nfs_lock.c revision 214048
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 */ 3075631Salfred 3183651Speter#include <sys/cdefs.h> 3283651Speter__FBSDID("$FreeBSD: head/sys/nfs/nfs_lock.c 214048 2010-10-19 00:20:00Z rmacklem $"); 3383651Speter 3475631Salfred#include <sys/param.h> 3575631Salfred#include <sys/systm.h> 36138430Sphk#include <sys/conf.h> 3775631Salfred#include <sys/fcntl.h> 3876166Smarkm#include <sys/kernel.h> /* for hz */ 39114216Skan#include <sys/limits.h> 4076166Smarkm#include <sys/lock.h> 4175631Salfred#include <sys/malloc.h> 4276166Smarkm#include <sys/lockf.h> /* for hz */ /* Must come after sys/malloc.h */ 4375631Salfred#include <sys/mbuf.h> 4475631Salfred#include <sys/mount.h> 4575631Salfred#include <sys/namei.h> 46168931Srwatson#include <sys/priv.h> 4775631Salfred#include <sys/proc.h> 4876166Smarkm#include <sys/resourcevar.h> 4975631Salfred#include <sys/socket.h> 5076166Smarkm#include <sys/socket.h> 5175631Salfred#include <sys/unistd.h> 5275631Salfred#include <sys/vnode.h> 5375631Salfred 5475631Salfred#include <net/if.h> 5575631Salfred 5675631Salfred#include <nfs/nfsproto.h> 57210455Srmacklem#include <nfs/nfs_lock.h> 5883651Speter#include <nfsclient/nfs.h> 5983651Speter#include <nfsclient/nfsmount.h> 6083651Speter#include <nfsclient/nfsnode.h> 6183651Speter#include <nfsclient/nlminfo.h> 6275631Salfred 63151695Sglebiusextern void (*nlminfo_release_p)(struct proc *p); 64151695Sglebius 65214048Srmacklemvop_advlock_t *nfs_advlock_p = nfs_dolock; 66214048Srmacklemvop_reclaim_t *nfs_reclaim_p = NULL; 67214048Srmacklem 68151897SrwatsonMALLOC_DEFINE(M_NFSLOCK, "nfsclient_lock", "NFS lock request"); 69151897SrwatsonMALLOC_DEFINE(M_NLMINFO, "nfsclient_nlminfo", "NFS lock process structure"); 70138430Sphk 71138430Sphkstatic int nfslockdans(struct thread *td, struct lockd_ans *ansp); 72151695Sglebiusstatic void nlminfo_release(struct proc *p); 7375631Salfred/* 74138430Sphk * -------------------------------------------------------------------- 75138430Sphk * A miniature device driver which the userland uses to talk to us. 76138430Sphk * 77138430Sphk */ 78138430Sphk 79138430Sphkstatic struct cdev *nfslock_dev; 80138430Sphkstatic struct mtx nfslock_mtx; 81138430Sphkstatic int nfslock_isopen; 82138430Sphkstatic TAILQ_HEAD(,__lock_msg) nfslock_list; 83138430Sphk 84138430Sphkstatic int 85138430Sphknfslock_open(struct cdev *dev, int oflags, int devtype, struct thread *td) 86138430Sphk{ 87138430Sphk int error; 88138430Sphk 89168931Srwatson error = priv_check(td, PRIV_NFS_LOCKD); 90168931Srwatson if (error) 91168931Srwatson return (error); 92168931Srwatson 93138430Sphk mtx_lock(&nfslock_mtx); 94138430Sphk if (!nfslock_isopen) { 95138430Sphk error = 0; 96138430Sphk nfslock_isopen = 1; 97138430Sphk } else { 98138430Sphk error = EOPNOTSUPP; 99138430Sphk } 100138430Sphk mtx_unlock(&nfslock_mtx); 101138430Sphk 102138430Sphk return (error); 103138430Sphk} 104138430Sphk 105138430Sphkstatic int 106138430Sphknfslock_close(struct cdev *dev, int fflag, int devtype, struct thread *td) 107138430Sphk{ 108138430Sphk struct __lock_msg *lm; 109138430Sphk 110138430Sphk mtx_lock(&nfslock_mtx); 111138430Sphk nfslock_isopen = 0; 112138430Sphk while (!TAILQ_EMPTY(&nfslock_list)) { 113138430Sphk lm = TAILQ_FIRST(&nfslock_list); 114138430Sphk /* XXX: answer request */ 115138430Sphk TAILQ_REMOVE(&nfslock_list, lm, lm_link); 116138430Sphk free(lm, M_NFSLOCK); 117138430Sphk } 118138430Sphk mtx_unlock(&nfslock_mtx); 