Home | History | Annotate | Line # | Download | only in kern
kern_veriexec.c revision 1.1.2.2
      1 /*	$NetBSD: kern_veriexec.c,v 1.1.2.2 2015/04/06 15:18:20 skrll Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2005, 2006 Elad Efrat <elad (at) NetBSD.org>
      5  * Copyright (c) 2005, 2006 Brett Lymn <blymn (at) NetBSD.org>
      6  * All rights reserved.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  * 3. The name of the authors may not be used to endorse or promote products
     17  *    derived from this software without specific prior written permission.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
     20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     22  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
     23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     29  */
     30 
     31 #include <sys/cdefs.h>
     32 __KERNEL_RCSID(0, "$NetBSD: kern_veriexec.c,v 1.1.2.2 2015/04/06 15:18:20 skrll Exp $");
     33 
     34 #include "opt_veriexec.h"
     35 
     36 #include <sys/param.h>
     37 #include <sys/mount.h>
     38 #include <sys/kmem.h>
     39 #include <sys/vnode.h>
     40 #include <sys/namei.h>
     41 #include <sys/exec.h>
     42 #include <sys/once.h>
     43 #include <sys/proc.h>
     44 #include <sys/rwlock.h>
     45 #include <sys/syslog.h>
     46 #include <sys/sysctl.h>
     47 #include <sys/inttypes.h>
     48 #include <sys/verified_exec.h>
     49 #if defined(__FreeBSD__)
     50 # include <sys/systm.h>
     51 # include <sys/imgact.h>
     52 # include <crypto/sha1.h>
     53 # include <crypto/sha2/sha2.h>
     54 # include <crypto/ripemd160/rmd160.h>
     55 #else
     56 # include <sys/sha1.h>
     57 # include <sys/sha2.h>
     58 # include <sys/rmd160.h>
     59 #endif
     60 #include <sys/md5.h>
     61 #include <uvm/uvm_extern.h>
     62 #include <sys/fileassoc.h>
     63 #include <sys/kauth.h>
     64 #include <sys/conf.h>
     65 #include <miscfs/specfs/specdev.h>
     66 #include <prop/proplib.h>
     67 #include <sys/fcntl.h>
     68 
     69 /* Readable values for veriexec_file_report(). */
     70 #define	REPORT_ALWAYS		0x01	/* Always print */
     71 #define	REPORT_VERBOSE		0x02	/* Print when verbose >= 1 */
     72 #define	REPORT_DEBUG		0x04	/* Print when verbose >= 2 (debug) */
     73 #define	REPORT_PANIC		0x08	/* Call panic() */
     74 #define	REPORT_ALARM		0x10	/* Alarm - also print pid/uid/.. */
     75 #define	REPORT_LOGMASK		(REPORT_ALWAYS|REPORT_VERBOSE|REPORT_DEBUG)
     76 
     77 /* state of locking for veriexec_file_verify */
     78 #define VERIEXEC_UNLOCKED	0x00	/* Nothing locked, callee does it */
     79 #define VERIEXEC_LOCKED		0x01	/* Global op lock held */
     80 
     81 
     82 #define VERIEXEC_RW_UPGRADE(lock)	while((rw_tryupgrade(lock)) == 0){};
     83 
     84 struct veriexec_fpops {
     85 	const char *type;
     86 	size_t hash_len;
     87 	size_t context_size;
     88 	veriexec_fpop_init_t init;
     89 	veriexec_fpop_update_t update;
     90 	veriexec_fpop_final_t final;
     91 	LIST_ENTRY(veriexec_fpops) entries;
     92 };
     93 
     94 /* Veriexec per-file entry data. */
     95 struct veriexec_file_entry {
     96 	krwlock_t lock;				/* r/w lock */
     97 	u_char *filename;			/* File name. */
     98 	u_char type;				/* Entry type. */
     99 	u_char status;				/* Evaluation status. */
    100 	u_char page_fp_status;			/* Per-page FP status. */
    101 	u_char *fp;				/* Fingerprint. */
    102 	void *page_fp;				/* Per-page fingerprints */
    103 	size_t npages;			    	/* Number of pages. */
    104 	size_t last_page_size;			/* To support < PAGE_SIZE */
    105 	struct veriexec_fpops *ops;		/* Fingerprint ops vector*/
    106 	size_t filename_len;			/* Length of filename. */
    107 };
    108 
    109 /* Veriexec per-table data. */
    110 struct veriexec_table_entry {
    111 	uint64_t vte_count;			/* Number of Veriexec entries. */
    112 	const struct sysctlnode *vte_node;
    113 };
    114 
    115 static int veriexec_verbose;
    116 static int veriexec_strict;
    117 static int veriexec_bypass = 1;
    118 
    119 static char *veriexec_fp_names = NULL;
    120 static size_t veriexec_name_max = 0;
    121 
    122 static const struct sysctlnode *veriexec_count_node;
    123 
    124 static fileassoc_t veriexec_hook;
    125 static specificdata_key_t veriexec_mountspecific_key;
    126 
    127 static LIST_HEAD(, veriexec_fpops) veriexec_fpops_list =
    128 	LIST_HEAD_INITIALIZER(veriexec_fpops_list);
    129 
    130 static int veriexec_raw_cb(kauth_cred_t, kauth_action_t, void *,
    131     void *, void *, void *, void *);
    132 static struct veriexec_fpops *veriexec_fpops_lookup(const char *);
    133 static void veriexec_file_free(struct veriexec_file_entry *);
    134 
    135 static unsigned int veriexec_tablecount = 0;
    136 
    137 /*
    138  * Veriexec operations global lock - most ops hold this as a read
    139  * lock, it is upgraded to a write lock when destroying veriexec file
    140  * table entries.
    141  */
    142 static krwlock_t veriexec_op_lock;
    143 
    144 /*
    145  * Sysctl helper routine for Veriexec.
    146  */
    147 static int
    148 sysctl_kern_veriexec_algorithms(SYSCTLFN_ARGS)
    149 {
    150 	size_t len;
    151 	int error;
    152 	const char *p;
    153 
    154 	if (newp != NULL)
    155 		return EPERM;
    156 
    157 	if (namelen != 0)
    158 		return EINVAL;
    159 
    160 	p = veriexec_fp_names == NULL ? "" : veriexec_fp_names;
    161 
    162 	len = strlen(p) + 1;
    163 
    164 	if (*oldlenp < len && oldp)
    165 		return ENOMEM;
    166 
    167 	if (oldp && (error = copyout(p, oldp, len)) != 0)
    168 		return error;
    169 
    170 	*oldlenp = len;
    171 	return 0;
    172 }
    173 
    174 static int
    175 sysctl_kern_veriexec_strict(SYSCTLFN_ARGS)
    176 {
    177 	struct sysctlnode node;
    178 	int error, newval;
    179 
    180 	node = *rnode;
    181 	node.sysctl_data = &newval;
    182 
    183 	newval = veriexec_strict;
    184 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
    185 	if (error || newp == NULL)
    186 		return error;
    187 
    188 	if (newval < veriexec_strict)
    189 		return EPERM;
    190 
    191 	veriexec_strict = newval;
    192 
    193 	return 0;
    194 }
    195 
    196 SYSCTL_SETUP(sysctl_kern_veriexec_setup, "sysctl kern.veriexec setup")
    197 {
    198 	const struct sysctlnode *rnode = NULL;
    199 
    200 	sysctl_createv(clog, 0, NULL, &rnode,
    201 		       CTLFLAG_PERMANENT,
    202 		       CTLTYPE_NODE, "veriexec",
    203 		       SYSCTL_DESCR("Veriexec"),
    204 		       NULL, 0, NULL, 0,
    205 		       CTL_KERN, CTL_CREATE, CTL_EOL);
    206 
    207 	sysctl_createv(clog, 0, &rnode, NULL,
    208 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    209 		       CTLTYPE_INT, "verbose",
    210 		       SYSCTL_DESCR("Veriexec verbose level"),
    211 		       NULL, 0, &veriexec_verbose, 0,
    212 		       CTL_CREATE, CTL_EOL);
    213 	sysctl_createv(clog, 0, &rnode, NULL,
    214 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    215 		       CTLTYPE_INT, "strict",
    216 		       SYSCTL_DESCR("Veriexec strict level"),
    217 		       sysctl_kern_veriexec_strict, 0, NULL, 0,
    218 		       CTL_CREATE, CTL_EOL);
    219 	sysctl_createv(clog, 0, &rnode, NULL,
    220 		       CTLFLAG_PERMANENT,
    221 		       CTLTYPE_STRING, "algorithms",
    222 		       SYSCTL_DESCR("Veriexec supported hashing "
    223 				    "algorithms"),
    224 		       sysctl_kern_veriexec_algorithms, 0, NULL, 0,
    225 		       CTL_CREATE, CTL_EOL);
    226 	sysctl_createv(clog, 0, &rnode, &veriexec_count_node,
    227 		       CTLFLAG_PERMANENT,
    228 		       CTLTYPE_NODE, "count",
    229 		       SYSCTL_DESCR("Number of fingerprints on mount(s)"),
    230 		       NULL, 0, NULL, 0,
    231 		       CTL_CREATE, CTL_EOL);
    232 }
    233 
    234 /*
    235  * Add ops to the fignerprint ops vector list.
