xref: /titanic_41/usr/src/uts/common/os/swapgeneric.c (revision 0a05e7057ae5537db2da83492d375e6524599463)
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 /* ONC_PLUS EXTRACT START */
22 /*
23  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 /* ONC_PLUS EXTRACT END */
29 
30 /*
31  * Configure root, swap and dump devices.
32  */
33 
34 #include <sys/types.h>
35 #include <sys/param.h>
36 #include <sys/sysmacros.h>
37 #include <sys/signal.h>
38 #include <sys/cred.h>
39 #include <sys/proc.h>
40 #include <sys/user.h>
41 #include <sys/conf.h>
42 #include <sys/buf.h>
43 #include <sys/systm.h>
44 #include <sys/vm.h>
45 #include <sys/reboot.h>
46 #include <sys/file.h>
47 #include <sys/vfs.h>
48 #include <sys/vnode.h>
49 #include <sys/errno.h>
50 #include <sys/kmem.h>
51 #include <sys/uio.h>
52 #include <sys/open.h>
53 #include <sys/mount.h>
54 #include <sys/kobj.h>
55 #include <sys/bootconf.h>
56 #include <sys/sysconf.h>
57 #include <sys/modctl.h>
58 #include <sys/autoconf.h>
59 #include <sys/debug.h>
60 #include <sys/fs/snode.h>
61 #include <fs/fs_subr.h>
62 #include <sys/socket.h>
63 #include <net/if.h>
64 
65 #include <sys/mkdev.h>
66 #include <sys/cmn_err.h>
67 #include <sys/console.h>
68 
69 #include <sys/conf.h>
70 #include <sys/ddi.h>
71 #include <sys/sunddi.h>
72 #include <sys/hwconf.h>
73 #include <sys/dc_ki.h>
74 #include <sys/promif.h>
75 
76 
77 /*
78  * Local routines
79  */
80 static int preload_module(struct sysparam *, void *);
81 static struct vfssw *getfstype(char *, char *, size_t);
82 static int getphysdev(char *, char *, size_t);
83 static int load_bootpath_drivers(char *bootpath);
84 static int load_boot_driver(char *drv);
85 static int load_boot_platform_modules(char *drv);
86 static dev_info_t *path_to_devinfo(char *path);
87 static boolean_t netboot_over_ib(char *bootpath);
88 
89 
90 /*
91  * Module linkage information for the kernel.
92  */
93 static struct modlmisc modlmisc = {
94 	&mod_miscops, "root and swap configuration"
95 };
96 
97 static struct modlinkage modlinkage = {
98 	MODREV_1, (void *)&modlmisc, NULL
99 };
100 
101 int
102 _init(void)
103 {
104 	return (mod_install(&modlinkage));
105 }
106 
107 int
108 _fini(void)
109 {
110 	return (mod_remove(&modlinkage));
111 }
112 
113 int
114 _info(struct modinfo *modinfop)
115 {
116 	return (mod_info(&modlinkage, modinfop));
117 }
118 
119 /*
120  * Configure root file system.
121  */
122 int
123 rootconf(void)
124 {
125 	int error;
126 	struct vfssw *vsw;
127 	extern void pm_init(void);
128 
129 	BMDPRINTF(("rootconf: fstype %s\n", rootfs.bo_fstype));
130 	BMDPRINTF(("rootconf: name %s\n", rootfs.bo_name));
131 	BMDPRINTF(("rootconf: flags 0x%x\n", rootfs.bo_flags));
132 	BMDPRINTF(("rootconf: obp_bootpath %s\n", obp_bootpath));
133 
134 	/*
135 	 * Install cluster modules that were only loaded during
136 	 * loadrootmodules().
137 	 */
138 	if (error = clboot_rootconf())
139 		return (error);
140 
141 	if (root_is_svm) {
142 		(void) strncpy(rootfs.bo_name, obp_bootpath, BO_MAXOBJNAME);
143 
144 		BMDPRINTF(("rootconf: svm: rootfs name %s\n", rootfs.bo_name));
145 		BMDPRINTF(("rootconf: svm: svm name %s\n", svm_bootpath));
146 	}
147 
148 	/*
149 	 * Run _init on the root filesystem (we already loaded it
150 	 * but we've been waiting until now to _init it) which will
151 	 * have the side-effect of running vsw_init() on this vfs.
152 	 * Because all the nfs filesystems are lumped into one
153 	 * module we need to special case it.
