xref: /freebsd/sys/fs/nfsclient/nfs_clsubs.c (revision 3c4ba5f55438f7afd4f4b0b56f88f2bb505fd6a6)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from nfs_subs.c  8.8 (Berkeley) 5/22/95
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 /*
41  * These functions support the macros and help fiddle mbuf chains for
42  * the nfs op functions. They do things like create the rpc header and
43  * copy data between mbuf chains and uio lists.
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/proc.h>
52 #include <sys/mount.h>
53 #include <sys/vnode.h>
54 #include <sys/namei.h>
55 #include <sys/mbuf.h>
56 #include <sys/socket.h>
57 #include <sys/stat.h>
58 #include <sys/malloc.h>
59 #include <sys/syscall.h>
60 #include <sys/sysproto.h>
61 #include <sys/taskqueue.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_extern.h>
66 #include <vm/uma.h>
67 
68 #include <fs/nfs/nfsport.h>
69 #include <fs/nfsclient/nfsnode.h>
70 #include <fs/nfsclient/nfsmount.h>
71 #include <fs/nfsclient/nfs.h>
72 #include <fs/nfsclient/nfs_kdtrace.h>
73 
74 #include <netinet/in.h>
75 
76 /*
77  * Note that stdarg.h and the ANSI style va_start macro is used for both
78  * ANSI and traditional C compilers.
79  */
80 #include <machine/stdarg.h>
81 
82 extern struct mtx ncl_iod_mutex;
83 extern enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON];
84 extern struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON];
85 extern int ncl_numasync;
86 extern unsigned int ncl_iodmax;
87 extern struct nfsstatsv1 nfsstatsv1;
88 
89 struct task	ncl_nfsiodnew_task;
90 
91 int
92 ncl_uninit(struct vfsconf *vfsp)
93 {
94 	/*
95 	 * XXX: Unloading of nfscl module is unsupported.
96 	 */
97 #if 0
98 	int i;
99 
100 	/*
101 	 * Tell all nfsiod processes to exit. Clear ncl_iodmax, and wakeup
102 	 * any sleeping nfsiods so they check ncl_iodmax and exit.
103 	 */
104 	NFSLOCKIOD();
105 	ncl_iodmax = 0;
106 	for (i = 0; i < ncl_numasync; i++)
107 		if (ncl_iodwant[i] == NFSIOD_AVAILABLE)
108 			wakeup(&ncl_iodwant[i]);
109 	/* The last nfsiod to exit will wake us up when ncl_numasync hits 0 */
110 	while (ncl_numasync)
111 		msleep(&ncl_numasync, &ncl_iod_mutex, PWAIT, "ioddie", 0);
112 	NFSUNLOCKIOD();
113 	ncl_nhuninit();
114 	return (0);
115 #else
116 	return (EOPNOTSUPP);
117 #endif
118 }
119 
120 /* Returns with NFSLOCKNODE() held. */
121 void
122 ncl_dircookie_lock(struct nfsnode *np)
123 {
124 	NFSLOCKNODE(np);
125 	while (np->n_flag & NDIRCOOKIELK)
126 		(void) msleep(&np->n_flag, &np->n_mtx, PZERO, "nfsdirlk", 0);
127 	np->n_flag |= NDIRCOOKIELK;
128 }
129 
130 void
131 ncl_dircookie_unlock(struct nfsnode *np)
132 {
133 	NFSLOCKNODE(np);
134 	np->n_flag &= ~NDIRCOOKIELK;
135 	wakeup(&np->n_flag);
136 	NFSUNLOCKNODE(np);
137 }
138 
139 bool
140 ncl_excl_start(struct vnode *vp)
141 {
142 	struct nfsnode *np;
143 	int vn_lk;
144 
145 	ASSERT_VOP_LOCKED(vp, "ncl_excl_start");
146 	vn_lk = NFSVOPISLOCKED(vp);
147 	if (vn_lk == LK_EXCLUSIVE)
148 		return (false);
149 	KASSERT(vn_lk == LK_SHARED,
150 	    ("ncl_excl_start: wrong vnode lock %d", vn_lk));
151 	/* Ensure exclusive access, this might block */
152 	np = VTONFS(vp);
153 	lockmgr(&np->n_excl, LK_EXCLUSIVE, NULL);
154 	return (true);
155 }
156 
157 void
