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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2015, Joyent Inc.
25 */
26
27 #include <assert.h>
28 #include <ctype.h>
29 #include <errno.h>
30 #include <fcntl.h>
31 #include <stdio.h>
32 #include <stdlib.h>
33 #include <string.h>
34 #include <strings.h>
35 #include <syslog.h>
36 #include <zone.h>
37 #include <sys/types.h>
38 #include <sys/stat.h>
39 #include <stropts.h>
40 #include <sys/conf.h>
41 #include <pthread.h>
42 #include <unistd.h>
43 #include <wait.h>
44 #include <libcontract.h>
45 #include <libcontract_priv.h>
46 #include <sys/contract/process.h>
47 #include <sys/vnic.h>
48 #include <zone.h>
49 #include "dlmgmt_impl.h"
50
51 typedef enum dlmgmt_db_op {
52 DLMGMT_DB_OP_WRITE,
53 DLMGMT_DB_OP_DELETE,
54 DLMGMT_DB_OP_READ
55 } dlmgmt_db_op_t;
56
57 typedef struct dlmgmt_db_req_s {
58 struct dlmgmt_db_req_s *ls_next;
59 dlmgmt_db_op_t ls_op;
60 char ls_link[MAXLINKNAMELEN];
61 datalink_id_t ls_linkid;
62 zoneid_t ls_zoneid;
63 uint32_t ls_flags; /* Either DLMGMT_ACTIVE or */
64 /* DLMGMT_PERSIST, not both. */
65 } dlmgmt_db_req_t;
66
67 /*
68 * List of pending db updates (e.g., because of a read-only filesystem).
69 */
70 static dlmgmt_db_req_t *dlmgmt_db_req_head = NULL;
71 static dlmgmt_db_req_t *dlmgmt_db_req_tail = NULL;
72
73 /*
74 * rewrite_needed is set to B_TRUE by process_link_line() if it encounters a
75 * line with an old format. This will cause the file being read to be
76 * re-written with the current format.
77 */
78 static boolean_t rewrite_needed;
79
80 static int dlmgmt_db_update(dlmgmt_db_op_t, const char *,
81 dlmgmt_link_t *, uint32_t);
82 static int dlmgmt_process_db_req(dlmgmt_db_req_t *);
83 static int dlmgmt_process_db_onereq(dlmgmt_db_req_t *, boolean_t);
84 static void *dlmgmt_db_update_thread(void *);
85 static boolean_t process_link_line(char *, dlmgmt_link_t *);
86 static int process_db_write(dlmgmt_db_req_t *, FILE *, FILE *);
87 static int process_db_read(dlmgmt_db_req_t *, FILE *);
88 static void generate_link_line(dlmgmt_link_t *, boolean_t, char *);
89
90 #define BUFLEN(lim, ptr) (((lim) > (ptr)) ? ((lim) - (ptr)) : 0)
91 #define MAXLINELEN 1024
92
93 typedef void db_walk_func_t(dlmgmt_link_t *);
94
95 /*
96 * Translator functions to go from dladm_datatype_t to character strings.
97 * Each function takes a pointer to a buffer, the size of the buffer,
98 * the name of the attribute, and the value to be written. The functions
99 * return the number of bytes written to the buffer. If the buffer is not big
100 * enough to hold the string representing the value, then nothing is written
101 * and 0 is returned.
102 */
103 typedef size_t write_func_t(char *, size_t, char *, void *);
104
105 /*
106 * Translator functions to read from a NULL terminated string buffer into
107 * something of the given DLADM_TYPE_*. The functions each return the number
108 * of bytes read from the string buffer. If there is an error reading data
109 * from the buffer, then 0 is returned. It is the caller's responsibility
110 * to free the data allocated by these functions.
111 */
112 typedef size_t read_func_t(char *, void **);
113
114 typedef struct translator_s {
115 const char *type_name;
116 write_func_t *write_func;
117 read_func_t *read_func;
118 } translator_t;
119
120 /*
121 * Translator functions, defined later but declared here so that
122 * the translator table can be defined.
123 */
124 static write_func_t write_str, write_boolean, write_uint64;
125 static read_func_t read_str, read_boolean, read_int64;
126
127 /*
128 * Translator table, indexed by dladm_datatype_t.
129 */
130 static translator_t translators[] = {
131 { "string", write_str, read_str },
132 { "boolean", write_boolean, read_boolean },
133 { "int", write_uint64, read_int64 }
134 };
135
136 static size_t ntranslators = sizeof (translators) / sizeof (translator_t);
137
138 #define LINK_PROPERTY_DELIMINATOR ";"
139 #define LINK_PROPERTY_TYPE_VALUE_SEP ","
140 #define BASE_PROPERTY_LENGTH(t, n) (strlen(translators[(t)].type_name) +\
141 strlen(LINK_PROPERTY_TYPE_VALUE_SEP) +\
142 strlen(LINK_PROPERTY_DELIMINATOR) +\
143 strlen((n)))
144 #define GENERATE_PROPERTY_STRING(buf, length, conv, name, type, val) \
145 (snprintf((buf), (length), "%s=%s%s" conv "%s", (name), \
146 translators[(type)].type_name, \
147 LINK_PROPERTY_TYPE_VALUE_SEP, (val), LINK_PROPERTY_DELIMINATOR))
148
149 /*
150 * Name of the cache file to keep the active <link name, linkid> mapping
151 */
152 char cachefile[MAXPATHLEN];
153
154 #define DLMGMT_PERSISTENT_DB_PATH "/etc/dladm/datalink.conf"
155 #define DLMGMT_MAKE_FILE_DB_PATH(buffer, persistent) \
156 (void) snprintf((buffer), MAXPATHLEN, "%s", \
157 (persistent) ? DLMGMT_PERSISTENT_DB_PATH : cachefile);
158
159 typedef struct zopen_arg {
160 const char *zopen_modestr;
161 int *zopen_pipe;
162 int zopen_fd;
163 } zopen_arg_t;
164
165 typedef struct zrename_arg {
166 const char *zrename_newname;
167 } zrename_arg_t;
168
169 typedef union zfoparg {
170 zopen_arg_t zfop_openarg;
171 zrename_arg_t zfop_renamearg;
172 } zfoparg_t;
173
174 typedef struct zfcbarg {
175 boolean_t zfarg_inglobalzone; /* is callback in global zone? */
176 zoneid_t zfarg_finglobalzone; /* is file in global zone? */
177 const char *zfarg_filename;
178 zfoparg_t *zfarg_oparg;
179 } zfarg_t;
180 #define zfarg_openarg zfarg_oparg->zfop_openarg
181 #define zfarg_renamearg zfarg_oparg->zfop_renamearg
182
183 /* zone file callback */
184 typedef int zfcb_t(zfarg_t *);
185
186 /*
187 * Execute an operation on filename relative to zoneid's zone root. If the
188 * file is in the global zone, then the zfcb() callback will simply be called
189 * directly. If the file is in a non-global zone, then zfcb() will be called
190 * both from the global zone's context, and from the non-global zone's context
191 * (from a fork()'ed child that has entered the non-global zone). This is
192 * done to allow the callback to communicate with itself if needed (e.g. to
193 * pass back the file descriptor of an opened file).
