2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
32 * @(#)nfs_subs.c 8.8 (Berkeley) 5/22/95
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
39 * These functions support the macros and help fiddle mbuf chains for
40 * the nfs op functions. They do things like create the rpc header and
41 * copy data between mbuf chains and uio lists.
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/namei.h>
54 #include <sys/socket.h>
56 #include <sys/malloc.h>
57 #include <sys/sysent.h>
58 #include <sys/syscall.h>
59 #include <sys/sysproto.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_extern.h>
66 #include <rpc/rpcclnt.h>
68 #include <nfs/rpcv2.h>
69 #include <nfs/nfsproto.h>
70 #include <nfsclient/nfs.h>
71 #include <nfsclient/nfsnode.h>
72 #include <nfs/xdr_subs.h>
73 #include <nfsclient/nfsm_subs.h>
74 #include <nfsclient/nfsmount.h>
76 #include <netinet/in.h>
79 * Note that stdarg.h and the ANSI style va_start macro is used for both
80 * ANSI and traditional C compilers.
82 #include <machine/stdarg.h>
85 * Data items converted to xdr at startup, since they are constant
86 * This is kinda hokey, but may save a little time doing byte swaps
88 u_int32_t nfs_xdrneg1;
89 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
90 rpc_mismatch, rpc_auth_unix, rpc_msgaccepted;
91 u_int32_t nfs_true, nfs_false;
93 /* And other global data */
94 static u_int32_t nfs_xid = 0;
95 static enum vtype nv2tov_type[8]= {
96 VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON
100 int nfs_pbuf_freecnt = -1; /* start out unlimited */
102 struct nfs_reqq nfs_reqq;
103 struct mtx nfs_reqq_mtx;
104 struct nfs_bufq nfs_bufq;
105 static struct mtx nfs_xid_mtx;
108 * and the reverse mapping from generic to Version 2 procedure numbers
110 int nfsv2_procid[NFS_NPROCS] = {
136 LIST_HEAD(nfsnodehashhead, nfsnode);
143 mtx_lock(&nfs_xid_mtx);
145 /* Get a pretty random xid to start with */
149 * Skip zero xid if it should ever happen.
154 mtx_unlock(&nfs_xid_mtx);
159 * Create the header for an rpc request packet
160 * The hsiz is the size of the rest of the nfs request header.
161 * (just used to decide if a cluster is a good idea)
164 nfsm_reqhead(struct vnode *vp, u_long procid, int hsiz)
168 MGET(mb, M_WAIT, MT_DATA);
169 if (hsiz >= MINCLSIZE)
176 * Build the RPC header and fill in the authorization info.
177 * The authorization string argument is only used when the credentials
178 * come from outside of the kernel.
179 * Returns the head of the mbuf list.
182 nfsm_rpchead(struct ucred *cr, int nmflag, int procid, int auth_type,
183 int auth_len, struct mbuf *mrest, int mrest_len, struct mbuf **mbp,
193 authsiz = nfsm_rndup(auth_len);
194 MGETHDR(mb, M_WAIT, MT_DATA);
195 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
197 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
198 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
200 MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
204 bpos = mtod(mb, caddr_t);
207 * First the RPC header.
209 tl = nfsm_build(u_int32_t *, 8 * NFSX_UNSIGNED);
212 *tl++ = txdr_unsigned(nfs_xid_gen());
215 *tl++ = txdr_unsigned(NFS_PROG);
216 if (nmflag & NFSMNT_NFSV3) {
217 *tl++ = txdr_unsigned(NFS_VER3);
218 *tl++ = txdr_unsigned(procid);
220 *tl++ = txdr_unsigned(NFS_VER2);
221 *tl++ = txdr_unsigned(nfsv2_procid[procid]);
225 * And then the authorization cred.
227 *tl++ = txdr_unsigned(auth_type);
228 *tl = txdr_unsigned(authsiz);
231 tl = nfsm_build(u_int32_t *, auth_len);
232 *tl++ = 0; /* stamp ?? */
233 *tl++ = 0; /* NULL hostname */
234 *tl++ = txdr_unsigned(cr->cr_uid);
235 *tl++ = txdr_unsigned(cr->cr_groups[0]);
236 grpsiz = (auth_len >> 2) - 5;
237 *tl++ = txdr_unsigned(grpsiz);
238 for (i = 1; i <= grpsiz; i++)
239 *tl++ = txdr_unsigned(cr->cr_groups[i]);
244 * And the verifier...
