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize) 971 && fsp->fs_dtpref >= NFS_DIRBLKSIZ) 972 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) & 973 ~(NFS_DIRBLKSIZ - 1); 974 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) { 975 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1); 976 if (nmp->nm_readdirsize == 0) 977 nmp->nm_readdirsize = fsp->fs_rtmax; 978 } 979 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ) 980 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 981 if (fsp->fs_maxfilesize > 0 && 982 fsp->fs_maxfilesize < nmp->nm_maxfilesize) 983 nmp->nm_maxfilesize = fsp->fs_maxfilesize; 984 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp); 985 nmp->nm_state |= NFSSTA_GOTFSINFO; 986 } 987 988 /* 989 * Lookups source address which should be used to communicate with 990 * @nmp and stores it inside @pdst. 991 * 992 * Returns 0 on success. 993 */ 994 u_int8_t * 995 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p) 996 { 997 #if defined(INET6) || defined(INET) 998 int fibnum; 999 1000 fibnum = curthread->td_proc->p_fibnum; 1001 #endif 1002 #ifdef INET 1003 if (nmp->nm_nam->sa_family == AF_INET) { 1004 struct epoch_tracker et; 1005 struct nhop_object *nh; 1006 struct sockaddr_in *sin; 1007 struct in_addr addr = {}; 1008 1009 sin = (struct sockaddr_in *)nmp->nm_nam; 1010 NET_EPOCH_ENTER(et); 1011 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); 1012 nh = fib4_lookup(fibnum, sin->sin_addr, 0, NHR_NONE, 0); 1013 CURVNET_RESTORE(); 1014 if (nh != NULL) 1015 addr = IA_SIN(ifatoia(nh->nh_ifa))->sin_addr; 1016 NET_EPOCH_EXIT(et); 1017 if (nh == NULL) 1018 return (NULL); 1019 1020 if (IN_LOOPBACK(ntohl(addr.s_addr))) { 1021 /* Ignore loopback addresses */ 1022 return (NULL); 1023 } 1024 1025 *isinet6p = 0; 1026 *((struct in_addr *)paddr) = addr; 1027 1028 return (u_int8_t *)paddr; 1029 } 1030 #endif 1031 #ifdef INET6 1032 if (nmp->nm_nam->sa_family == AF_INET6) { 1033 struct epoch_tracker et; 1034 struct sockaddr_in6 *sin6; 1035 int error; 1036 1037 sin6 = (struct sockaddr_in6 *)nmp->nm_nam; 1038 1039 NET_EPOCH_ENTER(et); 1040 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); 1041 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr, 1042 sin6->sin6_scope_id, NULL, paddr, NULL); 1043 CURVNET_RESTORE(); 1044 NET_EPOCH_EXIT(et); 1045 if (error != 0) 1046 return (NULL); 1047 1048 if (IN6_IS_ADDR_LOOPBACK(paddr)) 1049 return (NULL); 1050 1051 /* Scope is embedded in */ 1052 *isinet6p = 1; 1053 1054 return (u_int8_t *)paddr; 1055 } 1056 #endif 1057 return (NULL); 1058 } 1059 1060 /* 1061 * Copy NFS uid, gids from the cred structure. 1062 */ 1063 void 1064 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr) 1065 { 1066 int i; 1067 1068 KASSERT(cr->cr_ngroups >= 0, 1069 ("newnfs_copyincred: negative cr_ngroups")); 1070 nfscr->nfsc_uid = cr->cr_uid; 1071 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1); 1072 for (i = 0; i < nfscr->nfsc_ngroups; i++) 1073 nfscr->nfsc_groups[i] = cr->cr_groups[i]; 1074 } 1075 1076 /* 1077 * Do any client specific initialization. 1078 */ 1079 void 1080 nfscl_init(void) 1081 { 1082 static int inited = 0; 1083 1084 if (inited) 1085 return; 1086 inited = 1; 1087 nfscl_inited = 1; 1088 ncl_pbuf_zone = pbuf_zsecond_create("nfspbuf", nswbuf / 2); 1089 } 1090 1091 /* 1092 * Check each of the attributes to be set, to ensure they aren't already 1093 * the correct value. Disable setting ones already correct. 