194 */
195 static int
196 dlmgmt_zfop(const char *filename, zoneid_t zoneid, zfcb_t *zfcb,
197 zfoparg_t *zfoparg)
198 {
199 int ctfd;
200 int err;
201 pid_t childpid;
202 siginfo_t info;
203 zfarg_t zfarg;
204 ctid_t ct;
205
206 if (zoneid != GLOBAL_ZONEID) {
207 /*
208 * We need to access a file that isn't in the global zone.
209 * Accessing non-global zone files from the global zone is
210 * unsafe (due to symlink attacks), we'll need to fork a child
211 * that enters the zone in question and executes the callback
212 * that will operate on the file.
213 *
214 * Before we proceed with this zone tango, we need to create a
215 * new process contract for the child, as required by
216 * zone_enter().
217 */
218 errno = 0;
219 ctfd = open64("/system/contract/process/template", O_RDWR);
220 if (ctfd == -1)
221 return (errno);
222 if ((err = ct_tmpl_set_critical(ctfd, 0)) != 0 ||
223 (err = ct_tmpl_set_informative(ctfd, 0)) != 0 ||
224 (err = ct_pr_tmpl_set_fatal(ctfd, CT_PR_EV_HWERR)) != 0 ||
225 (err = ct_pr_tmpl_set_param(ctfd, CT_PR_PGRPONLY)) != 0 ||
226 (err = ct_tmpl_activate(ctfd)) != 0) {
227 (void) close(ctfd);
228 return (err);
229 }
230 childpid = fork();
231 switch (childpid) {
232 case -1:
233 (void) ct_tmpl_clear(ctfd);
234 (void) close(ctfd);
235 return (err);
236 case 0:
237 (void) ct_tmpl_clear(ctfd);
238 (void) close(ctfd);
239 /*
240 * Elevate our privileges as zone_enter() requires all
241 * privileges.
242 */
243 if ((err = dlmgmt_elevate_privileges()) != 0)
244 _exit(err);
245 if (zone_enter(zoneid) == -1)
246 _exit(errno);
247 if ((err = dlmgmt_drop_privileges()) != 0)
248 _exit(err);
249 break;
250 default:
251 if (contract_latest(&ct) == -1)
252 ct = -1;
253 (void) ct_tmpl_clear(ctfd);
254 (void) close(ctfd);
255 if (waitid(P_PID, childpid, &info, WEXITED) == -1) {
256 (void) contract_abandon_id(ct);
257 return (errno);
258 }
259 (void) contract_abandon_id(ct);
260 if (info.si_status != 0)
261 return (info.si_status);
262 }
263 }
264
265 zfarg.zfarg_inglobalzone = (zoneid == GLOBAL_ZONEID || childpid != 0);
266 zfarg.zfarg_finglobalzone = (zoneid == GLOBAL_ZONEID);
267 zfarg.zfarg_filename = filename;
268 zfarg.zfarg_oparg = zfoparg;
269 err = zfcb(&zfarg);
270 if (!zfarg.zfarg_inglobalzone)
271 _exit(err);
272 return (err);
273 }
274
275 static int
276 dlmgmt_zopen_cb(zfarg_t *zfarg)
277 {
278 struct strrecvfd recvfd;
279 boolean_t newfile = B_FALSE;
280 boolean_t inglobalzone = zfarg->zfarg_inglobalzone;
281 zoneid_t finglobalzone = zfarg->zfarg_finglobalzone;
282 const char *filename = zfarg->zfarg_filename;
283 const char *modestr = zfarg->zfarg_openarg.zopen_modestr;
284 int *p = zfarg->zfarg_openarg.zopen_pipe;
285 struct stat statbuf;
286 int oflags;
287 mode_t mode;
288 int fd = -1;
289 int err;
290
291 /* We only ever open a file for reading or writing, not both. */
292 oflags = (modestr[0] == 'r') ? O_RDONLY : O_WRONLY | O_CREAT | O_TRUNC;
293 mode = (modestr[0] == 'r') ? 0 : S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH;
294
295 /* Open the file if we're in the same zone as the file. */
296 if (inglobalzone == finglobalzone) {
297 /*
298 * First determine if we will be creating the file as part of
299 * opening it. If so, then we'll need to ensure that it has
300 * the proper ownership after having opened it.
301 */
302 if (oflags & O_CREAT) {
303 if (stat(filename, &statbuf) == -1) {
304 if (errno == ENOENT)
305 newfile = B_TRUE;
306 else
307 return (errno);
308 }
309 }
310 if ((fd = open(filename, oflags, mode)) == -1)
311 return (errno);
312 if (newfile) {
313 if (chown(filename, UID_DLADM, GID_NETADM) == -1) {
314 err = errno;
315 (void) close(fd);
316 return (err);
317 }
318 }
319 }
320
321 /*
322 * If we're not in the global zone, send the file-descriptor back to
323 * our parent in the global zone.
324 */
325 if (!inglobalzone) {
326 assert(!finglobalzone);
327 assert(fd != -1);
328 return (ioctl(p[1], I_SENDFD, fd) == -1 ? errno : 0);
329 }
330
331 /*
332 * At this point, we know we're in the global zone. If the file was
333 * in a non-global zone, receive the file-descriptor from our child in
334 * the non-global zone.
335 */
336 if (!finglobalzone) {
337 if (ioctl(p[0], I_RECVFD, &recvfd) == -1)
338 return (errno);
339 fd = recvfd.fd;
340 }
341
342 zfarg->zfarg_openarg.zopen_fd = fd;
343 return (0);
344 }
345
346 static int
347 dlmgmt_zunlink_cb(zfarg_t *zfarg)
348 {
349 if (zfarg->zfarg_inglobalzone != zfarg->zfarg_finglobalzone)
350 return (0);
351 return (unlink(zfarg->zfarg_filename) == 0 ? 0 : errno);
352 }
353
354 static int
355 dlmgmt_zrename_cb(zfarg_t *zfarg)
356 {
357 if (zfarg->zfarg_inglobalzone != zfarg->zfarg_finglobalzone)
358 return (0);
359 return (rename(zfarg->zfarg_filename,
360 zfarg->zfarg_renamearg.zrename_newname) == 0 ? 0 : errno);
361 }
362
363 /*
364 * Same as fopen(3C), except that it opens the file relative to zoneid's zone
365 * root.
366 */
367 static FILE *
368 dlmgmt_zfopen(const char *filename, const char *modestr, zoneid_t zoneid,
369 int *err)
370 {
371 int p[2];
372 zfoparg_t zfoparg;
373 FILE *fp = NULL;
374
375 if (zoneid != GLOBAL_ZONEID && pipe(p) == -1) {
376 *err = errno;
377 return (NULL);
378 }
379
380 zfoparg.zfop_openarg.zopen_modestr = modestr;
381 zfoparg.zfop_openarg.zopen_pipe = p;
382 *err = dlmgmt_zfop(filename, zoneid, dlmgmt_zopen_cb, &zfoparg);
383 if (zoneid != GLOBAL_ZONEID) {
384 (void) close(p[0]);
385 (void) close(p[1]);
386 }
387 if (*err == 0) {
388 fp = fdopen(zfoparg.zfop_openarg.zopen_fd, modestr);
389 if (fp == NULL) {
390 *err = errno;
391 (void) close(zfoparg.zfop_openarg.zopen_fd);
392 }
393 }
394 return (fp);
395 }
396
397 /*
398 * Same as rename(2), except that old and new are relative to zoneid's zone
399 * root.