246 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
247 *tl++ = txdr_unsigned(RPCAUTH_NULL);
250 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
251 mreq->m_pkthdr.rcvif = NULL;
257 * copies a uio scatter/gather list to an mbuf chain.
258 * NOTE: can ony handle iovcnt == 1
261 nfsm_uiotombuf(struct uio *uiop, struct mbuf **mq, int siz, caddr_t *bpos)
264 struct mbuf *mp, *mp2;
265 int xfer, left, mlen;
266 int uiosiz, clflg, rem;
270 if (uiop->uio_iovcnt != 1)
271 panic("nfsm_uiotombuf: iovcnt != 1");
274 if (siz > MLEN) /* or should it >= MCLBYTES ?? */
278 rem = nfsm_rndup(siz)-siz;
281 left = uiop->uio_iov->iov_len;
282 uiocp = uiop->uio_iov->iov_base;
287 mlen = M_TRAILINGSPACE(mp);
289 MGET(mp, M_WAIT, MT_DATA);
295 mlen = M_TRAILINGSPACE(mp);
297 xfer = (left > mlen) ? mlen : left;
300 if (uiop->uio_iov->iov_op != NULL)
301 (*(uiop->uio_iov->iov_op))
302 (uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
305 if (uiop->uio_segflg == UIO_SYSSPACE)
306 bcopy(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
308 copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
312 uiop->uio_offset += xfer;
313 uiop->uio_resid -= xfer;
315 uiop->uio_iov->iov_base =
316 (char *)uiop->uio_iov->iov_base + uiosiz;
317 uiop->uio_iov->iov_len -= uiosiz;
321 if (rem > M_TRAILINGSPACE(mp)) {
322 MGET(mp, M_WAIT, MT_DATA);
326 cp = mtod(mp, caddr_t)+mp->m_len;
327 for (left = 0; left < rem; left++)
332 *bpos = mtod(mp, caddr_t)+mp->m_len;
338 * Copy a string into mbufs for the hard cases...
341 nfsm_strtmbuf(struct mbuf **mb, char **bpos, const char *cp, long siz)
343 struct mbuf *m1 = NULL, *m2;
344 long left, xfer, len, tlen;
350 left = M_TRAILINGSPACE(m2);
352 tl = ((u_int32_t *)(*bpos));
353 *tl++ = txdr_unsigned(siz);
355 left -= NFSX_UNSIGNED;
356 m2->m_len += NFSX_UNSIGNED;
358 bcopy(cp, (caddr_t) tl, left);
365 /* Loop around adding mbufs */
367 MGET(m1, M_WAIT, MT_DATA);
370 m1->m_len = NFSMSIZ(m1);
373 tl = mtod(m1, u_int32_t *);
376 *tl++ = txdr_unsigned(siz);
377 m1->m_len -= NFSX_UNSIGNED;
378 tlen = NFSX_UNSIGNED;
381 if (siz < m1->m_len) {
382 len = nfsm_rndup(siz);
387 xfer = len = m1->m_len;
389 bcopy(cp, (caddr_t) tl, xfer);
390 m1->m_len = len+tlen;
395 *bpos = mtod(m1, caddr_t)+m1->m_len;
400 * Called once to initialize data structures...
