nfs_lock.c revision 154316
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/nfsclient/nfs_lock.c 154316 2006-01-13 23:16:29Z rwatson $"); 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> 4675631Salfred#include <sys/proc.h> 4776166Smarkm#include <sys/resourcevar.h> 4875631Salfred#include <sys/socket.h> 4976166Smarkm#include <sys/socket.h> 5075631Salfred#include <sys/unistd.h> 5175631Salfred#include <sys/vnode.h> 5275631Salfred 5375631Salfred#include <net/if.h> 5475631Salfred 55122698Salfred#include <rpc/rpcclnt.h> 56122698Salfred 5775631Salfred#include <nfs/rpcv2.h> 5875631Salfred#include <nfs/nfsproto.h> 5983651Speter#include <nfsclient/nfs.h> 6083651Speter#include <nfsclient/nfsmount.h> 6183651Speter#include <nfsclient/nfsnode.h> 6283651Speter#include <nfsclient/nfs_lock.h> 6383651Speter#include <nfsclient/nlminfo.h> 6475631Salfred 65151695Sglebiusextern void (*nlminfo_release_p)(struct proc *p); 66151695Sglebius 67151897SrwatsonMALLOC_DEFINE(M_NFSLOCK, "nfsclient_lock", "NFS lock request"); 68151897SrwatsonMALLOC_DEFINE(M_NLMINFO, "nfsclient_nlminfo", "NFS lock process structure"); 69138430Sphk 70138430Sphkstatic int nfslockdans(struct thread *td, struct lockd_ans *ansp); 71151695Sglebiusstatic void nlminfo_release(struct proc *p); 7275631Salfred/* 73138430Sphk * -------------------------------------------------------------------- 74138430Sphk * A miniature device driver which the userland uses to talk to us. 75138430Sphk * 76138430Sphk */ 77138430Sphk 78138430Sphkstatic struct cdev *nfslock_dev; 79138430Sphkstatic struct mtx nfslock_mtx; 80138430Sphkstatic int nfslock_isopen; 81138430Sphkstatic TAILQ_HEAD(,__lock_msg) nfslock_list; 82138430Sphk 83138430Sphkstatic int 84138430Sphknfslock_open(struct cdev *dev, int oflags, int devtype, struct thread *td) 85138430Sphk{ 86138430Sphk int error; 87138430Sphk 88138430Sphk mtx_lock(&nfslock_mtx); 89138430Sphk if (!nfslock_isopen) { 90138430Sphk error = 0; 91138430Sphk nfslock_isopen = 1; 92138430Sphk } else { 93138430Sphk error = EOPNOTSUPP; 94138430Sphk } 95138430Sphk mtx_unlock(&nfslock_mtx); 96138430Sphk 97138430Sphk return (error); 98138430Sphk} 99138430Sphk 100138430Sphkstatic int 101138430Sphknfslock_close(struct cdev *dev, int fflag, int devtype, struct thread *td) 102138430Sphk{ 103138430Sphk struct __lock_msg *lm; 104138430Sphk 105138430Sphk mtx_lock(&nfslock_mtx); 106138430Sphk nfslock_isopen = 0; 107138430Sphk while (!TAILQ_EMPTY(&nfslock_list)) { 108138430Sphk lm = TAILQ_FIRST(&nfslock_list); 109138430Sphk /* XXX: answer request */ 110138430Sphk TAILQ_REMOVE(&nfslock_list, lm, lm_link); 111138430Sphk free(lm, M_NFSLOCK); 112138430Sphk } 113138430Sphk mtx_unlock(&nfslock_mtx); 114138430Sphk return (0); 115138430Sphk} 116138430Sphk 117138430Sphkstatic int 118138430Sphknfslock_read(struct cdev *dev, struct uio *uio, int ioflag) 119138430Sphk{ 120138430Sphk int error; 121138430Sphk struct __lock_msg *lm; 122138430Sphk 123138430Sphk if (uio->uio_resid != sizeof *lm) 124138430Sphk return (EOPNOTSUPP); 125138430Sphk lm = NULL; 126138430Sphk error = 0; 127138430Sphk mtx_lock(&nfslock_mtx); 128138430Sphk while (TAILQ_EMPTY(&nfslock_list)) { 129138430Sphk error = msleep(&nfslock_list, &nfslock_mtx, PSOCK | PCATCH, 130138430Sphk "nfslockd", 0); 131138430Sphk if (error) 132138430Sphk break; 133138430Sphk } 134138430Sphk