nfs_nfsdport.c revision 232018
1/*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 */ 33 34#include <sys/cdefs.h> 35__FBSDID("$FreeBSD: stable/9/sys/fs/nfsserver/nfs_nfsdport.c 232018 2012-02-23 04:32:41Z rmacklem $"); 36 37#include <sys/capability.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 <sys/hash.h> 47#include <sys/sysctl.h> 48#include <nlm/nlm_prot.h> 49#include <nlm/nlm.h> 50 51FEATURE(nfsd, "NFSv4 server"); 52 53extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; 54extern int nfsrv_useacl; 55extern int newnfs_numnfsd; 56extern struct mount nfsv4root_mnt; 57extern struct nfsrv_stablefirst nfsrv_stablefirst; 58extern void (*nfsd_call_servertimer)(void); 59extern SVCPOOL *nfsrvd_pool; 60struct vfsoptlist nfsv4root_opt, nfsv4root_newopt; 61NFSDLOCKMUTEX; 62struct mtx nfs_cache_mutex; 63struct mtx nfs_v4root_mutex; 64struct nfsrvfh nfs_rootfh, nfs_pubfh; 65int nfs_pubfhset = 0, nfs_rootfhset = 0; 66struct proc *nfsd_master_proc = NULL; 67static pid_t nfsd_master_pid = (pid_t)-1; 68static char nfsd_master_comm[MAXCOMLEN + 1]; 69static struct timeval nfsd_master_start; 70static uint32_t nfsv4_sysid = 0; 71 72static int nfssvc_srvcall(struct thread *, struct nfssvc_args *, 73 struct ucred *); 74 75int nfsrv_enable_crossmntpt = 1; 76static int nfs_commit_blks; 77static int nfs_commit_miss; 78extern int nfsrv_issuedelegs; 79extern int nfsrv_dolocallocks; 80 81SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server"); 82SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW, 83 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points"); 84SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 85 0, ""); 86SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 87 0, ""); 88SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW, 89 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations"); 90SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW, 91 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files"); 92 93#define MAX_REORDERED_RPC 16 94#define NUM_HEURISTIC 1031 95#define NHUSE_INIT 64 96#define NHUSE_INC 16 97#define NHUSE_MAX 2048 98 99static struct nfsheur { 100 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */ 101 off_t nh_nextoff; /* next offset for sequential detection */ 102 int nh_use; /* use count for selection */ 103 int nh_seqcount; /* heuristic */ 104} nfsheur[NUM_HEURISTIC]; 105 106 107/* 108 * Heuristic to detect sequential operation. 109 */ 110static struct nfsheur * 111nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp) 112{ 113 struct nfsheur *nh; 114 int hi, try; 115 116 /* Locate best candidate. */ 117 try = 32; 118 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC; 119 nh = &nfsheur[hi]; 120 while (try--) { 121 if (nfsheur[hi].nh_vp == vp) { 122 nh = &nfsheur[hi]; 123 break; 124 } 125 if (nfsheur[hi].nh_use > 0) 126 --nfsheur[hi].nh_use; 127 hi = (hi + 1) % NUM_HEURISTIC; 128 if (nfsheur[hi].nh_use < nh->nh_use) 129 nh = &nfsheur[hi]; 130 } 131 132 /* Initialize hint if this is a new file. */ 133 if (nh->nh_vp != vp) { 134 nh->nh_vp = vp; 135 nh->nh_nextoff = uio->uio_offset; 136 nh->nh_use = NHUSE_INIT; 137 if (uio->uio_offset == 0) 138 nh->nh_seqcount = 4; 139 else 140 nh->nh_seqcount = 1; 141 } 142 143 /* Calculate heuristic. */ 144 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) || 145 uio->uio_offset == nh->nh_nextoff) { 146 /* See comments in vfs_vnops.c:sequential_heuristic(). */ 147 nh->nh_seqcount += howmany(uio->uio_resid, 16384); 148 if (nh->nh_seqcount > IO_SEQMAX) 149 nh->nh_seqcount = IO_SEQMAX; 150 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC * 151 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) { 152 /* Probably a reordered RPC, leave seqcount alone. */ 153 } else if (nh->nh_seqcount > 1) { 154 nh->nh_seqcount /= 2; 155 } else { 156 nh->nh_seqcount = 0; 157 } 158 nh->nh_use += NHUSE_INC; 159 if (nh->nh_use > NHUSE_MAX) 160 nh->nh_use = NHUSE_MAX; 161 return (nh); 162} 163 164/* 165 * Get attributes into nfsvattr structure. 166 */ 167int 168nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 169 struct thread *p, int vpislocked) 170{ 171 int error, lockedit = 0; 172 173 if (vpislocked == 0) { 174 /* 175 * When vpislocked == 0, the vnode is either exclusively 176 * locked by this thread or not locked by this thread. 177 * As such, shared lock it, if not exclusively locked. 178 */ 179 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) { 180 lockedit = 1; 181 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY); 182 } 183 } 184 error = VOP_GETATTR(vp, &nvap->na_vattr, cred); 185 if (lockedit != 0) 186 NFSVOPUNLOCK(vp, 0); 187 188 NFSEXITCODE(error); 189 return (error); 190} 191 192/* 193 * Get a file handle for a vnode. 194 */ 195int 196nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p) 197{ 198 int error; 199 200 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t)); 201 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid; 202 error = VOP_VPTOFH(vp, &fhp->fh_fid); 203 204 NFSEXITCODE(error); 205 return (error); 206} 207 208/* 209 * Perform access checking for vnodes obtained from file handles that would 210 * refer to files already opened by a Unix client. You cannot just use 211 * vn_writechk() and VOP_ACCESSX() for two reasons. 212 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write 213 * case. 214 * 2 - The owner is to be given access irrespective of mode bits for some 215 * operations, so that processes that chmod after opening a file don't 216 * break. 217 */ 218int 219nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred, 220 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked, 221 u_int32_t *supportedtypep) 222{ 223 struct vattr vattr; 224 int error = 0, getret = 0; 225 226 if (vpislocked == 0) { 227 if (NFSVOPLOCK(vp, LK_SHARED) != 0) { 228 error = EPERM; 229 goto out; 230 } 231 } 232 if (accmode & VWRITE) { 233 /* Just vn_writechk() changed to check rdonly */ 234 /* 235 * Disallow write attempts on read-only file systems; 236 * unless the file is a socket or a block or character 237 * device resident on the file system. 238 */ 239 if (NFSVNO_EXRDONLY(exp) || 240 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 241 switch (vp->v_type) { 242 case VREG: 243 case VDIR: 244 case VLNK: 245 error = EROFS; 246 default: 247 break; 248 } 249 } 250 /* 251 * If there's shared text associated with 252 * the inode, try to free it up once. If 253 * we fail, we can't allow writing. 254 */ 255 if ((vp->v_vflag & VV_TEXT) != 0 && error == 0) 256 error = ETXTBSY; 257 } 258 if (error != 0) { 259 if (vpislocked == 0) 260 NFSVOPUNLOCK(vp, 0); 261 goto out; 262 } 263 264 /* 265 * Should the override still be applied when ACLs are enabled? 266 */ 267 error = VOP_ACCESSX(vp, accmode, cred, p); 268 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) { 269 /* 270 * Try again with VEXPLICIT_DENY, to see if the test for 271 * deletion is supported. 272 */ 273 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p); 274 if (error == 0) { 275 if (vp->v_type == VDIR) { 276 accmode &= ~(VDELETE | VDELETE_CHILD); 277 accmode |= VWRITE; 278 error = VOP_ACCESSX(vp, accmode, cred, p); 279 } else if (supportedtypep != NULL) { 280 *supportedtypep &= ~NFSACCESS_DELETE; 281 } 282 } 283 } 284 285 /* 286 * Allow certain operations for the owner (reads and writes 287 * on files that are already open). 288 */ 289 if (override != NFSACCCHK_NOOVERRIDE && 290 (error == EPERM || error == EACCES)) { 291 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT)) 292 error = 0; 293 else if (override & NFSACCCHK_ALLOWOWNER) { 294 getret = VOP_GETATTR(vp, &vattr, cred); 295 if (getret == 0 && cred->cr_uid == vattr.va_uid) 296 error = 0; 297 } 298 } 299 if (vpislocked == 0) 300 NFSVOPUNLOCK(vp, 0); 301 302out: 303 NFSEXITCODE(error); 304 return (error); 305} 306 307/* 308 * Set attribute(s) vnop. 309 */ 310int 311nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred, 312 struct thread *p, struct nfsexstuff *exp) 313{ 314 int error; 315 316 error = VOP_SETATTR(vp, &nvap->na_vattr, cred); 317 NFSEXITCODE(error); 318 return (error); 319} 320 321/* 322 * Set up nameidata for a lookup() call and do it. 323 */ 324int 325nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp, 326 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p, 327 struct vnode **retdirp) 328{ 329 struct componentname *cnp = &ndp->ni_cnd; 330 int i; 331 struct iovec aiov; 332 struct uio auio; 333 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen; 334 int error = 0, crossmnt; 335 char *cp; 336 337 *retdirp = NULL; 338 cnp->cn_nameptr = cnp->cn_pnbuf; 339 ndp->ni_strictrelative = 0; 340 /* 341 * Extract and set starting directory. 342 */ 343 if (dp->v_type != VDIR) { 344 if (islocked) 345 vput(dp); 346 else 347 vrele(dp); 348 nfsvno_relpathbuf(ndp); 349 error = ENOTDIR; 350 goto out1; 351 } 352 if (islocked) 353 NFSVOPUNLOCK(dp, 0); 354 VREF(dp); 355 *retdirp = dp; 356 if (NFSVNO_EXRDONLY(exp)) 357 cnp->cn_flags |= RDONLY; 358 ndp->ni_segflg = UIO_SYSSPACE; 359 crossmnt = 1; 360 361 if (nd->nd_flag & ND_PUBLOOKUP) { 362 ndp->ni_loopcnt = 0; 363 if (cnp->cn_pnbuf[0] == '/') { 364 vrele(dp); 365 /* 366 * Check for degenerate pathnames here, since lookup() 367 * panics on them. 368 */ 369 for (i = 1; i < ndp->ni_pathlen; i++) 370 if (cnp->cn_pnbuf[i] != '/') 371 break; 372 if (i == ndp->ni_pathlen) { 373 error = NFSERR_ACCES; 374 goto out; 375 } 376 dp = rootvnode; 377 VREF(dp); 378 } 379 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) || 380 (nd->nd_flag & ND_NFSV4) == 0) { 381 /* 382 * Only cross mount points for NFSv4 when doing a 383 * mount while traversing the file system above 384 * the mount point, unless nfsrv_enable_crossmntpt is set. 385 */ 386 cnp->cn_flags |= NOCROSSMOUNT; 387 crossmnt = 0; 388 } 389 390 /* 391 * Initialize for scan, set ni_startdir and bump ref on dp again 392 * becuase lookup() will dereference ni_startdir. 393 */ 394 395 cnp->cn_thread = p; 396 ndp->ni_startdir = dp; 397 ndp->ni_rootdir = rootvnode; 398 399 if (!lockleaf) 400 cnp->cn_flags |= LOCKLEAF; 401 for (;;) { 402 cnp->cn_nameptr = cnp->cn_pnbuf; 403 /* 404 * Call lookup() to do the real work. If an error occurs, 405 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and 406 * we do not have to dereference anything before returning. 407 * In either case ni_startdir will be dereferenced and NULLed 408 * out. 409 */ 410 error = lookup(ndp); 411 if (error) 412 break; 413 414 /* 415 * Check for encountering a symbolic link. Trivial 416 * termination occurs if no symlink encountered. 417 */ 418 if ((cnp->cn_flags & ISSYMLINK) == 0) { 419 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) 420 nfsvno_relpathbuf(ndp); 421 if (ndp->ni_vp && !lockleaf) 422 NFSVOPUNLOCK(ndp->ni_vp, 0); 423 break; 424 } 425 426 /* 427 * Validate symlink 428 */ 429 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1) 430 NFSVOPUNLOCK(ndp->ni_dvp, 0); 431 if (!