xref: /freebsd/sys/fs/nfsclient/nfs_clsubs.c (revision 35ae9291c2621d66ac66ed4a4996761946ac3e2d)
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 
61 #include <vm/vm.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_extern.h>
64 #include <vm/uma.h>
65 
66 #include <fs/nfs/nfsport.h>
67 #include <fs/nfsclient/nfsnode.h>
68 #include <fs/nfsclient/nfsmount.h>
69 #include <fs/nfsclient/nfs.h>
70 #include <fs/nfsclient/nfs_lock.h>
71 
72 #include <netinet/in.h>
73 
74 /*
75  * Note that stdarg.h and the ANSI style va_start macro is used for both
76  * ANSI and traditional C compilers.
77  */
78 #include <machine/stdarg.h>
79 
80 extern struct mtx ncl_iod_mutex;
81 extern struct proc *ncl_iodwant[NFS_MAXRAHEAD];
82 extern struct nfsmount *ncl_iodmount[NFS_MAXRAHEAD];
83 extern int ncl_numasync;
84 extern unsigned int ncl_iodmax;
85 extern struct nfsstats newnfsstats;
86 
87 int
88 ncl_uninit(struct vfsconf *vfsp)
89 {
90 	int i;
91 
92 	/*
93 	 * Tell all nfsiod processes to exit. Clear ncl_iodmax, and wakeup
94 	 * any sleeping nfsiods so they check ncl_iodmax and exit.
95 	 */
96 	mtx_lock(&ncl_iod_mutex);
97 	ncl_iodmax = 0;
98 	for (i = 0; i < ncl_numasync; i++)
99 		if (ncl_iodwant[i])
100 			wakeup(&ncl_iodwant[i]);
101 	/* The last nfsiod to exit will wake us up when ncl_numasync hits 0 */
102 	while (ncl_numasync)
103 		msleep(&ncl_numasync, &ncl_iod_mutex, PWAIT, "ioddie", 0);
104 	mtx_unlock(&ncl_iod_mutex);
105 	ncl_nhuninit();
106 	return (0);
107 }
108 
109 void
110 ncl_dircookie_lock(struct nfsnode *np)
111 {
112 	mtx_lock(&np->n_mtx);
113 	while (np->n_flag & NDIRCOOKIELK)
114 		(void) msleep(&np->n_flag, &np->n_mtx, PZERO, "nfsdirlk", 0);
115 	np->n_flag |= NDIRCOOKIELK;
116 	mtx_unlock(&np->n_mtx);
117 }
118 
119 void
120 ncl_dircookie_unlock(struct nfsnode *np)
121 {
122 	mtx_lock(&np->n_mtx);
123 	np->n_flag &= ~NDIRCOOKIELK;
124 	wakeup(&np->n_flag);
125 	mtx_unlock(&np->n_mtx);
126 }
127 
128 int
129 ncl_upgrade_vnlock(struct vnode *vp)
130 {
131 	int old_lock;
132 
133 	ASSERT_VOP_LOCKED(vp, "ncl_upgrade_vnlock");
134 	old_lock = VOP_ISLOCKED(vp);
135 	if (old_lock != LK_EXCLUSIVE) {
136 		KASSERT(old_lock == LK_SHARED,
137 		    ("ncl_upgrade_vnlock: wrong old_lock %d", old_lock));
138 		/* Upgrade to exclusive lock, this might block */
139 		vn_lock(vp, LK_UPGRADE | LK_RETRY);
140   	}
141 	return (old_lock);
142 }
143 
144 void
145 ncl_downgrade_vnlock(struct vnode *vp, int old_lock)
146 {
147 	if (old_lock != LK_EXCLUSIVE) {
148 		KASSERT(old_lock == LK_SHARED, ("wrong old_lock %d", old_lock));
149 		/* Downgrade from exclusive lock. */
150 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
151   	}
152 }
153 
154 void
155 ncl_printf(const char *fmt, ...)
