Home | History | Annotate | Line # | Download | only in kern
kern_veriexec.c revision 1.6
      1 /*	$NetBSD: kern_veriexec.c,v 1.6 2015/04/26 09:16:06 maxv 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 1.6 2015/04/26 09:16:06 maxv 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 /* state of file locking for veriexec_file_verify */
     82 #define VERIEXEC_FILE_UNLOCKED	0x02	/* Nothing locked, callee does it */
     83 #define VERIEXEC_FILE_LOCKED	0x04	/* File locked */
     84 
     85 #define VERIEXEC_RW_UPGRADE(lock)	while((rw_tryupgrade(lock)) == 0){};
     86 
     87 struct veriexec_fpops {
     88 	const char *type;
     89 	size_t hash_len;
     90 	size_t context_size;
     91 	veriexec_fpop_init_t init;
     92 	veriexec_fpop_update_t update;
     93 	veriexec_fpop_final_t final;
     94 	LIST_ENTRY(veriexec_fpops) entries;
     95 };
     96 
     97 /* Veriexec per-file entry data. */
     98 struct veriexec_file_entry {
     99 	krwlock_t lock;				/* r/w lock */
    100 	u_char *filename;			/* File name. */
    101 	u_char type;				/* Entry type. */
    102 	u_char status;				/* Evaluation status. */
    103 	u_char page_fp_status;			/* Per-page FP status. */
    104 	u_char *fp;				/* Fingerprint. */
    105 	void *page_fp;				/* Per-page fingerprints */
    106 	size_t npages;			    	/* Number of pages. */
    107 	size_t last_page_size;			/* To support < PAGE_SIZE */
    108 	struct veriexec_fpops *ops;		/* Fingerprint ops vector*/
    109 	size_t filename_len;			/* Length of filename. */
    110 };
    111 
    112 /* Veriexec per-table data. */
    113 struct veriexec_table_entry {
    114 	uint64_t vte_count;			/* Number of Veriexec entries. */
    115 	const struct sysctlnode *vte_node;
    116 };
    117 
    118 static int veriexec_verbose;
    119 static int veriexec_strict;
    120 static int veriexec_bypass = 1;
    121 
    122 static char *veriexec_fp_names = NULL;
    123 static size_t veriexec_name_max = 0;
    124 
    125 static const struct sysctlnode *veriexec_count_node;
    126 
    127 static fileassoc_t veriexec_hook;
    128 static specificdata_key_t veriexec_mountspecific_key;
    129 
    130 static LIST_HEAD(, veriexec_fpops) veriexec_fpops_list =
    131 	LIST_HEAD_INITIALIZER(veriexec_fpops_list);
    132 
    133 static int veriexec_raw_cb(kauth_cred_t, kauth_action_t, void *,
    134     void *, void *, void *, void *);
    135 static struct veriexec_fpops *veriexec_fpops_lookup(const char *);
    136 static void veriexec_file_free(struct veriexec_file_entry *);
    137 
    138 static unsigned int veriexec_tablecount = 0;
    139 
    140 /*
    141  * Veriexec operations global lock - most ops hold this as a read
    142  * lock, it is upgraded to a write lock when destroying veriexec file
    143  * table entries.
    144  */
    145 static krwlock_t veriexec_op_lock;
    146 
    147 /*
    148  * Sysctl helper routine for Veriexec.
    149  */
    150 static int
    151 sysctl_kern_veriexec_algorithms(SYSCTLFN_ARGS)
    152 {
    153 	size_t len;
    154 	int error;
    155 	const char *p;
    156 
    157 	if (newp != NULL)
    158 		return EPERM;
    159 
    160 	if (namelen != 0)
    161 		return EINVAL;
    162 
    163 	p = veriexec_fp_names == NULL ? "" : veriexec_fp_names;
    164 
    165 	len = strlen(p) + 1;
    166 
    167 	if (*oldlenp < len && oldp)
    168 		return ENOMEM;
    169 
    170 	if (oldp && (error = copyout(p, oldp, len)) != 0)
    171 		return error;
    172 
    173 	*oldlenp = len;
    174 	return 0;
    175 }
    176 
    177 static int
    178 sysctl_kern_veriexec_strict(SYSCTLFN_ARGS)
    179 {
    180 	struct sysctlnode node;
    181 	int error, newval;
    182 
    183 	node = *rnode;
    184 	node.sysctl_data = &newval;
    185 
    186 	newval = veriexec_strict;
    187 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
    188 	if (error || newp == NULL)
    189 		return error;
    190 
    191 	if (newval < veriexec_strict)
    192 		return EPERM;
    193 
    194 	veriexec_strict = newval;
    195 
    196 	return 0;
    197 }
    198 
    199 SYSCTL_SETUP(sysctl_kern_veriexec_setup, "sysctl kern.veriexec setup")
    200 {
    201 	const struct sysctlnode *rnode = NULL;
    202 
    203 	sysctl_createv(clog, 0, NULL, &rnode,
    204 		       CTLFLAG_PERMANENT,
    205 		       CTLTYPE_NODE, "veriexec",
    206 		       SYSCTL_DESCR("Veriexec"),
    207 		       NULL, 0, NULL, 0,
    208 		       CTL_KERN, CTL_CREATE, CTL_EOL);
    209 
    210 	sysctl_createv(clog, 0, &rnode, NULL,
    211 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    212 		       CTLTYPE_INT, "verbose",
    213 		       SYSCTL_DESCR("Veriexec verbose level"),
    214 		       NULL, 0, &veriexec_verbose, 0,
    215 		       CTL_CREATE, CTL_EOL);
    216 	sysctl_createv(clog, 0, &rnode, NULL,
    217 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    218 		       CTLTYPE_INT, "strict",
    219 		       SYSCTL_DESCR("Veriexec strict level"),
    220 		       sysctl_kern_veriexec_strict, 0, NULL, 0,
    221 		       CTL_CREATE, CTL_EOL);
    222 	sysctl_createv(clog, 0, &rnode, NULL,
    223 		       CTLFLAG_PERMANENT,
    224 		       CTLTYPE_STRING, "algorithms",
    225 		       SYSCTL_DESCR("Veriexec supported hashing "
    226 				    "algorithms"),
    227 		       sysctl_kern_veriexec_algorithms, 0, NULL, 0,
    228 		       CTL_CREATE, CTL_EOL);
    229 	sysctl_createv(clog, 0, &rnode, &veriexec_count_node,
    230 		       CTLFLAG_PERMANENT,
    231 		       CTLTYPE_NODE, "count",
    232 		       SYSCTL_DESCR("Number of fingerprints on mount(s)"),
    233 		       NULL, 0, NULL, 0,
    234 		       CTL_CREATE, CTL_EOL);
    235 }
    236 
    237 /*
    238  * Add ops to the fignerprint ops vector list.
    239  */
    240 int
    241 veriexec_fpops_add(const char *fp_type, size_t hash_len, size_t ctx_size,
    242     veriexec_fpop_init_t init, veriexec_fpop_update_t update,
    243     veriexec_fpop_final_t final)
    244 {
    245 	struct veriexec_fpops *ops;
    246 
    247 	/* Sanity check all parameters. */
    248 	if ((fp_type == NULL) || (hash_len == 0) || (ctx_size == 0) ||
    249 	    (init == NULL) || (update == NULL) || (final == NULL))
    250 		return (EFAULT);
    251 
    252 	if (veriexec_fpops_lookup(fp_type) != NULL)
    253 		return (EEXIST);
    254 
    255 	ops = kmem_alloc(sizeof(*ops), KM_SLEEP);
    256 
    257 	ops->type = fp_type;
    258 	ops->hash_len = hash_len;
    259 	ops->context_size = ctx_size;
    260 	ops->init = init;
    261 	ops->update = update;
    262 	ops->final = final;
    263 
    264 	LIST_INSERT_HEAD(&veriexec_fpops_list, ops, entries);
    265 
    266 	/*
    267 	 * If we don't have space for any names, allocate enough for six
    268 	 * which should be sufficient. (it's also enough for all algorithms
    269 	 * we can support at the moment)
    270 	 */
    271 	if (veriexec_fp_names == NULL) {
    272 		veriexec_name_max = 64;
    273 		veriexec_fp_names = kmem_zalloc(veriexec_name_max, KM_SLEEP);
    274 	}
    275 
    276 	/*
    277 	 * If we're running out of space for storing supported algorithms,
    278 	 * extend the buffer with space for four names.