(nd->nd_flag & ND_PUBLOOKUP)) { 432 error = EINVAL; 433 goto badlink2; 434 } 435 436 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 437 error = ELOOP; 438 goto badlink2; 439 } 440 if (ndp->ni_pathlen > 1) 441 cp = uma_zalloc(namei_zone, M_WAITOK); 442 else 443 cp = cnp->cn_pnbuf; 444 aiov.iov_base = cp; 445 aiov.iov_len = MAXPATHLEN; 446 auio.uio_iov = &aiov; 447 auio.uio_iovcnt = 1; 448 auio.uio_offset = 0; 449 auio.uio_rw = UIO_READ; 450 auio.uio_segflg = UIO_SYSSPACE; 451 auio.uio_td = NULL; 452 auio.uio_resid = MAXPATHLEN; 453 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 454 if (error) { 455 badlink1: 456 if (ndp->ni_pathlen > 1) 457 uma_zfree(namei_zone, cp); 458 badlink2: 459 vrele(ndp->ni_dvp); 460 vput(ndp->ni_vp); 461 break; 462 } 463 linklen = MAXPATHLEN - auio.uio_resid; 464 if (linklen == 0) { 465 error = ENOENT; 466 goto badlink1; 467 } 468 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 469 error = ENAMETOOLONG; 470 goto badlink1; 471 } 472 473 /* 474 * Adjust or replace path 475 */ 476 if (ndp->ni_pathlen > 1) { 477 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 478 uma_zfree(namei_zone, cnp->cn_pnbuf); 479 cnp->cn_pnbuf = cp; 480 } else 481 cnp->cn_pnbuf[linklen] = '\0'; 482 ndp->ni_pathlen += linklen; 483 484 /* 485 * Cleanup refs for next loop and check if root directory 486 * should replace current directory. Normally ni_dvp 487 * becomes the new base directory and is cleaned up when 488 * we loop. Explicitly null pointers after invalidation 489 * to clarify operation. 490 */ 491 vput(ndp->ni_vp); 492 ndp->ni_vp = NULL; 493 494 if (cnp->cn_pnbuf[0] == '/') { 495 vrele(ndp->ni_dvp); 496 ndp->ni_dvp = ndp->ni_rootdir; 497 VREF(ndp->ni_dvp); 498 } 499 ndp->ni_startdir = ndp->ni_dvp; 500 ndp->ni_dvp = NULL; 501 } 502 if (!lockleaf) 503 cnp->cn_flags &= ~LOCKLEAF; 504 505out: 506 if (error) { 507 uma_zfree(namei_zone, cnp->cn_pnbuf); 508 ndp->ni_vp = NULL; 509 ndp->ni_dvp = NULL; 510 ndp->ni_startdir = NULL; 511 cnp->cn_flags &= ~HASBUF; 512 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) { 513 ndp->ni_dvp = NULL; 514 } 515 516out1: 517 NFSEXITCODE2(error, nd); 518 return (error); 519} 520 521/* 522 * Set up a pathname buffer and return a pointer to it and, optionally 523 * set a hash pointer. 524 */ 525void 526nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp) 527{ 528 struct componentname *cnp = &ndp->ni_cnd; 529 530 cnp->cn_flags |= (NOMACCHECK | HASBUF); 531 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 532 if (hashpp != NULL) 533 *hashpp = NULL; 534 *bufpp = cnp->cn_pnbuf; 535} 536 537/* 538 * Release the above path buffer, if not released by nfsvno_namei(). 539 */ 540void 541nfsvno_relpathbuf(struct nameidata *ndp) 542{ 543 544 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0) 545 panic("nfsrelpath"); 546 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 547 ndp->ni_cnd.cn_flags &= ~HASBUF; 548} 549 550/* 551 * Readlink vnode op into an mbuf list. 552 */ 553int 554nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p, 555 struct mbuf **mpp, struct mbuf **mpendp, int *lenp) 556{ 557 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN]; 558 struct iovec *ivp = iv; 559 struct uio io, *uiop = &io; 560 struct mbuf *mp, *mp2 = NULL, *mp3 = NULL; 561 int i, len, tlen, error = 0; 562 563 len = 0; 564 i = 0; 565 while (len < NFS_MAXPATHLEN) { 566 NFSMGET(mp); 567 MCLGET(mp, M_WAIT); 568 mp->m_len = NFSMSIZ(mp); 569 if (len == 0) { 570 mp3 = mp2 = mp; 571 } else { 572 mp2->m_next = mp; 573 mp2 = mp; 574 } 575 if ((len + mp->m_len) > NFS_MAXPATHLEN) { 576 mp->m_len = NFS_MAXPATHLEN - len; 577 len = NFS_MAXPATHLEN; 578 } else { 579 len += mp->m_len; 580 } 581 ivp->iov_base = mtod(mp, caddr_t); 582 ivp->iov_len = mp->m_len; 583 i++; 584 ivp++; 585 } 586 uiop->uio_iov = iv; 587 uiop->uio_iovcnt = i; 588 uiop->uio_offset = 0; 589 uiop->uio_resid = len; 590 uiop->uio_rw = UIO_READ; 591 uiop->uio_segflg = UIO_SYSSPACE; 592 uiop->uio_td = NULL; 593 error = VOP_READLINK(vp, uiop, cred); 594 if (error) { 595 m_freem(mp3); 596 *lenp = 0; 597 goto out; 598 } 599 if (uiop->uio_resid > 0) { 600 len -= uiop->uio_resid; 601 tlen = NFSM_RNDUP(len); 602 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len); 603 } 604 *lenp = len; 605 *mpp = mp3; 606 *mpendp = mp; 607 608out: 609 NFSEXITCODE(error); 610 return (error); 611} 612 613/* 614 * Read vnode op call into mbuf list. 615 */ 616int 617nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred, 618 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp) 619{ 620 struct mbuf *m; 621 int i; 622 struct iovec *iv; 623 struct iovec *iv2; 624 int error = 0, len, left, siz, tlen, ioflag = 0; 625 struct mbuf *m2 = NULL, *m3; 626 struct uio io, *uiop = &io; 627 struct nfsheur *nh; 628 629 len = left = NFSM_RNDUP(cnt); 630 m3 = NULL; 631 /* 632 * Generate the mbuf list with the uio_iov ref. to it. 633 */ 634 i = 0; 635 while (left > 0) { 636 NFSMGET(m); 637 MCLGET(m, M_WAIT); 638 m->m_len = 0; 639 siz = min(M_TRAILINGSPACE(m), left); 640 left -= siz; 641 i++; 642 if (m3) 643 m2->m_next = m; 644 else 645 m3 = m; 646 m2 = m; 647 } 648 MALLOC(iv, struct iovec *, i * sizeof (struct iovec), 649 M_TEMP, M_WAITOK); 650 uiop->uio_iov = iv2 = iv; 651 m = m3; 652 left = len; 653 i = 0; 654 while (left > 0) { 655 if (m == NULL) 656 panic("nfsvno_read iov"); 657 siz = min(M_TRAILINGSPACE(m), left); 658 if (siz > 0) { 659 iv->iov_base = mtod(m, caddr_t) + m->m_len; 660 iv->iov_len = siz; 661 m->m_len += siz; 662 left -= siz; 663 iv++; 664 i++; 665 } 666 m = m->m_next; 667 } 668 uiop->uio_iovcnt = i; 669 uiop->uio_offset = off; 670 uiop->uio_resid = len; 671 uiop->uio_rw = UIO_READ; 672 uiop->uio_segflg = UIO_SYSSPACE; 673 nh = nfsrv_sequential_heuristic(uiop, vp); 674 ioflag |= nh->nh_seqcount << IO_SEQSHIFT; 675 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred); 676 FREE((caddr_t)iv2, M_TEMP); 677 if (error) { 678 m_freem(m3); 679 *mpp = NULL; 680 goto out; 681 } 682 nh->nh_nextoff = uiop->uio_offset; 683 tlen = len - uiop->uio_resid; 684 cnt = cnt < tlen ? cnt : tlen; 685 tlen = NFSM_RNDUP(cnt); 686 if (tlen == 0) { 687 m_freem(m3); 688 m3 = NULL; 689 } else if (len != tlen || tlen != cnt) 690 nfsrv_adj(m3, len - tlen, tlen - cnt); 691 *mpp = m3; 692 *mpendp = m2; 693 694out: 695 NFSEXITCODE(error); 696 return (error); 697} 698 699/* 700 * Write vnode op from an mbuf list. 701 */ 702int 703nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable, 704 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p) 705{ 706 struct iovec *ivp; 707 int i, len; 708 struct iovec *iv; 709 int ioflags, error; 710 struct uio io, *uiop = &io; 711 struct nfsheur *nh; 712 713 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP, 714 M_WAITOK); 715 uiop->uio_iov = iv = ivp; 716 uiop->uio_iovcnt = cnt; 717 i = mtod(mp, caddr_t) + mp->m_len - cp; 718 len = retlen; 719 while (len > 0) { 720 if (mp == NULL) 721 panic("nfsvno_write"); 722 if (i > 0) { 723 i = min(i, len); 724 ivp->iov_base = cp; 725 ivp->iov_len = i; 726 ivp++; 727 len -= i; 728 } 729 mp = mp->m_next; 730 if (mp) { 731 i = mp->m_len; 732 cp = mtod(mp, caddr_t); 733 } 734 } 735 736 if (stable == NFSWRITE_UNSTABLE) 737 ioflags = IO_NODELOCKED; 738 else 739 ioflags = (IO_SYNC | IO_NODELOCKED); 740 uiop->uio_resid = retlen; 741 uiop->uio_rw = UIO_WRITE; 742 uiop->uio_segflg = UIO_SYSSPACE; 743 NFSUIOPROC(uiop, p); 744 uiop->uio_offset = off; 745 nh = nfsrv_sequential_heuristic(uiop, vp); 746 ioflags |= nh->nh_seqcount << IO_SEQSHIFT; 747 error = VOP_WRITE(vp, uiop, ioflags, cred); 748 if (error == 0) 749 nh->nh_nextoff = uiop->uio_offset; 750 FREE((caddr_t)iv, M_TEMP); 751 752 NFSEXITCODE(error); 753 return (error); 754} 755 756/* 757 * Common code for creating a regular file (plus special files for V2). 758 */ 759int 760nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp, 761 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp, 762 int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp) 763{ 764 u_quad_t tempsize; 765 int error; 766 767 error = nd->nd_repstat; 768 if (!error && ndp->ni_vp == NULL) { 769 if (nvap->na_type == VREG || nvap->na_type == VSOCK) { 770 vrele(ndp->ni_startdir); 771 error = VOP_CREATE(ndp->ni_dvp, 772 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 773 vput(ndp->ni_dvp); 774 nfsvno_relpathbuf(ndp); 775 if (!error) { 776 if (*exclusive_flagp) { 777 *exclusive_flagp = 0; 778 NFSVNO_ATTRINIT(nvap); 779 nvap->na_atime.tv_sec = cverf[0]; 780 nvap->na_atime.tv_nsec = cverf[1]; 781 error = VOP_SETATTR(ndp->ni_vp, 782 &nvap->na_vattr, nd->nd_cred); 783 } 784 } 785 /* 786 * NFS V2 Only. nfsrvd_mknod() does this for V3. 787 * (This implies, just get out on an error.) 788 */ 789 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK || 790 nvap->na_type == VFIFO) { 791 if (nvap->na_type == VCHR && rdev == 0xffffffff) 792 nvap->na_type = VFIFO; 793 if (nvap->na_type != VFIFO && 794 (error = priv_check_cred(nd->nd_cred, 795 PRIV_VFS_MKNOD_DEV, 0))) { 796 vrele(ndp->ni_startdir); 797 nfsvno_relpathbuf(ndp); 798 vput(ndp->ni_dvp); 799 goto out; 800 } 801 nvap->na_rdev = rdev; 802 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 803 &ndp->ni_cnd, &nvap->na_vattr); 804 vput(ndp->ni_dvp); 805 nfsvno_relpathbuf(ndp); 806 vrele(ndp->ni_startdir); 807 if (error) 808 goto out; 809 } else { 810 vrele(ndp->ni_startdir); 811 nfsvno_relpathbuf(ndp); 812 vput(ndp->ni_dvp); 813 error = ENXIO; 814 goto out; 815 } 816 *vpp = ndp->ni_vp; 817 } else { 818 /* 819 * Handle cases where error is already set and/or 820 * the file exists. 821 * 1 - clean up the lookup 822 * 2 - iff !error and na_size set, truncate it 823 */ 824 vrele(ndp->ni_startdir); 825 nfsvno_relpathbuf(ndp); 826 *vpp = ndp->ni_vp; 827 if (ndp->ni_dvp == *vpp) 828 vrele(ndp->ni_dvp); 829 else 830 vput(ndp->ni_dvp); 831 if (!error && nvap->na_size != VNOVAL) { 832 error = nfsvno_accchk(*vpp, VWRITE, 833 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 834 NFSACCCHK_VPISLOCKED, NULL); 835 if (!error) { 836 tempsize = nvap->na_size; 837 NFSVNO_ATTRINIT(nvap); 838 nvap->na_size = tempsize; 839 error = VOP_SETATTR(*vpp, 840 &nvap->na_vattr, nd->nd_cred); 841 } 842 } 843 if (error) 844 vput(*vpp); 845 } 846 847out: 848 NFSEXITCODE(error); 849 return (error); 850} 851 852/* 853 * Do a mknod vnode op. 854 */ 855int 856nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred, 857 struct thread *p) 858{ 859 int error = 0; 860 enum vtype vtyp; 861 862 vtyp = nvap->na_type; 863 /* 864 * Iff doesn't exist, create it. 865 */ 866 if (ndp->ni_vp) { 867 vrele(ndp->ni_startdir); 868 nfsvno_relpathbuf(ndp); 869 vput(ndp->ni_dvp); 870 vrele(ndp->ni_vp); 871 error = EEXIST; 872 goto out; 873 } 874 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) { 875 vrele(ndp->ni_startdir); 876 nfsvno_relpathbuf(ndp); 877 vput(ndp->ni_dvp); 878 error = NFSERR_BADTYPE; 879 goto out; 880 } 881 if (vtyp == VSOCK) { 882 vrele(ndp->ni_startdir); 883 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, 884 &ndp->ni_cnd, &nvap->na_vattr); 885 vput(ndp->ni_dvp); 886 nfsvno_relpathbuf(ndp); 887 } else { 888 if (nvap->na_type != VFIFO && 889 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) { 890 vrele(ndp->ni_startdir); 891 nfsvno_relpathbuf(ndp); 892 vput(ndp->ni_dvp); 893 goto out; 894 } 895 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp, 896 &ndp->ni_cnd, &nvap->na_vattr); 897 vput(ndp->ni_dvp); 898 nfsvno_relpathbuf(ndp); 899 vrele(ndp->ni_startdir); 900 /* 901 * Since VOP_MKNOD returns the ni_vp, I can't 902 * see any reason to do the lookup. 