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 57#include <fs/nfsclient/nfs_kdtrace.h> 58 59#ifdef KDTRACE_HOOKS 60dtrace_nfsclient_attrcache_flush_probe_func_t 61 dtrace_nfscl_attrcache_flush_done_probe; 62uint32_t nfscl_attrcache_flush_done_id; 63 64dtrace_nfsclient_attrcache_get_hit_probe_func_t 65 dtrace_nfscl_attrcache_get_hit_probe; 66uint32_t nfscl_attrcache_get_hit_id; 67 68dtrace_nfsclient_attrcache_get_miss_probe_func_t 69 dtrace_nfscl_attrcache_get_miss_probe; 70uint32_t nfscl_attrcache_get_miss_id; 71 72dtrace_nfsclient_attrcache_load_probe_func_t 73 dtrace_nfscl_attrcache_load_done_probe; 74uint32_t nfscl_attrcache_load_done_id; 75#endif /* !KDTRACE_HOOKS */ 76 77extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 78extern struct vop_vector newnfs_vnodeops; 79extern struct vop_vector newnfs_fifoops; 80extern uma_zone_t newnfsnode_zone; 81extern struct buf_ops buf_ops_newnfs; 82extern int ncl_pbuf_freecnt; 83extern short nfsv4_cbport; 84extern int nfscl_enablecallb; 85extern int nfs_numnfscbd; 86extern int nfscl_inited; 87struct mtx ncl_iod_mutex; 88NFSDLOCKMUTEX; 89extern struct mtx nfsrv_dslock_mtx; 90 91extern void (*ncl_call_invalcaches)(struct vnode *); 92 93SYSCTL_DECL(_vfs_nfs); 94static int ncl_fileid_maxwarnings = 10; 95SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN, 96 &ncl_fileid_maxwarnings, 0, 97 "Limit fileid corruption warnings; 0 is off; -1 is unlimited"); 98static volatile int ncl_fileid_nwarnings; 99 100static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *, 101 struct nfsvattr *); 102 103/* 104 * Comparison function for vfs_hash functions. 105 */ 106int 107newnfs_vncmpf(struct vnode *vp, void *arg) 108{ 109 struct nfsfh *nfhp = (struct nfsfh *)arg; 110 struct nfsnode *np = VTONFS(vp); 111 112 if (np->n_fhp->nfh_len != nfhp->nfh_len || 113 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len)) 114 return (1); 115 return (0); 116} 117 118/* 119 * Look up a vnode/nfsnode by file handle. 120 * Callers must check for mount points!! 121 * In all cases, a pointer to a 122 * nfsnode structure is returned. 123 * This variant takes a "struct nfsfh *" as second argument and uses 124 * that structure up, either by hanging off the nfsnode or FREEing it. 125 */ 126int 127nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp, 128 struct componentname *cnp, struct thread *td, struct nfsnode **npp, 129 void *stuff, int lkflags) 130{ 131 struct nfsnode *np, *dnp; 132 struct vnode *vp, *nvp; 133 struct nfsv4node *newd, *oldd; 134 int error; 135 u_int hash; 136 struct nfsmount *nmp; 137 138 nmp = VFSTONFS(mntp); 139 dnp = VTONFS(dvp); 140 *npp = NULL; 141 142 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT); 143 144 error = vfs_hash_get(mntp, hash, lkflags, 145 td, &nvp, newnfs_vncmpf, nfhp); 146 if (error == 0 && nvp != NULL) { 147 /* 148 * I believe there is a slight chance that vgonel() could 149 * get called on this vnode between when NFSVOPLOCK() drops 150 * the VI_LOCK() and vget() acquires it again, so that it 151 * hasn't yet had v_usecount incremented. If this were to 152 * happen, the VI_DOOMED flag would be set, so check for 153 * that here. Since we now have the v_usecount incremented, 154 * we should be ok until we vrele() it, if the VI_DOOMED 155 * flag isn't set now. 156 */ 157 VI_LOCK(nvp); 158 if ((nvp->v_iflag & VI_DOOMED)) { 159 VI_UNLOCK(nvp); 160 vrele(nvp); 161 error = ENOENT; 162 } else { 163 VI_UNLOCK(nvp); 164 } 165 } 166 if (error) { 167 free(nfhp, M_NFSFH); 168 return (error); 169 } 170 if (nvp != NULL) { 171 np = VTONFS(nvp); 172 /* 173 * For NFSv4, check to see if it is the same name and 174 * replace the name, if it is different. 175 */ 176 oldd = newd = NULL; 177 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL && 178 nvp->v_type == VREG && 179 (np->n_v4->n4_namelen != cnp->cn_namelen || 180 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 181 cnp->cn_namelen) || 182 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 183 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 184 dnp->n_fhp->nfh_len))) { 185 newd = malloc( 186 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + 187 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); 188 NFSLOCKNODE(np); 189 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG 190 && (np->n_v4->n4_namelen != cnp->cn_namelen || 191 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 192 cnp->cn_namelen) || 193 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || 194 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 195 dnp->n_fhp->nfh_len))) { 196 oldd = np->n_v4; 197 np->n_v4 = newd; 198 newd = NULL; 199 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 200 np->n_v4->n4_namelen = cnp->cn_namelen; 201 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 202 dnp->n_fhp->nfh_len); 203 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 204 cnp->cn_namelen); 205 } 206 NFSUNLOCKNODE(np); 207 } 208 if (newd != NULL) 209 free(newd, M_NFSV4NODE); 210 if (oldd != NULL) 211 free(oldd, M_NFSV4NODE); 212 *npp = np; 213 free(nfhp, M_NFSFH); 214 return (0); 215 } 216 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO); 217 218 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp); 219 if (error) { 220 uma_zfree(newnfsnode_zone, np); 221 free(nfhp, M_NFSFH); 222 return (error); 223 } 224 vp = nvp; 225 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0")); 226 vp->v_bufobj.bo_ops = &buf_ops_newnfs; 227 vp->v_data = np; 228 np->n_vnode = vp; 229 /* 230 * Initialize the mutex even if the vnode is going to be a loser. 