1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 1995, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 #include <sys/types.h>
27 #include <sys/t_lock.h>
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/buf.h>
31 #include <sys/conf.h>
32 #include <sys/cred.h>
33 #include <sys/kmem.h>
34 #include <sys/sysmacros.h>
35 #include <sys/vfs.h>
36 #include <sys/vfs_opreg.h>
37 #include <sys/vnode.h>
38 #include <sys/debug.h>
39 #include <sys/errno.h>
40 #include <sys/time.h>
41 #include <sys/file.h>
42 #include <sys/open.h>
43 #include <sys/user.h>
44 #include <sys/termios.h>
45 #include <sys/stream.h>
46 #include <sys/strsubr.h>
47 #include <sys/strsun.h>
48 #include <sys/esunddi.h>
49 #include <sys/flock.h>
50 #include <sys/modctl.h>
51 #include <sys/cmn_err.h>
52 #include <sys/mkdev.h>
53 #include <sys/pathname.h>
54 #include <sys/ddi.h>
55 #include <sys/stat.h>
56 #include <sys/fs/snode.h>
57 #include <sys/fs/dv_node.h>
58 #include <sys/zone.h>
59
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <netinet/in.h>
63 #include <sys/un.h>
64 #include <sys/ucred.h>
65
66 #include <sys/tiuser.h>
67 #define _SUN_TPI_VERSION 2
68 #include <sys/tihdr.h>
69
70 #include <c2/audit.h>
71
72 #include <fs/sockfs/nl7c.h>
73 #include <fs/sockfs/sockcommon.h>
74 #include <fs/sockfs/sockfilter_impl.h>
75 #include <fs/sockfs/socktpi.h>
76 #include <fs/sockfs/socktpi_impl.h>
77 #include <fs/sockfs/sodirect.h>
78
79 /*
80 * Macros that operate on struct cmsghdr.
81 * The CMSG_VALID macro does not assume that the last option buffer is padded.
82 */
83 #define CMSG_CONTENT(cmsg) (&((cmsg)[1]))
84 #define CMSG_CONTENTLEN(cmsg) ((cmsg)->cmsg_len - sizeof (struct cmsghdr))
85 #define CMSG_VALID(cmsg, start, end) \
86 (ISALIGNED_cmsghdr(cmsg) && \
87 ((uintptr_t)(cmsg) >= (uintptr_t)(start)) && \
88 ((uintptr_t)(cmsg) < (uintptr_t)(end)) && \
89 ((ssize_t)(cmsg)->cmsg_len >= sizeof (struct cmsghdr)) && \
90 ((uintptr_t)(cmsg) + (cmsg)->cmsg_len <= (uintptr_t)(end)))
91 #define SO_LOCK_WAKEUP_TIME 3000 /* Wakeup time in milliseconds */
92
93 dev_t sockdev; /* For fsid in getattr */
94 int sockfs_defer_nl7c_init = 0;
95
96 struct socklist socklist;
97
98 struct kmem_cache *socket_cache;
99
100 /*
101 * sockconf_lock protects the socket configuration (socket types and
102 * socket filters) which is changed via the sockconfig system call.
103 */
104 krwlock_t sockconf_lock;
105
106 static int sockfs_update(kstat_t *, int);
107 static int sockfs_snapshot(kstat_t *, void *, int);
108 extern smod_info_t *sotpi_smod_create(void);
109
110 extern void sendfile_init();
111
112 extern void nl7c_init(void);
113
114 extern int modrootloaded;
115
116 /*
117 * Translate from a device pathname (e.g. "/dev/tcp") to a vnode.
118 * Returns with the vnode held.
119 */
120 int
121 sogetvp(char *devpath, vnode_t **vpp, int uioflag)
122 {
123 struct snode *csp;
124 vnode_t *vp, *dvp;
125 major_t maj;
126 int error;
127
128 ASSERT(uioflag == UIO_SYSSPACE || uioflag == UIO_USERSPACE);
129
130 /*
131 * Lookup the underlying filesystem vnode.
132 */
133 error = lookupname(devpath, uioflag, FOLLOW, NULLVPP, &vp);
134 if (error)
135 return (error);
136
137 /* Check that it is the correct vnode */
138 if (vp->v_type != VCHR) {
139 VN_RELE(vp);
140 return (ENOTSOCK);
141 }
142
143 /*
144 * If devpath went through devfs, the device should already
145 * be configured. If devpath is a mknod file, however, we
146 * need to make sure the device is properly configured.
147 * To do this, we do something similar to spec_open()
148 * except that we resolve to the minor/leaf level since
149 * we need to return a vnode.
150 */
151 csp = VTOS(VTOS(vp)->s_commonvp);
152 if (!(csp->s_flag & SDIPSET)) {
153 char *pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
154 error = ddi_dev_pathname(vp->v_rdev, S_IFCHR, pathname);
155 if (error == 0)
156 error = devfs_lookupname(pathname, NULLVPP, &dvp);
157 VN_RELE(vp);
158 kmem_free(pathname, MAXPATHLEN);
159 if (error != 0)
160 return (ENXIO);
161 vp = dvp; /* use the devfs vp */
162 }
163
164 /* device is configured at this point */
165 maj = getmajor(vp->v_rdev);
166 if (!STREAMSTAB(maj)) {
167 VN_RELE(vp);
168 return (ENOSTR);
169 }
170
171 *vpp = vp;
172 return (0);
173 }
174
175 /*
176 * Update the accessed, updated, or changed times in an sonode
177 * with the current time.
178 *
179 * Note that both SunOS 4.X and 4.4BSD sockets do not present reasonable
180 * attributes in a fstat call. (They return the current time and 0 for
181 * all timestamps, respectively.) We maintain the current timestamps
182 * here primarily so that should sockmod be popped the resulting
183 * file descriptor will behave like a stream w.r.t. the timestamps.
184 */
185 void
186 so_update_attrs(struct sonode *so, int flag)
187 {
188 time_t now = gethrestime_sec();
189
190 if (SOCK_IS_NONSTR(so))
191 return;
192
193 mutex_enter(&so->so_lock);
194 so->so_flag |= flag;
195 if (flag & SOACC)
196 SOTOTPI(so)->sti_atime = now;
197 if (flag & SOMOD)
198 SOTOTPI(so)->sti_mtime = now;
199 mutex_exit(&so->so_lock);
200 }
201
202 extern so_create_func_t sock_comm_create_function;
203 extern so_destroy_func_t sock_comm_destroy_function;
204 /*
205 * Init function called when sockfs is loaded.
206 */
207 int
208 sockinit(int fstype, char *name)
209 {
210 static const fs_operation_def_t sock_vfsops_template[] = {
211 NULL, NULL
212 };
213 int error;
214 major_t dev;
215 char *err_str;
216
217 error = vfs_setfsops(fstype, sock_vfsops_template, NULL);
218 if (error != 0) {
219 zcmn_err(GLOBAL_ZONEID, CE_WARN,
220 "sockinit: bad vfs ops template");
221 return (error);
222 }
223
224 error = vn_make_ops(name, socket_vnodeops_template,
225 &socket_vnodeops);
226 if (error != 0) {
227 err_str = "sockinit: bad socket vnode ops template";
228 /* vn_make_ops() does not reset socktpi_vnodeops on failure. */
229 socket_vnodeops = NULL;
230 goto failure;
231 }
232
233 socket_cache = kmem_cache_create("socket_cache",
234 sizeof (struct sonode), 0, sonode_constructor,
235 sonode_destructor, NULL, NULL, NULL, 0);
236
237 rw_init(&sockconf_lock, NULL, RW_DEFAULT, NULL);
238
239 error = socktpi_init();
240 if (error != 0) {
241 err_str = NULL;
242 goto failure;
243 }
244
245 error = sod_init();
246 if (error != 0) {
247 err_str = NULL;
248 goto failure;
249 }
250
251 /*
252 * Set up the default create and destroy functions
253 */
254 sock_comm_create_function = socket_sonode_create;
255 sock_comm_destroy_function = socket_sonode_destroy;
256
257 /*
258 * Build initial list mapping socket parameters to vnode.
259 */
260 smod_init();
261 smod_add(sotpi_smod_create());
262
263 sockparams_init();
264
265 /*
266 * If sockets are needed before init runs /sbin/soconfig
267 * it is possible to preload the sockparams list here using
268 * calls like:
269 * sockconfig(1,2,3, "/dev/tcp", 0);
270 */
271
272 /*
273 * Create a unique dev_t for use in so_fsid.
274 */
275
276 if ((dev = getudev()) == (major_t)-1)
277 dev = 0;
278 sockdev = makedevice(dev, 0);
279
280 mutex_init(&socklist.sl_lock, NULL, MUTEX_DEFAULT, NULL);
281 sendfile_init();
282 if (!modrootloaded) {
283 sockfs_defer_nl7c_init = 1;
284 } else {
285 nl7c_init();
286 }
287
288 /* Initialize socket filters */
289 sof_init();
290
291 return (0);
292
293 failure:
294 (void) vfs_freevfsops_by_type(fstype);
295 if (socket_vnodeops != NULL)
296 vn_freevnodeops(socket_vnodeops);
297 if (err_str != NULL)
298 zcmn_err(GLOBAL_ZONEID, CE_WARN, err_str);
299 return (error);
300 }
301
302 /*
303 * Caller must hold the mutex. Used to set SOLOCKED.
