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 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26 #include <sys/types.h>
27 #include <sys/stat.h>
28 #include <sys/errno.h>
29 #include <sys/kmem.h>
30 #include <sys/t_lock.h>
31 #include <sys/ksynch.h>
32 #include <sys/buf.h>
33 #include <sys/vfs.h>
34 #include <sys/vnode.h>
35 #include <sys/mode.h>
36 #include <sys/systm.h>
37 #include <vm/seg.h>
38 #include <sys/file.h>
39 #include <sys/acl.h>
40 #include <sys/fs/ufs_inode.h>
41 #include <sys/fs/ufs_acl.h>
42 #include <sys/fs/ufs_quota.h>
43 #include <sys/sysmacros.h>
44 #include <sys/debug.h>
45 #include <sys/policy.h>
46
47 /* Cache routines */
48 static int si_signature(si_t *);
49 static int si_cachei_get(struct inode *, si_t **);
50 static int si_cachea_get(struct inode *, si_t *, si_t **);
51 static int si_cmp(si_t *, si_t *);
52 static void si_cache_put(si_t *);
53 void si_cache_del(si_t *, int);
54 void si_cache_init(void);
55
56 static void ufs_si_free_mem(si_t *);
57 static int ufs_si_store(struct inode *, si_t *, int, cred_t *);
58 static si_t *ufs_acl_cp(si_t *);
59 static int ufs_sectobuf(si_t *, caddr_t *, size_t *);
60 static int acl_count(ufs_ic_acl_t *);
61 static int acl_validate(aclent_t *, int, int);
62 static int vsecattr2aclentry(vsecattr_t *, si_t **);
63 static int aclentry2vsecattr(si_t *, vsecattr_t *);
64
65 krwlock_t si_cache_lock; /* Protects si_cache */
66 int si_cachecnt = 64; /* # buckets in si_cache[a|i] */
67 si_t **si_cachea; /* The 'by acl' cache chains */
68 si_t **si_cachei; /* The 'by inode' cache chains */
69 long si_cachehit = 0;
70 long si_cachemiss = 0;
71
72 #define SI_HASH(S) ((int)(S) & (si_cachecnt - 1))
73
74 /*
75 * Store the new acls in aclp. Attempts to make things atomic.
76 * Search the acl cache for an identical sp and, if found, attach
77 * the cache'd acl to ip. If the acl is new (not in the cache),
78 * add it to the cache, then attach it to ip. Last, remove and
79 * decrement the reference count of any prior acl list attached
80 * to the ip.
81 *
82 * Parameters:
83 * ip - Ptr to inode to receive the acl list
84 * sp - Ptr to in-core acl structure to attach to the inode.
85 * puship - 0 do not push the object inode(ip) 1 push the ip
86 * cr - Ptr to credentials
87 *
88 * Returns: 0 - Success
89 * N - From errno.h
90 */
91 static int
ufs_si_store(struct inode * ip,si_t * sp,int puship,cred_t * cr)92 ufs_si_store(struct inode *ip, si_t *sp, int puship, cred_t *cr)
93 {
94 struct vfs *vfsp;
95 struct inode *sip;
96 si_t *oldsp;
97 si_t *csp;
98 caddr_t acldata;
99 ino_t oldshadow;
100 size_t acldatalen;
101 off_t offset;
102 int shadow;
103 int err;
104 int refcnt;
105 int usecnt;
106 int signature;
107 int resid;
108 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
109 struct fs *fs = ufsvfsp->vfs_fs;
110
111 ASSERT(RW_WRITE_HELD(&ip->i_contents));
112 ASSERT(ip->i_ufs_acl != sp);
113
114 if (!CHECK_ACL_ALLOWED(ip->i_mode & IFMT))
115 return (ENOSYS);
116
117 /*
118 * if there are only the three owner/group/other then do not
119 * create a shadow inode. If there is already a shadow with
120 * the file, remove it.
121 *
122 */
123 if (!sp->ausers &&
124 !sp->agroups &&
125 !sp->downer &&
126 !sp->dgroup &&
127 !sp->dother &&
128 sp->dclass.acl_ismask == 0 &&
129 !sp->dusers &&
130 !sp->dgroups) {
131 if (ip->i_ufs_acl)
132 err = ufs_si_free(ip->i_ufs_acl, ITOV(ip)->v_vfsp, cr);
133 ip->i_ufs_acl = NULL;
134 ip->i_shadow = 0;
135 ip->i_flag |= IMOD | IACC;
136 ip->i_mode = (ip->i_smode & ~0777) |
137 ((sp->aowner->acl_ic_perm & 07) << 6) |
138 (MASK2MODE(sp)) |
139 (sp->aother->acl_ic_perm & 07);
140 TRANS_INODE(ip->i_ufsvfs, ip);
141 ufs_iupdat(ip, 1);
142 ufs_si_free_mem(sp);
143 return (0);
144 }
145
146 loop:
147
148 /*
149 * Check cache. If in cache, use existing shadow inode.
150 * Increment the shadow link count, then attach to the
151 * cached ufs_acl_entry struct, and increment it's reference
152 * count. Then discard the passed-in ufs_acl_entry and
153 * return.
154 */
155 if (si_cachea_get(ip, sp, &csp) == 0) {
156 ASSERT(RW_WRITE_HELD(&csp->s_lock));
157 if (ip->i_ufs_acl == csp) {
158 rw_exit(&csp->s_lock);
159 (void) ufs_si_free_mem(sp);
160 return (0);
161 }
162 vfsp = ITOV(ip)->v_vfsp;
163 ASSERT(csp->s_shadow <= INT_MAX);
164 shadow = (int)csp->s_shadow;
165 /*
166 * We can't call ufs_iget while holding the csp locked,
167 * because we might deadlock. So we drop the
168 * lock on csp, then go search the si_cache again
169 * to see if the csp is still there.
170 */
171 rw_exit(&csp->s_lock);
172 if ((err = ufs_iget(vfsp, shadow, &sip, cr)) != 0) {
173 (void) ufs_si_free_mem(sp);
174 return (EIO);
175 }
176 rw_enter(&sip->i_contents, RW_WRITER);
177 if ((sip->i_mode & IFMT) != IFSHAD || sip->i_nlink <= 0) {
178 rw_exit(&sip->i_contents);
179 VN_RELE(ITOV(sip));
180 goto loop;
181 }
182 /* Get the csp again */
183 if (si_cachea_get(ip, sp, &csp) != 0) {
184 rw_exit(&sip->i_contents);
185 VN_RELE(ITOV(sip));
186 goto loop;
187 }
188 ASSERT(RW_WRITE_HELD(&csp->s_lock));
189 /* See if we got the right shadow */
190 if (csp->s_shadow != shadow) {
191 rw_exit(&csp->s_lock);
192 rw_exit(&sip->i_contents);
193 VN_RELE(ITOV(sip));
194 goto loop;
195 }
196 ASSERT(RW_WRITE_HELD(&sip->i_contents));
197 ASSERT(sip->i_dquot == 0);
198 /* Increment link count */
199 ASSERT(sip->i_nlink > 0);
200 sip->i_nlink++;
201 TRANS_INODE(ufsvfsp, sip);
202 csp->s_use = sip->i_nlink;
203 csp->s_ref++;
204 ASSERT(sp->s_ref >= 0 && sp->s_ref <= sp->s_use);
205 sip->i_flag |= ICHG | IMOD;
206 sip->i_seq++;
207 ITIMES_NOLOCK(sip);
208 /*
209 * Always release s_lock before both releasing i_contents
210 * and calling VN_RELE.
211 */
212 rw_exit(&csp->s_lock);
213 rw_exit(&sip->i_contents);
214 VN_RELE(ITOV(sip));
215 (void) ufs_si_free_mem(sp);
216 sp = csp;
217 si_cachehit++;
218 goto switchshadows;
219 }
220
221 /* Alloc a shadow inode and fill it in */
222 err = ufs_ialloc(ip, ip->i_number, (mode_t)IFSHAD, &sip, cr);
223 if (err) {
224 (void) ufs_si_free_mem(sp);
225 return (err);
226 }
227 rw_enter(&sip->i_contents, RW_WRITER);
228 sip->i_flag |= IACC | IUPD | ICHG;
229 sip->i_seq++;
230 sip->i_mode = (o_mode_t)IFSHAD;
231 ITOV(sip)->v_type = VREG;
232 ufs_reset_vnode(ITOV(sip));
233 sip->i_nlink = 1;
234 sip->i_uid = crgetuid(cr);
235 sip->i_suid = (ulong_t)sip->i_uid > (ulong_t)USHRT_MAX ?
236 UID_LONG : sip->i_uid;
237 sip->i_gid = crgetgid(cr);
238 sip->i_sgid = (ulong_t)sip->i_gid > (ulong_t)USHRT_MAX ?
239 GID_LONG : sip->i_gid;
240 sip->i_shadow = 0;
241 TRANS_INODE(ufsvfsp, sip);
242 sip->i_ufs_acl = NULL;
243 ASSERT(sip->i_size == 0);
244
245 sp->s_shadow = sip->i_number;
246
247 if ((err = ufs_sectobuf(sp, &acldata, &acldatalen)) != 0)
248 goto errout;
249 offset = 0;
250
251 /*
252 * We don't actually care about the residual count upon failure,
253 * but giving ufs_rdwri() the pointer means it won't translate
254 * all failures to EIO. Our caller needs to know when ENOSPC
255 * gets hit.
256 */
257 resid = 0;
258 if (((err = ufs_rdwri(UIO_WRITE, FWRITE|FSYNC, sip, acldata,
259 acldatalen, (offset_t)0, UIO_SYSSPACE, &resid, cr)) != 0) ||
260 (resid != 0)) {
261 kmem_free(acldata, acldatalen);
262 if ((resid != 0) && (err == 0))
263 err = ENOSPC;
264 goto errout;
265 }
266
267 offset += acldatalen;
268 if ((acldatalen + fs->fs_bsize) > ufsvfsp->vfs_maxacl)
269 ufsvfsp->vfs_maxacl = acldatalen + fs->fs_bsize;
270
271 kmem_free(acldata, acldatalen);
272 /* Sync & free the shadow inode */
273 ufs_iupdat(sip, 1);
274 rw_exit(&sip->i_contents);
275 VN_RELE(ITOV(sip));
276
277 /* We're committed to using this sp */
278 sp->s_use = 1;
279 sp->s_ref = 1;
280
281 /* Now put the new acl stuff in the cache */
282 /* XXX Might make a duplicate */
283 si_cache_put(sp);
284 si_cachemiss++;
285
286 switchshadows:
287 /* Now switch the parent inode to use the new shadow inode */
288 ASSERT(RW_WRITE_HELD(&ip->i_contents));
289 rw_enter(&sp->s_lock, RW_READER);
290 oldsp = ip->i_ufs_acl;
291 oldshadow = ip->i_shadow;
292 ip->i_ufs_acl = sp;
293 ASSERT(sp->s_shadow <= INT_MAX);
294 ip->i_shadow = (int32_t)sp->s_shadow;
295 ASSERT(oldsp != sp);
296 ASSERT(oldshadow != ip->i_number);
297 ASSERT(ip->i_number != ip->i_shadow);
298 /*
299 * Change the mode bits to follow the acl list
300 *
301 * NOTE: a directory is not required to have a "regular" acl
302 * bug id's 1238908, 1257173, 1263171 and 1263188
303 *
304 * but if a "regular" acl is present, it must contain
305 * an "owner", "group", and "other" acl
306 *
307 * If an ACL mask exists, the effective group rights are
308 * set to the mask. Otherwise, the effective group rights
309 * are set to the object group bits.