231 * This simplifies the logic in reclaim, which can then unconditionally 232 * destroy the mutex (in the case of the loser, or if hash_insert 233 * happened to return an error no special casing is needed). 234 */ 235 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK); 236 lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE | 237 LK_CANRECURSE); 238 239 /* 240 * Are we getting the root? If so, make sure the vnode flags 241 * are correct 242 */ 243 if ((nfhp->nfh_len == nmp->nm_fhsize) && 244 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) { 245 if (vp->v_type == VNON) 246 vp->v_type = VDIR; 247 vp->v_vflag |= VV_ROOT; 248 } 249 250 vp->v_vflag |= VV_VMSIZEVNLOCK; 251 252 np->n_fhp = nfhp; 253 /* 254 * For NFSv4, we have to attach the directory file handle and 255 * file name, so that Open Ops can be done later. 256 */ 257 if (nmp->nm_flag & NFSMNT_NFSV4) { 258 np->n_v4 = malloc(sizeof (struct nfsv4node) 259 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE, 260 M_WAITOK); 261 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; 262 np->n_v4->n4_namelen = cnp->cn_namelen; 263 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, 264 dnp->n_fhp->nfh_len); 265 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), 266 cnp->cn_namelen); 267 } else { 268 np->n_v4 = NULL; 269 } 270 271 /* 272 * NFS supports recursive and shared locking. 273 */ 274 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL); 275 VN_LOCK_AREC(vp); 276 VN_LOCK_ASHARE(vp); 277 error = insmntque(vp, mntp); 278 if (error != 0) { 279 *npp = NULL; 280 mtx_destroy(&np->n_mtx); 281 lockdestroy(&np->n_excl); 282 free(nfhp, M_NFSFH); 283 if (np->n_v4 != NULL) 284 free(np->n_v4, M_NFSV4NODE); 285 uma_zfree(newnfsnode_zone, np); 286 return (error); 287 } 288 error = vfs_hash_insert(vp, hash, lkflags, 289 td, &nvp, newnfs_vncmpf, nfhp); 290 if (error) 291 return (error); 292 if (nvp != NULL) { 293 *npp = VTONFS(nvp); 294 /* vfs_hash_insert() vput()'s the losing vnode */ 295 return (0); 296 } 297 *npp = np; 298 299 return (0); 300} 301 302/* 303 * Another variant of nfs_nget(). This one is only used by reopen. It 304 * takes almost the same args as nfs_nget(), but only succeeds if an entry 305 * exists in the cache. (Since files should already be "open" with a 306 * vnode ref cnt on the node when reopen calls this, it should always 307 * succeed.) 308 * Also, don't get a vnode lock, since it may already be locked by some 309 * other process that is handling it. This is ok, since all other threads 310 * on the client are blocked by the nfsc_lock being exclusively held by the 311 * caller of this function. 312 */ 313int 314nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize, 315 struct thread *td, struct nfsnode **npp) 316{ 317 struct vnode *nvp; 318 u_int hash; 319 struct nfsfh *nfhp; 320 int error; 321 322 *npp = NULL; 323 /* For forced dismounts, just return error. */ 324 if (NFSCL_FORCEDISM(mntp)) 325 return (EINTR); 326 nfhp = malloc(sizeof (struct nfsfh) + fhsize, 327 M_NFSFH, M_WAITOK); 328 bcopy(fhp, &nfhp->nfh_fh[0], fhsize); 329 nfhp->nfh_len = fhsize; 330 331 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT); 332 333 /* 334 * First, try to get the vnode locked, but don't block for the lock. 335 */ 336 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp, 337 newnfs_vncmpf, nfhp); 338 if (error == 0 && nvp != NULL) { 339 NFSVOPUNLOCK(nvp, 0); 340 } else if (error == EBUSY) { 341 /* 342 * It is safe so long as a vflush() with 343 * FORCECLOSE has not been done. Since the Renew thread is 344 * stopped and the MNTK_UNMOUNTF flag is set before doing 345 * a vflush() with FORCECLOSE, we should be ok here. 346 */ 347 if (NFSCL_FORCEDISM(mntp)) 348 error = EINTR; 349 else { 350 vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp); 351 if (nvp == NULL) { 352 error = ENOENT; 353 } else if ((nvp->v_iflag & VI_DOOMED) != 0) { 354 error = ENOENT; 355 vrele(nvp); 356 } else { 357 error = 0; 358 } 359 } 360 } 361 free(nfhp, M_NFSFH); 362 if (error) 363 return (error); 364 if (nvp != NULL) { 365 *npp = VTONFS(nvp); 366 return (0); 367 } 368 return (EINVAL); 369} 370 371static void 372nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap, 373 struct nfsvattr *newnap) 374{ 375 int off; 376 377 if (ncl_fileid_maxwarnings >= 0 && 378 ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) 379 return; 380 off = 0; 381 if (ncl_fileid_maxwarnings >= 0) { 382 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) 383 off = 1; 384 } 385 386 printf("newnfs: server '%s' error: fileid changed. " 387 "fsid %jx:%jx: expected fileid %#jx, got %#jx. " 388 "(BROKEN NFS SERVER OR MIDDLEWARE)\n", 389 nmp->nm_com.nmcom_hostname, 390 (uintmax_t)nmp->nm_fsid[0], 391 (uintmax_t)nmp->nm_fsid[1], 392 (uintmax_t)oldnap->na_fileid, 393 (uintmax_t)newnap->na_fileid); 394 395 if (off) 396 printf("newnfs: Logged %d times about fileid corruption; " 397 "going quiet to avoid spamming logs excessively. (Limit " 398 "is: %d).