    236  */
    237 int
    238 veriexec_fpops_add(const char *fp_type, size_t hash_len, size_t ctx_size,
    239     veriexec_fpop_init_t init, veriexec_fpop_update_t update,
    240     veriexec_fpop_final_t final)
    241 {
    242 	struct veriexec_fpops *ops;
    243 
    244 	/* Sanity check all parameters. */
    245 	if ((fp_type == NULL) || (hash_len == 0) || (ctx_size == 0) ||
    246 	    (init == NULL) || (update == NULL) || (final == NULL))
    247 		return (EFAULT);
    248 
    249 	if (veriexec_fpops_lookup(fp_type) != NULL)
    250 		return (EEXIST);
    251 
    252 	ops = kmem_alloc(sizeof(*ops), KM_SLEEP);
    253 
    254 	ops->type = fp_type;
    255 	ops->hash_len = hash_len;
    256 	ops->context_size = ctx_size;
    257 	ops->init = init;
    258 	ops->update = update;
    259 	ops->final = final;
    260 
    261 	LIST_INSERT_HEAD(&veriexec_fpops_list, ops, entries);
    262 
    263 	/*
    264 	 * If we don't have space for any names, allocate enough for six
    265 	 * which should be sufficient. (it's also enough for all algorithms
    266 	 * we can support at the moment)
    267 	 */
    268 	if (veriexec_fp_names == NULL) {
    269 		veriexec_name_max = 64;
    270 		veriexec_fp_names = kmem_zalloc(veriexec_name_max, KM_SLEEP);
    271 	}
    272 
    273 	/*
    274 	 * If we're running out of space for storing supported algorithms,
    275 	 * extend the buffer with space for four names.
    276 	 */
    277 	while (veriexec_name_max - (strlen(veriexec_fp_names) + 1) <
    278 	    strlen(fp_type)) {
    279 		char *newp;
    280 		unsigned int new_max;
    281 
    282 		/* Add space for four algorithm names. */
    283 		new_max = veriexec_name_max + 64;
    284 		newp = kmem_zalloc(new_max, KM_SLEEP);
    285 		strlcpy(newp, veriexec_fp_names, new_max);
    286 		kmem_free(veriexec_fp_names, veriexec_name_max);
    287 		veriexec_fp_names = newp;
    288 		veriexec_name_max = new_max;
    289 	}
    290 
    291 	if (*veriexec_fp_names != '\0')
    292 		strlcat(veriexec_fp_names, " ", veriexec_name_max);
    293 
    294 	strlcat(veriexec_fp_names, fp_type, veriexec_name_max);
    295 
    296 	return (0);
    297 }
    298 
    299 static void
    300 veriexec_mountspecific_dtor(void *v)
    301 {
    302 	struct veriexec_table_entry *vte = v;
    303 
    304 	if (vte == NULL) {
    305 		return;
    306 	}
    307 	sysctl_free(__UNCONST(vte->vte_node));
    308 	veriexec_tablecount--;
    309 	kmem_free(vte, sizeof(*vte));
    310 }
    311 
    312 static int
    313 veriexec_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
    314     void *arg0, void *arg1, void *arg2, void *arg3)
    315 {
    316 	int result;
    317 	enum kauth_system_req req;
    318 
    319 	if (action != KAUTH_SYSTEM_VERIEXEC)
    320 		return KAUTH_RESULT_DEFER;
    321 
    322 	result = KAUTH_RESULT_DEFER;
    323 	req = (enum kauth_system_req)arg0;
    324 
    325 	if (req == KAUTH_REQ_SYSTEM_VERIEXEC_MODIFY &&
    326 	    veriexec_strict > VERIEXEC_LEARNING) {
    327 		log(LOG_WARNING, "Veriexec: Strict mode, modifying "
    328 		    "tables not permitted.\n");
    329 
    330 		result = KAUTH_RESULT_DENY;
    331 	}
    332 
    333 	return result;
    334 }
    335 
    336 /*
    337  * Initialise Veriexec.
    338  */
    339 void
    340 veriexec_init(void)
    341 {
    342 	int error;
    343 
    344 	/* Register a fileassoc for Veriexec. */
    345 	error = fileassoc_register("veriexec",
    346 	    (fileassoc_cleanup_cb_t)veriexec_file_free, &veriexec_hook);
    347 	if (error)
    348 		panic("Veriexec: Can't register fileassoc: error=%d", error);
    349 
    350 	/* Register listener to handle raw disk access. */
    351 	if (kauth_listen_scope(KAUTH_SCOPE_DEVICE, veriexec_raw_cb, NULL) ==
    352 	    NULL)
    353 		panic("Veriexec: Can't listen on device scope");
    354 
    355 	error = mount_specific_key_create(&veriexec_mountspecific_key,
    356 	    veriexec_mountspecific_dtor);
    357 	if (error)
    358 		panic("Veriexec: Can't create mountspecific key");
    359 
    360 	if (kauth_listen_scope(KAUTH_SCOPE_SYSTEM, veriexec_listener_cb,
    361 	    NULL) == NULL)
    362 		panic("Veriexec: Can't listen on system scope");
    363 
    364 	rw_init(&veriexec_op_lock);
    365 
    366 #define	FPOPS_ADD(a, b, c, d, e, f)	\
    367 	veriexec_fpops_add(a, b, c, (veriexec_fpop_init_t)d, \
    368 	 (veriexec_fpop_update_t)e, (veriexec_fpop_final_t)f)
    369 
    370 #ifdef VERIFIED_EXEC_FP_RMD160
    371 	FPOPS_ADD("RMD160", RMD160_DIGEST_LENGTH, sizeof(RMD160_CTX),
    372 	    RMD160Init, RMD160Update, RMD160Final);
    373 #endif /* VERIFIED_EXEC_FP_RMD160 */
    374 
    375 #ifdef VERIFIED_EXEC_FP_SHA256
    376 	FPOPS_ADD("SHA256", SHA256_DIGEST_LENGTH, sizeof(SHA256_CTX),
    377 	    SHA256_Init, SHA256_Update, SHA256_Final);
    378 #endif /* VERIFIED_EXEC_FP_SHA256 */
    379 
    380 #ifdef VERIFIED_EXEC_FP_SHA384
    381 	FPOPS_ADD("SHA384", SHA384_DIGEST_LENGTH, sizeof(SHA384_CTX),
    382 	    SHA384_Init, SHA384_Update, SHA384_Final);
    383 #endif /* VERIFIED_EXEC_FP_SHA384 */
    384 
    385 #ifdef VERIFIED_EXEC_FP_SHA512
    386 	FPOPS_ADD("SHA512", SHA512_DIGEST_LENGTH, sizeof(SHA512_CTX),
    387 	    SHA512_Init, SHA512_Update, SHA512_Final);
    388 #endif /* VERIFIED_EXEC_FP_SHA512 */
    389 
    390 #ifdef VERIFIED_EXEC_FP_SHA1
    391 	FPOPS_ADD("SHA1", SHA1_DIGEST_LENGTH, sizeof(SHA1_CTX),
    392 	    SHA1Init, SHA1Update, SHA1Final);
    393 #endif /* VERIFIED_EXEC_FP_SHA1 */
    394 
    395 #ifdef VERIFIED_EXEC_FP_MD5
    396 	FPOPS_ADD("MD5", MD5_DIGEST_LENGTH, sizeof(MD5_CTX),
    397 	    MD5Init, MD5Update, MD5Final);
    398 #endif /* VERIFIED_EXEC_FP_MD5 */
    399 
    400 #undef FPOPS_ADD
    401 }
    402 
    403 static struct veriexec_fpops *
    404 veriexec_fpops_lookup(const char *name)
    405 {
    406 	struct veriexec_fpops *ops;
    407 
    408 	if (name == NULL)
    409 		return (NULL);
    410 
    411 	LIST_FOREACH(ops, &veriexec_fpops_list, entries) {
    412 		if (strcasecmp(name, ops->type) == 0)
    413 			return (ops);
    414 	}
    415 
    416 	return (NULL);
    417 }
    418 
    419 /*
    420  * Calculate fingerprint. Information on hash length and routines used is
    421  * extracted from veriexec_hash_list according to the hash type.