310 */
311 if (sp->aowner) { /* Owner */
312 ip->i_mode &= ~0700; /* clear Owner */
313 ip->i_mode |= (sp->aowner->acl_ic_perm & 07) << 6;
314 ip->i_uid = sp->aowner->acl_ic_who;
315 }
316
317 if (sp->agroup) { /* Group */
318 ip->i_mode &= ~0070; /* clear Group */
319 ip->i_mode |= MASK2MODE(sp); /* apply mask */
320 ip->i_gid = sp->agroup->acl_ic_who;
321 }
322
323 if (sp->aother) { /* Other */
324 ip->i_mode &= ~0007; /* clear Other */
325 ip->i_mode |= (sp->aother->acl_ic_perm & 07);
326 }
327
328 if (sp->aclass.acl_ismask)
329 ip->i_mode = (ip->i_mode & ~070) |
330 (((sp->aclass.acl_maskbits & 07) << 3) &
331 ip->i_mode);
332
333 TRANS_INODE(ufsvfsp, ip);
334 rw_exit(&sp->s_lock);
335 ip->i_flag |= ICHG;
336 ip->i_seq++;
337 /*
338 * when creating a file there is no need to push the inode, it
339 * is pushed later
340 */
341 if (puship == 1)
342 ufs_iupdat(ip, 1);
343
344 /*
345 * Decrement link count on the old shadow inode,
346 * and decrement reference count on the old aclp,
347 */
348 if (oldshadow) {
349 /* Get the shadow inode */
350 ASSERT(RW_WRITE_HELD(&ip->i_contents));
351 vfsp = ITOV(ip)->v_vfsp;
352 if ((err = ufs_iget_alloced(vfsp, oldshadow, &sip, cr)) != 0) {
353 return (EIO);
354 }
355 /* Decrement link count */
356 rw_enter(&sip->i_contents, RW_WRITER);
357 if (oldsp)
358 rw_enter(&oldsp->s_lock, RW_WRITER);
359 ASSERT(sip->i_dquot == 0);
360 ASSERT(sip->i_nlink > 0);
361 usecnt = --sip->i_nlink;
362 ufs_setreclaim(sip);
363 TRANS_INODE(ufsvfsp, sip);
364 sip->i_flag |= ICHG | IMOD;
365 sip->i_seq++;
366 ITIMES_NOLOCK(sip);
367 if (oldsp) {
368 oldsp->s_use = usecnt;
369 refcnt = --oldsp->s_ref;
370 signature = oldsp->s_signature;
371 /*
372 * Always release s_lock before both releasing
373 * i_contents and calling VN_RELE.
374 */
375 rw_exit(&oldsp->s_lock);
376 }
377 rw_exit(&sip->i_contents);
378 VN_RELE(ITOV(sip));
379 if (oldsp && (refcnt == 0))
380 si_cache_del(oldsp, signature);
381 }
382 return (0);
383
384 errout:
385 /* Throw the newly alloc'd inode away */
386 sip->i_nlink = 0;
387 ufs_setreclaim(sip);
388 TRANS_INODE(ufsvfsp, sip);
389 ITIMES_NOLOCK(sip);
390 rw_exit(&sip->i_contents);
391 VN_RELE(ITOV(sip));
392 ASSERT(!sp->s_use && !sp->s_ref && !(sp->s_flags & SI_CACHED));
393 (void) ufs_si_free_mem(sp);
394 return (err);
395 }
396
397 /*
398 * Load the acls for inode ip either from disk (adding to the cache),
399 * or search the cache and attach the cache'd acl list to the ip.
400 * In either case, maintain the proper reference count on the cached entry.
401 *
402 * Parameters:
403 * ip - Ptr to the inode which needs the acl list loaded
404 * cr - Ptr to credentials
405 *
406 * Returns: 0 - Success
407 * N - From errno.h
408 */
409 int
ufs_si_load(struct inode * ip,cred_t * cr)410 ufs_si_load(struct inode *ip, cred_t *cr)
411 /*
412 * ip parent inode in
413 * cr credentials in
414 */
415 {
416 struct vfs *vfsp;
417 struct inode *sip;
418 ufs_fsd_t *fsdp;
419 si_t *sp;
420 vsecattr_t vsecattr = {
421 (uint_t)0,
422 (int)0,
423 (void *)NULL,
424 (int)0,
425 (void *)NULL};
426 aclent_t *aclp;
427 ufs_acl_t *ufsaclp;
428 caddr_t acldata = NULL;
429 ino_t maxino;
430 int err;
431 size_t acldatalen;
432 int numacls;
433 int shadow;
434 int usecnt;
435 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
436 struct fs *fs = ufsvfsp->vfs_fs;
437
438 ASSERT(ip != NULL);
439 ASSERT(RW_WRITE_HELD(&ip->i_contents));
440 ASSERT(ip->i_shadow && ip->i_ufs_acl == NULL);
441 ASSERT((ip->i_mode & IFMT) != IFSHAD);
442
443 if (!CHECK_ACL_ALLOWED(ip->i_mode & IFMT))
444 return (ENOSYS);
445
446 if (ip->i_shadow == ip->i_number)
447 return (EIO);
448
449 maxino = (ino_t)(ITOF(ip)->fs_ncg * ITOF(ip)->fs_ipg);
450 if (ip->i_shadow < UFSROOTINO || ip->i_shadow > maxino)
451 return (EIO);
452
453 /*
454 * XXX Check cache. If in cache, link to it and increment
455 * the reference count, then return.
456 */
457 if (si_cachei_get(ip, &sp) == 0) {
458 ASSERT(RW_WRITE_HELD(&sp->s_lock));
459 ip->i_ufs_acl = sp;
460 sp->s_ref++;
461 ASSERT(sp->s_ref >= 0 && sp->s_ref <= sp->s_use);
462 rw_exit(&sp->s_lock);
463 si_cachehit++;
464 return (0);
465 }
466
467 /* Get the shadow inode */
468 vfsp = ITOV(ip)->v_vfsp;
469 shadow = ip->i_shadow;
470 if ((err = ufs_iget_alloced(vfsp, shadow, &sip, cr)) != 0) {
471 return (err);
472 }
473 rw_enter(&sip->i_contents, RW_WRITER);
474
475 if ((sip->i_mode & IFMT) != IFSHAD) {
476 rw_exit(&sip->i_contents);
477 err = EINVAL;
478 goto alldone;
479 }
480
481 ASSERT(sip->i_dquot == 0);
482 usecnt = sip->i_nlink;
483 if ((!ULOCKFS_IS_NOIACC(&ufsvfsp->vfs_ulockfs)) &&
484 (!(sip)->i_ufsvfs->vfs_noatime)) {
485 sip->i_flag |= IACC;
486 }
487 rw_downgrade(&sip->i_contents);
488
489 ASSERT(sip->i_size <= MAXOFF_T);
490 /* Read the acl's and other stuff from disk */
491 acldata = kmem_zalloc((size_t)sip->i_size, KM_SLEEP);
492 acldatalen = sip->i_size;
493
494 err = ufs_rdwri(UIO_READ, FREAD, sip, acldata, acldatalen, (offset_t)0,
495 UIO_SYSSPACE, (int *)0, cr);
496
497 rw_exit(&sip->i_contents);
498
499 if (err)
500 goto alldone;
501
502 /*
503 * Convert from disk format
504 * Result is a vsecattr struct which we then convert to the
505 * si struct.
506 */
507 bzero((caddr_t)&vsecattr, sizeof (vsecattr_t));
508 for (fsdp = (ufs_fsd_t *)acldata;
509 fsdp < (ufs_fsd_t *)(acldata + acldatalen);
510 fsdp = (ufs_fsd_t *)((caddr_t)fsdp +
511 FSD_RECSZ(fsdp, fsdp->fsd_size))) {
512 if (fsdp->fsd_size <= 0)
513 break;
514 switch (fsdp->fsd_type) {
515 case FSD_ACL:
516 numacls = vsecattr.vsa_aclcnt =
517 (int)((fsdp->fsd_size - 2 * sizeof (int)) /
518 sizeof (ufs_acl_t));
519 aclp = vsecattr.vsa_aclentp =
520 kmem_zalloc(numacls * sizeof (aclent_t), KM_SLEEP);
521 for (ufsaclp = (ufs_acl_t *)fsdp->fsd_data;
522 numacls; ufsaclp++) {
523 aclp->a_type = ufsaclp->acl_tag;
524 aclp->a_id = ufsaclp->acl_who;
525 aclp->a_perm = ufsaclp->acl_perm;
526 aclp++;
527 numacls--;
528 }
529 break;
530 case FSD_DFACL:
531 numacls = vsecattr.vsa_dfaclcnt =
532 (int)((fsdp->fsd_size - 2 * sizeof (int)) /
533 sizeof (ufs_acl_t));
534 aclp = vsecattr.vsa_dfaclentp =
535 kmem_zalloc(numacls * sizeof (aclent_t), KM_SLEEP);
536 for (ufsaclp = (ufs_acl_t *)fsdp->fsd_data;
537 numacls; ufsaclp++) {
538 aclp->a_type = ufsaclp->acl_tag;
539 aclp->a_id = ufsaclp->acl_who;
540 aclp->a_perm = ufsaclp->acl_perm;
541 aclp++;
542 numacls--;
543 }
544 break;
545 }
546 }
547 /* Sort the lists */
548 if (vsecattr.vsa_aclentp) {
549 ksort((caddr_t)vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt,
550 sizeof (aclent_t), cmp2acls);
551 if ((err = acl_validate(vsecattr.vsa_aclentp,
552 vsecattr.vsa_aclcnt, ACL_CHECK)) != 0) {
553 goto alldone;
554 }
555 }
556 if (vsecattr.vsa_dfaclentp) {
557 ksort((caddr_t)vsecattr.vsa_dfaclentp, vsecattr.vsa_dfaclcnt,
558 sizeof (aclent_t), cmp2acls);
559 if ((err = acl_validate(vsecattr.vsa_dfaclentp,
560 vsecattr.vsa_dfaclcnt, DEF_ACL_CHECK)) != 0) {
561 goto alldone;
562 }
563 }
564
565 /* ignore shadow inodes without ACLs */
566 if (!vsecattr.vsa_aclentp && !vsecattr.vsa_dfaclentp) {
567 err = 0;
568 goto alldone;
569 }
570
571 /* Convert from vsecattr struct to ufs_acl_entry struct */
572 if ((err = vsecattr2aclentry(&vsecattr, &sp)) != 0) {
573 goto alldone;
574 }
575
576 /* There aren't filled in by vsecattr2aclentry */
577 sp->s_shadow = ip->i_shadow;
578 sp->s_dev = ip->i_dev;
579 sp->s_use = usecnt;
580 sp->s_ref = 1;
581 ASSERT(sp->s_ref >= 0 && sp->s_ref <= sp->s_use);
582
583 /* XXX Might make a duplicate */
584 si_cache_put(sp);
585
586 /* Signal anyone waiting on this shadow to be loaded */
587 ip->i_ufs_acl = sp;
588 err = 0;
589 si_cachemiss++;
590 if ((acldatalen + fs->fs_bsize) > ufsvfsp->vfs_maxacl)
591 ufsvfsp->vfs_maxacl = acldatalen + fs->fs_bsize;
592 alldone:
593 /*
594 * Common exit point. Mark shadow inode as ISTALE
595 * if we detect an internal inconsistency, to
596 * prevent stray inodes appearing in the cache.
