xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs4_srv_ns.c (revision 06260e34c28df0a3a191e42e1b51c9de745b3510)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 /*
27  * Copyright 2018 Nexenta Systems, Inc.
28  * Copyright (c) 2015, Joyent, Inc.
29  */
30 
31 #include <sys/systm.h>
32 
33 #include <nfs/nfs.h>
34 #include <nfs/export.h>
35 #include <sys/cmn_err.h>
36 #include <sys/avl.h>
37 
38 #define	PSEUDOFS_SUFFIX		" (pseudo)"
39 
40 /*
41  * A version of VOP_FID that deals with a remote VOP_FID for nfs.
42  * If vp is an nfs node, nfs4_fid() returns EREMOTE, nfs3_fid() and nfs_fid()
43  * returns the filehandle of vp as its fid. When nfs uses fid to set the
44  * exportinfo filehandle template, a remote nfs filehandle would be too big for
45  * the fid of the exported directory. This routine remaps the value of the
46  * attribute va_nodeid of vp to be the fid of vp, so that the fid can fit.
47  *
48  * We need this fid mainly for setting up NFSv4 server namespace where an
49  * nfs filesystem is also part of it. Thus, need to be able to setup a pseudo
50  * exportinfo for an nfs node.
51  *
52  * e.g. mount a filesystem on top of a nfs dir, and then share the new mount
53  *      (like exporting a local disk from a "diskless" client)
54  */
55 int
vop_fid_pseudo(vnode_t * vp,fid_t * fidp)56 vop_fid_pseudo(vnode_t *vp, fid_t *fidp)
57 {
58 	struct vattr va;
59 	int error;
60 
61 	error = VOP_FID(vp, fidp, NULL);
62 
63 	/*
64 	 * XXX nfs4_fid() does nothing and returns EREMOTE.
65 	 * XXX nfs3_fid()/nfs_fid() returns nfs filehandle as its fid
66 	 * which has a bigger length than local fid.
67 	 * NFS_FH4MAXDATA is the size of
68 	 * fhandle4_t.fh_xdata[NFS_FH4MAXDATA].
69 	 *
70 	 * Note: nfs[2,3,4]_fid() only gets called for diskless clients.
71 	 */
72 	if (error == EREMOTE ||
73 	    (error == 0 && fidp->fid_len > NFS_FH4MAXDATA)) {
74 
75 		va.va_mask = AT_NODEID;
76 		error = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
77 		if (error)
78 			return (error);
79 
80 		fidp->fid_len = sizeof (va.va_nodeid);
81 		bcopy(&va.va_nodeid, fidp->fid_data, fidp->fid_len);
82 		return (0);
83 	}
84 
85 	return (error);
86 }
87 
88 /*
89  * Get an nfsv4 vnode of the given fid from the visible list of an
90  * nfs filesystem or get the exi_vp if it is the root node.
91  */
92 int
nfs4_vget_pseudo(struct exportinfo * exi,vnode_t ** vpp,fid_t * fidp)93 nfs4_vget_pseudo(struct exportinfo *exi, vnode_t **vpp, fid_t *fidp)
94 {
95 	fid_t exp_fid;
96 	struct exp_visible *visp;
97 	int error;
98 
99 	/* check if the given fid is in the visible list */
100 
101 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
102 		if (EQFID(fidp, &visp->vis_fid)) {
103 			VN_HOLD(visp->vis_vp);
104 			*vpp = visp->vis_vp;
105 			return (0);
106 		}
107 	}
108 
109 	/* check if the given fid is the same as the exported node */
110 
111 	bzero(&exp_fid, sizeof (exp_fid));
112 	exp_fid.fid_len = MAXFIDSZ;
113 	error = vop_fid_pseudo(exi->exi_vp, &exp_fid);
114 	if (error)
115 		return (error);
116 
117 	if (EQFID(fidp, &exp_fid)) {
118 		VN_HOLD(exi->exi_vp);
119 		*vpp = exi->exi_vp;
120 		return (0);
121 	}
122 
123 	return (ENOENT);
124 }
125 
126 /*
127  * Create a pseudo export entry
128  *
129  * This is an export entry that's created as the
130  * side-effect of a "real" export.  As a part of
131  * a real export, the pathname to the export is
132  * checked to see if all the directory components
133  * are accessible via an NFSv4 client, i.e. are
134  * exported.  If treeclimb_export() finds an unexported
135  * mountpoint along the path, then it calls this
136  * function to export it.
137  *
138  * This pseudo export differs from a real export in that
139  * it only allows read-only access.  A "visible" list of
140  * directories is added to filter lookup and readdir results
141  * to only contain dirnames which lead to descendant shares.