903 */ 904 } 905 906out: 907 NFSEXITCODE(error); 908 return (error); 909} 910 911/* 912 * Mkdir vnode op. 913 */ 914int 915nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid, 916 struct ucred *cred, struct thread *p, struct nfsexstuff *exp) 917{ 918 int error = 0; 919 920 if (ndp->ni_vp != NULL) { 921 if (ndp->ni_dvp == ndp->ni_vp) 922 vrele(ndp->ni_dvp); 923 else 924 vput(ndp->ni_dvp); 925 vrele(ndp->ni_vp); 926 nfsvno_relpathbuf(ndp); 927 error = EEXIST; 928 goto out; 929 } 930 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 931 &nvap->na_vattr); 932 vput(ndp->ni_dvp); 933 nfsvno_relpathbuf(ndp); 934 935out: 936 NFSEXITCODE(error); 937 return (error); 938} 939 940/* 941 * symlink vnode op. 942 */ 943int 944nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp, 945 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p, 946 struct nfsexstuff *exp) 947{ 948 int error = 0; 949 950 if (ndp->ni_vp) { 951 vrele(ndp->ni_startdir); 952 nfsvno_relpathbuf(ndp); 953 if (ndp->ni_dvp == ndp->ni_vp) 954 vrele(ndp->ni_dvp); 955 else 956 vput(ndp->ni_dvp); 957 vrele(ndp->ni_vp); 958 error = EEXIST; 959 goto out; 960 } 961 962 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, 963 &nvap->na_vattr, pathcp); 964 vput(ndp->ni_dvp); 965 vrele(ndp->ni_startdir); 966 nfsvno_relpathbuf(ndp); 967 /* 968 * Although FreeBSD still had the lookup code in 969 * it for 7/current, there doesn't seem to be any 970 * point, since VOP_SYMLINK() returns the ni_vp. 971 * Just vput it for v2. 972 */ 973 if (!not_v2 && !error) 974 vput(ndp->ni_vp); 975 976out: 977 NFSEXITCODE(error); 978 return (error); 979} 980 981/* 982 * Parse symbolic link arguments. 983 * This function has an ugly side effect. It will MALLOC() an area for 984 * the symlink and set iov_base to point to it, only if it succeeds. 985 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must 986 * be FREE'd later. 987 */ 988int 989nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap, 990 struct thread *p, char **pathcpp, int *lenp) 991{ 992 u_int32_t *tl; 993 char *pathcp = NULL; 994 int error = 0, len; 995 struct nfsv2_sattr *sp; 996 997 *pathcpp = NULL; 998 *lenp = 0; 999 if ((nd->nd_flag & ND_NFSV3) && 1000 (error = nfsrv_sattr(nd, nvap, NULL, NULL, p))) 1001 goto nfsmout; 1002 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 1003 len = fxdr_unsigned(int, *tl); 1004 if (len > NFS_MAXPATHLEN || len <= 0) { 1005 error = EBADRPC; 1006 goto nfsmout; 1007 } 1008 MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK); 1009 error = nfsrv_mtostr(nd, pathcp, len); 1010 if (error) 1011 goto nfsmout; 1012 if (nd->nd_flag & ND_NFSV2) { 1013 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 1014 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode); 1015 } 1016 *pathcpp = pathcp; 1017 *lenp = len; 1018 NFSEXITCODE2(0, nd); 1019 return (0); 1020nfsmout: 1021 if (pathcp) 1022 free(pathcp, M_TEMP); 1023 NFSEXITCODE2(error, nd); 1024 return (error); 1025} 1026 1027/* 1028 * Remove a non-directory object. 1029 */ 1030int 1031nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1032 struct thread *p, struct nfsexstuff *exp) 1033{ 1034 struct vnode *vp; 1035 int error = 0; 1036 1037 vp = ndp->ni_vp; 1038 if (vp->v_type == VDIR) 1039 error = NFSERR_ISDIR; 1040 else if (is_v4) 1041 error = nfsrv_checkremove(vp, 1, p); 1042 if (!error) 1043 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd); 1044 if (ndp->ni_dvp == vp) 1045 vrele(ndp->ni_dvp); 1046 else 1047 vput(ndp->ni_dvp); 1048 vput(vp); 1049 NFSEXITCODE(error); 1050 return (error); 1051} 1052 1053/* 1054 * Remove a directory. 1055 */ 1056int 1057nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred, 1058 struct thread *p, struct nfsexstuff *exp) 1059{ 1060 struct vnode *vp; 1061 int error = 0; 1062 1063 vp = ndp->ni_vp; 1064 if (vp->v_type != VDIR) { 1065 error = ENOTDIR; 1066 goto out; 1067 } 1068 /* 1069 * No rmdir "." please. 1070 */ 1071 if (ndp->ni_dvp == vp) { 1072 error = EINVAL; 1073 goto out; 1074 } 1075 /* 1076 * The root of a mounted filesystem cannot be deleted. 1077 */ 1078 if (vp->v_vflag & VV_ROOT) 1079 error = EBUSY; 1080out: 1081 if (!error) 1082 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd); 1083 if (ndp->ni_dvp == vp) 1084 vrele(ndp->ni_dvp); 1085 else 1086 vput(ndp->ni_dvp); 1087 vput(vp); 1088 NFSEXITCODE(error); 1089 return (error); 1090} 1091 1092/* 1093 * Rename vnode op. 1094 */ 1095int 1096nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp, 1097 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p) 1098{ 1099 struct vnode *fvp, *tvp, *tdvp; 1100 int error = 0; 1101 1102 fvp = fromndp->ni_vp; 1103 if (ndstat) { 1104 vrele(fromndp->ni_dvp); 1105 vrele(fvp); 1106 error = ndstat; 1107 goto out1; 1108 } 1109 tdvp = tondp->ni_dvp; 1110 tvp = tondp->ni_vp; 1111 if (tvp != NULL) { 1112 if (fvp->v_type == VDIR && tvp->v_type != VDIR) { 1113 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST; 1114 goto out; 1115 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { 1116 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST; 1117 goto out; 1118 } 1119 if (tvp->v_type == VDIR && tvp->v_mountedhere) { 1120 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1121 goto out; 1122 } 1123 1124 /* 1125 * A rename to '.' or '..' results in a prematurely 1126 * unlocked vnode on FreeBSD5, so I'm just going to fail that 1127 * here. 1128 */ 1129 if ((tondp->ni_cnd.cn_namelen == 1 && 1130 tondp->ni_cnd.cn_nameptr[0] == '.') || 1131 (tondp->ni_cnd.cn_namelen == 2 && 1132 tondp->ni_cnd.cn_nameptr[0] == '.' && 1133 tondp->ni_cnd.cn_nameptr[1] == '.')) { 1134 error = EINVAL; 1135 goto out; 1136 } 1137 } 1138 if (fvp->v_type == VDIR && fvp->v_mountedhere) { 1139 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1140 goto out; 1141 } 1142 if (fvp->v_mount != tdvp->v_mount) { 1143 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV; 1144 goto out; 1145 } 1146 if (fvp == tdvp) { 1147 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL; 1148 goto out; 1149 } 1150 if (fvp == tvp) { 1151 /* 1152 * If source and destination are the same, there is nothing to 1153 * do. Set error to -1 to indicate this. 1154 */ 1155 error = -1; 1156 goto out; 1157 } 1158 if (ndflag & ND_NFSV4) { 1159 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) { 1160 error = nfsrv_checkremove(fvp, 0, p); 1161 NFSVOPUNLOCK(fvp, 0); 1162 } else 1163 error = EPERM; 1164 if (tvp && !error) 1165 error = nfsrv_checkremove(tvp, 1, p); 1166 } else { 1167 /* 1168 * For NFSv2 and NFSv3, try to get rid of the delegation, so 1169 * that the NFSv4 client won't be confused by the rename. 1170 * Since nfsd_recalldelegation() can only be called on an 1171 * unlocked vnode at this point and fvp is the file that will 1172 * still exist after the rename, just do fvp. 1173 */ 1174 nfsd_recalldelegation(fvp, p); 1175 } 1176out: 1177 if (!error) { 1178 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp, 1179 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp, 1180 &tondp->ni_cnd); 1181 } else { 1182 if (tdvp == tvp) 1183 vrele(tdvp); 1184 else 1185 vput(tdvp); 1186 if (tvp) 1187 vput(tvp); 1188 vrele(fromndp->ni_dvp); 1189 vrele(fvp); 1190 if (error == -1) 1191 error = 0; 1192 } 1193 vrele(tondp->ni_startdir); 1194 nfsvno_relpathbuf(tondp); 1195out1: 1196 vrele(fromndp->ni_startdir); 1197 nfsvno_relpathbuf(fromndp); 1198 NFSEXITCODE(error); 1199 return (error); 1200} 1201 1202/* 1203 * Link vnode op. 1204 */ 1205int 1206nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred, 1207 struct thread *p, struct nfsexstuff *exp) 1208{ 1209 struct vnode *xp; 1210 int error = 0; 1211 1212 xp = ndp->ni_vp; 1213 if (xp != NULL) { 1214 error = EEXIST; 1215 } else { 1216 xp = ndp->ni_dvp; 1217 if (vp->v_mount != xp->v_mount) 1218 error = EXDEV; 1219 } 1220 if (!error) { 1221 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1222 if ((vp->v_iflag & VI_DOOMED) == 0) 1223 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd); 1224 else 1225 error = EPERM; 1226 if (ndp->ni_dvp == vp) 1227 vrele(ndp->ni_dvp); 1228 else 1229 vput(ndp->ni_dvp); 1230 NFSVOPUNLOCK(vp, 0); 1231 } else { 1232 if (ndp->ni_dvp == ndp->ni_vp) 1233 vrele(ndp->ni_dvp); 1234 else 1235 vput(ndp->ni_dvp); 1236 if (ndp->ni_vp) 1237 vrele(ndp->ni_vp); 1238 } 1239 nfsvno_relpathbuf(ndp); 1240 NFSEXITCODE(error); 1241 return (error); 1242} 1243 1244/* 1245 * Do the fsync() appropriate for the commit. 1246 */ 1247int 1248nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred, 1249 struct thread *td) 1250{ 1251 int error = 0; 1252 1253 /* 1254 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of 1255 * file is done. At this time VOP_FSYNC does not accept offset and 1256 * byte count parameters so call VOP_FSYNC the whole file for now. 1257 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3. 1258 */ 1259 if (cnt == 0 || cnt > MAX_COMMIT_COUNT) { 1260 /* 1261 * Give up and do the whole thing 1262 */ 1263 if (vp->v_object && 1264 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 1265 VM_OBJECT_LOCK(vp->v_object); 1266 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); 1267 VM_OBJECT_UNLOCK(vp->v_object); 1268 } 1269 error = VOP_FSYNC(vp, MNT_WAIT, td); 1270 } else { 1271 /* 1272 * Locate and synchronously write any buffers that fall 1273 * into the requested range. Note: we are assuming that 1274 * f_iosize is a power of 2. 1275 */ 1276 int iosize = vp->v_mount->mnt_stat.f_iosize; 1277 int iomask = iosize - 1; 1278 struct bufobj *bo; 1279 daddr_t lblkno; 1280 1281 /* 1282 * Align to iosize boundry, super-align to page boundry. 1283 */ 1284 if (off & iomask) { 1285 cnt += off & iomask; 1286 off &= ~(u_quad_t)iomask; 1287 } 1288 if (off & PAGE_MASK) { 1289 cnt += off & PAGE_MASK; 1290 off &= ~(u_quad_t)PAGE_MASK; 1291 } 1292 lblkno = off / iosize; 1293 1294 if (vp->v_object && 1295 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 1296 VM_OBJECT_LOCK(vp->v_object); 1297 vm_object_page_clean(vp->v_object, off, off + cnt, 1298 OBJPC_SYNC); 1299 VM_OBJECT_UNLOCK(vp->v_object); 1300 } 1301 1302 bo = &vp->v_bufobj; 1303 BO_LOCK(bo); 1304 while (cnt > 0) { 1305 struct buf *bp; 1306 1307 /* 1308 * If we have a buffer and it is marked B_DELWRI we 1309 * have to lock and write it. Otherwise the prior 1310 * write is assumed to have already been committed. 1311 * 1312 * gbincore() can return invalid buffers now so we 1313 * have to check that bit as well (though B_DELWRI 1314 * should not be set if B_INVAL is set there could be 1315 * a race here since we haven't locked the buffer). 1316 */ 1317 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) { 1318 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | 1319 LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) { 1320 BO_LOCK(bo); 1321 continue; /* retry */ 1322 } 1323 if ((bp->b_flags & (B_DELWRI|B_INVAL)) == 1324 B_DELWRI) { 1325 bremfree(bp); 1326 bp->b_flags &= ~B_ASYNC; 1327 bwrite(bp); 1328 ++nfs_commit_miss; 1329 } else 1330 BUF_UNLOCK(bp); 1331 BO_LOCK(bo); 1332 } 1333 ++nfs_commit_blks; 1334 if (cnt < iosize) 1335 break; 1336 cnt -= iosize; 1337 ++lblkno; 1338 } 1339 BO_UNLOCK(bo); 1340 } 1341 NFSEXITCODE(error); 1342 return (error); 1343} 1344 1345/* 1346 * Statfs vnode op. 1347 */ 1348int 1349nfsvno_statfs(struct vnode *vp, struct statfs *sf) 1350{ 1351 int error; 1352 1353 error = VFS_STATFS(vp->v_mount, sf); 1354 if (error == 0) { 1355 /* 1356 * Since NFS handles these values as unsigned on the 1357 * wire, there is no way to represent negative values, 1358 * so set them to 0. Without this, they will appear 1359 * to be very large positive values for clients like 1360 * Solaris10. 1361 */ 1362 if (sf->f_bavail < 0) 1363 sf->f_bavail = 0; 1364 if (sf->f_ffree < 0) 1365 sf->f_ffree = 0; 1366 } 1367 NFSEXITCODE(error); 1368 return (error); 1369} 1370 1371/* 1372 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but 1373 * must handle nfsrv_opencheck() calls after any other access checks. 