597 */
598 if (err) {
599 rw_enter(&sip->i_contents, RW_READER);
600 mutex_enter(&sip->i_tlock);
601 sip->i_flag |= ISTALE;
602 mutex_exit(&sip->i_tlock);
603 rw_exit(&sip->i_contents);
604 }
605 VN_RELE(ITOV(sip));
606
607 /*
608 * Cleanup of data structures allocated
609 * on the fly.
610 */
611 if (acldata)
612 kmem_free(acldata, acldatalen);
613
614 if (vsecattr.vsa_aclentp)
615 kmem_free(vsecattr.vsa_aclentp,
616 vsecattr.vsa_aclcnt * sizeof (aclent_t));
617 if (vsecattr.vsa_dfaclentp)
618 kmem_free(vsecattr.vsa_dfaclentp,
619 vsecattr.vsa_dfaclcnt * sizeof (aclent_t));
620 return (err);
621 }
622
623 /*
624 * Check the inode's ACL's to see if this mode of access is
625 * allowed; return 0 if allowed, EACCES if not.
626 *
627 * We follow the procedure defined in Sec. 3.3.5, ACL Access
628 * Check Algorithm, of the POSIX 1003.6 Draft Standard.
629 */
630 int
ufs_acl_access(struct inode * ip,int mode,cred_t * cr)631 ufs_acl_access(struct inode *ip, int mode, cred_t *cr)
632 /*
633 * ip parent inode
634 * mode mode of access read, write, execute/examine
635 * cr credentials
636 */
637 {
638 ufs_ic_acl_t *acl;
639 int ismask, mask = 0;
640 int gperm = 0;
641 int ngroup = 0;
642 si_t *sp = NULL;
643 uid_t uid = crgetuid(cr);
644 uid_t owner;
645
646 ASSERT(ip->i_ufs_acl != NULL);
647 ASSERT(RW_LOCK_HELD(&ip->i_contents));
648
649 sp = ip->i_ufs_acl;
650
651 ismask = sp->aclass.acl_ismask ?
652 sp->aclass.acl_ismask : NULL;
653
654 if (ismask)
655 mask = sp->aclass.acl_maskbits;
656 else
657 mask = -1;
658
659 /*
660 * (1) If user owns the file, obey user mode bits
661 */
662 owner = sp->aowner->acl_ic_who;
663 if (uid == owner) {
664 return (MODE_CHECK(owner, mode, (sp->aowner->acl_ic_perm << 6),
665 cr, ip));
666 }
667
668 /*
669 * (2) Obey any matching ACL_USER entry
670 */
671 if (sp->ausers)
672 for (acl = sp->ausers; acl != NULL; acl = acl->acl_ic_next) {
673 if (acl->acl_ic_who == uid) {
674 return (MODE_CHECK(owner, mode,
675 (mask & acl->acl_ic_perm) << 6, cr, ip));
676 }
677 }
678
679 /*
680 * (3) If user belongs to file's group, obey group mode bits
681 * if no ACL mask is defined; if there is an ACL mask, we look
682 * at both the group mode bits and any ACL_GROUP entries.
683 */
684 if (groupmember((uid_t)sp->agroup->acl_ic_who, cr)) {
685 ngroup++;
686 gperm = (sp->agroup->acl_ic_perm);
687 if (!ismask)
688 return (MODE_CHECK(owner, mode, (gperm << 6), cr, ip));
689 }
690
691 /*
692 * (4) Accumulate the permissions in matching ACL_GROUP entries
693 */
694 if (sp->agroups)
695 for (acl = sp->agroups; acl != NULL; acl = acl->acl_ic_next)
696 {
697 if (groupmember(acl->acl_ic_who, cr)) {
698 ngroup++;
699 gperm |= acl->acl_ic_perm;
700 }
701 }
702
703 if (ngroup != 0)
704 return (MODE_CHECK(owner, mode, ((gperm & mask) << 6), cr, ip));
705
706 /*
707 * (5) Finally, use the "other" mode bits
708 */
709 return (MODE_CHECK(owner, mode, sp->aother->acl_ic_perm << 6, cr, ip));
710 }
711
712 /*ARGSUSED2*/
713 int
ufs_acl_get(struct inode * ip,vsecattr_t * vsap,int flag,cred_t * cr)714 ufs_acl_get(struct inode *ip, vsecattr_t *vsap, int flag, cred_t *cr)
715 {
716 aclent_t *aclentp;
717
718 ASSERT(RW_LOCK_HELD(&ip->i_contents));
719
720 /* XXX Range check, sanity check, shadow check */
721 /* If an ACL is present, get the data from the shadow inode info */
722 if (ip->i_ufs_acl)
723 return (aclentry2vsecattr(ip->i_ufs_acl, vsap));
724
725 /*
726 * If no ACLs are present, fabricate one from the mode bits.
727 * This code is almost identical to fs_fab_acl(), but we
728 * already have the mode bits handy, so we'll avoid going
729 * through VOP_GETATTR() again.
730 */
731
732 vsap->vsa_aclcnt = 0;
733 vsap->vsa_aclentp = NULL;
734 vsap->vsa_dfaclcnt = 0; /* Default ACLs are not fabricated */
735 vsap->vsa_dfaclentp = NULL;
736
737 if (vsap->vsa_mask & (VSA_ACLCNT | VSA_ACL))
738 vsap->vsa_aclcnt = 4; /* USER, GROUP, OTHER, and CLASS */
739
740 if (vsap->vsa_mask & VSA_ACL) {
741 vsap->vsa_aclentp = kmem_zalloc(4 * sizeof (aclent_t),
742 KM_SLEEP);
743 if (vsap->vsa_aclentp == NULL)
744 return (ENOMEM);
745 aclentp = vsap->vsa_aclentp;
746
747 /* Owner */
748 aclentp->a_type = USER_OBJ;
749 aclentp->a_perm = ((ushort_t)(ip->i_mode & 0700)) >> 6;
750 aclentp->a_id = ip->i_uid; /* Really undefined */
751 aclentp++;
752
753 /* Group */
754 aclentp->a_type = GROUP_OBJ;
755 aclentp->a_perm = ((ushort_t)(ip->i_mode & 0070)) >> 3;
756 aclentp->a_id = ip->i_gid; /* Really undefined */
757 aclentp++;
758
759 /* Other */
760 aclentp->a_type = OTHER_OBJ;
761 aclentp->a_perm = ip->i_mode & 0007;
762 aclentp->a_id = 0; /* Really undefined */
763 aclentp++;
764
765 /* Class */
766 aclentp->a_type = CLASS_OBJ;
767 aclentp->a_perm = ((ushort_t)(ip->i_mode & 0070)) >> 3;
768 aclentp->a_id = 0; /* Really undefined */
769 ksort((caddr_t)vsap->vsa_aclentp, vsap->vsa_aclcnt,
770 sizeof (aclent_t), cmp2acls);
771 }
772
773 return (0);
774 }
775
776 /*ARGSUSED2*/
777 int
ufs_acl_set(struct inode * ip,vsecattr_t * vsap,int flag,cred_t * cr)778 ufs_acl_set(struct inode *ip, vsecattr_t *vsap, int flag, cred_t *cr)
779 {
780 si_t *sp;
781 int err;
782
783 ASSERT(RW_WRITE_HELD(&ip->i_contents));
784
785 if (!CHECK_ACL_ALLOWED(ip->i_mode & IFMT))
786 return (ENOSYS);
787
788 /*
789 * only the owner of the file or privileged users can change the ACLs
790 */
791 if (secpolicy_vnode_setdac(cr, ip->i_uid) != 0)
792 return (EPERM);
793
794 /* Convert from vsecattr struct to ufs_acl_entry struct */
795 if ((err = vsecattr2aclentry(vsap, &sp)) != 0)
796 return (err);
797 sp->s_dev = ip->i_dev;
798
799 /*
800 * Make the user & group objs in the acl list follow what's
801 * in the inode.
802 */
803 #ifdef DEBUG
804 if (vsap->vsa_mask == VSA_ACL) {
805 ASSERT(sp->aowner);
806 ASSERT(sp->agroup);
807 ASSERT(sp->aother);
808 }
809 #endif /* DEBUG */
810
811 if (sp->aowner)
812 sp->aowner->acl_ic_who = ip->i_uid;
813 if (sp->agroup)
814 sp->agroup->acl_ic_who = ip->i_gid;
815
816 /*
817 * Write and cache the new acl list
818 */
819 err = ufs_si_store(ip, sp, 1, cr);
820
821 return (err);
822 }
823
824 /*
825 * XXX Scan sorted array of acl's, checking for:
826 * 1) Any duplicate/conflicting entries (same type and id)
827 * 2) More than 1 of USER_OBJ, GROUP_OBJ, OTHER_OBJ, CLASS_OBJ
828 * 3) More than 1 of DEF_USER_OBJ, DEF_GROUP_OBJ, DEF_OTHER_OBJ, DEF_CLASS_OBJ
829 *
830 * Parameters:
831 * aclentp - ptr to sorted list of acl entries.