400 */
401 static int
402 dlmgmt_zrename(const char *old, const char *new, zoneid_t zoneid)
403 {
404 zfoparg_t zfoparg;
405
406 zfoparg.zfop_renamearg.zrename_newname = new;
407 return (dlmgmt_zfop(old, zoneid, dlmgmt_zrename_cb, &zfoparg));
408 }
409
410 /*
411 * Same as unlink(2), except that filename is relative to zoneid's zone root.
412 */
413 static int
414 dlmgmt_zunlink(const char *filename, zoneid_t zoneid)
415 {
416 return (dlmgmt_zfop(filename, zoneid, dlmgmt_zunlink_cb, NULL));
417 }
418
419 static size_t
420 write_str(char *buffer, size_t buffer_length, char *name, void *value)
421 {
422 char *ptr = value;
423 size_t data_length = strnlen(ptr, buffer_length);
424
425 /*
426 * Strings are assumed to be NULL terminated. In order to fit in
427 * the buffer, the string's length must be less then buffer_length.
428 * If the value is empty, there's no point in writing it, in fact,
429 * we shouldn't even see that case.
430 */
431 if (data_length + BASE_PROPERTY_LENGTH(DLADM_TYPE_STR, name) ==
432 buffer_length || data_length == 0)
433 return (0);
434
435 /*
436 * Since we know the string will fit in the buffer, snprintf will
437 * always return less than buffer_length, so we can just return
438 * whatever snprintf returns.
439 */
440 return (GENERATE_PROPERTY_STRING(buffer, buffer_length, "%s",
441 name, DLADM_TYPE_STR, ptr));
442 }
443
444 static size_t
445 write_boolean(char *buffer, size_t buffer_length, char *name, void *value)
446 {
447 boolean_t *ptr = value;
448
449 /*
450 * Booleans are either zero or one, so we only need room for two
451 * characters in the buffer.
452 */
453 if (buffer_length <= 1 + BASE_PROPERTY_LENGTH(DLADM_TYPE_BOOLEAN, name))
454 return (0);
455
456 return (GENERATE_PROPERTY_STRING(buffer, buffer_length, "%d",
457 name, DLADM_TYPE_BOOLEAN, *ptr));
458 }
459
460 static size_t
461 write_uint64(char *buffer, size_t buffer_length, char *name, void *value)
462 {
463 uint64_t *ptr = value;
464
465 /*
466 * Limit checking for uint64_t is a little trickier.
467 */
468 if (snprintf(NULL, 0, "%lld", *ptr) +
469 BASE_PROPERTY_LENGTH(DLADM_TYPE_UINT64, name) >= buffer_length)
470 return (0);
471
472 return (GENERATE_PROPERTY_STRING(buffer, buffer_length, "%lld",
473 name, DLADM_TYPE_UINT64, *ptr));
474 }
475
476 static size_t
477 read_str(char *buffer, void **value)
478 {
479 char *ptr = calloc(MAXLINKATTRVALLEN, sizeof (char));
480 ssize_t len;
481
482 if (ptr == NULL || (len = strlcpy(ptr, buffer, MAXLINKATTRVALLEN))
483 >= MAXLINKATTRVALLEN) {
484 free(ptr);
485 return (0);
486 }
487
488 *(char **)value = ptr;
489
490 /* Account for NULL terminator */
491 return (len + 1);
492 }
493
494 static size_t
495 read_boolean(char *buffer, void **value)
496 {
497 boolean_t *ptr = calloc(1, sizeof (boolean_t));
498
499 if (ptr == NULL)
500 return (0);
501
502 *ptr = atoi(buffer);
503 *(boolean_t **)value = ptr;
504
505 return (sizeof (boolean_t));
506 }
507
508 static size_t
509 read_int64(char *buffer, void **value)
510 {
511 int64_t *ptr = calloc(1, sizeof (int64_t));
512
513 if (ptr == NULL)
514 return (0);
515
516 *ptr = (int64_t)atoll(buffer);
517 *(int64_t **)value = ptr;
518
519 return (sizeof (int64_t));
520 }
521
522 static dlmgmt_db_req_t *
523 dlmgmt_db_req_alloc(dlmgmt_db_op_t op, const char *linkname,
524 datalink_id_t linkid, zoneid_t zoneid, uint32_t flags, int *err)
525 {
526 dlmgmt_db_req_t *req;
527
528 if ((req = calloc(1, sizeof (dlmgmt_db_req_t))) == NULL) {
529 *err = errno;
530 } else {
531 req->ls_op = op;
532 if (linkname != NULL)
533 (void) strlcpy(req->ls_link, linkname, MAXLINKNAMELEN);
534 req->ls_linkid = linkid;
535 req->ls_zoneid = zoneid;
536 req->ls_flags = flags;
537 }
538 return (req);
539 }
540
541 /*
542 * Update the db entry with name "entryname" using information from "linkp".
543 */
544 static int
545 dlmgmt_db_update(dlmgmt_db_op_t op, const char *entryname, dlmgmt_link_t *linkp,
546 uint32_t flags)
547 {
548 dlmgmt_db_req_t *req;
549 int err;
550
551 /* It is either a persistent request or an active request, not both. */
552 assert((flags == DLMGMT_PERSIST) || (flags == DLMGMT_ACTIVE));
553
554 if ((req = dlmgmt_db_req_alloc(op, entryname, linkp->ll_linkid,
555 linkp->ll_zoneid, flags, &err)) == NULL)
556 return (err);
557
558 /*
559 * If the return error is EINPROGRESS, this request is handled
560 * asynchronously; return success.
561 */
562 err = dlmgmt_process_db_req(req);
563 if (err != EINPROGRESS)
564 free(req);
565 else
566 err = 0;
567 return (err);
568 }
569
570 #define DLMGMT_DB_OP_STR(op) \
571 (((op) == DLMGMT_DB_OP_READ) ? "read" : \
572 (((op) == DLMGMT_DB_OP_WRITE) ? "write" : "delete"))
573
574 #define DLMGMT_DB_CONF_STR(flag) \
575 (((flag) == DLMGMT_ACTIVE) ? "active" : \
576 (((flag) == DLMGMT_PERSIST) ? "persistent" : ""))
577
578 static int
579 dlmgmt_process_db_req(dlmgmt_db_req_t *req)
580 {
581 pthread_t tid;
582 boolean_t writeop;
583 int err;
584
585 /*
586 * If there are already pending "write" requests, queue this request in
587 * the pending list. Note that this function is called while the
588 * dlmgmt_rw_lock is held, so it is safe to access the global variables.
589 */
590 writeop = (req->ls_op != DLMGMT_DB_OP_READ);
591 if (writeop && (req->ls_flags == DLMGMT_PERSIST) &&
592 (dlmgmt_db_req_head != NULL)) {
593 dlmgmt_db_req_tail->ls_next = req;
594 dlmgmt_db_req_tail = req;
595 return (EINPROGRESS);
596 }
597
598 err = dlmgmt_process_db_onereq(req, writeop);
599 if (err != EINPROGRESS && err != 0 && err != ENOENT) {
600 /*
601 * Log the error unless the request processing is still in
602 * progress or if the configuration file hasn't been created
603 * yet (ENOENT).