403 nfs_init(struct vfsconf *vfsp)
407 nfsmount_zone = uma_zcreate("NFSMOUNT", sizeof(struct nfsmount),
408 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
409 rpc_vers = txdr_unsigned(RPC_VER2);
410 rpc_call = txdr_unsigned(RPC_CALL);
411 rpc_reply = txdr_unsigned(RPC_REPLY);
412 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
413 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
414 rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
415 rpc_autherr = txdr_unsigned(RPC_AUTHERR);
416 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
417 nfs_true = txdr_unsigned(TRUE);
418 nfs_false = txdr_unsigned(FALSE);
419 nfs_xdrneg1 = txdr_unsigned(-1);
420 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
423 /* Ensure async daemons disabled */
424 for (i = 0; i < NFS_MAXASYNCDAEMON; i++) {
425 nfs_iodwant[i] = NULL;
426 nfs_iodmount[i] = NULL;
428 nfs_nhinit(); /* Init the nfsnode table */
431 * Initialize reply list and start timer
433 TAILQ_INIT(&nfs_reqq);
434 callout_init(&nfs_callout, CALLOUT_MPSAFE);
435 mtx_init(&nfs_reqq_mtx, "NFS reqq lock", NULL, MTX_DEF);
436 mtx_init(&nfs_iod_mtx, "NFS iod lock", NULL, MTX_DEF);
437 mtx_init(&nfs_xid_mtx, "NFS xid lock", NULL, MTX_DEF);
439 nfs_pbuf_freecnt = nswbuf / 2 + 1;
445 nfs_uninit(struct vfsconf *vfsp)
449 callout_stop(&nfs_callout);
451 KASSERT(TAILQ_EMPTY(&nfs_reqq),
452 ("nfs_uninit: request queue not empty"));
455 * Tell all nfsiod processes to exit. Clear nfs_iodmax, and wakeup
456 * any sleeping nfsiods so they check nfs_iodmax and exit.
458 mtx_lock(&nfs_iod_mtx);
460 for (i = 0; i < nfs_numasync; i++)
462 wakeup(&nfs_iodwant[i]);
463 /* The last nfsiod to exit will wake us up when nfs_numasync hits 0 */
465 msleep(&nfs_numasync, &nfs_iod_mtx, PWAIT, "ioddie", 0);
466 mtx_unlock(&nfs_iod_mtx);
468 uma_zdestroy(nfsmount_zone);
473 nfs_dircookie_lock(struct nfsnode *np)
475 mtx_lock(&np->n_mtx);
476 while (np->n_flag & NDIRCOOKIELK)
477 (void) msleep(&np->n_flag, &np->n_mtx, PZERO, "nfsdirlk", 0);
478 np->n_flag |= NDIRCOOKIELK;
479 mtx_unlock(&np->n_mtx);
483 nfs_dircookie_unlock(struct nfsnode *np)
485 mtx_lock(&np->n_mtx);
486 np->n_flag &= ~NDIRCOOKIELK;
488 mtx_unlock(&np->n_mtx);
492 nfs_upgrade_vnlock(struct vnode *vp)
496 if ((old_lock = VOP_ISLOCKED(vp)) != LK_EXCLUSIVE) {
497 if (old_lock == LK_SHARED) {
498 /* Upgrade to exclusive lock, this might block */
499 vn_lock(vp, LK_UPGRADE | LK_RETRY);
501 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
508 nfs_downgrade_vnlock(struct vnode *vp, int old_lock)
510 if (old_lock != LK_EXCLUSIVE) {
511 if (old_lock == LK_SHARED) {
512 /* Downgrade from exclusive lock, this might block */
513 vn_lock(vp, LK_DOWNGRADE);
521 nfs_printf(const char *fmt, ...)
533 * Attribute cache routines.
534 * nfs_loadattrcache() - loads or updates the cache contents from attributes
535 * that are on the mbuf list
536 * nfs_getattrcache() - returns valid attributes if found in cache, returns
541 * Load the attribute cache (that lives in the nfsnode entry) with
542 * the values on the mbuf list and
544 * copy the attributes to *vaper
547 nfs_loadattrcache(struct vnode **vpp, struct mbuf **mdp, caddr_t *dposp,
548 struct vattr *vaper, int dontshrink)
550 struct vnode *vp = *vpp;
552 struct nfs_fattr *fp;
560 struct timespec mtime, mtime_save;
561 int v3 = NFS_ISV3(vp);
562 struct thread *td = curthread;
565 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
566 cp2 = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, M_WAIT);
569 fp = (struct nfs_fattr *)cp2;
571 vtyp = nfsv3tov_type(fp->fa_type);
572 vmode = fxdr_unsigned(u_short, fp->fa_mode);
573 rdev = makedev(fxdr_unsigned(int, fp->fa3_rdev.specdata1),
574 fxdr_unsigned(int, fp->fa3_rdev.specdata2));
575 fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
577 vtyp = nfsv2tov_type(fp->fa_type);
578 vmode = fxdr_unsigned(u_short, fp->fa_mode);
582 * The duplicate information returned in fa_type and fa_mode
583 * is an ambiguity in the NFS version 2 protocol.