832 * nentries - # acl entries on the list
833 * flag - Bitmap (ACL_CHECK and/or DEF_ACL_CHECK) indicating whether the
834 * list contains regular acls, default acls, or both.
835 *
836 * Returns: 0 - Success
837 * EINVAL - Invalid list (dups or multiple entries of type USER_OBJ, etc)
838 */
839 static int
acl_validate(aclent_t * aclentp,int nentries,int flag)840 acl_validate(aclent_t *aclentp, int nentries, int flag)
841 {
842 int i;
843 int nuser_objs = 0;
844 int ngroup_objs = 0;
845 int nother_objs = 0;
846 int nclass_objs = 0;
847 int ndef_user_objs = 0;
848 int ndef_group_objs = 0;
849 int ndef_other_objs = 0;
850 int ndef_class_objs = 0;
851 int nusers = 0;
852 int ngroups = 0;
853 int ndef_users = 0;
854 int ndef_groups = 0;
855 int numdefs = 0;
856
857 /* Null list or list of one */
858 if (aclentp == NULL)
859 return (0);
860
861 if (nentries <= 0)
862 return (EINVAL);
863
864 for (i = 1; i < nentries; i++) {
865 if (((aclentp[i - 1].a_type == aclentp[i].a_type) &&
866 (aclentp[i - 1].a_id == aclentp[i].a_id)) ||
867 (aclentp[i - 1].a_perm > 07)) {
868 return (EINVAL);
869 }
870 }
871
872 if (flag == 0 || (flag != ACL_CHECK && flag != DEF_ACL_CHECK))
873 return (EINVAL);
874
875 /* Count types */
876 for (i = 0; i < nentries; i++) {
877 switch (aclentp[i].a_type) {
878 case USER_OBJ: /* Owner */
879 nuser_objs++;
880 break;
881 case GROUP_OBJ: /* Group */
882 ngroup_objs++;
883 break;
884 case OTHER_OBJ: /* Other */
885 nother_objs++;
886 break;
887 case CLASS_OBJ: /* Mask */
888 nclass_objs++;
889 break;
890 case DEF_USER_OBJ: /* Default Owner */
891 ndef_user_objs++;
892 break;
893 case DEF_GROUP_OBJ: /* Default Group */
894 ndef_group_objs++;
895 break;
896 case DEF_OTHER_OBJ: /* Default Other */
897 ndef_other_objs++;
898 break;
899 case DEF_CLASS_OBJ: /* Default Mask */
900 ndef_class_objs++;
901 break;
902 case USER: /* Users */
903 nusers++;
904 break;
905 case GROUP: /* Groups */
906 ngroups++;
907 break;
908 case DEF_USER: /* Default Users */
909 ndef_users++;
910 break;
911 case DEF_GROUP: /* Default Groups */
912 ndef_groups++;
913 break;
914 default: /* Unknown type */
915 return (EINVAL);
916 }
917 }
918
919 /*
920 * For normal acl's, we require there be one (and only one)
921 * USER_OBJ, GROUP_OBJ and OTHER_OBJ. There is either zero
922 * or one CLASS_OBJ.
923 */
924 if (flag & ACL_CHECK) {
925 if (nuser_objs != 1 || ngroup_objs != 1 ||
926 nother_objs != 1 || nclass_objs > 1) {
927 return (EINVAL);
928 }
929 /*
930 * If there are ANY group acls, there MUST be a
931 * class_obj(mask) acl (1003.6/D12 p. 29 lines 75-80).
932 */
933 if (ngroups && !nclass_objs) {
934 return (EINVAL);
935 }
936 if (nuser_objs + ngroup_objs + nother_objs + nclass_objs +
937 ngroups + nusers > MAX_ACL_ENTRIES)
938 return (EINVAL);
939 }
940
941 /*
942 * For default acl's, we require that there be either one (and only one)
943 * DEF_USER_OBJ, DEF_GROUP_OBJ and DEF_OTHER_OBJ
944 * or there be none of them.
945 */
946 if (flag & DEF_ACL_CHECK) {
947 if (ndef_other_objs > 1 || ndef_user_objs > 1 ||
948 ndef_group_objs > 1 || ndef_class_objs > 1) {
949 return (EINVAL);
950 }
951
952 numdefs = ndef_other_objs + ndef_user_objs + ndef_group_objs;
953
954 if (numdefs != 0 && numdefs != 3) {
955 return (EINVAL);
956 }
957 /*
958 * If there are ANY def_group acls, there MUST be a
959 * def_class_obj(mask) acl (1003.6/D12 P. 29 lines 75-80).
960 * XXX(jimh) This is inferred.
961 */
962 if (ndef_groups && !ndef_class_objs) {
963 return (EINVAL);
964 }
965 if ((ndef_users || ndef_groups) &&
966 ((numdefs != 3) && !ndef_class_objs)) {
967 return (EINVAL);
968 }
969 if (ndef_user_objs + ndef_group_objs + ndef_other_objs +
970 ndef_class_objs + ndef_users + ndef_groups >
971 MAX_ACL_ENTRIES)
972 return (EINVAL);
973 }
974 return (0);
975 }
976
977 static int
formacl(ufs_ic_acl_t ** aclpp,aclent_t * aclentp)978 formacl(ufs_ic_acl_t **aclpp, aclent_t *aclentp)
979 {
980 ufs_ic_acl_t *uaclp;
981
982 uaclp = kmem_alloc(sizeof (ufs_ic_acl_t), KM_SLEEP);
983 uaclp->acl_ic_perm = aclentp->a_perm;
984 uaclp->acl_ic_who = aclentp->a_id;
985 uaclp->acl_ic_next = *aclpp;
986 *aclpp = uaclp;
987 return (0);
988 }
989
990 /*
991 * XXX - Make more efficient
992 * Convert from the vsecattr struct, used by the VOP interface, to
993 * the ufs_acl_entry struct used for in-core storage of acl's.
994 *
995 * Parameters:
996 * vsap - Ptr to array of security attributes.
997 * spp - Ptr to ptr to si struct for the results
998 *
999 * Returns: 0 - Success
1000 * N - From errno.h
1001 */
1002 static int
vsecattr2aclentry(vsecattr_t * vsap,si_t ** spp)1003 vsecattr2aclentry(vsecattr_t *vsap, si_t **spp)
1004 {
1005 aclent_t *aclentp, *aclp;
1006 si_t *sp;
1007 int err;
1008 int i;
1009
1010 /* Sort & validate the lists on the vsap */
1011 ksort((caddr_t)vsap->vsa_aclentp, vsap->vsa_aclcnt,
1012 sizeof (aclent_t), cmp2acls);
1013 ksort((caddr_t)vsap->vsa_dfaclentp, vsap->vsa_dfaclcnt,
1014 sizeof (aclent_t), cmp2acls);
1015 if ((err = acl_validate(vsap->vsa_aclentp,
1016 vsap->vsa_aclcnt, ACL_CHECK)) != 0)
1017 return (err);
1018 if ((err = acl_validate(vsap->vsa_dfaclentp,
1019 vsap->vsa_dfaclcnt, DEF_ACL_CHECK)) != 0)
1020 return (err);
1021
1022 /* Create new si struct and hang acl's off it */
1023 sp = kmem_zalloc(sizeof (si_t), KM_SLEEP);
1024 rw_init(&sp->s_lock, NULL, RW_DEFAULT, NULL);
1025
1026 /* Process acl list */
1027 aclp = (aclent_t *)vsap->vsa_aclentp;
1028 aclentp = aclp + vsap->vsa_aclcnt - 1;
1029 for (i = 0; i < vsap->vsa_aclcnt; i++) {
1030 switch (aclentp->a_type) {
1031 case USER_OBJ: /* Owner */
1032 if (err = formacl(&sp->aowner, aclentp))
1033 goto error;
1034 break;
1035 case GROUP_OBJ: /* Group */
1036 if (err = formacl(&sp->agroup, aclentp))
1037 goto error;
1038 break;
1039 case OTHER_OBJ: /* Other */
1040 if (err = formacl(&sp->aother, aclentp))
1041 goto error;
1042 break;
1043 case USER:
1044 if (err = formacl(&sp->ausers, aclentp))
1045 goto error;
1046 break;
1047 case CLASS_OBJ: /* Mask */
1048 sp->aclass.acl_ismask = 1;
1049 sp->aclass.acl_maskbits = aclentp->a_perm;
1050 break;
1051 case GROUP:
1052 if (err = formacl(&sp->agroups, aclentp))
1053 goto error;
1054 break;
1055 default:
1056 break;
1057 }
1058 aclentp--;
1059 }
1060
1061 /* Process default acl list */
1062 aclp = (aclent_t *)vsap->vsa_dfaclentp;
1063 aclentp = aclp + vsap->vsa_dfaclcnt - 1;
1064 for (i = 0; i < vsap->vsa_dfaclcnt; i++) {
1065 switch (aclentp->a_type) {
1066 case DEF_USER_OBJ: /* Default Owner */
1067 if (err = formacl(&sp->downer, aclentp))
1068 goto error;
1069 break;
1070 case DEF_GROUP_OBJ: /* Default Group */
1071 if (err = formacl(&sp->dgroup, aclentp))
1072 goto error;
1073 break;
1074 case DEF_OTHER_OBJ: /* Default Other */
1075 if (err = formacl(&sp->dother, aclentp))
1076 goto error;
1077 break;
1078 case DEF_USER:
1079 if (err = formacl(&sp->dusers, aclentp))
1080 goto error;
1081 break;
1082 case DEF_CLASS_OBJ: /* Default Mask */
1083 sp->dclass.acl_ismask = 1;
1084 sp->dclass.acl_maskbits = aclentp->a_perm;
1085 break;
1086 case DEF_GROUP:
1087 if (err = formacl(&sp->dgroups, aclentp))
1088 goto error;
1089 break;
1090 default:
1091 break;
1092 }
1093 aclentp--;
1094 }
1095 *spp = sp;
1096 return (0);
1097
1098 error:
1099 ufs_si_free_mem(sp);
1100 return (err);
1101 }
1102
1103 void
formvsec(int obj_type,ufs_ic_acl_t * aclp,aclent_t ** aclentpp)1104 formvsec(int obj_type, ufs_ic_acl_t *aclp, aclent_t **aclentpp)
1105 {
1106 for (; aclp; aclp = aclp->acl_ic_next) {
1107 (*aclentpp)->a_type = obj_type;
1108 (*aclentpp)->a_perm = aclp->acl_ic_perm;
1109 (*aclentpp)->a_id = aclp->acl_ic_who;
1110 (*aclentpp)++;
1111 }
1112 }
1113
1114 /*
1115 * XXX - Make more efficient
1116 * Convert from the ufs_acl_entry struct used for in-core storage of acl's
1117 * to the vsecattr struct, used by the VOP interface.