\n", ncl_fileid_nwarnings, 399 ncl_fileid_maxwarnings); 400} 401 402void 403ncl_copy_vattr(struct vattr *dst, struct vattr *src) 404{ 405 dst->va_type = src->va_type; 406 dst->va_mode = src->va_mode; 407 dst->va_nlink = src->va_nlink; 408 dst->va_uid = src->va_uid; 409 dst->va_gid = src->va_gid; 410 dst->va_fsid = src->va_fsid; 411 dst->va_fileid = src->va_fileid; 412 dst->va_size = src->va_size; 413 dst->va_blocksize = src->va_blocksize; 414 dst->va_atime = src->va_atime; 415 dst->va_mtime = src->va_mtime; 416 dst->va_ctime = src->va_ctime; 417 dst->va_gen = src->va_gen; 418 dst->va_flags = src->va_flags; 419 dst->va_rdev = src->va_rdev; 420 dst->va_bytes = src->va_bytes; 421 dst->va_filerev = src->va_filerev; 422} 423 424/* 425 * Load the attribute cache (that lives in the nfsnode entry) with 426 * the attributes of the second argument and 427 * Iff vaper not NULL 428 * copy the attributes to *vaper 429 * Similar to nfs_loadattrcache(), except the attributes are passed in 430 * instead of being parsed out of the mbuf list. 431 */ 432int 433nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper, 434 void *stuff, int writeattr, int dontshrink) 435{ 436 struct vnode *vp = *vpp; 437 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper; 438 struct nfsnode *np; 439 struct nfsmount *nmp; 440 struct timespec mtime_save; 441 int error, force_fid_err; 442 443 error = 0; 444 445 /* 446 * If v_type == VNON it is a new node, so fill in the v_type, 447 * n_mtime fields. Check to see if it represents a special 448 * device, and if so, check for a possible alias. Once the 449 * correct vnode has been obtained, fill in the rest of the 450 * information. 451 */ 452 np = VTONFS(vp); 453 NFSLOCKNODE(np); 454 if (vp->v_type != nvap->va_type) { 455 vp->v_type = nvap->va_type; 456 if (vp->v_type == VFIFO) 457 vp->v_op = &newnfs_fifoops; 458 np->n_mtime = nvap->va_mtime; 459 } 460 nmp = VFSTONFS(vp->v_mount); 461 vap = &np->n_vattr.na_vattr; 462 mtime_save = vap->va_mtime; 463 if (writeattr) { 464 np->n_vattr.na_filerev = nap->na_filerev; 465 np->n_vattr.na_size = nap->na_size; 466 np->n_vattr.na_mtime = nap->na_mtime; 467 np->n_vattr.na_ctime = nap->na_ctime; 468 np->n_vattr.na_fsid = nap->na_fsid; 469 np->n_vattr.na_mode = nap->na_mode; 470 } else { 471 force_fid_err = 0; 472 KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning, 473 force_fid_err); 474 /* 475 * BROKEN NFS SERVER OR MIDDLEWARE 476 * 477 * Certain NFS servers (certain old proprietary filers ca. 478 * 2006) or broken middleboxes (e.g. WAN accelerator products) 479 * will respond to GETATTR requests with results for a 480 * different fileid. 481 * 482 * The WAN accelerator we've observed not only serves stale 483 * cache results for a given file, it also occasionally serves 484 * results for wholly different files. This causes surprising 485 * problems; for example the cached size attribute of a file 486 * may truncate down and then back up, resulting in zero 487 * regions in file contents read by applications. We observed 488 * this reliably with Clang and .c files during parallel build. 489 * A pcap revealed packet fragmentation and GETATTR RPC 490 * responses with wholly wrong fileids. 491 */ 492 if ((np->n_vattr.na_fileid != 0 && 493 np->n_vattr.na_fileid != nap->na_fileid) || 494 force_fid_err) { 495 nfscl_warn_fileid(nmp, &np->n_vattr, nap); 496 error = EIDRM; 497 goto out; 498 } 499 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr, 500 sizeof (struct nfsvattr)); 501 } 502 503 /* 504 * For NFSv4, if the node's fsid is not equal to the mount point's 505 * fsid, return the low order 32bits of the node's fsid. This 506 * allows getcwd(3) to work. There is a chance that the fsid might 507 * be the same as a local fs, but since this is in an NFS mount 508 * point, I don't think that will cause any problems? 509 */ 510 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) && 511 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] || 512 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) { 513 /* 514 * va_fsid needs to be set to some value derived from 515 * np->n_vattr.na_filesid that is not equal 516 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes 517 * from the value used for the top level server volume 518 * in the mounted subtree. 519 */ 520 vn_fsid(vp, vap); 521 if ((uint32_t)vap->va_fsid == np->n_vattr.na_filesid[0]) 522 vap->va_fsid = hash32_buf( 523 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0); 524 } else 525 vn_fsid(vp, vap); 526 np->n_attrstamp = time_second; 527 if (vap->va_size != np->n_size) { 528 if (vap->va_type == VREG) { 529 if (dontshrink && vap->va_size < np->n_size) { 530 /* 531 * We've been told not to shrink the file; 532 * zero np->n_attrstamp to indicate that 533 * the attributes are stale. 