304 */
305 void
306 so_lock_single(struct sonode *so)
307 {
308 ASSERT(MUTEX_HELD(&so->so_lock));
309
310 while (so->so_flag & (SOLOCKED | SOASYNC_UNBIND)) {
311 cv_wait_stop(&so->so_single_cv, &so->so_lock,
312 SO_LOCK_WAKEUP_TIME);
313 }
314 so->so_flag |= SOLOCKED;
315 }
316
317 /*
318 * Caller must hold the mutex and pass in SOLOCKED or SOASYNC_UNBIND.
319 * Used to clear SOLOCKED or SOASYNC_UNBIND.
320 */
321 void
322 so_unlock_single(struct sonode *so, int flag)
323 {
324 ASSERT(MUTEX_HELD(&so->so_lock));
325 ASSERT(flag & (SOLOCKED|SOASYNC_UNBIND));
326 ASSERT((flag & ~(SOLOCKED|SOASYNC_UNBIND)) == 0);
327 ASSERT(so->so_flag & flag);
328 /*
329 * Process the T_DISCON_IND on sti_discon_ind_mp.
330 *
331 * Call to so_drain_discon_ind will result in so_lock
332 * being dropped and re-acquired later.
333 */
334 if (!SOCK_IS_NONSTR(so)) {
335 sotpi_info_t *sti = SOTOTPI(so);
336
337 if (sti->sti_discon_ind_mp != NULL)
338 so_drain_discon_ind(so);
339 }
340
341 cv_signal(&so->so_single_cv);
342 so->so_flag &= ~flag;
343 }
344
345 /*
346 * Caller must hold the mutex. Used to set SOREADLOCKED.
347 * If the caller wants nonblocking behavior it should set fmode.
348 */
349 int
350 so_lock_read(struct sonode *so, int fmode)
351 {
352 ASSERT(MUTEX_HELD(&so->so_lock));
353
354 while (so->so_flag & SOREADLOCKED) {
355 if (fmode & (FNDELAY|FNONBLOCK))
356 return (EWOULDBLOCK);
357 cv_wait_stop(&so->so_read_cv, &so->so_lock,
358 SO_LOCK_WAKEUP_TIME);
359 }
360 so->so_flag |= SOREADLOCKED;
361 return (0);
362 }
363
364 /*
365 * Like so_lock_read above but allows signals.
366 */
367 int
368 so_lock_read_intr(struct sonode *so, int fmode)
369 {
370 ASSERT(MUTEX_HELD(&so->so_lock));
371
372 while (so->so_flag & SOREADLOCKED) {
373 if (fmode & (FNDELAY|FNONBLOCK))
374 return (EWOULDBLOCK);
375 if (!cv_wait_sig(&so->so_read_cv, &so->so_lock))
376 return (EINTR);
377 }
378 so->so_flag |= SOREADLOCKED;
379 return (0);
380 }
381
382 /*
383 * Caller must hold the mutex. Used to clear SOREADLOCKED,
384 * set in so_lock_read() or so_lock_read_intr().
385 */
386 void
387 so_unlock_read(struct sonode *so)
388 {
389 ASSERT(MUTEX_HELD(&so->so_lock));
390 ASSERT(so->so_flag & SOREADLOCKED);
391
392 cv_signal(&so->so_read_cv);
393 so->so_flag &= ~SOREADLOCKED;
394 }
395
396 /*
397 * Verify that the specified offset falls within the mblk and
398 * that the resulting pointer is aligned.
399 * Returns NULL if not.
400 */
401 void *
402 sogetoff(mblk_t *mp, t_uscalar_t offset,
403 t_uscalar_t length, uint_t align_size)
404 {
405 uintptr_t ptr1, ptr2;
406
407 ASSERT(mp && mp->b_wptr >= mp->b_rptr);
408 ptr1 = (uintptr_t)mp->b_rptr + offset;
409 ptr2 = (uintptr_t)ptr1 + length;
410 if (ptr1 < (uintptr_t)mp->b_rptr || ptr2 > (uintptr_t)mp->b_wptr) {
411 eprintline(0);
412 return (NULL);
413 }
414 if ((ptr1 & (align_size - 1)) != 0) {
415 eprintline(0);
416 return (NULL);
417 }
418 return ((void *)ptr1);
419 }
420
421 /*
422 * Return the AF_UNIX underlying filesystem vnode matching a given name.
423 * Makes sure the sending and the destination sonodes are compatible.
424 * The vnode is returned held.
425 *
426 * The underlying filesystem VSOCK vnode has a v_stream pointer that
427 * references the actual stream head (hence indirectly the actual sonode).
428 */
429 static int
430 so_ux_lookup(struct sonode *so, struct sockaddr_un *soun, int checkaccess,
431 vnode_t **vpp)
432 {
433 vnode_t *vp; /* Underlying filesystem vnode */
434 vnode_t *rvp; /* real vnode */
435 vnode_t *svp; /* sockfs vnode */
436 struct sonode *so2;
437 int error;
438
439 dprintso(so, 1, ("so_ux_lookup(%p) name <%s>\n", (void *)so,
440 soun->sun_path));
441
442 error = lookupname(soun->sun_path, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp);
443 if (error) {
444 eprintsoline(so, error);
445 return (error);
446 }
447
448 /*
449 * Traverse lofs mounts get the real vnode
450 */
451 if (VOP_REALVP(vp, &rvp, NULL) == 0) {
452 VN_HOLD(rvp); /* hold the real vnode */
453 VN_RELE(vp); /* release hold from lookup */
454 vp = rvp;
455 }
456
457 if (vp->v_type != VSOCK) {
458 error = ENOTSOCK;
459 eprintsoline(so, error);
460 goto done2;
461 }
462
463 if (checkaccess) {
464 /*
465 * Check that we have permissions to access the destination
466 * vnode. This check is not done in BSD but it is required
467 * by X/Open.
468 */
469 if (error = VOP_ACCESS(vp, VREAD|VWRITE, 0, CRED(), NULL)) {
470 eprintsoline(so, error);
471 goto done2;
472 }
473 }
474
475 /*
476 * Check if the remote socket has been closed.
477 *
478 * Synchronize with vn_rele_stream by holding v_lock while traversing
479 * v_stream->sd_vnode.
480 */
481 mutex_enter(&vp->v_lock);
482 if (vp->v_stream == NULL) {
483 mutex_exit(&vp->v_lock);
484 if (so->so_type == SOCK_DGRAM)
485 error = EDESTADDRREQ;
486 else
487 error = ECONNREFUSED;
488
489 eprintsoline(so, error);
490 goto done2;
491 }
492 ASSERT(vp->v_stream->sd_vnode);
493 svp = vp->v_stream->sd_vnode;
494 /*
495 * holding v_lock on underlying filesystem vnode and acquiring
496 * it on sockfs vnode. Assumes that no code ever attempts to
497 * acquire these locks in the reverse order.
498 */
499 VN_HOLD(svp);
500 mutex_exit(&vp->v_lock);
501
502 if (svp->v_type != VSOCK) {
503 error = ENOTSOCK;
504 eprintsoline(so, error);
505 goto done;
506 }
507
508 so2 = VTOSO(svp);
509
510 if (so->so_type != so2->so_type) {
511 error = EPROTOTYPE;
512 eprintsoline(so, error);
513 goto done;
514 }
515
516 VN_RELE(svp);
517 *vpp = vp;
518 return (0);
519
520 done:
521 VN_RELE(svp);
522 done2:
523 VN_RELE(vp);
524 return (error);
525 }
526
527 /*
528 * Verify peer address for connect and sendto/sendmsg.
529 * Since sendto/sendmsg would not get synchronous errors from the transport
530 * provider we have to do these ugly checks in the socket layer to
531 * preserve compatibility with SunOS 4.X.