1118 *
1119 * Parameters:
1120 * sp - Ptr to si struct with the acls
1121 * vsap - Ptr to a vsecattr struct which will take the results.
1122 *
1123 * Returns: 0 - Success
1124 * N - From errno table
1125 */
1126 static int
aclentry2vsecattr(si_t * sp,vsecattr_t * vsap)1127 aclentry2vsecattr(si_t *sp, vsecattr_t *vsap)
1128 {
1129 aclent_t *aclentp;
1130 int numacls = 0;
1131 int err;
1132
1133 vsap->vsa_aclentp = vsap->vsa_dfaclentp = NULL;
1134
1135 numacls = acl_count(sp->aowner) +
1136 acl_count(sp->agroup) +
1137 acl_count(sp->aother) +
1138 acl_count(sp->ausers) +
1139 acl_count(sp->agroups);
1140 if (sp->aclass.acl_ismask)
1141 numacls++;
1142
1143 if (vsap->vsa_mask & (VSA_ACLCNT | VSA_ACL))
1144 vsap->vsa_aclcnt = numacls;
1145
1146 if (numacls == 0)
1147 goto do_defaults;
1148
1149 if (vsap->vsa_mask & VSA_ACL) {
1150 vsap->vsa_aclentp = kmem_zalloc(numacls * sizeof (aclent_t),
1151 KM_SLEEP);
1152 aclentp = vsap->vsa_aclentp;
1153
1154 formvsec(USER_OBJ, sp->aowner, &aclentp);
1155 formvsec(USER, sp->ausers, &aclentp);
1156 formvsec(GROUP_OBJ, sp->agroup, &aclentp);
1157 formvsec(GROUP, sp->agroups, &aclentp);
1158 formvsec(OTHER_OBJ, sp->aother, &aclentp);
1159
1160 if (sp->aclass.acl_ismask) {
1161 aclentp->a_type = CLASS_OBJ; /* Mask */
1162 aclentp->a_perm = sp->aclass.acl_maskbits;
1163 aclentp->a_id = 0;
1164 aclentp++;
1165 }
1166
1167 /* Sort the acl list */
1168 ksort((caddr_t)vsap->vsa_aclentp, vsap->vsa_aclcnt,
1169 sizeof (aclent_t), cmp2acls);
1170 /* Check the acl list */
1171 if ((err = acl_validate(vsap->vsa_aclentp,
1172 vsap->vsa_aclcnt, ACL_CHECK)) != 0) {
1173 kmem_free(vsap->vsa_aclentp,
1174 numacls * sizeof (aclent_t));
1175 vsap->vsa_aclentp = NULL;
1176 return (err);
1177 }
1178
1179 }
1180 do_defaults:
1181 /* Process Defaults */
1182
1183 numacls = acl_count(sp->downer) +
1184 acl_count(sp->dgroup) +
1185 acl_count(sp->dother) +
1186 acl_count(sp->dusers) +
1187 acl_count(sp->dgroups);
1188 if (sp->dclass.acl_ismask)
1189 numacls++;
1190
1191 if (vsap->vsa_mask & (VSA_DFACLCNT | VSA_DFACL))
1192 vsap->vsa_dfaclcnt = numacls;
1193
1194 if (numacls == 0)
1195 goto do_others;
1196
1197 if (vsap->vsa_mask & VSA_DFACL) {
1198 vsap->vsa_dfaclentp =
1199 kmem_zalloc(numacls * sizeof (aclent_t), KM_SLEEP);
1200 aclentp = vsap->vsa_dfaclentp;
1201 formvsec(DEF_USER_OBJ, sp->downer, &aclentp);
1202 formvsec(DEF_USER, sp->dusers, &aclentp);
1203 formvsec(DEF_GROUP_OBJ, sp->dgroup, &aclentp);
1204 formvsec(DEF_GROUP, sp->dgroups, &aclentp);
1205 formvsec(DEF_OTHER_OBJ, sp->dother, &aclentp);
1206
1207 if (sp->dclass.acl_ismask) {
1208 aclentp->a_type = DEF_CLASS_OBJ; /* Mask */
1209 aclentp->a_perm = sp->dclass.acl_maskbits;
1210 aclentp->a_id = 0;
1211 aclentp++;
1212 }
1213
1214 /* Sort the default acl list */
1215 ksort((caddr_t)vsap->vsa_dfaclentp, vsap->vsa_dfaclcnt,
1216 sizeof (aclent_t), cmp2acls);
1217 if ((err = acl_validate(vsap->vsa_dfaclentp,
1218 vsap->vsa_dfaclcnt, DEF_ACL_CHECK)) != 0) {
1219 if (vsap->vsa_aclentp != NULL)
1220 kmem_free(vsap->vsa_aclentp,
1221 vsap->vsa_aclcnt * sizeof (aclent_t));
1222 kmem_free(vsap->vsa_dfaclentp,
1223 vsap->vsa_dfaclcnt * sizeof (aclent_t));
1224 vsap->vsa_aclentp = vsap->vsa_dfaclentp = NULL;
1225 return (err);
1226 }
1227 }
1228
1229 do_others:
1230 return (0);
1231 }
1232
1233 static void
acl_free(ufs_ic_acl_t * aclp)1234 acl_free(ufs_ic_acl_t *aclp)
1235 {
1236 while (aclp != NULL) {
1237 ufs_ic_acl_t *nextaclp = aclp->acl_ic_next;
1238 kmem_free(aclp, sizeof (ufs_ic_acl_t));
1239 aclp = nextaclp;
1240 }
1241 }
1242
1243 /*
1244 * ufs_si_free_mem will discard the sp, and the acl hanging off of the
1245 * sp. It is required that the sp not be locked, and not be in the
1246 * cache.
1247 *
1248 * input: pointer to sp to discard.
1249 *
1250 * return - nothing.
1251 *
1252 */
1253 static void
ufs_si_free_mem(si_t * sp)1254 ufs_si_free_mem(si_t *sp)
1255 {
1256 ASSERT(!(sp->s_flags & SI_CACHED));
1257 ASSERT(!RW_LOCK_HELD(&sp->s_lock));
1258 /*
1259 * remove from the cache
1260 * free the acl entries
1261 */
1262 acl_free(sp->aowner);
1263 acl_free(sp->agroup);
1264 acl_free(sp->aother);
1265 acl_free(sp->ausers);
1266 acl_free(sp->agroups);
1267
1268 acl_free(sp->downer);
1269 acl_free(sp->dgroup);
1270 acl_free(sp->dother);
1271 acl_free(sp->dusers);
1272 acl_free(sp->dgroups);
1273
1274 rw_destroy(&sp->s_lock);
1275 kmem_free(sp, sizeof (si_t));
1276 }
1277
1278 void
acl_cpy(ufs_ic_acl_t * saclp,ufs_ic_acl_t * daclp)1279 acl_cpy(ufs_ic_acl_t *saclp, ufs_ic_acl_t *daclp)
1280 {
1281 ufs_ic_acl_t *aclp, *prev_aclp = NULL, *aclp1;
1282
1283 if (saclp == NULL) {
1284 daclp = NULL;
1285 return;
1286 }
1287 prev_aclp = daclp;
1288
1289 for (aclp = saclp; aclp != NULL; aclp = aclp->acl_ic_next) {
1290 aclp1 = kmem_alloc(sizeof (ufs_ic_acl_t), KM_SLEEP);
1291 aclp1->acl_ic_next = NULL;
1292 aclp1->acl_ic_who = aclp->acl_ic_who;
1293 aclp1->acl_ic_perm = aclp->acl_ic_perm;
1294 prev_aclp->acl_ic_next = aclp1;
1295 prev_aclp = (ufs_ic_acl_t *)&aclp1->acl_ic_next;
1296 }
1297 }
1298
1299 /*
1300 * ufs_si_inherit takes a parent acl structure (saclp) and the inode
1301 * of the object that is inheriting an acl and returns the inode
1302 * with the acl linked to it. It also writes the acl to disk if
1303 * it is a unique inode.
1304 *
1305 * ip - pointer to inode of object inheriting the acl (contents lock)
1306 * tdp - parent inode (rw_lock and contents lock)
1307 * mode - creation modes
1308 * cr - credentials pointer
1309 */
1310 int
ufs_si_inherit(struct inode * ip,struct inode * tdp,o_mode_t mode,cred_t * cr)1311 ufs_si_inherit(struct inode *ip, struct inode *tdp, o_mode_t mode, cred_t *cr)
1312 {
1313 si_t *tsp, *sp = tdp->i_ufs_acl;
1314 int error;
1315 o_mode_t old_modes, old_uid, old_gid;
1316 int mask;
1317
1318 ASSERT(RW_WRITE_HELD(&ip->i_contents));
1319 ASSERT(RW_WRITE_HELD(&tdp->i_rwlock));
1320 ASSERT(RW_WRITE_HELD(&tdp->i_contents));
1321
1322 /*
1323 * if links/symbolic links, or other invalid acl objects are copied
1324 * or moved to a directory with a default acl do not allow inheritance
1325 * just return.
