xref: /freebsd/sys/fs/nfsclient/nfs_clsubs.c (revision 640235e2c2ba32947f7c59d168437ffa1280f1e6)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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.
19  *
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
30  * SUCH DAMAGE.
31  *
32  *	from nfs_subs.c  8.8 (Berkeley) 5/22/95
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 /*
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.
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/proc.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/namei.h>
53 #include <sys/mbuf.h>
54 #include <sys/socket.h>
55 #include <sys/stat.h>
56 #include <sys/malloc.h>
57 #include <sys/sysent.h>
58 #include <sys/syscall.h>
59 #include <sys/sysproto.h>
60 #include <sys/taskqueue.h>
61 
62 #include <vm/vm.h>
63 #include <vm/vm_object.h>
64 #include <vm/vm_extern.h>
65 #include <vm/uma.h>
66 
67 #include <fs/nfs/nfsport.h>
68 #include <fs/nfsclient/nfsnode.h>
69 #include <fs/nfsclient/nfsmount.h>
70 #include <fs/nfsclient/nfs.h>
71 #include <fs/nfsclient/nfs_kdtrace.h>
72 
73 #include <netinet/in.h>
74 
75 /*
76  * Note that stdarg.h and the ANSI style va_start macro is used for both
77  * ANSI and traditional C compilers.
78  */
79 #include <machine/stdarg.h>
80 
81 extern struct mtx ncl_iod_mutex;
82 extern enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON];
83 extern struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON];
84 extern int ncl_numasync;
85 extern unsigned int ncl_iodmax;
86 extern struct nfsstats newnfsstats;
87 
88 struct task	ncl_nfsiodnew_task;
89 
90 int
91 ncl_uninit(struct vfsconf *vfsp)
92 {
93 	/*
94 	 * XXX: Unloading of nfscl module is unsupported.
95 	 */
96 #if 0
97 	int i;
98 
99 	/*
100 	 * Tell all nfsiod processes to exit. Clear ncl_iodmax, and wakeup
101 	 * any sleeping nfsiods so they check ncl_iodmax and exit.
102 	 */
103 	mtx_lock(&ncl_iod_mutex);
104 	ncl_iodmax = 0;
105 	for (i = 0; i < ncl_numasync; i++)
106 		if (ncl_iodwant[i] == NFSIOD_AVAILABLE)
107 			wakeup(&ncl_iodwant[i]);
108 	/* The last nfsiod to exit will wake us up when ncl_numasync hits 0 */
109 	while (ncl_numasync)
110 		msleep(&ncl_numasync, &ncl_iod_mutex, PWAIT, "ioddie", 0);
111 	mtx_unlock(&ncl_iod_mutex);
112 	ncl_nhuninit();
113 	return (0);
114 #else
115 	return (EOPNOTSUPP);
116 #endif
117 }
118 
119 void
120 ncl_dircookie_lock(struct nfsnode *np)
121 {
122 	mtx_lock(&np->n_mtx);
123 	while (np->n_flag & NDIRCOOKIELK)
124 		(void) msleep(&np->n_flag, &np->n_mtx, PZERO, "nfsdirlk", 0);
125 	np->n_flag |= NDIRCOOKIELK;
126 	mtx_unlock(&np->n_mtx);
127 }
128 
129 void
130 ncl_dircookie_unlock(struct nfsnode *np)
131 {
132 	mtx_lock(&np->n_mtx);
133 	np->n_flag &= ~NDIRCOOKIELK;
134 	wakeup(&np->n_flag);
135 	mtx_unlock(&np->n_mtx);
136 }
137 
138 int
139 ncl_upgrade_vnlock(struct vnode *vp)
140 {
141 	int old_lock;
142 
143 	ASSERT_VOP_LOCKED(vp, "ncl_upgrade_vnlock");
144 	old_lock = NFSVOPISLOCKED(vp);
145 	if (old_lock != LK_EXCLUSIVE) {
146 		KASSERT(old_lock == LK_SHARED,
147 		    ("ncl_upgrade_vnlock: wrong old_lock %d", old_lock));
148 		/* Upgrade to exclusive lock, this might block */
149 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
150   	}