534 */ 535 vap->va_size = np->n_size; 536 np->n_attrstamp = 0; 537 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 538 } else if (np->n_flag & NMODIFIED) { 539 /* 540 * We've modified the file: Use the larger 541 * of our size, and the server's size. 542 */ 543 if (vap->va_size < np->n_size) { 544 vap->va_size = np->n_size; 545 } else { 546 np->n_size = vap->va_size; 547 np->n_flag |= NSIZECHANGED; 548 } 549 } else { 550 np->n_size = vap->va_size; 551 np->n_flag |= NSIZECHANGED; 552 } 553 } else { 554 np->n_size = vap->va_size; 555 } 556 } 557 /* 558 * The following checks are added to prevent a race between (say) 559 * a READDIR+ and a WRITE. 560 * READDIR+, WRITE requests sent out. 561 * READDIR+ resp, WRITE resp received on client. 562 * However, the WRITE resp was handled before the READDIR+ resp 563 * causing the post op attrs from the write to be loaded first 564 * and the attrs from the READDIR+ to be loaded later. If this 565 * happens, we have stale attrs loaded into the attrcache. 566 * We detect this by for the mtime moving back. We invalidate the 567 * attrcache when this happens. 568 */ 569 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) { 570 /* Size changed or mtime went backwards */ 571 np->n_attrstamp = 0; 572 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); 573 } 574 if (vaper != NULL) { 575 ncl_copy_vattr(vaper, vap); 576 if (np->n_flag & NCHG) { 577 if (np->n_flag & NACC) 578 vaper->va_atime = np->n_atim; 579 if (np->n_flag & NUPD) 580 vaper->va_mtime = np->n_mtim; 581 } 582 } 583 584out: 585#ifdef KDTRACE_HOOKS 586 if (np->n_attrstamp != 0) 587 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error); 588#endif 589 (void)ncl_pager_setsize(vp, NULL); 590 return (error); 591} 592 593/* 594 * Call vnode_pager_setsize() if the size of the node changed, as 595 * recorded in nfsnode vs. v_object, or delay the call if notifying 596 * the pager is not possible at the moment. 597 * 598 * If nsizep is non-NULL, the call is delayed and the new node size is 599 * provided. Caller should itself call vnode_pager_setsize() if 600 * function returned true. If nsizep is NULL, function tries to call 601 * vnode_pager_setsize() itself if needed and possible, and the nfs 602 * node is unlocked unconditionally, the return value is not useful. 603 */ 604bool 605ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep) 606{ 607 struct nfsnode *np; 608 vm_object_t object; 609 struct vattr *vap; 610 u_quad_t nsize; 611 bool setnsize; 612 613 np = VTONFS(vp); 614 NFSASSERTNODE(np); 615 616 vap = &np->n_vattr.na_vattr; 617 nsize = vap->va_size; 618 object = vp->v_object; 619 setnsize = false; 620 621 if (object != NULL && nsize != object->un_pager.vnp.vnp_size) { 622 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && 623 (curthread->td_pflags2 & TDP2_SBPAGES) == 0) 624 setnsize = true; 625 else 626 np->n_flag |= NVNSETSZSKIP; 627 } 628 if (nsizep == NULL) { 629 NFSUNLOCKNODE(np); 630 if (setnsize) 631 vnode_pager_setsize(vp, nsize); 632 setnsize = false; 633 } else { 634 *nsizep = nsize; 635 } 636 return (setnsize); 637} 638 639/* 640 * Fill in the client id name. For these bytes: 641 * 1 - they must be unique 642 * 2 - they should be persistent across client reboots 643 * 1 is more critical than 2 644 * Use the mount point's unique id plus either the uuid or, if that 645 * isn't set, random junk. 646 */ 647void 648nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen) 649{ 650 int uuidlen; 651 652 /* 653 * First, put in the 64bit mount point identifier. 654 */ 655 if (idlen >= sizeof (u_int64_t)) { 656 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t)); 657 cp += sizeof (u_int64_t); 658 idlen -= sizeof (u_int64_t); 659 } 660 661 /* 662 * If uuid is non-zero length, use it. 663 */ 664 uuidlen = strlen(uuid); 665 if (uuidlen > 0 && idlen >= uuidlen) { 666 NFSBCOPY(uuid, cp, uuidlen); 667 cp += uuidlen; 668 idlen -= uuidlen; 669 } 670 671 /* 672 * This only normally happens if the uuid isn't set. 673 */ 674 while (idlen > 0) { 675 *cp++ = (u_int8_t)(arc4random() % 256); 676 idlen--; 677 } 678} 679 680/* 681 * Fill in a lock owner name. For now, pid + the process's creation time. 682 */ 683void 684nfscl_filllockowner(void *id, u_int8_t *cp, int flags) 685{ 686 union { 687 u_int32_t lval; 688 u_int8_t cval[4]; 689 } tl; 690 struct proc *p; 691 692 if (id == NULL) { 693 /* Return the single open_owner of all 0 bytes. */ 694 bzero(cp, NFSV4CL_LOCKNAMELEN); 695 return; 696 } 697 if ((flags & F_POSIX) != 0) { 698 p = (struct proc *)id; 699 tl.lval = p->p_pid; 700 *cp++ = tl.cval[0]; 701 *cp++ = tl.cval[1]; 702 *cp++ = tl.cval[2]; 703 *cp++ = tl.cval[3]; 704 tl.lval = p->p_stats->p_start.tv_sec; 705 *cp++ = tl.cval[0]; 706 *cp++ = tl.cval[1]; 707 *cp++ = tl.cval[2]; 708 *cp++ = tl.cval[3]; 709 tl.lval = p->p_stats->p_start.tv_usec; 710 *cp++ = tl.cval[0]; 711 *cp++ = tl.cval[1]; 712 *cp++ = tl.cval[2]; 713 *cp = tl.cval[3]; 714 } else if ((flags & F_FLOCK) != 0) { 715 bcopy(&id, cp, sizeof(id)); 716 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id)); 717 } else { 718 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n"); 719 bzero(cp, NFSV4CL_LOCKNAMELEN); 720 } 721} 722 723/* 724 * Find the parent process for the thread passed in as an argument. 