532 */
533 int
534 so_addr_verify(struct sonode *so, const struct sockaddr *name,
535 socklen_t namelen)
536 {
537 int family;
538
539 dprintso(so, 1, ("so_addr_verify(%p, %p, %d)\n",
540 (void *)so, (void *)name, namelen));
541
542 ASSERT(name != NULL);
543
544 family = so->so_family;
545 switch (family) {
546 case AF_INET:
547 if (name->sa_family != family) {
548 eprintsoline(so, EAFNOSUPPORT);
549 return (EAFNOSUPPORT);
550 }
551 if (namelen != (socklen_t)sizeof (struct sockaddr_in)) {
552 eprintsoline(so, EINVAL);
553 return (EINVAL);
554 }
555 break;
556 case AF_INET6: {
557 #ifdef DEBUG
558 struct sockaddr_in6 *sin6;
559 #endif /* DEBUG */
560
561 if (name->sa_family != family) {
562 eprintsoline(so, EAFNOSUPPORT);
563 return (EAFNOSUPPORT);
564 }
565 if (namelen != (socklen_t)sizeof (struct sockaddr_in6)) {
566 eprintsoline(so, EINVAL);
567 return (EINVAL);
568 }
569 #ifdef DEBUG
570 /* Verify that apps don't forget to clear sin6_scope_id etc */
571 sin6 = (struct sockaddr_in6 *)name;
572 if (sin6->sin6_scope_id != 0 &&
573 !IN6_IS_ADDR_LINKSCOPE(&sin6->sin6_addr)) {
574 zcmn_err(getzoneid(), CE_WARN,
575 "connect/send* with uninitialized sin6_scope_id "
576 "(%d) on socket. Pid = %d\n",
577 (int)sin6->sin6_scope_id, (int)curproc->p_pid);
578 }
579 #endif /* DEBUG */
580 break;
581 }
582 case AF_UNIX:
583 if (SOTOTPI(so)->sti_faddr_noxlate) {
584 return (0);
585 }
586 if (namelen < (socklen_t)sizeof (short)) {
587 eprintsoline(so, ENOENT);
588 return (ENOENT);
589 }
590 if (name->sa_family != family) {
591 eprintsoline(so, EAFNOSUPPORT);
592 return (EAFNOSUPPORT);
593 }
594 /* MAXPATHLEN + soun_family + nul termination */
595 if (namelen > (socklen_t)(MAXPATHLEN + sizeof (short) + 1)) {
596 eprintsoline(so, ENAMETOOLONG);
597 return (ENAMETOOLONG);
598 }
599
600 break;
601
602 default:
603 /*
604 * Default is don't do any length or sa_family check
605 * to allow non-sockaddr style addresses.
606 */
607 break;
608 }
609
610 return (0);
611 }
612
613
614 /*
615 * Translate an AF_UNIX sockaddr_un to the transport internal name.
616 * Assumes caller has called so_addr_verify first.
617 */
618 /*ARGSUSED*/
619 int
620 so_ux_addr_xlate(struct sonode *so, struct sockaddr *name,
621 socklen_t namelen, int checkaccess,
622 void **addrp, socklen_t *addrlenp)
623 {
624 int error;
625 struct sockaddr_un *soun;
626 vnode_t *vp;
627 void *addr;
628 socklen_t addrlen;
629 sotpi_info_t *sti = SOTOTPI(so);
630
631 dprintso(so, 1, ("so_ux_addr_xlate(%p, %p, %d, %d)\n",
632 (void *)so, (void *)name, namelen, checkaccess));
633
634 ASSERT(name != NULL);
635 ASSERT(so->so_family == AF_UNIX);
636 ASSERT(!sti->sti_faddr_noxlate);
637 ASSERT(namelen >= (socklen_t)sizeof (short));
638 ASSERT(name->sa_family == AF_UNIX);
639 soun = (struct sockaddr_un *)name;
640 /*
641 * Lookup vnode for the specified path name and verify that
642 * it is a socket.
643 */
644 error = so_ux_lookup(so, soun, checkaccess, &vp);
645 if (error) {
646 eprintsoline(so, error);
647 return (error);
648 }
649 /*
650 * Use the address of the peer vnode as the address to send
651 * to. We release the peer vnode here. In case it has been
652 * closed by the time the T_CONN_REQ or T_UNIDATA_REQ reaches the
653 * transport the message will get an error or be dropped.
654 */
655 sti->sti_ux_faddr.soua_vp = vp;
656 sti->sti_ux_faddr.soua_magic = SOU_MAGIC_EXPLICIT;
657 addr = &sti->sti_ux_faddr;
658 addrlen = (socklen_t)sizeof (sti->sti_ux_faddr);
659 dprintso(so, 1, ("ux_xlate UNIX: addrlen %d, vp %p\n",
660 addrlen, (void *)vp));
661 VN_RELE(vp);
662 *addrp = addr;
663 *addrlenp = (socklen_t)addrlen;
664 return (0);
665 }
666
667 /*
668 * Esballoc free function for messages that contain SO_FILEP option.
669 * Decrement the reference count on the file pointers using closef.
670 */
671 void
672 fdbuf_free(struct fdbuf *fdbuf)
673 {
674 int i;
675 struct file *fp;
676
677 dprint(1, ("fdbuf_free: %d fds\n", fdbuf->fd_numfd));
678 for (i = 0; i < fdbuf->fd_numfd; i++) {
679 /*
680 * We need pointer size alignment for fd_fds. On a LP64
681 * kernel, the required alignment is 8 bytes while
682 * the option headers and values are only 4 bytes
683 * aligned. So its safer to do a bcopy compared to
684 * assigning fdbuf->fd_fds[i] to fp.
685 */
686 bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp));
687 dprint(1, ("fdbuf_free: [%d] = %p\n", i, (void *)fp));
688 (void) closef(fp);
689 }
690 if (fdbuf->fd_ebuf != NULL)
691 kmem_free(fdbuf->fd_ebuf, fdbuf->fd_ebuflen);
692 kmem_free(fdbuf, fdbuf->fd_size);
693 }
694
695 /*
696 * Allocate an esballoc'ed message for AF_UNIX file descriptor passing.
697 * Waits if memory is not available.
698 */
699 mblk_t *
700 fdbuf_allocmsg(int size, struct fdbuf *fdbuf)
701 {
702 uchar_t *buf;
703 mblk_t *mp;
704
705 dprint(1, ("fdbuf_allocmsg: size %d, %d fds\n", size, fdbuf->fd_numfd));
706 buf = kmem_alloc(size, KM_SLEEP);
707 fdbuf->fd_ebuf = (caddr_t)buf;
708 fdbuf->fd_ebuflen = size;
709 fdbuf->fd_frtn.free_func = fdbuf_free;
710 fdbuf->fd_frtn.free_arg = (caddr_t)fdbuf;
711
712 mp = esballoc_wait(buf, size, BPRI_MED, &fdbuf->fd_frtn);
713 mp->b_datap->db_type = M_PROTO;
714 return (mp);
715 }
716
717 /*
718 * Extract file descriptors from a fdbuf.
719 * Return list in rights/rightslen.
720 */
721 /*ARGSUSED*/
722 static int
723 fdbuf_extract(struct fdbuf *fdbuf, void *rights, int rightslen)
724 {
725 int i, fd;
726 int *rp;
727 struct file *fp;
728 int numfd;
729
730 dprint(1, ("fdbuf_extract: %d fds, len %d\n",
731 fdbuf->fd_numfd, rightslen));
732
733 numfd = fdbuf->fd_numfd;
734 ASSERT(rightslen == numfd * (int)sizeof (int));
735
736 /*
737 * Allocate a file descriptor and increment the f_count.
738 * The latter is needed since we always call fdbuf_free
739 * which performs a closef.
740 */
741 rp = (int *)rights;
742 for (i = 0; i < numfd; i++) {
743 if ((fd = ufalloc(0)) == -1)
744 goto cleanup;
745 /*
746 * We need pointer size alignment for fd_fds. On a LP64
747 * kernel, the required alignment is 8 bytes while
748 * the option headers and values are only 4 bytes
749 * aligned. So its safer to do a bcopy compared to
750 * assigning fdbuf->fd_fds[i] to fp.
751 */
752 bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp));
753 mutex_enter(&fp->f_tlock);
754 fp->f_count++;
755 mutex_exit(&fp->f_tlock);
756 setf(fd, fp);
757 *rp++ = fd;
758 /* add curproc to the pid list associated with that file */
759 if (fp->f_vnode != NULL)
760 (void) VOP_IOCTL(fp->f_vnode, F_FORKED,
761 (intptr_t)curproc, FKIOCTL, kcred, NULL, NULL);
762
763 if (AU_AUDITING())
764 audit_fdrecv(fd, fp);
765 dprint(1, ("fdbuf_extract: [%d] = %d, %p refcnt %d\n",
766 i, fd, (void *)fp, fp->f_count));
767 }
768 return (0);
769
770 cleanup:
771 /*
772 * Undo whatever partial work the loop above has done.
773 */
774 {
775 int j;
776
777 rp = (int *)rights;
778 for (j = 0; j < i; j++) {
779 dprint(0,
780 ("fdbuf_extract: cleanup[%d] = %d\n", j, *rp));
781 (void) closeandsetf(*rp++, NULL);
782 }
783 }
784
785 return (EMFILE);
786 }
787
788 /*
789 * Insert file descriptors into an fdbuf.
790 * Returns a kmem_alloc'ed fdbuf. The fdbuf should be freed
791 * by calling fdbuf_free().