    422  *
    423  * NOTE: vfe is assumed to be locked for writing on entry.
    424  */
    425 static int
    426 veriexec_fp_calc(struct lwp *l, struct vnode *vp, int lock_state,
    427     struct veriexec_file_entry *vfe, u_char *fp)
    428 {
    429 	struct vattr va;
    430 	void *ctx, *page_ctx;
    431 	u_char *buf, *page_fp;
    432 	off_t offset, len;
    433 	size_t resid, npages;
    434 	int error, do_perpage, pagen;
    435 
    436 	if (lock_state == VERIEXEC_UNLOCKED)
    437 		vn_lock(vp, LK_SHARED | LK_RETRY);
    438 	error = VOP_GETATTR(vp, &va, l->l_cred);
    439 	if (lock_state == VERIEXEC_UNLOCKED)
    440 		VOP_UNLOCK(vp);
    441 	if (error)
    442 		return (error);
    443 
    444 #ifdef notyet /* XXX - for now */
    445 	if ((vfe->type & VERIEXEC_UNTRUSTED) &&
    446 	    (vfe->page_fp_status == PAGE_FP_NONE))
    447 		do_perpage = 1;
    448 	else
    449 #endif  /* notyet */
    450 		do_perpage = 0;
    451 
    452 	ctx = kmem_alloc(vfe->ops->context_size, KM_SLEEP);
    453 	buf = kmem_alloc(PAGE_SIZE, KM_SLEEP);
    454 
    455 	page_ctx = NULL;
    456 	page_fp = NULL;
    457 	npages = 0;
    458 	if (do_perpage) {
    459 		npages = (va.va_size >> PAGE_SHIFT) + 1;
    460 		page_fp = kmem_alloc(vfe->ops->hash_len * npages, KM_SLEEP);
    461 		vfe->page_fp = page_fp;
    462 		page_ctx = kmem_alloc(vfe->ops->context_size, KM_SLEEP);
    463 	}
    464 
    465 	(vfe->ops->init)(ctx);
    466 
    467 	len = 0;
    468 	error = 0;
    469 	pagen = 0;
    470 	for (offset = 0; offset < va.va_size; offset += PAGE_SIZE) {
    471 		len = ((va.va_size - offset) < PAGE_SIZE) ?
    472 		    (va.va_size - offset) : PAGE_SIZE;
    473 
    474 		error = vn_rdwr(UIO_READ, vp, buf, len, offset,
    475 				UIO_SYSSPACE,
    476 				((lock_state == VERIEXEC_LOCKED)?
    477 				 IO_NODELOCKED : 0),
    478 				l->l_cred, &resid, NULL);
    479 
    480 		if (error) {
    481 			if (do_perpage) {
    482 				kmem_free(vfe->page_fp,
    483 				    vfe->ops->hash_len * npages);
    484 				vfe->page_fp = NULL;
    485 			}
    486 
    487 			goto bad;
    488 		}
    489 
    490 		(vfe->ops->update)(ctx, buf, (unsigned int) len);
    491 
    492 		if (do_perpage) {
    493 			(vfe->ops->init)(page_ctx);
    494 			(vfe->ops->update)(page_ctx, buf, (unsigned int)len);
    495 			(vfe->ops->final)(page_fp, page_ctx);
    496 
    497 			if (veriexec_verbose >= 2) {
    498 				int i;
    499 
    500 				printf("hash for page %d: ", pagen);
    501 				for (i = 0; i < vfe->ops->hash_len; i++)
    502 					printf("%02x", page_fp[i]);
    503 				printf("\n");
    504 			}
    505 
    506 			page_fp += vfe->ops->hash_len;
    507 			pagen++;
    508 		}
    509 
    510 		if (len != PAGE_SIZE)
    511 			break;
    512 	}
    513 
    514 	(vfe->ops->final)(fp, ctx);
    515 
    516 	if (do_perpage) {
    517 		vfe->last_page_size = len;
    518 		vfe->page_fp_status = PAGE_FP_READY;
    519 		vfe->npages = npages;
    520 	}
    521 
    522 bad:
    523 	if (do_perpage)
    524 		kmem_free(page_ctx, vfe->ops->context_size);
    525 
    526 	kmem_free(ctx, vfe->ops->context_size);
    527 	kmem_free(buf, PAGE_SIZE);
    528 
    529 	return (error);
    530 }
    531 
    532 /* Compare two fingerprints of the same type. */
    533 static int
    534 veriexec_fp_cmp(struct veriexec_fpops *ops, u_char *fp1, u_char *fp2)
    535 {
    536 	if (veriexec_verbose >= 2) {
    537 		int i;
    538 
    539 		printf("comparing hashes...\n");
    540 		printf("fp1: ");
    541 		for (i = 0; i < ops->hash_len; i++) {
    542 			printf("%02x", fp1[i]);
    543 		}
    544 		printf("\nfp2: ");
    545 		for (i = 0; i < ops->hash_len; i++) {
    546 			printf("%02x", fp2[i]);
    547 		}
    548 		printf("\n");
    549 	}
    550 
    551 	return (memcmp(fp1, fp2, ops->hash_len));
    552 }
    553 
    554 static struct veriexec_table_entry *
    555 veriexec_table_lookup(struct mount *mp)
    556 {
    557 	/* XXX: From raidframe init */
    558 	if (mp == NULL)
    559 		return NULL;
    560 
    561 	return mount_getspecific(mp, veriexec_mountspecific_key);
    562 }
    563 
    564 static struct veriexec_file_entry *
    565 veriexec_get(struct vnode *vp)
    566 {
    567 	return (fileassoc_lookup(vp, veriexec_hook));
    568 }
    569 
    570 bool
    571 veriexec_lookup(struct vnode *vp)
    572 {
    573 	return (veriexec_get(vp) == NULL ? false : true);
    574 }
    575 
    576 /*
    577  * Routine for maintaining mostly consistent message formats in Veriexec.
    578  */
    579 static void
    580 veriexec_file_report(struct veriexec_file_entry *vfe, const u_char *msg,
    581     const u_char *filename, struct lwp *l, int f)
    582 {
    583 	if (vfe != NULL && vfe->filename != NULL)
    584 		filename = vfe->filename;
    585 
    586 	if (filename == NULL)
    587 		return;
    588 
    589 	if (((f & REPORT_LOGMASK) >> 1) <= veriexec_verbose) {
    590 		if (!(f & REPORT_ALARM) || (l == NULL))
    591 			log(LOG_NOTICE, "Veriexec: %s [%s]\n", msg,
    592 			    filename);
    593 		else
    594 			log(LOG_ALERT, "Veriexec: %s [%s, prog=%s pid=%u, "
    595 			    "uid=%u, gid=%u]\n", msg, filename,
    596 			    l->l_proc->p_comm, l->l_proc->p_pid,
    597 			    kauth_cred_getuid(l->l_cred),
    598 			    kauth_cred_getgid(l->l_cred));
    599 	}
    600 
    601 	if (f & REPORT_PANIC)
    602 		panic("Veriexec: Unrecoverable error.");
    603 }
    604 
    605 /*
    606  * Verify the fingerprint of the given file. If we're called directly from
    607  * sys_execve(), 'flag' will be VERIEXEC_DIRECT. If we're called from
    608  * exec_script(), 'flag' will be VERIEXEC_INDIRECT.  If we are called from
    609  * vn_open(), 'flag' will be VERIEXEC_FILE.