    279 	 */
    280 	while (veriexec_name_max - (strlen(veriexec_fp_names) + 1) <
    281 	    strlen(fp_type)) {
    282 		char *newp;
    283 		unsigned int new_max;
    284 
    285 		/* Add space for four algorithm names. */
    286 		new_max = veriexec_name_max + 64;
    287 		newp = kmem_zalloc(new_max, KM_SLEEP);
    288 		strlcpy(newp, veriexec_fp_names, new_max);
    289 		kmem_free(veriexec_fp_names, veriexec_name_max);
    290 		veriexec_fp_names = newp;
    291 		veriexec_name_max = new_max;
    292 	}
    293 
    294 	if (*veriexec_fp_names != '\0')
    295 		strlcat(veriexec_fp_names, " ", veriexec_name_max);
    296 
    297 	strlcat(veriexec_fp_names, fp_type, veriexec_name_max);
    298 
    299 	return (0);
    300 }
    301 
    302 static void
    303 veriexec_mountspecific_dtor(void *v)
    304 {
    305 	struct veriexec_table_entry *vte = v;
    306 
    307 	if (vte == NULL) {
    308 		return;
    309 	}
    310 	sysctl_free(__UNCONST(vte->vte_node));
    311 	veriexec_tablecount--;
    312 	kmem_free(vte, sizeof(*vte));
    313 }
    314 
    315 static int
    316 veriexec_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
    317     void *arg0, void *arg1, void *arg2, void *arg3)
    318 {
    319 	int result;
    320 	enum kauth_system_req req;
    321 
    322 	if (action != KAUTH_SYSTEM_VERIEXEC)
    323 		return KAUTH_RESULT_DEFER;
    324 
    325 	result = KAUTH_RESULT_DEFER;
    326 	req = (enum kauth_system_req)arg0;
    327 
    328 	if (req == KAUTH_REQ_SYSTEM_VERIEXEC_MODIFY &&
    329 	    veriexec_strict > VERIEXEC_LEARNING) {
    330 		log(LOG_WARNING, "Veriexec: Strict mode, modifying "
    331 		    "tables not permitted.\n");
    332 
    333 		result = KAUTH_RESULT_DENY;
    334 	}
    335 
    336 	return result;
    337 }
    338 
    339 /*
    340  * Initialise Veriexec.
    341  */
    342 void
    343 veriexec_init(void)
    344 {
    345 	int error;
    346 
    347 	/* Register a fileassoc for Veriexec. */
    348 	error = fileassoc_register("veriexec",
    349 	    (fileassoc_cleanup_cb_t)veriexec_file_free, &veriexec_hook);
    350 	if (error)
    351 		panic("Veriexec: Can't register fileassoc: error=%d", error);
    352 
    353 	/* Register listener to handle raw disk access. */
    354 	if (kauth_listen_scope(KAUTH_SCOPE_DEVICE, veriexec_raw_cb, NULL) ==
    355 	    NULL)
    356 		panic("Veriexec: Can't listen on device scope");
    357 
    358 	error = mount_specific_key_create(&veriexec_mountspecific_key,
    359 	    veriexec_mountspecific_dtor);
    360 	if (error)
    361 		panic("Veriexec: Can't create mountspecific key");
    362 
    363 	if (kauth_listen_scope(KAUTH_SCOPE_SYSTEM, veriexec_listener_cb,
    364 	    NULL) == NULL)
    365 		panic("Veriexec: Can't listen on system scope");
    366 
    367 	rw_init(&veriexec_op_lock);
    368 
    369 #define	FPOPS_ADD(a, b, c, d, e, f)	\
    370 	veriexec_fpops_add(a, b, c, (veriexec_fpop_init_t)d, \
    371 	 (veriexec_fpop_update_t)e, (veriexec_fpop_final_t)f)
    372 
    373 #ifdef VERIFIED_EXEC_FP_RMD160
    374 	FPOPS_ADD("RMD160", RMD160_DIGEST_LENGTH, sizeof(RMD160_CTX),
    375 	    RMD160Init, RMD160Update, RMD160Final);
    376 #endif /* VERIFIED_EXEC_FP_RMD160 */
    377 
    378 #ifdef VERIFIED_EXEC_FP_SHA256
    379 	FPOPS_ADD("SHA256", SHA256_DIGEST_LENGTH, sizeof(SHA256_CTX),
    380 	    SHA256_Init, SHA256_Update, SHA256_Final);
    381 #endif /* VERIFIED_EXEC_FP_SHA256 */
    382 
    383 #ifdef VERIFIED_EXEC_FP_SHA384
    384 	FPOPS_ADD("SHA384", SHA384_DIGEST_LENGTH, sizeof(SHA384_CTX),
    385 	    SHA384_Init, SHA384_Update, SHA384_Final);
    386 #endif /* VERIFIED_EXEC_FP_SHA384 */
    387 
    388 #ifdef VERIFIED_EXEC_FP_SHA512
    389 	FPOPS_ADD("SHA512", SHA512_DIGEST_LENGTH, sizeof(SHA512_CTX),
    390 	    SHA512_Init, SHA512_Update, SHA512_Final);
    391 #endif /* VERIFIED_EXEC_FP_SHA512 */
    392 
    393 #ifdef VERIFIED_EXEC_FP_SHA1
    394 	FPOPS_ADD("SHA1", SHA1_DIGEST_LENGTH, sizeof(SHA1_CTX),
    395 	    SHA1Init, SHA1Update, SHA1Final);
    396 #endif /* VERIFIED_EXEC_FP_SHA1 */
    397 
    398 #ifdef VERIFIED_EXEC_FP_MD5
    399 	FPOPS_ADD("MD5", MD5_DIGEST_LENGTH, sizeof(MD5_CTX),
    400 	    MD5Init, MD5Update, MD5Final);
    401 #endif /* VERIFIED_EXEC_FP_MD5 */
    402 
    403 #undef FPOPS_ADD
    404 }
    405 
    406 static struct veriexec_fpops *
    407 veriexec_fpops_lookup(const char *name)
    408 {
    409 	struct veriexec_fpops *ops;
    410 
    411 	if (name == NULL)
    412 		return (NULL);
    413 
    414 	LIST_FOREACH(ops, &veriexec_fpops_list, entries) {
    415 		if (strcasecmp(name, ops->type) == 0)
    416 			return (ops);
    417 	}
    418 
    419 	return (NULL);
    420 }
    421 
    422 /*
    423  * Calculate fingerprint. Information on hash length and routines used is
    424  * extracted from veriexec_hash_list according to the hash type.
    425  *
    426  * NOTE: vfe is assumed to be locked for writing on entry.