1094 */ 1095 int 1096 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap) 1097 { 1098 1099 if (vap->va_mode != (mode_t)VNOVAL) { 1100 if (vap->va_mode == nvap->na_mode) 1101 vap->va_mode = (mode_t)VNOVAL; 1102 } 1103 if (vap->va_uid != (uid_t)VNOVAL) { 1104 if (vap->va_uid == nvap->na_uid) 1105 vap->va_uid = (uid_t)VNOVAL; 1106 } 1107 if (vap->va_gid != (gid_t)VNOVAL) { 1108 if (vap->va_gid == nvap->na_gid) 1109 vap->va_gid = (gid_t)VNOVAL; 1110 } 1111 if (vap->va_size != VNOVAL) { 1112 if (vap->va_size == nvap->na_size) 1113 vap->va_size = VNOVAL; 1114 } 1115 1116 /* 1117 * We are normally called with only a partially initialized 1118 * VAP. Since the NFSv3 spec says that server may use the 1119 * file attributes to store the verifier, the spec requires 1120 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1121 * in atime, but we can't really assume that all servers will 1122 * so we ensure that our SETATTR sets both atime and mtime. 1123 * Set the VA_UTIMES_NULL flag for this case, so that 1124 * the server's time will be used. This is needed to 1125 * work around a bug in some Solaris servers, where 1126 * setting the time TOCLIENT causes the Setattr RPC 1127 * to return NFS_OK, but not set va_mode. 1128 */ 1129 if (vap->va_mtime.tv_sec == VNOVAL) { 1130 vfs_timestamp(&vap->va_mtime); 1131 vap->va_vaflags |= VA_UTIMES_NULL; 1132 } 1133 if (vap->va_atime.tv_sec == VNOVAL) 1134 vap->va_atime = vap->va_mtime; 1135 return (1); 1136 } 1137 1138 /* 1139 * Map nfsv4 errors to errno.h errors. 1140 * The uid and gid arguments are only used for NFSERR_BADOWNER and that 1141 * error should only be returned for the Open, Create and Setattr Ops. 1142 * As such, most calls can just pass in 0 for those arguments. 1143 */ 1144 int 1145 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid) 1146 { 1147 struct proc *p; 1148 1149 if (error < 10000 || error >= NFSERR_STALEWRITEVERF) 1150 return (error); 1151 if (td != NULL) 1152 p = td->td_proc; 1153 else 1154 p = NULL; 1155 switch (error) { 1156 case NFSERR_BADOWNER: 1157 tprintf(p, LOG_INFO, 1158 "No name and/or group mapping for uid,gid:(%d,%d)\n", 1159 uid, gid); 1160 return (EPERM); 1161 case NFSERR_BADNAME: 1162 case NFSERR_BADCHAR: 1163 printf("nfsv4 char/name not handled by server\n"); 1164 return (ENOENT); 1165 case NFSERR_STALECLIENTID: 1166 case NFSERR_STALESTATEID: 1167 case NFSERR_EXPIRED: 1168 case NFSERR_BADSTATEID: 1169 case NFSERR_BADSESSION: 1170 printf("nfsv4 recover err returned %d\n", error); 1171 return (EIO); 1172 case NFSERR_BADHANDLE: 1173 case NFSERR_SERVERFAULT: 1174 case NFSERR_BADTYPE: 1175 case NFSERR_FHEXPIRED: 1176 case NFSERR_RESOURCE: 1177 case NFSERR_MOVED: 1178 case NFSERR_NOFILEHANDLE: 1179 case NFSERR_MINORVERMISMATCH: 1180 case NFSERR_OLDSTATEID: 1181 case NFSERR_BADSEQID: 1182 case NFSERR_LEASEMOVED: 1183 case NFSERR_RECLAIMBAD: 1184 case NFSERR_BADXDR: 1185 case NFSERR_OPILLEGAL: 1186 printf("nfsv4 client/server protocol prob err=%d\n", 1187 error); 1188 return (EIO); 1189 default: 1190 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error); 1191 return (EIO); 1192 }; 1193 } 1194 1195 /* 1196 * Check to see if the process for this owner exists. Return 1 if it doesn't 1197 * and 0 otherwise. 