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 2016 Joyent, Inc.
25 * Copyright 2023 Oxide Computer Company
26 */
27
28 #include <assert.h>
29 #include <ctype.h>
30 #include <errno.h>
31 #include <fcntl.h>
32 #include <stdio.h>
33 #include <stdlib.h>
34 #include <string.h>
35 #include <strings.h>
36 #include <syslog.h>
37 #include <zone.h>
38 #include <sys/types.h>
39 #include <sys/stat.h>
40 #include <stropts.h>
41 #include <sys/conf.h>
42 #include <pthread.h>
43 #include <unistd.h>
44 #include <wait.h>
45 #include <libcontract.h>
46 #include <libcontract_priv.h>
47 #include <sys/contract/process.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 /*
560 * If this is a transient link, then use the global zone cache file.
561 * This is in order to allow recovery from a dlmgmtd failure that
562 * leaves a zone in a 'down' state. In that state it is not possible
563 * to read the zone's cache file (since it is always done from a sub
564 * process running in the zone's context). As a result, datalinks would
565 * otherwise remain stuck in the zone.
566 */
567 if (flags == DLMGMT_ACTIVE && linkp->ll_transient)
568 req->ls_zoneid = GLOBAL_ZONEID;
569
570 /* If transient op and onloan, use the global zone cache file. */
571 if (flags == DLMGMT_ACTIVE && linkp->ll_onloan)
572 req->ls_zoneid = GLOBAL_ZONEID;
573
574 /*
575 * If the return error is EINPROGRESS, this request is handled
576 * asynchronously; return success.
577 */
578 err = dlmgmt_process_db_req(req);
579 if (err != EINPROGRESS)
580 free(req);
581 else
582 err = 0;
583 return (err);
584 }
585
586 #define DLMGMT_DB_OP_STR(op) \
587 (((op) == DLMGMT_DB_OP_READ) ? "read" : \
588 (((op) == DLMGMT_DB_OP_WRITE) ? "write" : "delete"))
589
590 #define DLMGMT_DB_CONF_STR(flag) \
591 (((flag) == DLMGMT_ACTIVE) ? "active" : \
592 (((flag) == DLMGMT_PERSIST) ? "persistent" : ""))
593
594 static int
595 dlmgmt_process_db_req(dlmgmt_db_req_t *req)
596 {
597 pthread_t tid;
598 boolean_t writeop;
599 int err;
600
601 /*
602 * If there are already pending "write" requests, queue this request in
603 * the pending list. Note that this function is called while the
604 * dlmgmt_rw_lock is held, so it is safe to access the global variables.
605 */
606 writeop = (req->ls_op != DLMGMT_DB_OP_READ);
607 if (writeop && (req->ls_flags == DLMGMT_PERSIST) &&
608 (dlmgmt_db_req_head != NULL)) {
609 dlmgmt_db_req_tail->ls_next = req;
610 dlmgmt_db_req_tail = req;
611 return (EINPROGRESS);
612 }
613
614 err = dlmgmt_process_db_onereq(req, writeop);
615 if (err != EINPROGRESS && err != 0 && err != ENOENT) {
616 /*
617 * Log the error unless the request processing is still in
618 * progress or if the configuration file hasn't been created
619 * yet (ENOENT).
620 */
621 dlmgmt_log(LOG_WARNING, "dlmgmt_process_db_onereq() %s "
622 "operation on %s configuration failed: %s",
623 DLMGMT_DB_OP_STR(req->ls_op),
624 DLMGMT_DB_CONF_STR(req->ls_flags), strerror(err));
625 }
626
627 if (err == EINPROGRESS) {
628 assert(req->ls_flags == DLMGMT_PERSIST);
629 assert(writeop && dlmgmt_db_req_head == NULL);
630 dlmgmt_db_req_tail = dlmgmt_db_req_head = req;
631 err = pthread_create(&tid, NULL, dlmgmt_db_update_thread, NULL);
632 if (err == 0)
633 return (EINPROGRESS);
634 }
635 return (err);
636 }
637
638 static int
639 dlmgmt_process_db_onereq(dlmgmt_db_req_t *req, boolean_t writeop)
640 {
641 int err = 0;
642 FILE *fp, *nfp = NULL;
643 char file[MAXPATHLEN];
644 char newfile[MAXPATHLEN];
645
646 DLMGMT_MAKE_FILE_DB_PATH(file, (req->ls_flags == DLMGMT_PERSIST));
647 fp = dlmgmt_zfopen(file, "r", req->ls_zoneid, &err);
648 /*
649 * Note that it is not an error if the file doesn't exist. If we're
650 * reading, we treat this case the same way as an empty file. If
651 * we're writing, the file will be created when we open the file for
652 * writing below.