792 */
793 int
794 fdbuf_create(void *rights, int rightslen, struct fdbuf **fdbufp)
795 {
796 int numfd, i;
797 int *fds;
798 struct file *fp;
799 struct fdbuf *fdbuf;
800 int fdbufsize;
801
802 dprint(1, ("fdbuf_create: len %d\n", rightslen));
803
804 numfd = rightslen / (int)sizeof (int);
805
806 fdbufsize = (int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *));
807 fdbuf = kmem_alloc(fdbufsize, KM_SLEEP);
808 fdbuf->fd_size = fdbufsize;
809 fdbuf->fd_numfd = 0;
810 fdbuf->fd_ebuf = NULL;
811 fdbuf->fd_ebuflen = 0;
812 fds = (int *)rights;
813 for (i = 0; i < numfd; i++) {
814 if ((fp = getf(fds[i])) == NULL) {
815 fdbuf_free(fdbuf);
816 return (EBADF);
817 }
818 dprint(1, ("fdbuf_create: [%d] = %d, %p refcnt %d\n",
819 i, fds[i], (void *)fp, fp->f_count));
820 mutex_enter(&fp->f_tlock);
821 fp->f_count++;
822 mutex_exit(&fp->f_tlock);
823 /*
824 * The maximum alignment for fdbuf (or any option header
825 * and its value) it 4 bytes. On a LP64 kernel, the alignment
826 * is not sufficient for pointers (fd_fds in this case). Since
827 * we just did a kmem_alloc (we get a double word alignment),
828 * we don't need to do anything on the send side (we loose
829 * the double word alignment because fdbuf goes after an
830 * option header (eg T_unitdata_req) which is only 4 byte
831 * aligned). We take care of this when we extract the file
832 * descriptor in fdbuf_extract or fdbuf_free.
833 */
834 fdbuf->fd_fds[i] = fp;
835 fdbuf->fd_numfd++;
836 releasef(fds[i]);
837 if (AU_AUDITING())
838 audit_fdsend(fds[i], fp, 0);
839 }
840 *fdbufp = fdbuf;
841 return (0);
842 }
843
844 static int
845 fdbuf_optlen(int rightslen)
846 {
847 int numfd;
848
849 numfd = rightslen / (int)sizeof (int);
850
851 return ((int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *)));
852 }
853
854 static t_uscalar_t
855 fdbuf_cmsglen(int fdbuflen)
856 {
857 return (t_uscalar_t)((fdbuflen - FDBUF_HDRSIZE) /
858 (int)sizeof (struct file *) * (int)sizeof (int));
859 }
860
861
862 /*
863 * Return non-zero if the mblk and fdbuf are consistent.
864 */
865 static int
866 fdbuf_verify(mblk_t *mp, struct fdbuf *fdbuf, int fdbuflen)
867 {
868 if (fdbuflen >= FDBUF_HDRSIZE &&
869 fdbuflen == fdbuf->fd_size) {
870 frtn_t *frp = mp->b_datap->db_frtnp;
871 /*
872 * Check that the SO_FILEP portion of the
873 * message has not been modified by
874 * the loopback transport. The sending sockfs generates
875 * a message that is esballoc'ed with the free function
876 * being fdbuf_free() and where free_arg contains the
877 * identical information as the SO_FILEP content.
878 *
879 * If any of these constraints are not satisfied we
880 * silently ignore the option.
881 */
882 ASSERT(mp);
883 if (frp != NULL &&
884 frp->free_func == fdbuf_free &&
885 frp->free_arg != NULL &&
886 bcmp(frp->free_arg, fdbuf, fdbuflen) == 0) {
887 dprint(1, ("fdbuf_verify: fdbuf %p len %d\n",
888 (void *)fdbuf, fdbuflen));
889 return (1);
890 } else {
891 zcmn_err(getzoneid(), CE_WARN,
892 "sockfs: mismatched fdbuf content (%p)",
893 (void *)mp);
894 return (0);
895 }
896 } else {
897 zcmn_err(getzoneid(), CE_WARN,
898 "sockfs: mismatched fdbuf len %d, %d\n",
899 fdbuflen, fdbuf->fd_size);
900 return (0);
901 }
902 }
903
904 /*
905 * When the file descriptors returned by sorecvmsg can not be passed
906 * to the application this routine will cleanup the references on
907 * the files. Start at startoff bytes into the buffer.
908 */
909 static void
910 close_fds(void *fdbuf, int fdbuflen, int startoff)
911 {
912 int *fds = (int *)fdbuf;
913 int numfd = fdbuflen / (int)sizeof (int);
914 int i;
915
916 dprint(1, ("close_fds(%p, %d, %d)\n", fdbuf, fdbuflen, startoff));
917
918 for (i = 0; i < numfd; i++) {
919 if (startoff < 0)
920 startoff = 0;
921 if (startoff < (int)sizeof (int)) {
922 /*
923 * This file descriptor is partially or fully after
924 * the offset
925 */
926 dprint(0,
927 ("close_fds: cleanup[%d] = %d\n", i, fds[i]));
928 (void) closeandsetf(fds[i], NULL);
929 }
930 startoff -= (int)sizeof (int);
931 }
932 }
933
934 /*
935 * Close all file descriptors contained in the control part starting at
936 * the startoffset.
937 */
938 void
939 so_closefds(void *control, t_uscalar_t controllen, int oldflg,
940 int startoff)
941 {
942 struct cmsghdr *cmsg;
943
944 if (control == NULL)
945 return;
946
947 if (oldflg) {
948 close_fds(control, controllen, startoff);
949 return;
950 }
951 /* Scan control part for file descriptors. */
952 for (cmsg = (struct cmsghdr *)control;
953 CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
954 cmsg = CMSG_NEXT(cmsg)) {
955 if (cmsg->cmsg_level == SOL_SOCKET &&
956 cmsg->cmsg_type == SCM_RIGHTS) {
957 close_fds(CMSG_CONTENT(cmsg),
958 (int)CMSG_CONTENTLEN(cmsg),
959 startoff - (int)sizeof (struct cmsghdr));
960 }
961 startoff -= cmsg->cmsg_len;
962 }
963 }
964
965 /*
966 * Returns a pointer/length for the file descriptors contained
967 * in the control buffer. Returns with *fdlenp == -1 if there are no
968 * file descriptor options present. This is different than there being
969 * a zero-length file descriptor option.
970 * Fail if there are multiple SCM_RIGHT cmsgs.
971 */
972 int
973 so_getfdopt(void *control, t_uscalar_t controllen, int oldflg,
974 void **fdsp, int *fdlenp)
975 {
976 struct cmsghdr *cmsg;
977 void *fds;
978 int fdlen;
979
980 if (control == NULL) {
981 *fdsp = NULL;
982 *fdlenp = -1;
983 return (0);
984 }
985
986 if (oldflg) {
987 *fdsp = control;
988 if (controllen == 0)
989 *fdlenp = -1;
990 else
991 *fdlenp = controllen;
992 dprint(1, ("so_getfdopt: old %d\n", *fdlenp));
993 return (0);
994 }
995
996 fds = NULL;
997 fdlen = 0;
998
999 for (cmsg = (struct cmsghdr *)control;
1000 CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1001 cmsg = CMSG_NEXT(cmsg)) {
1002 if (cmsg->cmsg_level == SOL_SOCKET &&
1003 cmsg->cmsg_type == SCM_RIGHTS) {
1004 if (fds != NULL)
1005 return (EINVAL);
1006 fds = CMSG_CONTENT(cmsg);
1007 fdlen = (int)CMSG_CONTENTLEN(cmsg);
1008 dprint(1, ("so_getfdopt: new %lu\n",
1009 (size_t)CMSG_CONTENTLEN(cmsg)));
1010 }
1011 }
1012 if (fds == NULL) {
1013 dprint(1, ("so_getfdopt: NONE\n"));
1014 *fdlenp = -1;
1015 } else
1016 *fdlenp = fdlen;
1017 *fdsp = fds;
1018 return (0);
1019 }
1020
1021 /*
1022 * Return the length of the options including any file descriptor options.
1023 */
1024 t_uscalar_t
1025 so_optlen(void *control, t_uscalar_t controllen, int oldflg)
1026 {
1027 struct cmsghdr *cmsg;
1028 t_uscalar_t optlen = 0;
1029 t_uscalar_t len;
1030
1031 if (control == NULL)
1032 return (0);
1033
1034 if (oldflg)
1035 return ((t_uscalar_t)(sizeof (struct T_opthdr) +
1036 fdbuf_optlen(controllen)));
1037
1038 for (cmsg = (struct cmsghdr *)control;
1039 CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1040 cmsg = CMSG_NEXT(cmsg)) {
1041 if (cmsg->cmsg_level == SOL_SOCKET &&
1042 cmsg->cmsg_type == SCM_RIGHTS) {
1043 len = fdbuf_optlen((int)CMSG_CONTENTLEN(cmsg));
1044 } else {
1045 len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg);
1046 }
1047 optlen += (t_uscalar_t)(_TPI_ALIGN_TOPT(len) +
1048 sizeof (struct T_opthdr));
1049 }
1050 dprint(1, ("so_optlen: controllen %d, flg %d -> optlen %d\n",
1051 controllen, oldflg, optlen));
1052 return (optlen);
1053 }
1054
1055 /*
1056 * Copy options from control to the mblk. Skip any file descriptor options.