142  *
143  * A visible list has a per-file-system scope.  Any exportinfo
144  * struct (real or pseudo) can have a visible list as long as
145  * a) its export root is VROOT, or is the zone's root for in-zone NFS service
146  * b) a descendant of the export root is shared
147  */
148 struct exportinfo *
pseudo_exportfs(nfs_export_t * ne,vnode_t * vp,fid_t * fid,struct exp_visible * vis_head,struct exportdata * exdata)149 pseudo_exportfs(nfs_export_t *ne, vnode_t *vp, fid_t *fid,
150     struct exp_visible *vis_head, struct exportdata *exdata)
151 {
152 	struct exportinfo *exi;
153 	struct exportdata *kex;
154 	fsid_t fsid;
155 	int vpathlen;
156 	int i;
157 
158 	ASSERT(RW_WRITE_HELD(&ne->exported_lock));
159 
160 	fsid = vp->v_vfsp->vfs_fsid;
161 	exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
162 	exi->exi_fsid = fsid;
163 	exi->exi_fid = *fid;
164 	exi->exi_vp = vp;
165 	VN_HOLD(exi->exi_vp);
166 	exi->exi_visible = vis_head;
167 	exi->exi_count = 1;
168 	exi->exi_zoneid = ne->ne_globals->nfs_zoneid;
169 	exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
170 	    VSW_VOLATILEDEV) ? 1 : 0;
171 	mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
172 
173 	/*
174 	 * Build up the template fhandle
175 	 */
176 	exi->exi_fh.fh_fsid = fsid;
177 	ASSERT(exi->exi_fid.fid_len <= sizeof (exi->exi_fh.fh_xdata));
178 	exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
179 	bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
180 	    exi->exi_fid.fid_len);
181 	exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
182 
183 	kex = &exi->exi_export;
184 	kex->ex_flags = EX_PSEUDO;
185 
186 	vpathlen = strlen(vp->v_path);
187 	kex->ex_pathlen = vpathlen + strlen(PSEUDOFS_SUFFIX);
188 	kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
189 
190 	if (vpathlen)
191 		(void) strncpy(kex->ex_path, vp->v_path, vpathlen);
192 	(void) strcpy(kex->ex_path + vpathlen, PSEUDOFS_SUFFIX);
193 
194 	/* Transfer the secinfo data from exdata to this new pseudo node */
195 	if (exdata)
196 		srv_secinfo_exp2pseu(&exi->exi_export, exdata);
197 
198 	/*
199 	 * Initialize auth cache and auth cache lock
200 	 */
201 	for (i = 0; i < AUTH_TABLESIZE; i++) {
202 		exi->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
203 		avl_create(exi->exi_cache[i], nfsauth_cache_clnt_compar,
204 		    sizeof (struct auth_cache_clnt),
205 		    offsetof(struct auth_cache_clnt, authc_link));
206 	}
207 	rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
208 
209 	/*
210 	 * Insert the new entry at the front of the export list
211 	 */
212 	export_link(ne, exi);
213 
214 	/*
215 	 * Initialize exi_id and exi_kstats
216 	 */
217 	mutex_enter(&nfs_exi_id_lock);
218 	exi->exi_id = exi_id_get_next();
219 	avl_add(&exi_id_tree, exi);
220 	mutex_exit(&nfs_exi_id_lock);
221 
222 	return (exi);
223 }
224 
225 /*
226  * Free a list of visible directories
227  */
228 void
free_visible(struct exp_visible * head)229 free_visible(struct exp_visible *head)
230 {
231 	struct exp_visible *visp, *next;
232 
233 	for (visp = head; visp; visp = next) {
234 		if (visp->vis_vp != NULL)
235 			VN_RELE(visp->vis_vp);
236 
237 		next = visp->vis_next;
238 		srv_secinfo_list_free(visp->vis_secinfo, visp->vis_seccnt);
239 		kmem_free(visp, sizeof (*visp));
240 	}
241 }
242 
243 /*
244  * Connects newchild (or subtree with newchild in head)
245  * to the parent node. We always add it to the beginning
246  * of sibling list.
247  */
248 static void
tree_add_child(treenode_t * parent,treenode_t * newchild)249 tree_add_child(treenode_t *parent, treenode_t *newchild)
250 {
251 	newchild->tree_parent = parent;
252 	newchild->tree_sibling = parent->tree_child_first;
253 	parent->tree_child_first = newchild;
254 }
255 
256 /* Look up among direct children a node with the exact tree_vis pointer */
257 static treenode_t *
tree_find_child_by_vis(treenode_t * t,exp_visible_t * vis)258 tree_find_child_by_vis(treenode_t *t, exp_visible_t *vis)
259 {
260 	for (t = t->tree_child_first; t; t = t->tree_sibling)
261 		if (t->tree_vis == vis)
262 			return (t);
263 	return (NULL);
264 }
265 
266 /*
267  * Add new node to the head of subtree pointed by 'n'. n can be NULL.
268  * Interconnects the new treenode with exp_visible and exportinfo
269  * if needed.
270  */
271 static treenode_t *
tree_prepend_node(treenode_t * n,exp_visible_t * v,exportinfo_t * e)272 tree_prepend_node(treenode_t *n, exp_visible_t *v, exportinfo_t *e)
273 {
274 	treenode_t *tnode = kmem_zalloc(sizeof (*tnode), KM_SLEEP);
275 
276 	if (n) {
277 		tnode->tree_child_first = n;
278 		n->tree_parent = tnode;
279 	}
280 	if (v) {
281 		tnode->tree_vis = v;
282 	}
283 	if (e) {
284 		tnode->tree_exi = e;
285 		e->exi_tree = tnode;
286 	}
287 	return (tnode);
288 }
289 
290 /*
291  * Removes node from the tree and frees the treenode struct.
292  * Does not free structures pointed by tree_exi and tree_vis,
293  * they should be already freed.
294  */
295 static void
tree_remove_node(nfs_export_t * ne,treenode_t * node)296 tree_remove_node(nfs_export_t *ne, treenode_t *node)
297 {
298 	treenode_t *parent = node->tree_parent;
299 	treenode_t *s; /* s for sibling */
300 
301 	if (parent == NULL) {
302 		kmem_free(node, sizeof (*node));
303 		ne->ns_root = NULL;
304 		return;
305 	}
306 	/* This node is first child */
307 	if (parent->tree_child_first == node) {
308 		parent->tree_child_first = node->tree_sibling;
309 	/* This node is not first child */
310 	} else {
311 		s = parent->tree_child_first;
312 		while (s->tree_sibling != node)
313 			s = s->tree_sibling;
314 		s->tree_sibling = s->tree_sibling->tree_sibling;
315 	}
316 	kmem_free(node, sizeof (*node));
317 }
318 
319 /*
320  * When we export a new directory we need to add a new
321  * path segment through the pseudofs to reach the new
322  * directory. This new path is reflected in a list of
323  * directories added to the "visible" list.
324  *
325  * Here there are two lists of visible fids: one hanging off the
326  * pseudo exportinfo, and the one we want to add.  It's possible
327  * that the two lists share a common path segment
328  * and have some common directories.  We need to combine
329  * the lists so there's no duplicate entries. Where a common
330  * path component is found, the vis_count field is bumped.
