1 /* $OpenBSD: nfs_kq.c,v 1.20 2014/07/12 18:43:52 tedu Exp $ */ 2 /* $NetBSD: nfs_kq.c,v 1.7 2003/10/30 01:43:10 simonb Exp $ */ 3 4 /*- 5 * Copyright (c) 2002 The NetBSD Foundation, Inc. 6 * All rights reserved. 7 * 8 * This code is derived from software contributed to The NetBSD Foundation 9 * by Jaromir Dolecek. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/kernel.h> 36 #include <sys/proc.h> 37 #include <sys/mount.h> 38 #include <sys/malloc.h> 39 #include <sys/vnode.h> 40 #include <sys/unistd.h> 41 #include <sys/file.h> 42 #include <sys/kthread.h> 43 #include <sys/rwlock.h> 44 #include <sys/queue.h> 45 46 #include <nfs/rpcv2.h> 47 #include <nfs/nfsproto.h> 48 #include <nfs/nfs.h> 49 #include <nfs/nfsnode.h> 50 #include <nfs/nfs_var.h> 51 52 void nfs_kqpoll(void *); 53 54 void filt_nfsdetach(struct knote *); 55 int filt_nfsread(struct knote *, long); 56 int filt_nfsvnode(struct knote *, long); 57 58 struct kevq { 59 SLIST_ENTRY(kevq) kev_link; 60 struct vnode *vp; 61 u_int usecount; 62 u_int flags; 63 #define KEVQ_BUSY 0x01 /* currently being processed */ 64 #define KEVQ_WANT 0x02 /* want to change this entry */ 65 struct timespec omtime; /* old modification time */ 66 struct timespec octime; /* old change time */ 67 nlink_t onlink; /* old number of references to file */ 68 }; 69 SLIST_HEAD(kevqlist, kevq); 70 71 struct rwlock nfskevq_lock = RWLOCK_INITIALIZER("nfskqlk"); 72 struct proc *pnfskq; 73 struct kevqlist kevlist = SLIST_HEAD_INITIALIZER(kevlist); 74 75 /* 76 * This quite simplistic routine periodically checks for server changes 77 * of any of the watched files every NFS_MINATTRTIMO/2 seconds. 78 * Only changes in size, modification time, change time and nlinks 79 * are being checked, everything else is ignored. 80 * The routine only calls VOP_GETATTR() when it's likely it would get 81 * some new data, i.e. when the vnode expires from attrcache. This 82 * should give same result as periodically running stat(2) from userland, 83 * while keeping CPU/network usage low, and still provide proper kevent 84 * semantics. 85 * The poller thread is created when first vnode is added to watch list, 86 * and exits when the watch list is empty. The overhead of thread creation 87 * isn't really important, neither speed of attach and detach of knote. 88 */ 89 /* ARGSUSED */ 90 void 91 nfs_kqpoll(void *arg) 92 { 93 struct kevq *ke; 94 struct vattr attr; 95 struct proc *p = pnfskq; 96 u_quad_t osize; 97 int error; 98 99 for(;;) { 100 rw_enter_write(&nfskevq_lock); 101 SLIST_FOREACH(ke, &kevlist, kev_link) { 102 struct nfsnode *np = VTONFS(ke->vp); 103 104 #ifdef DEBUG 105 printf("nfs_kqpoll on: "); 106 VOP_PRINT(ke->vp); 107 #endif 108 /* skip if still in attrcache */ 109 if (nfs_getattrcache(ke->vp, &attr) != ENOENT) 110 continue; 111 112 /* 113 * Mark entry busy, release lock and check 114 * for changes. 115 */ 116 ke->flags |= KEVQ_BUSY; 117 rw_exit_write(&nfskevq_lock); 118 119 /* save v_size, nfs_getattr() updates it */ 120 osize = np->n_size; 121 122 error = VOP_GETATTR(ke->vp, &attr, p->p_ucred, p); 123 if (error == ESTALE) { 124 NFS_INVALIDATE_ATTRCACHE(np); 125 VN_KNOTE(ke->vp, NOTE_DELETE); 126 goto next; 127 } 128 129 /* following is a bit fragile, but about best 130 * we can get */ 131 if (attr.va_size != osize) { 132 int flags = NOTE_WRITE; 133 134 if (attr.va_size > osize) 135 flags |= NOTE_EXTEND; 136 else 137 flags |= NOTE_TRUNCATE; 138 139 VN_KNOTE(ke->vp, flags); 140 ke->omtime = attr.va_mtime; 141 } else if (attr.va_mtime.tv_sec != ke->omtime.tv_sec 142 || attr.va_mtime.tv_nsec != ke->omtime.tv_nsec) { 143 VN_KNOTE(ke->vp, NOTE_WRITE); 144 ke->omtime = attr.va_mtime; 145 } 146 147 if (attr.va_ctime.tv_sec != ke->octime.tv_sec 148 || attr.va_ctime.tv_nsec != ke->octime.tv_nsec) { 149 VN_KNOTE(ke->vp, NOTE_ATTRIB); 150 ke->octime = attr.va_ctime; 151 } 152 153 if (attr.va_nlink != ke->onlink) { 154 VN_KNOTE(ke->vp, NOTE_LINK); 155 ke->onlink = attr.va_nlink; 156 } 157 158 next: 159 rw_enter_write(&nfskevq_lock); 160 ke->flags &= ~KEVQ_BUSY; 161 if (ke->flags & KEVQ_WANT) { 162 ke->flags &= ~KEVQ_WANT; 163 wakeup(ke); 164 } 165 } 166 167 if (SLIST_EMPTY(&kevlist)) { 168 /* Nothing more to watch, exit */ 169 pnfskq = NULL; 170 rw_exit_write(&nfskevq_lock); 171 kthread_exit(0); 172 } 173 rw_exit_write(&nfskevq_lock); 174 175 /* wait a while before checking for changes again */ 176 tsleep(pnfskq, PSOCK, "nfskqpw", NFS_MINATTRTIMO * hz / 2); 177 178 } 179 } 180 181 void 182 filt_nfsdetach(struct knote *kn) 183 { 184 struct vnode *vp = (struct vnode *)kn->kn_hook; 185 struct kevq *ke; 186 187 SLIST_REMOVE(&vp->v_selectinfo.si_note, kn, knote, kn_selnext); 188 189 /* Remove the vnode from watch list */ 190 rw_enter_write(&nfskevq_lock); 191 SLIST_FOREACH(ke, &kevlist, kev_link) { 192 if (ke->vp == vp) { 193 while (ke->flags & KEVQ_BUSY) { 194 ke->flags |= KEVQ_WANT; 195 rw_exit_write(&nfskevq_lock); 196 (void) tsleep(ke, PSOCK, "nfskqdet", 0); 197 rw_enter_write(&nfskevq_lock); 198 } 199 200 if (ke->usecount > 1) { 201 /* keep, other kevents need this */ 202 ke->usecount--; 203 } else { 204 /* last user, g/c */ 205 SLIST_REMOVE(&kevlist, ke, kevq, kev_link); 206 free(ke, M_KEVENT, 0); 207 } 208 break; 209 } 210 } 211 rw_exit_write(&nfskevq_lock); 212 } 213 214 int 215 filt_nfsread(struct knote *kn, long hint) 216 { 217 struct vnode *vp = (struct vnode *)kn->kn_hook; 218 struct nfsnode *np = VTONFS(vp); 219 220 /* 221 * filesystem is gone, so set the EOF flag and schedule 222 * the knote for deletion. 223 */ 224 if (hint == NOTE_REVOKE) { 225 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 226 return (1); 227 } 228 229 kn->kn_data = np->n_size - kn->kn_fp->f_offset; 230 #ifdef DEBUG 231 printf("nfsread event. %lld\n", kn->kn_data); 232 #endif 233 return (kn->kn_data != 0); 234 } 235 236 int 237 filt_nfsvnode(struct knote *kn, long hint) 238 { 239 if (kn->kn_sfflags & hint) 240 kn->kn_fflags |= hint; 241 if (hint == NOTE_REVOKE) { 242 kn->kn_flags |= EV_EOF; 243 return (1); 244 } 245 return (kn->kn_fflags != 0); 246 } 247 248 static const struct filterops nfsread_filtops = 249 { 1, NULL, filt_nfsdetach, filt_nfsread }; 250 static const struct filterops nfsvnode_filtops = 251 { 1, NULL, filt_nfsdetach, filt_nfsvnode }; 252 253 int 254 nfs_kqfilter(void *v) 255 { 256 struct vop_kqfilter_args *ap = v; 257 struct vnode *vp; 258 struct knote *kn; 259 struct kevq *ke; 260 int error = 0; 261 struct vattr attr; 262 struct proc *p = curproc; /* XXX */ 263 264 vp = ap->a_vp; 265 kn = ap->a_kn; 266 267 #ifdef DEBUG 268 printf("nfs_kqfilter(%d) on: ", kn->kn_filter); 269 VOP_PRINT(vp); 270 #endif 271 272 switch (kn->kn_filter) { 273 case EVFILT_READ: 274 kn->kn_fop = &nfsread_filtops; 275 break; 276 case EVFILT_VNODE: 277 kn->kn_fop = &nfsvnode_filtops; 278 break; 279 default: 280 return (EINVAL); 281 } 282 283 kn->kn_hook = vp; 284 285 /* 286 * Put the vnode to watched list. 287 */ 288 289 /* 290 * Fetch current attributes. It's only needed when the vnode 291 * is not watched yet, but we need to do this without lock 292 * held. This is likely cheap due to attrcache, so do it now. 293 */ 294 memset(&attr, 0, sizeof(attr)); 295 (void) VOP_GETATTR(vp, &attr, p->p_ucred, p); 296 297 rw_enter_write(&nfskevq_lock); 298 299 /* ensure the poller is running */ 300 if (!pnfskq) { 301 error = kthread_create(nfs_kqpoll, NULL, &pnfskq, 302 "nfskqpoll"); 303 if (error) 304 goto out; 305 } 306 307 SLIST_FOREACH(ke, &kevlist, kev_link) 308 if (ke->vp == vp) 309 break; 310 311 if (ke) { 312 /* already watched, so just bump usecount */ 313 ke->usecount++; 314 } else { 315 /* need a new one */ 316 ke = malloc(sizeof(struct kevq), M_KEVENT, M_WAITOK); 317 ke->vp = vp; 318 ke->usecount = 1; 319 ke->flags = 0; 320 ke->omtime = attr.va_mtime; 321 ke->octime = attr.va_ctime; 322 ke->onlink = attr.va_nlink; 323 SLIST_INSERT_HEAD(&kevlist, ke, kev_link); 324 } 325 326 /* kick the poller */ 327 wakeup(pnfskq); 328 329 SLIST_INSERT_HEAD(&vp->v_selectinfo.si_note, kn, kn_selnext); 330 331 out: 332 rw_exit_write(&nfskevq_lock); 333 return (error); 334 } 335