1057 */
1058 void
1059 so_cmsg2opt(void *control, t_uscalar_t controllen, int oldflg, mblk_t *mp)
1060 {
1061 struct T_opthdr toh;
1062 struct cmsghdr *cmsg;
1063
1064 if (control == NULL)
1065 return;
1066
1067 if (oldflg) {
1068 /* No real options - caller has handled file descriptors */
1069 return;
1070 }
1071 for (cmsg = (struct cmsghdr *)control;
1072 CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1073 cmsg = CMSG_NEXT(cmsg)) {
1074 /*
1075 * Note: The caller handles file descriptors prior
1076 * to calling this function.
1077 */
1078 t_uscalar_t len;
1079
1080 if (cmsg->cmsg_level == SOL_SOCKET &&
1081 cmsg->cmsg_type == SCM_RIGHTS)
1082 continue;
1083
1084 len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg);
1085 toh.level = cmsg->cmsg_level;
1086 toh.name = cmsg->cmsg_type;
1087 toh.len = len + (t_uscalar_t)sizeof (struct T_opthdr);
1088 toh.status = 0;
1089
1090 soappendmsg(mp, &toh, sizeof (toh));
1091 soappendmsg(mp, CMSG_CONTENT(cmsg), len);
1092 mp->b_wptr += _TPI_ALIGN_TOPT(len) - len;
1093 ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
1094 }
1095 }
1096
1097 /*
1098 * Return the length of the control message derived from the options.
1099 * Exclude SO_SRCADDR and SO_UNIX_CLOSE options. Include SO_FILEP.
1100 * When oldflg is set only include SO_FILEP.
1101 * so_opt2cmsg and so_cmsglen are inter-related since so_cmsglen
1102 * allocates the space that so_opt2cmsg fills. If one changes, the other should
1103 * also be checked for any possible impacts.
1104 */
1105 t_uscalar_t
1106 so_cmsglen(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg)
1107 {
1108 t_uscalar_t cmsglen = 0;
1109 struct T_opthdr *tohp;
1110 t_uscalar_t len;
1111 t_uscalar_t last_roundup = 0;
1112
1113 ASSERT(__TPI_TOPT_ISALIGNED(opt));
1114
1115 for (tohp = (struct T_opthdr *)opt;
1116 tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1117 tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1118 dprint(1, ("so_cmsglen: level 0x%x, name %d, len %d\n",
1119 tohp->level, tohp->name, tohp->len));
1120 if (tohp->level == SOL_SOCKET &&
1121 (tohp->name == SO_SRCADDR ||
1122 tohp->name == SO_UNIX_CLOSE)) {
1123 continue;
1124 }
1125 if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) {
1126 struct fdbuf *fdbuf;
1127 int fdbuflen;
1128
1129 fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp);
1130 fdbuflen = (int)_TPI_TOPT_DATALEN(tohp);
1131
1132 if (!fdbuf_verify(mp, fdbuf, fdbuflen))
1133 continue;
1134 if (oldflg) {
1135 cmsglen += fdbuf_cmsglen(fdbuflen);
1136 continue;
1137 }
1138 len = fdbuf_cmsglen(fdbuflen);
1139 } else if (tohp->level == SOL_SOCKET &&
1140 tohp->name == SCM_TIMESTAMP) {
1141 if (oldflg)
1142 continue;
1143
1144 if (get_udatamodel() == DATAMODEL_NATIVE) {
1145 len = sizeof (struct timeval);
1146 } else {
1147 len = sizeof (struct timeval32);
1148 }
1149 } else {
1150 if (oldflg)
1151 continue;
1152 len = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp);
1153 }
1154 /*
1155 * Exclude roundup for last option to not set
1156 * MSG_CTRUNC when the cmsg fits but the padding doesn't fit.
1157 */
1158 last_roundup = (t_uscalar_t)
1159 (ROUNDUP_cmsglen(len + (int)sizeof (struct cmsghdr)) -
1160 (len + (int)sizeof (struct cmsghdr)));
1161 cmsglen += (t_uscalar_t)(len + (int)sizeof (struct cmsghdr)) +
1162 last_roundup;
1163 }
1164 cmsglen -= last_roundup;
1165 dprint(1, ("so_cmsglen: optlen %d, flg %d -> cmsglen %d\n",
1166 optlen, oldflg, cmsglen));
1167 return (cmsglen);
1168 }
1169
1170 /*
1171 * Copy options from options to the control. Convert SO_FILEP to
1172 * file descriptors.
1173 * Returns errno or zero.
1174 * so_opt2cmsg and so_cmsglen are inter-related since so_cmsglen
1175 * allocates the space that so_opt2cmsg fills. If one changes, the other should
1176 * also be checked for any possible impacts.
1177 */
1178 int
1179 so_opt2cmsg(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg,
1180 void *control, t_uscalar_t controllen)
1181 {
1182 struct T_opthdr *tohp;
1183 struct cmsghdr *cmsg;
1184 struct fdbuf *fdbuf;
1185 int fdbuflen;
1186 int error;
1187 #if defined(DEBUG) || defined(__lint)
1188 struct cmsghdr *cend = (struct cmsghdr *)
1189 (((uint8_t *)control) + ROUNDUP_cmsglen(controllen));
1190 #endif
1191 cmsg = (struct cmsghdr *)control;
1192
1193 ASSERT(__TPI_TOPT_ISALIGNED(opt));
1194
1195 for (tohp = (struct T_opthdr *)opt;
1196 tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1197 tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1198 dprint(1, ("so_opt2cmsg: level 0x%x, name %d, len %d\n",
1199 tohp->level, tohp->name, tohp->len));
1200
1201 if (tohp->level == SOL_SOCKET &&
1202 (tohp->name == SO_SRCADDR ||
1203 tohp->name == SO_UNIX_CLOSE)) {
1204 continue;
1205 }
1206 ASSERT((uintptr_t)cmsg <= (uintptr_t)control + controllen);
1207 if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) {
1208 fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp);
1209 fdbuflen = (int)_TPI_TOPT_DATALEN(tohp);
1210
1211 if (!fdbuf_verify(mp, fdbuf, fdbuflen))
1212 return (EPROTO);
1213 if (oldflg) {
1214 error = fdbuf_extract(fdbuf, control,
1215 (int)controllen);
1216 if (error != 0)
1217 return (error);
1218 continue;
1219 } else {
1220 int fdlen;
1221
1222 fdlen = (int)fdbuf_cmsglen(
1223 (int)_TPI_TOPT_DATALEN(tohp));
1224
1225 cmsg->cmsg_level = tohp->level;
1226 cmsg->cmsg_type = SCM_RIGHTS;
1227 cmsg->cmsg_len = (socklen_t)(fdlen +
1228 sizeof (struct cmsghdr));
1229
1230 error = fdbuf_extract(fdbuf,
1231 CMSG_CONTENT(cmsg), fdlen);
1232 if (error != 0)
1233 return (error);
1234 }
1235 } else if (tohp->level == SOL_SOCKET &&
1236 tohp->name == SCM_TIMESTAMP) {
1237 timestruc_t *timestamp;
1238
1239 if (oldflg)
1240 continue;
1241
1242 cmsg->cmsg_level = tohp->level;
1243 cmsg->cmsg_type = tohp->name;
1244
1245 timestamp =
1246 (timestruc_t *)P2ROUNDUP((intptr_t)&tohp[1],
1247 sizeof (intptr_t));
1248
1249 if (get_udatamodel() == DATAMODEL_NATIVE) {
1250 struct timeval tv;
1251
1252 cmsg->cmsg_len = sizeof (struct timeval) +
1253 sizeof (struct cmsghdr);
1254 tv.tv_sec = timestamp->tv_sec;
1255 tv.tv_usec = timestamp->tv_nsec /
1256 (NANOSEC / MICROSEC);
1257 /*
1258 * on LP64 systems, the struct timeval in
1259 * the destination will not be 8-byte aligned,
1260 * so use bcopy to avoid alignment trouble
1261 */
1262 bcopy(&tv, CMSG_CONTENT(cmsg), sizeof (tv));
1263 } else {
1264 struct timeval32 *time32;
1265
1266 cmsg->cmsg_len = sizeof (struct timeval32) +
1267 sizeof (struct cmsghdr);
1268 time32 = (struct timeval32 *)CMSG_CONTENT(cmsg);
1269 time32->tv_sec = (time32_t)timestamp->tv_sec;
1270 time32->tv_usec =
1271 (int32_t)(timestamp->tv_nsec /
1272 (NANOSEC / MICROSEC));
1273 }
1274
1275 } else {
1276 if (oldflg)
1277 continue;
1278
1279 cmsg->cmsg_level = tohp->level;
1280 cmsg->cmsg_type = tohp->name;
1281 cmsg->cmsg_len = (socklen_t)(_TPI_TOPT_DATALEN(tohp) +
1282 sizeof (struct cmsghdr));
1283
1284 /* copy content to control data part */
1285 bcopy(&tohp[1], CMSG_CONTENT(cmsg),
1286 CMSG_CONTENTLEN(cmsg));
1287 }
1288 /* move to next CMSG structure! */
1289 cmsg = CMSG_NEXT(cmsg);
1290 }
1291 dprint(1, ("so_opt2cmsg: buf %p len %d; cend %p; final cmsg %p\n",
1292 control, controllen, (void *)cend, (void *)cmsg));
1293 ASSERT(cmsg <= cend);
1294 return (0);
1295 }
1296
1297 /*
1298 * Extract the SO_SRCADDR option value if present.