119138430Sphk return (0); 120138430Sphk} 121138430Sphk 122138430Sphkstatic int 123138430Sphknfslock_read(struct cdev *dev, struct uio *uio, int ioflag) 124138430Sphk{ 125138430Sphk int error; 126138430Sphk struct __lock_msg *lm; 127138430Sphk 128138430Sphk if (uio->uio_resid != sizeof *lm) 129138430Sphk return (EOPNOTSUPP); 130138430Sphk lm = NULL; 131138430Sphk error = 0; 132138430Sphk mtx_lock(&nfslock_mtx); 133138430Sphk while (TAILQ_EMPTY(&nfslock_list)) { 134138430Sphk error = msleep(&nfslock_list, &nfslock_mtx, PSOCK | PCATCH, 135138430Sphk "nfslockd", 0); 136138430Sphk if (error) 137138430Sphk break; 138138430Sphk } 139138430Sphk if (!error) { 140138430Sphk lm = TAILQ_FIRST(&nfslock_list); 141138430Sphk TAILQ_REMOVE(&nfslock_list, lm, lm_link); 142138430Sphk } 143138430Sphk mtx_unlock(&nfslock_mtx); 144138430Sphk if (!error) { 145138430Sphk error = uiomove(lm, sizeof *lm, uio); 146138430Sphk free(lm, M_NFSLOCK); 147138430Sphk } 148138430Sphk return (error); 149138430Sphk} 150138430Sphk 151138430Sphkstatic int 152138430Sphknfslock_write(struct cdev *dev, struct uio *uio, int ioflag) 153138430Sphk{ 154138430Sphk struct lockd_ans la; 155138430Sphk int error; 156138430Sphk 157138430Sphk if (uio->uio_resid != sizeof la) 158138430Sphk return (EOPNOTSUPP); 159138430Sphk error = uiomove(&la, sizeof la, uio); 160138430Sphk if (!error) 161138430Sphk error = nfslockdans(curthread, &la); 162138430Sphk return (error); 163138430Sphk} 164138430Sphk 165138430Sphkstatic int 166138430Sphknfslock_send(struct __lock_msg *lm) 167138430Sphk{ 168138430Sphk struct __lock_msg *lm2; 169138430Sphk int error; 170138430Sphk 171138430Sphk error = 0; 172138430Sphk lm2 = malloc(sizeof *lm2, M_NFSLOCK, M_WAITOK); 173138430Sphk mtx_lock(&nfslock_mtx); 174138430Sphk if (nfslock_isopen) { 175138430Sphk memcpy(lm2, lm, sizeof *lm2); 176138430Sphk TAILQ_INSERT_TAIL(&nfslock_list, lm2, lm_link); 177138430Sphk wakeup(&nfslock_list); 178138430Sphk } else { 179138430Sphk error = EOPNOTSUPP; 180138430Sphk } 181138430Sphk mtx_unlock(&nfslock_mtx); 182138430Sphk if (error) 183138430Sphk free(lm2, M_NFSLOCK); 184138430Sphk return (error); 185138430Sphk} 186138430Sphk 187138430Sphkstatic struct cdevsw nfslock_cdevsw = { 188138430Sphk .d_version = D_VERSION, 189138430Sphk .d_open = nfslock_open, 190138430Sphk .d_close = nfslock_close, 191138430Sphk .d_read = nfslock_read, 192138430Sphk .d_write = nfslock_write, 193138430Sphk .d_name = "nfslock" 194138430Sphk}; 195138430Sphk 196138430Sphkstatic int 197138430Sphknfslock_modevent(module_t mod __unused, int type, void *data __unused) 198138430Sphk{ 199138430Sphk 200138430Sphk switch (type) { 201138430Sphk case MOD_LOAD: 202138430Sphk if (bootverbose) 203138430Sphk printf("nfslock: pseudo-device\n"); 204138430Sphk mtx_init(&nfslock_mtx, "nfslock", NULL, MTX_DEF); 205138430Sphk TAILQ_INIT(&nfslock_list); 206151695Sglebius nlminfo_release_p = nlminfo_release; 207138430Sphk nfslock_dev = make_dev(&nfslock_cdevsw, 0, 208138430Sphk UID_ROOT, GID_KMEM, 0600, _PATH_NFSLCKDEV); 209138430Sphk return (0); 210138430Sphk default: 211138430Sphk return (EOPNOTSUPP); 212138430Sphk } 213138430Sphk} 214138430Sphk 215138430SphkDEV_MODULE(nfslock, nfslock_modevent, NULL); 216138430SphkMODULE_VERSION(nfslock, 1); 217138430Sphk 218138430Sphk 219138430Sphk/* 22075631Salfred * XXX 22175631Salfred * We have to let the process know if the call succeeded. I'm using an extra 22283651Speter * field in the p_nlminfo field in the proc structure, as it is already for 22375631Salfred * lockd stuff. 