154 	 */
155 	if (strncmp(rootfs.bo_fstype, "nfs", 3) == 0) {
156 		if (modload("fs", "nfs") == -1) {
157 			cmn_err(CE_CONT, "Cannot initialize %s filesystem\n",
158 			    rootfs.bo_fstype);
159 			return (ENXIO);
160 		}
161 	} else {
162 		if (modload("fs", rootfs.bo_fstype) == -1) {
163 			cmn_err(CE_CONT, "Cannot initialize %s filesystem\n",
164 			    rootfs.bo_fstype);
165 			return (ENXIO);
166 		}
167 	}
168 	RLOCK_VFSSW();
169 	vsw = vfs_getvfsswbyname(rootfs.bo_fstype);
170 	RUNLOCK_VFSSW();
171 	VFS_INIT(rootvfs, &vsw->vsw_vfsops, (caddr_t)0);
172 	VFS_HOLD(rootvfs);
173 
174 	if (root_is_svm) {
175 		rootvfs->vfs_flag |= VFS_RDONLY;
176 	}
177 
178 	/*
179 	 * This pm-releated call has to occur before root is mounted since we
180 	 * need to power up all devices.  It is placed after VFS_INIT() such
181 	 * that opening a device via ddi_lyr_ interface just before root has
182 	 * been mounted would work.
183 	 */
184 	pm_init();
185 
186 	if (netboot) {
187 		if ((error = strplumb()) != 0) {
188 			cmn_err(CE_CONT, "Cannot plumb network device\n");
189 			return (error);
190 		}
191 	}
192 
193 	/*
194 	 * ufs_mountroot() ends up calling getrootdev()
195 	 * (below) which actually triggers the _init, identify,
196 	 * probe and attach of the drivers that make up root device
197 	 * bush; these are also quietly waiting in memory.
198 	 */
199 	BMDPRINTF(("rootconf: calling VFS_MOUNTROOT %s\n", rootfs.bo_fstype));
200 
201 	error = VFS_MOUNTROOT(rootvfs, ROOT_INIT);
202 	vfs_unrefvfssw(vsw);
203 	rootdev = rootvfs->vfs_dev;
204 
205 	if (error)
206 		cmn_err(CE_CONT, "Cannot mount root on %s fstype %s\n",
207 		    rootfs.bo_name, rootfs.bo_fstype);
208 	else
209 		cmn_err(CE_CONT, "?root on %s fstype %s\n",
210 		    rootfs.bo_name, rootfs.bo_fstype);
211 	return (error);
212 }
213 
214 /*
215  * Remount root on an SVM mirror root device
216  * Only supported on UFS filesystems at present
217  */
218 int
219 svm_rootconf(void)
220 {
221 	int	error;
222 	extern int ufs_remountroot(struct vfs *vfsp);
223 
224 	ASSERT(root_is_svm == 1);
225 
226 	if (strcmp(rootfs.bo_fstype, "ufs") != 0) {
227 		cmn_err(CE_CONT, "Mounting root on %s with filesystem "
228 		    "type %s is not supported\n",
229 		    rootfs.bo_name, rootfs.bo_fstype);
230 		return (EINVAL);
231 	}
232 
233 	(void) strncpy(rootfs.bo_name, svm_bootpath, BO_MAXOBJNAME);
234 
235 	BMDPRINTF(("svm_rootconf: rootfs %s\n", rootfs.bo_name));
236 
237 	error = ufs_remountroot(rootvfs);
238 
239 	if (error) {
240 		cmn_err(CE_CONT, "Cannot remount root on %s fstype %s\n",
241 		    rootfs.bo_name, rootfs.bo_fstype);
242 	} else {
243 		cmn_err(CE_CONT, "?root remounted on %s fstype %s\n",
244 		    rootfs.bo_name, rootfs.bo_fstype);
245 	}
246 	return (error);
247 }
248 
249 /*
250  * Under the assumption that our root file system is on a
251  * disk partition, get the dev_t of the partition in question.
252  *
253  * By now, boot has faithfully loaded all our modules into memory, and
254  * we've taken over resource management.  Before we go any further, we
255  * have to fire up the device drivers and stuff we need to mount the
256  * root filesystem.  That's what we do here.  Fingers crossed.
257  */
258 dev_t
259 getrootdev(void)
260 {
261 	dev_t	d;
262 
263 	if ((d = ddi_pathname_to_dev_t(rootfs.bo_name)) == NODEV)
264 		cmn_err(CE_CONT, "Cannot assemble drivers for root %s\n",
265 		    rootfs.bo_name);
266 	return (d);
267 }
268 
269 /*
270  * If booted with ASKNAME, prompt on the console for a filesystem
271  * name and return it.
272  */
273 void
274 getfsname(char *askfor, char *name, size_t namelen)
275 {
276 	if (boothowto & RB_ASKNAME) {
277 		printf("%s name: ", askfor);
278 		console_gets(name, namelen);
279 	}
280 }
281 
282 /*ARGSUSED1*/
283 static int
284 preload_module(struct sysparam *sysp, void *p)
285 {
286 	static char *wmesg = "forceload of %s failed";
287 	char *name;
288 
289 	name = sysp->sys_ptr;
290 	BMDPRINTF(("preload_module: %s\n", name));
291 	if (modloadonly(NULL, name) < 0)
292 		cmn_err(CE_WARN, wmesg, name);
293 	return (0);
294 }
295 
296 /* ONC_PLUS EXTRACT START */
297 /*
298  * We want to load all the modules needed to mount the root filesystem,
299  * so that when we start the ball rolling in 'getrootdev', every module
300  * should already be in memory, just waiting to be init-ed.