653 */
654 if (fp == NULL && !writeop)
655 return (err);
656
657 if (writeop) {
658 (void) snprintf(newfile, MAXPATHLEN, "%s.new", file);
659 nfp = dlmgmt_zfopen(newfile, "w", req->ls_zoneid, &err);
660 if (nfp == NULL) {
661 /*
662 * EROFS can happen at boot when the file system is
663 * read-only. Return EINPROGRESS so that the caller
664 * can add this request to the pending request list
665 * and start a retry thread.
666 */
667 err = (errno == EROFS ? EINPROGRESS : errno);
668 goto done;
669 }
670 }
671 if (writeop) {
672 if ((err = process_db_write(req, fp, nfp)) == 0)
673 err = dlmgmt_zrename(newfile, file, req->ls_zoneid);
674 } else {
675 err = process_db_read(req, fp);
676 }
677
678 done:
679 if (nfp != NULL) {
680 (void) fclose(nfp);
681 if (err != 0)
682 (void) dlmgmt_zunlink(newfile, req->ls_zoneid);
683 }
684 (void) fclose(fp);
685 return (err);
686 }
687
688 /*ARGSUSED*/
689 static void *
690 dlmgmt_db_update_thread(void *arg)
691 {
692 dlmgmt_db_req_t *req;
693
694 dlmgmt_table_lock(B_TRUE);
695
696 assert(dlmgmt_db_req_head != NULL);
697 while ((req = dlmgmt_db_req_head) != NULL) {
698 assert(req->ls_flags == DLMGMT_PERSIST);
699 if (dlmgmt_process_db_onereq(req, B_TRUE) == EINPROGRESS) {
700 /*
701 * The filesystem is still read only. Go to sleep and
702 * try again.
703 */
704 dlmgmt_table_unlock();
705 (void) sleep(5);
706 dlmgmt_table_lock(B_TRUE);
707 continue;
708 }
709
710 /*
711 * The filesystem is no longer read only. Continue processing
712 * and remove the request from the pending list.
713 */
714 dlmgmt_db_req_head = req->ls_next;
715 if (dlmgmt_db_req_tail == req) {
716 assert(dlmgmt_db_req_head == NULL);
717 dlmgmt_db_req_tail = NULL;
718 }
719 free(req);
720 }
721
722 dlmgmt_table_unlock();
723 return (NULL);
724 }
725
726 static int
727 parse_linkprops(char *buf, dlmgmt_link_t *linkp)
728 {
729 boolean_t found_type = B_FALSE;
730 dladm_datatype_t type = DLADM_TYPE_STR;
731 int i, len;
732 char *curr;
733 char attr_name[MAXLINKATTRLEN];
734 size_t attr_buf_len = 0;
735 void *attr_buf = NULL;
736
737 curr = buf;
738 len = strlen(buf);
739 attr_name[0] = '\0';
740 for (i = 0; i < len; i++) {
741 char c = buf[i];
742 boolean_t match = (c == '=' ||
743 (c == ',' && !found_type) || c == ';');
744 boolean_t rename = B_FALSE;
745
746 /*
747 * Move to the next character if there is no match and
748 * if we have not reached the last character.
749 */
750 if (!match && i != len - 1)
751 continue;
752
753 if (match) {
754 /*
755 * NUL-terminate the string pointed to by 'curr'.
756 */
757 buf[i] = '\0';
758 if (*curr == '\0')
759 goto parse_fail;
760 }
761
762 if (attr_name[0] != '\0' && found_type) {
763 /*
764 * We get here after we have processed the "<prop>="
765 * pattern. The pattern we are now interested in is
766 * "<val>;".
767 */
768 if (c == '=')
769 goto parse_fail;
770
771 if (strcmp(attr_name, "linkid") == 0) {
772 if (read_int64(curr, &attr_buf) == 0)
773 goto parse_fail;
774 linkp->ll_linkid =
775 (datalink_class_t)*(int64_t *)attr_buf;
776 } else if (strcmp(attr_name, "name") == 0) {
777 if (read_str(curr, &attr_buf) == 0)
778 goto parse_fail;
779 (void) snprintf(linkp->ll_link,
780 MAXLINKNAMELEN, "%s", attr_buf);
781 } else if (strcmp(attr_name, "class") == 0) {
782 if (read_int64(curr, &attr_buf) == 0)
783 goto parse_fail;
784 linkp->ll_class =
785 (datalink_class_t)*(int64_t *)attr_buf;
786 } else if (strcmp(attr_name, "media") == 0) {
787 if (read_int64(curr, &attr_buf) == 0)
788 goto parse_fail;
789 linkp->ll_media =
790 (uint32_t)*(int64_t *)attr_buf;
791 } else if (strcmp(attr_name, "zone") == 0) {
792 if (read_str(curr, &attr_buf) == 0)
793 goto parse_fail;
794 linkp->ll_zoneid = getzoneidbyname(attr_buf);
795 if (linkp->ll_zoneid == -1) {
796 if (errno == EFAULT)
797 abort();
798 /*
799 * If we can't find the zone, assign the
800 * link to the GZ and mark it for being
801 * renamed.
