nfs_lock.c revision 77563
175631Salfred/*- 275631Salfred * Copyright (c) 1997 Berkeley Software Design, Inc. All rights reserved. 375631Salfred * 475631Salfred * Redistribution and use in source and binary forms, with or without 575631Salfred * modification, are permitted provided that the following conditions 675631Salfred * are met: 775631Salfred * 1. Redistributions of source code must retain the above copyright 875631Salfred * notice, this list of conditions and the following disclaimer. 975631Salfred * 2. Redistributions in binary form must reproduce the above copyright 1075631Salfred * notice, this list of conditions and the following disclaimer in the 1175631Salfred * documentation and/or other materials provided with the distribution. 1275631Salfred * 3. Berkeley Software Design Inc's name may not be used to endorse or 1375631Salfred * promote products derived from this software without specific prior 1475631Salfred * written permission. 1575631Salfred * 1675631Salfred * THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``AS IS'' AND 1775631Salfred * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1875631Salfred * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1975631Salfred * ARE DISCLAIMED. IN NO EVENT SHALL BERKELEY SOFTWARE DESIGN INC BE LIABLE 2075631Salfred * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2175631Salfred * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2275631Salfred * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2375631Salfred * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2475631Salfred * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2575631Salfred * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2675631Salfred * SUCH DAMAGE. 2775631Salfred * 2875631Salfred * from BSDI nfs_lock.c,v 2.4 1998/12/14 23:49:56 jch Exp 2975631Salfred * $FreeBSD: head/sys/nfsclient/nfs_lock.c 77563 2001-06-01 01:30:51Z jake $ 3075631Salfred */ 3175631Salfred 3275631Salfred#include <sys/param.h> 3375631Salfred#include <sys/systm.h> 3475631Salfred#include <sys/fcntl.h> 3576166Smarkm#include <sys/kernel.h> /* for hz */ 3676166Smarkm#include <sys/lock.h> 3775631Salfred#include <sys/malloc.h> 3876166Smarkm#include <sys/lockf.h> /* for hz */ /* Must come after sys/malloc.h */ 3975631Salfred#include <sys/mbuf.h> 4075631Salfred#include <sys/mount.h> 4175631Salfred#include <sys/namei.h> 4275631Salfred#include <sys/proc.h> 4376166Smarkm#include <sys/resourcevar.h> 4475631Salfred#include <sys/socket.h> 4576166Smarkm#include <sys/socket.h> 4675631Salfred#include <sys/unistd.h> 4775631Salfred#include <sys/vnode.h> 4875631Salfred 4975631Salfred#include <net/if.h> 5075631Salfred 5175631Salfred#include <nfs/rpcv2.h> 5275631Salfred#include <nfs/nfsproto.h> 5375631Salfred#include <nfs/nfs.h> 5475631Salfred#include <nfs/nfsmount.h> 5575631Salfred#include <nfs/nfsnode.h> 5675631Salfred#include <nfs/nfs_lock.h> 5775631Salfred#include <nfs/nlminfo.h> 5875631Salfred 5975631Salfred#define NFSOWNER_1ST_LEVEL_START 1 /* initial entries */ 6075631Salfred#define NFSOWNER_2ND_LEVEL 256 /* some power of 2 */ 6175631Salfred 6275631Salfred#define NFSOWNER(tbl, i) \ 6375631Salfred (tbl)[(i) / NFSOWNER_2ND_LEVEL][(i) % NFSOWNER_2ND_LEVEL] 6475631Salfred 6575631Salfred/* 6675631Salfred * XXX 6775631Salfred * We have to let the process know if the call succeeded. I'm using an extra 6875631Salfred * field in the p_nlminfo field in the proc structure, as it is already for 6975631Salfred * lockd stuff. 7075631Salfred */ 7175631Salfred 7275631Salfred/* 7375631Salfred * nfs_advlock -- 7475631Salfred * NFS advisory byte-level locks. 