604 */
605 dlmgmt_log(LOG_WARNING, "dlmgmt_process_db_onereq() %s "
606 "operation on %s configuration failed: %s",
607 DLMGMT_DB_OP_STR(req->ls_op),
608 DLMGMT_DB_CONF_STR(req->ls_flags), strerror(err));
609 }
610
611 if (err == EINPROGRESS) {
612 assert(req->ls_flags == DLMGMT_PERSIST);
613 assert(writeop && dlmgmt_db_req_head == NULL);
614 dlmgmt_db_req_tail = dlmgmt_db_req_head = req;
615 err = pthread_create(&tid, NULL, dlmgmt_db_update_thread, NULL);
616 if (err == 0)
617 return (EINPROGRESS);
618 }
619 return (err);
620 }
621
622 static int
623 dlmgmt_process_db_onereq(dlmgmt_db_req_t *req, boolean_t writeop)
624 {
625 int err = 0;
626 FILE *fp, *nfp = NULL;
627 char file[MAXPATHLEN];
628 char newfile[MAXPATHLEN];
629
630 DLMGMT_MAKE_FILE_DB_PATH(file, (req->ls_flags == DLMGMT_PERSIST));
631 fp = dlmgmt_zfopen(file, "r", req->ls_zoneid, &err);
632 /*
633 * Note that it is not an error if the file doesn't exist. If we're
634 * reading, we treat this case the same way as an empty file. If
635 * we're writing, the file will be created when we open the file for
636 * writing below.
637 */
638 if (fp == NULL && !writeop)
639 return (err);
640
641 if (writeop) {
642 (void) snprintf(newfile, MAXPATHLEN, "%s.new", file);
643 nfp = dlmgmt_zfopen(newfile, "w", req->ls_zoneid, &err);
644 if (nfp == NULL) {
645 /*
646 * EROFS can happen at boot when the file system is
647 * read-only. Return EINPROGRESS so that the caller
648 * can add this request to the pending request list
649 * and start a retry thread.
650 */
651 err = (errno == EROFS ? EINPROGRESS : errno);
652 goto done;
653 }
654 }
655 if (writeop) {
656 if ((err = process_db_write(req, fp, nfp)) == 0)
657 err = dlmgmt_zrename(newfile, file, req->ls_zoneid);
658 } else {
659 err = process_db_read(req, fp);
660 }
661
662 done:
663 if (nfp != NULL) {
664 (void) fclose(nfp);
665 if (err != 0)
666 (void) dlmgmt_zunlink(newfile, req->ls_zoneid);
667 }
668 (void) fclose(fp);
669 return (err);
670 }
671
672 /*ARGSUSED*/
673 static void *
674 dlmgmt_db_update_thread(void *arg)
675 {
676 dlmgmt_db_req_t *req;
677
678 dlmgmt_table_lock(B_TRUE);
679
680 assert(dlmgmt_db_req_head != NULL);
681 while ((req = dlmgmt_db_req_head) != NULL) {
682 assert(req->ls_flags == DLMGMT_PERSIST);
683 if (dlmgmt_process_db_onereq(req, B_TRUE) == EINPROGRESS) {
684 /*
685 * The filesystem is still read only. Go to sleep and
686 * try again.
687 */
688 dlmgmt_table_unlock();
689 (void) sleep(5);
690 dlmgmt_table_lock(B_TRUE);
691 continue;
692 }
693
694 /*
695 * The filesystem is no longer read only. Continue processing
696 * and remove the request from the pending list.
697 */
698 dlmgmt_db_req_head = req->ls_next;
699 if (dlmgmt_db_req_tail == req) {
700 assert(dlmgmt_db_req_head == NULL);
701 dlmgmt_db_req_tail = NULL;
702 }
703 free(req);
704 }
705
706 dlmgmt_table_unlock();
707 return (NULL);
708 }
709
710 static int
711 parse_linkprops(char *buf, dlmgmt_link_t *linkp)
712 {
713 boolean_t found_type = B_FALSE;
714 dladm_datatype_t type = DLADM_TYPE_STR;
715 int i, len;
716 char *curr;
717 char attr_name[MAXLINKATTRLEN];
718 size_t attr_buf_len = 0;
719 void *attr_buf = NULL;
720 boolean_t rename;
721
722 curr = buf;
723 len = strlen(buf);
724 attr_name[0] = '\0';
725 for (i = 0; i < len; i++) {
726 char c = buf[i];
727 boolean_t match = (c == '=' ||
728 (c == ',' && !found_type) || c == ';');
729
730 rename = B_FALSE;
731 /*
732 * Move to the next character if there is no match and
733 * if we have not reached the last character.
734 */
735 if (!match && i != len - 1)
736 continue;
737
738 if (match) {
739 /*
740 * NUL-terminate the string pointed to by 'curr'.
741 */
742 buf[i] = '\0';
743 if (*curr == '\0')
744 goto parse_fail;
745 }
746
747 if (attr_name[0] != '\0' && found_type) {
748 /*
749 * We get here after we have processed the "<prop>="
750 * pattern. The pattern we are now interested in is
751 * "<val>;".
752 */
753 if (c == '=')
754 goto parse_fail;
755
756 if (strcmp(attr_name, "linkid") == 0) {
757 if (read_int64(curr, &attr_buf) == 0)
758 goto parse_fail;
759 linkp->ll_linkid =
760 (datalink_class_t)*(int64_t *)attr_buf;
761 } else if (strcmp(attr_name, "name") == 0) {
762 if (read_str(curr, &attr_buf) == 0)
763 goto parse_fail;
764 (void) snprintf(linkp->ll_link,
765 MAXLINKNAMELEN, "%s", attr_buf);
766 } else if (strcmp(attr_name, "class") == 0) {
767 if (read_int64(curr, &attr_buf) == 0)
768 goto parse_fail;
769 linkp->ll_class =
770 (datalink_class_t)*(int64_t *)attr_buf;
771 } else if (strcmp(attr_name, "media") == 0) {
772 if (read_int64(curr, &attr_buf) == 0)
773 goto parse_fail;
774 linkp->ll_media =
775 (uint32_t)*(int64_t *)attr_buf;
776 } else if (strcmp(attr_name, "zone") == 0) {
777 if (read_str(curr, &attr_buf) == 0)
778 goto parse_fail;
779 linkp->ll_zoneid = getzoneidbyname(attr_buf);
780 if (linkp->ll_zoneid == -1) {
781 if (errno == EFAULT)
782 abort();
783 /*
784 * If we can't find the zone, assign the
785 * link to the GZ and mark it for being
786 * renamed.
787 */
788 linkp->ll_zoneid = 0;
789 rename = B_TRUE;
790 }
791 } else {
792 attr_buf_len = translators[type].read_func(curr,
793 &attr_buf);
794 if (attr_buf_len == 0)
795 goto parse_fail;
796
797 if (linkattr_set(&(linkp->ll_head), attr_name,
798 attr_buf, attr_buf_len, type) != 0) {
799 free(attr_buf);
800 goto parse_fail;
801 }
802 }
803
804 free(attr_buf);
805 attr_name[0] = '\0';
806 found_type = B_FALSE;
807 } else if (attr_name[0] != '\0') {
808 /*
809 * Non-zero length attr_name and found_type of false
810 * indicates that we have not found the type for this
811 * attribute. The pattern now is "<type>,<val>;", we
812 * want the <type> part of the pattern.
813 */
814 for (type = 0; type < ntranslators; type++) {
815 if (strcmp(curr,
816 translators[type].type_name) == 0) {
817 found_type = B_TRUE;
818 break;
819 }
820 }
821
822 if (!found_type)
823 goto parse_fail;
824 } else {
825 /*
826 * A zero length attr_name indicates we are looking
827 * at the beginning of a link attribute.