22475631Salfred */ 22575631Salfred 22675631Salfred/* 22775631Salfred * nfs_advlock -- 22875631Salfred * NFS advisory byte-level locks. 229178243Skib * 230178243Skib * The vnode shall be (shared) locked on the entry, it is 231178243Skib * unconditionally unlocked after. 23275631Salfred */ 23375631Salfredint 23483651Speternfs_dolock(struct vop_advlock_args *ap) 23575631Salfred{ 23675631Salfred LOCKD_MSG msg; 23783366Sjulian struct thread *td; 238138430Sphk struct vnode *vp; 239138430Sphk int error; 24075631Salfred struct flock *fl; 24183366Sjulian struct proc *p; 242214048Srmacklem struct nfsmount *nmp; 24375631Salfred 24483366Sjulian td = curthread; 24583366Sjulian p = td->td_proc; 24683366Sjulian 24775631Salfred vp = ap->a_vp; 24875631Salfred fl = ap->a_fl; 249214048Srmacklem nmp = VFSTONFS(vp->v_mount); 25075631Salfred 251178243Skib ASSERT_VOP_LOCKED(vp, "nfs_dolock"); 252178243Skib 253214048Srmacklem nmp->nm_getinfo(vp, msg.lm_fh, &msg.lm_fh_len, &msg.lm_addr, 254214048Srmacklem &msg.lm_nfsv3, NULL); 255178243Skib VOP_UNLOCK(vp, 0); 256178243Skib 25775631Salfred /* 25875631Salfred * the NLM protocol doesn't allow the server to return an error 25982174Sache * on ranges, so we do it. 26075631Salfred */ 26182194Sache if (fl->l_whence != SEEK_END) { 26282213Sache if ((fl->l_whence != SEEK_CUR && fl->l_whence != SEEK_SET) || 26382204Sache fl->l_start < 0 || 26482204Sache (fl->l_len < 0 && 26582204Sache (fl->l_start == 0 || fl->l_start + fl->l_len < 0))) 26682194Sache return (EINVAL); 26782204Sache if (fl->l_len > 0 && 26882204Sache (fl->l_len - 1 > OFF_MAX - fl->l_start)) 26982194Sache return (EOVERFLOW); 27082194Sache } 27175631Salfred 27275631Salfred /* 27375631Salfred * Fill in the information structure. 27475631Salfred */ 27575631Salfred msg.lm_version = LOCKD_MSG_VERSION; 27675631Salfred msg.lm_msg_ident.pid = p->p_pid; 277178243Skib 278178243Skib mtx_lock(&Giant); 27975631Salfred /* 28075631Salfred * if there is no nfsowner table yet, allocate one. 28175631Salfred */ 28275631Salfred if (p->p_nlminfo == NULL) { 283184214Sdes p->p_nlminfo = malloc(sizeof(struct nlminfo), 284184214Sdes M_NLMINFO, M_WAITOK | M_ZERO); 28575631Salfred p->p_nlminfo->pid_start = p->p_stats->p_start; 286114434Sdes timevaladd(&p->p_nlminfo->pid_start, &boottime); 28775631Salfred } 28875631Salfred msg.lm_msg_ident.pid_start = p->p_nlminfo->pid_start; 28975631Salfred msg.lm_msg_ident.msg_seq = ++(p->p_nlminfo->msg_seq); 29075631Salfred 29175631Salfred msg.lm_fl = *fl; 29275631Salfred msg.lm_wait = ap->a_flags & F_WAIT; 29375631Salfred msg.lm_getlk = ap->a_op == F_GETLK; 294101947Salfred cru2x(td->td_ucred, &msg.lm_cred); 29575631Salfred 29675631Salfred for (;;) { 297138430Sphk error = nfslock_send(&msg); 298138430Sphk if (error) 299178243Skib goto out; 30075631Salfred 301138430Sphk /* Unlocks succeed immediately. */ 30275631Salfred if (fl->l_type == F_UNLCK) 303178243Skib goto out; 30475631Salfred 30575631Salfred /* 306161371Sthomas * Retry after 20 seconds if we haven't gotten a response yet. 30775631Salfred * This number was picked out of thin air... but is longer 30875631Salfred * then even a reasonably loaded system should take (at least 30983651Speter * on a local network). XXX Probably should use a back-off 31075631Salfred * scheme. 