301  */
302 
303 int
304 loadrootmodules(void)
305 {
306 	struct vfssw	*vsw;
307 	char		*this;
308 	char		*name;
309 	int		err;
310 /* ONC_PLUS EXTRACT END */
311 	int		i, proplen;
312 	extern char	*impl_module_list[];
313 	extern char	*platform_module_list[];
314 
315 	/* Make sure that the PROM's devinfo tree has been created */
316 	ASSERT(ddi_root_node());
317 
318 	BMDPRINTF(("loadrootmodules: fstype %s\n", rootfs.bo_fstype));
319 	BMDPRINTF(("loadrootmodules: name %s\n", rootfs.bo_name));
320 	BMDPRINTF(("loadrootmodules: flags 0x%x\n", rootfs.bo_flags));
321 
322 	/*
323 	 * zzz We need to honor what's in rootfs if it's not null.
324 	 * non-null means use what's there.  This way we can
325 	 * change rootfs with /etc/system AND with tunetool.
326 	 */
327 	if (root_is_svm) {
328 		/* user replaced rootdev, record obp_bootpath */
329 		obp_bootpath[0] = '\0';
330 		(void) getphysdev("root", obp_bootpath, BO_MAXOBJNAME);
331 		BMDPRINTF(("loadrootmodules: obp_bootpath %s\n", obp_bootpath));
332 	} else {
333 		/*
334 		 * Get the root fstype and root device path from boot.
335 		 */
336 		rootfs.bo_fstype[0] = '\0';
337 		rootfs.bo_name[0] = '\0';
338 	}
339 
340 	/*
341 	 * This lookup will result in modloadonly-ing the root
342 	 * filesystem module - it gets _init-ed in rootconf()
343 	 */
344 	if ((vsw = getfstype("root", rootfs.bo_fstype, BO_MAXFSNAME)) == NULL)
345 		return (ENXIO);	/* in case we have no file system types */
346 
347 	(void) strcpy(rootfs.bo_fstype, vsw->vsw_name);
348 
349 	vfs_unrefvfssw(vsw);
350 
351 	/*
352 	 * Load the favored drivers of the implementation.
353 	 * e.g. 'sbus' and possibly 'zs' (even).
354 	 *
355 	 * Called whilst boot is still loaded (because boot does
356 	 * the i/o for us), and DDI services are unavailable.
357 	 */
358 	BMDPRINTF(("loadrootmodules: impl_module_list\n"));
359 	for (i = 0; (this = impl_module_list[i]) != NULL; i++) {
360 		if ((err = load_boot_driver(this)) != 0) {
361 			cmn_err(CE_WARN, "Cannot load drv/%s", this);
362 			return (err);
363 			/* NOTREACHED */
364 		}
365 	}
366 	/*
367 	 * Now load the platform modules (if any)
368 	 */
369 	BMDPRINTF(("loadrootmodules: platform_module_list\n"));
370 	for (i = 0; (this = platform_module_list[i]) != NULL; i++) {
371 		if ((err = load_boot_platform_modules(this)) != 0) {
372 			cmn_err(CE_WARN, "Cannot load drv/%s", this);
373 			return (err);
374 			/* NOTREACHED */
375 		}
376 	}
377 
378 loop:
379 	(void) getphysdev("root", rootfs.bo_name, BO_MAXOBJNAME);
380 
381 	/*
382 	 * Given a physical pathname, load the correct set of driver
383 	 * modules into memory, including all possible parents.
384 	 *
385 	 * NB: The code sets the variable 'name' for error reporting.
386 	 */
387 	err = 0;
388 	BMDPRINTF(("loadrootmodules: rootfs %s\n", rootfs.bo_name));
389 	if (root_is_svm == 0) {
390 		BMDPRINTF(("loadrootmodules: rootfs %s\n", rootfs.bo_name));
391 		name = rootfs.bo_name;
392 		err = load_bootpath_drivers(rootfs.bo_name);
393 	}
394 
395 	/*
396 	 * Load driver modules in obp_bootpath, this is always
397 	 * required for mountroot to succeed. obp_bootpath is
398 	 * is set if rootdev is set via /etc/system, which is
399 	 * the case if booting of a SVM/VxVM mirror.
400 	 */
401 	if ((err == 0) && obp_bootpath[0] != '\0') {
402 		BMDPRINTF(("loadrootmodules: obp_bootpath %s\n", obp_bootpath));
403 		name = obp_bootpath;
404 		err = load_bootpath_drivers(obp_bootpath);
405 	}
406 
407 	if (err != 0) {
408 		cmn_err(CE_CONT, "Cannot load drivers for %s\n", name);
409 		goto out;
410 		/* NOTREACHED */
411 	}
412 
413 	/*
414 	 * Check to see if the booter performed DHCP configuration
415 	 * ("bootp-response" boot property exists). If so, then before
416 	 * bootops disappears we need to save the value of this property
417 	 * such that the userland dhcpagent can adopt the DHCP management
418 	 * of our primary network interface.