1374 */ 1375void 1376nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp, 1377 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp, 1378 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create, 1379 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p, 1380 struct nfsexstuff *exp, struct vnode **vpp) 1381{ 1382 struct vnode *vp = NULL; 1383 u_quad_t tempsize; 1384 struct nfsexstuff nes; 1385 1386 if (ndp->ni_vp == NULL) 1387 nd->nd_repstat = nfsrv_opencheck(clientid, 1388 stateidp, stp, NULL, nd, p, nd->nd_repstat); 1389 if (!nd->nd_repstat) { 1390 if (ndp->ni_vp == NULL) { 1391 vrele(ndp->ni_startdir); 1392 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp, 1393 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr); 1394 vput(ndp->ni_dvp); 1395 nfsvno_relpathbuf(ndp); 1396 if (!nd->nd_repstat) { 1397 if (*exclusive_flagp) { 1398 *exclusive_flagp = 0; 1399 NFSVNO_ATTRINIT(nvap); 1400 nvap->na_atime.tv_sec = cverf[0]; 1401 nvap->na_atime.tv_nsec = cverf[1]; 1402 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp, 1403 &nvap->na_vattr, cred); 1404 } else { 1405 nfsrv_fixattr(nd, ndp->ni_vp, nvap, 1406 aclp, p, attrbitp, exp); 1407 } 1408 } 1409 vp = ndp->ni_vp; 1410 } else { 1411 if (ndp->ni_startdir) 1412 vrele(ndp->ni_startdir); 1413 nfsvno_relpathbuf(ndp); 1414 vp = ndp->ni_vp; 1415 if (create == NFSV4OPEN_CREATE) { 1416 if (ndp->ni_dvp == vp) 1417 vrele(ndp->ni_dvp); 1418 else 1419 vput(ndp->ni_dvp); 1420 } 1421 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) { 1422 if (ndp->ni_cnd.cn_flags & RDONLY) 1423 NFSVNO_SETEXRDONLY(&nes); 1424 else 1425 NFSVNO_EXINIT(&nes); 1426 nd->nd_repstat = nfsvno_accchk(vp, 1427 VWRITE, cred, &nes, p, 1428 NFSACCCHK_NOOVERRIDE, 1429 NFSACCCHK_VPISLOCKED, NULL); 1430 nd->nd_repstat = nfsrv_opencheck(clientid, 1431 stateidp, stp, vp, nd, p, nd->nd_repstat); 1432 if (!nd->nd_repstat) { 1433 tempsize = nvap->na_size; 1434 NFSVNO_ATTRINIT(nvap); 1435 nvap->na_size = tempsize; 1436 nd->nd_repstat = VOP_SETATTR(vp, 1437 &nvap->na_vattr, cred); 1438 } 1439 } else if (vp->v_type == VREG) { 1440 nd->nd_repstat = nfsrv_opencheck(clientid, 1441 stateidp, stp, vp, nd, p, nd->nd_repstat); 1442 } 1443 } 1444 } else { 1445 if (ndp->ni_cnd.cn_flags & HASBUF) 1446 nfsvno_relpathbuf(ndp); 1447 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) { 1448 vrele(ndp->ni_startdir); 1449 if (ndp->ni_dvp == ndp->ni_vp) 1450 vrele(ndp->ni_dvp); 1451 else 1452 vput(ndp->ni_dvp); 1453 if (ndp->ni_vp) 1454 vput(ndp->ni_vp); 1455 } 1456 } 1457 *vpp = vp; 1458 1459 NFSEXITCODE2(0, nd); 1460} 1461 1462/* 1463 * Updates the file rev and sets the mtime and ctime 1464 * to the current clock time, returning the va_filerev and va_Xtime 1465 * values. 1466 */ 1467void 1468nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap, 1469 struct ucred *cred, struct thread *p) 1470{ 1471 struct vattr va; 1472 1473 VATTR_NULL(&va); 1474 getnanotime(&va.va_mtime); 1475 (void) VOP_SETATTR(vp, &va, cred); 1476 (void) nfsvno_getattr(vp, nvap, cred, p, 1); 1477} 1478 1479/* 1480 * Glue routine to nfsv4_fillattr(). 1481 */ 1482int 1483nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp, 1484 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, 1485 struct ucred *cred, struct thread *p, int isdgram, int reterr, 1486 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) 1487{ 1488 int error; 1489 1490 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror, 1491 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root, 1492 mounted_on_fileno); 1493 NFSEXITCODE2(0, nd); 1494 return (error); 1495} 1496 1497/* Since the Readdir vnode ops vary, put the entire functions in here. */ 1498/* 1499 * nfs readdir service 1500 * - mallocs what it thinks is enough to read 1501 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR 1502 * - calls VOP_READDIR() 1503 * - loops around building the reply 1504 * if the output generated exceeds count break out of loop 1505 * The NFSM_CLGET macro is used here so that the reply will be packed 1506 * tightly in mbuf clusters. 1507 * - it trims out records with d_fileno == 0 1508 * this doesn't matter for Unix clients, but they might confuse clients 1509 * for other os'. 1510 * - it trims out records with d_type == DT_WHT 1511 * these cannot be seen through NFS (unless we extend the protocol) 1512 * The alternate call nfsrvd_readdirplus() does lookups as well. 1513 * PS: The NFS protocol spec. does not clarify what the "count" byte 1514 * argument is a count of.. just name strings and file id's or the 1515 * entire reply rpc or ... 1516 * I tried just file name and id sizes and it confused the Sun client, 1517 * so I am using the full rpc size now. The "paranoia.." comment refers 1518 * to including the status longwords that are not a part of the dir. 1519 * "entry" structures, but are in the rpc. 1520 */ 1521int 1522nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram, 1523 struct vnode *vp, struct thread *p, struct nfsexstuff *exp) 1524{ 1525 struct dirent *dp; 1526 u_int32_t *tl; 1527 int dirlen; 1528 char *cpos, *cend, *rbuf; 1529 struct nfsvattr at; 1530 int nlen, error = 0, getret = 1; 1531 int siz, cnt, fullsiz, eofflag, ncookies; 1532 u_int64_t off, toff, verf; 1533 u_long *cookies = NULL, *cookiep; 1534 struct uio io; 1535 struct iovec iv; 1536 int not_zfs; 1537 1538 if (nd->nd_repstat) { 1539 nfsrv_postopattr(nd, getret, &at); 1540 goto out; 1541 } 1542 if (nd->nd_flag & ND_NFSV2) { 1543 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1544 off = fxdr_unsigned(u_quad_t, *tl++); 1545 } else { 1546 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); 1547 off = fxdr_hyper(tl); 1548 tl += 2; 1549 verf = fxdr_hyper(tl); 1550 tl += 2; 1551 } 1552 toff = off; 1553 cnt = fxdr_unsigned(int, *tl); 1554 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 1555 cnt = NFS_SRVMAXDATA(nd); 1556 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 1557 fullsiz = siz; 1558 if (nd->nd_flag & ND_NFSV3) { 1559 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, 1560 p, 1); 1561#if 0 1562 /* 1563 * va_filerev is not sufficient as a cookie verifier, 1564 * since it is not supposed to change when entries are 1565 * removed/added unless that offset cookies returned to 1566 * the client are no longer valid. 1567 */ 1568 if (!nd->nd_repstat && toff && verf != at.na_filerev) 1569 nd->nd_repstat = NFSERR_BAD_COOKIE; 1570#endif 1571 } 1572 if (nd->nd_repstat == 0 && cnt == 0) { 1573 if (nd->nd_flag & ND_NFSV2) 1574 /* NFSv2 does not have NFSERR_TOOSMALL */ 1575 nd->nd_repstat = EPERM; 1576 else 1577 nd->nd_repstat = NFSERR_TOOSMALL; 1578 } 1579 if (!nd->nd_repstat) 1580 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 1581 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1582 NFSACCCHK_VPISLOCKED, NULL); 1583 if (nd->nd_repstat) { 1584 vput(vp); 1585 if (nd->nd_flag & ND_NFSV3) 1586 nfsrv_postopattr(nd, getret, &at); 1587 goto out; 1588 } 1589 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs"); 1590 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK); 1591again: 1592 eofflag = 0; 1593 if (cookies) { 1594 free((caddr_t)cookies, M_TEMP); 1595 cookies = NULL; 1596 } 1597 1598 iv.iov_base = rbuf; 1599 iv.iov_len = siz; 1600 io.uio_iov = &iv; 1601 io.uio_iovcnt = 1; 1602 io.uio_offset = (off_t)off; 1603 io.uio_resid = siz; 1604 io.uio_segflg = UIO_SYSSPACE; 1605 io.uio_rw = UIO_READ; 1606 io.uio_td = NULL; 1607 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 1608 &cookies); 1609 off = (u_int64_t)io.uio_offset; 1610 if (io.uio_resid) 1611 siz -= io.uio_resid; 1612 1613 if (!cookies && !nd->nd_repstat) 1614 nd->nd_repstat = NFSERR_PERM; 1615 if (nd->nd_flag & ND_NFSV3) { 1616 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1); 1617 if (!nd->nd_repstat) 1618 nd->nd_repstat = getret; 1619 } 1620 1621 /* 1622 * Handles the failed cases. nd->nd_repstat == 0 past here. 1623 */ 1624 if (nd->nd_repstat) { 1625 vput(vp); 1626 free((caddr_t)rbuf, M_TEMP); 1627 if (cookies) 1628 free((caddr_t)cookies, M_TEMP); 1629 if (nd->nd_flag & ND_NFSV3) 1630 nfsrv_postopattr(nd, getret, &at); 1631 goto out; 1632 } 1633 /* 1634 * If nothing read, return eof 1635 * rpc reply 1636 */ 1637 if (siz == 0) { 1638 vput(vp); 1639 if (nd->nd_flag & ND_NFSV2) { 1640 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1641 } else { 1642 nfsrv_postopattr(nd, getret, &at); 1643 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1644 txdr_hyper(at.na_filerev, tl); 1645 tl += 2; 1646 } 1647 *tl++ = newnfs_false; 1648 *tl = newnfs_true; 1649 FREE((caddr_t)rbuf, M_TEMP); 1650 FREE((caddr_t)cookies, M_TEMP); 1651 goto out; 1652 } 1653 1654 /* 1655 * Check for degenerate cases of nothing useful read. 1656 * If so go try again 1657 */ 1658 cpos = rbuf; 1659 cend = rbuf + siz; 1660 dp = (struct dirent *)cpos; 1661 cookiep = cookies; 1662 1663 /* 1664 * For some reason FreeBSD's ufs_readdir() chooses to back the 1665 * directory offset up to a block boundary, so it is necessary to 1666 * skip over the records that precede the requested offset. This 1667 * requires the assumption that file offset cookies monotonically 1668 * increase. 1669 * Since the offset cookies don't monotonically increase for ZFS, 1670 * this is not done when ZFS is the file system. 1671 */ 1672 while (cpos < cend && ncookies > 0 && 1673 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 1674 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) { 1675 cpos += dp->d_reclen; 1676 dp = (struct dirent *)cpos; 1677 cookiep++; 1678 ncookies--; 1679 } 1680 if (cpos >= cend || ncookies == 0) { 1681 siz = fullsiz; 1682 toff = off; 1683 goto again; 1684 } 1685 vput(vp); 1686 1687 /* 1688 * dirlen is the size of the reply, including all XDR and must 1689 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate 1690 * if the XDR should be included in "count", but to be safe, we do. 1691 * (Include the two booleans at the end of the reply in dirlen now.) 1692 */ 1693 if (nd->nd_flag & ND_NFSV3) { 1694 nfsrv_postopattr(nd, getret, &at); 1695 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1696 txdr_hyper(at.na_filerev, tl); 1697 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 1698 } else { 1699 dirlen = 2 * NFSX_UNSIGNED; 1700 } 1701 1702 /* Loop through the records and build reply */ 1703 while (cpos < cend && ncookies > 0) { 1704 nlen = dp->d_namlen; 1705 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 1706 nlen <= NFS_MAXNAMLEN) { 1707 if (nd->nd_flag & ND_NFSV3) 1708 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 1709 else 1710 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen)); 1711 if (dirlen > cnt) { 1712 eofflag = 0; 1713 break; 1714 } 1715 1716 /* 1717 * Build the directory record xdr from 1718 * the dirent entry. 