7575631Salfred */ 7675631Salfredint 7775631Salfrednfs_dolock(ap) 7875631Salfred struct vop_advlock_args /* { 7975631Salfred struct vnode *a_vp; 8075631Salfred caddr_t a_id; 8175631Salfred int a_op; 8275631Salfred struct flock *a_fl; 8375631Salfred int a_flags; 8475631Salfred } */ *ap; 8575631Salfred{ 8675631Salfred LOCKD_MSG msg; 8775631Salfred struct nameidata nd; 8875631Salfred struct proc *p; 8975631Salfred uid_t saved_uid; 9075631Salfred struct vnode *vp, *wvp; 9175631Salfred int error, error1; 9275631Salfred struct flock *fl; 9375631Salfred int fmode, ioflg; 9475631Salfred 9575631Salfred p = curproc; 9675631Salfred vp = ap->a_vp; 9775631Salfred fl = ap->a_fl; 9875631Salfred 9975631Salfred /* 10075631Salfred * the NLM protocol doesn't allow the server to return an error 10175631Salfred * on ranges, so we do it. Note that we should be returning 10275631Salfred * EOVERFLOW in some cases, but we don't have it. 10375631Salfred */ 10475631Salfred if (fl->l_start < 0 || fl->l_len < 0 || 10575631Salfred ((fl->l_len != 0 && 10675631Salfred (fl->l_start + fl->l_len - 1) < 0))) 10775631Salfred return (EINVAL); 10875631Salfred 10975631Salfred /* 11075631Salfred * Fill in the information structure. 11175631Salfred */ 11275631Salfred msg.lm_version = LOCKD_MSG_VERSION; 11375631Salfred msg.lm_msg_ident.pid = p->p_pid; 11475631Salfred /* 11575631Salfred * if there is no nfsowner table yet, allocate one. 11675631Salfred */ 11775631Salfred if (p->p_nlminfo == NULL) { 11875631Salfred MALLOC(p->p_nlminfo, struct nlminfo *, 11975631Salfred sizeof(struct nlminfo), M_LOCKF, M_WAITOK | M_ZERO); 12075631Salfred p->p_nlminfo->pid_start = p->p_stats->p_start; 12175631Salfred } 12275631Salfred msg.lm_msg_ident.pid_start = p->p_nlminfo->pid_start; 12375631Salfred msg.lm_msg_ident.msg_seq = ++(p->p_nlminfo->msg_seq); 12475631Salfred 12575631Salfred msg.lm_fl = *fl; 12675631Salfred msg.lm_wait = ap->a_flags & F_WAIT; 12775631Salfred msg.lm_getlk = ap->a_op == F_GETLK; 12875631Salfred /* 12975631Salfred * XXX -- I think this is wrong for anything other AF_INET. 13075631Salfred */ 13175631Salfred msg.lm_addr = *(VFSTONFS(vp->v_mount)->nm_nam); 13275631Salfred msg.lm_fh_len = NFS_ISV3(vp) ? VTONFS(vp)->n_fhsize : NFSX_V2FH; 13375631Salfred bcopy(VTONFS(vp)->n_fhp, msg.lm_fh, msg.lm_fh_len); 13475631Salfred msg.lm_nfsv3 = NFS_ISV3(vp); 13575631Salfred msg.lm_cred = *(p->p_ucred); 13675631Salfred 13775631Salfred /* 13875631Salfred * Open the lock fifo. If for any reason we don't find the fifo, it 13975631Salfred * means that the lock daemon isn't running. Translate any missing 14075631Salfred * file error message for the user, otherwise the application will 14175631Salfred * complain that the user's file is missing, which isn't the case. 14275631Salfred * Note that we use proc0's cred, so the fifo is opened as root. 14375631Salfred */ 14475631Salfred NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, _PATH_LCKFIFO, p); 14575631Salfred 14675631Salfred /* 14775631Salfred * XXX Hack to temporarily allow this process (regardless of it's creds) 14875631Salfred * to open the fifo we need to write to. vn_open() really should 14975631Salfred * take a ucred (and once it does, this code should be fixed to use 15075631Salfred * proc0's ucred. 15175631Salfred */ 15275631Salfred saved_uid = p->p_ucred->cr_uid; 15375631Salfred p->p_ucred->cr_uid = 0; /* temporarly run the vn_open as root */ 15475631Salfred 15575631Salfred fmode = FFLAGS(O_WRONLY); 15675631Salfred error = vn_open(&nd, &fmode, 0); 15775631Salfred p->p_ucred->cr_uid = saved_uid; 15875631Salfred if (error != 0) { 15975631Salfred return (error == ENOENT ? EOPNOTSUPP : error); 16075631Salfred } 16175631Salfred wvp = nd.ni_vp; 16275631Salfred VOP_UNLOCK(wvp, 0, p); /* vn_open leaves it locked */ 16375631Salfred 16475631Salfred 16575631Salfred ioflg = IO_UNIT; 16675631Salfred for (;;) { 16775631Salfred VOP_LEASE(wvp, p, proc0.p_ucred, LEASE_WRITE); 16875631Salfred 16975631Salfred error = vn_rdwr(UIO_WRITE, wvp, (caddr_t)&msg, sizeof(msg), 0, 17075631Salfred UIO_SYSSPACE, ioflg, proc0.p_ucred, NULL, p); 17175631Salfred 17275631Salfred if (error && (((ioflg & IO_NDELAY) == 0) || error != EAGAIN)) { 17375631Salfred break; 17475631Salfred } 17575631Salfred /* 17675631Salfred * If we're locking a file, wait for an answer. Unlocks succeed 17775631Salfred * immediately. 