    427  */
    428 static int
    429 veriexec_fp_calc(struct lwp *l, struct vnode *vp, int file_lock_state,
    430     struct veriexec_file_entry *vfe, u_char *fp)
    431 {
    432 	struct vattr va;
    433 	void *ctx, *page_ctx;
    434 	u_char *buf, *page_fp;
    435 	off_t offset, len;
    436 	size_t resid, npages;
    437 	int error, do_perpage, pagen;
    438 
    439 	KASSERT((file_lock_state != VERIEXEC_LOCKED) &&
    440 	    (file_lock_state != VERIEXEC_UNLOCKED));
    441 
    442 	if (file_lock_state == VERIEXEC_FILE_UNLOCKED)
    443 		vn_lock(vp, LK_SHARED | LK_RETRY);
    444 	error = VOP_GETATTR(vp, &va, l->l_cred);
    445 	if (file_lock_state == VERIEXEC_FILE_UNLOCKED)
    446 		VOP_UNLOCK(vp);
    447 	if (error)
    448 		return (error);
    449 
    450 #ifdef notyet /* XXX - for now */
    451 	if ((vfe->type & VERIEXEC_UNTRUSTED) &&
    452 	    (vfe->page_fp_status == PAGE_FP_NONE))
    453 		do_perpage = 1;
    454 	else
    455 #endif  /* notyet */
    456 		do_perpage = 0;
    457 
    458 	ctx = kmem_alloc(vfe->ops->context_size, KM_SLEEP);
    459 	buf = kmem_alloc(PAGE_SIZE, KM_SLEEP);
    460 
    461 	page_ctx = NULL;
    462 	page_fp = NULL;
    463 	npages = 0;
    464 	if (do_perpage) {
    465 		npages = (va.va_size >> PAGE_SHIFT) + 1;
    466 		page_fp = kmem_alloc(vfe->ops->hash_len * npages, KM_SLEEP);
    467 		vfe->page_fp = page_fp;
    468 		page_ctx = kmem_alloc(vfe->ops->context_size, KM_SLEEP);
    469 	}
    470 
    471 	(vfe->ops->init)(ctx);
    472 
    473 	len = 0;
    474 	error = 0;
    475 	pagen = 0;
    476 	for (offset = 0; offset < va.va_size; offset += PAGE_SIZE) {
    477 		len = ((va.va_size - offset) < PAGE_SIZE) ?
    478 		    (va.va_size - offset) : PAGE_SIZE;
    479 
    480 		error = vn_rdwr(UIO_READ, vp, buf, len, offset,
    481 				UIO_SYSSPACE,
    482 				((file_lock_state == VERIEXEC_FILE_LOCKED)?
    483 				 IO_NODELOCKED : 0),
    484 				l->l_cred, &resid, NULL);
    485 
    486 		if (error) {
    487 			if (do_perpage) {
    488 				kmem_free(vfe->page_fp,
    489 				    vfe->ops->hash_len * npages);
    490 				vfe->page_fp = NULL;
    491 			}
    492 
    493 			goto bad;
    494 		}
    495 
    496 		(vfe->ops->update)(ctx, buf, (unsigned int) len);
    497 
    498 		if (do_perpage) {
    499 			(vfe->ops->init)(page_ctx);
    500 			(vfe->ops->update)(page_ctx, buf, (unsigned int)len);
    501 			(vfe->ops->final)(page_fp, page_ctx);
    502 
    503 			if (veriexec_verbose >= 2) {
    504 				int i;
    505 
    506 				printf("hash for page %d: ", pagen);
    507 				for (i = 0; i < vfe->ops->hash_len; i++)
    508 					printf("%02x", page_fp[i]);
    509 				printf("\n");
    510 			}
    511 
    512 			page_fp += vfe->ops->hash_len;
    513 			pagen++;
    514 		}
    515 
    516 		if (len != PAGE_SIZE)
    517 			break;
    518 	}
    519 
    520 	(vfe->ops->final)(fp, ctx);
    521 
    522 	if (do_perpage) {
    523 		vfe->last_page_size = len;
    524 		vfe->page_fp_status = PAGE_FP_READY;
    525 		vfe->npages = npages;
    526 	}
    527 
    528 bad:
    529 	if (do_perpage)
    530 		kmem_free(page_ctx, vfe->ops->context_size);
    531 
    532 	kmem_free(ctx, vfe->ops->context_size);
    533 	kmem_free(buf, PAGE_SIZE);
    534 
    535 	return (error);
    536 }
    537 
    538 /* Compare two fingerprints of the same type. */
    539 static int
    540 veriexec_fp_cmp(struct veriexec_fpops *ops, u_char *fp1, u_char *fp2)
    541 {
    542 	if (veriexec_verbose >= 2) {
    543 		int i;
    544 
    545 		printf("comparing hashes...\n");
    546 		printf("fp1: ");
    547 		for (i = 0; i < ops->hash_len; i++) {
    548 			printf("%02x", fp1[i]);
    549 		}
    550 		printf("\nfp2: ");
    551 		for (i = 0; i < ops->hash_len; i++) {
    552 			printf("%02x", fp2[i]);
    553 		}
    554 		printf("\n");
    555 	}
    556 
    557 	return (memcmp(fp1, fp2, ops->hash_len));
    558 }
    559 
    560 static int
    561 veriexec_fp_status(struct lwp *l, struct vnode *vp, int file_lock_state,
    562     struct veriexec_file_entry *vfe, u_char *status)
    563 {
    564 	size_t hash_len = vfe->ops->hash_len;
    565 	u_char *digest;
    566 	int error = 0;
    567 
    568 	digest = kmem_zalloc(hash_len, KM_SLEEP);
    569 
    570 	error = veriexec_fp_calc(l, vp, file_lock_state, vfe, digest);
    571 	if (error)
    572 		goto out;
    573 
    574 	/* Compare fingerprint with loaded data. */
    575 	if (veriexec_fp_cmp(vfe->ops, vfe->fp, digest) == 0)
    576 		*status = FINGERPRINT_VALID;
    577 	else
    578 		*status = FINGERPRINT_NOMATCH;
    579 
    580 out:
    581 	kmem_free(digest, hash_len);
    582 	return error;
    583 }
    584 
    585 
    586 static struct veriexec_table_entry *
    587 veriexec_table_lookup(struct mount *mp)
    588 {
    589 	/* XXX: From raidframe init */
    590 	if (mp == NULL)
    591 		return NULL;
    592 
    593 	return mount_getspecific(mp, veriexec_mountspecific_key);
    594 }
    595 
    596 static struct veriexec_file_entry *
    597 veriexec_get(struct vnode *vp)
    598 {
    599 	return (fileassoc_lookup(vp, veriexec_hook));
    600 }
    601 
    602 bool
    603 veriexec_lookup(struct vnode *vp)
    604 {
    605 	return (veriexec_get(vp) == NULL ? false : true);
    606 }
    607 
    608 /*
    609  * Routine for maintaining mostly consistent message formats in Veriexec.
    610  */
    611 static void
    612 veriexec_file_report(struct veriexec_file_entry *vfe, const u_char *msg,
    613     const u_char *filename, struct lwp *l, int f)
    614 {
    615 	if (vfe != NULL && vfe->filename != NULL)
    616 		filename = vfe->filename;
    617 
    618 	if (filename == NULL)
    619 		return;
    620 
    621 	if (((f & REPORT_LOGMASK) >> 1) <= veriexec_verbose) {
    622 		if (!(f & REPORT_ALARM) || (l == NULL))
    623 			log(LOG_NOTICE, "Veriexec: %s [%s]\n", msg,
    624 			    filename);
    625 		else
    626 			log(LOG_ALERT, "Veriexec: %s [%s, prog=%s pid=%u, "
    627 			    "uid=%u, gid=%u]\n", msg, filename,
    628 			    l->l_proc->p_comm, l->l_proc->p_pid,
    629 			    kauth_cred_getuid(l->l_cred),
    630 			    kauth_cred_getgid(l->l_cred));
    631 	}
    632 
    633 	if (f & REPORT_PANIC)
    634 		panic("Veriexec: Unrecoverable error.");
    635 }
    636 
    637 /*
    638  * Verify the fingerprint of the given file. If we're called directly from
    639  * sys_execve(), 'flag' will be VERIEXEC_DIRECT. If we're called from
    640  * exec_script(), 'flag' will be VERIEXEC_INDIRECT.  If we are called from
    641  * vn_open(), 'flag' will be VERIEXEC_FILE.