1198 */ 1199 int 1200 nfscl_procdoesntexist(u_int8_t *own) 1201 { 1202 union { 1203 u_int32_t lval; 1204 u_int8_t cval[4]; 1205 } tl; 1206 struct proc *p; 1207 pid_t pid; 1208 int i, ret = 0; 1209 1210 /* For the single open_owner of all 0 bytes, just return 0. */ 1211 for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++) 1212 if (own[i] != 0) 1213 break; 1214 if (i == NFSV4CL_LOCKNAMELEN) 1215 return (0); 1216 1217 tl.cval[0] = *own++; 1218 tl.cval[1] = *own++; 1219 tl.cval[2] = *own++; 1220 tl.cval[3] = *own++; 1221 pid = tl.lval; 1222 p = pfind_any_locked(pid); 1223 if (p == NULL) 1224 return (1); 1225 if (p->p_stats == NULL) { 1226 PROC_UNLOCK(p); 1227 return (0); 1228 } 1229 tl.cval[0] = *own++; 1230 tl.cval[1] = *own++; 1231 tl.cval[2] = *own++; 1232 tl.cval[3] = *own++; 1233 if (tl.lval != p->p_stats->p_start.tv_sec) { 1234 ret = 1; 1235 } else { 1236 tl.cval[0] = *own++; 1237 tl.cval[1] = *own++; 1238 tl.cval[2] = *own++; 1239 tl.cval[3] = *own; 1240 if (tl.lval != p->p_stats->p_start.tv_usec) 1241 ret = 1; 1242 } 1243 PROC_UNLOCK(p); 1244 return (ret); 1245 } 1246 1247 /* 1248 * - nfs pseudo system call for the client 1249 */ 1250 /* 1251 * MPSAFE 1252 */ 1253 static int 1254 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap) 1255 { 1256 struct file *fp; 1257 struct nfscbd_args nfscbdarg; 1258 struct nfsd_nfscbd_args nfscbdarg2; 1259 struct nameidata nd; 1260 struct nfscl_dumpmntopts dumpmntopts; 1261 cap_rights_t rights; 1262 char *buf; 1263 int error; 1264 struct mount *mp; 1265 struct nfsmount *nmp; 1266 1267 if (uap->flag & NFSSVC_CBADDSOCK) { 1268 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg)); 1269 if (error) 1270 return (error); 1271 /* 1272 * Since we don't know what rights might be required, 1273 * pretend that we need them all. It is better to be too 1274 * careful than too reckless. 1275 */ 1276 error = fget(td, nfscbdarg.sock, 1277 cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp); 1278 if (error) 1279 return (error); 1280 if (fp->f_type != DTYPE_SOCKET) { 1281 fdrop(fp, td); 1282 return (EPERM); 1283 } 1284 error = nfscbd_addsock(fp); 1285 fdrop(fp, td); 1286 if (!error && nfscl_enablecallb == 0) { 1287 nfsv4_cbport = nfscbdarg.port; 1288 nfscl_enablecallb = 1; 1289 } 1290 } else if (uap->flag & NFSSVC_NFSCBD) { 1291 if (uap->argp == NULL) 1292 return (EINVAL); 1293 error = copyin(uap->argp, (caddr_t)&nfscbdarg2, 1294 sizeof(nfscbdarg2)); 1295 if (error) 1296 return (error); 1297 error = nfscbd_nfsd(td, &nfscbdarg2); 1298 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) { 1299 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts)); 1300 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 || 1301 dumpmntopts.ndmnt_blen > 1024)) 1302 error = EINVAL; 1303 if (error == 0) 1304 error = nfsrv_lookupfilename(&nd, 1305 dumpmntopts.ndmnt_fname, td); 1306 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, 1307 "nfs") != 0) { 1308 vput(nd.ni_vp); 1309 error = EINVAL; 1310 } 1311 if (error == 0) { 1312 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK); 1313 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf, 1314 dumpmntopts.ndmnt_blen); 1315 vput(nd.ni_vp); 1316 error = copyout(buf, dumpmntopts.ndmnt_buf, 1317 dumpmntopts.ndmnt_blen); 1318 free(buf, M_TEMP); 1319 } 1320 } else if (uap->flag & NFSSVC_FORCEDISM) { 1321 buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK); 1322 error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL); 1323 if (error == 0) { 1324 nmp = NULL; 1325 mtx_lock(&mountlist_mtx); 1326 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1327 if (strcmp(mp->mnt_stat.f_mntonname, buf) == 