151 	return (old_lock);
152 }
153 
154 void
155 ncl_downgrade_vnlock(struct vnode *vp, int old_lock)
156 {
157 	if (old_lock != LK_EXCLUSIVE) {
158 		KASSERT(old_lock == LK_SHARED, ("wrong old_lock %d", old_lock));
159 		/* Downgrade from exclusive lock. */
160 		NFSVOPLOCK(vp, LK_DOWNGRADE | LK_RETRY);
161   	}
162 }
163 
164 #ifdef NFS_ACDEBUG
165 #include <sys/sysctl.h>
166 SYSCTL_DECL(_vfs_nfs);
167 static int nfs_acdebug;
168 SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
169 #endif
170 
171 /*
172  * Check the time stamp
173  * If the cache is valid, copy contents to *vap and return 0
174  * otherwise return an error
175  */
176 int
177 ncl_getattrcache(struct vnode *vp, struct vattr *vaper)
178 {
179 	struct nfsnode *np;
180 	struct vattr *vap;
181 	struct nfsmount *nmp;
182 	int timeo, mustflush;
183 
184 	np = VTONFS(vp);
185 	vap = &np->n_vattr.na_vattr;
186 	nmp = VFSTONFS(vp->v_mount);
187 	mustflush = nfscl_mustflush(vp);	/* must be before mtx_lock() */
188 	mtx_lock(&np->n_mtx);
189 	/* XXX n_mtime doesn't seem to be updated on a miss-and-reload */
190 	timeo = (time_second - np->n_mtime.tv_sec) / 10;
191 
192 #ifdef NFS_ACDEBUG
193 	if (nfs_acdebug>1)
194 		printf("ncl_getattrcache: initial timeo = %d\n", timeo);
195 #endif
196 
197 	if (vap->va_type == VDIR) {
198 		if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acdirmin)
199 			timeo = nmp->nm_acdirmin;
200 		else if (timeo > nmp->nm_acdirmax)
201 			timeo = nmp->nm_acdirmax;
202 	} else {
203 		if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acregmin)
204 			timeo = nmp->nm_acregmin;
205 		else if (timeo > nmp->nm_acregmax)
206 			timeo = nmp->nm_acregmax;
207 	}
208 
209 #ifdef NFS_ACDEBUG
210 	if (nfs_acdebug > 2)
211 		printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
212 		    nmp->nm_acregmin, nmp->nm_acregmax,
213 		    nmp->nm_acdirmin, nmp->nm_acdirmax);
214 
215 	if (nfs_acdebug)
216 		printf("ncl_getattrcache: age = %d; final timeo = %d\n",
217 		    (time_second - np->n_attrstamp), timeo);
218 #endif
219 
220 	if ((time_second - np->n_attrstamp) >= timeo &&
221 	    (mustflush != 0 || np->n_attrstamp == 0)) {
222 		newnfsstats.attrcache_misses++;
223 		mtx_unlock(&np->n_mtx);
224 		KDTRACE_NFS_ATTRCACHE_GET_MISS(vp);
225 		return( ENOENT);
226 	}
227 	newnfsstats.attrcache_hits++;
228 	if (vap->va_size != np->n_size) {
229 		if (vap->va_type == VREG) {
230 			if (np->n_flag & NMODIFIED) {
231 				if (vap->va_size < np->n_size)
232 					vap->va_size = np->n_size;
233 				else
234 					np->n_size = vap->va_size;
235 			} else {
236 				np->n_size = vap->va_size;
237 			}
238 			vnode_pager_setsize(vp, np->n_size);
239 		} else {
240 			np->n_size = vap->va_size;
241 		}
242 	}
243 	bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
244 	if (np->n_flag & NCHG) {
245 		if (np->n_flag & NACC)
246 			vaper->va_atime = np->n_atim;
247 		if (np->n_flag & NUPD)
248 			vaper->va_mtime = np->n_mtim;
249 	}
250 	mtx_unlock(&np->n_mtx);
251 	KDTRACE_NFS_ATTRCACHE_GET_HIT(vp, vap);
252 	return (0);
253 }
254 
255 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
256 /*
257  * This function finds the directory cookie that corresponds to the
258  * logical byte offset given.