331  *
332  * This example shows that the treenode chain (tree_head) and
333  * exp_visible chain (vis_head) can differ in length. The latter
334  * can be shorter. The outer loop must loop over the vis_head chain.
335  *
336  * share /x/a
337  * mount -F ufs /dev/dsk/... /x/y
338  * mkdir -p /x/y/a/b
339  * share  /x/y/a/b
340  *
341  * When more_visible() is called during the second share,
342  * the existing namespace is following:
343  *                                   exp_visible_t
344  *   treenode_t       exportinfo_t      v0     v1
345  * ns_root+---+        +------------+  +---+  +---+
346  *      t0| / |........| E0 pseudo  |->| x |->| a |
347  *        +---+        +------------+  +---+  +---+
348  *          |                           /    /
349  *        +---+                        /    /
350  *      t1| x |------------------------    /
351  *        +---+                           /
352  *          |                            /
353  *        +---+                         /
354  *      t2| a |-------------------------
355  *        +---+........+------------+
356  *                     | E1 real    |
357  *                     +------------+
358  *
359  * This is being added:
360  *
361  *    tree_head  vis_head
362  *        +---+  +---+
363  *      t3| x |->| x |v2
364  *        +---+  +---+
365  *          |      |
366  *        +---+  +---+                     v4     v5
367  *      t4| y |->| y |v3  +------------+  +---+  +---+
368  *        +---+\ +---+    | E2 pseudo  |->| a |->| b |
369  *          |   \....... >+------------+  +---+  +---+
370  *        +---+                           /      /
371  *      t5| a |---------------------------      /
372  *        +---+                                /
373  *          |                                 /
374  *        +---+-------------------------------
375  *      t6| b |           +------------+
376  *        +---+..........>| E3 real    |
377  *                        +------------+
378  *
379  * more_visible() will:
380  * - kmem_free() t3 and v2
381  * - add t4, t5, t6 as a child of t1 (t4 will become sibling of t2)
382  * - add v3 to the end of E0->exi_visible
383  *
384  * Note that v4 and v5 were already processed in pseudo_exportfs() and
385  * added to E2. The outer loop of more_visible() will loop only over v2
386  * and v3. The inner loop of more_visible() always loops over v0 and v1.
387  *
388  * Illustration for this scenario:
389  *
390  * mkdir -p /v/a/b/c
391  * share /v/a/b/c
392  * mkdir /v/a/b/c1
393  * mkdir -p /v/a1
394  * mv /v/a/b /v/a1
395  * share /v/a1/b/c1
396  *
397  *           EXISTING
398  *           treenode
399  *           namespace:    +-----------+   visibles
400  *                         |exportinfo |-->v->a->b->c
401  * connect_point->+---+--->+-----------+
402  *                | / |T0
403  *                +---+
404  *                  |                            NEW treenode chain:
405  *         child->+---+
406  *                | v |T1                          +---+<-curr
407  *                +---+                          N1| v |
408  *                  |                              +---+
409  *                +---+                              |
410  *                | a |T2                          +---+<-tree_head
411  *                +---+                          N2| a1|
412  *                  |                              +---+
413  *                +---+                              |
414  *                | b |T3                          +---+
415  *                +---+                          N3| b |
416  *                  |                              +---+
417  *                +---+                              |
418  *                | c |T4                          +---+
419  *                +---+                          N4| c1|
420  *                                                 +---+
421  *
422  * The picture above illustrates the position of following pointers after line
423  * 'child = tree_find_child_by_vis(connect_point, curr->tree_vis);'
424  * was executed for the first time in the outer 'for' loop:
425  *
426  * connect_point..parent treenode in the EXISTING namespace to which the 'curr'
427  *                should be connected. If 'connect_point' already has a child
428  *                with the same value of tree_vis as the curr->tree_vis is,
429  *                the 'curr' will not be added, but kmem_free()d.
430  * child..........the result of tree_find_child_by_vis()
431  * curr...........currently processed treenode from the NEW treenode chain
432  * tree_head......current head of the NEW treenode chain, in this case it was
433  *                already moved down to its child - preparation for another loop
434  *
435  * What will happen to NEW treenodes N1, N2, N3, N4 in more_visible() later:
436  *
437  * N1: is merged - i.e. N1 is kmem_free()d. T0 has a child T1 with the same
438  *     tree_vis as N1
439  * N2: is added as a new child of T1
440  *     Note: not just N2, but the whole chain N2->N3->N4 is added
441  * N3: not processed separately (it was added together with N2)
442  *     Even that N3 and T3 have same tree_vis, they are NOT merged, but will
443  *     become duplicates.
444  * N4: not processed separately
445  */
446 static void
more_visible(struct exportinfo * exi,treenode_t * tree_head)447 more_visible(struct exportinfo *exi, treenode_t *tree_head)
448 {
449 	struct exp_visible *vp1, *vp2, *vis_head, *tail, *next;
450 	int found;
451 	treenode_t *child, *curr, *connect_point;
452 	nfs_export_t *ne = nfs_get_export();
453 
454 	vis_head = tree_head->tree_vis;
455 	connect_point = exi->exi_tree;
456 
457 	/*
458 	 * If exportinfo doesn't already have a visible
459 	 * list just assign the entire supplied list.