    642  *
    643  * 'veriexec_op_lock' must be locked (and remains locked).
    644  *
    645  * NOTE: The veriexec file entry pointer (vfep) will be returned LOCKED
    646  *       on no error.
    647  */
    648 static int
    649 veriexec_file_verify(struct lwp *l, struct vnode *vp, const u_char *name,
    650     int flag, int file_lock_state, struct veriexec_file_entry **vfep)
    651 {
    652 	struct veriexec_file_entry *vfe;
    653 	int error = 0;
    654 
    655 	KASSERT(rw_lock_held(&veriexec_op_lock));
    656 	KASSERT((file_lock_state != VERIEXEC_LOCKED) &&
    657 	    (file_lock_state != VERIEXEC_UNLOCKED));
    658 
    659 #define VFE_NEEDS_EVAL(vfe) ((vfe->status == FINGERPRINT_NOTEVAL) || \
    660 			     (vfe->type & VERIEXEC_UNTRUSTED))
    661 
    662 	if (vfep != NULL)
    663 		*vfep = NULL;
    664 
    665 	if (vp->v_type != VREG)
    666 		return (0);
    667 
    668 	/* Lookup veriexec table entry, save pointer if requested. */
    669 	vfe = veriexec_get(vp);
    670 	if (vfep != NULL)
    671 		*vfep = vfe;
    672 
    673 	/* No entry in the veriexec tables. */
    674 	if (vfe == NULL) {
    675 		veriexec_file_report(NULL, "No entry.", name,
    676 		    l, REPORT_VERBOSE);
    677 
    678 		/*
    679 		 * Lockdown mode: Deny access to non-monitored files.
    680 		 * IPS mode: Deny execution of non-monitored files.
    681 		 */
    682 		if ((veriexec_strict >= VERIEXEC_LOCKDOWN) ||
    683 		    ((veriexec_strict >= VERIEXEC_IPS) &&
    684 		     (flag != VERIEXEC_FILE)))
    685 			return (EPERM);
    686 
    687 		return (0);
    688 	}
    689 
    690 	/*
    691 	 * Grab the lock for the entry, if we need to do an evaluation
    692 	 * then the lock is a write lock, after we have the write
    693 	 * lock, check if we really need it - some other thread may
    694 	 * have already done the work for us.
    695 	 */
    696 	if (VFE_NEEDS_EVAL(vfe)) {
    697 		rw_enter(&vfe->lock, RW_WRITER);
    698 		if (!VFE_NEEDS_EVAL(vfe))
    699 			rw_downgrade(&vfe->lock);
    700 	} else
    701 		rw_enter(&vfe->lock, RW_READER);
    702 
    703 	/* Evaluate fingerprint if needed. */
    704 	if (VFE_NEEDS_EVAL(vfe)) {
    705 		u_char status;
    706 
    707 		error = veriexec_fp_status(l, vp, file_lock_state, vfe, &status);
    708 		if (error) {
    709 			veriexec_file_report(vfe, "Fingerprint calculation error.",
    710 			    name, NULL, REPORT_ALWAYS);
    711 			rw_exit(&vfe->lock);
    712 			return (error);
    713 		}
    714 		vfe->status = status;
    715 		rw_downgrade(&vfe->lock);
    716 	}
    717 
    718 	if (!(vfe->type & flag)) {
    719 		veriexec_file_report(vfe, "Incorrect access type.", name, l,
    720 		    REPORT_ALWAYS|REPORT_ALARM);
    721 
    722 		/* IPS mode: Enforce access type. */
    723 		if (veriexec_strict >= VERIEXEC_IPS) {
    724 			rw_exit(&vfe->lock);
    725 			return (EPERM);
    726 		}
    727 	}
    728 
    729 	switch (vfe->status) {
    730 	case FINGERPRINT_NOTEVAL:
    731 		/* Should not happen. */
    732 		rw_exit(&vfe->lock);
    733 		veriexec_file_report(vfe, "Not-evaluated status "
    734 		    "post evaluation; inconsistency detected.", name,
    735 		    NULL, REPORT_ALWAYS|REPORT_PANIC);
    736 		/* NOTREACHED */
    737 
    738 	case FINGERPRINT_VALID:
    739 		/* Valid fingerprint. */
    740 		veriexec_file_report(vfe, "Match.", name, NULL,
    741 		    REPORT_VERBOSE);
    742 
    743 		break;
    744 
    745 	case FINGERPRINT_NOMATCH:
    746 		/* Fingerprint mismatch. */
    747 		veriexec_file_report(vfe, "Mismatch.", name,
    748 		    NULL, REPORT_ALWAYS|REPORT_ALARM);
    749 
    750 		/* IDS mode: Deny access on fingerprint mismatch. */
    751 		if (veriexec_strict >= VERIEXEC_IDS) {
    752 			rw_exit(&vfe->lock);
    753 			error = EPERM;
    754 		}
    755 
    756 		break;
    757 
    758 	default:
    759 		/* Should never happen. */
    760 		rw_exit(&vfe->lock);
    761 		veriexec_file_report(vfe, "Invalid status "
    762 		    "post evaluation.", name, NULL, REPORT_ALWAYS|REPORT_PANIC);
    763 		/* NOTREACHED */
    764 	}
    765 
    766 	return (error);
    767 }
    768 
    769 int
    770 veriexec_verify(struct lwp *l, struct vnode *vp, const u_char *name, int flag,
    771     bool *found)
    772 {
    773 	struct veriexec_file_entry *vfe;
    774 	int r;
    775 
    776 	if (veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
    777 		return 0;
    778 
    779 	rw_enter(&veriexec_op_lock, RW_READER);
    780 	r = veriexec_file_verify(l, vp, name, flag, VERIEXEC_FILE_UNLOCKED,
    781 	    &vfe);
    782 	rw_exit(&veriexec_op_lock);
    783 
    784 	if ((r  == 0) && (vfe != NULL))
    785 		rw_exit(&vfe->lock);
    786 
    787 	if (found != NULL)
    788 		*found = (vfe != NULL) ? true : false;
    789 
    790 	return (r);
    791 }
    792 
    793 #ifdef notyet
    794 /*
    795  * Evaluate per-page fingerprints.