419 	 */
420 	proplen = BOP_GETPROPLEN(bootops, "bootp-response");
421 	if (proplen > 0) {
422 		dhcack = kmem_zalloc(proplen, KM_SLEEP);
423 		if (BOP_GETPROP(bootops, "bootp-response", dhcack) == -1) {
424 			cmn_err(CE_WARN, "BOP_GETPROP of  "
425 			    "\"bootp-response\" failed\n");
426 			kmem_free(dhcack, dhcacklen);
427 			dhcack = NULL;
428 			goto out;
429 		}
430 		dhcacklen = proplen;
431 
432 		/*
433 		 * Fetch the "netdev-path" boot property (if it exists), and
434 		 * stash it for later use by sysinfo(SI_DHCP_CACHE, ...).
435 		 */
436 		proplen = BOP_GETPROPLEN(bootops, "netdev-path");
437 		if (proplen > 0) {
438 			netdev_path = kmem_zalloc(proplen, KM_SLEEP);
439 			if (BOP_GETPROP(bootops, "netdev-path",
440 			    (uchar_t *)netdev_path) == -1) {
441 				cmn_err(CE_WARN, "BOP_GETPROP of  "
442 				    "\"netdev-path\" failed\n");
443 				kmem_free(netdev_path, proplen);
444 				goto out;
445 			}
446 		}
447 	}
448 
449 	/*
450 	 * Preload (load-only, no init) all modules which
451 	 * were added to the /etc/system file with the
452 	 * FORCELOAD keyword.
453 	 */
454 	BMDPRINTF(("loadrootmodules: preload_module\n"));
455 	(void) mod_sysctl_type(MOD_FORCELOAD, preload_module, NULL);
456 
457 /* ONC_PLUS EXTRACT START */
458 	/*
459 	 * If we booted otw then load in the plumbing
460 	 * routine now while we still can. If we didn't
461 	 * boot otw then we will load strplumb in main().
462 	 *
463 	 * NFS is actually a set of modules, the core routines,
464 	 * a diskless helper module, rpcmod, and the tli interface.  Load
465 	 * them now while we still can.
466 	 *
467 	 * Because we glomb all versions of nfs into a single module
468 	 * we check based on the initial string "nfs".
469 	 *
470 	 * XXX: A better test for this is to see if device_type
471 	 * XXX: from the PROM is "network".
472 	 */
473 
474 	if (strncmp(rootfs.bo_fstype, "nfs", 3) == 0) {
475 		++netboot;
476 
477 		if ((err = modload("misc", "tlimod")) < 0)  {
478 			cmn_err(CE_CONT, "Cannot load misc/tlimod\n");
479 			goto out;
480 			/* NOTREACHED */
481 		}
482 		if ((err = modload("strmod", "rpcmod")) < 0)  {
483 			cmn_err(CE_CONT, "Cannot load strmod/rpcmod\n");
484 			goto out;
485 			/* NOTREACHED */
486 		}
487 		if ((err = modload("misc", "nfs_dlboot")) < 0)  {
488 			cmn_err(CE_CONT, "Cannot load misc/nfs_dlboot\n");
489 			goto out;
490 			/* NOTREACHED */
491 		}
492 		if ((err = modload("mac", "mac_ether")) < 0)  {
493 			cmn_err(CE_CONT, "Cannot load mac/mac_ether\n");
494 			goto out;
495 			/* NOTREACHED */
496 		}
497 		if ((err = modload("misc", "strplumb")) < 0)  {
498 			cmn_err(CE_CONT, "Cannot load misc/strplumb\n");
499 			goto out;
500 			/* NOTREACHED */
501 		}
502 		if ((err = strplumb_load()) < 0) {
503 			goto out;
504 			/* NOTREACHED */
505 		}
506 	}
507 
508 	/*
509 	 * Preload modules needed for booting as a cluster.
510 	 */
511 	err = clboot_loadrootmodules();
512 
513 out:
514 	if (err != 0 && (boothowto & RB_ASKNAME))
515 		goto loop;
516 
517 	return (err);
518 }
519 /* ONC_PLUS EXTRACT END */
520 
521 static int
522 get_bootpath_prop(char *bootpath)
523 {
524 	if (root_is_ramdisk) {
525 		if (BOP_GETPROP(bootops, "bootarchive", bootpath) == -1)
526 			return (-1);
527 		(void) strlcat(bootpath, ":a", BO_MAXOBJNAME);
528 	} else {
529 		/*
530 		 * Look for the 1275 compliant name 'bootpath' first,
531 		 * but make certain it has a non-NULL value as well.