17875631Salfred */ 17975631Salfred if (fl->l_type == F_UNLCK) 18075631Salfred /* 18175631Salfred * XXX this isn't exactly correct. The client side 18275631Salfred * needs to continue sending it's unlock until 18375631Salfred * it gets a responce back. 18475631Salfred */ 18575631Salfred break; 18675631Salfred 18775631Salfred /* 18875631Salfred * retry after 20 seconds if we haven't gotten a responce yet. 18975631Salfred * This number was picked out of thin air... but is longer 19075631Salfred * then even a reasonably loaded system should take (at least 19175631Salfred * on a local network). XXX Probably should use a back-off 19275631Salfred * scheme. 19375631Salfred */ 19475631Salfred if ((error = tsleep((void *)p->p_nlminfo, 19575631Salfred PCATCH | PUSER, "lockd", 20*hz)) != 0) { 19675631Salfred if (error == EWOULDBLOCK) { 19775631Salfred /* 19875631Salfred * We timed out, so we rewrite the request 19975631Salfred * to the fifo, but only if it isn't already 20075631Salfred * full. 20175631Salfred */ 20275631Salfred ioflg |= IO_NDELAY; 20375631Salfred continue; 20475631Salfred } 20575631Salfred 20675631Salfred break; 20775631Salfred } 20875631Salfred 20975631Salfred if (msg.lm_getlk && p->p_nlminfo->retcode == 0) { 21075631Salfred if (p->p_nlminfo->set_getlk_pid) { 21175631Salfred fl->l_pid = p->p_nlminfo->getlk_pid; 21275631Salfred } else { 21375631Salfred fl->l_type = F_UNLCK; 21475631Salfred } 21575631Salfred } 21675631Salfred error = p->p_nlminfo->retcode; 21775631Salfred break; 21875631Salfred } 21975631Salfred 22075631Salfred if ((error1 = vn_close(wvp, FWRITE, proc0.p_ucred, p)) && error == 0) 22175631Salfred return (error1); 22275631Salfred 22375631Salfred return (error); 22475631Salfred} 22575631Salfred 22675631Salfred/* 22775631Salfred * nfslockdans -- 22875631Salfred * NFS advisory byte-level locks answer from the lock daemon. 22975631Salfred */ 23075631Salfredint 23175631Salfrednfslockdans(p, ansp) 23275631Salfred struct proc *p; 23375631Salfred struct lockd_ans *ansp; 23475631Salfred{ 23575631Salfred int error; 23675631Salfred 23775631Salfred /* Let root, or someone who once was root (lockd generally 23875631Salfred * switches to the daemon uid once it is done setting up) make 23977183Srwatson * this call. 24077183Srwatson * 24177183Srwatson * XXX This authorization check is probably not right. 24275631Salfred */ 24377183Srwatson if ((error = suser(p)) != 0 && p->p_ucred->cr_svuid != 0) 24475631Salfred return (error); 24575631Salfred 24675631Salfred /* the version should match, or we're out of sync */ 24775631Salfred if (ansp->la_vers != LOCKD_ANS_VERSION) 24875631Salfred return (EINVAL); 24975631Salfred 25075631Salfred /* Find the process, set its return errno and wake it up. */ 25175631Salfred if ((p = pfind(ansp->la_msg_ident.pid)) == NULL) 25275631Salfred return (ESRCH); 25375631Salfred 25475631Salfred /* verify the pid hasn't been reused (if we can), and it isn't waiting 25575631Salfred * for an answer from a more recent request. We return an EPIPE if 25675631Salfred * the match fails, because we've already used ESRCH above, and this 25775631Salfred * is sort of like writing on a pipe after the reader has closed it. 25875631Salfred */ 25975631Salfred if (p->p_nlminfo == NULL || 26075631Salfred ((ansp->la_msg_ident.msg_seq != -1) && 26175631Salfred (timevalcmp(&p->p_nlminfo->pid_start, 26275631Salfred &ansp->la_msg_ident.pid_start, !=) || 26377563Sjake p->p_nlminfo->msg_seq != ansp->la_msg_ident.msg_seq))) { 26477563Sjake PROC_UNLOCK(p); 26575631Salfred return (EPIPE); 26677563Sjake } 26775631Salfred 26875631Salfred p->p_nlminfo->retcode = ansp->la_errno; 26975631Salfred p->p_nlminfo->set_getlk_pid = ansp->la_set_getlk_pid; 27075631Salfred p->p_nlminfo->getlk_pid = ansp->la_getlk_pid; 27175631Salfred 27275631Salfred (void)wakeup((void *)p->p_nlminfo); 27375631Salfred 27477563Sjake PROC_UNLOCK(p); 27575631Salfred return (0); 27675631Salfred} 277