1299 */
1300 void
1301 so_getopt_srcaddr(void *opt, t_uscalar_t optlen, void **srcp,
1302 t_uscalar_t *srclenp)
1303 {
1304 struct T_opthdr *tohp;
1305
1306 ASSERT(__TPI_TOPT_ISALIGNED(opt));
1307
1308 ASSERT(srcp != NULL && srclenp != NULL);
1309 *srcp = NULL;
1310 *srclenp = 0;
1311
1312 for (tohp = (struct T_opthdr *)opt;
1313 tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1314 tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1315 dprint(1, ("so_getopt_srcaddr: level 0x%x, name %d, len %d\n",
1316 tohp->level, tohp->name, tohp->len));
1317 if (tohp->level == SOL_SOCKET &&
1318 tohp->name == SO_SRCADDR) {
1319 *srcp = _TPI_TOPT_DATA(tohp);
1320 *srclenp = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp);
1321 }
1322 }
1323 }
1324
1325 /*
1326 * Verify if the SO_UNIX_CLOSE option is present.
1327 */
1328 int
1329 so_getopt_unix_close(void *opt, t_uscalar_t optlen)
1330 {
1331 struct T_opthdr *tohp;
1332
1333 ASSERT(__TPI_TOPT_ISALIGNED(opt));
1334
1335 for (tohp = (struct T_opthdr *)opt;
1336 tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1337 tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1338 dprint(1,
1339 ("so_getopt_unix_close: level 0x%x, name %d, len %d\n",
1340 tohp->level, tohp->name, tohp->len));
1341 if (tohp->level == SOL_SOCKET &&
1342 tohp->name == SO_UNIX_CLOSE)
1343 return (1);
1344 }
1345 return (0);
1346 }
1347
1348 /*
1349 * Allocate an M_PROTO message.
1350 *
1351 * If allocation fails the behavior depends on sleepflg:
1352 * _ALLOC_NOSLEEP fail immediately
1353 * _ALLOC_INTR sleep for memory until a signal is caught
1354 * _ALLOC_SLEEP sleep forever. Don't return NULL.
1355 */
1356 mblk_t *
1357 soallocproto(size_t size, int sleepflg, cred_t *cr)
1358 {
1359 mblk_t *mp;
1360
1361 /* Round up size for reuse */
1362 size = MAX(size, 64);
1363 if (cr != NULL)
1364 mp = allocb_cred(size, cr, curproc->p_pid);
1365 else
1366 mp = allocb(size, BPRI_MED);
1367
1368 if (mp == NULL) {
1369 int error; /* Dummy - error not returned to caller */
1370
1371 switch (sleepflg) {
1372 case _ALLOC_SLEEP:
1373 if (cr != NULL) {
1374 mp = allocb_cred_wait(size, STR_NOSIG, &error,
1375 cr, curproc->p_pid);
1376 } else {
1377 mp = allocb_wait(size, BPRI_MED, STR_NOSIG,
1378 &error);
1379 }
1380 ASSERT(mp);
1381 break;
1382 case _ALLOC_INTR:
1383 if (cr != NULL) {
1384 mp = allocb_cred_wait(size, 0, &error, cr,
1385 curproc->p_pid);
1386 } else {
1387 mp = allocb_wait(size, BPRI_MED, 0, &error);
1388 }
1389 if (mp == NULL) {
1390 /* Caught signal while sleeping for memory */
1391 eprintline(ENOBUFS);
1392 return (NULL);
1393 }
1394 break;
1395 case _ALLOC_NOSLEEP:
1396 default:
1397 eprintline(ENOBUFS);
1398 return (NULL);
1399 }
1400 }
1401 DB_TYPE(mp) = M_PROTO;
1402 return (mp);
1403 }
1404
1405 /*
1406 * Allocate an M_PROTO message with a single component.
1407 * len is the length of buf. size is the amount to allocate.
1408 *
1409 * buf can be NULL with a non-zero len.
1410 * This results in a bzero'ed chunk being placed the message.
1411 */
1412 mblk_t *
1413 soallocproto1(const void *buf, ssize_t len, ssize_t size, int sleepflg,
1414 cred_t *cr)
1415 {
1416 mblk_t *mp;
1417
1418 if (size == 0)
1419 size = len;
1420
1421 ASSERT(size >= len);
1422 /* Round up size for reuse */
1423 size = MAX(size, 64);
1424 mp = soallocproto(size, sleepflg, cr);
1425 if (mp == NULL)
1426 return (NULL);
1427 mp->b_datap->db_type = M_PROTO;
1428 if (len != 0) {
1429 if (buf != NULL)
1430 bcopy(buf, mp->b_wptr, len);
1431 else
1432 bzero(mp->b_wptr, len);
1433 mp->b_wptr += len;
1434 }
1435 return (mp);
1436 }
1437
1438 /*
1439 * Append buf/len to mp.
1440 * The caller has to ensure that there is enough room in the mblk.
1441 *
1442 * buf can be NULL with a non-zero len.
1443 * This results in a bzero'ed chunk being placed the message.
1444 */
1445 void
1446 soappendmsg(mblk_t *mp, const void *buf, ssize_t len)
1447 {
1448 ASSERT(mp);
1449
1450 if (len != 0) {
1451 /* Assert for room left */
1452 ASSERT(mp->b_datap->db_lim - mp->b_wptr >= len);
1453 if (buf != NULL)
1454 bcopy(buf, mp->b_wptr, len);
1455 else
1456 bzero(mp->b_wptr, len);
1457 }
1458 mp->b_wptr += len;
1459 }
1460
1461 /*
1462 * Create a message using two kernel buffers.
1463 * If size is set that will determine the allocation size (e.g. for future
1464 * soappendmsg calls). If size is zero it is derived from the buffer
1465 * lengths.
1466 */
1467 mblk_t *
1468 soallocproto2(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2,
1469 ssize_t size, int sleepflg, cred_t *cr)
1470 {
1471 mblk_t *mp;
1472
1473 if (size == 0)
1474 size = len1 + len2;
1475 ASSERT(size >= len1 + len2);
1476
1477 mp = soallocproto1(buf1, len1, size, sleepflg, cr);
1478 if (mp)
1479 soappendmsg(mp, buf2, len2);
1480 return (mp);
1481 }
1482
1483 /*
1484 * Create a message using three kernel buffers.
1485 * If size is set that will determine the allocation size (for future
1486 * soappendmsg calls). If size is zero it is derived from the buffer
1487 * lengths.
