xref: /freebsd/sys/fs/nfsclient/nfs_clsubs.c (revision f4f8f02054f3abb6ceb84aefcdecc78d5c8b462f)
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  	if ((old_lock = VOP_ISLOCKED(vp)) != LK_EXCLUSIVE) {
134  		if (old_lock == LK_SHARED) {
135  			/* Upgrade to exclusive lock, this might block */
136  			vn_lock(vp, LK_UPGRADE | LK_RETRY);
137  		} else {
138  			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
139  		}
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  		if (old_lock == LK_SHARED) {
149  			/* Downgrade from exclusive lock, this might block */
150  			vn_lock(vp, LK_DOWNGRADE);
151  		} else {
152  			VOP_UNLOCK(vp, 0);
153  		}
154   	}
155 }
156 
157 void
158 ncl_printf(const char *fmt, ...)
159 {
160 	va_list ap;
161 
162 	mtx_lock(&Giant);
163 	va_start(ap, fmt);
164 	printf(fmt, ap);
165 	va_end(ap);
166 	mtx_unlock(&Giant);
167 }
168 
169 #ifdef NFS_ACDEBUG
170 #include <sys/sysctl.h>
171 SYSCTL_DECL(_vfs_newnfs);
172 static int nfs_acdebug;
173 SYSCTL_INT(_vfs_newnfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
174 #endif
175 
176 /*
177  * Check the time stamp
178  * If the cache is valid, copy contents to *vap and return 0
179  * otherwise return an error
180  */
181 int
182 ncl_getattrcache(struct vnode *vp, struct vattr *vaper)
183 {
184 	struct nfsnode *np;
185 	struct vattr *vap;
186 	struct nfsmount *nmp;
187 	int timeo;
188 
189 	np = VTONFS(vp);
190 	vap = &np->n_vattr.na_vattr;
191 	nmp = VFSTONFS(vp->v_mount);
192 #ifdef NFS_ACDEBUG
193 	mtx_lock(&Giant);	/* ncl_printf() */
194 #endif
195 	mtx_lock(&np->n_mtx);
196 	/* XXX n_mtime doesn't seem to be updated on a miss-and-reload */
197 	timeo = (time_second - np->n_mtime.tv_sec) / 10;
198 
199 #ifdef NFS_ACDEBUG
200 	if (nfs_acdebug>1)
201 		ncl_printf("nfs_getattrcache: initial timeo = %d\n", timeo);
202 #endif
203 
204 	if (vap->va_type == VDIR) {
205 		if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acdirmin)
206 			timeo = nmp->nm_acdirmin;
207 		else if (timeo > nmp->nm_acdirmax)
208 			timeo = nmp->nm_acdirmax;
209 	} else {
210 		if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acregmin)
211 			timeo = nmp->nm_acregmin;
212 		else if (timeo > nmp->nm_acregmax)
213 			timeo = nmp->nm_acregmax;
214 	}
215 
216 #ifdef NFS_ACDEBUG
217 	if (nfs_acdebug > 2)
218 		ncl_printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
219 			   nmp->nm_acregmin, nmp->nm_acregmax,
220 			   nmp->nm_acdirmin, nmp->nm_acdirmax);
221 
222 	if (nfs_acdebug)
223 		ncl_printf("nfs_getattrcache: age = %d; final timeo = %d\n",
224 			   (time_second - np->n_attrstamp), timeo);
225 #endif
226 
227 	if ((time_second - np->n_attrstamp) >= timeo) {
228 		newnfsstats.attrcache_misses++;
229 		mtx_unlock(&np->n_mtx);
230 		return( ENOENT);
231 	}
232 	newnfsstats.attrcache_hits++;
233 	if (vap->va_size != np->n_size) {
234 		if (vap->va_type == VREG) {
235 			if (np->n_flag & NMODIFIED) {
236 				if (vap->va_size < np->n_size)
237 					vap->va_size = np->n_size;
238 				else
239 					np->n_size = vap->va_size;
240 			} else {
241 				np->n_size = vap->va_size;
242 			}
243 			vnode_pager_setsize(vp, np->n_size);
244 		} else {
245 			np->n_size = vap->va_size;
246 		}
247 	}
248 	bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
249 	if (np->n_flag & NCHG) {
250 		if (np->n_flag & NACC)
251 			vaper->va_atime = np->n_atim;
252 		if (np->n_flag & NUPD)
253 			vaper->va_mtime = np->n_mtim;
254 	}
255 	mtx_unlock(&np->n_mtx);
256 #ifdef NFS_ACDEBUG
257 	mtx_unlock(&Giant);	/* ncl_printf() */
258 #endif
259 	return (0);
260 }
261 
262 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
263 /*
264  * This function finds the directory cookie that corresponds to the
265  * logical byte offset given.