802 */
803 linkp->ll_zoneid = 0;
804 rename = B_TRUE;
805 }
806 } else if (strcmp(attr_name, "transient") == 0) {
807 if (read_boolean(curr, &attr_buf) == 0)
808 goto parse_fail;
809 linkp->ll_transient = *(boolean_t *)attr_buf;
810 } else {
811 attr_buf_len = translators[type].read_func(curr,
812 &attr_buf);
813 if (attr_buf_len == 0)
814 goto parse_fail;
815
816 if (linkattr_set(&(linkp->ll_head), attr_name,
817 attr_buf, attr_buf_len, type) != 0) {
818 free(attr_buf);
819 goto parse_fail;
820 }
821 }
822
823 free(attr_buf);
824 attr_name[0] = '\0';
825 found_type = B_FALSE;
826 } else if (attr_name[0] != '\0') {
827 /*
828 * Non-zero length attr_name and found_type of false
829 * indicates that we have not found the type for this
830 * attribute. The pattern now is "<type>,<val>;", we
831 * want the <type> part of the pattern.
832 */
833 for (type = 0; type < ntranslators; type++) {
834 if (strcmp(curr,
835 translators[type].type_name) == 0) {
836 found_type = B_TRUE;
837 break;
838 }
839 }
840
841 if (!found_type)
842 goto parse_fail;
843 } else {
844 /*
845 * A zero length attr_name indicates we are looking
846 * at the beginning of a link attribute.
847 */
848 if (c != '=')
849 goto parse_fail;
850
851 (void) snprintf(attr_name, MAXLINKATTRLEN, "%s", curr);
852 }
853
854 /*
855 * The zone that this link belongs to has died, we are
856 * reparenting it to the GZ and renaming it to avoid name
857 * collisions.
858 */
859 if (rename) {
860 (void) snprintf(linkp->ll_link, MAXLINKNAMELEN,
861 "SUNWorphan%u", (uint16_t)(gethrtime() / 1000));
862 }
863
864 curr = buf + i + 1;
865 }
866
867 /* Correct any erroneous IPTUN datalink class constant in the file */
868 if (linkp->ll_class == 0x60) {
869 linkp->ll_class = DATALINK_CLASS_IPTUN;
870 rewrite_needed = B_TRUE;
871 }
872
873 return (0);
874
875 parse_fail:
876 /*
877 * Free linkp->ll_head (link attribute list)
878 */
879 linkattr_destroy(linkp);
880 return (-1);
881 }
882
883 static boolean_t
884 process_link_line(char *buf, dlmgmt_link_t *linkp)
885 {
886 int i, len, llen;
887 char *str, *lasts;
888 char tmpbuf[MAXLINELEN];
889
890 bzero(linkp, sizeof (*linkp));
891 linkp->ll_linkid = DATALINK_INVALID_LINKID;
892
893 /*
894 * Use a copy of buf for parsing so that we can do whatever we want.
895 */
896 (void) strlcpy(tmpbuf, buf, MAXLINELEN);
897
898 /*
899 * Skip leading spaces, blank lines, and comments.
900 */
901 len = strlen(tmpbuf);
902 for (i = 0; i < len; i++) {
903 if (!isspace(tmpbuf[i]))
904 break;
905 }
906 if (i == len || tmpbuf[i] == '#')
907 return (B_TRUE);
908
909 str = tmpbuf + i;
910 /*
911 * Find the link name and assign it to the link structure.
912 */
913 if (strtok_r(str, " \n\t", &lasts) == NULL)
914 goto fail;
915
916 llen = strlen(str);
917 /*
918 * Note that a previous version of the persistent datalink.conf file
919 * stored the linkid as the first field. In that case, the name will
920 * be obtained through parse_linkprops from a property with the format
921 * "name=<linkname>". If we encounter such a format, we set
922 * rewrite_needed so that dlmgmt_db_init() can rewrite the file with
923 * the new format after it's done reading in the data.
924 */
925 if (isdigit(str[0])) {
926 linkp->ll_linkid = atoi(str);
927 rewrite_needed = B_TRUE;
928 } else {
929 if (strlcpy(linkp->ll_link, str, sizeof (linkp->ll_link)) >=
930 sizeof (linkp->ll_link))
931 goto fail;
932 }
933
934 str += llen + 1;
935 if (str >= tmpbuf + len)
936 goto fail;
937
938 /*
939 * Now find the list of link properties.
940 */
941 if ((str = strtok_r(str, " \n\t", &lasts)) == NULL)
942 goto fail;
943
944 if (parse_linkprops(str, linkp) < 0)
945 goto fail;
946
947 return (B_TRUE);
948
949 fail:
950 /*
951 * Delete corrupted line.