158 ncl_excl_finish(struct vnode *vp, bool old_lock)
159 {
160 	struct nfsnode *np;
161 
162 	if (!old_lock)
163 		return;
164 	np = VTONFS(vp);
165 	lockmgr(&np->n_excl, LK_RELEASE, NULL);
166 }
167 
168 #ifdef NFS_ACDEBUG
169 #include <sys/sysctl.h>
170 SYSCTL_DECL(_vfs_nfs);
171 static int nfs_acdebug;
172 SYSCTL_INT(_vfs_nfs, OID_AUTO, acdebug, CTLFLAG_RW, &nfs_acdebug, 0, "");
173 #endif
174 
175 /*
176  * Check the time stamp
177  * If the cache is valid, copy contents to *vap and return 0
178  * otherwise return an error
179  */
180 int
181 ncl_getattrcache(struct vnode *vp, struct vattr *vaper)
182 {
183 	struct nfsnode *np;
184 	struct vattr *vap;
185 	struct nfsmount *nmp;
186 	int timeo, mustflush;
187 	u_quad_t nsize;
188 	bool setnsize;
189 
190 	np = VTONFS(vp);
191 	vap = &np->n_vattr.na_vattr;
192 	nmp = VFSTONFS(vp->v_mount);
193 	mustflush = nfscl_mustflush(vp);	/* must be before mtx_lock() */
194 	NFSLOCKNODE(np);
195 	/* XXX n_mtime doesn't seem to be updated on a miss-and-reload */
196 	timeo = (time_second - np->n_mtime.tv_sec) / 10;
197 
198 #ifdef NFS_ACDEBUG
199 	if (nfs_acdebug>1)
200 		printf("ncl_getattrcache: initial timeo = %d\n", timeo);
201 #endif
202 
203 	if (vap->va_type == VDIR) {
204 		if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acdirmin)
205 			timeo = nmp->nm_acdirmin;
206 		else if (timeo > nmp->nm_acdirmax)
207 			timeo = nmp->nm_acdirmax;
208 	} else {
209 		if ((np->n_flag & NMODIFIED) || timeo < nmp->nm_acregmin)
210 			timeo = nmp->nm_acregmin;
211 		else if (timeo > nmp->nm_acregmax)
212 			timeo = nmp->nm_acregmax;
213 	}
214 
215 #ifdef NFS_ACDEBUG
216 	if (nfs_acdebug > 2)
217 		printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
218 		    nmp->nm_acregmin, nmp->nm_acregmax,
219 		    nmp->nm_acdirmin, nmp->nm_acdirmax);
220 
221 	if (nfs_acdebug)
222 		printf("ncl_getattrcache: age = %d; final timeo = %d\n",
223 		    (time_second - np->n_attrstamp), timeo);
224 #endif
225 
226 	if ((time_second - np->n_attrstamp) >= timeo &&
227 	    (mustflush != 0 || np->n_attrstamp == 0)) {
228 		nfsstatsv1.attrcache_misses++;
229 		NFSUNLOCKNODE(np);
230 		KDTRACE_NFS_ATTRCACHE_GET_MISS(vp);
231 		return( ENOENT);
232 	}
233 	nfsstatsv1.attrcache_hits++;
234 	setnsize = false;
235 	if (vap->va_size != np->n_size) {
236 		if (vap->va_type == VREG) {
237 			if (np->n_flag & NMODIFIED) {
238 				if (vap->va_size < np->n_size)
239 					vap->va_size = np->n_size;
240 				else
241 					np->n_size = vap->va_size;
242 			} else {
243 				np->n_size = vap->va_size;
244 			}
245 			setnsize = ncl_pager_setsize(vp, &nsize);
246 		} else {
247 			np->n_size = vap->va_size;
248 		}
249 	}
250 	bcopy((caddr_t)vap, (caddr_t)vaper, sizeof(struct vattr));
251 	if (np->n_flag & NCHG) {
252 		if (np->n_flag & NACC)
253 			vaper->va_atime = np->n_atim;
254 		if (np->n_flag & NUPD)
255 			vaper->va_mtime = np->n_mtim;
256 	}
257 	NFSUNLOCKNODE(np);
258 	if (setnsize)
259 		vnode_pager_setsize(vp, nsize);
260 	KDTRACE_NFS_ATTRCACHE_GET_HIT(vp, vap);
261 	return (0);
262 }
263 
264 static nfsuint64 nfs_nullcookie = { { 0, 0 } };
265 /*
266  * This function finds the directory cookie that corresponds to the
267  * logical byte offset given.
