Home | History | Annotate | Line # | Download | only in nfs
nfs_node.c revision 1.116
      1 /*	$NetBSD: nfs_node.c,v 1.116 2011/06/12 03:35:59 rmind Exp $	*/
      2 
      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  *	@(#)nfs_node.c	8.6 (Berkeley) 5/22/95
     35  */
     36 
     37 #include <sys/cdefs.h>
     38 __KERNEL_RCSID(0, "$NetBSD: nfs_node.c,v 1.116 2011/06/12 03:35:59 rmind Exp $");
     39 
     40 #ifdef _KERNEL_OPT
     41 #include "opt_nfs.h"
     42 #endif
     43 
     44 #include <sys/param.h>
     45 #include <sys/systm.h>
     46 #include <sys/proc.h>
     47 #include <sys/mount.h>
     48 #include <sys/namei.h>
     49 #include <sys/vnode.h>
     50 #include <sys/kernel.h>
     51 #include <sys/pool.h>
     52 #include <sys/lock.h>
     53 #include <sys/hash.h>
     54 #include <sys/kauth.h>
     55 
     56 #include <nfs/rpcv2.h>
     57 #include <nfs/nfsproto.h>
     58 #include <nfs/nfs.h>
     59 #include <nfs/nfsnode.h>
     60 #include <nfs/nfsmount.h>
     61 #include <nfs/nfs_var.h>
     62 
     63 struct pool nfs_node_pool;
     64 struct pool nfs_vattr_pool;
     65 static struct workqueue *nfs_sillyworkq;
     66 
     67 extern int prtactive;
     68 
     69 static void nfs_gop_size(struct vnode *, off_t, off_t *, int);
     70 static int nfs_gop_alloc(struct vnode *, off_t, off_t, int, kauth_cred_t);
     71 static int nfs_gop_write(struct vnode *, struct vm_page **, int, int);
     72 static void nfs_sillyworker(struct work *, void *);
     73 
     74 static const struct genfs_ops nfs_genfsops = {
     75 	.gop_size = nfs_gop_size,
     76 	.gop_alloc = nfs_gop_alloc,
     77 	.gop_write = nfs_gop_write,
     78 };
     79 
     80 /*
     81  * Reinitialize inode hash table.
     82  */
     83 void
     84 nfs_node_init(void)
     85 {
     86 
     87 	pool_init(&nfs_node_pool, sizeof(struct nfsnode), 0, 0, 0, "nfsnodepl",
     88 	    &pool_allocator_nointr, IPL_NONE);
     89 	pool_init(&nfs_vattr_pool, sizeof(struct vattr), 0, 0, 0, "nfsvapl",
     90 	    &pool_allocator_nointr, IPL_NONE);
     91 	if (workqueue_create(&nfs_sillyworkq, "nfssilly", nfs_sillyworker,
     92 	    NULL, PRI_NONE, IPL_NONE, 0) != 0) {
     93 	    	panic("nfs_node_init");
     94 	}
     95 }
     96 
     97 /*
     98  * Free resources previously allocated in nfs_node_reinit().
     99  */
    100 void
    101 nfs_node_done(void)
    102 {
    103 
    104 	pool_destroy(&nfs_node_pool);
    105 	pool_destroy(&nfs_vattr_pool);
    106 	workqueue_destroy(nfs_sillyworkq);
    107 }
    108 
    109 struct fh_match {
    110 	nfsfh_t *fhm_fhp;
    111 	size_t fhm_fhsize;
    112 	size_t fhm_fhoffset;
    113 };
    114 
    115 static int
    116 nfs_compare_nodes(void *ctx, const void *parent, const void *node)
    117 {
    118 	const struct nfsnode * const pnp = parent;
    119 	const struct nfsnode * const np = node;
    120 
    121 	if (pnp->n_fhsize != np->n_fhsize)
    122 		return np->n_fhsize - pnp->n_fhsize;
    123 
    124 	return memcmp(np->n_fhp, pnp->n_fhp, np->n_fhsize);
    125 }
    126 
    127 static int
    128 nfs_compare_node_fh(void *ctx, const void *b, const void *key)
    129 {
    130 	const struct nfsnode * const pnp = b;
    131 	const struct fh_match * const fhm = key;
    132 
    133 	if (pnp->n_fhsize != fhm->fhm_fhsize)
    134 		return fhm->fhm_fhsize - pnp->n_fhsize;
    135 
    136 	return memcmp(fhm->fhm_fhp, pnp->n_fhp, pnp->n_fhsize);
    137 }
    138 
    139 static const rb_tree_ops_t nfs_node_rbtree_ops = {
    140 	.rbto_compare_nodes = nfs_compare_nodes,
    141 	.rbto_compare_key = nfs_compare_node_fh,
    142 	.rbto_node_offset = offsetof(struct nfsnode, n_rbnode),
    143 	.rbto_context = NULL
    144 };
    145 
    146 void
    147 nfs_rbtinit(struct nfsmount *nmp)
    148 {
    149 
    150 	rb_tree_init(&nmp->nm_rbtree, &nfs_node_rbtree_ops);
    151 }
    152 
    153 /*
    154  * Look up a vnode/nfsnode by file handle.