460 	 */
461 	if (exi->exi_visible == NULL) {
462 		tree_add_child(connect_point, tree_head);
463 		exi->exi_visible = vis_head;
464 
465 		/* Update the change timestamp */
466 		tree_update_change(ne, connect_point, &vis_head->vis_change);
467 
468 		return;
469 	}
470 
471 	/* The outer loop traverses the supplied list. */
472 	for (vp1 = vis_head; vp1; vp1 = next) {
473 		found = 0;
474 		next = vp1->vis_next;
475 
476 		/* The inner loop searches the exportinfo visible list. */
477 		for (vp2 = exi->exi_visible; vp2; vp2 = vp2->vis_next) {
478 			tail = vp2;
479 			if (EQFID(&vp1->vis_fid, &vp2->vis_fid)) {
480 				found = 1;
481 				vp2->vis_count++;
482 				VN_RELE(vp1->vis_vp);
483 				/* Transfer vis_exported from vp1 to vp2. */
484 				if (vp1->vis_exported && !vp2->vis_exported)
485 					vp2->vis_exported = 1;
486 				kmem_free(vp1, sizeof (*vp1));
487 				tree_head->tree_vis = vp2;
488 				break;
489 			}
490 		}
491 
492 		/* If not found - add to the end of the list */
493 		if (! found) {
494 			tail->vis_next = vp1;
495 			vp1->vis_next = NULL;
496 		}
497 
498 		curr = tree_head;
499 		tree_head = tree_head->tree_child_first;
500 
501 		if (! connect_point) /* No longer merging */
502 			continue;
503 		/*
504 		 * The inner loop could set curr->tree_vis to the EXISTING
505 		 * exp_visible vp2, so we can search among the children of
506 		 * connect_point for the curr->tree_vis. No need for EQFID.
507 		 */
508 		child = tree_find_child_by_vis(connect_point, curr->tree_vis);
509 
510 		/*
511 		 * Merging cannot be done if a valid child->tree_exi would
512 		 * be overwritten by a new curr->tree_exi.
513 		 */
514 		if (child &&
515 		    (child->tree_exi == NULL || curr->tree_exi == NULL)) {
516 			if (curr->tree_exi) { /* Transfer the exportinfo */
517 				child->tree_exi = curr->tree_exi;
518 				child->tree_exi->exi_tree = child;
519 			}
520 			kmem_free(curr, sizeof (treenode_t));
521 			connect_point = child;
522 		} else { /* Branching */
523 			tree_add_child(connect_point, curr);
524 
525 			/* Update the change timestamp */
526 			tree_update_change(ne, connect_point,
527 			    &curr->tree_vis->vis_change);
528 
529 			connect_point = NULL;
530 		}
531 	}
532 }
533 
534 /*
535  * Remove one visible entry from the pseudo exportfs.
536  *
537  * When we unexport a directory, we have to remove path
538  * components from the visible list in the pseudo exportfs
539  * entry. The supplied visible contains one fid of one path
540  * component. The visible list of the export
541  * is checked against provided visible, matching fid has its
542  * reference count decremented.  If a reference count drops to
543  * zero, then it means no paths now use this directory, so its
544  * fid can be removed from the visible list.
545  *
546  * When the last path is removed, the visible list will be null.
547  */
548 static void
less_visible(struct exportinfo * exi,struct exp_visible * vp1)549 less_visible(struct exportinfo *exi, struct exp_visible *vp1)
550 {
551 	struct exp_visible *vp2;
552 	struct exp_visible *prev, *next;
553 
554 	for (vp2 = exi->exi_visible, prev = NULL; vp2; vp2 = next) {
555 
556 		next = vp2->vis_next;
557 
558 		if (vp1 == vp2) {
559 			/*
560 			 * Decrement the ref count.
561 			 * Remove the entry if it's zero.
562 			 */
563 			if (--vp2->vis_count <= 0) {
564 				if (prev == NULL)
565 					exi->exi_visible = next;
566 				else
567 					prev->vis_next = next;
568 				VN_RELE(vp2->vis_vp);
569 				srv_secinfo_list_free(vp2->vis_secinfo,
570 				    vp2->vis_seccnt);
571 				kmem_free(vp2, sizeof (*vp1));
572 			}
573 			break;
574 		}
575 		prev = vp2;
576 	}
577 }
578 
579 /*
580  * This function checks the path to a new export to
581  * check whether all the pathname components are
582  * exported. It works by climbing the file tree one
583  * component at a time via "..", crossing mountpoints
584  * if necessary until an export entry is found, or the
585  * system root is reached.
586  *
587  * If an unexported mountpoint is found, then
588  * a new pseudo export is added and the pathname from
589  * the mountpoint down to the export is added to the
590  * visible list for the new pseudo export.  If an existing
591  * pseudo export is found, then the pathname is added
592  * to its visible list.
593  *
594  * Note that there's some tests for exportdir.
595  * The exportinfo entry that's passed as a parameter
596  * is that of the real export and exportdir is set
597  * for this case.
598  *
599  * Here is an example of a possible setup:
600  *
601  * () - a new fs; fs mount point
602  * EXPORT - a real exported node
603  * PSEUDO - a pseudo node
604  * vis - visible list
605  * f# - security flavor#
606  * (f#) - security flavor# propagated from its descendents
607  * "" - covered vnode
608  *
609  *
610  *                 /
611  *                 |
612  *                 (a) PSEUDO (f1,f2)
613  *                 |   vis: b,b,"c","n"
614  *                 |
615  *                 b
616  *        ---------|------------------
617  *        |                          |
618  *        (c) EXPORT,f1(f2)          (n) PSEUDO (f1,f2)
619  *        |   vis: "e","d"           |   vis: m,m,,p,q,"o"
620  *        |                          |
621  *  ------------------          -------------------
622  *  |        |        |         |                  |
623  *  (d)      (e)      f         m EXPORT,f1(f2)    p
624  *  EXPORT   EXPORT             |                  |
625  *  f1       f2                 |                  |
626  *           |                  |                  |
627  *           j                 (o) EXPORT,f2       q EXPORT f2
628  *
629  */
630 int
treeclimb_export(struct exportinfo * exip)631 treeclimb_export(struct exportinfo *exip)
632 {
633 	vnode_t *dvp, *vp;
634 	fid_t fid;
635 	int error;
636 	int exportdir;
637 	struct exportinfo *new_exi = exip;
638 	struct exp_visible *visp;
639 	struct exp_visible *vis_head = NULL;
640 	struct vattr va;
641 	treenode_t *tree_head = NULL;
642 	timespec_t now;
643 	nfs_export_t *ne;
644 
645 	ne = exip->exi_ne;
646 	ASSERT3P(ne, ==, nfs_get_export());	/* curzone reality check */
647 	ASSERT(RW_WRITE_HELD(&ne->exported_lock));
648 
649 	gethrestime(&now);
650 
651 	vp = exip->exi_vp;
652 	VN_HOLD(vp);
653 	exportdir = 1;
654 
655 	for (;;) {
656 
657 		bzero(&fid, sizeof (fid));
658 		fid.fid_len = MAXFIDSZ;
659 		error = vop_fid_pseudo(vp, &fid);
660 		if (error)
661 			break;
662 
663 		ASSERT3U(exip->exi_zoneid, ==, curzone->zone_id);
664 		/*
665 		 * The root of the file system, or the zone's root for
666 		 * in-zone NFS service needs special handling
667 		 */
668 		if (vp->v_flag & VROOT || vp == EXI_TO_ZONEROOTVP(exip)) {
669 			if (!exportdir) {
670 				struct exportinfo *exi;
671 
672 				/*
673 				 * Check if this VROOT dir is already exported.