532 		 */
533 		if ((BOP_GETPROP(bootops, "bootpath", bootpath) == -1) ||
534 		    strlen(bootpath) == 0) {
535 			if (BOP_GETPROP(bootops,
536 			    "boot-path", bootpath) == -1)
537 				return (-1);
538 		}
539 	}
540 	return (0);
541 }
542 
543 static int
544 get_fstype_prop(char *fstype)
545 {
546 	char *prop = (root_is_ramdisk) ? "archive-fstype" : "fstype";
547 
548 	return (BOP_GETPROP(bootops, prop, fstype));
549 }
550 
551 /*
552  * Get the name of the root or swap filesystem type, and return
553  * the corresponding entry in the vfs switch.
554  *
555  * If we're not asking the user, and we're trying to find the
556  * root filesystem type, we ask boot for the filesystem
557  * type that it came from and use that.  Similarly, if we're
558  * trying to find the swap filesystem, we try and derive it from
559  * the root filesystem type.
560  *
561  * If we are booting via NFS we currently have these options:
562  *	nfs -	dynamically choose NFS V2. V3, or V4 (default)
563  *	nfs2 -	force NFS V2
564  *	nfs3 -	force NFS V3
565  *	nfs4 -	force NFS V4
566  * Because we need to maintain backward compatibility with the naming
567  * convention that the NFS V2 filesystem name is "nfs" (see vfs_conf.c)
568  * we need to map "nfs" => "nfsdyn" and "nfs2" => "nfs".  The dynamic
569  * nfs module will map the type back to either "nfs", "nfs3", or "nfs4".
570  * This is only for root filesystems, all other uses such as cachefs
571  * will expect that "nfs" == NFS V2.
572  *
573  * If the filesystem isn't already loaded, vfs_getvfssw() will load
574  * it for us, but if (at the time we call it) modrootloaded is
575  * still not set, it won't run the filesystems _init routine (and
576  * implicitly it won't run the filesystems vsw_init() entry either).
577  * We do that explicitly in rootconf().
578  */
579 static struct vfssw *
580 getfstype(char *askfor, char *fsname, size_t fsnamelen)
581 {
582 	struct vfssw *vsw;
583 	static char defaultfs[BO_MAXFSNAME];
584 	int root = 0;
585 
586 	if (strcmp(askfor, "root") == 0) {
587 		(void) get_fstype_prop(defaultfs);
588 		root++;
589 	} else {
590 		(void) strcpy(defaultfs, "swapfs");
591 	}
592 
593 	if (boothowto & RB_ASKNAME) {
594 		for (*fsname = '\0'; *fsname == '\0'; *fsname = '\0') {
595 			printf("%s filesystem type [%s]: ", askfor, defaultfs);
596 			console_gets(fsname, fsnamelen);
597 			if (*fsname == '\0')
598 				(void) strcpy(fsname, defaultfs);
599 			if (root) {
600 				if (strcmp(fsname, "nfs2") == 0)
601 					(void) strcpy(fsname, "nfs");
602 				else if (strcmp(fsname, "nfs") == 0)
603 					(void) strcpy(fsname, "nfsdyn");
604 			}
605 			if ((vsw = vfs_getvfssw(fsname)) != NULL)
606 				return (vsw);
607 			printf("Unknown filesystem type '%s'\n", fsname);
608 		}
609 	} else if (*fsname == '\0') {
610 		fsname = defaultfs;
611 	}
612 	if (*fsname == '\0') {
613 		return (NULL);
614 	}
615 
616 	if (root) {
617 		if (strcmp(fsname, "nfs2") == 0)
618 			(void) strcpy(fsname, "nfs");
619 		else if (strcmp(fsname, "nfs") == 0)
620 			(void) strcpy(fsname, "nfsdyn");
621 	}
622 
623 	return (vfs_getvfssw(fsname));
624 }
625 
626 
627 /*
628  * Get a physical device name, and maybe load and attach
629  * the driver.
630  *
631  * XXX	Need better checking of whether or not a device
632  *	actually exists if the user typed in a pathname.
633  *
634  * XXX	Are we sure we want to expose users to this sort
635  *	of physical namespace gobbledygook (now there's
636  *	a word to conjure with..)
637  *
638  * XXX	Note that on an OBP machine, we can easily ask the
639  *	prom and pretty-print some plausible set of bootable
640  *	devices.  We can also user the prom to verify any
641  *	such device.  Later tim.. later.
642  */
643 static int
644 getphysdev(char *askfor, char *name, size_t namelen)
645 {
646 	static char fmt[] = "Enter physical name of %s device\n[%s]: ";
647 	dev_t dev;
648 	static char defaultpath[BO_MAXOBJNAME];
649 
650 	/*
651 	 * Establish 'default' values - we get the root device from
652 	 * boot, and we infer the swap device is the same but with
653 	 * a 'b' on the end instead of an 'a'.  A first stab at
654 	 * ease-of-use ..