725 * If none exists, return NULL, otherwise return a thread for the parent. 726 * (Can be any of the threads, since it is only used for td->td_proc.) 727 */ 728NFSPROC_T * 729nfscl_getparent(struct thread *td) 730{ 731 struct proc *p; 732 struct thread *ptd; 733 734 if (td == NULL) 735 return (NULL); 736 p = td->td_proc; 737 if (p->p_pid == 0) 738 return (NULL); 739 p = p->p_pptr; 740 if (p == NULL) 741 return (NULL); 742 ptd = TAILQ_FIRST(&p->p_threads); 743 return (ptd); 744} 745 746/* 747 * Start up the renew kernel thread. 748 */ 749static void 750start_nfscl(void *arg) 751{ 752 struct nfsclclient *clp; 753 struct thread *td; 754 755 clp = (struct nfsclclient *)arg; 756 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads); 757 nfscl_renewthread(clp, td); 758 kproc_exit(0); 759} 760 761void 762nfscl_start_renewthread(struct nfsclclient *clp) 763{ 764 765 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0, 766 "nfscl"); 767} 768 769/* 770 * Handle wcc_data. 771 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr 772 * as the first Op after PutFH. 773 * (For NFSv4, the postop attributes are after the Op, so they can't be 774 * parsed here. A separate call to nfscl_postop_attr() is required.) 775 */ 776int 777nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp, 778 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff) 779{ 780 u_int32_t *tl; 781 struct nfsnode *np = VTONFS(vp); 782 struct nfsvattr nfsva; 783 int error = 0; 784 785 if (wccflagp != NULL) 786 *wccflagp = 0; 787 if (nd->nd_flag & ND_NFSV3) { 788 *flagp = 0; 789 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 790 if (*tl == newnfs_true) { 791 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 792 if (wccflagp != NULL) { 793 NFSLOCKNODE(np); 794 *wccflagp = (np->n_mtime.tv_sec == 795 fxdr_unsigned(u_int32_t, *(tl + 2)) && 796 np->n_mtime.tv_nsec == 797 fxdr_unsigned(u_int32_t, *(tl + 3))); 798 NFSUNLOCKNODE(np); 799 } 800 } 801 error = nfscl_postop_attr(nd, nap, flagp, stuff); 802 if (wccflagp != NULL && *flagp == 0) 803 *wccflagp = 0; 804 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) 805 == (ND_NFSV4 | ND_V4WCCATTR)) { 806 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, 807 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, 808 NULL, NULL, NULL, NULL, NULL); 809 if (error) 810 return (error); 811 /* 812 * Get rid of Op# and status for next op. 813 */ 814 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 815 if (*++tl) 816 nd->nd_flag |= ND_NOMOREDATA; 817 if (wccflagp != NULL && 818 nfsva.na_vattr.va_mtime.tv_sec != 0) { 819 NFSLOCKNODE(np); 820 *wccflagp = (np->n_mtime.tv_sec == 821 nfsva.na_vattr.va_mtime.tv_sec && 822 np->n_mtime.tv_nsec == 823 nfsva.na_vattr.va_mtime.tv_sec); 824 NFSUNLOCKNODE(np); 825 } 826 } 827nfsmout: 828 return (error); 829} 830 831/* 832 * Get postop attributes. 833 */ 834int 835nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp, 836 void *stuff) 837{ 838 u_int32_t *tl; 839 int error = 0; 840 841 *retp = 0; 842 if (nd->nd_flag & ND_NOMOREDATA) 843 return (error); 844 if (nd->nd_flag & ND_NFSV3) { 845 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 846 *retp = fxdr_unsigned(int, *tl); 847 } else if (nd->nd_flag & ND_NFSV4) { 848 /* 849 * For NFSv4, the postop attr are at the end, so no point 850 * in looking if nd_repstat != 0. 851 */ 852 if (!nd->nd_repstat) { 853 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 854 if (*(tl + 1)) 855 /* should never happen since nd_repstat != 0 */ 856 nd->nd_flag |= ND_NOMOREDATA; 857 else 858 *retp = 1; 859 } 860 } else if (!nd->nd_repstat) { 861 /* For NFSv2, the attributes are here iff nd_repstat == 0 */ 862 *retp = 1; 863 } 864 if (*retp) { 865 error = nfsm_loadattr(nd, nap); 866 if (error) 867 *retp = 0; 868 } 869nfsmout: 870 return (error); 871} 872 873/* 874 * nfscl_request() - mostly a wrapper for newnfs_request(). 875 */ 876int 877nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p, 878 struct ucred *cred, void *stuff) 879{ 880 int ret, vers; 881 struct nfsmount *nmp; 882 883 nmp = VFSTONFS(vp->v_mount); 884 if (nd->nd_flag & ND_NFSV4) 885 vers = NFS_VER4; 886 else if (nd->nd_flag & ND_NFSV3) 887 vers = NFS_VER3; 888 else 889 vers = NFS_VER2; 890 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred, 891 NFS_PROG, vers, NULL, 1, NULL, NULL); 892 return (ret); 893} 894 895/* 896 * fill in this bsden's variant of statfs using nfsstatfs. 897 */ 898void 899nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs) 900{ 901 struct statfs *sbp = (struct statfs *)statfs; 902 903 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) { 904 sbp->f_bsize = NFS_FABLKSIZE; 905 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE; 906 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE; 907 /* 908 * Although sf_abytes is uint64_t and f_bavail is int64_t, 909 * the value after dividing by NFS_FABLKSIZE is small 910 * enough that it will fit in 63bits, so it is ok to 911 * assign it to f_bavail without fear that it will become 912 * negative. 