674 				 * If so, then attach the pseudonodes.  If not,
675 				 * then continue .. traversal until we hit a
676 				 * VROOT export (pseudo or real).
677 				 */
678 				exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid,
679 				    vp);
680 				if (exi != NULL) {
681 					/*
682 					 * Found an export info
683 					 *
684 					 * Extend the list of visible
685 					 * directories whether it's a pseudo
686 					 * or a real export.
687 					 */
688 					more_visible(exi, tree_head);
689 					break;	/* and climb no further */
690 				}
691 
692 				/*
693 				 * Found the root directory of a filesystem
694 				 * that isn't exported.  Need to export
695 				 * this as a pseudo export so that an NFS v4
696 				 * client can do lookups in it.
697 				 */
698 				new_exi = pseudo_exportfs(ne, vp, &fid,
699 				    vis_head, NULL);
700 				vis_head = NULL;
701 			}
702 
703 			if (VN_IS_CURZONEROOT(vp)) {
704 				/* at system root */
705 				/*
706 				 * If sharing "/", new_exi is shared exportinfo
707 				 * (exip). Otherwise, new_exi is exportinfo
708 				 * created by pseudo_exportfs() above.
709 				 */
710 				ne->ns_root = tree_prepend_node(tree_head, NULL,
711 				    new_exi);
712 
713 				/* Update the change timestamp */
714 				tree_update_change(ne, ne->ns_root, &now);
715 
716 				break;
717 			}
718 
719 			/*
720 			 * Traverse across the mountpoint and continue the
721 			 * climb on the mounted-on filesystem.
722 			 */
723 			vp = untraverse(vp, ne->exi_root->exi_vp);
724 			exportdir = 0;
725 			continue;
726 		}
727 
728 		/*
729 		 * Do a getattr to obtain the nodeid (inode num)
730 		 * for this vnode.
731 		 */
732 		va.va_mask = AT_NODEID;
733 		error = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
734 		if (error)
735 			break;
736 
737 		/*
738 		 *  Add this directory fid to visible list
739 		 */
740 		visp = kmem_alloc(sizeof (*visp), KM_SLEEP);
741 		VN_HOLD(vp);
742 		visp->vis_vp = vp;
743 		visp->vis_fid = fid;		/* structure copy */
744 		visp->vis_ino = va.va_nodeid;
745 		visp->vis_count = 1;
746 		visp->vis_exported = exportdir;
747 		visp->vis_secinfo = NULL;
748 		visp->vis_seccnt = 0;
749 		visp->vis_change = now;		/* structure copy */
750 		visp->vis_next = vis_head;
751 		vis_head = visp;
752 
753 		/*
754 		 * Will set treenode's pointer to exportinfo to
755 		 * 1. shared exportinfo (exip) - if first visit here
756 		 * 2. freshly allocated pseudo export (if any)
757 		 * 3. null otherwise
758 		 */
759 		tree_head = tree_prepend_node(tree_head, visp, new_exi);
760 		new_exi = NULL;
761 
762 		/*
763 		 * Now, do a ".." to find parent dir of vp.
764 		 */
765 		error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, CRED(),
766 		    NULL, NULL, NULL);
767 
768 		if (error == ENOTDIR && exportdir) {
769 			dvp = exip->exi_dvp;
770 			ASSERT(dvp != NULL);
771 			VN_HOLD(dvp);
772 			error = 0;
773 		}
774 
775 		if (error)
776 			break;
777 
778 		exportdir = 0;
779 		VN_RELE(vp);
780 		vp = dvp;
781 	}
782 
783 	VN_RELE(vp);
784 
785 	/*
786 	 * We can have set error due to error in:
787 	 * 1. vop_fid_pseudo()
788 	 * 2. VOP_GETATTR()
789 	 * 3. VOP_LOOKUP()
790 	 * We must free pseudo exportinfos, visibles and treenodes.
791 	 * Visibles are referenced from treenode_t::tree_vis and
792 	 * exportinfo_t::exi_visible. To avoid double freeing, only
793 	 * exi_visible pointer is used, via exi_rele(), for the clean-up.
794 	 */
795 	if (error) {
796 		/* Free unconnected visibles, if there are any. */
797 		if (vis_head)
798 			free_visible(vis_head);
799 
800 		/* Connect unconnected exportinfo, if there is any. */
801 		if (new_exi && new_exi != exip)
802 			tree_head = tree_prepend_node(tree_head, NULL, new_exi);
803 
804 		while (tree_head) {
805 			treenode_t *t2 = tree_head;
806 			exportinfo_t *e  = tree_head->tree_exi;
807 			/* exip will be freed in exportfs() */
808 			if (e && e != exip) {
809 				mutex_enter(&nfs_exi_id_lock);
810 				avl_remove(&exi_id_tree, e);
811 				mutex_exit(&nfs_exi_id_lock);
812 				export_unlink(ne, e);
813 				exi_rele(e);
814 			}
815 			tree_head = tree_head->tree_child_first;
816 			kmem_free(t2, sizeof (*t2));
817 		}
818 	}
819 
820 	return (error);
821 }
822 
823 /*
824  * Walk up the tree and:
825  * 1. release pseudo exportinfo if it has no child
826  * 2. release visible in parent's exportinfo
827  * 3. delete non-exported leaf nodes from tree
828  *
829  * Deleting of nodes will start only if the unshared
830  * node was a leaf node.