    610  *
    611  * NOTE: The veriexec file entry pointer (vfep) will be returned LOCKED
    612  *       on no error.
    613  */
    614 static int
    615 veriexec_file_verify(struct lwp *l, struct vnode *vp, const u_char *name,
    616     int flag, int lockstate, struct veriexec_file_entry **vfep)
    617 {
    618 	struct veriexec_file_entry *vfe;
    619 	int error;
    620 
    621 #define VFE_NEEDS_EVAL(vfe) ((vfe->status == FINGERPRINT_NOTEVAL) || \
    622 			     (vfe->type & VERIEXEC_UNTRUSTED))
    623 
    624 	if (vfep != NULL)
    625 		*vfep = NULL;
    626 
    627 	if (vp->v_type != VREG)
    628 		return (0);
    629 
    630 	if (lockstate == VERIEXEC_UNLOCKED)
    631 		rw_enter(&veriexec_op_lock, RW_READER);
    632 
    633 	/* Lookup veriexec table entry, save pointer if requested. */
    634 	vfe = veriexec_get(vp);
    635 	if (vfep != NULL)
    636 		*vfep = vfe;
    637 	if (vfe == NULL)
    638 		goto out;
    639 
    640 	error = 0;
    641 
    642 	/*
    643 	 * Grab the lock for the entry, if we need to do an evaluation
    644 	 * then the lock is a write lock, after we have the write
    645 	 * lock, check if we really need it - some other thread may
    646 	 * have already done the work for us.
    647 	 */
    648 	if (VFE_NEEDS_EVAL(vfe)) {
    649 		rw_enter(&vfe->lock, RW_WRITER);
    650 		if (!VFE_NEEDS_EVAL(vfe))
    651 			rw_downgrade(&vfe->lock);
    652 	} else
    653 		rw_enter(&vfe->lock, RW_READER);
    654 
    655 	/* Evaluate fingerprint if needed. */
    656 	if (VFE_NEEDS_EVAL(vfe)) {
    657 		u_char *digest;
    658 
    659 		/* Calculate fingerprint for on-disk file. */
    660 		digest = kmem_zalloc(vfe->ops->hash_len, KM_SLEEP);
    661 
    662 		error = veriexec_fp_calc(l, vp, lockstate, vfe, digest);
    663 		if (error) {
    664 			veriexec_file_report(vfe, "Fingerprint calculation error.",
    665 			    name, NULL, REPORT_ALWAYS);
    666 			kmem_free(digest, vfe->ops->hash_len);
    667 			rw_exit(&vfe->lock);
    668 			if (lockstate == VERIEXEC_UNLOCKED)
    669 				rw_exit(&veriexec_op_lock);
    670 			return (error);
    671 		}
    672 
    673 		/* Compare fingerprint with loaded data. */
    674 		if (veriexec_fp_cmp(vfe->ops, vfe->fp, digest) == 0)
    675 			vfe->status = FINGERPRINT_VALID;
    676 		else
    677 			vfe->status = FINGERPRINT_NOMATCH;
    678 
    679 		kmem_free(digest, vfe->ops->hash_len);
    680 		rw_downgrade(&vfe->lock);
    681 	}
    682 
    683 	if (!(vfe->type & flag)) {
    684 		veriexec_file_report(vfe, "Incorrect access type.", name, l,
    685 		    REPORT_ALWAYS|REPORT_ALARM);
    686 
    687 		/* IPS mode: Enforce access type. */
    688 		if (veriexec_strict >= VERIEXEC_IPS) {
    689 			rw_exit(&vfe->lock);
    690 			if (lockstate == VERIEXEC_UNLOCKED)
    691 				rw_exit(&veriexec_op_lock);
    692 			return (EPERM);
    693 		}
    694 	}
    695 
    696  out:
    697 	/* No entry in the veriexec tables. */
    698 	if (vfe == NULL) {
    699 		veriexec_file_report(NULL, "No entry.", name,
    700 		    l, REPORT_VERBOSE);
    701 
    702 		if (lockstate == VERIEXEC_UNLOCKED)
    703 			rw_exit(&veriexec_op_lock);
    704 		/*
    705 		 * Lockdown mode: Deny access to non-monitored files.
    706 		 * IPS mode: Deny execution of non-monitored files.
    707 		 */
    708 		if ((veriexec_strict >= VERIEXEC_LOCKDOWN) ||
    709 		    ((veriexec_strict >= VERIEXEC_IPS) &&
    710 		     (flag != VERIEXEC_FILE)))
    711 			return (EPERM);
    712 
    713 		return (0);
    714 	}
    715 
    716         switch (vfe->status) {
    717 	case FINGERPRINT_NOTEVAL:
    718 		/* Should not happen. */
    719 		rw_exit(&vfe->lock);
    720 		if (lockstate == VERIEXEC_UNLOCKED)
    721 			rw_exit(&veriexec_op_lock);
    722 		veriexec_file_report(vfe, "Not-evaluated status "
    723 		    "post evaluation; inconsistency detected.", name,
    724 		    NULL, REPORT_ALWAYS|REPORT_PANIC);
    725 
    726 		/*NOTREACHED*/
    727 
    728 	case FINGERPRINT_VALID:
    729 		/* Valid fingerprint. */
    730 		veriexec_file_report(vfe, "Match.", name, NULL,
    731 		    REPORT_VERBOSE);
    732 
    733 		break;
    734 
    735 	case FINGERPRINT_NOMATCH:
    736 		/* Fingerprint mismatch. */
    737 		veriexec_file_report(vfe, "Mismatch.", name,
    738 		    NULL, REPORT_ALWAYS|REPORT_ALARM);
    739 
    740 		/* IDS mode: Deny access on fingerprint mismatch. */
    741 		if (veriexec_strict >= VERIEXEC_IDS) {
    742 			rw_exit(&vfe->lock);
    743 			error = EPERM;
    744 		}
    745 
    746 		break;
    747 
    748 	default:
    749 		/* Should never happen. */
    750 		rw_exit(&vfe->lock);
    751 		if (lockstate == VERIEXEC_UNLOCKED)
    752 			rw_exit(&veriexec_op_lock);
    753 		veriexec_file_report(vfe, "Invalid status "
    754 		    "post evaluation.", name, NULL, REPORT_ALWAYS|REPORT_PANIC);
    755         }
    756 
    757 	if (lockstate == VERIEXEC_UNLOCKED)
    758 		rw_exit(&veriexec_op_lock);
    759 	return (error);
    760 }
    761 
    762 int
    763 veriexec_verify(struct lwp *l, struct vnode *vp, const u_char *name, int flag,
    764     bool *found)
    765 {
    766 	struct veriexec_file_entry *vfe;
    767 	int r;
    768 
    769 	if (veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
    770 		return 0;
    771 
    772 	r = veriexec_file_verify(l, vp, name, flag, VERIEXEC_UNLOCKED, &vfe);
    773 
    774 	if ((r  == 0) && (vfe != NULL))
    775 		rw_exit(&vfe->lock);
    776 
    777 	if (found != NULL)
    778 		*found = (vfe != NULL) ? true : false;
    779 
    780 	return (r);
    781 }
    782 
    783 #ifdef notyet
    784 /*
    785  * Evaluate per-page fingerprints.