913 */ 914 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE; 915 sbp->f_files = sfp->sf_tfiles; 916 /* Since f_ffree is int64_t, clip it to 63bits. */ 917 if (sfp->sf_ffiles > INT64_MAX) 918 sbp->f_ffree = INT64_MAX; 919 else 920 sbp->f_ffree = sfp->sf_ffiles; 921 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) { 922 /* 923 * The type casts to (int32_t) ensure that this code is 924 * compatible with the old NFS client, in that it will 925 * propagate bit31 to the high order bits. This may or may 926 * not be correct for NFSv2, but since it is a legacy 927 * environment, I'd rather retain backwards compatibility. 928 */ 929 sbp->f_bsize = (int32_t)sfp->sf_bsize; 930 sbp->f_blocks = (int32_t)sfp->sf_blocks; 931 sbp->f_bfree = (int32_t)sfp->sf_bfree; 932 sbp->f_bavail = (int32_t)sfp->sf_bavail; 933 sbp->f_files = 0; 934 sbp->f_ffree = 0; 935 } 936} 937 938/* 939 * Use the fsinfo stuff to update the mount point. 940 */ 941void 942nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp) 943{ 944 945 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) && 946 fsp->fs_wtpref >= NFS_FABLKSIZE) 947 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) & 948 ~(NFS_FABLKSIZE - 1); 949 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) { 950 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1); 951 if (nmp->nm_wsize == 0) 952 nmp->nm_wsize = fsp->fs_wtmax; 953 } 954 if (nmp->nm_wsize < NFS_FABLKSIZE) 955 nmp->nm_wsize = NFS_FABLKSIZE; 956 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) && 957 fsp->fs_rtpref >= NFS_FABLKSIZE) 958 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) & 959 ~(NFS_FABLKSIZE - 1); 960 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) { 961 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1); 962 if (nmp->nm_rsize == 0) 963 nmp->nm_rsize = fsp->fs_rtmax; 964 } 965 if (nmp->nm_rsize < NFS_FABLKSIZE) 966 nmp->nm_rsize = NFS_FABLKSIZE; 967 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize) 968 && fsp->fs_dtpref >= NFS_DIRBLKSIZ) 969 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) & 970 ~(NFS_DIRBLKSIZ - 1); 971 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) { 972 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1); 973 if (nmp->nm_readdirsize == 0) 974 nmp->nm_readdirsize = fsp->fs_rtmax; 975 } 976 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ) 977 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 978 if (fsp->fs_maxfilesize > 0 && 979 fsp->fs_maxfilesize < nmp->nm_maxfilesize) 980 nmp->nm_maxfilesize = fsp->fs_maxfilesize; 981 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp); 982 nmp->nm_state |= NFSSTA_GOTFSINFO; 983} 984 985/* 986 * Lookups source address which should be used to communicate with 987 * @nmp and stores it inside @pdst. 988 * 989 * Returns 0 on success. 990 */ 991u_int8_t * 992nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p) 993{ 994#if defined(INET6) || defined(INET) 995 int error, fibnum; 996 997 fibnum = curthread->td_proc->p_fibnum; 998#endif 999#ifdef INET 1000 if (nmp->nm_nam->sa_family == AF_INET) { 1001 struct sockaddr_in *sin; 1002 struct nhop4_extended nh_ext; 1003 1004 sin = (struct sockaddr_in *)nmp->nm_nam; 1005 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); 1006 error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0, 1007 &nh_ext); 1008 CURVNET_RESTORE(); 1009 if (error != 0) 1010 return (NULL); 1011 1012 if (IN_LOOPBACK(ntohl(nh_ext.nh_src.s_addr))) { 1013 /* Ignore loopback addresses */ 1014 return (NULL); 1015 } 1016 1017 *isinet6p = 0; 1018 *((struct in_addr *)paddr) = nh_ext.nh_src; 1019 1020 return (u_int8_t *)paddr; 1021 } 1022#endif 1023#ifdef INET6 1024 if (nmp->nm_nam->sa_family == AF_INET6) { 1025 struct sockaddr_in6 *sin6; 1026 1027 sin6 = (struct sockaddr_in6 *)nmp->nm_nam; 1028 1029 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); 1030 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr, 1031 sin6->sin6_scope_id, NULL, paddr, NULL); 1032 CURVNET_RESTORE(); 1033 if (error != 0) 1034 return (NULL); 1035 1036 if (IN6_IS_ADDR_LOOPBACK(paddr)) 1037 return (NULL); 1038 1039 /* Scope is embedded in */ 1040 *isinet6p = 1; 1041 1042 return (u_int8_t *)paddr; 1043 } 1044#endif 1045 return (NULL); 1046} 1047 1048/* 1049 * Copy NFS uid, gids from the cred structure. 1050 */ 1051void 1052newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr) 1053{ 1054 int i; 1055 1056 KASSERT(cr->cr_ngroups >= 0, 1057 ("newnfs_copyincred: negative cr_ngroups")); 1058 nfscr->nfsc_uid = cr->cr_uid; 1059 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1); 1060 for (i = 0; i < nfscr->nfsc_ngroups; i++) 1061 nfscr->nfsc_groups[i] = cr->cr_groups[i]; 1062} 1063 1064 1065/* 1066 * Do any client specific initialization. 