156 {
157 	va_list ap;
158 
159 	mtx_lock(&Giant);
160 	va_start(ap, fmt);
161 	printf(fmt, ap);
162 	va_end(ap);
163 	mtx_unlock(&Giant);
164 }
165 
166 #ifdef NFS_ACDEBUG
167 #include <sys/sysctl.h>
168 SYSCTL_DECL(_vfs_newnfs);
169 static int nfs_acdebug;
170 SYSCTL_INT(_vfs_newnfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
171 #endif
172 
173 /*
174  * Check the time stamp
175  * If the cache is valid, copy contents to *vap and return 0
176  * otherwise return an error
177  */
178 int
179 ncl_getattrcache(struct vnode *vp, struct vattr *vaper)
180 {
181 	struct nfsnode *np;
182 	struct vattr *vap;
183 	struct nfsmount *nmp;
184 	int timeo;
185 
186 	np = VTONFS(vp);
187 	vap = &np->n_vattr.na_vattr;
188 	nmp = VFSTONFS(vp->v_mount);
189 #ifdef NFS_ACDEBUG
190 	mtx_lock(&Giant);	/* ncl_printf() */
191 #endif
192 	mtx_lock(&np->n_mtx);
193 	/* XXX n_mtime doesn't seem to be updated on a miss-and-reload */
194 	timeo = (time_second - np->n_mtime.tv_sec) / 10;
195 
196 #ifdef NFS_ACDEBUG
197 	if (nfs_acdebug>1)
198 		ncl_printf("nfs_getattrcache: initial timeo = %d\n", timeo);
199 #endif
200 
201 	if (vap->va_type == VDIR) {
202 		if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acdirmin)
203 			timeo = nmp->nm_acdirmin;
204 		else if (timeo > nmp->nm_acdirmax)
205 			timeo = nmp->nm_acdirmax;
206 	} else {
207 		if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acregmin)
208 			timeo = nmp->nm_acregmin;
209 		else if (timeo > nmp->nm_acregmax)
210 			timeo = nmp->nm_acregmax;
211 	}
212 
213 #ifdef NFS_ACDEBUG
214 	if (nfs_acdebug > 2)
215 		ncl_printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
216 			   nmp->nm_acregmin, nmp->nm_acregmax,
217 			   nmp->nm_acdirmin, nmp->nm_acdirmax);
218 
219 	if (nfs_acdebug)
220 		ncl_printf("nfs_getattrcache: age = %d; final timeo = %d\n",
221 			   (time_second - np->n_attrstamp), timeo);
222 #endif
223 
224 	if ((time_second - np->n_attrstamp) >= timeo) {
225 		newnfsstats.attrcache_misses++;
226 		mtx_unlock(&np->n_mtx);
227 		return( ENOENT);
228 	}
229 	newnfsstats.attrcache_hits++;
230 	if (vap->va_size != np->n_size) {
231 		if (vap->va_type == VREG) {
232 			if (np->n_flag & NMODIFIED) {
233 				if (vap->va_size < np->n_size)
234 					vap->va_size = np->n_size;
235 				else
236 					np->n_size = vap->va_size;
237 			} else {
238 				np->n_size = vap->va_size;
239 			}
240 			vnode_pager_setsize(vp, np->n_size);
241 		} else {
242 			np->n_size = vap->va_size;
243 		}
244 	}
245 	bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
246 	if (np->n_flag & NCHG) {
247 		if (np->n_flag & NACC)
248 			vaper->va_atime = np->n_atim;
249 		if (np->n_flag & NUPD)
250 			vaper->va_mtime = np->n_mtim;
251 	}
252 	mtx_unlock(&np->n_mtx);
253 #ifdef NFS_ACDEBUG
254 	mtx_unlock(&Giant);	/* ncl_printf() */
255 #endif
256 	return (0);
257 }
258 
259 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
260 /*
261  * This function finds the directory cookie that corresponds to the
262  * logical byte offset given.