1488 */
1489 mblk_t *
1490 soallocproto3(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2,
1491 const void *buf3, ssize_t len3, ssize_t size, int sleepflg, cred_t *cr)
1492 {
1493 mblk_t *mp;
1494
1495 if (size == 0)
1496 size = len1 + len2 +len3;
1497 ASSERT(size >= len1 + len2 + len3);
1498
1499 mp = soallocproto1(buf1, len1, size, sleepflg, cr);
1500 if (mp != NULL) {
1501 soappendmsg(mp, buf2, len2);
1502 soappendmsg(mp, buf3, len3);
1503 }
1504 return (mp);
1505 }
1506
1507 #ifdef DEBUG
1508 char *
1509 pr_state(uint_t state, uint_t mode)
1510 {
1511 static char buf[1024];
1512
1513 buf[0] = 0;
1514 if (state & SS_ISCONNECTED)
1515 (void) strcat(buf, "ISCONNECTED ");
1516 if (state & SS_ISCONNECTING)
1517 (void) strcat(buf, "ISCONNECTING ");
1518 if (state & SS_ISDISCONNECTING)
1519 (void) strcat(buf, "ISDISCONNECTING ");
1520 if (state & SS_CANTSENDMORE)
1521 (void) strcat(buf, "CANTSENDMORE ");
1522
1523 if (state & SS_CANTRCVMORE)
1524 (void) strcat(buf, "CANTRCVMORE ");
1525 if (state & SS_ISBOUND)
1526 (void) strcat(buf, "ISBOUND ");
1527 if (state & SS_NDELAY)
1528 (void) strcat(buf, "NDELAY ");
1529 if (state & SS_NONBLOCK)
1530 (void) strcat(buf, "NONBLOCK ");
1531
1532 if (state & SS_ASYNC)
1533 (void) strcat(buf, "ASYNC ");
1534 if (state & SS_ACCEPTCONN)
1535 (void) strcat(buf, "ACCEPTCONN ");
1536 if (state & SS_SAVEDEOR)
1537 (void) strcat(buf, "SAVEDEOR ");
1538
1539 if (state & SS_RCVATMARK)
1540 (void) strcat(buf, "RCVATMARK ");
1541 if (state & SS_OOBPEND)
1542 (void) strcat(buf, "OOBPEND ");
1543 if (state & SS_HAVEOOBDATA)
1544 (void) strcat(buf, "HAVEOOBDATA ");
1545 if (state & SS_HADOOBDATA)
1546 (void) strcat(buf, "HADOOBDATA ");
1547
1548 if (mode & SM_PRIV)
1549 (void) strcat(buf, "PRIV ");
1550 if (mode & SM_ATOMIC)
1551 (void) strcat(buf, "ATOMIC ");
1552 if (mode & SM_ADDR)
1553 (void) strcat(buf, "ADDR ");
1554 if (mode & SM_CONNREQUIRED)
1555 (void) strcat(buf, "CONNREQUIRED ");
1556
1557 if (mode & SM_FDPASSING)
1558 (void) strcat(buf, "FDPASSING ");
1559 if (mode & SM_EXDATA)
1560 (void) strcat(buf, "EXDATA ");
1561 if (mode & SM_OPTDATA)
1562 (void) strcat(buf, "OPTDATA ");
1563 if (mode & SM_BYTESTREAM)
1564 (void) strcat(buf, "BYTESTREAM ");
1565 return (buf);
1566 }
1567
1568 char *
1569 pr_addr(int family, struct sockaddr *addr, t_uscalar_t addrlen)
1570 {
1571 static char buf[1024];
1572
1573 if (addr == NULL || addrlen == 0) {
1574 (void) sprintf(buf, "(len %d) %p", addrlen, (void *)addr);
1575 return (buf);
1576 }
1577 switch (family) {
1578 case AF_INET: {
1579 struct sockaddr_in sin;
1580
1581 bcopy(addr, &sin, sizeof (sin));
1582
1583 (void) sprintf(buf, "(len %d) %x/%d",
1584 addrlen, ntohl(sin.sin_addr.s_addr), ntohs(sin.sin_port));
1585 break;
1586 }
1587 case AF_INET6: {
1588 struct sockaddr_in6 sin6;
1589 uint16_t *piece = (uint16_t *)&sin6.sin6_addr;
1590
1591 bcopy((char *)addr, (char *)&sin6, sizeof (sin6));
1592 (void) sprintf(buf, "(len %d) %x:%x:%x:%x:%x:%x:%x:%x/%d",
1593 addrlen,
1594 ntohs(piece[0]), ntohs(piece[1]),
1595 ntohs(piece[2]), ntohs(piece[3]),
1596 ntohs(piece[4]), ntohs(piece[5]),
1597 ntohs(piece[6]), ntohs(piece[7]),
1598 ntohs(sin6.sin6_port));
1599 break;
1600 }
1601 case AF_UNIX: {
1602 struct sockaddr_un *soun = (struct sockaddr_un *)addr;
1603
1604 (void) sprintf(buf, "(len %d) %s", addrlen,
1605 (soun == NULL) ? "(none)" : soun->sun_path);
1606 break;
1607 }
1608 default:
1609 (void) sprintf(buf, "(unknown af %d)", family);
1610 break;
1611 }
1612 return (buf);
1613 }
1614
1615 /* The logical equivalence operator (a if-and-only-if b) */
1616 #define EQUIVALENT(a, b) (((a) && (b)) || (!(a) && (!(b))))
1617
1618 /*
1619 * Verify limitations and invariants on oob state.
1620 * Return 1 if OK, otherwise 0 so that it can be used as
1621 * ASSERT(verify_oobstate(so));
1622 */
1623 int
1624 so_verify_oobstate(struct sonode *so)
1625 {
1626 boolean_t havemark;
1627
1628 ASSERT(MUTEX_HELD(&so->so_lock));
1629
1630 /*
1631 * The possible state combinations are:
1632 * 0
1633 * SS_OOBPEND
1634 * SS_OOBPEND|SS_HAVEOOBDATA
1635 * SS_OOBPEND|SS_HADOOBDATA
1636 * SS_HADOOBDATA
1637 */
1638 switch (so->so_state & (SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA)) {
1639 case 0:
1640 case SS_OOBPEND:
1641 case SS_OOBPEND|SS_HAVEOOBDATA:
1642 case SS_OOBPEND|SS_HADOOBDATA:
1643 case SS_HADOOBDATA:
1644 break;
1645 default:
1646 printf("Bad oob state 1 (%p): state %s\n",
1647 (void *)so, pr_state(so->so_state, so->so_mode));
1648 return (0);
1649 }
1650
1651 /* SS_RCVATMARK should only be set when SS_OOBPEND is set */
1652 if ((so->so_state & (SS_RCVATMARK|SS_OOBPEND)) == SS_RCVATMARK) {
1653 printf("Bad oob state 2 (%p): state %s\n",
1654 (void *)so, pr_state(so->so_state, so->so_mode));
1655 return (0);
1656 }
1657
1658 /*
1659 * (havemark != 0 or SS_RCVATMARK) iff SS_OOBPEND
1660 * For TPI, the presence of a "mark" is indicated by sti_oobsigcnt.
1661 */
1662 havemark = (SOCK_IS_NONSTR(so)) ? so->so_oobmark > 0 :
1663 SOTOTPI(so)->sti_oobsigcnt > 0;
1664
1665 if (!EQUIVALENT(havemark || (so->so_state & SS_RCVATMARK),
1666 so->so_state & SS_OOBPEND)) {
1667 printf("Bad oob state 3 (%p): state %s\n",
1668 (void *)so, pr_state(so->so_state, so->so_mode));
1669 return (0);
1670 }
1671
1672 /*
1673 * Unless SO_OOBINLINE we have so_oobmsg != NULL iff SS_HAVEOOBDATA
1674 */
1675 if (!(so->so_options & SO_OOBINLINE) &&
1676 !EQUIVALENT(so->so_oobmsg != NULL, so->so_state & SS_HAVEOOBDATA)) {
1677 printf("Bad oob state 4 (%p): state %s\n",
1678 (void *)so, pr_state(so->so_state, so->so_mode));
1679 return (0);
1680 }
1681
1682 if (!SOCK_IS_NONSTR(so) &&
1683 SOTOTPI(so)->sti_oobsigcnt < SOTOTPI(so)->sti_oobcnt) {
1684 printf("Bad oob state 5 (%p): counts %d/%d state %s\n",
1685 (void *)so, SOTOTPI(so)->sti_oobsigcnt,
1686 SOTOTPI(so)->sti_oobcnt,
1687 pr_state(so->so_state, so->so_mode));
1688 return (0);
1689 }
1690
1691 return (1);
1692 }
1693 #undef EQUIVALENT
1694 #endif /* DEBUG */
1695
1696 /* initialize sockfs zone specific kstat related items */
1697 void *
1698 sock_kstat_init(zoneid_t zoneid)
1699 {
1700 kstat_t *ksp;
1701
1702 ksp = kstat_create_zone("sockfs", 0, "sock_unix_list", "misc",
1703 KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VAR_SIZE|KSTAT_FLAG_VIRTUAL, zoneid);
1704
1705 if (ksp != NULL) {
1706 ksp->ks_update = sockfs_update;
1707 ksp->ks_snapshot = sockfs_snapshot;
1708 ksp->ks_lock = &socklist.sl_lock;
1709 ksp->ks_private = (void *)(uintptr_t)zoneid;
1710 kstat_install(ksp);
1711 }
1712
1713 return (ksp);
1714 }
1715
1716 /* tear down sockfs zone specific kstat related items */
1717 /*ARGSUSED*/
1718 void
1719 sock_kstat_fini(zoneid_t zoneid, void *arg)
1720 {
1721 kstat_t *ksp = (kstat_t *)arg;
1722
1723 if (ksp != NULL) {
1724 ASSERT(zoneid == (zoneid_t)(uintptr_t)ksp->ks_private);
1725 kstat_delete(ksp);
1726 }
1727 }
1728
1729 /*
1730 * Zones:
1731 * Note that nactive is going to be different for each zone.
1732 * This means we require kstat to call sockfs_update and then sockfs_snapshot
1733 * for the same zone, or sockfs_snapshot will be taken into the wrong size
1734 * buffer. This is safe, but if the buffer is too small, user will not be
1735 * given details of all sockets. However, as this kstat has a ks_lock, kstat
1736 * driver will keep it locked between the update and the snapshot, so no
1737 * other process (zone) can currently get inbetween resulting in a wrong size
1738 * buffer allocation.