585 * VREG should be taken literally as a regular file. If a
586 * server intents to return some type information differently
587 * in the upper bits of the mode field (e.g. for sockets, or
588 * FIFOs), NFSv2 mandates fa_type to be VNON. Anyway, we
589 * leave the examination of the mode bits even in the VREG
590 * case to avoid breakage for bogus servers, but we make sure
591 * that there are actually type bits set in the upper part of
592 * fa_mode (and failing that, trust the va_type field).
594 * NFSv3 cleared the issue, and requires fa_mode to not
595 * contain any type information (while also introduing sockets
596 * and FIFOs for fa_type).
598 if (vtyp == VNON || (vtyp == VREG && (vmode & S_IFMT) != 0))
599 vtyp = IFTOVT(vmode);
600 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
601 fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
604 * Really ugly NFSv2 kludge.
606 if (vtyp == VCHR && rdev == 0xffffffff)
611 * If v_type == VNON it is a new node, so fill in the v_type,
612 * n_mtime fields. Check to see if it represents a special
613 * device, and if so, check for a possible alias. Once the
614 * correct vnode has been obtained, fill in the rest of the
618 mtx_lock(&np->n_mtx);
619 if (vp->v_type != vtyp) {
621 if (vp->v_type == VFIFO)
622 vp->v_op = &nfs_fifoops;
627 vap->va_mode = (vmode & 07777);
629 mtime_save = vap->va_mtime;
630 vap->va_mtime = mtime;
631 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
633 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
634 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
635 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
636 vap->va_size = fxdr_hyper(&fp->fa3_size);
637 vap->va_blocksize = NFS_FABLKSIZE;
638 vap->va_bytes = fxdr_hyper(&fp->fa3_used);
639 vap->va_fileid = fxdr_unsigned(int32_t,
640 fp->fa3_fileid.nfsuquad[1]);
641 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
642 fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
646 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
647 vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
648 vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
649 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
650 vap->va_blocksize = fxdr_unsigned(int32_t, fp->fa2_blocksize);
651 vap->va_bytes = (u_quad_t)fxdr_unsigned(int32_t, fp->fa2_blocks)
653 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
654 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
656 vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
657 fp->fa2_ctime.nfsv2_sec);
658 vap->va_ctime.tv_nsec = 0;
659 vap->va_gen = fxdr_unsigned(u_int32_t, fp->fa2_ctime.nfsv2_usec);
662 np->n_attrstamp = time_second;
663 /* Timestamp the NFS otw getattr fetch */
665 np->n_ac_ts_tid = td->td_tid;
666 np->n_ac_ts_pid = td->td_proc->p_pid;
667 np->n_ac_ts_syscalls = td->td_syscalls;
669 bzero(&np->n_ac_ts, sizeof(struct nfs_attrcache_timestamp));
671 if (vap->va_size != np->n_size) {
672 if (vap->va_type == VREG) {
673 if (dontshrink && vap->va_size < np->n_size) {
675 * We've been told not to shrink the file;
676 * zero np->n_attrstamp to indicate that
677 * the attributes are stale.
679 vap->va_size = np->n_size;
681 } else if (np->n_flag & NMODIFIED) {
683 * We've modified the file: Use the larger
684 * of our size, and the server's size.
686 if (vap->va_size < np->n_size) {
687 vap->va_size = np->n_size;
689 np->n_size = vap->va_size;
690 np->n_flag |= NSIZECHANGED;
693 np->n_size = vap->va_size;
694 np->n_flag |= NSIZECHANGED;
696 vnode_pager_setsize(vp, np->n_size);
698 np->n_size = vap->va_size;
702 * The following checks are added to prevent a race between (say)
703 * a READDIR+ and a WRITE.