952 */
953 buf[0] = '\0';
954 return (B_FALSE);
955 }
956
957 /*
958 * Find any properties in linkp that refer to "old", and rename to "new".
959 * Return B_TRUE if any renaming occurred.
960 */
961 static int
962 dlmgmt_attr_rename(dlmgmt_link_t *linkp, const char *old, const char *new,
963 boolean_t *renamed)
964 {
965 dlmgmt_linkattr_t *attrp;
966 char *newval = NULL, *pname;
967 char valcp[MAXLINKATTRVALLEN];
968 size_t newsize;
969
970 *renamed = B_FALSE;
971
972 if ((attrp = linkattr_find(linkp->ll_head, "linkover")) != NULL ||
973 (attrp = linkattr_find(linkp->ll_head, "simnetpeer")) != NULL) {
974 if (strcmp(old, (char *)attrp->lp_val) == 0) {
975 newsize = strlen(new) + 1;
976 if ((newval = malloc(newsize)) == NULL)
977 return (errno);
978 (void) strcpy(newval, new);
979 free(attrp->lp_val);
980 attrp->lp_val = newval;
981 attrp->lp_sz = newsize;
982 *renamed = B_TRUE;
983 }
984 return (0);
985 }
986
987 if ((attrp = linkattr_find(linkp->ll_head, "portnames")) == NULL)
988 return (0);
989
990 /* <linkname>:[<linkname>:]... */
991 if ((newval = calloc(MAXLINKATTRVALLEN, sizeof (char))) == NULL)
992 return (errno);
993
994 bcopy(attrp->lp_val, valcp, sizeof (valcp));
995 pname = strtok(valcp, ":");
996 while (pname != NULL) {
997 if (strcmp(pname, old) == 0) {
998 (void) strcat(newval, new);
999 *renamed = B_TRUE;
1000 } else {
1001 (void) strcat(newval, pname);
1002 }
1003 (void) strcat(newval, ":");
1004 pname = strtok(NULL, ":");
1005 }
1006 if (*renamed) {
1007 free(attrp->lp_val);
1008 attrp->lp_val = newval;
1009 attrp->lp_sz = strlen(newval) + 1;
1010 } else {
1011 free(newval);
1012 }
1013 return (0);
1014 }
1015
1016 static int
1017 process_db_write(dlmgmt_db_req_t *req, FILE *fp, FILE *nfp)
1018 {
1019 boolean_t done = B_FALSE;
1020 int err = 0;
1021 dlmgmt_link_t link_in_file, *linkp = NULL, *dblinkp;
1022 boolean_t persist = (req->ls_flags == DLMGMT_PERSIST);
1023 boolean_t writeall, rename, attr_renamed;
1024 char buf[MAXLINELEN];
1025
1026 writeall = (req->ls_linkid == DATALINK_ALL_LINKID);
1027
1028 if (req->ls_op == DLMGMT_DB_OP_WRITE && !writeall) {
1029 /*
1030 * find the link in the avl tree with the given linkid.
1031 */
1032 linkp = link_by_id(req->ls_linkid, req->ls_zoneid);
1033 if (linkp == NULL || (linkp->ll_flags & req->ls_flags) == 0) {
1034 /*
1035 * This link has already been changed. This could
1036 * happen if the request is pending because of
1037 * read-only file-system. If so, we are done.
1038 */
1039 return (0);
1040 }
1041 /*
1042 * In the case of a rename, linkp's name has been updated to
1043 * the new name, and req->ls_link is the old link name.
1044 */
1045 rename = (strcmp(req->ls_link, linkp->ll_link) != 0);
1046 }
1047
1048 /*
1049 * fp can be NULL if the file didn't initially exist and we're
1050 * creating it as part of this write operation.
1051 */
1052 if (fp == NULL)
1053 goto write;
1054
1055 while (err == 0 && fgets(buf, sizeof (buf), fp) != NULL &&
1056 process_link_line(buf, &link_in_file)) {
1057 /*
1058 * Only the link name is needed. Free the memory allocated for
1059 * the link attributes list of link_in_file.
1060 */
1061 linkattr_destroy(&link_in_file);
1062
1063 if (link_in_file.ll_link[0] == '\0' || done) {
1064 /*
1065 * this is a comment line or we are done updating the
1066 * line for the specified link, write the rest of
1067 * lines out.
1068 */
1069 if (fputs(buf, nfp) == EOF)
1070 err = errno;
1071 continue;
1072 }
1073
1074 switch (req->ls_op) {
1075 case DLMGMT_DB_OP_WRITE:
1076 /*
1077 * For write operations, we generate a new output line
1078 * if we're either writing all links (writeall) or if
1079 * the name of the link in the file matches the one
1080 * we're looking for. Otherwise, we write out the
1081 * buffer as-is.
1082 *
1083 * If we're doing a rename operation, ensure that any
1084 * references to the link being renamed in link
1085 * properties are also updated before we write
1086 * anything.