1719 */ 1720 if (nd->nd_flag & ND_NFSV3) { 1721 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 1722 *tl++ = newnfs_true; 1723 *tl++ = 0; 1724 } else { 1725 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1726 *tl++ = newnfs_true; 1727 } 1728 *tl = txdr_unsigned(dp->d_fileno); 1729 (void) nfsm_strtom(nd, dp->d_name, nlen); 1730 if (nd->nd_flag & ND_NFSV3) { 1731 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1732 *tl++ = 0; 1733 } else 1734 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 1735 *tl = txdr_unsigned(*cookiep); 1736 } 1737 cpos += dp->d_reclen; 1738 dp = (struct dirent *)cpos; 1739 cookiep++; 1740 ncookies--; 1741 } 1742 if (cpos < cend) 1743 eofflag = 0; 1744 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 1745 *tl++ = newnfs_false; 1746 if (eofflag) 1747 *tl = newnfs_true; 1748 else 1749 *tl = newnfs_false; 1750 FREE((caddr_t)rbuf, M_TEMP); 1751 FREE((caddr_t)cookies, M_TEMP); 1752 1753out: 1754 NFSEXITCODE2(0, nd); 1755 return (0); 1756nfsmout: 1757 vput(vp); 1758 NFSEXITCODE2(error, nd); 1759 return (error); 1760} 1761 1762/* 1763 * Readdirplus for V3 and Readdir for V4. 1764 */ 1765int 1766nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram, 1767 struct vnode *vp, struct thread *p, struct nfsexstuff *exp) 1768{ 1769 struct dirent *dp; 1770 u_int32_t *tl; 1771 int dirlen; 1772 char *cpos, *cend, *rbuf; 1773 struct vnode *nvp; 1774 fhandle_t nfh; 1775 struct nfsvattr nva, at, *nvap = &nva; 1776 struct mbuf *mb0, *mb1; 1777 struct nfsreferral *refp; 1778 int nlen, r, error = 0, getret = 1, usevget = 1; 1779 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt; 1780 caddr_t bpos0, bpos1; 1781 u_int64_t off, toff, verf; 1782 u_long *cookies = NULL, *cookiep; 1783 nfsattrbit_t attrbits, rderrbits, savbits; 1784 struct uio io; 1785 struct iovec iv; 1786 struct componentname cn; 1787 int at_root, needs_unbusy, not_zfs, supports_nfsv4acls; 1788 struct mount *mp, *new_mp; 1789 uint64_t mounted_on_fileno; 1790 1791 if (nd->nd_repstat) { 1792 nfsrv_postopattr(nd, getret, &at); 1793 goto out; 1794 } 1795 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); 1796 off = fxdr_hyper(tl); 1797 toff = off; 1798 tl += 2; 1799 verf = fxdr_hyper(tl); 1800 tl += 2; 1801 siz = fxdr_unsigned(int, *tl++); 1802 cnt = fxdr_unsigned(int, *tl); 1803 1804 /* 1805 * Use the server's maximum data transfer size as the upper bound 1806 * on reply datalen. 1807 */ 1808 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0) 1809 cnt = NFS_SRVMAXDATA(nd); 1810 1811 /* 1812 * siz is a "hint" of how much directory information (name, fileid, 1813 * cookie) should be in the reply. At least one client "hints" 0, 1814 * so I set it to cnt for that case. I also round it up to the 1815 * next multiple of DIRBLKSIZ. 1816 */ 1817 if (siz <= 0) 1818 siz = cnt; 1819 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1)); 1820 1821 if (nd->nd_flag & ND_NFSV4) { 1822 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); 1823 if (error) 1824 goto nfsmout; 1825 NFSSET_ATTRBIT(&savbits, &attrbits); 1826 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits); 1827 NFSZERO_ATTRBIT(&rderrbits); 1828 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR); 1829 } else { 1830 NFSZERO_ATTRBIT(&attrbits); 1831 } 1832 fullsiz = siz; 1833 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1); 1834 if (!nd->nd_repstat) { 1835 if (off && verf != at.na_filerev) { 1836 /* 1837 * va_filerev is not sufficient as a cookie verifier, 1838 * since it is not supposed to change when entries are 1839 * removed/added unless that offset cookies returned to 1840 * the client are no longer valid. 1841 */ 1842#if 0 1843 if (nd->nd_flag & ND_NFSV4) { 1844 nd->nd_repstat = NFSERR_NOTSAME; 1845 } else { 1846 nd->nd_repstat = NFSERR_BAD_COOKIE; 1847 } 1848#endif 1849 } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) { 1850 nd->nd_repstat = NFSERR_BAD_COOKIE; 1851 } 1852 } 1853 if (!nd->nd_repstat && vp->v_type != VDIR) 1854 nd->nd_repstat = NFSERR_NOTDIR; 1855 if (!nd->nd_repstat && cnt == 0) 1856 nd->nd_repstat = NFSERR_TOOSMALL; 1857 if (!nd->nd_repstat) 1858 nd->nd_repstat = nfsvno_accchk(vp, VEXEC, 1859 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE, 1860 NFSACCCHK_VPISLOCKED, NULL); 1861 if (nd->nd_repstat) { 1862 vput(vp); 1863 if (nd->nd_flag & ND_NFSV3) 1864 nfsrv_postopattr(nd, getret, &at); 1865 goto out; 1866 } 1867 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs"); 1868 1869 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK); 1870again: 1871 eofflag = 0; 1872 if (cookies) { 1873 free((caddr_t)cookies, M_TEMP); 1874 cookies = NULL; 1875 } 1876 1877 iv.iov_base = rbuf; 1878 iv.iov_len = siz; 1879 io.uio_iov = &iv; 1880 io.uio_iovcnt = 1; 1881 io.uio_offset = (off_t)off; 1882 io.uio_resid = siz; 1883 io.uio_segflg = UIO_SYSSPACE; 1884 io.uio_rw = UIO_READ; 1885 io.uio_td = NULL; 1886 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies, 1887 &cookies); 1888 off = (u_int64_t)io.uio_offset; 1889 if (io.uio_resid) 1890 siz -= io.uio_resid; 1891 1892 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1); 1893 1894 if (!cookies && !nd->nd_repstat) 1895 nd->nd_repstat = NFSERR_PERM; 1896 if (!nd->nd_repstat) 1897 nd->nd_repstat = getret; 1898 if (nd->nd_repstat) { 1899 vput(vp); 1900 if (cookies) 1901 free((caddr_t)cookies, M_TEMP); 1902 free((caddr_t)rbuf, M_TEMP); 1903 if (nd->nd_flag & ND_NFSV3) 1904 nfsrv_postopattr(nd, getret, &at); 1905 goto out; 1906 } 1907 /* 1908 * If nothing read, return eof 1909 * rpc reply 1910 */ 1911 if (siz == 0) { 1912 vput(vp); 1913 if (nd->nd_flag & ND_NFSV3) 1914 nfsrv_postopattr(nd, getret, &at); 1915 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 1916 txdr_hyper(at.na_filerev, tl); 1917 tl += 2; 1918 *tl++ = newnfs_false; 1919 *tl = newnfs_true; 1920 free((caddr_t)cookies, M_TEMP); 1921 free((caddr_t)rbuf, M_TEMP); 1922 goto out; 1923 } 1924 1925 /* 1926 * Check for degenerate cases of nothing useful read. 1927 * If so go try again 1928 */ 1929 cpos = rbuf; 1930 cend = rbuf + siz; 1931 dp = (struct dirent *)cpos; 1932 cookiep = cookies; 1933 1934 /* 1935 * For some reason FreeBSD's ufs_readdir() chooses to back the 1936 * directory offset up to a block boundary, so it is necessary to 1937 * skip over the records that precede the requested offset. This 1938 * requires the assumption that file offset cookies monotonically 1939 * increase. 1940 * Since the offset cookies don't monotonically increase for ZFS, 1941 * this is not done when ZFS is the file system. 1942 */ 1943 while (cpos < cend && ncookies > 0 && 1944 (dp->d_fileno == 0 || dp->d_type == DT_WHT || 1945 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) || 1946 ((nd->nd_flag & ND_NFSV4) && 1947 ((dp->d_namlen == 1 && dp->d_name[0] == '.') || 1948 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) { 1949 cpos += dp->d_reclen; 1950 dp = (struct dirent *)cpos; 1951 cookiep++; 1952 ncookies--; 1953 } 1954 if (cpos >= cend || ncookies == 0) { 1955 siz = fullsiz; 1956 toff = off; 1957 goto again; 1958 } 1959 1960 /* 1961 * Busy the file system so that the mount point won't go away 1962 * and, as such, VFS_VGET() can be used safely. 1963 */ 1964 mp = vp->v_mount; 1965 vfs_ref(mp); 1966 NFSVOPUNLOCK(vp, 0); 1967 nd->nd_repstat = vfs_busy(mp, 0); 1968 vfs_rel(mp); 1969 if (nd->nd_repstat != 0) { 1970 vrele(vp); 1971 free(cookies, M_TEMP); 1972 free(rbuf, M_TEMP); 1973 if (nd->nd_flag & ND_NFSV3) 1974 nfsrv_postopattr(nd, getret, &at); 1975 goto out; 1976 } 1977 1978 /* 1979 * Save this position, in case there is an error before one entry 1980 * is created. 1981 */ 1982 mb0 = nd->nd_mb; 1983 bpos0 = nd->nd_bpos; 1984 1985 /* 1986 * Fill in the first part of the reply. 1987 * dirlen is the reply length in bytes and cannot exceed cnt. 1988 * (Include the two booleans at the end of the reply in dirlen now, 1989 * so we recognize when we have exceeded cnt.) 1990 */ 1991 if (nd->nd_flag & ND_NFSV3) { 1992 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED; 1993 nfsrv_postopattr(nd, getret, &at); 1994 } else { 1995 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED; 1996 } 1997 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); 1998 txdr_hyper(at.na_filerev, tl); 1999 2000 /* 2001 * Save this position, in case there is an empty reply needed. 2002 */ 2003 mb1 = nd->nd_mb; 2004 bpos1 = nd->nd_bpos; 2005 2006 /* Loop through the records and build reply */ 2007 entrycnt = 0; 2008 while (cpos < cend && ncookies > 0 && dirlen < cnt) { 2009 nlen = dp->d_namlen; 2010 if (dp->d_fileno != 0 && dp->d_type != DT_WHT && 2011 nlen <= NFS_MAXNAMLEN && 2012 ((nd->nd_flag & ND_NFSV3) || nlen > 2 || 2013 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.')) 2014 || (nlen == 1 && dp->d_name[0] != '.'))) { 2015 /* 2016 * Save the current position in the reply, in case 2017 * this entry exceeds cnt. 2018 */ 2019 mb1 = nd->nd_mb; 2020 bpos1 = nd->nd_bpos; 2021 2022 /* 2023 * For readdir_and_lookup get the vnode using 2024 * the file number. 2025 */ 2026 nvp = NULL; 2027 refp = NULL; 2028 r = 0; 2029 at_root = 0; 2030 needs_unbusy = 0; 2031 new_mp = mp; 2032 mounted_on_fileno = (uint64_t)dp->d_fileno; 2033 if ((nd->nd_flag & ND_NFSV3) || 2034 NFSNONZERO_ATTRBIT(&savbits)) { 2035 if (nd->nd_flag & ND_NFSV4) 2036 refp = nfsv4root_getreferral(NULL, 2037 vp, dp->d_fileno); 2038 if (refp == NULL) { 2039 if (usevget) 2040 r = VFS_VGET(mp, dp->d_fileno, 2041 LK_SHARED, &nvp); 2042 else 2043 r = EOPNOTSUPP; 2044 if (r == EOPNOTSUPP) { 2045 if (usevget) { 2046 usevget = 0; 2047 cn.cn_nameiop = LOOKUP; 2048 cn.cn_lkflags = 2049 LK_SHARED | 2050 LK_RETRY; 2051 cn.cn_cred = 2052 nd->nd_cred; 2053 cn.cn_thread = p; 2054 } 2055 cn.cn_nameptr = dp->d_name; 2056 cn.cn_namelen = nlen; 2057 cn.cn_flags = ISLASTCN | 2058 NOFOLLOW | LOCKLEAF | 2059 MPSAFE; 2060 if (nlen == 2 && 2061 dp->d_name[0] == '.' && 2062 dp->d_name[1] == '.') 2063 cn.cn_flags |= 2064 ISDOTDOT; 2065 if (NFSVOPLOCK(vp, LK_SHARED) 2066 != 0) { 2067 nd->nd_repstat = EPERM; 2068 break; 2069 } 2070 if ((vp->v_vflag & VV_ROOT) != 0 2071 && (cn.cn_flags & ISDOTDOT) 2072 != 0) { 2073 vref(vp); 2074 nvp = vp; 2075 r = 0; 2076 } else { 2077 r = VOP_LOOKUP(vp, &nvp, 2078 &cn); 2079 if (vp != nvp) 2080 NFSVOPUNLOCK(vp, 2081 0); 2082 } 2083 } 2084 2085 /* 2086 * For NFSv4, check to see if nvp is 2087 * a mount point and get the mount 2088 * point vnode, as required. 2089 */ 2090 if (r == 0 && 2091 nfsrv_enable_crossmntpt != 0 && 2092 (nd->nd_flag & ND_NFSV4) != 0 && 2093 nvp->v_type == VDIR && 2094 nvp->v_mountedhere != NULL) { 2095 new_mp = nvp->v_mountedhere; 2096 r = vfs_busy(new_mp, 0); 2097 vput(nvp); 2098 nvp = NULL; 2099 if (r == 0) { 2100 r = VFS_ROOT(new_mp, 2101 LK_SHARED, &nvp); 2102 needs_unbusy = 1; 2103 if (r == 0) 2104 at_root = 1; 2105 } 2106 } 2107 } 2108 if (!r) { 2109 if (refp == NULL && 2110 ((nd->nd_flag & ND_NFSV3) || 2111 NFSNONZERO_ATTRBIT(&attrbits))) { 2112 r = nfsvno_getfh(nvp, &nfh, p); 2113 if (!r) 2114 r = nfsvno_getattr(nvp, nvap, 2115 nd->nd_cred, p, 1); 2116 } 2117 } else { 2118 nvp = NULL; 2119 } 2120 if (r) { 2121 if (!NFSISSET_ATTRBIT(&attrbits, 2122 NFSATTRBIT_RDATTRERROR)) { 2123 if (nvp != NULL) 2124 vput(nvp); 2125 if (needs_unbusy != 0) 2126 vfs_unbusy(new_mp); 2127 nd->nd_repstat = r; 2128 break; 2129 } 2130 } 2131 } 2132 2133 /* 2134 * Build the directory record xdr 2135 */ 2136 if (nd->nd_flag & ND_NFSV3) { 2137 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2138 *tl++ = newnfs_true; 2139 *tl++ = 0; 2140 *tl = txdr_unsigned(dp->d_fileno); 2141 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2142 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2143 *tl++ = 0; 2144 *tl = txdr_unsigned(*cookiep); 2145 nfsrv_postopattr(nd, 0, nvap); 2146 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1); 2147 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR); 2148 if (nvp != NULL) 2149 vput(nvp); 2150 } else { 2151 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); 2152 *tl++ = newnfs_true; 2153 *tl++ = 0; 2154 *tl = txdr_unsigned(*cookiep); 2155 dirlen += nfsm_strtom(nd, dp->d_name, nlen); 2156 if (nvp != NULL) { 2157 supports_nfsv4acls = 2158 nfs_supportsnfsv4acls(nvp); 2159 NFSVOPUNLOCK(nvp, 0); 2160 } else 2161 supports_nfsv4acls = 0; 2162 if (refp != NULL) { 2163 dirlen += nfsrv_putreferralattr(nd, 2164 &savbits, refp, 0, 2165 &nd->nd_repstat); 2166 if (nd->nd_repstat) { 2167 if (nvp != NULL) 2168 vrele(nvp); 2169 if (needs_unbusy != 0) 2170 vfs_unbusy(new_mp); 2171 break; 2172 } 2173 } else if (r) { 2174 dirlen += nfsvno_fillattr(nd, new_mp, 2175 nvp, nvap, &nfh, r, &rderrbits, 2176 nd->nd_cred, p, isdgram, 0, 2177 supports_nfsv4acls, at_root, 2178 mounted_on_fileno); 2179 } else { 2180 dirlen += nfsvno_fillattr(nd, new_mp, 2181 nvp, nvap, &nfh, r, &attrbits, 2182 nd->nd_cred, p, isdgram, 0, 2183 supports_nfsv4acls, at_root, 2184 mounted_on_fileno); 2185 } 2186 if (nvp != NULL) 2187 vrele(nvp); 2188 dirlen += (3 * NFSX_UNSIGNED); 2189 } 2190 if (needs_unbusy != 0) 2191 vfs_unbusy(new_mp); 2192 if (dirlen <= cnt) 2193 entrycnt++; 2194 } 2195 cpos += dp->d_reclen; 2196 dp = (struct dirent *)cpos; 2197 cookiep++; 2198 ncookies--; 2199 } 2200 vrele(vp); 2201 vfs_unbusy(mp); 2202 2203 /* 2204 * If dirlen > cnt, we must strip off the last entry. If that 2205 * results in an empty reply, report NFSERR_TOOSMALL. 