if (!error) { 135138430Sphk lm = TAILQ_FIRST(&nfslock_list); 136138430Sphk TAILQ_REMOVE(&nfslock_list, lm, lm_link); 137138430Sphk } 138138430Sphk mtx_unlock(&nfslock_mtx); 139138430Sphk if (!error) { 140138430Sphk error = uiomove(lm, sizeof *lm, uio); 141138430Sphk free(lm, M_NFSLOCK); 142138430Sphk } 143138430Sphk return (error); 144138430Sphk} 145138430Sphk 146138430Sphkstatic int 147138430Sphknfslock_write(struct cdev *dev, struct uio *uio, int ioflag) 148138430Sphk{ 149138430Sphk struct lockd_ans la; 150138430Sphk int error; 151138430Sphk 152138430Sphk if (uio->uio_resid != sizeof la) 153138430Sphk return (EOPNOTSUPP); 154138430Sphk error = uiomove(&la, sizeof la, uio); 155138430Sphk if (!error) 156138430Sphk error = nfslockdans(curthread, &la); 157138430Sphk return (error); 158138430Sphk} 159138430Sphk 160138430Sphkstatic int 161138430Sphknfslock_send(struct __lock_msg *lm) 162138430Sphk{ 163138430Sphk struct __lock_msg *lm2; 164138430Sphk int error; 165138430Sphk 166138430Sphk error = 0; 167138430Sphk lm2 = malloc(sizeof *lm2, M_NFSLOCK, M_WAITOK); 168138430Sphk mtx_lock(&nfslock_mtx); 169138430Sphk if (nfslock_isopen) { 170138430Sphk memcpy(lm2, lm, sizeof *lm2); 171138430Sphk TAILQ_INSERT_TAIL(&nfslock_list, lm2, lm_link); 172138430Sphk wakeup(&nfslock_list); 173138430Sphk } else { 174138430Sphk error = EOPNOTSUPP; 175138430Sphk } 176138430Sphk mtx_unlock(&nfslock_mtx); 177138430Sphk if (error) 178138430Sphk free(lm2, M_NFSLOCK); 179138430Sphk return (error); 180138430Sphk} 181138430Sphk 182138430Sphkstatic struct cdevsw nfslock_cdevsw = { 183138430Sphk .d_version = D_VERSION, 184138430Sphk .d_open = nfslock_open, 185138430Sphk .d_close = nfslock_close, 186138430Sphk .d_read = nfslock_read, 187138430Sphk .d_write = nfslock_write, 188138430Sphk .d_name = "nfslock" 189138430Sphk}; 190138430Sphk 191138430Sphkstatic int 192138430Sphknfslock_modevent(module_t mod __unused, int type, void *data __unused) 193138430Sphk{ 194138430Sphk 195138430Sphk switch (type) { 196138430Sphk case MOD_LOAD: 197138430Sphk if (bootverbose) 198138430Sphk printf("nfslock: pseudo-device\n"); 199138430Sphk mtx_init(&nfslock_mtx, "nfslock", NULL, MTX_DEF); 200138430Sphk TAILQ_INIT(&nfslock_list); 201151695Sglebius nlminfo_release_p = nlminfo_release; 202138430Sphk nfslock_dev = make_dev(&nfslock_cdevsw, 0, 203138430Sphk UID_ROOT, GID_KMEM, 0600, _PATH_NFSLCKDEV); 204138430Sphk return (0); 205138430Sphk default: 206138430Sphk return (EOPNOTSUPP); 207138430Sphk } 208138430Sphk} 209138430Sphk 210138430SphkDEV_MODULE(nfslock, nfslock_modevent, NULL); 211138430SphkMODULE_VERSION(nfslock, 1); 212138430Sphk 213138430Sphk 214138430Sphk/* 21575631Salfred * XXX 21675631Salfred * We have to let the process know if the call succeeded. I'm using an extra 21783651Speter * field in the p_nlminfo field in the proc structure, as it is already for 21875631Salfred * lockd stuff. 21975631Salfred */ 22075631Salfred 22175631Salfred/* 22275631Salfred * nfs_advlock -- 22375631Salfred * NFS advisory byte-level locks. 