263  */
264 nfsuint64 *
265 ncl_getcookie(struct nfsnode *np, off_t off, int add)
266 {
267 	struct nfsdmap *dp, *dp2;
268 	int pos;
269 	nfsuint64 *retval = NULL;
270 
271 	pos = (uoff_t)off / NFS_DIRBLKSIZ;
272 	if (pos == 0 || off < 0) {
273 #ifdef DIAGNOSTIC
274 		if (add)
275 			panic("nfs getcookie add at <= 0");
276 #endif
277 		return (&nfs_nullcookie);
278 	}
279 	pos--;
280 	dp = LIST_FIRST(&np->n_cookies);
281 	if (!dp) {
282 		if (add) {
283 			MALLOC(dp, struct nfsdmap *, sizeof (struct nfsdmap),
284 				M_NFSDIROFF, M_WAITOK);
285 			dp->ndm_eocookie = 0;
286 			LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
287 		} else
288 			goto out;
289 	}
290 	while (pos >= NFSNUMCOOKIES) {
291 		pos -= NFSNUMCOOKIES;
292 		if (LIST_NEXT(dp, ndm_list)) {
293 			if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
294 			    pos >= dp->ndm_eocookie)
295 				goto out;
296 			dp = LIST_NEXT(dp, ndm_list);
297 		} else if (add) {
298 			MALLOC(dp2, struct nfsdmap *, sizeof (struct nfsdmap),
299 				M_NFSDIROFF, M_WAITOK);
300 			dp2->ndm_eocookie = 0;
301 			LIST_INSERT_AFTER(dp, dp2, ndm_list);
302 			dp = dp2;
303 		} else
304 			goto out;
305 	}
306 	if (pos >= dp->ndm_eocookie) {
307 		if (add)
308 			dp->ndm_eocookie = pos + 1;
309 		else
310 			goto out;
311 	}
312 	retval = &dp->ndm_cookies[pos];
313 out:
314 	return (retval);
315 }
316 
317 /*
318  * Invalidate cached directory information, except for the actual directory
319  * blocks (which are invalidated separately).
320  * Done mainly to avoid the use of stale offset cookies.
321  */
322 void
323 ncl_invaldir(struct vnode *vp)
324 {
325 	struct nfsnode *np = VTONFS(vp);
326 
327 #ifdef DIAGNOSTIC
328 	if (vp->v_type != VDIR)
329 		panic("nfs: invaldir not dir");
330 #endif
331 	ncl_dircookie_lock(np);
332 	np->n_direofoffset = 0;
333 	np->n_cookieverf.nfsuquad[0] = 0;
334 	np->n_cookieverf.nfsuquad[1] = 0;
335 	if (LIST_FIRST(&np->n_cookies))
336 		LIST_FIRST(&np->n_cookies)->ndm_eocookie = 0;
337 	ncl_dircookie_unlock(np);
338 }
339 
340 /*
341  * The write verifier has changed (probably due to a server reboot), so all
342  * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
343  * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
344  * and B_CLUSTEROK flags.  Once done the new write verifier can be set for the
345  * mount point.
346  *
347  * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
348  * writes are not clusterable.
349  */
350 void
351 ncl_clearcommit(struct mount *mp)
352 {
353 	struct vnode *vp, *nvp;
354 	struct buf *bp, *nbp;
355 	struct bufobj *bo;
356 
357 	MNT_ILOCK(mp);
358 	MNT_VNODE_FOREACH(vp, mp, nvp) {
359 		bo = &vp->v_bufobj;
360 		VI_LOCK(vp);
361 		if (vp->v_iflag & VI_DOOMED) {
362 			VI_UNLOCK(vp);
363 			continue;
364 		}
365 		vholdl(vp);
366 		VI_UNLOCK(vp);
367 		MNT_IUNLOCK(mp);
368 		BO_LOCK(bo);
369 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
370 			if (!BUF_ISLOCKED(bp) &&
371 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
372 				== (B_DELWRI | B_NEEDCOMMIT))
373 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
374 		}
375 		BO_UNLOCK(bo);
376 		vdrop(vp);
377 		MNT_ILOCK(mp);
378 	}
379 	MNT_IUNLOCK(mp);
380 }
381 
382 /*
383  * Called once to initialize data structures...
384  */
385 int
386 ncl_init(struct vfsconf *vfsp)
387 {
388 	int i;
389 
390 	/* Ensure async daemons disabled */
391 	for (i = 0; i < NFS_MAXRAHEAD; i++) {
392 		ncl_iodwant[i] = NULL;
393 		ncl_iodmount[i] = NULL;
394 	}
395 	ncl_nhinit();			/* Init the nfsnode table */
396 
397 	return (0);
398 }
399 
400