831  * Deleting of nodes will finish when we reach a node which
832  * has children or is a real export, then we might still need
833  * to continue releasing visibles, until we reach VROOT or zone's root node.
834  */
835 void
treeclimb_unexport(nfs_export_t * ne,struct exportinfo * exip)836 treeclimb_unexport(nfs_export_t *ne, struct exportinfo *exip)
837 {
838 	treenode_t *tnode, *old_nd;
839 	treenode_t *connect_point = NULL;
840 
841 	ASSERT(RW_WRITE_HELD(&ne->exported_lock));
842 	ASSERT(curzone->zone_id == exip->exi_zoneid ||
843 	    curzone->zone_id == global_zone->zone_id);
844 
845 	/*
846 	 * exi_tree can be null for the zone root
847 	 * which means we're already at the "top"
848 	 * and there's nothing more to "climb".
849 	 */
850 	tnode = exip->exi_tree;
851 	if (tnode == NULL) {
852 		/* Should only happen for... */
853 		ASSERT(exip == ne->exi_root);
854 		return;
855 	}
856 
857 	/*
858 	 * The unshared exportinfo was unlinked in unexport().
859 	 * Zeroing tree_exi ensures that we will skip it.
860 	 */
861 	tnode->tree_exi = NULL;
862 
863 	if (tnode->tree_vis != NULL) /* system root has tree_vis == NULL */
864 		tnode->tree_vis->vis_exported = 0;
865 
866 	while (tnode != NULL) {
867 
868 		/*
869 		 * Stop at VROOT (or zone root) node which is exported or has
870 		 * child.
871 		 */
872 		if (TREE_ROOT(tnode) &&
873 		    (TREE_EXPORTED(tnode) || tnode->tree_child_first != NULL))
874 			break;
875 
876 		/* Release pseudo export if it has no child */
877 		if (TREE_ROOT(tnode) && !TREE_EXPORTED(tnode) &&
878 		    tnode->tree_child_first == NULL) {
879 			mutex_enter(&nfs_exi_id_lock);
880 			avl_remove(&exi_id_tree, tnode->tree_exi);
881 			mutex_exit(&nfs_exi_id_lock);
882 			export_unlink(ne, tnode->tree_exi);
883 			exi_rele(tnode->tree_exi);
884 			tnode->tree_exi = NULL;
885 		}
886 
887 		/* Release visible in parent's exportinfo */
888 		if (tnode->tree_vis != NULL)
889 			less_visible(vis2exi(tnode), tnode->tree_vis);
890 
891 		/* Continue with parent */
892 		old_nd = tnode;
893 		tnode = tnode->tree_parent;
894 
895 		/* Remove itself, if this is a leaf and non-exported node */
896 		if (old_nd->tree_child_first == NULL &&
897 		    !TREE_EXPORTED(old_nd)) {
898 			tree_remove_node(ne, old_nd);
899 			connect_point = tnode;
900 		}
901 	}
902 
903 	/* Update the change timestamp */
904 	if (connect_point != NULL)
905 		tree_update_change(ne, connect_point, NULL);
906 }
907 
908 /*
909  * Traverse backward across mountpoint from the
910  * root vnode of a filesystem to its mounted-on
911  * vnode.
912  */
913 vnode_t *
untraverse(vnode_t * vp,vnode_t * zone_rootvp)914 untraverse(vnode_t *vp, vnode_t *zone_rootvp)
915 {
916 	vnode_t *tvp, *nextvp;
917 
918 	tvp = vp;
919 	for (;;) {
920 		if (!(tvp->v_flag & VROOT) && !VN_CMP(tvp, zone_rootvp))
921 			break;
922 
923 		/* lock vfs to prevent unmount of this vfs */
924 		vfs_lock_wait(tvp->v_vfsp);
925 
926 		if ((nextvp = tvp->v_vfsp->vfs_vnodecovered) == NULL) {
927 			vfs_unlock(tvp->v_vfsp);
928 			break;
929 		}
930 
931 		/*
932 		 * Hold nextvp to prevent unmount.  After unlock vfs and
933 		 * rele tvp, any number of overlays could be unmounted.
934 		 * Putting a hold on vfs_vnodecovered will only allow
935 		 * tvp's vfs to be unmounted. Of course if caller placed
936 		 * extra hold on vp before calling untraverse, the following
937 		 * hold would not be needed.  Since prev actions of caller
938 		 * are unknown, we need to hold here just to be safe.
939 		 */
940 		VN_HOLD(nextvp);
941 		vfs_unlock(tvp->v_vfsp);
942 		VN_RELE(tvp);
943 		tvp = nextvp;
944 	}
945 
946 	return (tvp);
947 }
948 
949 /*
950  * Given an exportinfo, climb up to find the exportinfo for the VROOT
951  * (or zone root) of the filesystem.
952  *
953  * e.g.         /
954  *              |
955  *              a (VROOT) pseudo-exportinfo
956  *		|
957  *		b
958  *		|
959  *		c  #share /a/b/c
960  *		|
961  *		d
962  *
963  * where c is in the same filesystem as a.
964  * So, get_root_export(*exportinfo_for_c) returns exportinfo_for_a
965  *
966  * If d is shared, then c will be put into a's visible list.
967  * Note: visible list is per filesystem and is attached to the
968  * VROOT exportinfo.  Returned exi does NOT have a new hold.