22475631Salfred */ 22575631Salfredint 22683651Speternfs_dolock(struct vop_advlock_args *ap) 22775631Salfred{ 22875631Salfred LOCKD_MSG msg; 22983366Sjulian struct thread *td; 230138430Sphk struct vnode *vp; 231138430Sphk int error; 23275631Salfred struct flock *fl; 23383366Sjulian struct proc *p; 23475631Salfred 23583366Sjulian td = curthread; 23683366Sjulian p = td->td_proc; 23783366Sjulian 23875631Salfred vp = ap->a_vp; 23975631Salfred fl = ap->a_fl; 24075631Salfred 24175631Salfred /* 24275631Salfred * the NLM protocol doesn't allow the server to return an error 24382174Sache * on ranges, so we do it. 24475631Salfred */ 24582194Sache if (fl->l_whence != SEEK_END) { 24682213Sache if ((fl->l_whence != SEEK_CUR && fl->l_whence != SEEK_SET) || 24782204Sache fl->l_start < 0 || 24882204Sache (fl->l_len < 0 && 24982204Sache (fl->l_start == 0 || fl->l_start + fl->l_len < 0))) 25082194Sache return (EINVAL); 25182204Sache if (fl->l_len > 0 && 25282204Sache (fl->l_len - 1 > OFF_MAX - fl->l_start)) 25382194Sache return (EOVERFLOW); 25482194Sache } 25575631Salfred 25675631Salfred /* 25775631Salfred * Fill in the information structure. 25875631Salfred */ 25975631Salfred msg.lm_version = LOCKD_MSG_VERSION; 26075631Salfred msg.lm_msg_ident.pid = p->p_pid; 26175631Salfred /* 26275631Salfred * if there is no nfsowner table yet, allocate one. 26375631Salfred */ 26475631Salfred if (p->p_nlminfo == NULL) { 26575631Salfred MALLOC(p->p_nlminfo, struct nlminfo *, 266151695Sglebius sizeof(struct nlminfo), M_NLMINFO, M_WAITOK | M_ZERO); 26775631Salfred p->p_nlminfo->pid_start = p->p_stats->p_start; 268114434Sdes timevaladd(&p->p_nlminfo->pid_start, &boottime); 26975631Salfred } 27075631Salfred msg.lm_msg_ident.pid_start = p->p_nlminfo->pid_start; 27175631Salfred msg.lm_msg_ident.msg_seq = ++(p->p_nlminfo->msg_seq); 27275631Salfred 27375631Salfred msg.lm_fl = *fl; 27475631Salfred msg.lm_wait = ap->a_flags & F_WAIT; 27575631Salfred msg.lm_getlk = ap->a_op == F_GETLK; 276100134Salfred bcopy(VFSTONFS(vp->v_mount)->nm_nam, &msg.lm_addr, 277100134Salfred min(sizeof msg.lm_addr, VFSTONFS(vp->v_mount)->nm_nam->sa_len)); 27875631Salfred msg.lm_fh_len = NFS_ISV3(vp) ? VTONFS(vp)->n_fhsize : NFSX_V2FH; 27975631Salfred bcopy(VTONFS(vp)->n_fhp, msg.lm_fh, msg.lm_fh_len); 28075631Salfred msg.lm_nfsv3 = NFS_ISV3(vp); 281101947Salfred cru2x(td->td_ucred, &msg.lm_cred); 28275631Salfred 28375631Salfred for (;;) { 284138430Sphk error = nfslock_send(&msg); 285138430Sphk if (error) 286138430Sphk return (error); 28775631Salfred 288138430Sphk /* Unlocks succeed immediately. */ 28975631Salfred if (fl->l_type == F_UNLCK) 290138430Sphk return (error); 29175631Salfred 29275631Salfred /* 29375631Salfred * retry after 20 seconds if we haven't gotten a responce yet. 29475631Salfred * This number was picked out of thin air... but is longer 29575631Salfred * then even a reasonably loaded system should take (at least 29683651Speter * on a local network). XXX Probably should use a back-off 29775631Salfred * scheme. 298116185Srwatson * 299116185Srwatson * XXX: No PCATCH here since we currently have no useful 300116185Srwatson * way to signal to the userland rpc.lockd that the request 301116185Srwatson * has been aborted. Once the rpc.lockd implementation 302116185Srwatson * can handle aborts, and we report them properly, 303116185Srwatson * PCATCH can be put back. In the mean time, if we did 304116185Srwatson * permit aborting, the lock attempt would "get lost" 305116185Srwatson * and the lock would get stuck in the locked state. 