828 */
829 if (c != '=')
830 goto parse_fail;
831
832 (void) snprintf(attr_name, MAXLINKATTRLEN, "%s", curr);
833 }
834
835 /*
836 * The zone that this link belongs to has died, we are
837 * reparenting it to the GZ and renaming it to avoid name
838 * collisions.
839 */
840 if (rename == B_TRUE) {
841 (void) snprintf(linkp->ll_link, MAXLINKNAMELEN,
842 "SUNWorphan%u", (uint16_t)(gethrtime() / 1000));
843 }
844 curr = buf + i + 1;
845 }
846
847 /* Correct any erroneous IPTUN datalink class constant in the file */
848 if (linkp->ll_class == 0x60) {
849 linkp->ll_class = DATALINK_CLASS_IPTUN;
850 rewrite_needed = B_TRUE;
851 }
852
853 return (0);
854
855 parse_fail:
856 /*
857 * Free linkp->ll_head (link attribute list)
858 */
859 linkattr_destroy(linkp);
860 return (-1);
861 }
862
863 static boolean_t
864 process_link_line(char *buf, dlmgmt_link_t *linkp)
865 {
866 int i, len, llen;
867 char *str, *lasts;
868 char tmpbuf[MAXLINELEN];
869
870 bzero(linkp, sizeof (*linkp));
871 linkp->ll_linkid = DATALINK_INVALID_LINKID;
872
873 /*
874 * Use a copy of buf for parsing so that we can do whatever we want.
875 */
876 (void) strlcpy(tmpbuf, buf, MAXLINELEN);
877
878 /*
879 * Skip leading spaces, blank lines, and comments.
880 */
881 len = strlen(tmpbuf);
882 for (i = 0; i < len; i++) {
883 if (!isspace(tmpbuf[i]))
884 break;
885 }
886 if (i == len || tmpbuf[i] == '#')
887 return (B_TRUE);
888
889 str = tmpbuf + i;
890 /*
891 * Find the link name and assign it to the link structure.
892 */
893 if (strtok_r(str, " \n\t", &lasts) == NULL)
894 goto fail;
895
896 llen = strlen(str);
897 /*
898 * Note that a previous version of the persistent datalink.conf file
899 * stored the linkid as the first field. In that case, the name will
900 * be obtained through parse_linkprops from a property with the format
901 * "name=<linkname>". If we encounter such a format, we set
902 * rewrite_needed so that dlmgmt_db_init() can rewrite the file with
903 * the new format after it's done reading in the data.
904 */
905 if (isdigit(str[0])) {
906 linkp->ll_linkid = atoi(str);
907 rewrite_needed = B_TRUE;
908 } else {
909 if (strlcpy(linkp->ll_link, str, sizeof (linkp->ll_link)) >=
910 sizeof (linkp->ll_link))
911 goto fail;
912 }
913
914 str += llen + 1;
915 if (str >= tmpbuf + len)
916 goto fail;
917
918 /*
919 * Now find the list of link properties.
920 */
921 if ((str = strtok_r(str, " \n\t", &lasts)) == NULL)
922 goto fail;
923
924 if (parse_linkprops(str, linkp) < 0)
925 goto fail;
926
927 return (B_TRUE);
928
929 fail:
930 /*
931 * Delete corrupted line.
932 */
933 buf[0] = '\0';
934 return (B_FALSE);
935 }
936
937 /*
938 * Find any properties in linkp that refer to "old", and rename to "new".
939 * Return B_TRUE if any renaming occurred.
940 */
941 static int
942 dlmgmt_attr_rename(dlmgmt_link_t *linkp, const char *old, const char *new,
943 boolean_t *renamed)
944 {
945 dlmgmt_linkattr_t *attrp;
946 char *newval = NULL, *pname;
947 char valcp[MAXLINKATTRVALLEN];
948 size_t newsize;
949
950 *renamed = B_FALSE;
951
952 if ((attrp = linkattr_find(linkp->ll_head, "linkover")) != NULL ||
953 (attrp = linkattr_find(linkp->ll_head, "simnetpeer")) != NULL) {
954 if (strcmp(old, (char *)attrp->lp_val) == 0) {
955 newsize = strlen(new) + 1;
956 if ((newval = malloc(newsize)) == NULL)
957 return (errno);
958 (void) strcpy(newval, new);
959 free(attrp->lp_val);
960 attrp->lp_val = newval;
961 attrp->lp_sz = newsize;
962 *renamed = B_TRUE;
963 }
964 return (0);
965 }
966
967 if ((attrp = linkattr_find(linkp->ll_head, "portnames")) == NULL)
968 return (0);
969
970 /* <linkname>:[<linkname>:]... */
971 if ((newval = calloc(MAXLINKATTRVALLEN, sizeof (char))) == NULL)
972 return (errno);
973
974 bcopy(attrp->lp_val, valcp, sizeof (valcp));
975 pname = strtok(valcp, ":");
976 while (pname != NULL) {
977 if (strcmp(pname, old) == 0) {
978 (void) strcat(newval, new);
979 *renamed = B_TRUE;
980 } else {
981 (void) strcat(newval, pname);
982 }
983 (void) strcat(newval, ":");
984 pname = strtok(NULL, ":");
985 }
986 if (*renamed) {
987 free(attrp->lp_val);
988 attrp->lp_val = newval;
989 attrp->lp_sz = strlen(newval) + 1;
990 } else {
991 free(newval);
992 }
993 return (0);
994 }
995
996 static int
997 process_db_write(dlmgmt_db_req_t *req, FILE *fp, FILE *nfp)
998 {
999 boolean_t done = B_FALSE;
1000 int err = 0;
1001 dlmgmt_link_t link_in_file, *linkp = NULL, *dblinkp;
1002 boolean_t persist = (req->ls_flags == DLMGMT_PERSIST);
1003 boolean_t writeall, rename, attr_renamed;
1004 char buf[MAXLINELEN];
1005
1006 writeall = (req->ls_linkid == DATALINK_ALL_LINKID);
1007
1008 if (req->ls_op == DLMGMT_DB_OP_WRITE && !writeall) {
1009 /*
1010 * find the link in the avl tree with the given linkid.
1011 */
1012 linkp = link_by_id(req->ls_linkid, req->ls_zoneid);
1013 if (linkp == NULL || (linkp->ll_flags & req->ls_flags) == 0) {
1014 /*
1015 * This link has already been changed. This could
1016 * happen if the request is pending because of
1017 * read-only file-system. If so, we are done.
1018 */
1019 return (0);
1020 }
1021 /*
1022 * In the case of a rename, linkp's name has been updated to
1023 * the new name, and req->ls_link is the old link name.
1024 */
1025 rename = (strcmp(req->ls_link, linkp->ll_link) != 0);
1026 }
1027
1028 /*
1029 * fp can be NULL if the file didn't initially exist and we're
1030 * creating it as part of this write operation.
1031 */
1032 if (fp == NULL)
1033 goto write;
1034
1035 while (err == 0 && fgets(buf, sizeof (buf), fp) != NULL &&
1036 process_link_line(buf, &link_in_file)) {
1037 /*
1038 * Only the link name is needed. Free the memory allocated for
1039 * the link attributes list of link_in_file.