268  */
269 nfsuint64 *
270 ncl_getcookie(struct nfsnode *np, off_t off, int add)
271 {
272 	struct nfsdmap *dp, *dp2;
273 	int pos;
274 	nfsuint64 *retval = NULL;
275 
276 	pos = (uoff_t)off / NFS_DIRBLKSIZ;
277 	if (pos == 0 || off < 0) {
278 		KASSERT(!add, ("nfs getcookie add at <= 0"));
279 		return (&nfs_nullcookie);
280 	}
281 	pos--;
282 	dp = LIST_FIRST(&np->n_cookies);
283 	if (!dp) {
284 		if (add) {
285 			dp = malloc(sizeof (struct nfsdmap),
286 				M_NFSDIROFF, M_WAITOK);
287 			dp->ndm_eocookie = 0;
288 			LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
289 		} else
290 			goto out;
291 	}
292 	while (pos >= NFSNUMCOOKIES) {
293 		pos -= NFSNUMCOOKIES;
294 		if (LIST_NEXT(dp, ndm_list)) {
295 			if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
296 			    pos >= dp->ndm_eocookie)
297 				goto out;
298 			dp = LIST_NEXT(dp, ndm_list);
299 		} else if (add) {
300 			dp2 = malloc(sizeof (struct nfsdmap),
301 				M_NFSDIROFF, M_WAITOK);
302 			dp2->ndm_eocookie = 0;
303 			LIST_INSERT_AFTER(dp, dp2, ndm_list);
304 			dp = dp2;
305 		} else
306 			goto out;
307 	}
308 	if (pos >= dp->ndm_eocookie) {
309 		if (add)
310 			dp->ndm_eocookie = pos + 1;
311 		else
312 			goto out;
313 	}
314 	retval = &dp->ndm_cookies[pos];
315 out:
316 	return (retval);
317 }
318 
319 /*
320  * Invalidate cached directory information, except for the actual directory
321  * blocks (which are invalidated separately).
322  * Done mainly to avoid the use of stale offset cookies.
323  */
324 void
325 ncl_invaldir(struct vnode *vp)
326 {
327 	struct nfsnode *np = VTONFS(vp);
328 
329 	KASSERT(vp->v_type == VDIR, ("nfs: invaldir not dir"));
330 	ncl_dircookie_lock(np);
331 	np->n_direofoffset = 0;
332 	NFSUNLOCKNODE(np);
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_VNODE_FOREACH_ALL(vp, mp, nvp) {
358 		bo = &vp->v_bufobj;
359 		vholdl(vp);
360 		VI_UNLOCK(vp);
361 		BO_LOCK(bo);
362 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
363 			if (!BUF_ISLOCKED(bp) &&
364 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
365 				== (B_DELWRI | B_NEEDCOMMIT))
366 				bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
367 		}
368 		BO_UNLOCK(bo);
369 		vdrop(vp);
370 	}
371 }
372 
373 /*
374  * Called once to initialize data structures...
375  */
376 int
377 ncl_init(struct vfsconf *vfsp)
378 {
379 	int i;
380 
381 	/* Ensure async daemons disabled */
382 	for (i = 0; i < NFS_MAXASYNCDAEMON; i++) {
383 		ncl_iodwant[i] = NFSIOD_NOT_AVAILABLE;
384 		ncl_iodmount[i] = NULL;
385 	}
386 	TASK_INIT(&ncl_nfsiodnew_task, 0, ncl_nfsiodnew_tq, NULL);
387 	ncl_nhinit();			/* Init the nfsnode table */
388 
389 	return (0);
390 }
391