1087 */
1088 if (writeall) {
1089 linkp = link_by_name(link_in_file.ll_link,
1090 req->ls_zoneid);
1091 }
1092 if (writeall || strcmp(req->ls_link,
1093 link_in_file.ll_link) == 0) {
1094 generate_link_line(linkp, persist, buf);
1095 if (!writeall && !rename)
1096 done = B_TRUE;
1097 } else if (rename && persist) {
1098 dblinkp = link_by_name(link_in_file.ll_link,
1099 req->ls_zoneid);
1100 err = dlmgmt_attr_rename(dblinkp, req->ls_link,
1101 linkp->ll_link, &attr_renamed);
1102 if (err != 0)
1103 break;
1104 if (attr_renamed) {
1105 generate_link_line(dblinkp, persist,
1106 buf);
1107 }
1108 }
1109 if (fputs(buf, nfp) == EOF)
1110 err = errno;
1111 break;
1112 case DLMGMT_DB_OP_DELETE:
1113 /*
1114 * Delete is simple. If buf does not represent the
1115 * link we're deleting, write it out.
1116 */
1117 if (strcmp(req->ls_link, link_in_file.ll_link) != 0) {
1118 if (fputs(buf, nfp) == EOF)
1119 err = errno;
1120 } else {
1121 done = B_TRUE;
1122 }
1123 break;
1124 case DLMGMT_DB_OP_READ:
1125 default:
1126 err = EINVAL;
1127 break;
1128 }
1129 }
1130
1131 write:
1132 /*
1133 * If we get to the end of the file and have not seen what linkid
1134 * points to, write it out then.
1135 */
1136 if (req->ls_op == DLMGMT_DB_OP_WRITE && !writeall && !rename && !done) {
1137 generate_link_line(linkp, persist, buf);
1138 done = B_TRUE;
1139 if (fputs(buf, nfp) == EOF)
1140 err = errno;
1141 }
1142
1143 return (err);
1144 }
1145
1146 static int
1147 process_db_read(dlmgmt_db_req_t *req, FILE *fp)
1148 {
1149 avl_index_t name_where, id_where;
1150 dlmgmt_link_t link_in_file, *newlink, *link_in_db;
1151 char buf[MAXLINELEN];
1152 int err = 0;
1153
1154 /*
1155 * This loop processes each line of the configuration file.
1156 */
1157 while (fgets(buf, MAXLINELEN, fp) != NULL) {
1158 if (!process_link_line(buf, &link_in_file)) {
1159 err = EINVAL;
1160 break;
1161 }
1162
1163 /*
1164 * Skip the comment line.
1165 */
1166 if (link_in_file.ll_link[0] == '\0') {
1167 linkattr_destroy(&link_in_file);
1168 continue;
1169 }
1170
1171 if ((req->ls_flags & DLMGMT_ACTIVE) &&
1172 link_in_file.ll_linkid == DATALINK_INVALID_LINKID) {
1173 linkattr_destroy(&link_in_file);
1174 continue;
1175 }
1176
1177 assert(req->ls_zoneid == 0 ||
1178 link_in_file.ll_zoneid == req->ls_zoneid);
1179 link_in_db = link_by_name(link_in_file.ll_link,
1180 link_in_file.ll_zoneid);
1181 if (link_in_db != NULL) {
1182 /*
1183 * If the link in the database already has the flag
1184 * for this request set, then the entry is a
1185 * duplicate. If it's not a duplicate, then simply
1186 * turn on the appropriate flag on the existing link.
1187 */
1188 if (link_in_db->ll_flags & req->ls_flags) {
1189 dlmgmt_log(LOG_WARNING, "Duplicate links "
1190 "in the repository: %s",
1191 link_in_file.ll_link);
1192 linkattr_destroy(&link_in_file);
1193 } else {
1194 if (req->ls_flags & DLMGMT_PERSIST) {
1195 /*
1196 * Save the newly read properties into
1197 * the existing link.
1198 */
1199 assert(link_in_db->ll_head == NULL);
1200 link_in_db->ll_head =
1201 link_in_file.ll_head;
1202 } else {
1203 linkattr_destroy(&link_in_file);
1204 }
1205 link_in_db->ll_flags |= req->ls_flags;
1206 }
1207 } else {
1208 /*
1209 * This is a new link. Allocate a new dlmgmt_link_t
1210 * and add it to the trees.