1040 */
1041 linkattr_destroy(&link_in_file);
1042
1043 if (link_in_file.ll_link[0] == '\0' || done) {
1044 /*
1045 * this is a comment line or we are done updating the
1046 * line for the specified link, write the rest of
1047 * lines out.
1048 */
1049 if (fputs(buf, nfp) == EOF)
1050 err = errno;
1051 continue;
1052 }
1053
1054 switch (req->ls_op) {
1055 case DLMGMT_DB_OP_WRITE:
1056 /*
1057 * For write operations, we generate a new output line
1058 * if we're either writing all links (writeall) or if
1059 * the name of the link in the file matches the one
1060 * we're looking for. Otherwise, we write out the
1061 * buffer as-is.
1062 *
1063 * If we're doing a rename operation, ensure that any
1064 * references to the link being renamed in link
1065 * properties are also updated before we write
1066 * anything.
1067 */
1068 if (writeall) {
1069 linkp = link_by_name(link_in_file.ll_link,
1070 req->ls_zoneid);
1071 }
1072 if (writeall || strcmp(req->ls_link,
1073 link_in_file.ll_link) == 0) {
1074 generate_link_line(linkp, persist, buf);
1075 if (!writeall && !rename)
1076 done = B_TRUE;
1077 } else if (rename && persist) {
1078 dblinkp = link_by_name(link_in_file.ll_link,
1079 req->ls_zoneid);
1080 err = dlmgmt_attr_rename(dblinkp, req->ls_link,
1081 linkp->ll_link, &attr_renamed);
1082 if (err != 0)
1083 break;
1084 if (attr_renamed) {
1085 generate_link_line(dblinkp, persist,
1086 buf);
1087 }
1088 }
1089 if (fputs(buf, nfp) == EOF)
1090 err = errno;
1091 break;
1092 case DLMGMT_DB_OP_DELETE:
1093 /*
1094 * Delete is simple. If buf does not represent the
1095 * link we're deleting, write it out.
1096 */
1097 if (strcmp(req->ls_link, link_in_file.ll_link) != 0) {
1098 if (fputs(buf, nfp) == EOF)
1099 err = errno;
1100 } else {
1101 done = B_TRUE;
1102 }
1103 break;
1104 case DLMGMT_DB_OP_READ:
1105 default:
1106 err = EINVAL;
1107 break;
1108 }
1109 }
1110
1111 write:
1112 /*
1113 * If we get to the end of the file and have not seen what linkid
1114 * points to, write it out then.
1115 */
1116 if (req->ls_op == DLMGMT_DB_OP_WRITE && !writeall && !rename && !done) {
1117 generate_link_line(linkp, persist, buf);
1118 done = B_TRUE;
1119 if (fputs(buf, nfp) == EOF)
1120 err = errno;
1121 }
1122
1123 return (err);
1124 }
1125
1126 static int
1127 process_db_read(dlmgmt_db_req_t *req, FILE *fp)
1128 {
1129 avl_index_t name_where, id_where;
1130 dlmgmt_link_t link_in_file, *newlink, *link_in_db;
1131 char buf[MAXLINELEN];
1132 int err = 0;
1133
1134 /*
1135 * This loop processes each line of the configuration file.
1136 */
1137 while (fgets(buf, MAXLINELEN, fp) != NULL) {
1138 if (!process_link_line(buf, &link_in_file)) {
1139 err = EINVAL;
1140 break;
1141 }
1142
1143 /*
1144 * Skip the comment line.
1145 */
1146 if (link_in_file.ll_link[0] == '\0') {
1147 linkattr_destroy(&link_in_file);
1148 continue;
1149 }
1150
1151 if ((req->ls_flags & DLMGMT_ACTIVE) &&
1152 link_in_file.ll_linkid == DATALINK_INVALID_LINKID) {
1153 linkattr_destroy(&link_in_file);
1154 continue;
1155 }
1156
1157 link_in_file.ll_zoneid = req->ls_zoneid;
1158 link_in_db = link_by_name(link_in_file.ll_link,
1159 link_in_file.ll_zoneid);
1160 if (link_in_db != NULL) {
1161 /*
1162 * If the link in the database already has the flag
1163 * for this request set, then the entry is a
1164 * duplicate. If it's not a duplicate, then simply
1165 * turn on the appropriate flag on the existing link.
1166 */
1167 if (link_in_db->ll_flags & req->ls_flags) {
1168 dlmgmt_log(LOG_WARNING, "Duplicate links "
1169 "in the repository: %s",
1170 link_in_file.ll_link);
1171 linkattr_destroy(&link_in_file);
1172 } else {
1173 if (req->ls_flags & DLMGMT_PERSIST) {
1174 /*
1175 * Save the newly read properties into
1176 * the existing link.
1177 */
1178 assert(link_in_db->ll_head == NULL);
1179 link_in_db->ll_head =
1180 link_in_file.ll_head;
1181 } else {
1182 linkattr_destroy(&link_in_file);
1183 }
1184 link_in_db->ll_flags |= req->ls_flags;
1185 }
1186 } else {
1187 /*
1188 * This is a new link. Allocate a new dlmgmt_link_t
1189 * and add it to the trees.
1190 */
1191 newlink = calloc(1, sizeof (*newlink));
1192 if (newlink == NULL) {
1193 dlmgmt_log(LOG_WARNING, "Unable to allocate "
1194 "memory to create new link %s",
1195 link_in_file.ll_link);
1196 linkattr_destroy(&link_in_file);
1197 continue;
1198 }
1199 bcopy(&link_in_file, newlink, sizeof (*newlink));
1200
1201 if (newlink->ll_linkid == DATALINK_INVALID_LINKID)
1202 newlink->ll_linkid = dlmgmt_nextlinkid;
1203 if (avl_find(&dlmgmt_id_avl, newlink, &id_where) !=
1204 NULL) {
1205 dlmgmt_log(LOG_WARNING, "Link ID %d is already"
1206 " in use, destroying link %s",
1207 newlink->ll_linkid, newlink->ll_link);
1208 link_destroy(newlink);
1209 continue;
1210 }
1211
1212 if ((req->ls_flags & DLMGMT_ACTIVE) &&
1213 link_activate(newlink) != 0) {
1214 dlmgmt_log(LOG_WARNING, "Unable to activate %s",
1215 newlink->ll_link);
1216 link_destroy(newlink);
1217 continue;
1218 }
1219
1220 avl_insert(&dlmgmt_id_avl, newlink, id_where);
1221 /*
1222 * link_activate call above can insert newlink in
1223 * dlmgmt_name_avl tree when activating a link that is
1224 * assigned to a NGZ.
1225 */
1226 if (avl_find(&dlmgmt_name_avl, newlink,
1227 &name_where) == NULL)
1228 avl_insert(&dlmgmt_name_avl, newlink,
1229 name_where);
1230
1231 dlmgmt_advance(newlink);
1232 newlink->ll_flags |= req->ls_flags;
1233 }
1234 }
1235
1236 return (err);
1237 }
1238
1239 /*
1240 * Generate an entry in the link database.