1739 */
1740
1741 static int
1742 sockfs_update(kstat_t *ksp, int rw)
1743 {
1744 uint_t n, nactive = 0; /* # of active AF_UNIX sockets */
1745 uint_t tsze, sze;
1746 struct sonode *so; /* current sonode on socklist */
1747 zoneid_t myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private;
1748
1749 tsze = sze = 0;
1750
1751 ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid());
1752
1753 if (rw == KSTAT_WRITE) { /* bounce all writes */
1754 return (EACCES);
1755 }
1756
1757 for (so = socklist.sl_list; so != NULL; so = SOTOTPI(so)->sti_next_so) {
1758 if (so->so_count != 0 && so->so_zoneid == myzoneid) {
1759
1760 nactive++;
1761
1762 mutex_enter(&so->so_pid_list_lock);
1763 n = list_size(&so->so_pid_list);
1764 mutex_exit(&so->so_pid_list_lock);
1765
1766 sze = sizeof (struct sockinfo);
1767 sze += (n > 1)?((n - 1) * sizeof (conn_pid_node_t)):0;
1768 tsze += sze;
1769
1770 }
1771 }
1772 ksp->ks_ndata = nactive;
1773 ksp->ks_data_size = tsze;
1774
1775 return (0);
1776 }
1777
1778 static int
1779 sockfs_snapshot(kstat_t *ksp, void *buf, int rw)
1780 {
1781 int ns; /* # of sonodes we've copied */
1782 struct sonode *so; /* current sonode on socklist */
1783 struct sockinfo *psi; /* where we put sockinfo data */
1784 t_uscalar_t sn_len; /* soa_len */
1785 zoneid_t myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private;
1786 sotpi_info_t *sti;
1787
1788 uint_t sze;
1789 conn_pid_node_list_hdr_t *cph;
1790
1791 ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid());
1792
1793 ksp->ks_snaptime = gethrtime();
1794
1795 if (rw == KSTAT_WRITE) { /* bounce all writes */
1796 return (EACCES);
1797 }
1798
1799 /*
1800 * for each sonode on the socklist, we massage the important
1801 * info into buf, in k_sockinfo format.
1802 */
1803 psi = (struct sockinfo *)buf;
1804 ns = 0;
1805 for (so = socklist.sl_list; so != NULL; so = SOTOTPI(so)->sti_next_so) {
1806 /* only stuff active sonodes and the same zone: */
1807 if (so->so_count == 0 || so->so_zoneid != myzoneid) {
1808 continue;
1809 }
1810
1811 /* get the pidnode list associated with this sonode */
1812 cph = so_get_sock_pid_list((sock_upper_handle_t)so);
1813
1814 /* calculate the size of this sockinfo structure */
1815 sze = sizeof (struct sockinfo);
1816 sze += (cph->cph_pn_cnt > 1)?
1817 ((cph->cph_pn_cnt - 1) * sizeof (conn_pid_node_t)):0;
1818
1819 /*
1820 * If the sonode was activated between the update and the
1821 * snapshot, we're done - as this is only a snapshot. We need
1822 * to make sure that we have space for this sockinfo.
1823 */
1824 if (((caddr_t)(psi) + sze) >
1825 ((caddr_t)buf + ksp->ks_data_size)) {
1826 break;
1827 }
1828
1829 sti = SOTOTPI(so);
1830 /* copy important info into buf: */
1831 psi->si_size = sze;
1832 psi->si_family = so->so_family;
1833 psi->si_type = so->so_type;
1834 psi->si_flag = so->so_flag;
1835 psi->si_state = so->so_state;
1836 psi->si_serv_type = sti->sti_serv_type;
1837 psi->si_ux_laddr_sou_magic =
1838 sti->sti_ux_laddr.soua_magic;
1839 psi->si_ux_faddr_sou_magic =
1840 sti->sti_ux_faddr.soua_magic;
1841 psi->si_laddr_soa_len = sti->sti_laddr.soa_len;
1842 psi->si_faddr_soa_len = sti->sti_faddr.soa_len;
1843 psi->si_szoneid = so->so_zoneid;
1844 psi->si_faddr_noxlate = sti->sti_faddr_noxlate;
1845
1846
1847 mutex_enter(&so->so_lock);
1848
1849 if (sti->sti_laddr_sa != NULL) {
1850 ASSERT(sti->sti_laddr_sa->sa_data != NULL);
1851 sn_len = sti->sti_laddr_len;
1852 ASSERT(sn_len <= sizeof (short) +
1853 sizeof (psi->si_laddr_sun_path));
1854
1855 psi->si_laddr_family =
1856 sti->sti_laddr_sa->sa_family;
1857 if (sn_len != 0) {
1858 /* AF_UNIX socket names are NULL terminated */
1859 (void) strncpy(psi->si_laddr_sun_path,
1860 sti->sti_laddr_sa->sa_data,
1861 sizeof (psi->si_laddr_sun_path));
1862 sn_len = strlen(psi->si_laddr_sun_path);
1863 }
1864 psi->si_laddr_sun_path[sn_len] = 0;
1865 }
1866
1867 if (sti->sti_faddr_sa != NULL) {
1868 ASSERT(sti->sti_faddr_sa->sa_data != NULL);
1869 sn_len = sti->sti_faddr_len;
1870 ASSERT(sn_len <= sizeof (short) +
1871 sizeof (psi->si_faddr_sun_path));
1872
1873 psi->si_faddr_family =
1874 sti->sti_faddr_sa->sa_family;
1875 if (sn_len != 0) {
1876 (void) strncpy(psi->si_faddr_sun_path,
1877 sti->sti_faddr_sa->sa_data,
1878 sizeof (psi->si_faddr_sun_path));
1879 sn_len = strlen(psi->si_faddr_sun_path);
1880 }
1881 psi->si_faddr_sun_path[sn_len] = 0;
1882 }
1883
1884 mutex_exit(&so->so_lock);
1885
1886 (void) sprintf(psi->si_son_straddr, "%p", (void *)so);
1887 (void) sprintf(psi->si_lvn_straddr, "%p",
1888 (void *)sti->sti_ux_laddr.soua_vp);
1889 (void) sprintf(psi->si_fvn_straddr, "%p",
1890 (void *)sti->sti_ux_faddr.soua_vp);
1891
1892 if ((psi->si_pn_cnt = cph->cph_pn_cnt) > 0)
1893 (void) memcpy(psi->si_pns, cph->cph_cpns,
1894 psi->si_pn_cnt * sizeof (conn_pid_node_t));
1895
1896 kmem_free(cph, cph->cph_tot_size);
1897
1898 psi = (struct sockinfo *)((char *)psi + psi->si_size);
1899 ns++;
1900 }
1901
1902 ksp->ks_ndata = ns;
1903 return (0);
1904 }
1905
1906 ssize_t
1907 soreadfile(file_t *fp, uchar_t *buf, u_offset_t fileoff, int *err, size_t size)
1908 {
1909 struct uio auio;
1910 struct iovec aiov[MSG_MAXIOVLEN];
1911 register vnode_t *vp;
1912 int ioflag, rwflag;
1913 ssize_t cnt;
1914 int error = 0;
1915 int iovcnt = 0;
1916 short fflag;
1917
1918 vp = fp->f_vnode;
1919 fflag = fp->f_flag;
1920
1921 rwflag = 0;
1922 aiov[0].iov_base = (caddr_t)buf;
1923 aiov[0].iov_len = size;
1924 iovcnt = 1;
1925 cnt = (ssize_t)size;
1926 (void) VOP_RWLOCK(vp, rwflag, NULL);
1927
1928 auio.uio_loffset = fileoff;
1929 auio.uio_iov = aiov;
1930 auio.uio_iovcnt = iovcnt;
1931 auio.uio_resid = cnt;
1932 auio.uio_segflg = UIO_SYSSPACE;
1933 auio.uio_llimit = MAXOFFSET_T;
1934 auio.uio_fmode = fflag;
1935 auio.uio_extflg = UIO_COPY_CACHED;
1936
1937 ioflag = auio.uio_fmode & (FAPPEND|FSYNC|FDSYNC|FRSYNC);
1938
1939 /* If read sync is not asked for, filter sync flags */
1940 if ((ioflag & FRSYNC) == 0)
1941 ioflag &= ~(FSYNC|FDSYNC);
1942 error = VOP_READ(vp, &auio, ioflag, fp->f_cred, NULL);
1943 cnt -= auio.uio_resid;
1944
1945 VOP_RWUNLOCK(vp, rwflag, NULL);
1946
1947 if (error == EINTR && cnt != 0)
1948 error = 0;
1949 out:
1950 if (error != 0) {
1951 *err = error;
1952 return (0);
1953 } else {
1954 *err = 0;
1955 return (cnt);
1956 }
1957 }
1958
1959 int
1960 so_copyin(const void *from, void *to, size_t size, int fromkernel)
1961 {
1962 if (fromkernel) {
1963 bcopy(from, to, size);
1964 return (0);
1965 }
1966 return (xcopyin(from, to, size));
1967 }
1968
1969 int
1970 so_copyout(const void *from, void *to, size_t size, int tokernel)
1971 {
1972 if (tokernel) {
1973 bcopy(from, to, size);
1974 return (0);
1975 }
1976 return (xcopyout(from, to, size));
1977 }