1328 0 && strcmp(mp->mnt_stat.f_fstypename, 1329 "nfs") == 0 && mp->mnt_data != NULL) { 1330 nmp = VFSTONFS(mp); 1331 NFSDDSLOCK(); 1332 if (nfsv4_findmirror(nmp) != NULL) { 1333 NFSDDSUNLOCK(); 1334 error = ENXIO; 1335 nmp = NULL; 1336 break; 1337 } 1338 mtx_lock(&nmp->nm_mtx); 1339 if ((nmp->nm_privflag & 1340 NFSMNTP_FORCEDISM) == 0) { 1341 nmp->nm_privflag |= 1342 (NFSMNTP_FORCEDISM | 1343 NFSMNTP_CANCELRPCS); 1344 mtx_unlock(&nmp->nm_mtx); 1345 } else { 1346 mtx_unlock(&nmp->nm_mtx); 1347 nmp = NULL; 1348 } 1349 NFSDDSUNLOCK(); 1350 break; 1351 } 1352 } 1353 mtx_unlock(&mountlist_mtx); 1354 1355 if (nmp != NULL) { 1356 /* 1357 * Call newnfs_nmcancelreqs() to cause 1358 * any RPCs in progress on the mount point to 1359 * fail. 1360 * This will cause any process waiting for an 1361 * RPC to complete while holding a vnode lock 1362 * on the mounted-on vnode (such as "df" or 1363 * a non-forced "umount") to fail. 1364 * This will unlock the mounted-on vnode so 1365 * a forced dismount can succeed. 1366 * Then clear NFSMNTP_CANCELRPCS and wakeup(), 1367 * so that nfs_unmount() can complete. 1368 */ 1369 newnfs_nmcancelreqs(nmp); 1370 mtx_lock(&nmp->nm_mtx); 1371 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 1372 wakeup(nmp); 1373 mtx_unlock(&nmp->nm_mtx); 1374 } else if (error == 0) 1375 error = EINVAL; 1376 } 1377 free(buf, M_TEMP); 1378 } else { 1379 error = EINVAL; 1380 } 1381 return (error); 1382 } 1383 1384 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *); 1385 1386 /* 1387 * Called once to initialize data structures... 1388 */ 1389 static int 1390 nfscl_modevent(module_t mod, int type, void *data) 1391 { 1392 int error = 0; 1393 static int loaded = 0; 1394 1395 switch (type) { 1396 case MOD_LOAD: 1397 if (loaded) 1398 return (0); 1399 newnfs_portinit(); 1400 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF); 1401 nfscl_init(); 1402 NFSD_LOCK(); 1403 nfsrvd_cbinit(0); 1404 NFSD_UNLOCK(); 1405 ncl_call_invalcaches = ncl_invalcaches; 1406 nfsd_call_nfscl = nfssvc_nfscl; 1407 loaded = 1; 1408 break; 1409 1410 case MOD_UNLOAD: 1411 if (nfs_numnfscbd != 0) { 1412 error = EBUSY; 1413 break; 1414 } 1415 1416 /* 1417 * XXX: Unloading of nfscl module is unsupported. 1418 */ 1419 #if 0 1420 ncl_call_invalcaches = NULL; 1421 nfsd_call_nfscl = NULL; 1422 uma_zdestroy(ncl_pbuf_zone); 1423 /* and get rid of the mutexes */ 1424 mtx_destroy(&ncl_iod_mutex); 1425 loaded = 0; 1426 break; 1427 #else 1428 /* FALLTHROUGH */ 1429 #endif 1430 default: 1431 error = EOPNOTSUPP; 1432 break; 1433 } 1434 return error; 1435 } 1436 static moduledata_t nfscl_mod = { 1437 "nfscl", 1438 nfscl_modevent, 1439 NULL, 1440 }; 1441 /* 1442 * This is the main module declaration for the NFS client. The 1443 * nfscl_modevent() function is needed to ensure that the module 1444 * cannot be unloaded, among other things. 1445 * There is also a module declaration in sys/fs/nfsclient/nfs_clvfsops.c 1446 * for the name "nfs" within the VFS_SET() macro that defines the "nfs" 1447 * file system type. 1448 */ 1449 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST); 1450 1451 /* So that loader and kldload(2) can find us, wherever we are.. */ 1452 MODULE_VERSION(nfscl, 1); 1453 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1); 1454 MODULE_DEPEND(nfscl, krpc, 1, 1, 1); 1455 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1); 1456 MODULE_DEPEND(nfscl, xdr, 1, 1, 1); 1457