30675631Salfred */ 307115415Srwatson error = tsleep(p->p_nlminfo, PUSER, "lockd", 20*hz); 308107104Salfred if (error != 0) { 30975631Salfred if (error == EWOULDBLOCK) { 31075631Salfred /* 31175631Salfred * We timed out, so we rewrite the request 312154316Srwatson * to the fifo. 31375631Salfred */ 31475631Salfred continue; 31575631Salfred } 31675631Salfred 31775631Salfred break; 31875631Salfred } 31975631Salfred 32075631Salfred if (msg.lm_getlk && p->p_nlminfo->retcode == 0) { 32175631Salfred if (p->p_nlminfo->set_getlk_pid) { 32275631Salfred fl->l_pid = p->p_nlminfo->getlk_pid; 32375631Salfred } else { 32475631Salfred fl->l_type = F_UNLCK; 32575631Salfred } 32675631Salfred } 32775631Salfred error = p->p_nlminfo->retcode; 32875631Salfred break; 32975631Salfred } 33075631Salfred 331138430Sphk return (error); 33275631Salfred} 33375631Salfred 33475631Salfred/* 33575631Salfred * nfslockdans -- 33675631Salfred * NFS advisory byte-level locks answer from the lock daemon. 33775631Salfred */ 338138430Sphkstatic int 33986363Srwatsonnfslockdans(struct thread *td, struct lockd_ans *ansp) 34075631Salfred{ 34186363Srwatson struct proc *targetp; 34275631Salfred int error; 34375631Salfred 34475631Salfred /* Let root, or someone who once was root (lockd generally 34583651Speter * switches to the daemon uid once it is done setting up) make 34677183Srwatson * this call. 34777183Srwatson * 34877183Srwatson * XXX This authorization check is probably not right. 34975631Salfred */ 35093593Sjhb if ((error = suser(td)) != 0 && 35191406Sjhb td->td_ucred->cr_svuid != 0) 35275631Salfred return (error); 35375631Salfred 35475631Salfred /* the version should match, or we're out of sync */ 35575631Salfred if (ansp->la_vers != LOCKD_ANS_VERSION) 35675631Salfred return (EINVAL); 35775631Salfred 35875631Salfred /* Find the process, set its return errno and wake it up. */ 35986363Srwatson if ((targetp = pfind(ansp->la_msg_ident.pid)) == NULL) 36075631Salfred return (ESRCH); 36175631Salfred 36283651Speter /* verify the pid hasn't been reused (if we can), and it isn't waiting 36375631Salfred * for an answer from a more recent request. We return an EPIPE if 36475631Salfred * the match fails, because we've already used ESRCH above, and this 36575631Salfred * is sort of like writing on a pipe after the reader has closed it. 36675631Salfred */ 36786363Srwatson if (targetp->p_nlminfo == NULL || 36875631Salfred ((ansp->la_msg_ident.msg_seq != -1) && 36986363Srwatson (timevalcmp(&targetp->p_nlminfo->pid_start, 37075631Salfred &ansp->la_msg_ident.pid_start, !=) || 37186363Srwatson targetp->p_nlminfo->msg_seq != ansp->la_msg_ident.msg_seq))) { 37286363Srwatson PROC_UNLOCK(targetp); 37375631Salfred return (EPIPE); 37477563Sjake } 37575631Salfred 37686363Srwatson targetp->p_nlminfo->retcode = ansp->la_errno; 37786363Srwatson targetp->p_nlminfo->set_getlk_pid = ansp->la_set_getlk_pid; 37886363Srwatson targetp->p_nlminfo->getlk_pid = ansp->la_getlk_pid; 37975631Salfred 380107104Salfred wakeup(targetp->p_nlminfo); 38175631Salfred 38286363Srwatson PROC_UNLOCK(targetp); 38375631Salfred return (0); 38475631Salfred} 385138430Sphk 386151695Sglebius/* 387151695Sglebius * Free nlminfo attached to process. 388151695Sglebius */ 389151695Sglebiusvoid 390151695Sglebiusnlminfo_release(struct proc *p) 391151695Sglebius{ 392151695Sglebius free(p->p_nlminfo, M_NLMINFO); 393151695Sglebius p->p_nlminfo = NULL; 394151695Sglebius} 395