2206 */ 2207 if (dirlen > cnt || nd->nd_repstat) { 2208 if (!nd->nd_repstat && entrycnt == 0) 2209 nd->nd_repstat = NFSERR_TOOSMALL; 2210 if (nd->nd_repstat) 2211 newnfs_trimtrailing(nd, mb0, bpos0); 2212 else 2213 newnfs_trimtrailing(nd, mb1, bpos1); 2214 eofflag = 0; 2215 } else if (cpos < cend) 2216 eofflag = 0; 2217 if (!nd->nd_repstat) { 2218 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2219 *tl++ = newnfs_false; 2220 if (eofflag) 2221 *tl = newnfs_true; 2222 else 2223 *tl = newnfs_false; 2224 } 2225 FREE((caddr_t)cookies, M_TEMP); 2226 FREE((caddr_t)rbuf, M_TEMP); 2227 2228out: 2229 NFSEXITCODE2(0, nd); 2230 return (0); 2231nfsmout: 2232 vput(vp); 2233 NFSEXITCODE2(error, nd); 2234 return (error); 2235} 2236 2237/* 2238 * Get the settable attributes out of the mbuf list. 2239 * (Return 0 or EBADRPC) 2240 */ 2241int 2242nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap, 2243 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2244{ 2245 u_int32_t *tl; 2246 struct nfsv2_sattr *sp; 2247 struct timeval curtime; 2248 int error = 0, toclient = 0; 2249 2250 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { 2251 case ND_NFSV2: 2252 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR); 2253 /* 2254 * Some old clients didn't fill in the high order 16bits. 2255 * --> check the low order 2 bytes for 0xffff 2256 */ 2257 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff) 2258 nvap->na_mode = nfstov_mode(sp->sa_mode); 2259 if (sp->sa_uid != newnfs_xdrneg1) 2260 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid); 2261 if (sp->sa_gid != newnfs_xdrneg1) 2262 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid); 2263 if (sp->sa_size != newnfs_xdrneg1) 2264 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size); 2265 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) { 2266#ifdef notyet 2267 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime); 2268#else 2269 nvap->na_atime.tv_sec = 2270 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec); 2271 nvap->na_atime.tv_nsec = 0; 2272#endif 2273 } 2274 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1) 2275 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime); 2276 break; 2277 case ND_NFSV3: 2278 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2279 if (*tl == newnfs_true) { 2280 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2281 nvap->na_mode = nfstov_mode(*tl); 2282 } 2283 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2284 if (*tl == newnfs_true) { 2285 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2286 nvap->na_uid = fxdr_unsigned(uid_t, *tl); 2287 } 2288 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2289 if (*tl == newnfs_true) { 2290 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2291 nvap->na_gid = fxdr_unsigned(gid_t, *tl); 2292 } 2293 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2294 if (*tl == newnfs_true) { 2295 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2296 nvap->na_size = fxdr_hyper(tl); 2297 } 2298 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2299 switch (fxdr_unsigned(int, *tl)) { 2300 case NFSV3SATTRTIME_TOCLIENT: 2301 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2302 fxdr_nfsv3time(tl, &nvap->na_atime); 2303 toclient = 1; 2304 break; 2305 case NFSV3SATTRTIME_TOSERVER: 2306 NFSGETTIME(&curtime); 2307 nvap->na_atime.tv_sec = curtime.tv_sec; 2308 nvap->na_atime.tv_nsec = curtime.tv_usec * 1000; 2309 nvap->na_vaflags |= VA_UTIMES_NULL; 2310 break; 2311 }; 2312 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2313 switch (fxdr_unsigned(int, *tl)) { 2314 case NFSV3SATTRTIME_TOCLIENT: 2315 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 2316 fxdr_nfsv3time(tl, &nvap->na_mtime); 2317 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2318 break; 2319 case NFSV3SATTRTIME_TOSERVER: 2320 NFSGETTIME(&curtime); 2321 nvap->na_mtime.tv_sec = curtime.tv_sec; 2322 nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000; 2323 if (!toclient) 2324 nvap->na_vaflags |= VA_UTIMES_NULL; 2325 break; 2326 }; 2327 break; 2328 case ND_NFSV4: 2329 error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p); 2330 }; 2331nfsmout: 2332 NFSEXITCODE2(error, nd); 2333 return (error); 2334} 2335 2336/* 2337 * Handle the setable attributes for V4. 2338 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise. 2339 */ 2340int 2341nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap, 2342 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p) 2343{ 2344 u_int32_t *tl; 2345 int attrsum = 0; 2346 int i, j; 2347 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0; 2348 int toclient = 0; 2349 u_char *cp, namestr[NFSV4_SMALLSTR + 1]; 2350 uid_t uid; 2351 gid_t gid; 2352 struct timeval curtime; 2353 2354 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup); 2355 if (error) 2356 goto nfsmout; 2357 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2358 attrsize = fxdr_unsigned(int, *tl); 2359 2360 /* 2361 * Loop around getting the setable attributes. If an unsupported 2362 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return. 2363 */ 2364 if (retnotsup) { 2365 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2366 bitpos = NFSATTRBIT_MAX; 2367 } else { 2368 bitpos = 0; 2369 } 2370 for (; bitpos < NFSATTRBIT_MAX; bitpos++) { 2371 if (attrsum > attrsize) { 2372 error = NFSERR_BADXDR; 2373 goto nfsmout; 2374 } 2375 if (NFSISSET_ATTRBIT(attrbitp, bitpos)) 2376 switch (bitpos) { 2377 case NFSATTRBIT_SIZE: 2378 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); 2379 nvap->na_size = fxdr_hyper(tl); 2380 attrsum += NFSX_HYPER; 2381 break; 2382 case NFSATTRBIT_ACL: 2383 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize, 2384 p); 2385 if (error) 2386 goto nfsmout; 2387 if (aceerr && !nd->nd_repstat) 2388 nd->nd_repstat = aceerr; 2389 attrsum += aclsize; 2390 break; 2391 case NFSATTRBIT_ARCHIVE: 2392 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2393 if (!nd->nd_repstat) 2394 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2395 attrsum += NFSX_UNSIGNED; 2396 break; 2397 case NFSATTRBIT_HIDDEN: 2398 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2399 if (!nd->nd_repstat) 2400 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2401 attrsum += NFSX_UNSIGNED; 2402 break; 2403 case NFSATTRBIT_MIMETYPE: 2404 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2405 i = fxdr_unsigned(int, *tl); 2406 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 2407 if (error) 2408 goto nfsmout; 2409 if (!nd->nd_repstat) 2410 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2411 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); 2412 break; 2413 case NFSATTRBIT_MODE: 2414 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2415 nvap->na_mode = nfstov_mode(*tl); 2416 attrsum += NFSX_UNSIGNED; 2417 break; 2418 case NFSATTRBIT_OWNER: 2419 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2420 j = fxdr_unsigned(int, *tl); 2421 if (j < 0) { 2422 error = NFSERR_BADXDR; 2423 goto nfsmout; 2424 } 2425 if (j > NFSV4_SMALLSTR) 2426 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 2427 else 2428 cp = namestr; 2429 error = nfsrv_mtostr(nd, cp, j); 2430 if (error) { 2431 if (j > NFSV4_SMALLSTR) 2432 free(cp, M_NFSSTRING); 2433 goto nfsmout; 2434 } 2435 if (!nd->nd_repstat) { 2436 nd->nd_repstat = nfsv4_strtouid(cp,j,&uid,p); 2437 if (!nd->nd_repstat) 2438 nvap->na_uid = uid; 2439 } 2440 if (j > NFSV4_SMALLSTR) 2441 free(cp, M_NFSSTRING); 2442 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 2443 break; 2444 case NFSATTRBIT_OWNERGROUP: 2445 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2446 j = fxdr_unsigned(int, *tl); 2447 if (j < 0) { 2448 error = NFSERR_BADXDR; 2449 goto nfsmout; 2450 } 2451 if (j > NFSV4_SMALLSTR) 2452 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK); 2453 else 2454 cp = namestr; 2455 error = nfsrv_mtostr(nd, cp, j); 2456 if (error) { 2457 if (j > NFSV4_SMALLSTR) 2458 free(cp, M_NFSSTRING); 2459 goto nfsmout; 2460 } 2461 if (!nd->nd_repstat) { 2462 nd->nd_repstat = nfsv4_strtogid(cp,j,&gid,p); 2463 if (!nd->nd_repstat) 2464 nvap->na_gid = gid; 2465 } 2466 if (j > NFSV4_SMALLSTR) 2467 free(cp, M_NFSSTRING); 2468 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 2469 break; 2470 case NFSATTRBIT_SYSTEM: 2471 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2472 if (!nd->nd_repstat) 2473 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2474 attrsum += NFSX_UNSIGNED; 2475 break; 2476 case NFSATTRBIT_TIMEACCESSSET: 2477 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2478 attrsum += NFSX_UNSIGNED; 2479 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 2480 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2481 fxdr_nfsv4time(tl, &nvap->na_atime); 2482 toclient = 1; 2483 attrsum += NFSX_V4TIME; 2484 } else { 2485 NFSGETTIME(&curtime); 2486 nvap->na_atime.tv_sec = curtime.tv_sec; 2487 nvap->na_atime.tv_nsec = curtime.tv_usec * 1000; 2488 nvap->na_vaflags |= VA_UTIMES_NULL; 2489 } 2490 break; 2491 case NFSATTRBIT_TIMEBACKUP: 2492 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2493 if (!nd->nd_repstat) 2494 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2495 attrsum += NFSX_V4TIME; 2496 break; 2497 case NFSATTRBIT_TIMECREATE: 2498 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2499 if (!nd->nd_repstat) 2500 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2501 attrsum += NFSX_V4TIME; 2502 break; 2503 case NFSATTRBIT_TIMEMODIFYSET: 2504 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 2505 attrsum += NFSX_UNSIGNED; 2506 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) { 2507 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); 2508 fxdr_nfsv4time(tl, &nvap->na_mtime); 2509 nvap->na_vaflags &= ~VA_UTIMES_NULL; 2510 attrsum += NFSX_V4TIME; 2511 } else { 2512 NFSGETTIME(&curtime); 2513 nvap->na_mtime.tv_sec = curtime.tv_sec; 2514 nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000; 2515 if (!toclient) 2516 nvap->na_vaflags |= VA_UTIMES_NULL; 2517 } 2518 break; 2519 default: 2520 nd->nd_repstat = NFSERR_ATTRNOTSUPP; 2521 /* 2522 * set bitpos so we drop out of the loop. 2523 */ 2524 bitpos = NFSATTRBIT_MAX; 2525 break; 2526 }; 2527 } 2528 2529 /* 2530 * some clients pad the attrlist, so we need to skip over the 2531 * padding. 