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