    796  */
    797 int
    798 veriexec_page_verify(struct veriexec_file_entry *vfe, struct vm_page *pg,
    799     size_t idx, struct lwp *l)
    800 {
    801 	void *ctx;
    802 	u_char *fp;
    803 	u_char *page_fp;
    804 	int error;
    805 	vaddr_t kva;
    806 
    807 	if (vfe->page_fp_status == PAGE_FP_NONE)
    808 		return (0);
    809 
    810 	if (vfe->page_fp_status == PAGE_FP_FAIL)
    811 		return (EPERM);
    812 
    813 	if (idx >= vfe->npages)
    814 		return (0);
    815 
    816 	ctx = kmem_alloc(vfe->ops->context_size, KM_SLEEP);
    817 	fp = kmem_alloc(vfe->ops->hash_len, KM_SLEEP);
    818 	kva = uvm_km_alloc(kernel_map, PAGE_SIZE, VM_PGCOLOR_BUCKET(pg),
    819 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
    820 	pmap_kenter_pa(kva, VM_PAGE_TO_PHYS(pg), VM_PROT_READ, 0);
    821 	pmap_update(pmap_kernel());
    822 
    823 	page_fp = (u_char *) vfe->page_fp + (vfe->ops->hash_len * idx);
    824 	(vfe->ops->init)(ctx);
    825 	(vfe->ops->update)(ctx, (void *) kva,
    826 			   ((vfe->npages - 1) == idx) ? vfe->last_page_size
    827 						      : PAGE_SIZE);
    828 	(vfe->ops->final)(fp, ctx);
    829 
    830 	pmap_kremove(kva, PAGE_SIZE);
    831 	pmap_update(pmap_kernel());
    832 	uvm_km_free(kernel_map, kva, PAGE_SIZE, UVM_KMF_VAONLY);
    833 
    834 	error = veriexec_fp_cmp(vfe->ops, page_fp, fp);
    835 	if (error) {
    836 		const char *msg;
    837 
    838 		if (veriexec_strict > VERIEXEC_LEARNING) {
    839 			msg = "Pages modified: Killing process.";
    840 		} else {
    841 			msg = "Pages modified.";
    842 			error = 0;
    843 		}
    844 
    845 		veriexec_file_report(msg, "[page_in]", l,
    846 		    REPORT_ALWAYS|REPORT_ALARM);
    847 
    848 		if (error) {
    849 			ksiginfo_t ksi;
    850 
    851 			KSI_INIT(&ksi);
    852 			ksi.ksi_signo = SIGKILL;
    853 			ksi.ksi_code = SI_NOINFO;
    854 			ksi.ksi_pid = l->l_proc->p_pid;
    855 			ksi.ksi_uid = 0;
    856 
    857 			kpsignal(l->l_proc, &ksi, NULL);
    858 		}
    859 	}
    860 
    861 	kmem_free(ctx, vfe->ops->context_size);
    862 	kmem_free(fp, vfe->ops->hash_len);
    863 
    864 	return (error);
    865 }
    866 #endif /* notyet */
    867 
    868 /*
    869  * Veriexec remove policy code.
    870  */
    871 int
    872 veriexec_removechk(struct lwp *l, struct vnode *vp, const char *pathbuf)
    873 {
    874 	struct veriexec_file_entry *vfe;
    875 	int error;
    876 
    877 	if (veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
    878 		return 0;
    879 
    880 	rw_enter(&veriexec_op_lock, RW_READER);
    881 	vfe = veriexec_get(vp);
    882 	rw_exit(&veriexec_op_lock);
    883 
    884 	if (vfe == NULL) {
    885 		/* Lockdown mode: Deny access to non-monitored files. */
    886 		if (veriexec_strict >= VERIEXEC_LOCKDOWN)
    887 			return (EPERM);
    888 
    889 		return (0);
    890 	}
    891 
    892 	veriexec_file_report(vfe, "Remove request.", pathbuf, l,
    893 	    REPORT_ALWAYS|REPORT_ALARM);
    894 
    895 	/* IDS mode: Deny removal of monitored files. */
    896 	if (veriexec_strict >= VERIEXEC_IDS)
    897 		error = EPERM;
    898 	else
    899 		error = veriexec_file_delete(l, vp);
    900 
    901 	return error;
    902 }
    903 
    904 /*
    905  * Veriexec rename policy.
    906  *
    907  * XXX: Once there's a way to hook after a successful rename, it would be
    908  * XXX: nice to update vfe->filename to the new name if it's not NULL and
    909  * XXX: the new name is absolute (ie., starts with a slash).
    910  */
    911 int
    912 veriexec_renamechk(struct lwp *l, struct vnode *fromvp, const char *fromname,
    913     struct vnode *tovp, const char *toname)
    914 {
    915 	struct veriexec_file_entry *fvfe = NULL, *tvfe = NULL;
    916 
    917 	if (veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
    918 		return 0;
    919 
    920 	rw_enter(&veriexec_op_lock, RW_READER);
    921 
    922 	if (veriexec_strict >= VERIEXEC_LOCKDOWN) {
    923 		log(LOG_ALERT, "Veriexec: Preventing rename of `%s' to "
    924 		    "`%s', uid=%u, pid=%u: Lockdown mode.\n", fromname, toname,
    925 		    kauth_cred_geteuid(l->l_cred), l->l_proc->p_pid);
    926 		rw_exit(&veriexec_op_lock);
    927 		return (EPERM);
    928 	}
    929 
    930 	fvfe = veriexec_get(fromvp);
    931 	if (tovp != NULL)
    932 		tvfe = veriexec_get(tovp);
    933 
    934 	if ((fvfe == NULL) && (tvfe == NULL)) {
    935 		/* None of them is monitored */
    936 		rw_exit(&veriexec_op_lock);
    937 		return 0;
    938 	}
    939 
    940 	if (veriexec_strict >= VERIEXEC_IPS) {
    941 		log(LOG_ALERT, "Veriexec: Preventing rename of `%s' "
    942 		    "to `%s', uid=%u, pid=%u: IPS mode, %s "
    943 		    "monitored.\n", fromname, toname,
    944 		    kauth_cred_geteuid(l->l_cred),
    945 		    l->l_proc->p_pid, (fvfe != NULL && tvfe != NULL) ?
    946 		    "files" : "file");
    947 		rw_exit(&veriexec_op_lock);
    948 		return (EPERM);
    949 	}
    950 
    951 	if (fvfe != NULL) {
    952 		/*
    953 		 * Monitored file is renamed; filename no longer relevant.
    954 		 */
    955 
    956 		/*
    957 		 * XXX: We could keep the buffer, and when (and if) updating the
    958 		 * XXX: filename post-rename, re-allocate it only if it's not
    959 		 * XXX: big enough for the new filename.
    960 		 */
    961 
    962 		/* XXX: Get write lock on fvfe here? */
    963 
    964 		VERIEXEC_RW_UPGRADE(&veriexec_op_lock);
    965 		/* once we have the op lock in write mode
    966 		 * there should be no locks on any file
    967 		 * entries so we can destroy the object.
    968 		 */
    969 
    970 		if (fvfe->filename_len > 0)
    971 			kmem_free(fvfe->filename, fvfe->filename_len);
    972 
    973 		fvfe->filename = NULL;
    974 		fvfe->filename_len = 0;
    975 
    976 		rw_downgrade(&veriexec_op_lock);
    977 	}
    978 
    979 	log(LOG_NOTICE, "Veriexec: %s file `%s' renamed to "
    980 	    "%s file `%s', uid=%u, pid=%u.\n", (fvfe != NULL) ?
    981 	    "Monitored" : "Non-monitored", fromname, (tvfe != NULL) ?
    982 	    "monitored" : "non-monitored", toname,
    983 	    kauth_cred_geteuid(l->l_cred), l->l_proc->p_pid);
    984 
    985 	rw_exit(&veriexec_op_lock);
    986 
    987 	if (tvfe != NULL) {
    988 		/*
    989 		 * Monitored file is overwritten. Remove the entry.