    786  */
    787 int
    788 veriexec_page_verify(struct veriexec_file_entry *vfe, struct vm_page *pg,
    789     size_t idx, struct lwp *l)
    790 {
    791 	void *ctx;
    792 	u_char *fp;
    793 	u_char *page_fp;
    794 	int error;
    795 	vaddr_t kva;
    796 
    797 	if (vfe->page_fp_status == PAGE_FP_NONE)
    798 		return (0);
    799 
    800 	if (vfe->page_fp_status == PAGE_FP_FAIL)
    801 		return (EPERM);
    802 
    803 	if (idx >= vfe->npages)
    804 		return (0);
    805 
    806 	ctx = kmem_alloc(vfe->ops->context_size, KM_SLEEP);
    807 	fp = kmem_alloc(vfe->ops->hash_len, KM_SLEEP);
    808 	kva = uvm_km_alloc(kernel_map, PAGE_SIZE, VM_PGCOLOR_BUCKET(pg),
    809 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
    810 	pmap_kenter_pa(kva, VM_PAGE_TO_PHYS(pg), VM_PROT_READ, 0);
    811 	pmap_update(pmap_kernel());
    812 
    813 	page_fp = (u_char *) vfe->page_fp + (vfe->ops->hash_len * idx);
    814 	(vfe->ops->init)(ctx);
    815 	(vfe->ops->update)(ctx, (void *) kva,
    816 			   ((vfe->npages - 1) == idx) ? vfe->last_page_size
    817 						      : PAGE_SIZE);
    818 	(vfe->ops->final)(fp, ctx);
    819 
    820 	pmap_kremove(kva, PAGE_SIZE);
    821 	pmap_update(pmap_kernel());
    822 	uvm_km_free(kernel_map, kva, PAGE_SIZE, UVM_KMF_VAONLY);
    823 
    824 	error = veriexec_fp_cmp(vfe->ops, page_fp, fp);
    825 	if (error) {
    826 		const char *msg;
    827 
    828 		if (veriexec_strict > VERIEXEC_LEARNING) {
    829 			msg = "Pages modified: Killing process.";
    830 		} else {
    831 			msg = "Pages modified.";
    832 			error = 0;
    833 		}
    834 
    835 		veriexec_file_report(msg, "[page_in]", l,
    836 		    REPORT_ALWAYS|REPORT_ALARM);
    837 
    838 		if (error) {
    839 			ksiginfo_t ksi;
    840 
    841 			KSI_INIT(&ksi);
    842 			ksi.ksi_signo = SIGKILL;
    843 			ksi.ksi_code = SI_NOINFO;
    844 			ksi.ksi_pid = l->l_proc->p_pid;
    845 			ksi.ksi_uid = 0;
    846 
    847 			kpsignal(l->l_proc, &ksi, NULL);
    848 		}
    849 	}
    850 
    851 	kmem_free(ctx, vfe->ops->context_size);
    852 	kmem_free(fp, vfe->ops->hash_len);
    853 
    854 	return (error);
    855 }
    856 #endif /* notyet */
    857 
    858 /*
    859  * Veriexec remove policy code.
    860  */
    861 int
    862 veriexec_removechk(struct lwp *l, struct vnode *vp, const char *pathbuf)
    863 {
    864 	struct veriexec_file_entry *vfe;
    865 	int error;
    866 
    867 	if (veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
    868 		return 0;
    869 
    870 	rw_enter(&veriexec_op_lock, RW_READER);
    871 
    872 	vfe = veriexec_get(vp);
    873 	rw_exit(&veriexec_op_lock);
    874 
    875 	if (vfe == NULL) {
    876 		/* Lockdown mode: Deny access to non-monitored files. */
    877 		if (veriexec_strict >= VERIEXEC_LOCKDOWN)
    878 			return (EPERM);
    879 
    880 		return (0);
    881 	}
    882 
    883 	veriexec_file_report(vfe, "Remove request.", pathbuf, l,
    884 	    REPORT_ALWAYS|REPORT_ALARM);
    885 
    886 	/* IDS mode: Deny removal of monitored files. */
    887 	if (veriexec_strict >= VERIEXEC_IDS)
    888 		error = EPERM;
    889 	else
    890 		error = veriexec_file_delete(l, vp);
    891 
    892 
    893 	return error;
    894 }
    895 
    896 /*
    897  * Veriexec rename policy.
    898  *
    899  * XXX: Once there's a way to hook after a successful rename, it would be
    900  * XXX: nice to update vfe->filename to the new name if it's not NULL and
    901  * XXX: the new name is absolute (ie., starts with a slash).
    902  */
    903 int
    904 veriexec_renamechk(struct lwp *l, struct vnode *fromvp, const char *fromname,
    905     struct vnode *tovp, const char *toname)
    906 {
    907 	struct veriexec_file_entry *vfe, *tvfe;
    908 
    909 	if (veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
    910 		return 0;
    911 
    912 	rw_enter(&veriexec_op_lock, RW_READER);
    913 
    914 	if (veriexec_strict >= VERIEXEC_LOCKDOWN) {
    915 		log(LOG_ALERT, "Veriexec: Preventing rename of `%s' to "
    916 		    "`%s', uid=%u, pid=%u: Lockdown mode.\n", fromname, toname,
    917 		    kauth_cred_geteuid(l->l_cred), l->l_proc->p_pid);
    918 
    919 		rw_exit(&veriexec_op_lock);
    920 		return (EPERM);
    921 	}
    922 
    923 	vfe = veriexec_get(fromvp);
    924 	tvfe = NULL;
    925 	if (tovp != NULL)
    926 		tvfe = veriexec_get(tovp);
    927 
    928 	if ((vfe != NULL) || (tvfe != NULL)) {
    929 		if (veriexec_strict >= VERIEXEC_IPS) {
    930 			log(LOG_ALERT, "Veriexec: Preventing rename of `%s' "
    931 			    "to `%s', uid=%u, pid=%u: IPS mode, %s "
    932 			    "monitored.\n", fromname, toname,
    933 			    kauth_cred_geteuid(l->l_cred),
    934 			    l->l_proc->p_pid, (vfe != NULL && tvfe != NULL) ?
    935 			    "files" : "file");
    936 
    937 			rw_exit(&veriexec_op_lock);
    938 			return (EPERM);
    939 		}
    940 
    941 		/*
    942 		 * Monitored file is renamed; filename no longer relevant.
    943 		 *
    944 		 * XXX: We could keep the buffer, and when (and if) updating the
    945 		 * XXX: filename post-rename, re-allocate it only if it's not
    946 		 * XXX: big enough for the new filename.
    947 		 */
    948 		if (vfe != NULL) {
    949 			/* XXXX get write lock on vfe here? */
    950 
    951 			VERIEXEC_RW_UPGRADE(&veriexec_op_lock);
    952 			/* once we have the op lock in write mode
    953 			 * there should be no locks on any file
    954 			 * entries so we can destroy the object.
    955 			 */
    956 
    957 			if (vfe->filename_len > 0)
    958 				kmem_free(vfe->filename, vfe->filename_len);
    959 
    960 			vfe->filename = NULL;
    961 			vfe->filename_len = 0;
    962 
    963 			rw_downgrade(&veriexec_op_lock);
    964 		}
    965 
    966 		log(LOG_NOTICE, "Veriexec: %s file `%s' renamed to "
    967 		    "%s file `%s', uid=%u, pid=%u.\n", (vfe != NULL) ?
    968 		    "Monitored" : "Non-monitored", fromname, (tvfe != NULL) ?
    969 		    "monitored" : "non-monitored", toname,
    970 		    kauth_cred_geteuid(l->l_cred), l->l_proc->p_pid);
    971 
    972 		rw_exit(&veriexec_op_lock);
    973 
    974 		/*
    975 		 * Monitored file is overwritten. Remove the entry.