259  */
260 nfsuint64 *
261 ncl_getcookie(struct nfsnode *np, off_t off, int add)
262 {
263 	struct nfsdmap *dp, *dp2;
264 	int pos;
265 	nfsuint64 *retval = NULL;
266 
267 	pos = (uoff_t)off / NFS_DIRBLKSIZ;
268 	if (pos == 0 || off < 0) {
269 		KASSERT(!add, ("nfs getcookie add at <= 0"));
270 		return (&nfs_nullcookie);
271 	}
272 	pos--;
273 	dp = LIST_FIRST(&np->n_cookies);
274 	if (!dp) {
275 		if (add) {
276 			MALLOC(dp, struct nfsdmap *, sizeof (struct nfsdmap),
277 				M_NFSDIROFF, M_WAITOK);
278 			dp->ndm_eocookie = 0;
279 			LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
280 		} else
281 			goto out;
282 	}
283 	while (pos >= NFSNUMCOOKIES) {
284 		pos -= NFSNUMCOOKIES;
285 		if (LIST_NEXT(dp, ndm_list)) {
286 			if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
287 			    pos >= dp->ndm_eocookie)
288 				goto out;
289 			dp = LIST_NEXT(dp, ndm_list);
290 		} else if (add) {
291 			MALLOC(dp2, struct nfsdmap *, sizeof (struct nfsdmap),
292 				M_NFSDIROFF, M_WAITOK);
293 			dp2->ndm_eocookie = 0;
294 			LIST_INSERT_AFTER(dp, dp2, ndm_list);
295 			dp = dp2;
296 		} else
297 			goto out;
298 	}
299 	if (pos >= dp->ndm_eocookie) {
300 		if (add)
301 			dp->ndm_eocookie = pos + 1;
302 		else
303 			goto out;
304 	}
305 	retval = &dp->ndm_cookies[pos];
306 out:
307 	return (retval);
308 }
309 
310 /*
311  * Invalidate cached directory information, except for the actual directory
312  * blocks (which are invalidated separately).
313  * Done mainly to avoid the use of stale offset cookies.
314  */
315 void
316 ncl_invaldir(struct vnode *vp)
317 {
318 	struct nfsnode *np = VTONFS(vp);
319 
320 	KASSERT(vp->v_type == VDIR, ("nfs: invaldir not dir"));
321 	ncl_dircookie_lock(np);
322 	np->n_direofoffset = 0;
323 	np->n_cookieverf.nfsuquad[0] = 0;
324 	np->n_cookieverf.nfsuquad[1] = 0;
325 	if (LIST_FIRST(&np->n_cookies))
326 		LIST_FIRST(&np->n_cookies)->ndm_eocookie = 0;
327 	ncl_dircookie_unlock(np);
328 }
329 
330 /*
331  * The write verifier has changed (probably due to a server reboot), so all
332  * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
333  * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
334  * and B_CLUSTEROK flags.  Once done the new write verifier can be set for the
335  * mount point.
336  *
337  * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
338  * writes are not clusterable.
339  */
340 void
341 ncl_clearcommit(struct mount *mp)
342 {
343 	struct vnode *vp, *nvp;
344 	struct buf *bp, *nbp;
345 	struct bufobj *bo;
346 
347 	MNT_VNODE_FOREACH_ALL(vp, mp, nvp) {
348 		bo = &vp->v_bufobj;
349 		vholdl(vp);
350 		VI_UNLOCK(vp);
351 		BO_LOCK(bo);
352 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
353 			if (!BUF_ISLOCKED(bp) &&
354 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
355 				== (B_DELWRI | B_NEEDCOMMIT))
356 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
357 		}
358 		BO_UNLOCK(bo);
359 		vdrop(vp);
360 	}
361 }
362 
363 /*
364  * Called once to initialize data structures...
365  */
366 int
367 ncl_init(struct vfsconf *vfsp)
368 {
369 	int i;
370 
371 	/* Ensure async daemons disabled */
372 	for (i = 0; i < NFS_MAXASYNCDAEMON; i++) {
373 		ncl_iodwant[i] = NFSIOD_NOT_AVAILABLE;
374 		ncl_iodmount[i] = NULL;
375 	}
376 	TASK_INIT(&ncl_nfsiodnew_task, 0, ncl_nfsiodnew_tq, NULL);
377 	ncl_nhinit();			/* Init the nfsnode table */
378 
379 	return (0);
380 }
381 
382