    990 		 */
    991 		(void)veriexec_file_delete(l, tovp);
    992 	}
    993 
    994 	return (0);
    995 }
    996 
    997 static void
    998 veriexec_file_free(struct veriexec_file_entry *vfe)
    999 {
   1000 	if (vfe != NULL) {
   1001 		if (vfe->fp != NULL)
   1002 			kmem_free(vfe->fp, vfe->ops->hash_len);
   1003 		if (vfe->page_fp != NULL)
   1004 			kmem_free(vfe->page_fp, vfe->ops->hash_len);
   1005 		if (vfe->filename != NULL)
   1006 			kmem_free(vfe->filename, vfe->filename_len);
   1007 		rw_destroy(&vfe->lock);
   1008 		kmem_free(vfe, sizeof(*vfe));
   1009 	}
   1010 }
   1011 
   1012 static void
   1013 veriexec_file_purge(struct veriexec_file_entry *vfe, int have_lock)
   1014 {
   1015 	if (vfe == NULL)
   1016 		return;
   1017 
   1018 	if (have_lock == VERIEXEC_UNLOCKED)
   1019 		rw_enter(&vfe->lock, RW_WRITER);
   1020 	else
   1021 		VERIEXEC_RW_UPGRADE(&vfe->lock);
   1022 
   1023 	vfe->status = FINGERPRINT_NOTEVAL;
   1024 	if (have_lock == VERIEXEC_UNLOCKED)
   1025 		rw_exit(&vfe->lock);
   1026 	else
   1027 		rw_downgrade(&vfe->lock);
   1028 }
   1029 
   1030 static void
   1031 veriexec_file_purge_cb(struct veriexec_file_entry *vfe, void *cookie)
   1032 {
   1033 	veriexec_file_purge(vfe, VERIEXEC_UNLOCKED);
   1034 }
   1035 
   1036 /*
   1037  * Invalidate a Veriexec file entry.
   1038  * XXX: This should be updated when per-page fingerprints are added.
   1039  */
   1040 void
   1041 veriexec_purge(struct vnode *vp)
   1042 {
   1043 	rw_enter(&veriexec_op_lock, RW_READER);
   1044 	veriexec_file_purge(veriexec_get(vp), VERIEXEC_UNLOCKED);
   1045 	rw_exit(&veriexec_op_lock);
   1046 }
   1047 
   1048 /*
   1049  * Enforce raw disk access policy.
   1050  *
   1051  * IDS mode: Invalidate fingerprints on a mount if it's opened for writing.
   1052  * IPS mode: Don't allow raw writing to disks we monitor.
   1053  * Lockdown mode: Don't allow raw writing to all disks.
   1054  *
   1055  * XXX: This is bogus. There's an obvious race condition between the time
   1056  * XXX: the disk is open for writing, in which an attacker can access a
   1057  * XXX: monitored file to get its signature cached again, and when the raw
   1058  * XXX: file is overwritten on disk.
   1059  * XXX:
   1060  * XXX: To solve this, we need something like the following:
   1061  * XXX:		open raw disk:
   1062  * XXX:		  - raise refcount,
   1063  * XXX:		  - invalidate fingerprints,
   1064  * XXX:		  - mark all entries for that disk with "no cache" flag
   1065  * XXX:
   1066  * XXX:		veriexec_verify:
   1067  * XXX:		  - if "no cache", don't cache evaluation result
   1068  * XXX:
   1069  * XXX:		close raw disk:
   1070  * XXX:		  - lower refcount,
   1071  * XXX:		  - if refcount == 0, remove "no cache" flag from all entries
   1072  */
   1073 static int
   1074 veriexec_raw_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
   1075     void *arg0, void *arg1, void *arg2, void *arg3)
   1076 {
   1077 	int result;
   1078 	enum kauth_device_req req;
   1079 	struct veriexec_table_entry *vte;
   1080 
   1081 	result = KAUTH_RESULT_DENY;
   1082 	req = (enum kauth_device_req)arg0;
   1083 
   1084 	switch (action) {
   1085 	case KAUTH_DEVICE_RAWIO_SPEC: {
   1086 		struct vnode *vp, *bvp;
   1087 		int error;
   1088 
   1089 		if (req == KAUTH_REQ_DEVICE_RAWIO_SPEC_READ) {
   1090 			result = KAUTH_RESULT_DEFER;
   1091 			break;
   1092 		}
   1093 
   1094 		vp = arg1;
   1095 		KASSERT(vp != NULL);
   1096 
   1097 		/* Handle /dev/mem and /dev/kmem. */
   1098 		if (iskmemvp(vp)) {
   1099 			if (veriexec_strict < VERIEXEC_IPS)
   1100 				result = KAUTH_RESULT_DEFER;
   1101 
   1102 			break;
   1103 		}
   1104 
   1105 		error = rawdev_mounted(vp, &bvp);
   1106 		if (error == EINVAL) {
   1107 			result = KAUTH_RESULT_DEFER;
   1108 			break;
   1109 		}
   1110 
   1111 		/*
   1112 		 * XXX: See vfs_mountedon() comment in rawdev_mounted().
   1113 		 */
   1114 		vte = veriexec_table_lookup(bvp->v_mount);
   1115 		if (vte == NULL) {
   1116 			result = KAUTH_RESULT_DEFER;
   1117 			break;
   1118 		}
   1119 
   1120 		switch (veriexec_strict) {
   1121 		case VERIEXEC_LEARNING:
   1122 		case VERIEXEC_IDS:
   1123 			result = KAUTH_RESULT_DEFER;
   1124 
   1125 			rw_enter(&veriexec_op_lock, RW_WRITER);
   1126 			fileassoc_table_run(bvp->v_mount, veriexec_hook,
   1127 			    (fileassoc_cb_t)veriexec_file_purge_cb, NULL);
   1128 			rw_exit(&veriexec_op_lock);
   1129 
   1130 			break;
   1131 		case VERIEXEC_IPS:
   1132 			result = KAUTH_RESULT_DENY;
   1133 			break;
   1134 		case VERIEXEC_LOCKDOWN:
   1135 			result = KAUTH_RESULT_DENY;
   1136 			break;
   1137 		}
   1138 
   1139 		break;
   1140 		}
   1141 
   1142 	case KAUTH_DEVICE_RAWIO_PASSTHRU:
   1143 		/* XXX What can we do here? */
   1144 		if (veriexec_strict < VERIEXEC_IPS)
   1145 			result = KAUTH_RESULT_DEFER;
   1146 
   1147 		break;
   1148 
   1149 	default:
   1150 		result = KAUTH_RESULT_DEFER;
   1151 		break;
   1152 	}
   1153 
   1154 	return (result);
   1155 }
   1156 
   1157 /*
   1158  * Create a new Veriexec table.
   1159  */
   1160 static struct veriexec_table_entry *
   1161 veriexec_table_add(struct lwp *l, struct mount *mp)
   1162 {
   1163 	struct veriexec_table_entry *vte;
   1164 	u_char buf[16];
   1165 
   1166 	vte = kmem_zalloc(sizeof(*vte), KM_SLEEP);
   1167 	mount_setspecific(mp, veriexec_mountspecific_key, vte);
   1168 
   1169 	snprintf(buf, sizeof(buf), "table%u", veriexec_tablecount++);
   1170 	sysctl_createv(NULL, 0, &veriexec_count_node, &vte->vte_node,
   1171 		       0, CTLTYPE_NODE, buf, NULL, NULL, 0, NULL,
   1172 		       0, CTL_CREATE, CTL_EOL);
   1173 
   1174 	sysctl_createv(NULL, 0, &vte->vte_node, NULL,
   1175 		       CTLFLAG_READONLY, CTLTYPE_STRING, "mntpt",
   1176 		       NULL, NULL, 0, mp->mnt_stat.f_mntonname,
   1177 		       0, CTL_CREATE, CTL_EOL);
   1178 	sysctl_createv(NULL, 0, &vte->vte_node, NULL,
   1179 		       CTLFLAG_READONLY, CTLTYPE_STRING, "fstype",
   1180 		       NULL, NULL, 0, mp->mnt_stat.f_fstypename,
   1181 		       0, CTL_CREATE, CTL_EOL);
   1182 	sysctl_createv(NULL, 0, &vte->vte_node, NULL,
   1183 		       CTLFLAG_READONLY, CTLTYPE_QUAD, "nentries",
   1184 		       NULL, NULL, 0, &vte->vte_count, 0, CTL_CREATE, CTL_EOL);
   1185 
   1186 	return (vte);
   1187 }
   1188 
   1189 /*
   1190  * Add a file to be monitored by Veriexec.