266  */
267 nfsuint64 *
268 ncl_getcookie(struct nfsnode *np, off_t off, int add)
269 {
270 	struct nfsdmap *dp, *dp2;
271 	int pos;
272 	nfsuint64 *retval = NULL;
273 
274 	pos = (uoff_t)off / NFS_DIRBLKSIZ;
275 	if (pos == 0 || off < 0) {
276 #ifdef DIAGNOSTIC
277 		if (add)
278 			panic("nfs getcookie add at <= 0");
279 #endif
280 		return (&nfs_nullcookie);
281 	}
282 	pos--;
283 	dp = LIST_FIRST(&np->n_cookies);
284 	if (!dp) {
285 		if (add) {
286 			MALLOC(dp, struct nfsdmap *, sizeof (struct nfsdmap),
287 				M_NFSDIROFF, M_WAITOK);
288 			dp->ndm_eocookie = 0;
289 			LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
290 		} else
291 			goto out;
292 	}
293 	while (pos >= NFSNUMCOOKIES) {
294 		pos -= NFSNUMCOOKIES;
295 		if (LIST_NEXT(dp, ndm_list)) {
296 			if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
297 			    pos >= dp->ndm_eocookie)
298 				goto out;
299 			dp = LIST_NEXT(dp, ndm_list);
300 		} else if (add) {
301 			MALLOC(dp2, struct nfsdmap *, sizeof (struct nfsdmap),
302 				M_NFSDIROFF, M_WAITOK);
303 			dp2->ndm_eocookie = 0;
304 			LIST_INSERT_AFTER(dp, dp2, ndm_list);
305 			dp = dp2;
306 		} else
307 			goto out;
308 	}
309 	if (pos >= dp->ndm_eocookie) {
310 		if (add)
311 			dp->ndm_eocookie = pos + 1;
312 		else
313 			goto out;
314 	}
315 	retval = &dp->ndm_cookies[pos];
316 out:
317 	return (retval);
318 }
319 
320 /*
321  * Invalidate cached directory information, except for the actual directory
322  * blocks (which are invalidated separately).
323  * Done mainly to avoid the use of stale offset cookies.
324  */
325 void
326 ncl_invaldir(struct vnode *vp)
327 {
328 	struct nfsnode *np = VTONFS(vp);
329 
330 #ifdef DIAGNOSTIC
331 	if (vp->v_type != VDIR)
332 		panic("nfs: invaldir not dir");
333 #endif
334 	ncl_dircookie_lock(np);
335 	np->n_direofoffset = 0;
336 	np->n_cookieverf.nfsuquad[0] = 0;
337 	np->n_cookieverf.nfsuquad[1] = 0;
338 	if (LIST_FIRST(&np->n_cookies))
339 		LIST_FIRST(&np->n_cookies)->ndm_eocookie = 0;
340 	ncl_dircookie_unlock(np);
341 }
342 
343 /*
344  * The write verifier has changed (probably due to a server reboot), so all
345  * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
346  * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
347  * and B_CLUSTEROK flags.  Once done the new write verifier can be set for the
348  * mount point.
349  *
350  * B_CLUSTEROK must be cleared along with B_NEEDCOMMIT because stage 1 data
351  * writes are not clusterable.
352  */
353 void
354 ncl_clearcommit(struct mount *mp)
355 {
356 	struct vnode *vp, *nvp;
357 	struct buf *bp, *nbp;
358 	struct bufobj *bo;
359 
360 	MNT_ILOCK(mp);
361 	MNT_VNODE_FOREACH(vp, mp, nvp) {
362 		bo = &vp->v_bufobj;
363 		VI_LOCK(vp);
364 		if (vp->v_iflag & VI_DOOMED) {
365 			VI_UNLOCK(vp);
366 			continue;
367 		}
368 		vholdl(vp);
369 		VI_UNLOCK(vp);
370 		MNT_IUNLOCK(mp);
371 		BO_LOCK(bo);
372 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
373 			if (!BUF_ISLOCKED(bp) &&
374 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
375 				== (B_DELWRI | B_NEEDCOMMIT))
376 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
377 		}
378 		BO_UNLOCK(bo);
379 		vdrop(vp);
380 		MNT_ILOCK(mp);
381 	}
382 	MNT_IUNLOCK(mp);
383 }
384 
385 /*
386  * Called once to initialize data structures...
387  */
388 int
389 ncl_init(struct vfsconf *vfsp)
390 {
391 	int i;
392 
393 	/* Ensure async daemons disabled */
394 	for (i = 0; i < NFS_MAXRAHEAD; i++) {
395 		ncl_iodwant[i] = NULL;
396 		ncl_iodmount[i] = NULL;
397 	}
398 	ncl_nhinit();			/* Init the nfsnode table */
399 
400 	return (0);
401 }
402 
403