    976 		 */
    977 		if (tvfe != NULL)
    978 			(void)veriexec_file_delete(l, tovp);
    979 
    980 	} else
    981 		rw_exit(&veriexec_op_lock);
    982 
    983 	return (0);
    984 }
    985 
    986 static void
    987 veriexec_file_free(struct veriexec_file_entry *vfe)
    988 {
    989 	if (vfe != NULL) {
    990 		if (vfe->fp != NULL)
    991 			kmem_free(vfe->fp, vfe->ops->hash_len);
    992 		if (vfe->page_fp != NULL)
    993 			kmem_free(vfe->page_fp, vfe->ops->hash_len);
    994 		if (vfe->filename != NULL)
    995 			kmem_free(vfe->filename, vfe->filename_len);
    996 		rw_destroy(&vfe->lock);
    997 		kmem_free(vfe, sizeof(*vfe));
    998 	}
    999 }
   1000 
   1001 static void
   1002 veriexec_file_purge(struct veriexec_file_entry *vfe, int have_lock)
   1003 {
   1004 	if (vfe == NULL)
   1005 		return;
   1006 
   1007 	if (have_lock == VERIEXEC_UNLOCKED)
   1008 		rw_enter(&vfe->lock, RW_WRITER);
   1009 	else
   1010 		VERIEXEC_RW_UPGRADE(&vfe->lock);
   1011 
   1012 	vfe->status = FINGERPRINT_NOTEVAL;
   1013 	if (have_lock == VERIEXEC_UNLOCKED)
   1014 		rw_exit(&vfe->lock);
   1015 	else
   1016 		rw_downgrade(&vfe->lock);
   1017 }
   1018 
   1019 static void
   1020 veriexec_file_purge_cb(struct veriexec_file_entry *vfe, void *cookie)
   1021 {
   1022 	veriexec_file_purge(vfe, VERIEXEC_UNLOCKED);
   1023 }
   1024 
   1025 /*
   1026  * Invalidate a Veriexec file entry.
   1027  * XXX: This should be updated when per-page fingerprints are added.
   1028  */
   1029 void
   1030 veriexec_purge(struct vnode *vp)
   1031 {
   1032 
   1033 	rw_enter(&veriexec_op_lock, RW_READER);
   1034 	veriexec_file_purge(veriexec_get(vp), VERIEXEC_UNLOCKED);
   1035 	rw_exit(&veriexec_op_lock);
   1036 }
   1037 
   1038 /*
   1039  * Enforce raw disk access policy.
   1040  *
   1041  * IDS mode: Invalidate fingerprints on a mount if it's opened for writing.
   1042  * IPS mode: Don't allow raw writing to disks we monitor.
   1043  * Lockdown mode: Don't allow raw writing to all disks.
   1044  *
   1045  * XXX: This is bogus. There's an obvious race condition between the time
   1046  * XXX: the disk is open for writing, in which an attacker can access a
   1047  * XXX: monitored file to get its signature cached again, and when the raw
   1048  * XXX: file is overwritten on disk.
   1049  * XXX:
   1050  * XXX: To solve this, we need something like the following:
   1051  * XXX:		open raw disk:
   1052  * XXX:		  - raise refcount,
   1053  * XXX:		  - invalidate fingerprints,
   1054  * XXX:		  - mark all entries for that disk with "no cache" flag
   1055  * XXX:
   1056  * XXX:		veriexec_verify:
   1057  * XXX:		  - if "no cache", don't cache evaluation result
   1058  * XXX:
   1059  * XXX:		close raw disk:
   1060  * XXX:		  - lower refcount,
   1061  * XXX:		  - if refcount == 0, remove "no cache" flag from all entries
   1062  */
   1063 static int
   1064 veriexec_raw_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
   1065     void *arg0, void *arg1, void *arg2, void *arg3)
   1066 {
   1067 	int result;
   1068 	enum kauth_device_req req;
   1069 	struct veriexec_table_entry *vte;
   1070 
   1071 	result = KAUTH_RESULT_DENY;
   1072 	req = (enum kauth_device_req)arg0;
   1073 
   1074 	switch (action) {
   1075 	case KAUTH_DEVICE_RAWIO_SPEC: {
   1076 		struct vnode *vp, *bvp;
   1077 		int error;
   1078 
   1079 		if (req == KAUTH_REQ_DEVICE_RAWIO_SPEC_READ) {
   1080 			result = KAUTH_RESULT_DEFER;
   1081 			break;
   1082 		}
   1083 
   1084 		vp = arg1;
   1085 		KASSERT(vp != NULL);
   1086 
   1087 		/* Handle /dev/mem and /dev/kmem. */
   1088 		if (iskmemvp(vp)) {
   1089 			if (veriexec_strict < VERIEXEC_IPS)
   1090 				result = KAUTH_RESULT_DEFER;
   1091 
   1092 			break;
   1093 		}
   1094 
   1095 		error = rawdev_mounted(vp, &bvp);
   1096 		if (error == EINVAL) {
   1097 			result = KAUTH_RESULT_DEFER;
   1098 			break;
   1099 		}
   1100 
   1101 		/*
   1102 		 * XXX: See vfs_mountedon() comment in rawdev_mounted().
   1103 		 */
   1104 		vte = veriexec_table_lookup(bvp->v_mount);
   1105 		if (vte == NULL) {
   1106 			result = KAUTH_RESULT_DEFER;
   1107 			break;
   1108 		}
   1109 
   1110 		switch (veriexec_strict) {
   1111 		case VERIEXEC_LEARNING:
   1112 		case VERIEXEC_IDS:
   1113 			result = KAUTH_RESULT_DEFER;
   1114 
   1115 			rw_enter(&veriexec_op_lock, RW_WRITER);
   1116 			fileassoc_table_run(bvp->v_mount, veriexec_hook,
   1117 			    (fileassoc_cb_t)veriexec_file_purge_cb, NULL);
   1118 			rw_exit(&veriexec_op_lock);
   1119 
   1120 			break;
   1121 		case VERIEXEC_IPS:
   1122 			result = KAUTH_RESULT_DENY;
   1123 			break;
   1124 		case VERIEXEC_LOCKDOWN:
   1125 			result = KAUTH_RESULT_DENY;
   1126 			break;
   1127 		}
   1128 
   1129 		break;
   1130 		}
   1131 
   1132 	case KAUTH_DEVICE_RAWIO_PASSTHRU:
   1133 		/* XXX What can we do here? */
   1134 		if (veriexec_strict < VERIEXEC_IPS)
   1135 			result = KAUTH_RESULT_DEFER;
   1136 
   1137 		break;
   1138 
   1139 	default:
   1140 		result = KAUTH_RESULT_DEFER;
   1141 		break;
   1142 	}
   1143 
   1144 	return (result);
   1145 }
   1146 
   1147 /*
   1148  * Create a new Veriexec table.
   1149  */
   1150 static struct veriexec_table_entry *
   1151 veriexec_table_add(struct lwp *l, struct mount *mp)
   1152 {
   1153 	struct veriexec_table_entry *vte;
   1154 	u_char buf[16];
   1155 
   1156 	vte = kmem_zalloc(sizeof(*vte), KM_SLEEP);
   1157 	mount_setspecific(mp, veriexec_mountspecific_key, vte);
   1158 
   1159 	snprintf(buf, sizeof(buf), "table%u", veriexec_tablecount++);
   1160 	sysctl_createv(NULL, 0, &veriexec_count_node, &vte->vte_node,
   1161 		       0, CTLTYPE_NODE, buf, NULL, NULL, 0, NULL,
   1162 		       0, CTL_CREATE, CTL_EOL);
   1163 
   1164 	sysctl_createv(NULL, 0, &vte->vte_node, NULL,
   1165 		       CTLFLAG_READONLY, CTLTYPE_STRING, "mntpt",
   1166 		       NULL, NULL, 0, mp->mnt_stat.f_mntonname,
   1167 		       0, CTL_CREATE, CTL_EOL);
   1168 	sysctl_createv(NULL, 0, &vte->vte_node, NULL,
   1169 		       CTLFLAG_READONLY, CTLTYPE_STRING, "fstype",
   1170 		       NULL, NULL, 0, mp->mnt_stat.f_fstypename,
   1171 		       0, CTL_CREATE, CTL_EOL);
   1172 	sysctl_createv(NULL, 0, &vte->vte_node, NULL,
   1173 		       CTLFLAG_READONLY, CTLTYPE_QUAD, "nentries",
   1174 		       NULL, NULL, 0, &vte->vte_count, 0, CTL_CREATE, CTL_EOL);
   1175 
   1176 	return (vte);
   1177 }
   1178 
   1179 /*
   1180  * Add a file to be monitored by Veriexec.
   1181  *
   1182  * Expected elements in dict: file, fp, fp-type, entry-type.