   1191  *
   1192  * Expected elements in dict: file, fp, fp-type, entry-type.
   1193  */
   1194 int
   1195 veriexec_file_add(struct lwp *l, prop_dictionary_t dict)
   1196 {
   1197 	struct veriexec_table_entry *vte;
   1198 	struct veriexec_file_entry *vfe = NULL;
   1199 	struct vnode *vp;
   1200 	const char *file, *fp_type;
   1201 	int error;
   1202 
   1203 	if (!prop_dictionary_get_cstring_nocopy(dict, "file", &file))
   1204 		return (EINVAL);
   1205 
   1206 	error = namei_simple_kernel(file, NSM_FOLLOW_NOEMULROOT, &vp);
   1207 	if (error)
   1208 		return (error);
   1209 
   1210 	/* Add only regular files. */
   1211 	if (vp->v_type != VREG) {
   1212 		log(LOG_ERR, "Veriexec: Not adding `%s': Not a regular file.\n",
   1213 		    file);
   1214 		error = EBADF;
   1215 		goto out;
   1216 	}
   1217 
   1218 	vfe = kmem_zalloc(sizeof(*vfe), KM_SLEEP);
   1219 	rw_init(&vfe->lock);
   1220 
   1221 	/* Lookup fingerprint hashing algorithm. */
   1222 	fp_type = prop_string_cstring_nocopy(prop_dictionary_get(dict,
   1223 	    "fp-type"));
   1224 	if ((vfe->ops = veriexec_fpops_lookup(fp_type)) == NULL) {
   1225 		log(LOG_ERR, "Veriexec: Invalid or unknown fingerprint type "
   1226 		    "`%s' for file `%s'.\n", fp_type, file);
   1227 		error = EOPNOTSUPP;
   1228 		goto out;
   1229 	}
   1230 
   1231 	if (prop_data_size(prop_dictionary_get(dict, "fp")) !=
   1232 	    vfe->ops->hash_len) {
   1233 		log(LOG_ERR, "Veriexec: Bad fingerprint length for `%s'.\n",
   1234 		    file);
   1235 		error = EINVAL;
   1236 		goto out;
   1237 	}
   1238 
   1239 	vfe->fp = kmem_alloc(vfe->ops->hash_len, KM_SLEEP);
   1240 	memcpy(vfe->fp, prop_data_data_nocopy(prop_dictionary_get(dict, "fp")),
   1241 	    vfe->ops->hash_len);
   1242 
   1243 	rw_enter(&veriexec_op_lock, RW_WRITER);
   1244 
   1245 	if (veriexec_get(vp)) {
   1246 		/* We already have an entry for this file. */
   1247 		error = EEXIST;
   1248 		goto unlock_out;
   1249 	}
   1250 
   1251 	/* Continue entry initialization. */
   1252 	if (prop_dictionary_get_uint8(dict, "entry-type", &vfe->type) == FALSE)
   1253 		vfe->type = 0;
   1254 	else {
   1255 		uint8_t extra_flags;
   1256 
   1257 		extra_flags = vfe->type & ~(VERIEXEC_DIRECT |
   1258 		    VERIEXEC_INDIRECT | VERIEXEC_FILE | VERIEXEC_UNTRUSTED);
   1259 		if (extra_flags) {
   1260 			log(LOG_NOTICE, "Veriexec: Contaminated flags `0x%x' "
   1261 			    "for `%s', skipping.\n", extra_flags, file);
   1262 			error = EINVAL;
   1263 			goto unlock_out;
   1264 		}
   1265 	}
   1266 	if (!(vfe->type & (VERIEXEC_DIRECT | VERIEXEC_INDIRECT |
   1267 	    VERIEXEC_FILE)))
   1268 		vfe->type |= VERIEXEC_DIRECT;
   1269 
   1270 	vfe->status = FINGERPRINT_NOTEVAL;
   1271 	if (prop_bool_true(prop_dictionary_get(dict, "keep-filename"))) {
   1272 		vfe->filename_len = strlen(file) + 1;
   1273 		vfe->filename = kmem_alloc(vfe->filename_len, KM_SLEEP);
   1274 		strlcpy(vfe->filename, file, vfe->filename_len);
   1275 	} else
   1276 		vfe->filename = NULL;
   1277 
   1278 	vfe->page_fp = NULL;
   1279 	vfe->page_fp_status = PAGE_FP_NONE;
   1280 	vfe->npages = 0;
   1281 	vfe->last_page_size = 0;
   1282 
   1283 	if (prop_bool_true(prop_dictionary_get(dict, "eval-on-load")) ||
   1284 	    (vfe->type & VERIEXEC_UNTRUSTED)) {
   1285 		u_char status;
   1286 
   1287 		error = veriexec_fp_status(l, vp, VERIEXEC_FILE_UNLOCKED,
   1288 		    vfe, &status);
   1289 		if (error)
   1290 			goto unlock_out;
   1291 		vfe->status = status;
   1292 	}
   1293 
   1294 	vte = veriexec_table_lookup(vp->v_mount);
   1295 	if (vte == NULL)
   1296 		vte = veriexec_table_add(l, vp->v_mount);
   1297 
   1298 	/* XXX if we bail below this, we might want to gc newly created vtes. */
   1299 
   1300 	error = fileassoc_add(vp, veriexec_hook, vfe);
   1301 	if (error)
   1302 		goto unlock_out;
   1303 
   1304 	vte->vte_count++;
   1305 
   1306 	veriexec_file_report(NULL, "New entry.", file, NULL, REPORT_DEBUG);
   1307 	veriexec_bypass = 0;
   1308 
   1309   unlock_out:
   1310 	rw_exit(&veriexec_op_lock);
   1311 
   1312   out:
   1313 	vrele(vp);
   1314 	if (error)
   1315 		veriexec_file_free(vfe);
   1316 
   1317 	return (error);
   1318 }
   1319 
   1320 int
   1321 veriexec_table_delete(struct lwp *l, struct mount *mp)
   1322 {
   1323 	struct veriexec_table_entry *vte;
   1324 
   1325 	vte = veriexec_table_lookup(mp);
   1326 	if (vte == NULL)
   1327 		return (ENOENT);
   1328 
   1329 	veriexec_mountspecific_dtor(vte);
   1330 	mount_setspecific(mp, veriexec_mountspecific_key, NULL);
   1331 
   1332 	return (fileassoc_table_clear(mp, veriexec_hook));
   1333 }
   1334 
   1335 int
   1336 veriexec_file_delete(struct lwp *l, struct vnode *vp)
   1337 {
   1338 	struct veriexec_table_entry *vte;
   1339 	int error;
   1340 
   1341 	vte = veriexec_table_lookup(vp->v_mount);
   1342 	if (vte == NULL)
   1343 		return (ENOENT);
   1344 
   1345 	rw_enter(&veriexec_op_lock, RW_WRITER);
   1346 	error = fileassoc_clear(vp, veriexec_hook);
   1347 	rw_exit(&veriexec_op_lock);
   1348 	if (!error) {
   1349 		KASSERT(vte->vte_count > 0);
   1350 		vte->vte_count--;
   1351 	}
   1352 
   1353 	return (error);
   1354 }
   1355 
   1356 /*
   1357  * Convert Veriexec entry data to a dictionary readable by userland tools.