1326 */
1327 if (!CHECK_ACL_ALLOWED(ip->i_mode & IFMT))
1328 return (0);
1329
1330 /* lock the parent security information */
1331 rw_enter(&sp->s_lock, RW_READER);
1332
1333 ASSERT(((tdp->i_mode & IFMT) == IFDIR) ||
1334 ((tdp->i_mode & IFMT) == IFATTRDIR));
1335
1336 mask = ((sp->downer != NULL) ? 1 : 0) |
1337 ((sp->dgroup != NULL) ? 2 : 0) |
1338 ((sp->dother != NULL) ? 4 : 0);
1339
1340 if (mask == 0) {
1341 rw_exit(&sp->s_lock);
1342 return (0);
1343 }
1344
1345 if (mask != 7) {
1346 rw_exit(&sp->s_lock);
1347 return (EINVAL);
1348 }
1349
1350 tsp = kmem_zalloc(sizeof (si_t), KM_SLEEP);
1351 rw_init(&tsp->s_lock, NULL, RW_DEFAULT, NULL);
1352
1353 /* copy the default acls */
1354
1355 ASSERT(RW_READ_HELD(&sp->s_lock));
1356 acl_cpy(sp->downer, (ufs_ic_acl_t *)&tsp->aowner);
1357 acl_cpy(sp->dgroup, (ufs_ic_acl_t *)&tsp->agroup);
1358 acl_cpy(sp->dother, (ufs_ic_acl_t *)&tsp->aother);
1359 acl_cpy(sp->dusers, (ufs_ic_acl_t *)&tsp->ausers);
1360 acl_cpy(sp->dgroups, (ufs_ic_acl_t *)&tsp->agroups);
1361 tsp->aclass.acl_ismask = sp->dclass.acl_ismask;
1362 tsp->aclass.acl_maskbits = sp->dclass.acl_maskbits;
1363
1364 /*
1365 * set the owner, group, and other values from the master
1366 * inode.
1367 */
1368
1369 MODE2ACL(tsp->aowner, (mode >> 6), ip->i_uid);
1370 MODE2ACL(tsp->agroup, (mode >> 3), ip->i_gid);
1371 MODE2ACL(tsp->aother, (mode), 0);
1372
1373 if (tsp->aclass.acl_ismask) {
1374 tsp->aclass.acl_maskbits &= mode >> 3;
1375 }
1376
1377
1378 /* copy default acl if necessary */
1379
1380 if (((ip->i_mode & IFMT) == IFDIR) ||
1381 ((ip->i_mode & IFMT) == IFATTRDIR)) {
1382 acl_cpy(sp->downer, (ufs_ic_acl_t *)&tsp->downer);
1383 acl_cpy(sp->dgroup, (ufs_ic_acl_t *)&tsp->dgroup);
1384 acl_cpy(sp->dother, (ufs_ic_acl_t *)&tsp->dother);
1385 acl_cpy(sp->dusers, (ufs_ic_acl_t *)&tsp->dusers);
1386 acl_cpy(sp->dgroups, (ufs_ic_acl_t *)&tsp->dgroups);
1387 tsp->dclass.acl_ismask = sp->dclass.acl_ismask;
1388 tsp->dclass.acl_maskbits = sp->dclass.acl_maskbits;
1389 }
1390 /*
1391 * save the new 9 mode bits in the inode (ip->ic_smode) for
1392 * ufs_getattr. Be sure the mode can be recovered if the store
1393 * fails.
1394 */
1395 old_modes = ip->i_mode;
1396 old_uid = ip->i_uid;
1397 old_gid = ip->i_gid;
1398 /*
1399 * store the acl, and get back a new security anchor if
1400 * it is a duplicate.
1401 */
1402 rw_exit(&sp->s_lock);
1403 rw_enter(&ip->i_rwlock, RW_WRITER);
1404
1405 /*
1406 * Suppress out of inodes messages if instructed in the
1407 * tdp inode.
1408 */
1409 ip->i_flag |= tdp->i_flag & IQUIET;
1410
1411 if ((error = ufs_si_store(ip, tsp, 0, cr)) != 0) {
1412 ip->i_mode = old_modes;
1413 ip->i_uid = old_uid;
1414 ip->i_gid = old_gid;
1415 }
1416 ip->i_flag &= ~IQUIET;
1417 rw_exit(&ip->i_rwlock);
1418 return (error);
1419 }
1420
1421 si_t *
ufs_acl_cp(si_t * sp)1422 ufs_acl_cp(si_t *sp)
1423 {
1424
1425 si_t *dsp;
1426
1427 ASSERT(RW_READ_HELD(&sp->s_lock));
1428 ASSERT(sp->s_ref && sp->s_use);
1429
1430 dsp = kmem_zalloc(sizeof (si_t), KM_SLEEP);
1431 rw_init(&dsp->s_lock, NULL, RW_DEFAULT, NULL);
1432
1433 acl_cpy(sp->aowner, (ufs_ic_acl_t *)&dsp->aowner);
1434 acl_cpy(sp->agroup, (ufs_ic_acl_t *)&dsp->agroup);
1435 acl_cpy(sp->aother, (ufs_ic_acl_t *)&dsp->aother);
1436 acl_cpy(sp->ausers, (ufs_ic_acl_t *)&dsp->ausers);
1437 acl_cpy(sp->agroups, (ufs_ic_acl_t *)&dsp->agroups);
1438
1439 dsp->aclass.acl_ismask = sp->aclass.acl_ismask;
1440 dsp->aclass.acl_maskbits = sp->aclass.acl_maskbits;
1441
1442 acl_cpy(sp->downer, (ufs_ic_acl_t *)&dsp->downer);
1443 acl_cpy(sp->dgroup, (ufs_ic_acl_t *)&dsp->dgroup);
1444 acl_cpy(sp->dother, (ufs_ic_acl_t *)&dsp->dother);
1445 acl_cpy(sp->dusers, (ufs_ic_acl_t *)&dsp->dusers);
1446 acl_cpy(sp->dgroups, (ufs_ic_acl_t *)&dsp->dgroups);
1447
1448 dsp->dclass.acl_ismask = sp->dclass.acl_ismask;
1449 dsp->dclass.acl_maskbits = sp->dclass.acl_maskbits;
1450
1451 return (dsp);
1452
1453 }
1454
1455 int
ufs_acl_setattr(struct inode * ip,struct vattr * vap,cred_t * cr)1456 ufs_acl_setattr(struct inode *ip, struct vattr *vap, cred_t *cr)
1457 {
1458
1459 si_t *sp;
1460 int mask = vap->va_mask;
1461 int error = 0;
1462
1463 ASSERT(RW_WRITE_HELD(&ip->i_contents));
1464
1465 if (!(mask & (AT_MODE|AT_UID|AT_GID)))
1466 return (0);
1467
1468 /*
1469 * if no regular acl's, nothing to do, so let's get out
1470 */
1471 if (!(ip->i_ufs_acl) || !(ip->i_ufs_acl->aowner))
1472 return (0);
1473
1474 rw_enter(&ip->i_ufs_acl->s_lock, RW_READER);
1475 sp = ufs_acl_cp(ip->i_ufs_acl);
1476 ASSERT(sp != ip->i_ufs_acl);
1477
1478 /*
1479 * set the mask to the group permissions if a mask entry
1480 * exists. Otherwise, set the group obj bits to the group
1481 * permissions. Since non-trivial ACLs always have a mask,
1482 * and the mask is the final arbiter of group permissions,
1483 * setting the mask has the effect of changing the effective
1484 * group permissions, even if the group_obj permissions in
1485 * the ACL aren't changed. Posix P1003.1e states that when
1486 * an ACL mask exists, chmod(2) must set the acl mask (NOT the
1487 * group_obj permissions) to the requested group permissions.
1488 */
1489 if (mask & AT_MODE) {
1490 sp->aowner->acl_ic_perm = (o_mode_t)(ip->i_mode & 0700) >> 6;
1491 if (sp->aclass.acl_ismask)
1492 sp->aclass.acl_maskbits =
1493 (o_mode_t)(ip->i_mode & 070) >> 3;
1494 else
1495 sp->agroup->acl_ic_perm =
1496 (o_mode_t)(ip->i_mode & 070) >> 3;
1497 sp->aother->acl_ic_perm = (o_mode_t)(ip->i_mode & 07);
1498 }
1499
1500 if (mask & AT_UID) {
1501 /* Caller has verified our privileges */
1502 sp->aowner->acl_ic_who = ip->i_uid;
1503 }
1504
1505 if (mask & AT_GID) {
1506 sp->agroup->acl_ic_who = ip->i_gid;
1507 }
1508
1509 rw_exit(&ip->i_ufs_acl->s_lock);
1510 error = ufs_si_store(ip, sp, 0, cr);
1511 return (error);
1512 }
1513
1514 static int
acl_count(ufs_ic_acl_t * p)1515 acl_count(ufs_ic_acl_t *p)
1516 {
1517 ufs_ic_acl_t *acl;
1518 int count;
1519
1520 for (count = 0, acl = p; acl; acl = acl->acl_ic_next, count++)
1521 ;
1522 return (count);
1523 }
1524
1525 /*
1526 * Takes as input a security structure and generates a buffer
1527 * with fsd's in a form which be written to the shadow inode.