969  */
970 struct exportinfo *
get_root_export(struct exportinfo * exip)971 get_root_export(struct exportinfo *exip)
972 {
973 	treenode_t *tnode = exip->exi_tree;
974 	exportinfo_t *exi = NULL;
975 
976 	while (tnode) {
977 		if (TREE_ROOT(tnode)) {
978 			exi = tnode->tree_exi;
979 			break;
980 		}
981 		tnode = tnode->tree_parent;
982 	}
983 	ASSERT(exi);
984 	return (exi);
985 }
986 
987 /*
988  * Return true if the supplied vnode has a sub-directory exported.
989  */
990 int
has_visible(struct exportinfo * exi,vnode_t * vp)991 has_visible(struct exportinfo *exi, vnode_t *vp)
992 {
993 	struct exp_visible *visp;
994 	fid_t fid;
995 	bool_t vp_is_exported;
996 
997 	vp_is_exported = VN_CMP(vp, exi->exi_vp);
998 
999 	/*
1000 	 * An exported root vnode has a sub-dir shared if it has a visible
1001 	 * list.  i.e. if it does not have a visible list, then there is no
1002 	 * node in this filesystem leads to any other shared node.
1003 	 */
1004 	ASSERT3P(curzone->zone_id, ==, exi->exi_zoneid);
1005 	if (vp_is_exported &&
1006 	    ((vp->v_flag & VROOT) || VN_IS_CURZONEROOT(vp))) {
1007 		return (exi->exi_visible ? 1 : 0);
1008 	}
1009 
1010 	/*
1011 	 * Only the exportinfo of a fs root node may have a visible list.
1012 	 * Either it is a pseudo root node, or a real exported root node.
1013 	 */
1014 	exi = get_root_export(exi);
1015 
1016 	if (!exi->exi_visible)
1017 		return (0);
1018 
1019 	/* Get the fid of the vnode */
1020 	bzero(&fid, sizeof (fid));
1021 	fid.fid_len = MAXFIDSZ;
1022 	if (vop_fid_pseudo(vp, &fid) != 0) {
1023 		return (0);
1024 	}
1025 
1026 	/*
1027 	 * See if vp is in the visible list of the root node exportinfo.
1028 	 */
1029 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
1030 		if (EQFID(&fid, &visp->vis_fid)) {
1031 			/*
1032 			 * If vp is an exported non-root node with only 1 path
1033 			 * count (for itself), it indicates no sub-dir shared
1034 			 * using this vp as a path.
1035 			 */
1036 			if (vp_is_exported && visp->vis_count < 2)
1037 				break;
1038 
1039 			return (1);
1040 		}
1041 	}
1042 
1043 	return (0);
1044 }
1045 
1046 /*
1047  * Returns true if the supplied vnode is visible
1048  * in this export.  If vnode is visible, return
1049  * vis_exported in expseudo.
1050  */
1051 int
nfs_visible(struct exportinfo * exi,vnode_t * vp,int * expseudo)1052 nfs_visible(struct exportinfo *exi, vnode_t *vp, int *expseudo)
1053 {
1054 	struct exp_visible *visp;
1055 	fid_t fid;
1056 
1057 	/*
1058 	 * First check to see if vp is export root.
1059 	 *
1060 	 * A pseudo export root can never be exported
1061 	 * (it would be a real export then); however,
1062 	 * it is always visible.  If a pseudo root object
1063 	 * was exported by server admin, then the entire
1064 	 * pseudo exportinfo (and all visible entries) would
1065 	 * be destroyed.  A pseudo exportinfo only exists
1066 	 * to provide access to real (descendant) export(s).
1067 	 *
1068 	 * Previously, rootdir was special cased here; however,
1069 	 * the export root special case handles the rootdir
1070 	 * case also.
1071 	 */
1072 	if (VN_CMP(vp, exi->exi_vp)) {
1073 		*expseudo = 0;
1074 		return (1);
1075 	}
1076 
1077 	/*
1078 	 * Only a PSEUDO node has a visible list or an exported VROOT
1079 	 * node may have a visible list.
1080 	 */
1081 	if (!PSEUDO(exi))
1082 		exi = get_root_export(exi);
1083 
1084 	/* Get the fid of the vnode */
1085 
1086 	bzero(&fid, sizeof (fid));
1087 	fid.fid_len = MAXFIDSZ;
1088 	if (vop_fid_pseudo(vp, &fid) != 0) {
1089 		*expseudo = 0;
1090 		return (0);
1091 	}
1092 
1093 	/*
1094 	 * We can't trust VN_CMP() above because of LOFS.
1095 	 * Even though VOP_CMP will do the right thing for LOFS
1096 	 * objects, VN_CMP will short circuit out early when the
1097 	 * vnode ops ptrs are different.  Just in case we're dealing
1098 	 * with LOFS, compare exi_fid/fsid here.
1099 	 *
1100 	 * expseudo is not set because this is not an export
1101 	 */
1102 	if (EQFID(&exi->exi_fid, &fid) &&
1103 	    EQFSID(&exi->exi_fsid, &vp->v_vfsp->vfs_fsid)) {
1104 		*expseudo = 0;
1105 		return (1);
1106 	}
1107 
1108 
1109 	/* See if it matches any fid in the visible list */
1110 
1111 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
1112 		if (EQFID(&fid, &visp->vis_fid)) {
1113 			*expseudo = visp->vis_exported;
1114 			return (1);
1115 		}
1116 	}
1117 
1118 	*expseudo = 0;
1119 
1120 	return (0);
1121 }
1122 
1123 /*
1124  * Returns true if the supplied vnode is the
1125  * directory of an export point.
1126  */
1127 int
nfs_exported(struct exportinfo * exi,vnode_t * vp)1128 nfs_exported(struct exportinfo *exi, vnode_t *vp)
1129 {
1130 	struct exp_visible *visp;
1131 	fid_t fid;
1132 
1133 	/*
1134 	 * First check to see if vp is the export root
1135 	 * This check required for the case of lookup ..
1136 	 * where .. is a V_ROOT vnode and a pseudo exportroot.