655 	 */
656 	if (strcmp(askfor, "root") == 0) {
657 		if (get_bootpath_prop(defaultpath) == -1)
658 			boothowto |= RB_ASKNAME | RB_VERBOSE;
659 	} else {
660 		(void) strcpy(defaultpath, rootfs.bo_name);
661 		defaultpath[strlen(defaultpath) - 1] = 'b';
662 	}
663 
664 retry:
665 	if (boothowto & RB_ASKNAME) {
666 		printf(fmt, askfor, defaultpath);
667 		console_gets(name, namelen);
668 	}
669 	if (*name == '\0')
670 		(void) strcpy(name, defaultpath);
671 
672 	if (strcmp(askfor, "swap") == 0)   {
673 
674 		/*
675 		 * Try to load and install the swap device driver.
676 		 */
677 		dev = ddi_pathname_to_dev_t(name);
678 
679 		if (dev == (dev_t)-1)  {
680 			printf("Not a supported device for swap.\n");
681 			boothowto |= RB_ASKNAME | RB_VERBOSE;
682 			goto retry;
683 		}
684 
685 		/*
686 		 * Ensure that we're not trying to swap on the floppy.
687 		 */
688 		if (strncmp(ddi_major_to_name(getmajor(dev)), "fd", 2) == 0) {
689 			printf("Too dangerous to swap on the floppy\n");
690 			if (boothowto & RB_ASKNAME)
691 				goto retry;
692 			return (-1);
693 		}
694 	}
695 
696 	return (0);
697 }
698 
699 
700 /*
701  * Load a driver needed to boot.
702  */
703 static int
704 load_boot_driver(char *drv)
705 {
706 	char		*drvname;
707 	major_t		major;
708 #ifdef	sparc
709 	struct devnames *dnp;
710 	ddi_prop_t	*propp;
711 	char		*module;
712 	char		*dir, *mf;
713 	int		plen;
714 	int		mlen;
715 #endif	/* sparc */
716 
717 	if ((major = ddi_name_to_major(drv)) == (major_t)-1) {
718 		cmn_err(CE_CONT, "%s: no major number\n", drv);
719 		return (-1);
720 	}
721 	/*
722 	 * resolve aliases
723 	 */
724 	drvname = ddi_major_to_name(major);
725 
726 #ifdef	DEBUG
727 	if (strcmp(drv, drvname) == 0) {
728 		BMDPRINTF(("load_boot_driver: %s\n", drv));
729 	} else {
730 		BMDPRINTF(("load_boot_driver: %s -> %s\n", drv, drvname));
731 	}
732 #endif	/* DEBUG */
733 
734 	if (modloadonly("drv", drvname) == -1) {
735 		cmn_err(CE_CONT, "%s: cannot load driver\n", drvname);
736 		return (-1);
737 	}
738 
739 #ifdef	sparc
740 	/*
741 	 * NOTE: this can be removed when newboot-sparc is delivered.
742 	 *
743 	 * Check to see if the driver had a 'ddi-forceload' global driver.conf
744 	 * property to identify additional modules that need to be loaded.
745 	 * The driver still needs to use ddi_modopen() to open these modules,
746 	 * but the 'ddi-forceload' property allows the modules to be loaded
747 	 * into memory prior to lights-out, so that driver ddi_modopen()
748 	 * calls during lights-out (when mounting root) will work correctly.
749 	 * Use of 'ddi-forceload' is only required for drivers involved in
750 	 * getting root mounted.
751 	 */
752 	dnp = &devnamesp[major];
753 	if (dnp->dn_global_prop_ptr && dnp->dn_global_prop_ptr->prop_list &&
754 	    ((propp = i_ddi_prop_search(DDI_DEV_T_ANY,
755 	    "ddi-forceload", DDI_PROP_TYPE_STRING,
756 	    &dnp->dn_global_prop_ptr->prop_list)) != NULL)) {
757 
758 		module = (char *)propp->prop_val;
759 		plen = propp->prop_len;
760 		while (plen > 0) {
761 			mlen = strlen(module);
762 			mf = strrchr(module, '/');
763 			if (mf) {
764 				dir = module;
765 				*mf++ = '\0';		/* '/' -> '\0' */
766 			} else {
767 				dir = "misc";
768 				mf = module;
769 			}
770 			if (modloadonly(dir, mf) == -1)
771 				cmn_err(CE_CONT,
772 				    "misc/%s: can't load module\n", mf);
773 			if (mf != module)
774 				*(mf - 1) = '/';	/* '\0' -> '/' */
775 
776 			module += mlen + 1;
777 			plen -= mlen + 1;
778 		}
779 	}
780 #endif	/* sparc */
781 
782 	return (0);
783 }
784 
785 
786 /*
787  * For a given instance, load that driver and its parents
788  */
789 static int
790 load_parent_drivers(dev_info_t *dip, char *path)
791 {
792 	int	rval = 0;
793 	major_t	major = (major_t)-1;
794 	char	*drv;
795 	char	*p;
796 
797 	while (dip) {
798 		/* check for path-oriented alias */
799 		if (path)
800 			major = ddi_name_to_major(path);
801 		else
802 			major = (major_t)-1;
803 
804 		if (major != (major_t)-1)
805 			drv = ddi_major_to_name(major);
806 		else
807 			drv = ddi_binding_name(dip);
808 
809 		if (load_boot_driver(drv) != 0)
810 			rval = -1;
811 
812 		dip = ddi_get_parent(dip);
813 		if (path) {
814 			p = strrchr(path, '/');
815 			if (p)
816 				*p = 0;
817 		}
818 	}
819 
820 	return (rval);
821 }
822 
823 
824 /*
825  * For a given path to a boot device,
826  * load that driver and all its parents.