   1358  */
   1359 static void
   1360 veriexec_file_convert(struct veriexec_file_entry *vfe, prop_dictionary_t rdict)
   1361 {
   1362 	if (vfe->filename)
   1363 		prop_dictionary_set(rdict, "file",
   1364 		    prop_string_create_cstring(vfe->filename));
   1365 	prop_dictionary_set_uint8(rdict, "entry-type", vfe->type);
   1366 	prop_dictionary_set_uint8(rdict, "status", vfe->status);
   1367 	prop_dictionary_set(rdict, "fp-type",
   1368 	    prop_string_create_cstring(vfe->ops->type));
   1369 	prop_dictionary_set(rdict, "fp",
   1370 	    prop_data_create_data(vfe->fp, vfe->ops->hash_len));
   1371 }
   1372 
   1373 int
   1374 veriexec_convert(struct vnode *vp, prop_dictionary_t rdict)
   1375 {
   1376 	struct veriexec_file_entry *vfe;
   1377 
   1378 	rw_enter(&veriexec_op_lock, RW_READER);
   1379 
   1380 	vfe = veriexec_get(vp);
   1381 	if (vfe == NULL) {
   1382 		rw_exit(&veriexec_op_lock);
   1383 		return (ENOENT);
   1384 	}
   1385 
   1386 	rw_enter(&vfe->lock, RW_READER);
   1387 	veriexec_file_convert(vfe, rdict);
   1388 	rw_exit(&vfe->lock);
   1389 
   1390 	rw_exit(&veriexec_op_lock);
   1391 	return (0);
   1392 }
   1393 
   1394 int
   1395 veriexec_unmountchk(struct mount *mp)
   1396 {
   1397 	int error;
   1398 
   1399 	if ((veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
   1400 	    || doing_shutdown)
   1401 		return (0);
   1402 
   1403 	rw_enter(&veriexec_op_lock, RW_READER);
   1404 
   1405 	switch (veriexec_strict) {
   1406 	case VERIEXEC_LEARNING:
   1407 		error = 0;
   1408 		break;
   1409 
   1410 	case VERIEXEC_IDS:
   1411 		if (veriexec_table_lookup(mp) != NULL) {
   1412 			log(LOG_INFO, "Veriexec: IDS mode, allowing unmount "
   1413 			    "of \"%s\".\n", mp->mnt_stat.f_mntonname);
   1414 		}
   1415 
   1416 		error = 0;
   1417 		break;
   1418 
   1419 	case VERIEXEC_IPS: {
   1420 		struct veriexec_table_entry *vte;
   1421 
   1422 		vte = veriexec_table_lookup(mp);
   1423 		if ((vte != NULL) && (vte->vte_count > 0)) {
   1424 			log(LOG_ALERT, "Veriexec: IPS mode, preventing"
   1425 			    " unmount of \"%s\" with monitored files.\n",
   1426 			    mp->mnt_stat.f_mntonname);
   1427 
   1428 			error = EPERM;
   1429 		} else
   1430 			error = 0;
   1431 		break;
   1432 		}
   1433 
   1434 	case VERIEXEC_LOCKDOWN:
   1435 	default:
   1436 		log(LOG_ALERT, "Veriexec: Lockdown mode, preventing unmount "
   1437 		    "of \"%s\".\n", mp->mnt_stat.f_mntonname);
   1438 		error = EPERM;
   1439 		break;
   1440 	}
   1441 
   1442 	rw_exit(&veriexec_op_lock);
   1443 	return (error);
   1444 }
   1445 
   1446 int
   1447 veriexec_openchk(struct lwp *l, struct vnode *vp, const char *path, int fmode)
   1448 {
   1449 	struct veriexec_file_entry *vfe = NULL;
   1450 	int error = 0;
   1451 
   1452 	if (veriexec_bypass && (veriexec_strict == VERIEXEC_LEARNING))
   1453 		return 0;
   1454 
   1455 	if (vp == NULL) {
   1456 		/* If no creation requested, let this fail normally. */
   1457 		if (!(fmode & O_CREAT))
   1458 			goto out;
   1459 
   1460 		/* Lockdown mode: Prevent creation of new files. */
   1461 		if (veriexec_strict >= VERIEXEC_LOCKDOWN) {
   1462 			log(LOG_ALERT, "Veriexec: Preventing new file "
   1463 			    "creation in `%s'.\n", path);
   1464 			error = EPERM;
   1465 		}
   1466 
   1467 		goto out;
   1468 	}
   1469 
   1470 	rw_enter(&veriexec_op_lock, RW_READER);
   1471 	error = veriexec_file_verify(l, vp, path, VERIEXEC_FILE,
   1472 				     VERIEXEC_FILE_LOCKED, &vfe);
   1473 
   1474 	if (error) {
   1475 		rw_exit(&veriexec_op_lock);
   1476 		goto out;
   1477 	}
   1478 
   1479 	if ((vfe != NULL) && ((fmode & FWRITE) || (fmode & O_TRUNC))) {
   1480 		veriexec_file_report(vfe, "Write access request.", path, l,
   1481 		    REPORT_ALWAYS | REPORT_ALARM);
   1482 
   1483 		/* IPS mode: Deny write access to monitored files. */
   1484 		if (veriexec_strict >= VERIEXEC_IPS)
   1485 			error = EPERM;
   1486 		else
   1487 			veriexec_file_purge(vfe, VERIEXEC_LOCKED);
   1488 	}
   1489 
   1490 	if (vfe != NULL)
   1491 		rw_exit(&vfe->lock);
   1492 
   1493 	rw_exit(&veriexec_op_lock);
   1494  out:
   1495 	return (error);
   1496 }
   1497 
   1498 static void
   1499 veriexec_file_dump(struct veriexec_file_entry *vfe, prop_array_t entries)
   1500 {
   1501 	prop_dictionary_t entry;
   1502 
   1503 	/* If we don't have a filename, this is meaningless. */
   1504 	if (vfe->filename == NULL)
   1505 		return;
   1506 
   1507 	entry = prop_dictionary_create();
   1508 
   1509 	veriexec_file_convert(vfe, entry);
   1510 
   1511 	prop_array_add(entries, entry);
   1512 }
   1513 
   1514 int
   1515 veriexec_dump(struct lwp *l, prop_array_t rarray)
   1516 {
   1517 	struct mount *mp, *nmp;
   1518 
   1519 	mutex_enter(&mountlist_lock);
   1520 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
   1521 		/* If it fails, the file-system is [being] unmounted. */
   1522 		if (vfs_busy(mp, &nmp) != 0)
   1523 			continue;
   1524 
   1525 		fileassoc_table_run(mp, veriexec_hook,
   1526 		    (fileassoc_cb_t)veriexec_file_dump, rarray);
   1527 
   1528 		vfs_unbusy(mp, false, &nmp);
   1529 	}
   1530 	mutex_exit(&mountlist_lock);
   1531 
   1532 	return (0);
   1533 }
   1534 
   1535 int
   1536 veriexec_flush(struct lwp *l)
   1537 {
   1538 	struct mount *mp, *nmp;
   1539 	int error = 0;
   1540 
   1541 	mutex_enter(&mountlist_lock);
   1542 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
   1543 		int lerror;
   1544 
   1545 		/* If it fails, the file-system is [being] unmounted. */
   1546 		if (vfs_busy(mp, &nmp) != 0)
   1547 			continue;
   1548 
   1549 		lerror = veriexec_table_delete(l, mp);
   1550 		if (lerror && lerror != ENOENT)
   1551 			error = lerror;
   1552 
   1553 		vfs_unbusy(mp, false, &nmp);
   1554 	}
   1555 	mutex_exit(&mountlist_lock);
   1556 
   1557 	return (error);
   1558 }
   1559