   1183  */
   1184 int
   1185 veriexec_file_add(struct lwp *l, prop_dictionary_t dict)
   1186 {
   1187 	struct veriexec_table_entry *vte;
   1188 	struct veriexec_file_entry *vfe = NULL, *hh;
   1189 	struct vnode *vp;
   1190 	const char *file, *fp_type;
   1191 	int error;
   1192 
   1193 	if (!prop_dictionary_get_cstring_nocopy(dict, "file", &file))
   1194 		return (EINVAL);
   1195 
   1196 	error = namei_simple_kernel(file, NSM_FOLLOW_NOEMULROOT, &vp);
   1197 	if (error)
   1198 		return (error);
   1199 
   1200 	/* Add only regular files. */
   1201 	if (vp->v_type != VREG) {
   1202 		log(LOG_ERR, "Veriexec: Not adding `%s': Not a regular file.\n",
   1203 		    file);
   1204 
   1205 		error = EBADF;
   1206 
   1207 		goto out;
   1208 	}
   1209 
   1210 	vfe = kmem_zalloc(sizeof(*vfe), KM_SLEEP);
   1211 
   1212 	rw_init(&vfe->lock);
   1213 
   1214 	/* Lookup fingerprint hashing algorithm. */
   1215 	fp_type = prop_string_cstring_nocopy(prop_dictionary_get(dict,
   1216 	    "fp-type"));
   1217 	if ((vfe->ops = veriexec_fpops_lookup(fp_type)) == NULL) {
   1218 		log(LOG_ERR, "Veriexec: Invalid or unknown fingerprint type "
   1219 		    "`%s' for file `%s'.\n", fp_type, file);
   1220 
   1221 		error = EOPNOTSUPP;
   1222 
   1223 		goto out;
   1224 	}
   1225 
   1226 	if (prop_data_size(prop_dictionary_get(dict, "fp")) !=
   1227 	    vfe->ops->hash_len) {
   1228 		log(LOG_ERR, "Veriexec: Bad fingerprint length for `%s'.\n",
   1229 		    file);
   1230 
   1231 		error = EINVAL;
   1232 
   1233 		goto out;
   1234 	}
   1235 
   1236 	vfe->fp = kmem_alloc(vfe->ops->hash_len, KM_SLEEP);
   1237 	memcpy(vfe->fp, prop_data_data_nocopy(prop_dictionary_get(dict, "fp")),
   1238 	    vfe->ops->hash_len);
   1239 
   1240 	rw_enter(&veriexec_op_lock, RW_WRITER);
   1241 
   1242 	/*
   1243 	 * See if we already have an entry for this file. If we do, then
   1244 	 * let the user know and silently pretend to succeed.
   1245 	 */
   1246 	hh = veriexec_get(vp);
   1247 	if (hh != NULL) {
   1248 		bool fp_mismatch;
   1249 
   1250 		if (strcmp(vfe->ops->type, fp_type) ||
   1251 		    memcmp(hh->fp, vfe->fp, hh->ops->hash_len))
   1252 			fp_mismatch = true;
   1253 		else
   1254 			fp_mismatch = false;
   1255 
   1256 		if ((veriexec_verbose >= 1) || fp_mismatch)
   1257 			log(LOG_NOTICE, "Veriexec: Duplicate entry for `%s' "
   1258 			    "ignored. (%s fingerprint)\n", file,
   1259 			    fp_mismatch ? "different" : "same");
   1260 
   1261 		veriexec_file_free(vfe);
   1262 
   1263 		/* XXX Should this be EEXIST if fp_mismatch is true? */
   1264 		error = 0;
   1265 
   1266 		goto unlock_out;
   1267 	}
   1268 
   1269 	/* Continue entry initialization. */
   1270 	if (prop_dictionary_get_uint8(dict, "entry-type", &vfe->type) == FALSE)
   1271 		vfe->type = 0;
   1272 	else {
   1273 		uint8_t extra_flags;
   1274 
   1275 		extra_flags = vfe->type & ~(VERIEXEC_DIRECT |
   1276 		    VERIEXEC_INDIRECT | VERIEXEC_FILE | VERIEXEC_UNTRUSTED);
   1277 		if (extra_flags) {
   1278 			log(LOG_NOTICE, "Veriexec: Contaminated flags `0x%x' "
   1279 			    "for `%s', skipping.\n", extra_flags, file);
   1280 
   1281 			error = EINVAL;
   1282 
   1283 			goto unlock_out;
   1284 		}
   1285 	}
   1286 	if (!(vfe->type & (VERIEXEC_DIRECT | VERIEXEC_INDIRECT |
   1287 	    VERIEXEC_FILE)))
   1288 		vfe->type |= VERIEXEC_DIRECT;
   1289 
   1290 	vfe->status = FINGERPRINT_NOTEVAL;
   1291 	if (prop_bool_true(prop_dictionary_get(dict, "keep-filename"))) {
   1292 		vfe->filename_len = strlen(file) + 1;
   1293 		vfe->filename = kmem_alloc(vfe->filename_len, KM_SLEEP);
   1294 		strlcpy(vfe->filename, file, vfe->filename_len);
   1295 	} else
   1296 		vfe->filename = NULL;
   1297 
   1298 	vfe->page_fp = NULL;
   1299 	vfe->page_fp_status = PAGE_FP_NONE;
   1300 	vfe->npages = 0;
   1301 	vfe->last_page_size = 0;
   1302 
   1303 	if (prop_bool_true(prop_dictionary_get(dict, "eval-on-load")) ||
   1304 	    (vfe->type & VERIEXEC_UNTRUSTED)) {
   1305 		u_char *digest;
   1306 
   1307 		digest = kmem_zalloc(vfe->ops->hash_len, KM_SLEEP);
   1308 
   1309 		error = veriexec_fp_calc(l, vp, VERIEXEC_UNLOCKED,
   1310 					 vfe, digest);
   1311 		if (error) {
   1312 			kmem_free(digest, vfe->ops->hash_len);
   1313 			goto unlock_out;
   1314 		}
   1315 
   1316 		if (veriexec_fp_cmp(vfe->ops, vfe->fp, digest) == 0)
   1317 			vfe->status = FINGERPRINT_VALID;
   1318 		else
   1319 			vfe->status = FINGERPRINT_NOMATCH;
   1320 
   1321 		kmem_free(digest, vfe->ops->hash_len);
   1322 	}
   1323 
   1324 	vte = veriexec_table_lookup(vp->v_mount);
   1325 	if (vte == NULL)
   1326 		vte = veriexec_table_add(l, vp->v_mount);
   1327 
   1328 	/* XXX if we bail below this, we might want to gc newly created vtes. */
   1329 
   1330 	error = fileassoc_add(vp, veriexec_hook, vfe);
   1331 	if (error)
   1332 		goto unlock_out;
   1333 
   1334 	vte->vte_count++;
   1335 
   1336 	veriexec_file_report(NULL, "New entry.", file, NULL, REPORT_DEBUG);
   1337 	veriexec_bypass = 0;
   1338 
   1339   unlock_out:
   1340 	rw_exit(&veriexec_op_lock);
   1341 
   1342   out:
   1343 	vrele(vp);
   1344 	if (error)
   1345 		veriexec_file_free(vfe);
   1346 
   1347 	return (error);
   1348 }
   1349 
   1350 int
   1351 veriexec_table_delete(struct lwp *l, struct mount *mp) {
   1352 	struct veriexec_table_entry *vte;
   1353 
   1354 	vte = veriexec_table_lookup(mp);
   1355 	if (vte == NULL)
   1356 		return (ENOENT);
   1357 
   1358 	veriexec_mountspecific_dtor(vte);
   1359 	mount_setspecific(mp, veriexec_mountspecific_key, NULL);
   1360 
   1361 	return (fileassoc_table_clear(mp, veriexec_hook));
   1362 }
   1363 
   1364 int
   1365 veriexec_file_delete(struct lwp *l, struct vnode *vp) {
   1366 	struct veriexec_table_entry *vte;
   1367 	int error;
   1368 
   1369 	vte = veriexec_table_lookup(vp->v_mount);
   1370 	if (vte == NULL)
   1371 		return (ENOENT);
   1372 
   1373 	rw_enter(&veriexec_op_lock, RW_WRITER);
   1374 	error = fileassoc_clear(vp, veriexec_hook);
   1375 	rw_exit(&veriexec_op_lock);
   1376 	if (!error)
   1377 		vte->vte_count--;
   1378 
   1379 	return (error);
   1380 }
   1381 
   1382 /*
   1383  * Convert Veriexec entry data to a dictionary readable by userland tools.