827  */
828 static int
829 load_bootpath_drivers(char *bootpath)
830 {
831 	dev_info_t	*dip;
832 	char		*pathcopy;
833 	int		pathcopy_len;
834 	int		rval;
835 	char		*p;
836 
837 	if (bootpath == NULL || *bootpath == 0)
838 		return (-1);
839 
840 	BMDPRINTF(("load_bootpath_drivers: %s\n", bootpath));
841 
842 	pathcopy = i_ddi_strdup(bootpath, KM_SLEEP);
843 	pathcopy_len = strlen(pathcopy) + 1;
844 
845 	dip = path_to_devinfo(pathcopy);
846 
847 #if defined(__i386) || defined(__amd64)
848 	/*
849 	 * i386 does not provide stub nodes for all boot devices,
850 	 * but we should be able to find the node for the parent,
851 	 * and the leaf of the boot path should be the driver name,
852 	 * which we go ahead and load here.
853 	 */
854 	if (dip == NULL) {
855 		char	*leaf;
856 
857 		/*
858 		 * Find last slash to build the full path to the
859 		 * parent of the leaf boot device
860 		 */
861 		p = strrchr(pathcopy, '/');
862 		*p++ = 0;
863 
864 		/*
865 		 * Now isolate the driver name of the leaf device
866 		 */
867 		leaf = p;
868 		p = strchr(leaf, '@');
869 		*p = 0;
870 
871 		BMDPRINTF(("load_bootpath_drivers: parent=%s leaf=%s\n",
872 		    bootpath, leaf));
873 
874 		dip = path_to_devinfo(pathcopy);
875 		if (leaf) {
876 			rval = load_boot_driver(leaf, NULL);
877 			if (rval == -1) {
878 				kmem_free(pathcopy, pathcopy_len);
879 				return (NULL);
880 			}
881 		}
882 	}
883 #endif
884 
885 	if (dip == NULL) {
886 		cmn_err(CE_WARN, "can't bind driver for boot path <%s>",
887 		    bootpath);
888 		kmem_free(pathcopy, pathcopy_len);
889 		return (NULL);
890 	}
891 
892 	/*
893 	 * Load IP over IB driver when netbooting over IB.
894 	 * As per IB 1275 binding, IP over IB is represented as
895 	 * service on the top of the HCA node. So, there is no
896 	 * PROM node and generic framework cannot pre-load
897 	 * IP over IB driver based on the bootpath. The following
898 	 * code preloads IP over IB driver when doing netboot over
899 	 * InfiniBand.
900 	 */
901 	if (netboot_over_ib(bootpath) &&
902 	    modloadonly("drv", "ibd") == -1) {
903 		cmn_err(CE_CONT, "ibd: cannot load platform driver\n");
904 		kmem_free(pathcopy, pathcopy_len);
905 		return (NULL);
906 	}
907 
908 	/* get rid of minor node at end of copy (if not already done above) */
909 	p = strrchr(pathcopy, '/');
910 	if (p) {
911 		p = strchr(p, ':');
912 		if (p)
913 			*p = 0;
914 	}
915 
916 	rval = load_parent_drivers(dip, pathcopy);
917 	kmem_free(pathcopy, pathcopy_len);
918 	return (rval);
919 }
920 
921 
922 
923 
924 /*
925  * Load drivers required for a platform
926  * Since all hardware nodes should be available in the device
927  * tree, walk the per-driver list and load the parents of
928  * each node found. If not a hardware node, try to load it.
929  * Pseudo nexus is already loaded.