2532 */ 2533 if (attrsum > attrsize) { 2534 error = NFSERR_BADXDR; 2535 } else { 2536 attrsize = NFSM_RNDUP(attrsize); 2537 if (attrsum < attrsize) 2538 error = nfsm_advance(nd, attrsize - attrsum, -1); 2539 } 2540nfsmout: 2541 NFSEXITCODE2(error, nd); 2542 return (error); 2543} 2544 2545/* 2546 * Check/setup export credentials. 2547 */ 2548int 2549nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp, 2550 struct ucred *credanon) 2551{ 2552 int error = 0; 2553 2554 /* 2555 * Check/setup credentials. 2556 */ 2557 if (nd->nd_flag & ND_GSS) 2558 exp->nes_exflag &= ~MNT_EXPORTANON; 2559 2560 /* 2561 * Check to see if the operation is allowed for this security flavor. 2562 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to 2563 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS. 2564 * Also, allow Secinfo, so that it can acquire the correct flavor(s). 2565 */ 2566 if (nfsvno_testexp(nd, exp) && 2567 nd->nd_procnum != NFSV4OP_SECINFO && 2568 nd->nd_procnum != NFSPROC_FSINFO) { 2569 if (nd->nd_flag & ND_NFSV4) 2570 error = NFSERR_WRONGSEC; 2571 else 2572 error = (NFSERR_AUTHERR | AUTH_TOOWEAK); 2573 goto out; 2574 } 2575 2576 /* 2577 * Check to see if the file system is exported V4 only. 2578 */ 2579 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) { 2580 error = NFSERR_PROGNOTV4; 2581 goto out; 2582 } 2583 2584 /* 2585 * Now, map the user credentials. 2586 * (Note that ND_AUTHNONE will only be set for an NFSv3 2587 * Fsinfo RPC. If set for anything else, this code might need 2588 * to change.) 2589 */ 2590 if (NFSVNO_EXPORTED(exp) && 2591 ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) || 2592 NFSVNO_EXPORTANON(exp) || 2593 (nd->nd_flag & ND_AUTHNONE))) { 2594 nd->nd_cred->cr_uid = credanon->cr_uid; 2595 nd->nd_cred->cr_gid = credanon->cr_gid; 2596 crsetgroups(nd->nd_cred, credanon->cr_ngroups, 2597 credanon->cr_groups); 2598 } 2599 2600out: 2601 NFSEXITCODE2(error, nd); 2602 return (error); 2603} 2604 2605/* 2606 * Check exports. 2607 */ 2608int 2609nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp, 2610 struct ucred **credp) 2611{ 2612 int i, error, *secflavors; 2613 2614 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 2615 &exp->nes_numsecflavor, &secflavors); 2616 if (error) { 2617 if (nfs_rootfhset) { 2618 exp->nes_exflag = 0; 2619 exp->nes_numsecflavor = 0; 2620 error = 0; 2621 } 2622 } else { 2623 /* Copy the security flavors. */ 2624 for (i = 0; i < exp->nes_numsecflavor; i++) 2625 exp->nes_secflavors[i] = secflavors[i]; 2626 } 2627 NFSEXITCODE(error); 2628 return (error); 2629} 2630 2631/* 2632 * Get a vnode for a file handle and export stuff. 2633 */ 2634int 2635nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam, 2636 int lktype, struct vnode **vpp, struct nfsexstuff *exp, 2637 struct ucred **credp) 2638{ 2639 int i, error, *secflavors; 2640 2641 *credp = NULL; 2642 exp->nes_numsecflavor = 0; 2643 if (VFS_NEEDSGIANT(mp)) 2644 error = ESTALE; 2645 else 2646 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, vpp); 2647 if (error != 0) 2648 /* Make sure the server replies ESTALE to the client. */ 2649 error = ESTALE; 2650 if (nam && !error) { 2651 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp, 2652 &exp->nes_numsecflavor, &secflavors); 2653 if (error) { 2654 if (nfs_rootfhset) { 2655 exp->nes_exflag = 0; 2656 exp->nes_numsecflavor = 0; 2657 error = 0; 2658 } else { 2659 vput(*vpp); 2660 } 2661 } else { 2662 /* Copy the security flavors. */ 2663 for (i = 0; i < exp->nes_numsecflavor; i++) 2664 exp->nes_secflavors[i] = secflavors[i]; 2665 } 2666 } 2667 if (error == 0 && lktype == LK_SHARED) 2668 /* 2669 * It would be much better to pass lktype to VFS_FHTOVP(), 2670 * but this will have to do until VFS_FHTOVP() has a lock 2671 * type argument like VFS_VGET(). 2672 */ 2673 NFSVOPLOCK(*vpp, LK_DOWNGRADE | LK_RETRY); 2674 2675 NFSEXITCODE(error); 2676 return (error); 2677} 2678 2679/* 2680 * nfsd_fhtovp() - convert a fh to a vnode ptr 2681 * - look up fsid in mount list (if not found ret error) 2682 * - get vp and export rights by calling nfsvno_fhtovp() 2683 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 2684 * for AUTH_SYS 2685 * - if mpp != NULL, return the mount point so that it can 2686 * be used for vn_finished_write() by the caller 2687 */ 2688void 2689nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype, 2690 struct vnode **vpp, struct nfsexstuff *exp, 2691 struct mount **mpp, int startwrite, struct thread *p) 2692{ 2693 struct mount *mp; 2694 struct ucred *credanon; 2695 fhandle_t *fhp; 2696 2697 fhp = (fhandle_t *)nfp->nfsrvfh_data; 2698 /* 2699 * Check for the special case of the nfsv4root_fh. 2700 */ 2701 mp = vfs_busyfs(&fhp->fh_fsid); 2702 if (mpp != NULL) 2703 *mpp = mp; 2704 if (mp == NULL) { 2705 *vpp = NULL; 2706 nd->nd_repstat = ESTALE; 2707 goto out; 2708 } 2709 2710 if (startwrite) 2711 vn_start_write(NULL, mpp, V_WAIT); 2712 2713 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp, 2714 &credanon); 2715 vfs_unbusy(mp); 2716 2717 /* 2718 * For NFSv4 without a pseudo root fs, unexported file handles 2719 * can be returned, so that Lookup works everywhere. 2720 */ 2721 if (!nd->nd_repstat && exp->nes_exflag == 0 && 2722 !(nd->nd_flag & ND_NFSV4)) { 2723 vput(*vpp); 2724 nd->nd_repstat = EACCES; 2725 } 2726 2727 /* 2728 * Personally, I've never seen any point in requiring a 2729 * reserved port#, since only in the rare case where the 2730 * clients are all boxes with secure system priviledges, 2731 * does it provide any enhanced security, but... some people 2732 * believe it to be useful and keep putting this code back in. 2733 * (There is also some "security checker" out there that 2734 * complains if the nfs server doesn't enforce this.) 2735 * However, note the following: 2736 * RFC3530 (NFSv4) specifies that a reserved port# not be 2737 * required. 2738 * RFC2623 recommends that, if a reserved port# is checked for, 2739 * that there be a way to turn that off--> ifdef'd. 2740 */ 2741#ifdef NFS_REQRSVPORT 2742 if (!nd->nd_repstat) { 2743 struct sockaddr_in *saddr; 2744 struct sockaddr_in6 *saddr6; 2745 2746 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *); 2747 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *); 2748 if (!(nd->nd_flag & ND_NFSV4) && 2749 ((saddr->sin_family == AF_INET && 2750 ntohs(saddr->sin_port) >= IPPORT_RESERVED) || 2751 (saddr6->sin6_family == AF_INET6 && 2752 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) { 2753 vput(*vpp); 2754 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK); 2755 } 2756 } 2757#endif /* NFS_REQRSVPORT */ 2758 2759 /* 2760 * Check/setup credentials. 2761 */ 2762 if (!nd->nd_repstat) { 2763 nd->nd_saveduid = nd->nd_cred->cr_uid; 2764 nd->nd_repstat = nfsd_excred(nd, exp, credanon); 2765 if (nd->nd_repstat) 2766 vput(*vpp); 2767 } 2768 if (credanon != NULL) 2769 crfree(credanon); 2770 if (nd->nd_repstat) { 2771 if (startwrite) 2772 vn_finished_write(mp); 2773 *vpp = NULL; 2774 if (mpp != NULL) 2775 *mpp = NULL; 2776 } 2777 2778out: 2779 NFSEXITCODE2(0, nd); 2780} 2781 2782/* 2783 * glue for fp. 2784 */ 2785int 2786fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp) 2787{ 2788 struct filedesc *fdp; 2789 struct file *fp; 2790 int error = 0; 2791 2792 fdp = p->td_proc->p_fd; 2793 if (fd >= fdp->fd_nfiles || 2794 (fp = fdp->fd_ofiles[fd]) == NULL) { 2795 error = EBADF; 2796 goto out; 2797 } 2798 *fpp = fp; 2799 2800out: 2801 NFSEXITCODE(error); 2802 return (error); 2803} 2804 2805/* 2806 * Called from nfssvc() to update the exports list. Just call 2807 * vfs_export(). This has to be done, since the v4 root fake fs isn't 2808 * in the mount list. 2809 */ 2810int 2811nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p) 2812{ 2813 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp; 2814 int error = 0; 2815 struct nameidata nd; 2816 fhandle_t fh; 2817 2818 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export); 2819 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0) 2820 nfs_rootfhset = 0; 2821 else if (error == 0) { 2822 if (nfsexargp->fspec == NULL) { 2823 error = EPERM; 2824 goto out; 2825 } 2826 /* 2827 * If fspec != NULL, this is the v4root path. 2828 */ 2829 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE, 2830 nfsexargp->fspec, p); 2831 if ((error = namei(&nd)) != 0) 2832 goto out; 2833 error = nfsvno_getfh(nd.ni_vp, &fh, p); 2834 vrele(nd.ni_vp); 2835 if (!error) { 2836 nfs_rootfh.nfsrvfh_len = NFSX_MYFH; 2837 NFSBCOPY((caddr_t)&fh, 2838 nfs_rootfh.nfsrvfh_data, 2839 sizeof (fhandle_t)); 2840 nfs_rootfhset = 1; 2841 } 2842 } 2843 2844out: 2845 NFSEXITCODE(error); 2846 return (error); 2847} 2848 2849/* 2850 * Get the tcp socket sequence numbers we need. 2851 * (Maybe this should be moved to the tcp sources?) 2852 */ 2853int 2854nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap) 2855{ 2856 struct inpcb *inp; 2857 struct tcpcb *tp; 2858 int error = 0; 2859 2860 inp = sotoinpcb(so); 2861 KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL")); 2862 INP_RLOCK(inp); 2863 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { 2864 INP_RUNLOCK(inp); 2865 error = EPIPE; 2866 goto out; 2867 } 2868 tp = intotcpcb(inp); 2869 if (tp->t_state != TCPS_ESTABLISHED) { 2870 INP_RUNLOCK(inp); 2871 error = EPIPE; 2872 goto out; 2873 } 2874 *maxp = tp->snd_max; 2875 *unap = tp->snd_una; 2876 INP_RUNLOCK(inp); 2877 2878out: 2879 NFSEXITCODE(error); 2880 return (error); 2881} 2882 2883/* 2884 * This function needs to test to see if the system is near its limit 2885 * for memory allocation via malloc() or mget() and return True iff 2886 * either of these resources are near their limit. 2887 * XXX (For now, this is just a stub.) 2888 */ 2889int nfsrv_testmalloclimit = 0; 2890int 2891nfsrv_mallocmget_limit(void) 2892{ 2893 static int printmesg = 0; 2894 static int testval = 1; 2895 2896 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) { 2897 if ((printmesg++ % 100) == 0) 2898 printf("nfsd: malloc/mget near limit\n"); 2899 return (1); 2900 } 2901 return (0); 2902} 2903 2904/* 2905 * BSD specific initialization of a mount point. 2906 */ 2907void 2908nfsd_mntinit(void) 2909{ 2910 static int inited = 0; 2911 2912 if (inited) 2913 return; 2914 inited = 1; 2915 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED); 2916 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist); 2917 nfsv4root_mnt.mnt_export = NULL; 2918 TAILQ_INIT(&nfsv4root_opt); 2919 TAILQ_INIT(&nfsv4root_newopt); 2920 nfsv4root_mnt.mnt_opt = &nfsv4root_opt; 2921 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt; 2922 nfsv4root_mnt.mnt_nvnodelistsize = 0; 2923} 2924 2925/* 2926 * Get a vnode for a file handle, without checking exports, etc. 2927 */ 2928struct vnode * 2929nfsvno_getvp(fhandle_t *fhp) 2930{ 2931 struct mount *mp; 2932 struct vnode *vp; 2933 int error; 2934 2935 mp = vfs_busyfs(&fhp->fh_fsid); 2936 if (mp == NULL) 2937 return (NULL); 2938 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp); 2939 vfs_unbusy(mp); 2940 if (error) 2941 return (NULL); 2942 return (vp); 2943} 2944 2945/* 2946 * Do a local VOP_ADVLOCK(). 