1211 */
1212 newlink = calloc(1, sizeof (*newlink));
1213 if (newlink == NULL) {
1214 dlmgmt_log(LOG_WARNING, "Unable to allocate "
1215 "memory to create new link %s",
1216 link_in_file.ll_link);
1217 linkattr_destroy(&link_in_file);
1218 continue;
1219 }
1220 bcopy(&link_in_file, newlink, sizeof (*newlink));
1221
1222 if (newlink->ll_linkid == DATALINK_INVALID_LINKID)
1223 newlink->ll_linkid = dlmgmt_nextlinkid;
1224 if (avl_find(&dlmgmt_id_avl, newlink, &id_where) !=
1225 NULL) {
1226 dlmgmt_log(LOG_WARNING, "Link ID %d is already"
1227 " in use, destroying link %s",
1228 newlink->ll_linkid, newlink->ll_link);
1229 link_destroy(newlink);
1230 continue;
1231 }
1232
1233 if ((req->ls_flags & DLMGMT_ACTIVE) &&
1234 link_activate(newlink) != 0) {
1235 dlmgmt_log(LOG_WARNING, "Unable to activate %s",
1236 newlink->ll_link);
1237 link_destroy(newlink);
1238 continue;
1239 }
1240
1241 avl_insert(&dlmgmt_id_avl, newlink, id_where);
1242 /*
1243 * link_activate call above can insert newlink in
1244 * dlmgmt_name_avl tree when activating a link that is
1245 * assigned to a NGZ.
1246 */
1247 if (avl_find(&dlmgmt_name_avl, newlink,
1248 &name_where) == NULL)
1249 avl_insert(&dlmgmt_name_avl, newlink,
1250 name_where);
1251
1252 dlmgmt_advance(newlink);
1253 newlink->ll_flags |= req->ls_flags;
1254 }
1255 }
1256
1257 return (err);
1258 }
1259
1260 /*
1261 * Generate an entry in the link database.
1262 * Each entry has this format:
1263 * <link name> <prop0>=<type>,<val>;...;<propn>=<type>,<val>;
1264 */
1265 static void
1266 generate_link_line(dlmgmt_link_t *linkp, boolean_t persist, char *buf)
1267 {
1268 char tmpbuf[MAXLINELEN];
1269 char *ptr = tmpbuf;
1270 char *lim = tmpbuf + MAXLINELEN;
1271 dlmgmt_linkattr_t *cur_p = NULL;
1272 uint64_t u64;
1273
1274 ptr += snprintf(ptr, BUFLEN(lim, ptr), "%s\t", linkp->ll_link);
1275 if (!persist) {
1276 char zname[ZONENAME_MAX];
1277
1278 /*
1279 * We store the linkid and the zone name in the active database
1280 * so that dlmgmtd can recover in the event that it is
1281 * restarted.
1282 */
1283 u64 = linkp->ll_linkid;
1284 ptr += write_uint64(ptr, BUFLEN(lim, ptr), "linkid", &u64);
1285
1286 if (getzonenamebyid(linkp->ll_zoneid, zname,
1287 sizeof (zname)) != -1) {
1288 ptr += write_str(ptr, BUFLEN(lim, ptr), "zone", zname);
1289 }
1290 }
1291 u64 = linkp->ll_class;
1292 ptr += write_uint64(ptr, BUFLEN(lim, ptr), "class", &u64);
1293 u64 = linkp->ll_media;
1294 ptr += write_uint64(ptr, BUFLEN(lim, ptr), "media", &u64);
1295
1296 if (!persist && linkp->ll_transient) {
1297 boolean_t b = B_TRUE;
1298 ptr += write_boolean(ptr, BUFLEN(lim, ptr), "transient", &b);
1299 }
1300
1301 /*
1302 * The daemon does not keep any active link attribute. Only store the
1303 * attributes if this request is for persistent configuration,
1304 */
1305 if (persist) {
1306 for (cur_p = linkp->ll_head; cur_p != NULL;
1307 cur_p = cur_p->lp_next) {
1308 ptr += translators[cur_p->lp_type].write_func(ptr,
1309 BUFLEN(lim, ptr), cur_p->lp_name, cur_p->lp_val);
1310 }
1311 }
1312
1313 if (ptr <= lim)
1314 (void) snprintf(buf, MAXLINELEN, "%s\n", tmpbuf);
1315 }
1316
1317 int
1318 dlmgmt_delete_db_entry(dlmgmt_link_t *linkp, uint32_t flags)
1319 {
1320 return (dlmgmt_db_update(DLMGMT_DB_OP_DELETE, linkp->ll_link, linkp,
1321 flags));
1322 }
1323
1324 int
1325 dlmgmt_write_db_entry(const char *entryname, dlmgmt_link_t *linkp,
1326 uint32_t flags)
1327 {
1328 int err;
1329
1330 if (flags & DLMGMT_PERSIST) {
1331 if ((err = dlmgmt_db_update(DLMGMT_DB_OP_WRITE, entryname,
1332 linkp, DLMGMT_PERSIST)) != 0) {
1333 return (err);
1334 }
1335 }
1336
1337 if (flags & DLMGMT_ACTIVE) {
1338 if (((err = dlmgmt_db_update(DLMGMT_DB_OP_WRITE, entryname,
1339 linkp, DLMGMT_ACTIVE)) != 0) && (flags & DLMGMT_PERSIST)) {
1340 (void) dlmgmt_db_update(DLMGMT_DB_OP_DELETE, entryname,
1341 linkp, DLMGMT_PERSIST);
1342 return (err);
1343 }
1344 }
1345
1346 return (0);
1347 }
1348
1349 /*
1350 * Upgrade properties that have link IDs as values to link names. Because '.'