   1384  */
   1385 static void
   1386 veriexec_file_convert(struct veriexec_file_entry *vfe, prop_dictionary_t rdict)
   1387 {
   1388 	if (vfe->filename)
   1389 		prop_dictionary_set(rdict, "file",
   1390 		    prop_string_create_cstring(vfe->filename));
   1391 	prop_dictionary_set_uint8(rdict, "entry-type", vfe->type);
   1392 	prop_dictionary_set_uint8(rdict, "status", vfe->status);
   1393 	prop_dictionary_set(rdict, "fp-type",
   1394 	    prop_string_create_cstring(vfe->ops->type));
   1395 	prop_dictionary_set(rdict, "fp",
   1396 	    prop_data_create_data(vfe->fp, vfe->ops->hash_len));
   1397 }
   1398 
   1399 int
   1400 veriexec_convert(struct vnode *vp, prop_dictionary_t rdict)
   1401 {
   1402 	struct veriexec_file_entry *vfe;
   1403 
   1404 	rw_enter(&veriexec_op_lock, RW_READER);
   1405 
   1406 	vfe = veriexec_get(vp);
   1407 	if (vfe == NULL) {
   1408 		rw_exit(&veriexec_op_lock);
   1409 		return (ENOENT);
   1410 	}
   1411 
   1412 	rw_enter(&vfe->lock, RW_READER);
   1413 	veriexec_file_convert(vfe, rdict);
   1414 
   1415 	rw_exit(&vfe->lock);
   1416 	rw_exit(&veriexec_op_lock);
   1417 	return (0);
   1418 }
   1419 
   1420 int
   1421 veriexec_unmountchk(struct mount *mp)
   1422 {
   1423 	int error;
   1424 
   1425 	if ((veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
   1426 	    || doing_shutdown)
   1427 		return (0);
   1428 
   1429 	rw_enter(&veriexec_op_lock, RW_READER);
   1430 
   1431 	switch (veriexec_strict) {
   1432 	case VERIEXEC_LEARNING:
   1433 		error = 0;
   1434 		break;
   1435 
   1436 	case VERIEXEC_IDS:
   1437 		if (veriexec_table_lookup(mp) != NULL) {
   1438 			log(LOG_INFO, "Veriexec: IDS mode, allowing unmount "
   1439 			    "of \"%s\".\n", mp->mnt_stat.f_mntonname);
   1440 		}
   1441 
   1442 		error = 0;
   1443 		break;
   1444 
   1445 	case VERIEXEC_IPS: {
   1446 		struct veriexec_table_entry *vte;
   1447 
   1448 		vte = veriexec_table_lookup(mp);
   1449 		if ((vte != NULL) && (vte->vte_count > 0)) {
   1450 			log(LOG_ALERT, "Veriexec: IPS mode, preventing"
   1451 			    " unmount of \"%s\" with monitored files.\n",
   1452 			    mp->mnt_stat.f_mntonname);
   1453 
   1454 			error = EPERM;
   1455 		} else
   1456 			error = 0;
   1457 		break;
   1458 		}
   1459 
   1460 	case VERIEXEC_LOCKDOWN:
   1461 	default:
   1462 		log(LOG_ALERT, "Veriexec: Lockdown mode, preventing unmount "
   1463 		    "of \"%s\".\n", mp->mnt_stat.f_mntonname);
   1464 		error = EPERM;
   1465 		break;
   1466 	}
   1467 
   1468 	rw_exit(&veriexec_op_lock);
   1469 	return (error);
   1470 }
   1471 
   1472 int
   1473 veriexec_openchk(struct lwp *l, struct vnode *vp, const char *path, int fmode)
   1474 {
   1475 	struct veriexec_file_entry *vfe = NULL;
   1476 	int error = 0;
   1477 
   1478 	if (veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
   1479 		return 0;
   1480 
   1481 	if (vp == NULL) {
   1482 		/* If no creation requested, let this fail normally. */
   1483 		if (!(fmode & O_CREAT))
   1484 			goto out;
   1485 
   1486 		/* Lockdown mode: Prevent creation of new files. */
   1487 		if (veriexec_strict >= VERIEXEC_LOCKDOWN) {
   1488 			log(LOG_ALERT, "Veriexec: Preventing new file "
   1489 			    "creation in `%s'.\n", path);
   1490 			error = EPERM;
   1491 		}
   1492 
   1493 		goto out;
   1494 	}
   1495 
   1496 	rw_enter(&veriexec_op_lock, RW_READER);
   1497 	error = veriexec_file_verify(l, vp, path, VERIEXEC_FILE,
   1498 				     VERIEXEC_LOCKED, &vfe);
   1499 
   1500 	if (error) {
   1501 		rw_exit(&veriexec_op_lock);
   1502 		goto out;
   1503 	}
   1504 
   1505 	if ((vfe != NULL) && ((fmode & FWRITE) || (fmode & O_TRUNC))) {
   1506 		veriexec_file_report(vfe, "Write access request.", path, l,
   1507 		    REPORT_ALWAYS | REPORT_ALARM);
   1508 
   1509 		/* IPS mode: Deny write access to monitored files. */
   1510 		if (veriexec_strict >= VERIEXEC_IPS)
   1511 			error = EPERM;
   1512 		else
   1513 			veriexec_file_purge(vfe, VERIEXEC_LOCKED);
   1514 	}
   1515 
   1516 	if (vfe != NULL)
   1517 		rw_exit(&vfe->lock);
   1518 
   1519 	rw_exit(&veriexec_op_lock);
   1520  out:
   1521 	return (error);
   1522 }
   1523 
   1524 static void
   1525 veriexec_file_dump(struct veriexec_file_entry *vfe, prop_array_t entries)
   1526 {
   1527 	prop_dictionary_t entry;
   1528 
   1529 	/* If we don't have a filename, this is meaningless. */
   1530 	if (vfe->filename == NULL)
   1531 		return;
   1532 
   1533 	entry = prop_dictionary_create();
   1534 
   1535 	veriexec_file_convert(vfe, entry);
   1536 
   1537 	prop_array_add(entries, entry);
   1538 }
   1539 
   1540 int
   1541 veriexec_dump(struct lwp *l, prop_array_t rarray)
   1542 {
   1543 	struct mount *mp, *nmp;
   1544 
   1545 	mutex_enter(&mountlist_lock);
   1546 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
   1547 		/* If it fails, the file-system is [being] unmounted. */
   1548 		if (vfs_busy(mp, &nmp) != 0)
   1549 			continue;
   1550 
   1551 		fileassoc_table_run(mp, veriexec_hook,
   1552 		    (fileassoc_cb_t)veriexec_file_dump, rarray);
   1553 
   1554 		vfs_unbusy(mp, false, &nmp);
   1555 	}
   1556 	mutex_exit(&mountlist_lock);
   1557 
   1558 	return (0);
   1559 }
   1560 
   1561 int
   1562 veriexec_flush(struct lwp *l)
   1563 {
   1564 	struct mount *mp, *nmp;
   1565 	int error = 0;
   1566 
   1567 	mutex_enter(&mountlist_lock);
   1568 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
   1569 		int lerror;
   1570 
   1571 		/* If it fails, the file-system is [being] unmounted. */
   1572 		if (vfs_busy(mp, &nmp) != 0)
   1573 			continue;
   1574 
   1575 		lerror = veriexec_table_delete(l, mp);
   1576 		if (lerror && lerror != ENOENT)
   1577 			error = lerror;
   1578 
   1579 		vfs_unbusy(mp, false, &nmp);
   1580 	}
   1581 	mutex_exit(&mountlist_lock);
   1582 
   1583 	return (error);
   1584 }
   1585