704 * READDIR+, WRITE requests sent out.
705 * READDIR+ resp, WRITE resp received on client.
706 * However, the WRITE resp was handled before the READDIR+ resp
707 * causing the post op attrs from the write to be loaded first
708 * and the attrs from the READDIR+ to be loaded later. If this
709 * happens, we have stale attrs loaded into the attrcache.
710 * We detect this by for the mtime moving back. We invalidate the
711 * attrcache when this happens.
713 if (timespeccmp(&mtime_save, &vap->va_mtime, >))
714 /* Size changed or mtime went backwards */
717 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
718 if (np->n_flag & NCHG) {
719 if (np->n_flag & NACC)
720 vaper->va_atime = np->n_atim;
721 if (np->n_flag & NUPD)
722 vaper->va_mtime = np->n_mtim;
725 mtx_unlock(&np->n_mtx);
730 #include <sys/sysctl.h>
731 SYSCTL_DECL(_vfs_nfs);
732 static int nfs_acdebug;
733 SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
737 * Check the time stamp
738 * If the cache is valid, copy contents to *vap and return 0
739 * otherwise return an error
742 nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
746 struct nfsmount *nmp;
751 nmp = VFSTONFS(vp->v_mount);
753 mtx_lock(&Giant); /* nfs_printf() */
755 mtx_lock(&np->n_mtx);
756 /* XXX n_mtime doesn't seem to be updated on a miss-and-reload */
757 timeo = (time_second - np->n_mtime.tv_sec) / 10;
761 nfs_printf("nfs_getattrcache: initial timeo = %d\n", timeo);
764 if (vap->va_type == VDIR) {
765 if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acdirmin)
766 timeo = nmp->nm_acdirmin;
767 else if (timeo > nmp->nm_acdirmax)
768 timeo = nmp->nm_acdirmax;
770 if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acregmin)
771 timeo = nmp->nm_acregmin;
772 else if (timeo > nmp->nm_acregmax)
773 timeo = nmp->nm_acregmax;
778 nfs_printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
779 nmp->nm_acregmin, nmp->nm_acregmax,
780 nmp->nm_acdirmin, nmp->nm_acdirmax);
783 nfs_printf("nfs_getattrcache: age = %d; final timeo = %d\n",
784 (time_second - np->n_attrstamp), timeo);
787 if ((time_second - np->n_attrstamp) >= timeo) {
788 nfsstats.attrcache_misses++;
789 mtx_unlock(&np->n_mtx);
792 nfsstats.attrcache_hits++;
793 if (vap->va_size != np->n_size) {
794 if (vap->va_type == VREG) {
795 if (np->n_flag & NMODIFIED) {
796 if (vap->va_size < np->n_size)
797 vap->va_size = np->n_size;
799 np->n_size = vap->va_size;
801 np->n_size = vap->va_size;
803 vnode_pager_setsize(vp, np->n_size);
805 np->n_size = vap->va_size;
808 bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
809 if (np->n_flag & NCHG) {
810 if (np->n_flag & NACC)
811 vaper->va_atime = np->n_atim;
812 if (np->n_flag & NUPD)
813 vaper->va_mtime = np->n_mtim;
815 mtx_unlock(&np->n_mtx);
817 mtx_unlock(&Giant); /* nfs_printf() */
822 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
824 * This function finds the directory cookie that corresponds to the
825 * logical byte offset given.
828 nfs_getcookie(struct nfsnode *np, off_t off, int add)
830 struct nfsdmap *dp, *dp2;
832 nfsuint64 *retval = NULL;
834 pos = (uoff_t)off / NFS_DIRBLKSIZ;
835 if (pos == 0 || off < 0) {
838 panic("nfs getcookie add at <= 0");
840 return (&nfs_nullcookie);
843 dp = LIST_FIRST(&np->n_cookies);
846 MALLOC(dp, struct nfsdmap *, sizeof (struct nfsdmap),
847 M_NFSDIROFF, M_WAITOK);
848 dp->ndm_eocookie = 0;
849 LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
853 while (pos >= NFSNUMCOOKIES) {
854 pos -= NFSNUMCOOKIES;
855 if (LIST_NEXT(dp, ndm_list)) {
856 if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
857 pos >= dp->ndm_eocookie)
859 dp = LIST_NEXT(dp, ndm_list);
861 MALLOC(dp2, struct nfsdmap *, sizeof (struct nfsdmap),
862 M_NFSDIROFF, M_WAITOK);
863 dp2->ndm_eocookie = 0;
864 LIST_INSERT_AFTER(dp, dp2, ndm_list);
869 if (pos >= dp->ndm_eocookie) {
871 dp->ndm_eocookie = pos + 1;
875 retval = &dp->ndm_cookies[pos];
881 * Invalidate cached directory information, except for the actual directory
882 * blocks (which are invalidated separately).