    155  * Callers must check for mount points!!
    156  * In all cases, a pointer to a
    157  * nfsnode structure is returned.
    158  */
    159 int
    160 nfs_nget1(struct mount *mntp, nfsfh_t *fhp, int fhsize, struct nfsnode **npp,
    161     int lkflags)
    162 {
    163 	struct nfsnode *np;
    164 	struct vnode *vp;
    165 	struct nfsmount *nmp = VFSTONFS(mntp);
    166 	int error;
    167 	struct fh_match fhm;
    168 
    169 	fhm.fhm_fhp = fhp;
    170 	fhm.fhm_fhsize = fhsize;
    171 
    172 loop:
    173 	rw_enter(&nmp->nm_rbtlock, RW_READER);
    174 	np = rb_tree_find_node(&nmp->nm_rbtree, &fhm);
    175 	if (np != NULL) {
    176 		vp = NFSTOV(np);
    177 		mutex_enter(vp->v_interlock);
    178 		rw_exit(&nmp->nm_rbtlock);
    179 		error = vget(vp, LK_EXCLUSIVE | lkflags);
    180 		if (error == EBUSY)
    181 			return error;
    182 		if (error)
    183 			goto loop;
    184 		*npp = np;
    185 		return(0);
    186 	}
    187 	rw_exit(&nmp->nm_rbtlock);
    188 
    189 	error = getnewvnode(VT_NFS, mntp, nfsv2_vnodeop_p, NULL, &vp);
    190 	if (error) {
    191 		*npp = 0;
    192 		return (error);
    193 	}
    194 	np = pool_get(&nfs_node_pool, PR_WAITOK);
    195 	memset(np, 0, sizeof *np);
    196 	np->n_vnode = vp;
    197 
    198 	/*
    199 	 * Insert the nfsnode in the hash queue for its new file handle
    200 	 */
    201 
    202 	if (fhsize > NFS_SMALLFH) {
    203 		np->n_fhp = kmem_alloc(fhsize, KM_SLEEP);
    204 	} else
    205 		np->n_fhp = &np->n_fh;
    206 	memcpy(np->n_fhp, fhp, fhsize);
    207 	np->n_fhsize = fhsize;
    208 	np->n_accstamp = -1;
    209 	np->n_vattr = pool_get(&nfs_vattr_pool, PR_WAITOK);
    210 
    211 	rw_enter(&nmp->nm_rbtlock, RW_WRITER);
    212 	if (NULL != rb_tree_find_node(&nmp->nm_rbtree, &fhm)) {
    213 		rw_exit(&nmp->nm_rbtlock);
    214 		if (fhsize > NFS_SMALLFH) {
    215 			kmem_free(np->n_fhp, fhsize);
    216 		}
    217 		pool_put(&nfs_vattr_pool, np->n_vattr);
    218 		pool_put(&nfs_node_pool, np);
    219 		ungetnewvnode(vp);
    220 		goto loop;
    221 	}
    222 	vp->v_data = np;
    223 	genfs_node_init(vp, &nfs_genfsops);
    224 	/*
    225 	 * Initalize read/write creds to useful values. VOP_OPEN will
    226 	 * overwrite these.
    227 	 */
    228 	np->n_rcred = curlwp->l_cred;
    229 	kauth_cred_hold(np->n_rcred);
    230 	np->n_wcred = curlwp->l_cred;
    231 	kauth_cred_hold(np->n_wcred);
    232 	VOP_LOCK(vp, LK_EXCLUSIVE);
    233 	NFS_INVALIDATE_ATTRCACHE(np);
    234 	uvm_vnp_setsize(vp, 0);
    235 	(void)rb_tree_insert_node(&nmp->nm_rbtree, np);
    236 	rw_exit(&nmp->nm_rbtlock);
    237 
    238 	*npp = np;
    239 	return (0);
    240 }
    241 
    242 int
    243 nfs_inactive(void *v)
    244 {
    245 	struct vop_inactive_args /* {
    246 		struct vnode *a_vp;
    247 		bool *a_recycle;
    248 	} */ *ap = v;
    249 	struct nfsnode *np;
    250 	struct sillyrename *sp;
    251 	struct vnode *vp = ap->a_vp;
    252 
    253 	np = VTONFS(vp);
    254 	if (vp->v_type != VDIR) {
    255 		sp = np->n_sillyrename;
    256 		np->n_sillyrename = (struct sillyrename *)0;
    257 	} else
    258 		sp = NULL;
    259 	if (sp != NULL)
    260 		nfs_vinvalbuf(vp, 0, sp->s_cred, curlwp, 1);
    261 	*ap->a_recycle = (np->n_flag & NREMOVED) != 0;
    262 	np->n_flag &=
    263 	    (NMODIFIED | NFLUSHINPROG | NFLUSHWANT | NEOFVALID | NTRUNCDELAYED);
    264 
    265 	if (vp->v_type == VDIR && np->n_dircache)
    266 		nfs_invaldircache(vp,
    267 		    NFS_INVALDIRCACHE_FORCE | NFS_INVALDIRCACHE_KEEPEOF);
    268 
    269 	VOP_UNLOCK(vp);
    270 
    271 	if (sp != NULL) {
    272 		workqueue_enqueue(nfs_sillyworkq, &sp->s_work, NULL);
    273 	}
    274 
    275 	return (0);
    276 }
    277 
    278 /*
    279  * Reclaim an nfsnode so that it can be used for other purposes.