1528 */
1529 static int
ufs_sectobuf(si_t * sp,caddr_t * buf,size_t * len)1530 ufs_sectobuf(si_t *sp, caddr_t *buf, size_t *len)
1531 {
1532 size_t acl_size;
1533 size_t def_acl_size;
1534 caddr_t buffer;
1535 struct ufs_fsd *fsdp;
1536 ufs_acl_t *bufaclp;
1537
1538 /*
1539 * Calc size of buffer to hold all the acls
1540 */
1541 acl_size = acl_count(sp->aowner) + /* owner */
1542 acl_count(sp->agroup) + /* owner group */
1543 acl_count(sp->aother) + /* owner other */
1544 acl_count(sp->ausers) + /* acl list */
1545 acl_count(sp->agroups); /* group alcs */
1546 if (sp->aclass.acl_ismask)
1547 acl_size++;
1548
1549 /* Convert to bytes */
1550 acl_size *= sizeof (ufs_acl_t);
1551
1552 /* Add fsd header */
1553 if (acl_size)
1554 acl_size += 2 * sizeof (int);
1555
1556 /*
1557 * Calc size of buffer to hold all the default acls
1558 */
1559 def_acl_size =
1560 acl_count(sp->downer) + /* def owner */
1561 acl_count(sp->dgroup) + /* def owner group */
1562 acl_count(sp->dother) + /* def owner other */
1563 acl_count(sp->dusers) + /* def users */
1564 acl_count(sp->dgroups); /* def group acls */
1565 if (sp->dclass.acl_ismask)
1566 def_acl_size++;
1567
1568 /*
1569 * Convert to bytes
1570 */
1571 def_acl_size *= sizeof (ufs_acl_t);
1572
1573 /*
1574 * Add fsd header
1575 */
1576 if (def_acl_size)
1577 def_acl_size += 2 * sizeof (int);
1578
1579 if (acl_size + def_acl_size == 0)
1580 return (0);
1581
1582 buffer = kmem_zalloc((acl_size + def_acl_size), KM_SLEEP);
1583 bufaclp = (ufs_acl_t *)buffer;
1584
1585 if (acl_size == 0)
1586 goto wrtdefs;
1587
1588 /* create fsd and copy acls */
1589 fsdp = (struct ufs_fsd *)bufaclp;
1590 fsdp->fsd_type = FSD_ACL;
1591 bufaclp = (ufs_acl_t *)&fsdp->fsd_data[0];
1592
1593 ACL_MOVE(sp->aowner, USER_OBJ, bufaclp);
1594 ACL_MOVE(sp->agroup, GROUP_OBJ, bufaclp);
1595 ACL_MOVE(sp->aother, OTHER_OBJ, bufaclp);
1596 ACL_MOVE(sp->ausers, USER, bufaclp);
1597 ACL_MOVE(sp->agroups, GROUP, bufaclp);
1598
1599 if (sp->aclass.acl_ismask) {
1600 bufaclp->acl_tag = CLASS_OBJ;
1601 bufaclp->acl_who = (uid_t)sp->aclass.acl_ismask;
1602 bufaclp->acl_perm = (o_mode_t)sp->aclass.acl_maskbits;
1603 bufaclp++;
1604 }
1605 ASSERT(acl_size <= INT_MAX);
1606 fsdp->fsd_size = (int)acl_size;
1607
1608 wrtdefs:
1609 if (def_acl_size == 0)
1610 goto alldone;
1611
1612 /* if defaults exist then create fsd and copy default acls */
1613 fsdp = (struct ufs_fsd *)bufaclp;
1614 fsdp->fsd_type = FSD_DFACL;
1615 bufaclp = (ufs_acl_t *)&fsdp->fsd_data[0];
1616
1617 ACL_MOVE(sp->downer, DEF_USER_OBJ, bufaclp);
1618 ACL_MOVE(sp->dgroup, DEF_GROUP_OBJ, bufaclp);
1619 ACL_MOVE(sp->dother, DEF_OTHER_OBJ, bufaclp);
1620 ACL_MOVE(sp->dusers, DEF_USER, bufaclp);
1621 ACL_MOVE(sp->dgroups, DEF_GROUP, bufaclp);
1622 if (sp->dclass.acl_ismask) {
1623 bufaclp->acl_tag = DEF_CLASS_OBJ;
1624 bufaclp->acl_who = (uid_t)sp->dclass.acl_ismask;
1625 bufaclp->acl_perm = (o_mode_t)sp->dclass.acl_maskbits;
1626 bufaclp++;
1627 }
1628 ASSERT(def_acl_size <= INT_MAX);
1629 fsdp->fsd_size = (int)def_acl_size;
1630
1631 alldone:
1632 *buf = buffer;
1633 *len = acl_size + def_acl_size;
1634
1635 return (0);
1636 }
1637
1638 /*
1639 * free a shadow inode on disk and in memory
1640 */
1641 int
ufs_si_free(si_t * sp,struct vfs * vfsp,cred_t * cr)1642 ufs_si_free(si_t *sp, struct vfs *vfsp, cred_t *cr)
1643 {
1644 struct inode *sip;
1645 int shadow;
1646 int err = 0;
1647 int refcnt;
1648 int signature;
1649
1650 ASSERT(vfsp);
1651 ASSERT(sp);
1652
1653 rw_enter(&sp->s_lock, RW_READER);
1654 ASSERT(sp->s_shadow <= INT_MAX);
1655 shadow = (int)sp->s_shadow;
1656 ASSERT(sp->s_ref);
1657 rw_exit(&sp->s_lock);
1658
1659 /*
1660 * Decrement link count on the shadow inode,
1661 * and decrement reference count on the sip.
1662 */
1663 if ((err = ufs_iget_alloced(vfsp, shadow, &sip, cr)) == 0) {
1664 rw_enter(&sip->i_contents, RW_WRITER);
1665 rw_enter(&sp->s_lock, RW_WRITER);
1666 ASSERT(sp->s_shadow == shadow);
1667 ASSERT(sip->i_dquot == 0);
1668 /* Decrement link count */
1669 ASSERT(sip->i_nlink > 0);
1670 /*
1671 * bug #1264710 assertion failure below
1672 */
1673 sp->s_use = --sip->i_nlink;
1674 ufs_setreclaim(sip);
1675 TRANS_INODE(sip->i_ufsvfs, sip);
1676 sip->i_flag |= ICHG | IMOD;
1677 sip->i_seq++;
1678 ITIMES_NOLOCK(sip);
1679 /* Dec ref counts on si referenced by this ip */
1680 refcnt = --sp->s_ref;
1681 signature = sp->s_signature;
1682 ASSERT(sp->s_ref >= 0 && sp->s_ref <= sp->s_use);
1683 /*
1684 * Release s_lock before calling VN_RELE
1685 * (which may want to acquire i_contents).
1686 */
1687 rw_exit(&sp->s_lock);
1688 rw_exit(&sip->i_contents);
1689 VN_RELE(ITOV(sip));
1690 } else {
1691 rw_enter(&sp->s_lock, RW_WRITER);
1692 /* Dec ref counts on si referenced by this ip */
1693 refcnt = --sp->s_ref;
1694 signature = sp->s_signature;
1695 ASSERT(sp->s_ref >= 0 && sp->s_ref <= sp->s_use);
1696 rw_exit(&sp->s_lock);
1697 }
1698
1699 if (refcnt == 0)
1700 si_cache_del(sp, signature);
1701 return (err);
1702 }
1703
1704 /*
1705 * Seach the si cache for an si structure by inode #.
1706 * Returns a locked si structure.
1707 *
1708 * Parameters:
1709 * ip - Ptr to an inode on this fs
1710 * spp - Ptr to ptr to si struct for the results, if found.
1711 *
1712 * Returns: 0 - Success (results in spp)
1713 * 1 - Failure (spp undefined)
1714 */
1715 static int
si_cachei_get(struct inode * ip,si_t ** spp)1716 si_cachei_get(struct inode *ip, si_t **spp)
1717 {
1718 si_t *sp;
1719
1720 rw_enter(&si_cache_lock, RW_READER);
1721 loop:
1722 for (sp = si_cachei[SI_HASH(ip->i_shadow)]; sp; sp = sp->s_forw)
1723 if (sp->s_shadow == ip->i_shadow && sp->s_dev == ip->i_dev)
1724 break;
1725
1726 if (sp == NULL) {
1727 /* Not in cache */
1728 rw_exit(&si_cache_lock);
1729 return (1);
1730 }
1731 /* Found it */
1732 rw_enter(&sp->s_lock, RW_WRITER);
1733 alldone:
1734 rw_exit(&si_cache_lock);
1735 *spp = sp;
1736 return (0);
1737 }
1738
1739 /*
1740 * Seach the si cache by si structure (ie duplicate of the one passed in).
1741 * In order for a match the signatures must be the same and
1742 * the devices must be the same, the acls must match and
1743 * link count of the cached shadow must be less than the
1744 * size of ic_nlink - 1. MAXLINK - 1 is used to allow the count
1745 * to be incremented one more time by the caller.
1746 * Returns a locked si structure.
1747 *
1748 * Parameters:
1749 * ip - Ptr to an inode on this fs
1750 * spi - Ptr to si the struct we're searching the cache for.
1751 * spp - Ptr to ptr to si struct for the results, if found.
1752 *
1753 * Returns: 0 - Success (results in spp)
1754 * 1 - Failure (spp undefined)
1755 */
1756 static int
si_cachea_get(struct inode * ip,si_t * spi,si_t ** spp)1757 si_cachea_get(struct inode *ip, si_t *spi, si_t **spp)
1758 {
1759 si_t *sp;
1760
1761 spi->s_dev = ip->i_dev;
1762 spi->s_signature = si_signature(spi);
1763 rw_enter(&si_cache_lock, RW_READER);
1764 loop:
1765 for (sp = si_cachea[SI_HASH(spi->s_signature)]; sp; sp = sp->s_next) {
1766 if (sp->s_signature == spi->s_signature &&
1767 sp->s_dev == spi->s_dev &&
1768 sp->s_use > 0 && /* deleting */
1769 sp->s_use <= (MAXLINK - 1) && /* Too many links */
1770 !si_cmp(sp, spi))
1771 break;
1772 }
1773
1774 if (sp == NULL) {
1775 /* Cache miss */
1776 rw_exit(&si_cache_lock);
1777 return (1);
1778 }
1779 /* Found it */
1780 rw_enter(&sp->s_lock, RW_WRITER);
1781 alldone:
1782 spi->s_shadow = sp->s_shadow; /* XXX For debugging */
1783 rw_exit(&si_cache_lock);
1784 *spp = sp;
1785 return (0);
1786 }
1787
1788 /*
1789 * Place an si structure in the si cache. May cause duplicates.
1790 *
1791 * Parameters:
1792 * sp - Ptr to the si struct to add to the cache.
1793 *
1794 * Returns: Nothing (void)
1795 */
1796 static void
si_cache_put(si_t * sp)1797 si_cache_put(si_t *sp)
1798 {
1799 si_t **tspp;
1800
1801 ASSERT(sp->s_fore == NULL);
1802 rw_enter(&si_cache_lock, RW_WRITER);
1803 if (!sp->s_signature)
1804 sp->s_signature = si_signature(sp);
1805 sp->s_flags |= SI_CACHED;
1806 sp->s_fore = NULL;
1807
1808 /* The 'by acl' chains */
1809 tspp = &si_cachea[SI_HASH(sp->s_signature)];
1810 sp->s_next = *tspp;
1811 *tspp = sp;
1812
1813 /* The 'by inode' chains */
1814 tspp = &si_cachei[SI_HASH(sp->s_shadow)];
1815 sp->s_forw = *tspp;
1816 *tspp = sp;
1817
1818 rw_exit(&si_cache_lock);
1819 }
1820
1821 /*
1822 * The sp passed in is a candidate for deletion from the cache. We acquire
1823 * the cache lock first, so no cache searches can be done. Then we search
1824 * for the acl in the cache, and if we find it we can lock it and check that
1825 * nobody else attached to it while we were acquiring the locks. If the acl
1826 * is in the cache and still has a zero reference count, then we remove it
1827 * from the cache and deallocate it. If the reference count is non-zero or
1828 * it is not found in the cache, then someone else attached to it or has
1829 * already freed it, so we just return.
1830 *
1831 * Parameters:
1832 * sp - Ptr to the sp struct which is the candicate for deletion.