883 * Done mainly to avoid the use of stale offset cookies.
886 nfs_invaldir(struct vnode *vp)
888 struct nfsnode *np = VTONFS(vp);
891 if (vp->v_type != VDIR)
892 panic("nfs: invaldir not dir");
894 nfs_dircookie_lock(np);
895 np->n_direofoffset = 0;
896 np->n_cookieverf.nfsuquad[0] = 0;
897 np->n_cookieverf.nfsuquad[1] = 0;
898 if (LIST_FIRST(&np->n_cookies))
899 LIST_FIRST(&np->n_cookies)->ndm_eocookie = 0;
900 nfs_dircookie_unlock(np);
904 * The write verifier has changed (probably due to a server reboot), so all
905 * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
906 * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
907 * and B_CLUSTEROK flags. Once done the new write verifier can be set for the
910 * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
911 * writes are not clusterable.
914 nfs_clearcommit(struct mount *mp)
916 struct vnode *vp, *nvp;
917 struct buf *bp, *nbp;
921 MNT_VNODE_FOREACH(vp, mp, nvp) {
924 if (vp->v_iflag & VI_DOOMED) {
932 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
933 if (!BUF_ISLOCKED(bp) &&
934 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
935 == (B_DELWRI | B_NEEDCOMMIT))
936 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
946 * Helper functions for former macros. Some of these should be
947 * moved to their callers.
951 nfsm_mtofh_xx(struct vnode *d, struct vnode **v, int v3, int *f,
952 struct mbuf **md, caddr_t *dpos)
954 struct nfsnode *ttnp;
962 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos);
965 *f = fxdr_unsigned(int, *tl);
969 t1 = nfsm_getfh_xx(&ttfhp, &ttfhsize, (v3), md, dpos);
972 t1 = nfs_nget(d->v_mount, ttfhp, ttfhsize, &ttnp, LK_EXCLUSIVE);
978 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos);
982 *f = fxdr_unsigned(int, *tl);
983 else if (fxdr_unsigned(int, *tl))
984 nfsm_adv_xx(NFSX_V3FATTR, md, dpos);
988 t1 = nfs_loadattrcache(&ttvp, md, dpos, NULL, 0);
997 nfsm_getfh_xx(nfsfh_t **f, int *s, int v3, struct mbuf **md, caddr_t *dpos)
1002 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos);
1005 *s = fxdr_unsigned(int, *tl);
1006 if (*s <= 0 || *s > NFSX_V3FHMAX)
1010 *f = nfsm_dissect_xx(nfsm_rndup(*s), md, dpos);
1019 nfsm_loadattr_xx(struct vnode **v, struct vattr *va, struct mbuf **md,
1024 struct vnode *ttvp = *v;
1025 t1 = nfs_loadattrcache(&ttvp, md, dpos, va, 0);
1033 nfsm_postop_attr_xx(struct vnode **v, int *f, struct mbuf **md,
1039 struct vnode *ttvp = *v;
1040 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos);
1043 *f = fxdr_unsigned(int, *tl);
1045 t1 = nfs_loadattrcache(&ttvp, md, dpos, NULL, 1);
1056 nfsm_wcc_data_xx(struct vnode **v, int *f, struct mbuf **md, caddr_t *dpos)
1059 int ttattrf, ttretf = 0;
1062 tl = nfsm_dissect_xx(NFSX_UNSIGNED, md, dpos);
1065 if (*tl == nfs_true) {
1066 tl = nfsm_dissect_xx(6 * NFSX_UNSIGNED, md, dpos);
1069 mtx_lock(&(VTONFS(*v))->n_mtx);
1071 ttretf = (VTONFS(*v)->n_mtime.tv_sec == fxdr_unsigned(u_int32_t, *(tl + 2)) &&