1067 */ 1068void 1069nfscl_init(void) 1070{ 1071 static int inited = 0; 1072 1073 if (inited) 1074 return; 1075 inited = 1; 1076 nfscl_inited = 1; 1077 ncl_pbuf_freecnt = nswbuf / 2 + 1; 1078} 1079 1080/* 1081 * Check each of the attributes to be set, to ensure they aren't already 1082 * the correct value. Disable setting ones already correct. 1083 */ 1084int 1085nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap) 1086{ 1087 1088 if (vap->va_mode != (mode_t)VNOVAL) { 1089 if (vap->va_mode == nvap->na_mode) 1090 vap->va_mode = (mode_t)VNOVAL; 1091 } 1092 if (vap->va_uid != (uid_t)VNOVAL) { 1093 if (vap->va_uid == nvap->na_uid) 1094 vap->va_uid = (uid_t)VNOVAL; 1095 } 1096 if (vap->va_gid != (gid_t)VNOVAL) { 1097 if (vap->va_gid == nvap->na_gid) 1098 vap->va_gid = (gid_t)VNOVAL; 1099 } 1100 if (vap->va_size != VNOVAL) { 1101 if (vap->va_size == nvap->na_size) 1102 vap->va_size = VNOVAL; 1103 } 1104 1105 /* 1106 * We are normally called with only a partially initialized 1107 * VAP. Since the NFSv3 spec says that server may use the 1108 * file attributes to store the verifier, the spec requires 1109 * us to do a SETATTR RPC. FreeBSD servers store the verifier 1110 * in atime, but we can't really assume that all servers will 1111 * so we ensure that our SETATTR sets both atime and mtime. 1112 * Set the VA_UTIMES_NULL flag for this case, so that 1113 * the server's time will be used. This is needed to 1114 * work around a bug in some Solaris servers, where 1115 * setting the time TOCLIENT causes the Setattr RPC 1116 * to return NFS_OK, but not set va_mode. 1117 */ 1118 if (vap->va_mtime.tv_sec == VNOVAL) { 1119 vfs_timestamp(&vap->va_mtime); 1120 vap->va_vaflags |= VA_UTIMES_NULL; 1121 } 1122 if (vap->va_atime.tv_sec == VNOVAL) 1123 vap->va_atime = vap->va_mtime; 1124 return (1); 1125} 1126 1127/* 1128 * Map nfsv4 errors to errno.h errors. 1129 * The uid and gid arguments are only used for NFSERR_BADOWNER and that 1130 * error should only be returned for the Open, Create and Setattr Ops. 1131 * As such, most calls can just pass in 0 for those arguments. 1132 */ 1133int 1134nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid) 1135{ 1136 struct proc *p; 1137 1138 if (error < 10000 || error >= NFSERR_STALEWRITEVERF) 1139 return (error); 1140 if (td != NULL) 1141 p = td->td_proc; 1142 else 1143 p = NULL; 1144 switch (error) { 1145 case NFSERR_BADOWNER: 1146 tprintf(p, LOG_INFO, 1147 "No name and/or group mapping for uid,gid:(%d,%d)\n", 1148 uid, gid); 1149 return (EPERM); 1150 case NFSERR_BADNAME: 1151 case NFSERR_BADCHAR: 1152 printf("nfsv4 char/name not handled by server\n"); 1153 return (ENOENT); 1154 case NFSERR_STALECLIENTID: 1155 case NFSERR_STALESTATEID: 1156 case NFSERR_EXPIRED: 1157 case NFSERR_BADSTATEID: 1158 case NFSERR_BADSESSION: 1159 printf("nfsv4 recover err returned %d\n", error); 1160 return (EIO); 1161 case NFSERR_BADHANDLE: 1162 case NFSERR_SERVERFAULT: 1163 case NFSERR_BADTYPE: 1164 case NFSERR_FHEXPIRED: 1165 case NFSERR_RESOURCE: 1166 case NFSERR_MOVED: 1167 case NFSERR_NOFILEHANDLE: 1168 case NFSERR_MINORVERMISMATCH: 1169 case NFSERR_OLDSTATEID: 1170 case NFSERR_BADSEQID: 1171 case NFSERR_LEASEMOVED: 1172 case NFSERR_RECLAIMBAD: 1173 case NFSERR_BADXDR: 1174 case NFSERR_OPILLEGAL: 1175 printf("nfsv4 client/server protocol prob err=%d\n", 1176 error); 1177 return (EIO); 1178 default: 1179 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error); 1180 return (EIO); 1181 }; 1182} 1183 1184/* 1185 * Check to see if the process for this owner exists. Return 1 if it doesn't 1186 * and 0 otherwise. 1187 */ 1188int 1189nfscl_procdoesntexist(u_int8_t *own) 1190{ 1191 union { 1192 u_int32_t lval; 1193 u_int8_t cval[4]; 1194 } tl; 1195 struct proc *p; 1196 pid_t pid; 1197 int i, ret = 0; 1198 1199 /* For the single open_owner of all 0 bytes, just return 0. */ 1200 for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++) 1201 if (own[i] != 0) 1202 break; 1203 if (i == NFSV4CL_LOCKNAMELEN) 1204 return (0); 1205 1206 tl.cval[0] = *own++; 1207 tl.cval[1] = *own++; 1208 tl.cval[2] = *own++; 1209 tl.cval[3] = *own++; 1210 pid = tl.lval; 1211 p = pfind_locked(pid); 1212 if (p == NULL) 1213 return (1); 1214 if (p->p_stats == NULL) { 1215 PROC_UNLOCK(p); 1216 return (0); 1217 } 1218 tl.cval[0] = *own++; 1219 tl.cval[1] = *own++; 1220 tl.cval[2] = *own++; 1221 tl.cval[3] = *own++; 1222 if (tl.lval != p->p_stats->p_start.tv_sec) { 1223 ret = 1; 1224 } else { 1225 tl.cval[0] = *own++; 1226 tl.cval[1] = *own++; 1227 tl.cval[2] = *own++; 1228 tl.cval[3] = *own; 1229 if (tl.lval != p->p_stats->p_start.tv_usec) 1230 ret = 1; 1231 } 1232 PROC_UNLOCK(p); 1233 return (ret); 1234} 1235 1236/* 1237 * - nfs pseudo system call for the client 1238 */ 1239/* 1240 * MPSAFE 1241 */ 1242static int 1243nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap) 1244{ 1245 struct file *fp; 1246 struct nfscbd_args nfscbdarg; 1247 struct nfsd_nfscbd_args nfscbdarg2; 1248 struct nameidata nd; 1249 struct nfscl_dumpmntopts dumpmntopts; 1250 cap_rights_t rights; 1251 char *buf; 1252 int error; 1253 struct mount *mp; 1254 struct nfsmount *nmp; 1255 1256 if (uap->flag & NFSSVC_CBADDSOCK) { 1257 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg)); 1258 if (error) 1259 return (error); 1260 /* 1261 * Since we don't know what rights might be required, 1262 * pretend that we need them all. It is better to be too 1263 * careful than too reckless. 