1137 	 * Pseudo export root objects do not have an entry
1138 	 * in the visible list even though every V_ROOT
1139 	 * pseudonode is visible.  It is safe to compare
1140 	 * vp here because pseudo_exportfs put a hold on
1141 	 * it when exi_vp was initialized.
1142 	 *
1143 	 * Note: VN_CMP() won't match for LOFS shares, but they're
1144 	 * handled below w/EQFID/EQFSID.
1145 	 */
1146 	if (VN_CMP(vp, exi->exi_vp))
1147 		return (1);
1148 
1149 	/* Get the fid of the vnode */
1150 
1151 	bzero(&fid, sizeof (fid));
1152 	fid.fid_len = MAXFIDSZ;
1153 	if (vop_fid_pseudo(vp, &fid) != 0)
1154 		return (0);
1155 
1156 	if (EQFID(&fid, &exi->exi_fid) &&
1157 	    EQFSID(&vp->v_vfsp->vfs_fsid, &exi->exi_fsid)) {
1158 		return (1);
1159 	}
1160 
1161 	/* See if it matches any fid in the visible list */
1162 
1163 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
1164 		if (EQFID(&fid, &visp->vis_fid))
1165 			return (visp->vis_exported);
1166 	}
1167 
1168 	return (0);
1169 }
1170 
1171 /*
1172  * Returns true if the supplied inode is visible
1173  * in this export.  This function is used by
1174  * readdir which uses inode numbers from the
1175  * directory.
1176  *
1177  * NOTE: this code does not match inode number for ".",
1178  * but it isn't required because NFS4 server rddir
1179  * skips . and .. entries.
1180  */
1181 int
nfs_visible_inode(struct exportinfo * exi,ino64_t ino,struct exp_visible ** visp)1182 nfs_visible_inode(struct exportinfo *exi, ino64_t ino,
1183     struct exp_visible **visp)
1184 {
1185 	/*
1186 	 * Only a PSEUDO node has a visible list or an exported VROOT
1187 	 * node may have a visible list.
1188 	 */
1189 	if (!PSEUDO(exi))
1190 		exi = get_root_export(exi);
1191 
1192 	for (*visp = exi->exi_visible; *visp != NULL; *visp = (*visp)->vis_next)
1193 		if ((u_longlong_t)ino == (*visp)->vis_ino) {
1194 			return (1);
1195 		}
1196 
1197 	return (0);
1198 }
1199 
1200 /*
1201  * Get the change attribute from visible and returns TRUE.
1202  * If the change value is not available returns FALSE.
1203  */
1204 bool_t
nfs_visible_change(struct exportinfo * exi,vnode_t * vp,timespec_t * change)1205 nfs_visible_change(struct exportinfo *exi, vnode_t *vp, timespec_t *change)
1206 {
1207 	struct exp_visible *visp;
1208 	fid_t fid;
1209 	treenode_t *node;
1210 	nfs_export_t *ne = nfs_get_export();
1211 
1212 	/*
1213 	 * First check to see if vp is export root.
1214 	 */
1215 	if (VN_CMP(vp, exi->exi_vp))
1216 		goto exproot;
1217 
1218 	/*
1219 	 * Only a PSEUDO node has a visible list or an exported VROOT
1220 	 * node may have a visible list.
1221 	 */
1222 	if (!PSEUDO(exi))
1223 		exi = get_root_export(exi);
1224 
1225 	/* Get the fid of the vnode */
1226 	bzero(&fid, sizeof (fid));
1227 	fid.fid_len = MAXFIDSZ;
1228 	if (vop_fid_pseudo(vp, &fid) != 0)
1229 		return (FALSE);
1230 
1231 	/*
1232 	 * We can't trust VN_CMP() above because of LOFS.
1233 	 * Even though VOP_CMP will do the right thing for LOFS
1234 	 * objects, VN_CMP will short circuit out early when the
1235 	 * vnode ops ptrs are different.  Just in case we're dealing
1236 	 * with LOFS, compare exi_fid/fsid here.
1237 	 */
1238 	if (EQFID(&exi->exi_fid, &fid) &&
1239 	    EQFSID(&exi->exi_fsid, &vp->v_vfsp->vfs_fsid))
1240 		goto exproot;
1241 
1242 	/* See if it matches any fid in the visible list */
1243 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
1244 		if (EQFID(&fid, &visp->vis_fid)) {
1245 			*change = visp->vis_change;
1246 			return (TRUE);
1247 		}
1248 	}
1249 
1250 	return (FALSE);
1251 
1252 exproot:
1253 	/* The VROOT export have its visible available through treenode */
1254 	node = exi->exi_tree;
1255 	if (node != ne->ns_root) {
1256 		ASSERT(node->tree_vis != NULL);
1257 		*change = node->tree_vis->vis_change;
1258 	} else {
1259 		ASSERT(node->tree_vis == NULL);
1260 		*change = ne->ns_root_change;
1261 	}
1262 	return (TRUE);
1263 }
1264 
1265 /*
1266  * Update the change attribute value for a particular treenode.  The change
1267  * attribute value is stored in the visible attached to the treenode, or in the
1268  * ns_root_change.
1269  *
1270  * If the change value is not supplied, the current time is used.
1271  */
1272 void
tree_update_change(nfs_export_t * ne,treenode_t * tnode,timespec_t * change)1273 tree_update_change(nfs_export_t *ne, treenode_t *tnode, timespec_t *change)
1274 {
1275 	timespec_t *vis_change;
1276 
1277 	ASSERT(tnode != NULL);
1278 	ASSERT((tnode != ne->ns_root && tnode->tree_vis != NULL) ||
1279 	    (tnode == ne->ns_root && tnode->tree_vis == NULL));
1280 
1281 	vis_change = tnode == ne->ns_root ? &ne->ns_root_change
1282 	    : &tnode->tree_vis->vis_change;
1283 
1284 	if (change != NULL)
1285 		*vis_change = *change;
1286 	else
1287 		gethrestime(vis_change);
1288 }
1289