2947 */ 2948int 2949nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first, 2950 u_int64_t end, struct thread *td) 2951{ 2952 int error = 0; 2953 struct flock fl; 2954 u_int64_t tlen; 2955 2956 if (nfsrv_dolocallocks == 0) 2957 goto out; 2958 2959 /* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */ 2960 if ((vp->v_iflag & VI_DOOMED) != 0) { 2961 error = EPERM; 2962 goto out; 2963 } 2964 2965 fl.l_whence = SEEK_SET; 2966 fl.l_type = ftype; 2967 fl.l_start = (off_t)first; 2968 if (end == NFS64BITSSET) { 2969 fl.l_len = 0; 2970 } else { 2971 tlen = end - first; 2972 fl.l_len = (off_t)tlen; 2973 } 2974 /* 2975 * For FreeBSD8, the l_pid and l_sysid must be set to the same 2976 * values for all calls, so that all locks will be held by the 2977 * nfsd server. (The nfsd server handles conflicts between the 2978 * various clients.) 2979 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024 2980 * bytes, so it can't be put in l_sysid. 2981 */ 2982 if (nfsv4_sysid == 0) 2983 nfsv4_sysid = nlm_acquire_next_sysid(); 2984 fl.l_pid = (pid_t)0; 2985 fl.l_sysid = (int)nfsv4_sysid; 2986 2987 NFSVOPUNLOCK(vp, 0); 2988 if (ftype == F_UNLCK) 2989 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl, 2990 (F_POSIX | F_REMOTE)); 2991 else 2992 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl, 2993 (F_POSIX | F_REMOTE)); 2994 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2995 2996out: 2997 NFSEXITCODE(error); 2998 return (error); 2999} 3000 3001/* 3002 * Check the nfsv4 root exports. 3003 */ 3004int 3005nfsvno_v4rootexport(struct nfsrv_descript *nd) 3006{ 3007 struct ucred *credanon; 3008 int exflags, error = 0, numsecflavor, *secflavors, i; 3009 3010 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags, 3011 &credanon, &numsecflavor, &secflavors); 3012 if (error) { 3013 error = NFSERR_PROGUNAVAIL; 3014 goto out; 3015 } 3016 if (credanon != NULL) 3017 crfree(credanon); 3018 for (i = 0; i < numsecflavor; i++) { 3019 if (secflavors[i] == AUTH_SYS) 3020 nd->nd_flag |= ND_EXAUTHSYS; 3021 else if (secflavors[i] == RPCSEC_GSS_KRB5) 3022 nd->nd_flag |= ND_EXGSS; 3023 else if (secflavors[i] == RPCSEC_GSS_KRB5I) 3024 nd->nd_flag |= ND_EXGSSINTEGRITY; 3025 else if (secflavors[i] == RPCSEC_GSS_KRB5P) 3026 nd->nd_flag |= ND_EXGSSPRIVACY; 3027 } 3028 3029out: 3030 NFSEXITCODE(error); 3031 return (error); 3032} 3033 3034/* 3035 * Nfs server psuedo system call for the nfsd's 3036 */ 3037/* 3038 * MPSAFE 3039 */ 3040static int 3041nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap) 3042{ 3043 struct file *fp; 3044 struct nfsd_addsock_args sockarg; 3045 struct nfsd_nfsd_args nfsdarg; 3046 int error; 3047 3048 if (uap->flag & NFSSVC_NFSDADDSOCK) { 3049 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg)); 3050 if (error) 3051 goto out; 3052 /* 3053 * Since we don't know what rights might be required, 3054 * pretend that we need them all. It is better to be too 3055 * careful than too reckless. 3056 */ 3057 if ((error = fget(td, sockarg.sock, CAP_SOCK_ALL, &fp)) != 0) 3058 goto out; 3059 if (fp->f_type != DTYPE_SOCKET) { 3060 fdrop(fp, td); 3061 error = EPERM; 3062 goto out; 3063 } 3064 error = nfsrvd_addsock(fp); 3065 fdrop(fp, td); 3066 } else if (uap->flag & NFSSVC_NFSDNFSD) { 3067 if (uap->argp == NULL) { 3068 error = EINVAL; 3069 goto out; 3070 } 3071 error = copyin(uap->argp, (caddr_t)&nfsdarg, 3072 sizeof (nfsdarg)); 3073 if (error) 3074 goto out; 3075 error = nfsrvd_nfsd(td, &nfsdarg); 3076 } else { 3077 error = nfssvc_srvcall(td, uap, td->td_ucred); 3078 } 3079 3080out: 3081 NFSEXITCODE(error); 3082 return (error); 3083} 3084 3085static int 3086nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred) 3087{ 3088 struct nfsex_args export; 3089 struct file *fp = NULL; 3090 int stablefd, len; 3091 struct nfsd_clid adminrevoke; 3092 struct nfsd_dumplist dumplist; 3093 struct nfsd_dumpclients *dumpclients; 3094 struct nfsd_dumplocklist dumplocklist; 3095 struct nfsd_dumplocks *dumplocks; 3096 struct nameidata nd; 3097 vnode_t vp; 3098 int error = EINVAL; 3099 struct proc *procp; 3100 3101 if (uap->flag & NFSSVC_PUBLICFH) { 3102 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data, 3103 sizeof (fhandle_t)); 3104 error = copyin(uap->argp, 3105 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t)); 3106 if (!error) 3107 nfs_pubfhset = 1; 3108 } else if (uap->flag & NFSSVC_V4ROOTEXPORT) { 3109 error = copyin(uap->argp,(caddr_t)&export, 3110 sizeof (struct nfsex_args)); 3111 if (!error) 3112 error = nfsrv_v4rootexport(&export, cred, p); 3113 } else if (uap->flag & NFSSVC_NOPUBLICFH) { 3114 nfs_pubfhset = 0; 3115 error = 0; 3116 } else if (uap->flag & NFSSVC_STABLERESTART) { 3117 error = copyin(uap->argp, (caddr_t)&stablefd, 3118 sizeof (int)); 3119 if (!error) 3120 error = fp_getfvp(p, stablefd, &fp, &vp); 3121 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE)) 3122 error = EBADF; 3123 if (!error && newnfs_numnfsd != 0) 3124 error = EPERM; 3125 if (!error) { 3126 nfsrv_stablefirst.nsf_fp = fp; 3127 nfsrv_setupstable(p); 3128 } 3129 } else if (uap->flag & NFSSVC_ADMINREVOKE) { 3130 error = copyin(uap->argp, (caddr_t)&adminrevoke, 3131 sizeof (struct nfsd_clid)); 3132 if (!error) 3133 error = nfsrv_adminrevoke(&adminrevoke, p); 3134 } else if (uap->flag & NFSSVC_DUMPCLIENTS) { 3135 error = copyin(uap->argp, (caddr_t)&dumplist, 3136 sizeof (struct nfsd_dumplist)); 3137 if (!error && (dumplist.ndl_size < 1 || 3138 dumplist.ndl_size > NFSRV_MAXDUMPLIST)) 3139 error = EPERM; 3140 if (!error) { 3141 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size; 3142 dumpclients = (struct nfsd_dumpclients *)malloc(len, 3143 M_TEMP, M_WAITOK); 3144 nfsrv_dumpclients(dumpclients, dumplist.ndl_size); 3145 error = copyout(dumpclients, 3146 CAST_USER_ADDR_T(dumplist.ndl_list), len); 3147 free((caddr_t)dumpclients, M_TEMP); 3148 } 3149 } else if (uap->flag & NFSSVC_DUMPLOCKS) { 3150 error = copyin(uap->argp, (caddr_t)&dumplocklist, 3151 sizeof (struct nfsd_dumplocklist)); 3152 if (!error && (dumplocklist.ndllck_size < 1 || 3153 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST)) 3154 error = EPERM; 3155 if (!error) 3156 error = nfsrv_lookupfilename(&nd, 3157 dumplocklist.ndllck_fname, p); 3158 if (!error) { 3159 len = sizeof (struct nfsd_dumplocks) * 3160 dumplocklist.ndllck_size; 3161 dumplocks = (struct nfsd_dumplocks *)malloc(len, 3162 M_TEMP, M_WAITOK); 3163 nfsrv_dumplocks(nd.ni_vp, dumplocks, 3164 dumplocklist.ndllck_size, p); 3165 vput(nd.ni_vp); 3166 error = copyout(dumplocks, 3167 CAST_USER_ADDR_T(dumplocklist.ndllck_list), len); 3168 free((caddr_t)dumplocks, M_TEMP); 3169 } 3170 } else if (uap->flag & NFSSVC_BACKUPSTABLE) { 3171 procp = p->td_proc; 3172 PROC_LOCK(procp); 3173 nfsd_master_pid = procp->p_pid; 3174 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1); 3175 nfsd_master_start = procp->p_stats->p_start; 3176 nfsd_master_proc = procp; 3177 PROC_UNLOCK(procp); 3178 } 3179 3180 NFSEXITCODE(error); 3181 return (error); 3182} 3183 3184/* 3185 * Check exports. 3186 * Returns 0 if ok, 1 otherwise. 3187 */ 3188int 3189nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp) 3190{ 3191 int i; 3192 3193 /* 3194 * This seems odd, but allow the case where the security flavor 3195 * list is empty. This happens when NFSv4 is traversing non-exported 3196 * file systems. Exported file systems should always have a non-empty 3197 * security flavor list. 3198 */ 3199 if (exp->nes_numsecflavor == 0) 3200 return (0); 3201 3202 for (i = 0; i < exp->nes_numsecflavor; i++) { 3203 /* 3204 * The tests for privacy and integrity must be first, 3205 * since ND_GSS is set for everything but AUTH_SYS. 3206 */ 3207 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P && 3208 (nd->nd_flag & ND_GSSPRIVACY)) 3209 return (0); 3210 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I && 3211 (nd->nd_flag & ND_GSSINTEGRITY)) 3212 return (0); 3213 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 && 3214 (nd->nd_flag & ND_GSS)) 3215 return (0); 3216 if (exp->nes_secflavors[i] == AUTH_SYS && 3217 (nd->nd_flag & ND_GSS) == 0) 3218 return (0); 3219 } 3220 return (1); 3221} 3222 3223/* 3224 * Calculate a hash value for the fid in a file handle. 3225 */ 3226uint32_t 3227nfsrv_hashfh(fhandle_t *fhp) 3228{ 3229 uint32_t hashval; 3230 3231 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0); 3232 return (hashval); 3233} 3234 3235/* 3236 * Signal the userland master nfsd to backup the stable restart file. 3237 */ 3238void 3239nfsrv_backupstable(void) 3240{ 3241 struct proc *procp; 3242 3243 if (nfsd_master_proc != NULL) { 3244 procp = pfind(nfsd_master_pid); 3245 /* Try to make sure it is the correct process. */ 3246 if (procp == nfsd_master_proc && 3247 procp->p_stats->p_start.tv_sec == 3248 nfsd_master_start.tv_sec && 3249 procp->p_stats->p_start.tv_usec == 3250 nfsd_master_start.tv_usec && 3251 strcmp(procp->p_comm, nfsd_master_comm) == 0) 3252 kern_psignal(procp, SIGUSR2); 3253 else 3254 nfsd_master_proc = NULL; 3255 3256 if (procp != NULL) 3257 PROC_UNLOCK(procp); 3258 } 3259} 3260 3261extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *); 3262 3263/* 3264 * Called once to initialize data structures... 3265 */ 3266static int 3267nfsd_modevent(module_t mod, int type, void *data) 3268{ 3269 int error = 0; 3270 static int loaded = 0; 3271 3272 switch (type) { 3273 case MOD_LOAD: 3274 if (loaded) 3275 goto out; 3276 newnfs_portinit(); 3277 mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF); 3278 mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF); 3279 mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL, 3280 MTX_DEF); 3281 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0); 3282 nfsrvd_initcache(); 3283 nfsd_init(); 3284 NFSD_LOCK(); 3285 nfsrvd_init(0); 3286 NFSD_UNLOCK(); 3287 nfsd_mntinit(); 3288#ifdef VV_DISABLEDELEG 3289 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation; 3290 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation; 3291#endif 3292 nfsd_call_servertimer = nfsrv_servertimer; 3293 nfsd_call_nfsd = nfssvc_nfsd; 3294 loaded = 1; 3295 break; 3296 3297 case MOD_UNLOAD: 3298 if (newnfs_numnfsd != 0) { 3299 error = EBUSY; 3300 break; 3301 } 3302 3303#ifdef VV_DISABLEDELEG 3304 vn_deleg_ops.vndeleg_recall = NULL; 3305 vn_deleg_ops.vndeleg_disable = NULL; 3306#endif 3307 nfsd_call_servertimer = NULL; 3308 nfsd_call_nfsd = NULL; 3309 3310 /* Clean out all NFSv4 state. */ 3311 nfsrv_throwawayallstate(curthread); 3312 3313 /* Clean the NFS server reply cache */ 3314 nfsrvd_cleancache(); 3315 3316 /* Free up the krpc server pool. */ 3317 if (nfsrvd_pool != NULL) 3318 svcpool_destroy(nfsrvd_pool); 3319 3320 /* and get rid of the locks */ 3321 mtx_destroy(&nfs_cache_mutex); 3322 mtx_destroy(&nfs_v4root_mutex); 3323 mtx_destroy(&nfsv4root_mnt.mnt_mtx); 3324 lockdestroy(&nfsv4root_mnt.mnt_explock); 3325 loaded = 0; 3326 break; 3327 default: 3328 error = EOPNOTSUPP; 3329 break; 3330 } 3331 3332out: 3333 NFSEXITCODE(error); 3334 return (error); 3335} 3336static moduledata_t nfsd_mod = { 3337 "nfsd", 3338 nfsd_modevent, 3339 NULL, 3340}; 3341DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY); 3342 3343/* So that loader and kldload(2) can find us, wherever we are.. */ 3344MODULE_VERSION(nfsd, 1); 3345MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1); 3346MODULE_DEPEND(nfsd, nfslock, 1, 1, 1); 3347MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1); 3348MODULE_DEPEND(nfsd, krpc, 1, 1, 1); 3349MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1); 3350 3351