1264 */ 1265 error = fget(td, nfscbdarg.sock, 1266 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp); 1267 if (error) 1268 return (error); 1269 if (fp->f_type != DTYPE_SOCKET) { 1270 fdrop(fp, td); 1271 return (EPERM); 1272 } 1273 error = nfscbd_addsock(fp); 1274 fdrop(fp, td); 1275 if (!error && nfscl_enablecallb == 0) { 1276 nfsv4_cbport = nfscbdarg.port; 1277 nfscl_enablecallb = 1; 1278 } 1279 } else if (uap->flag & NFSSVC_NFSCBD) { 1280 if (uap->argp == NULL) 1281 return (EINVAL); 1282 error = copyin(uap->argp, (caddr_t)&nfscbdarg2, 1283 sizeof(nfscbdarg2)); 1284 if (error) 1285 return (error); 1286 error = nfscbd_nfsd(td, &nfscbdarg2); 1287 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) { 1288 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts)); 1289 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 || 1290 dumpmntopts.ndmnt_blen > 1024)) 1291 error = EINVAL; 1292 if (error == 0) 1293 error = nfsrv_lookupfilename(&nd, 1294 dumpmntopts.ndmnt_fname, td); 1295 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, 1296 "nfs") != 0) { 1297 vput(nd.ni_vp); 1298 error = EINVAL; 1299 } 1300 if (error == 0) { 1301 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK); 1302 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf, 1303 dumpmntopts.ndmnt_blen); 1304 vput(nd.ni_vp); 1305 error = copyout(buf, dumpmntopts.ndmnt_buf, 1306 dumpmntopts.ndmnt_blen); 1307 free(buf, M_TEMP); 1308 } 1309 } else if (uap->flag & NFSSVC_FORCEDISM) { 1310 buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK); 1311 error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL); 1312 if (error == 0) { 1313 nmp = NULL; 1314 mtx_lock(&mountlist_mtx); 1315 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1316 if (strcmp(mp->mnt_stat.f_mntonname, buf) == 1317 0 && strcmp(mp->mnt_stat.f_fstypename, 1318 "nfs") == 0 && mp->mnt_data != NULL) { 1319 nmp = VFSTONFS(mp); 1320 NFSDDSLOCK(); 1321 if (nfsv4_findmirror(nmp) != NULL) { 1322 NFSDDSUNLOCK(); 1323 error = ENXIO; 1324 nmp = NULL; 1325 break; 1326 } 1327 mtx_lock(&nmp->nm_mtx); 1328 if ((nmp->nm_privflag & 1329 NFSMNTP_FORCEDISM) == 0) { 1330 nmp->nm_privflag |= 1331 (NFSMNTP_FORCEDISM | 1332 NFSMNTP_CANCELRPCS); 1333 mtx_unlock(&nmp->nm_mtx); 1334 } else { 1335 mtx_unlock(&nmp->nm_mtx); 1336 nmp = NULL; 1337 } 1338 NFSDDSUNLOCK(); 1339 break; 1340 } 1341 } 1342 mtx_unlock(&mountlist_mtx); 1343 1344 if (nmp != NULL) { 1345 /* 1346 * Call newnfs_nmcancelreqs() to cause 1347 * any RPCs in progress on the mount point to 1348 * fail. 1349 * This will cause any process waiting for an 1350 * RPC to complete while holding a vnode lock 1351 * on the mounted-on vnode (such as "df" or 1352 * a non-forced "umount") to fail. 1353 * This will unlock the mounted-on vnode so 1354 * a forced dismount can succeed. 1355 * Then clear NFSMNTP_CANCELRPCS and wakeup(), 1356 * so that nfs_unmount() can complete. 1357 */ 1358 newnfs_nmcancelreqs(nmp); 1359 mtx_lock(&nmp->nm_mtx); 1360 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 1361 wakeup(nmp); 1362 mtx_unlock(&nmp->nm_mtx); 1363 } else if (error == 0) 1364 error = EINVAL; 1365 } 1366 free(buf, M_TEMP); 1367 } else { 1368 error = EINVAL; 1369 } 1370 return (error); 1371} 1372 1373extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *); 1374 1375/* 1376 * Called once to initialize data structures... 1377 */ 1378static int 1379nfscl_modevent(module_t mod, int type, void *data) 1380{ 1381 int error = 0; 1382 static int loaded = 0; 1383 1384 switch (type) { 1385 case MOD_LOAD: 1386 if (loaded) 1387 return (0); 1388 newnfs_portinit(); 1389 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF); 1390 nfscl_init(); 1391 NFSD_LOCK(); 1392 nfsrvd_cbinit(0); 1393 NFSD_UNLOCK(); 1394 ncl_call_invalcaches = ncl_invalcaches; 1395 nfsd_call_nfscl = nfssvc_nfscl; 1396 loaded = 1; 1397 break; 1398 1399 case MOD_UNLOAD: 1400 if (nfs_numnfscbd != 0) { 1401 error = EBUSY; 1402 break; 1403 } 1404 1405 /* 1406 * XXX: Unloading of nfscl module is unsupported. 1407 */ 1408#if 0 1409 ncl_call_invalcaches = NULL; 1410 nfsd_call_nfscl = NULL; 1411 /* and get rid of the mutexes */ 1412 mtx_destroy(&ncl_iod_mutex); 1413 loaded = 0; 1414 break; 1415#else 1416 /* FALLTHROUGH */ 1417#endif 1418 default: 1419 error = EOPNOTSUPP; 1420 break; 1421 } 1422 return error; 1423} 1424static moduledata_t nfscl_mod = { 1425 "nfscl", 1426 nfscl_modevent, 1427 NULL, 1428}; 1429DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST); 1430 1431/* So that loader and kldload(2) can find us, wherever we are.. */ 1432MODULE_VERSION(nfscl, 1); 1433MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1); 1434MODULE_DEPEND(nfscl, krpc, 1, 1, 1); 1435MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1); 1436MODULE_DEPEND(nfscl, nfslock, 1, 1, 1); 1437 1438