1072 VTONFS(*v)->n_mtime.tv_nsec == fxdr_unsigned(u_int32_t, *(tl + 3)));
1073 mtx_unlock(&(VTONFS(*v))->n_mtx);
1075 t1 = nfsm_postop_attr_xx(v, &ttattrf, md, dpos);
1086 nfsm_strtom_xx(const char *a, int s, int m, struct mbuf **mb, caddr_t *bpos)
1092 return ENAMETOOLONG;
1093 t1 = nfsm_rndup(s) + NFSX_UNSIGNED;
1094 if (t1 <= M_TRAILINGSPACE(*mb)) {
1095 tl = nfsm_build_xx(t1, mb, bpos);
1096 *tl++ = txdr_unsigned(s);
1097 *(tl + ((t1 >> 2) - 2)) = 0;
1100 t1 = nfsm_strtmbuf(mb, bpos, a, s);
1108 nfsm_fhtom_xx(struct vnode *v, int v3, struct mbuf **mb, caddr_t *bpos)
1115 t1 = nfsm_rndup(VTONFS(v)->n_fhsize) + NFSX_UNSIGNED;
1116 if (t1 < M_TRAILINGSPACE(*mb)) {
1117 tl = nfsm_build_xx(t1, mb, bpos);
1118 *tl++ = txdr_unsigned(VTONFS(v)->n_fhsize);
1119 *(tl + ((t1 >> 2) - 2)) = 0;
1120 bcopy(VTONFS(v)->n_fhp, tl, VTONFS(v)->n_fhsize);
1122 t1 = nfsm_strtmbuf(mb, bpos,
1123 (const char *)VTONFS(v)->n_fhp,
1124 VTONFS(v)->n_fhsize);
1129 cp = nfsm_build_xx(NFSX_V2FH, mb, bpos);
1130 bcopy(VTONFS(v)->n_fhp, cp, NFSX_V2FH);
1136 nfsm_v3attrbuild_xx(struct vattr *va, int full, struct mbuf **mb,
1141 if (va->va_mode != (mode_t)VNOVAL) {
1142 tl = nfsm_build_xx(2 * NFSX_UNSIGNED, mb, bpos);
1144 *tl = txdr_unsigned(va->va_mode);
1146 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos);
1149 if (full && va->va_uid != (uid_t)VNOVAL) {
1150 tl = nfsm_build_xx(2 * NFSX_UNSIGNED, mb, bpos);
1152 *tl = txdr_unsigned(va->va_uid);
1154 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos);
1157 if (full && va->va_gid != (gid_t)VNOVAL) {
1158 tl = nfsm_build_xx(2 * NFSX_UNSIGNED, mb, bpos);
1160 *tl = txdr_unsigned(va->va_gid);
1162 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos);
1165 if (full && va->va_size != VNOVAL) {
1166 tl = nfsm_build_xx(3 * NFSX_UNSIGNED, mb, bpos);
1168 txdr_hyper(va->va_size, tl);
1170 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos);
1173 if (va->va_atime.tv_sec != VNOVAL) {
1174 if (va->va_atime.tv_sec != time_second) {
1175 tl = nfsm_build_xx(3 * NFSX_UNSIGNED, mb, bpos);
1176 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
1177 txdr_nfsv3time(&va->va_atime, tl);
1179 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos);
1180 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
1183 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos);
1184 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
1186 if (va->va_mtime.tv_sec != VNOVAL) {
1187 if (va->va_mtime.tv_sec != time_second) {
1188 tl = nfsm_build_xx(3 * NFSX_UNSIGNED, mb, bpos);
1189 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
1190 txdr_nfsv3time(&va->va_mtime, tl);
1192 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos);
1193 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
1196 tl = nfsm_build_xx(NFSX_UNSIGNED, mb, bpos);
1197 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);