1241 * Each entry has this format:
1242 * <link name> <prop0>=<type>,<val>;...;<propn>=<type>,<val>;
1243 */
1244 static void
1245 generate_link_line(dlmgmt_link_t *linkp, boolean_t persist, char *buf)
1246 {
1247 char tmpbuf[MAXLINELEN];
1248 char *ptr = tmpbuf;
1249 char *lim = tmpbuf + MAXLINELEN;
1250 dlmgmt_linkattr_t *cur_p = NULL;
1251 uint64_t u64;
1252
1253 ptr += snprintf(ptr, BUFLEN(lim, ptr), "%s\t", linkp->ll_link);
1254 if (!persist) {
1255 char zname[ZONENAME_MAX];
1256 /*
1257 * We store the linkid and the zone name in the active database
1258 * so that dlmgmtd can recover in the event that it is
1259 * restarted.
1260 */
1261 u64 = linkp->ll_linkid;
1262 ptr += write_uint64(ptr, BUFLEN(lim, ptr), "linkid", &u64);
1263
1264 if (getzonenamebyid(linkp->ll_zoneid, zname,
1265 sizeof (zname)) != -1) {
1266 ptr += write_str(ptr, BUFLEN(lim, ptr), "zone", zname);
1267 }
1268 }
1269 u64 = linkp->ll_class;
1270 ptr += write_uint64(ptr, BUFLEN(lim, ptr), "class", &u64);
1271 u64 = linkp->ll_media;
1272 ptr += write_uint64(ptr, BUFLEN(lim, ptr), "media", &u64);
1273
1274 /*
1275 * The daemon does not keep any active link attribute. Only store the
1276 * attributes if this request is for persistent configuration,
1277 */
1278 if (persist) {
1279 for (cur_p = linkp->ll_head; cur_p != NULL;
1280 cur_p = cur_p->lp_next) {
1281 ptr += translators[cur_p->lp_type].write_func(ptr,
1282 BUFLEN(lim, ptr), cur_p->lp_name, cur_p->lp_val);
1283 }
1284 }
1285
1286 if (ptr <= lim)
1287 (void) snprintf(buf, MAXLINELEN, "%s\n", tmpbuf);
1288 }
1289
1290 int
1291 dlmgmt_delete_db_entry(dlmgmt_link_t *linkp, uint32_t flags)
1292 {
1293 return (dlmgmt_db_update(DLMGMT_DB_OP_DELETE, linkp->ll_link, linkp,
1294 flags));
1295 }
1296
1297 int
1298 dlmgmt_write_db_entry(const char *entryname, dlmgmt_link_t *linkp,
1299 uint32_t flags)
1300 {
1301 int err;
1302
1303 if (flags & DLMGMT_PERSIST) {
1304 if ((err = dlmgmt_db_update(DLMGMT_DB_OP_WRITE, entryname,
1305 linkp, DLMGMT_PERSIST)) != 0) {
1306 return (err);
1307 }
1308 }
1309
1310 if (flags & DLMGMT_ACTIVE) {
1311 if (((err = dlmgmt_db_update(DLMGMT_DB_OP_WRITE, entryname,
1312 linkp, DLMGMT_ACTIVE)) != 0) && (flags & DLMGMT_PERSIST)) {
1313 (void) dlmgmt_db_update(DLMGMT_DB_OP_DELETE, entryname,
1314 linkp, DLMGMT_PERSIST);
1315 return (err);
1316 }
1317 }
1318
1319 return (0);
1320 }
1321
1322 /*
1323 * Upgrade properties that have link IDs as values to link names. Because '.'
1324 * is a valid linkname character, the port separater for link aggregations
1325 * must be changed to ':'.
1326 */
1327 static void
1328 linkattr_upgrade(dlmgmt_linkattr_t *attrp)
1329 {
1330 datalink_id_t linkid;
1331 char *portidstr;
1332 char portname[MAXLINKNAMELEN + 1];
1333 dlmgmt_link_t *linkp;
1334 char *new_attr_val;
1335 size_t new_attr_sz;
1336 boolean_t upgraded = B_FALSE;
1337
1338 if (strcmp(attrp->lp_name, "linkover") == 0 ||
1339 strcmp(attrp->lp_name, "simnetpeer") == 0) {
1340 if (attrp->lp_type == DLADM_TYPE_UINT64) {
1341 linkid = (datalink_id_t)*(uint64_t *)attrp->lp_val;
1342 if ((linkp = link_by_id(linkid, GLOBAL_ZONEID)) == NULL)
1343 return;
1344 new_attr_sz = strlen(linkp->ll_link) + 1;
1345 if ((new_attr_val = malloc(new_attr_sz)) == NULL)
1346 return;
1347 (void) strcpy(new_attr_val, linkp->ll_link);
1348 upgraded = B_TRUE;
1349 }
1350 } else if (strcmp(attrp->lp_name, "portnames") == 0) {
1351 /*
1352 * The old format for "portnames" was
1353 * "<linkid>.[<linkid>.]...". The new format is
1354 * "<linkname>:[<linkname>:]...".
1355 */
1356 if (!isdigit(((char *)attrp->lp_val)[0]))
1357 return;
1358 new_attr_val = calloc(MAXLINKATTRVALLEN, sizeof (char));
1359 if (new_attr_val == NULL)
1360 return;
1361 portidstr = (char *)attrp->lp_val;
1362 while (*portidstr != '\0') {
1363 errno = 0;
1364 linkid = strtol(portidstr, &portidstr, 10);
1365 if (linkid == 0 || *portidstr != '.' ||
1366 (linkp = link_by_id(linkid, GLOBAL_ZONEID)) ==
1367 NULL) {
1368 free(new_attr_val);
1369 return;
1370 }
1371 (void) snprintf(portname, sizeof (portname), "%s:",
1372 linkp->ll_link);
1373 if (strlcat(new_attr_val, portname,
1374 MAXLINKATTRVALLEN) >= MAXLINKATTRVALLEN) {
1375 free(new_attr_val);
1376 return;
1377 }
1378 /* skip the '.' delimiter */
1379 portidstr++;
1380 }
1381 new_attr_sz = strlen(new_attr_val) + 1;
1382 upgraded = B_TRUE;
1383 }
1384
1385 if (upgraded) {
1386 attrp->lp_type = DLADM_TYPE_STR;
1387 attrp->lp_sz = new_attr_sz;
1388 free(attrp->lp_val);
1389 attrp->lp_val = new_attr_val;
1390 }
1391 }
1392
1393 static void
1394 dlmgmt_db_upgrade(dlmgmt_link_t *linkp)
1395 {
1396 dlmgmt_linkattr_t *attrp;
1397
1398 for (attrp = linkp->ll_head; attrp != NULL; attrp = attrp->lp_next)
1399 linkattr_upgrade(attrp);
1400 }
1401
1402 static void
1403 dlmgmt_db_phys_activate(dlmgmt_link_t *linkp)
1404 {
1405 linkp->ll_flags |= DLMGMT_ACTIVE;
1406 (void) dlmgmt_write_db_entry(linkp->ll_link, linkp, DLMGMT_ACTIVE);
1407 }
1408
1409 static void
1410 dlmgmt_db_walk(zoneid_t zoneid, datalink_class_t class, db_walk_func_t *func)
1411 {
1412 dlmgmt_link_t *linkp;
1413
1414 for (linkp = avl_first(&dlmgmt_id_avl); linkp != NULL;
1415 linkp = AVL_NEXT(&dlmgmt_id_avl, linkp)) {
1416 if (linkp->ll_zoneid == zoneid && (linkp->ll_class & class))
1417 func(linkp);
1418 }
1419 }
1420
1421 /*
1422 * Attempt to mitigate one of the deadlocks in the dlmgmtd architecture.