1351 * is a valid linkname character, the port separater for link aggregations
1352 * must be changed to ':'.
1353 */
1354 static void
1355 linkattr_upgrade(dlmgmt_linkattr_t *attrp)
1356 {
1357 datalink_id_t linkid;
1358 char *portidstr;
1359 char portname[MAXLINKNAMELEN + 1];
1360 dlmgmt_link_t *linkp;
1361 char *new_attr_val;
1362 size_t new_attr_sz;
1363 boolean_t upgraded = B_FALSE;
1364
1365 if (strcmp(attrp->lp_name, "linkover") == 0 ||
1366 strcmp(attrp->lp_name, "simnetpeer") == 0) {
1367 if (attrp->lp_type == DLADM_TYPE_UINT64) {
1368 linkid = (datalink_id_t)*(uint64_t *)attrp->lp_val;
1369 if ((linkp = link_by_id(linkid, GLOBAL_ZONEID)) == NULL)
1370 return;
1371 new_attr_sz = strlen(linkp->ll_link) + 1;
1372 if ((new_attr_val = malloc(new_attr_sz)) == NULL)
1373 return;
1374 (void) strcpy(new_attr_val, linkp->ll_link);
1375 upgraded = B_TRUE;
1376 }
1377 } else if (strcmp(attrp->lp_name, "portnames") == 0) {
1378 /*
1379 * The old format for "portnames" was
1380 * "<linkid>.[<linkid>.]...". The new format is
1381 * "<linkname>:[<linkname>:]...".
1382 */
1383 if (!isdigit(((char *)attrp->lp_val)[0]))
1384 return;
1385 new_attr_val = calloc(MAXLINKATTRVALLEN, sizeof (char));
1386 if (new_attr_val == NULL)
1387 return;
1388 portidstr = (char *)attrp->lp_val;
1389 while (*portidstr != '\0') {
1390 errno = 0;
1391 linkid = strtol(portidstr, &portidstr, 10);
1392 if (linkid == 0 || *portidstr != '.' ||
1393 (linkp = link_by_id(linkid, GLOBAL_ZONEID)) ==
1394 NULL) {
1395 free(new_attr_val);
1396 return;
1397 }
1398 (void) snprintf(portname, sizeof (portname), "%s:",
1399 linkp->ll_link);
1400 if (strlcat(new_attr_val, portname,
1401 MAXLINKATTRVALLEN) >= MAXLINKATTRVALLEN) {
1402 free(new_attr_val);
1403 return;
1404 }
1405 /* skip the '.' delimiter */
1406 portidstr++;
1407 }
1408 new_attr_sz = strlen(new_attr_val) + 1;
1409 upgraded = B_TRUE;
1410 }
1411
1412 if (upgraded) {
1413 attrp->lp_type = DLADM_TYPE_STR;
1414 attrp->lp_sz = new_attr_sz;
1415 free(attrp->lp_val);
1416 attrp->lp_val = new_attr_val;
1417 }
1418 }
1419
1420 static void
1421 dlmgmt_db_upgrade(dlmgmt_link_t *linkp)
1422 {
1423 dlmgmt_linkattr_t *attrp;
1424
1425 for (attrp = linkp->ll_head; attrp != NULL; attrp = attrp->lp_next)
1426 linkattr_upgrade(attrp);
1427 }
1428
1429 static void
1430 dlmgmt_db_phys_activate(dlmgmt_link_t *linkp)
1431 {
1432 linkp->ll_flags |= DLMGMT_ACTIVE;
1433 (void) dlmgmt_write_db_entry(linkp->ll_link, linkp, DLMGMT_ACTIVE);
1434 }
1435
1436 static void
1437 dlmgmt_db_walk(zoneid_t zoneid, datalink_class_t class, db_walk_func_t *func)
1438 {
1439 dlmgmt_link_t *linkp;
1440
1441 for (linkp = avl_first(&dlmgmt_id_avl); linkp != NULL;
1442 linkp = AVL_NEXT(&dlmgmt_id_avl, linkp)) {
1443 if (linkp->ll_zoneid == zoneid && (linkp->ll_class & class))
1444 func(linkp);
1445 }
1446 }
1447
1448 /*
1449 * Attempt to mitigate one of the deadlocks in the dlmgmtd architecture.
1450 *
1451 * dlmgmt_db_init() calls dlmgmt_process_db_req() which eventually gets to
1452 * dlmgmt_zfop() which tries to fork, enter the zone and read the file.