311116185Srwatson * 312116185Srwatson * XXX: No PCATCH here since we currently have no useful 313116185Srwatson * way to signal to the userland rpc.lockd that the request 314116185Srwatson * has been aborted. Once the rpc.lockd implementation 315116185Srwatson * can handle aborts, and we report them properly, 316116185Srwatson * PCATCH can be put back. In the mean time, if we did 317116185Srwatson * permit aborting, the lock attempt would "get lost" 318116185Srwatson * and the lock would get stuck in the locked state. 31975631Salfred */ 320115415Srwatson error = tsleep(p->p_nlminfo, PUSER, "lockd", 20*hz); 321107104Salfred if (error != 0) { 32275631Salfred if (error == EWOULDBLOCK) { 32375631Salfred /* 32475631Salfred * We timed out, so we rewrite the request 325154316Srwatson * to the fifo. 32675631Salfred */ 32775631Salfred continue; 32875631Salfred } 32975631Salfred 33075631Salfred break; 33175631Salfred } 33275631Salfred 33375631Salfred if (msg.lm_getlk && p->p_nlminfo->retcode == 0) { 33475631Salfred if (p->p_nlminfo->set_getlk_pid) { 335177633Sdfr fl->l_sysid = 0; /* XXX */ 33675631Salfred fl->l_pid = p->p_nlminfo->getlk_pid; 33775631Salfred } else { 33875631Salfred fl->l_type = F_UNLCK; 33975631Salfred } 34075631Salfred } 34175631Salfred error = p->p_nlminfo->retcode; 34275631Salfred break; 34375631Salfred } 344178243Skib out: 345178243Skib mtx_unlock(&Giant); 346138430Sphk return (error); 34775631Salfred} 34875631Salfred 34975631Salfred/* 35075631Salfred * nfslockdans -- 35175631Salfred * NFS advisory byte-level locks answer from the lock daemon. 35275631Salfred */ 353138430Sphkstatic int 35486363Srwatsonnfslockdans(struct thread *td, struct lockd_ans *ansp) 35575631Salfred{ 35686363Srwatson struct proc *targetp; 35775631Salfred 35875631Salfred /* the version should match, or we're out of sync */ 35975631Salfred if (ansp->la_vers != LOCKD_ANS_VERSION) 36075631Salfred return (EINVAL); 36175631Salfred 36275631Salfred /* Find the process, set its return errno and wake it up. */ 36386363Srwatson if ((targetp = pfind(ansp->la_msg_ident.pid)) == NULL) 36475631Salfred return (ESRCH); 36575631Salfred 36683651Speter /* verify the pid hasn't been reused (if we can), and it isn't waiting 36775631Salfred * for an answer from a more recent request. We return an EPIPE if 36875631Salfred * the match fails, because we've already used ESRCH above, and this 36975631Salfred * is sort of like writing on a pipe after the reader has closed it. 37075631Salfred */ 37186363Srwatson if (targetp->p_nlminfo == NULL || 37275631Salfred ((ansp->la_msg_ident.msg_seq != -1) && 37386363Srwatson (timevalcmp(&targetp->p_nlminfo->pid_start, 37475631Salfred &ansp->la_msg_ident.pid_start, !=) || 37586363Srwatson targetp->p_nlminfo->msg_seq != ansp->la_msg_ident.msg_seq))) { 37686363Srwatson PROC_UNLOCK(targetp); 37775631Salfred return (EPIPE); 37877563Sjake } 37975631Salfred 38086363Srwatson targetp->p_nlminfo->retcode = ansp->la_errno; 38186363Srwatson targetp->p_nlminfo->set_getlk_pid = ansp->la_set_getlk_pid; 38286363Srwatson targetp->p_nlminfo->getlk_pid = ansp->la_getlk_pid; 38375631Salfred 384107104Salfred wakeup(targetp->p_nlminfo); 38575631Salfred 38686363Srwatson PROC_UNLOCK(targetp); 38775631Salfred return (0); 38875631Salfred} 389138430Sphk 390151695Sglebius/* 391151695Sglebius * Free nlminfo attached to process. 392151695Sglebius */ 393151695Sglebiusvoid 394151695Sglebiusnlminfo_release(struct proc *p) 395151695Sglebius{ 396151695Sglebius free(p->p_nlminfo, M_NLMINFO); 397151695Sglebius p->p_nlminfo = NULL; 398151695Sglebius} 399