1423 *
1424 * dlmgmt_db_init() calls dlmgmt_process_db_req() which eventually gets to
1425 * dlmgmt_zfop() which tries to fork, enter the zone and read the file.
1426 * Because of the upcall architecture of dlmgmtd this can lead to deadlock
1427 * with the following scenario:
1428 * a) the thread preparing to fork will have acquired the malloc locks
1429 * then attempt to suspend every thread in preparation to fork.
1430 * b) all of the upcalls will be blocked in door_ucred() trying to malloc()
1431 * and get the credentials of their caller.
1432 * c) we can't suspend the in-kernel thread making the upcall.
1433 *
1434 * Thus, we cannot serve door requests because we're blocked in malloc()
1435 * which fork() owns, but fork() is in turn blocked on the in-kernel thread
1436 * making the door upcall. This is a fundamental architectural problem with
1437 * any server handling upcalls and also trying to fork().
1438 *
1439 * To minimize the chance of this deadlock occuring, we check ahead of time to
1440 * see if the file we want to read actually exists in the zone (which it almost
1441 * never does), so we don't need fork in that case (i.e. rarely to never).
1442 */
1443 static boolean_t
1444 zone_file_exists(char *zoneroot, char *filename)
1445 {
1446 struct stat sb;
1447 char fname[MAXPATHLEN];
1448
1449 (void) snprintf(fname, sizeof (fname), "%s/%s", zoneroot, filename);
1450
1451 if (stat(fname, &sb) == -1)
1452 return (B_FALSE);
1453
1454 return (B_TRUE);
1455 }
1456
1457 /*
1458 * Initialize the datalink <link name, linkid> mapping and the link's
1459 * attributes list based on the configuration file /etc/dladm/datalink.conf
1460 * and the active configuration cache file
1461 * /etc/svc/volatile/dladm/datalink-management:default.cache.
1462 */
1463 int
1464 dlmgmt_db_init(zoneid_t zoneid, char *zoneroot)
1465 {
1466 dlmgmt_db_req_t *req;
1467 int err;
1468 boolean_t boot = B_FALSE;
1469 char tdir[MAXPATHLEN];
1470 char *path = cachefile;
1471
1472 if ((req = dlmgmt_db_req_alloc(DLMGMT_DB_OP_READ, NULL,
1473 DATALINK_INVALID_LINKID, zoneid, DLMGMT_ACTIVE, &err)) == NULL)
1474 return (err);
1475
1476 /* Handle running in a non-native branded zone (i.e. has /native) */
1477 if (zone_file_exists(zoneroot, "/native" DLMGMT_TMPFS_DIR)) {
1478 (void) snprintf(tdir, sizeof (tdir), "/native%s", cachefile);
1479 path = tdir;
1480 }
1481
1482 if (zone_file_exists(zoneroot, path)) {
1483 if ((err = dlmgmt_process_db_req(req)) != 0) {
1484 /*
1485 * If we get back ENOENT, that means that the active
1486 * configuration file doesn't exist yet, and is not an
1487 * error. We'll create it down below after we've
1488 * loaded the persistent configuration.
1489 */
1490 if (err != ENOENT)
1491 goto done;
1492 boot = B_TRUE;
1493 }
1494 } else {
1495 boot = B_TRUE;
1496 }
1497
1498 req->ls_flags = DLMGMT_PERSIST;
1499 err = dlmgmt_process_db_req(req);
1500 if (err != 0 && err != ENOENT)
1501 goto done;
1502 err = 0;
1503 if (rewrite_needed) {
1504 /*
1505 * First update links in memory, then dump the entire db to
1506 * disk.
1507 */
1508 dlmgmt_db_walk(zoneid, DATALINK_CLASS_ALL, dlmgmt_db_upgrade);
1509 req->ls_op = DLMGMT_DB_OP_WRITE;
1510 req->ls_linkid = DATALINK_ALL_LINKID;
1511 if ((err = dlmgmt_process_db_req(req)) != 0 &&
1512 err != EINPROGRESS)
1513 goto done;
1514 }
1515 if (boot) {
1516 dlmgmt_db_walk(zoneid, DATALINK_CLASS_PHYS,
1517 dlmgmt_db_phys_activate);
1518 }
1519
1520 done:
1521 if (err == EINPROGRESS)
1522 err = 0;
1523 else
1524 free(req);
1525 return (err);
1526 }
1527
1528 /*
1529 * Remove all links in the given zoneid.
1530 *
1531 * We do this work in two different passes. In the first pass, we remove any
1532 * entry that hasn't been loaned and mark every entry that has been loaned as
1533 * something that is going to be tombstomed. In the second pass, we drop the
1534 * table lock for every entry and remove the tombstombed entry for our zone.
1535 */
1536 void
1537 dlmgmt_db_fini(zoneid_t zoneid)
1538 {
1539 dlmgmt_link_t *linkp = avl_first(&dlmgmt_name_avl), *next_linkp;
1540
1541 while (linkp != NULL) {
1542 next_linkp = AVL_NEXT(&dlmgmt_name_avl, linkp);
1543 if (linkp->ll_zoneid == zoneid) {
1544 boolean_t onloan = linkp->ll_onloan;
1545
1546 /*
1547 * Cleanup any VNICs that were loaned to the zone
1548 * before the zone goes away and we can no longer
1549 * refer to the VNIC by the name/zoneid.
1550 */
1551 if (onloan) {
1552 (void) dlmgmt_delete_db_entry(linkp,
1553 DLMGMT_ACTIVE);
1554 linkp->ll_tomb = B_TRUE;
1555 } else {
1556 (void) dlmgmt_destroy_common(linkp,
1557 DLMGMT_ACTIVE | DLMGMT_PERSIST);
1558 }
1559 }
1560 linkp = next_linkp;
1561 }
1562
1563 again:
1564 linkp = avl_first(&dlmgmt_name_avl);
1565 while (linkp != NULL) {
1566 vnic_ioc_delete_t ioc;
1567
1568 next_linkp = AVL_NEXT(&dlmgmt_name_avl, linkp);
1569
1570 if (linkp->ll_zoneid != zoneid) {
1571 linkp = next_linkp;
1572 continue;
1573 }
1574 ioc.vd_vnic_id = linkp->ll_linkid;
1575 if (linkp->ll_tomb != B_TRUE)
1576 abort();
1577
1578 /*
1579 * We have to drop the table lock while going up into the
1580 * kernel. If we hold the table lock while deleting a vnic, we
1581 * may get blocked on the mac perimeter and the holder of it may
1582 * want something from dlmgmtd.
1583 */
1584 dlmgmt_table_unlock();
1585
1586 if (ioctl(dladm_dld_fd(dld_handle),
1587 VNIC_IOC_DELETE, &ioc) < 0)
1588 dlmgmt_log(LOG_WARNING, "dlmgmt_db_fini "
1589 "delete VNIC ioctl failed %d %d",
1590 ioc.vd_vnic_id, errno);
1591
1592 /*
1593 * Even though we've dropped the lock, we know that nothing else
1594 * could have removed us. Therefore, it should be safe to go
1595 * through and delete ourselves, but do nothing else. We'll have
1596 * to restart iteration from the beginning. This can be painful.
1597 */
1598 dlmgmt_table_lock(B_TRUE);
1599
1600 (void) dlmgmt_destroy_common(linkp,
1601 DLMGMT_ACTIVE | DLMGMT_PERSIST);
1602 goto again;
1603 }
1604
1605 }