1453 * Because of the upcall architecture of dlmgmtd this can lead to deadlock
1454 * with the following scenario:
1455 * a) the thread preparing to fork will have acquired the malloc locks
1456 * then attempt to suspend every thread in preparation to fork.
1457 * b) all of the upcalls will be blocked in door_ucred() trying to malloc()
1458 * and get the credentials of their caller.
1459 * c) we can't suspend the in-kernel thread making the upcall.
1460 *
1461 * Thus, we cannot serve door requests because we're blocked in malloc()
1462 * which fork() owns, but fork() is in turn blocked on the in-kernel thread
1463 * making the door upcall. This is a fundamental architectural problem with
1464 * any server handling upcalls and also trying to fork().
1465 *
1466 * To minimize the chance of this deadlock occuring, we check ahead of time to
1467 * see if the file we want to read actually exists in the zone (which it almost
1468 * never does), so we don't need fork in that case (i.e. rarely to never).
1469 */
1470 static boolean_t
1471 zone_file_exists(char *zoneroot, char *filename)
1472 {
1473 struct stat sb;
1474 char fname[MAXPATHLEN];
1475
1476 (void) snprintf(fname, sizeof (fname), "%s/%s", zoneroot, filename);
1477
1478 if (stat(fname, &sb) == -1)
1479 return (B_FALSE);
1480
1481 return (B_TRUE);
1482 }
1483
1484 /*
1485 * Initialize the datalink <link name, linkid> mapping and the link's
1486 * attributes list based on the configuration file /etc/dladm/datalink.conf
1487 * and the active configuration cache file
1488 * /etc/svc/volatile/dladm/datalink-management:default.cache.
1489 */
1490 int
1491 dlmgmt_db_init(zoneid_t zoneid, char *zoneroot)
1492 {
1493 dlmgmt_db_req_t *req;
1494 int err;
1495 boolean_t boot = B_FALSE;
1496 char tdir[MAXPATHLEN];
1497 char *path = cachefile;
1498
1499 if ((req = dlmgmt_db_req_alloc(DLMGMT_DB_OP_READ, NULL,
1500 DATALINK_INVALID_LINKID, zoneid, DLMGMT_ACTIVE, &err)) == NULL)
1501 return (err);
1502
1503 /* Handle running in a non-native branded zone (i.e. has /native) */
1504 if (zone_file_exists(zoneroot, "/native" DLMGMT_TMPFS_DIR)) {
1505 (void) snprintf(tdir, sizeof (tdir), "/native%s", cachefile);
1506 path = tdir;
1507 }
1508
1509 if (zone_file_exists(zoneroot, path)) {
1510 if ((err = dlmgmt_process_db_req(req)) != 0) {
1511 /*
1512 * If we get back ENOENT, that means that the active
1513 * configuration file doesn't exist yet, and is not an
1514 * error. We'll create it down below after we've
1515 * loaded the persistent configuration.
1516 */
1517 if (err != ENOENT)
1518 goto done;
1519 boot = B_TRUE;
1520 }
1521 } else {
1522 boot = B_TRUE;
1523 }
1524
1525 if (zone_file_exists(zoneroot, DLMGMT_PERSISTENT_DB_PATH)) {
1526 req->ls_flags = DLMGMT_PERSIST;
1527 err = dlmgmt_process_db_req(req);
1528 if (err != 0 && err != ENOENT)
1529 goto done;
1530 }
1531 err = 0;
1532 if (rewrite_needed) {
1533 /*
1534 * First update links in memory, then dump the entire db to
1535 * disk.
1536 */
1537 dlmgmt_db_walk(zoneid, DATALINK_CLASS_ALL, dlmgmt_db_upgrade);
1538 req->ls_op = DLMGMT_DB_OP_WRITE;
1539 req->ls_linkid = DATALINK_ALL_LINKID;
1540 if ((err = dlmgmt_process_db_req(req)) != 0 &&
1541 err != EINPROGRESS)
1542 goto done;
1543 }
1544 if (boot) {
1545 dlmgmt_db_walk(zoneid, DATALINK_CLASS_PHYS,
1546 dlmgmt_db_phys_activate);
1547 }
1548
1549 done:
1550 if (err == EINPROGRESS)
1551 err = 0;
1552 else
1553 free(req);
1554 return (err);
1555 }
1556
1557 /*
1558 * Remove all links in the given zoneid.
1559 */
1560 void
1561 dlmgmt_db_fini(zoneid_t zoneid)
1562 {
1563 dlmgmt_link_t *linkp = avl_first(&dlmgmt_name_avl), *next_linkp;
1564
1565 while (linkp != NULL) {
1566 next_linkp = AVL_NEXT(&dlmgmt_name_avl, linkp);
1567 if (linkp->ll_zoneid == zoneid) {
1568 (void) dlmgmt_destroy_common(linkp,
1569 DLMGMT_ACTIVE | DLMGMT_PERSIST);
1570 }
1571 linkp = next_linkp;
1572 }
1573 }