930  */
931 static int
932 load_boot_platform_modules(char *drv)
933 {
934 	major_t	major;
935 	dev_info_t *dip;
936 	char	*drvname;
937 	int	rval = 0;
938 
939 	if ((major = ddi_name_to_major(drv)) == (major_t)-1) {
940 		cmn_err(CE_CONT, "%s: no major number\n", drv);
941 		return (-1);
942 	}
943 
944 	/*
945 	 * resolve aliases
946 	 */
947 	drvname = ddi_major_to_name(major);
948 	if ((major = ddi_name_to_major(drvname)) == (major_t)-1)
949 		return (-1);
950 
951 #ifdef	DEBUG
952 	if (strcmp(drv, drvname) == 0) {
953 		BMDPRINTF(("load_boot_platform_modules: %s\n", drv));
954 	} else {
955 		BMDPRINTF(("load_boot_platform_modules: %s -> %s\n",
956 		    drv, drvname));
957 	}
958 #endif	/* DEBUG */
959 
960 	dip = devnamesp[major].dn_head;
961 	if (dip == NULL) {
962 		/* pseudo node, not-enumerated, needs to be loaded */
963 		if (modloadonly("drv", drvname) == -1) {
964 			cmn_err(CE_CONT, "%s: cannot load platform driver\n",
965 			    drvname);
966 			rval = -1;
967 		}
968 	} else {
969 		while (dip) {
970 			if (load_parent_drivers(dip, NULL) != 0)
971 				rval = -1;
972 			dip = ddi_get_next(dip);
973 		}
974 	}
975 
976 	return (rval);
977 }
978 
979 
980 /*
981  * i_find_node: Internal routine used by path_to_devinfo
982  * to locate a given nodeid in the device tree.
983  */
984 struct i_path_findnode {
985 	pnode_t nodeid;
986 	dev_info_t *dip;
987 };
988 
989 static int
990 i_path_find_node(dev_info_t *dev, void *arg)
991 {
992 	struct i_path_findnode *f = (struct i_path_findnode *)arg;
993 
994 
995 	if (ddi_get_nodeid(dev) == (int)f->nodeid) {
996 		f->dip = dev;
997 		return (DDI_WALK_TERMINATE);
998 	}
999 	return (DDI_WALK_CONTINUE);
1000 }
1001 
1002 /*
1003  * Return the devinfo node to a boot device
1004  */
1005 static dev_info_t *
1006 path_to_devinfo(char *path)
1007 {
1008 	struct i_path_findnode fn;
1009 	extern dev_info_t *top_devinfo;
1010 
1011 	/*
1012 	 * Get the nodeid of the given pathname, if such a mapping exists.
1013 	 */
1014 	fn.dip = NULL;
1015 	fn.nodeid = prom_finddevice(path);
1016 	if (fn.nodeid != OBP_BADNODE) {
1017 		/*
1018 		 * Find the nodeid in our copy of the device tree and return
1019 		 * whatever name we used to bind this node to a driver.
1020 		 */
1021 		ddi_walk_devs(top_devinfo, i_path_find_node, (void *)(&fn));
1022 	}
1023 
1024 #ifdef	DEBUG
1025 	/*
1026 	 * If we're bound to something other than the nodename,
1027 	 * note that in the message buffer and system log.
1028 	 */
1029 	if (fn.dip) {
1030 		char *p, *q;
1031 
1032 		p = ddi_binding_name(fn.dip);
1033 		q = ddi_node_name(fn.dip);
1034 		if (p && q && (strcmp(p, q) != 0)) {
1035 			BMDPRINTF(("path_to_devinfo: %s bound to %s\n",
1036 			    path, p));
1037 		}
1038 	}
1039 #endif	/* DEBUG */
1040 
1041 	return (fn.dip);
1042 }
1043 
1044 /*
1045  * This routine returns B_TRUE if the bootpath corresponds to
1046  * IP over IB driver.
1047  *
1048  * The format of the bootpath for the IP over IB looks like
1049  * /pci@1f,700000/pci@1/ib@0:port=1,pkey=8001,protocol=ip
1050  *
1051  * The minor node portion "port=1,pkey=8001,protocol=ip" represents
1052  * IP over IB driver.
1053  */
1054 static boolean_t
1055 netboot_over_ib(char *bootpath)
1056 {
1057 
1058 	char		*temp;
1059 	boolean_t	ret = B_FALSE;
1060 	pnode_t		node = prom_finddevice(bootpath);
1061 	int		len;
1062 	char		devicetype[OBP_MAXDRVNAME];
1063 
1064 	/* Is this IB node ? */
1065 	len = prom_getproplen(node, OBP_DEVICETYPE);
1066 	if (len <= 1 || len >= OBP_MAXDRVNAME)
1067 		return (B_FALSE);
1068 
1069 	(void) prom_getprop(node, OBP_DEVICETYPE, (caddr_t)devicetype);
1070 
1071 	if (strncmp("ib", devicetype, 2) == 0) {
1072 		/* Check for proper IP over IB string */
1073 		if ((temp = strstr(bootpath, ":port=")) != NULL) {
1074 			if ((temp = strstr(temp, ",pkey=")) != NULL)
1075 				if ((temp = strstr(temp,
1076 				    ",protocol=ip")) != NULL) {
1077 					ret = B_TRUE;
1078 				}
1079 		}
1080 	}
1081 	return (ret);
1082 }
1083