    280  */
    281 int
    282 nfs_reclaim(void *v)
    283 {
    284 	struct vop_reclaim_args /* {
    285 		struct vnode *a_vp;
    286 	} */ *ap = v;
    287 	struct vnode *vp = ap->a_vp;
    288 	struct nfsnode *np = VTONFS(vp);
    289 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
    290 
    291 	if (prtactive && vp->v_usecount > 1)
    292 		vprint("nfs_reclaim: pushing active", vp);
    293 
    294 	rw_enter(&nmp->nm_rbtlock, RW_WRITER);
    295 	rb_tree_remove_node(&nmp->nm_rbtree, np);
    296 	rw_exit(&nmp->nm_rbtlock);
    297 
    298 	/*
    299 	 * Free up any directory cookie structures and
    300 	 * large file handle structures that might be associated with
    301 	 * this nfs node.
    302 	 */
    303 	if (vp->v_type == VDIR && np->n_dircache != NULL) {
    304 		nfs_invaldircache(vp, NFS_INVALDIRCACHE_FORCE);
    305 		hashdone(np->n_dircache, HASH_LIST, nfsdirhashmask);
    306 	}
    307 	KASSERT(np->n_dirgens == NULL);
    308 
    309 	if (np->n_fhsize > NFS_SMALLFH)
    310 		kmem_free(np->n_fhp, np->n_fhsize);
    311 
    312 	pool_put(&nfs_vattr_pool, np->n_vattr);
    313 	if (np->n_rcred)
    314 		kauth_cred_free(np->n_rcred);
    315 
    316 	if (np->n_wcred)
    317 		kauth_cred_free(np->n_wcred);
    318 
    319 	if (vp->v_type == VREG) {
    320 		mutex_destroy(&np->n_commitlock);
    321 	}
    322 	genfs_node_destroy(vp);
    323 	pool_put(&nfs_node_pool, np);
    324 	vp->v_data = NULL;
    325 	return (0);
    326 }
    327 
    328 void
    329 nfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags)
    330 {
    331 
    332 	*eobp = MAX(size, vp->v_size);
    333 }
    334 
    335 int
    336 nfs_gop_alloc(struct vnode *vp, off_t off, off_t len, int flags,
    337     kauth_cred_t cred)
    338 {
    339 
    340 	return 0;
    341 }
    342 
    343 int
    344 nfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
    345 {
    346 	int i;
    347 
    348 	mutex_enter(vp->v_interlock);
    349 	for (i = 0; i < npages; i++) {
    350 		pmap_page_protect(pgs[i], VM_PROT_READ);
    351 	}
    352 	mutex_exit(vp->v_interlock);
    353 
    354 	return genfs_gop_write(vp, pgs, npages, flags);
    355 }
    356 
    357 /*
    358  * Remove a silly file that was rename'd earlier
    359  */
    360 static void
    361 nfs_sillyworker(struct work *work, void *arg)
    362 {
    363 	struct sillyrename *sp;
    364 	int error;
    365 
    366 	sp = (struct sillyrename *)work;
    367 	error = vn_lock(sp->s_dvp, LK_EXCLUSIVE);
    368 	if (error || sp->s_dvp->v_data == NULL) {
    369 		/* XXX should recover */
    370 		printf("%s: vp=%p error=%d\n", __func__, sp->s_dvp, error);
    371 		if (error == 0) {
    372 			vput(sp->s_dvp);
    373 		} else {
    374 			vrele(sp->s_dvp);
    375 		}
    376 	} else {
    377 		nfs_removeit(sp);
    378 		vput(sp->s_dvp);
    379 	}
    380 	kauth_cred_free(sp->s_cred);
    381 	kmem_free(sp, sizeof(*sp));
    382 }
    383