1833 * signature - the signature for the acl for lookup in the hash table
1834 *
1835 * Returns: Nothing (void)
1836 */
1837 void
si_cache_del(si_t * sp,int signature)1838 si_cache_del(si_t *sp, int signature)
1839 {
1840 si_t **tspp;
1841 int hash;
1842 int foundacl = 0;
1843
1844 /*
1845 * Unlink & free the sp from the other queues, then destroy it.
1846 * Search the 'by acl' chain first, then the 'by inode' chain
1847 * after the acl is locked.
1848 */
1849 rw_enter(&si_cache_lock, RW_WRITER);
1850 hash = SI_HASH(signature);
1851 for (tspp = &si_cachea[hash]; *tspp; tspp = &(*tspp)->s_next) {
1852 if (*tspp == sp) {
1853 /*
1854 * Wait to grab the acl lock until after the acl has
1855 * been found in the cache. Otherwise it might try to
1856 * grab a lock that has already been destroyed, or
1857 * delete an acl that has already been freed.
1858 */
1859 rw_enter(&sp->s_lock, RW_WRITER);
1860 /* See if someone else attached to it */
1861 if (sp->s_ref) {
1862 rw_exit(&sp->s_lock);
1863 rw_exit(&si_cache_lock);
1864 return;
1865 }
1866 ASSERT(sp->s_fore == NULL);
1867 ASSERT(sp->s_flags & SI_CACHED);
1868 foundacl = 1;
1869 *tspp = sp->s_next;
1870 break;
1871 }
1872 }
1873
1874 /*
1875 * If the acl was not in the cache, we assume another thread has
1876 * deleted it already. This could happen if another thread attaches to
1877 * the acl and then releases it after this thread has already found the
1878 * reference count to be zero but has not yet taken the cache lock.
1879 * Both threads end up seeing a reference count of zero, and call into
1880 * si_cache_del. See bug 4244827 for details on the race condition.
1881 */
1882 if (foundacl == 0) {
1883 rw_exit(&si_cache_lock);
1884 return;
1885 }
1886
1887 /* Now check the 'by inode' chain */
1888 hash = SI_HASH(sp->s_shadow);
1889 for (tspp = &si_cachei[hash]; *tspp; tspp = &(*tspp)->s_forw) {
1890 if (*tspp == sp) {
1891 *tspp = sp->s_forw;
1892 break;
1893 }
1894 }
1895
1896 /*
1897 * At this point, we can unlock everything because this si
1898 * is no longer in the cache, thus cannot be attached to.
1899 */
1900 rw_exit(&sp->s_lock);
1901 rw_exit(&si_cache_lock);
1902 sp->s_flags &= ~SI_CACHED;
1903 (void) ufs_si_free_mem(sp);
1904 }
1905
1906 /*
1907 * Alloc the hash buckets for the si cache & initialize
1908 * the unreferenced anchor and the cache lock.
1909 */
1910 void
si_cache_init(void)1911 si_cache_init(void)
1912 {
1913 rw_init(&si_cache_lock, NULL, RW_DEFAULT, NULL);
1914
1915 /* The 'by acl' headers */
1916 si_cachea = kmem_zalloc(si_cachecnt * sizeof (si_t *), KM_SLEEP);
1917 /* The 'by inode' headers */
1918 si_cachei = kmem_zalloc(si_cachecnt * sizeof (si_t *), KM_SLEEP);
1919 }
1920
1921 /*
1922 * aclcksum takes an acl and generates a checksum. It takes as input
1923 * the acl to start at.
1924 *
1925 * s_aclp - pointer to starting acl
1926 *
1927 * returns checksum
1928 */
1929 static int
aclcksum(ufs_ic_acl_t * s_aclp)1930 aclcksum(ufs_ic_acl_t *s_aclp)
1931 {
1932 ufs_ic_acl_t *aclp;
1933 int signature = 0;
1934 for (aclp = s_aclp; aclp; aclp = aclp->acl_ic_next) {
1935 signature += aclp->acl_ic_perm;
1936 signature += aclp->acl_ic_who;
1937 }
1938 return (signature);
1939 }
1940
1941 /*
1942 * Generate a unique signature for an si structure. Used by the
1943 * search routine si_cachea_get() to quickly identify candidates
1944 * prior to calling si_cmp().
1945 * Parameters:
1946 * sp - Ptr to the si struct to generate the signature for.
1947 *
1948 * Returns: A signature for the si struct (really a checksum)
1949 */
1950 static int
si_signature(si_t * sp)1951 si_signature(si_t *sp)
1952 {
1953 int signature = sp->s_dev;
1954
1955 signature += aclcksum(sp->aowner) + aclcksum(sp->agroup) +
1956 aclcksum(sp->aother) + aclcksum(sp->ausers) +
1957 aclcksum(sp->agroups) + aclcksum(sp->downer) +
1958 aclcksum(sp->dgroup) + aclcksum(sp->dother) +
1959 aclcksum(sp->dusers) + aclcksum(sp->dgroups);
1960 if (sp->aclass.acl_ismask)
1961 signature += sp->aclass.acl_maskbits;
1962 if (sp->dclass.acl_ismask)
1963 signature += sp->dclass.acl_maskbits;
1964
1965 return (signature);
1966 }
1967
1968 /*
1969 * aclcmp compares to acls to see if they are identical.
1970 *
1971 * sp1 is source
1972 * sp2 is sourceb
1973 *
1974 * returns 0 if equal and 1 if not equal
1975 */
1976 static int
aclcmp(ufs_ic_acl_t * aclin1p,ufs_ic_acl_t * aclin2p)1977 aclcmp(ufs_ic_acl_t *aclin1p, ufs_ic_acl_t *aclin2p)
1978 {
1979 ufs_ic_acl_t *aclp1;
1980 ufs_ic_acl_t *aclp2;
1981
1982 /*
1983 * if the starting pointers are equal then they are equal so
1984 * just return.
1985 */
1986 if (aclin1p == aclin2p)
1987 return (0);
1988 /*
1989 * check element by element
1990 */
1991 for (aclp1 = aclin1p, aclp2 = aclin2p; aclp1 && aclp2;
1992 aclp1 = aclp1->acl_ic_next, aclp2 = aclp2->acl_ic_next) {
1993 if (aclp1->acl_ic_perm != aclp2->acl_ic_perm ||
1994 aclp1->acl_ic_who != aclp2->acl_ic_who)
1995 return (1);
1996 }
1997 /*
1998 * both must be zero (at the end of the acl)
1999 */
2000 if (aclp1 || aclp2)
2001 return (1);
2002
2003 return (0);
2004 }
2005
2006 /*
2007 * Do extensive, field-by-field compare of two si structures. Returns
2008 * 0 if they are exactly identical, 1 otherwise.
2009 *
2010 * Paramters:
2011 * sp1 - Ptr to 1st si struct
2012 * sp2 - Ptr to 2nd si struct
2013 *
2014 * Returns:
2015 * 0 - Not identical
2016 * 1 - Identical
2017 */
2018 static int
si_cmp(si_t * sp1,si_t * sp2)2019 si_cmp(si_t *sp1, si_t *sp2)
2020 {
2021 if (sp1->s_dev != sp2->s_dev)
2022 return (1);
2023 if (aclcmp(sp1->aowner, sp2->aowner) ||
2024 aclcmp(sp1->agroup, sp2->agroup) ||
2025 aclcmp(sp1->aother, sp2->aother) ||
2026 aclcmp(sp1->ausers, sp2->ausers) ||
2027 aclcmp(sp1->agroups, sp2->agroups) ||
2028 aclcmp(sp1->downer, sp2->downer) ||
2029 aclcmp(sp1->dgroup, sp2->dgroup) ||
2030 aclcmp(sp1->dother, sp2->dother) ||
2031 aclcmp(sp1->dusers, sp2->dusers) ||
2032 aclcmp(sp1->dgroups, sp2->dgroups))
2033 return (1);
2034 if (sp1->aclass.acl_ismask != sp2->aclass.acl_ismask)
2035 return (1);
2036 if (sp1->dclass.acl_ismask != sp2->dclass.acl_ismask)
2037 return (1);
2038 if (sp1->aclass.acl_ismask &&
2039 sp1->aclass.acl_maskbits != sp2->aclass.acl_maskbits)
2040 return (1);
2041 if (sp1->dclass.acl_ismask &&
2042 sp1->dclass.acl_maskbits != sp2->dclass.acl_maskbits)
2043 return (1);
2044
2045 return (0);
2046 }
2047
2048 /*
2049 * Remove all acls associated with a device. All acls must have
2050 * a reference count of zero.
2051 *
2052 * inputs:
2053 * device - device to remove from the cache
2054 *
2055 * outputs:
2056 * none
2057 */
2058 void
ufs_si_cache_flush(dev_t dev)2059 ufs_si_cache_flush(dev_t dev)
2060 {
2061 si_t *tsp, **tspp;
2062 int i;
2063
2064 rw_enter(&si_cache_lock, RW_WRITER);
2065 for (i = 0; i < si_cachecnt; i++) {
2066 tspp = &si_cachea[i];
2067 while (*tspp) {
2068 if ((*tspp)->s_dev == dev) {
2069 *tspp = (*tspp)->s_next;
2070 } else {
2071 tspp = &(*tspp)->s_next;
2072 }
2073 }
2074 }
2075 for (i = 0; i < si_cachecnt; i++) {
2076 tspp = &si_cachei[i];
2077 while (*tspp) {
2078 if ((*tspp)->s_dev == dev) {
2079 tsp = *tspp;
2080 *tspp = (*tspp)->s_forw;
2081 tsp->s_flags &= ~SI_CACHED;
2082 ufs_si_free_mem(tsp);
2083 } else {
2084 tspp = &(*tspp)->s_forw;
2085 }
2086 }
2087 }
2088 rw_exit(&si_cache_lock);
2089 }
2090
2091 /*
2092 * ufs_si_del is used to unhook a sp from a inode in memory
2093 *
2094 * ip is the inode to remove the sp from.
2095 */
2096 void
ufs_si_del(struct inode * ip)2097 ufs_si_del(struct inode *ip)
2098 {
2099 si_t *sp = ip->i_ufs_acl;
2100 int refcnt;
2101 int signature;
2102
2103 if (sp) {
2104 rw_enter(&sp->s_lock, RW_WRITER);
2105 refcnt = --sp->s_ref;
2106 signature = sp->s_signature;
2107 ASSERT(sp->s_ref >= 0 && sp->s_ref <= sp->s_use);
2108 rw_exit(&sp->s_lock);
2109 if (refcnt == 0)
2110 si_cache_del(sp, signature);
2111 ip->i_ufs_acl = NULL;
2112 }
2113 }
2114