Print this page
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/cmd/dlmgmtd/dlmgmt_db.c
+++ new/usr/src/cmd/dlmgmtd/dlmgmt_db.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright 2015, Joyent Inc.
25 25 */
26 26
27 27 #include <assert.h>
28 28 #include <ctype.h>
29 29 #include <errno.h>
30 30 #include <fcntl.h>
31 31 #include <stdio.h>
32 32 #include <stdlib.h>
33 33 #include <string.h>
34 34 #include <strings.h>
35 35 #include <syslog.h>
36 36 #include <zone.h>
37 37 #include <sys/types.h>
38 38 #include <sys/stat.h>
39 39 #include <stropts.h>
40 40 #include <sys/conf.h>
41 41 #include <pthread.h>
42 42 #include <unistd.h>
43 43 #include <wait.h>
44 44 #include <libcontract.h>
45 45 #include <libcontract_priv.h>
46 46 #include <sys/contract/process.h>
47 47 #include <sys/vnic.h>
48 48 #include <zone.h>
49 49 #include "dlmgmt_impl.h"
50 50
51 51 typedef enum dlmgmt_db_op {
52 52 DLMGMT_DB_OP_WRITE,
53 53 DLMGMT_DB_OP_DELETE,
54 54 DLMGMT_DB_OP_READ
55 55 } dlmgmt_db_op_t;
56 56
57 57 typedef struct dlmgmt_db_req_s {
58 58 struct dlmgmt_db_req_s *ls_next;
59 59 dlmgmt_db_op_t ls_op;
60 60 char ls_link[MAXLINKNAMELEN];
61 61 datalink_id_t ls_linkid;
62 62 zoneid_t ls_zoneid;
63 63 uint32_t ls_flags; /* Either DLMGMT_ACTIVE or */
64 64 /* DLMGMT_PERSIST, not both. */
65 65 } dlmgmt_db_req_t;
66 66
67 67 /*
68 68 * List of pending db updates (e.g., because of a read-only filesystem).
69 69 */
70 70 static dlmgmt_db_req_t *dlmgmt_db_req_head = NULL;
71 71 static dlmgmt_db_req_t *dlmgmt_db_req_tail = NULL;
72 72
73 73 /*
74 74 * rewrite_needed is set to B_TRUE by process_link_line() if it encounters a
75 75 * line with an old format. This will cause the file being read to be
76 76 * re-written with the current format.
77 77 */
78 78 static boolean_t rewrite_needed;
79 79
80 80 static int dlmgmt_db_update(dlmgmt_db_op_t, const char *,
81 81 dlmgmt_link_t *, uint32_t);
82 82 static int dlmgmt_process_db_req(dlmgmt_db_req_t *);
83 83 static int dlmgmt_process_db_onereq(dlmgmt_db_req_t *, boolean_t);
84 84 static void *dlmgmt_db_update_thread(void *);
85 85 static boolean_t process_link_line(char *, dlmgmt_link_t *);
86 86 static int process_db_write(dlmgmt_db_req_t *, FILE *, FILE *);
87 87 static int process_db_read(dlmgmt_db_req_t *, FILE *);
88 88 static void generate_link_line(dlmgmt_link_t *, boolean_t, char *);
89 89
90 90 #define BUFLEN(lim, ptr) (((lim) > (ptr)) ? ((lim) - (ptr)) : 0)
91 91 #define MAXLINELEN 1024
92 92
93 93 typedef void db_walk_func_t(dlmgmt_link_t *);
94 94
95 95 /*
96 96 * Translator functions to go from dladm_datatype_t to character strings.
97 97 * Each function takes a pointer to a buffer, the size of the buffer,
98 98 * the name of the attribute, and the value to be written. The functions
99 99 * return the number of bytes written to the buffer. If the buffer is not big
100 100 * enough to hold the string representing the value, then nothing is written
101 101 * and 0 is returned.
102 102 */
103 103 typedef size_t write_func_t(char *, size_t, char *, void *);
104 104
105 105 /*
106 106 * Translator functions to read from a NULL terminated string buffer into
107 107 * something of the given DLADM_TYPE_*. The functions each return the number
108 108 * of bytes read from the string buffer. If there is an error reading data
109 109 * from the buffer, then 0 is returned. It is the caller's responsibility
110 110 * to free the data allocated by these functions.
111 111 */
112 112 typedef size_t read_func_t(char *, void **);
113 113
114 114 typedef struct translator_s {
115 115 const char *type_name;
116 116 write_func_t *write_func;
117 117 read_func_t *read_func;
118 118 } translator_t;
119 119
120 120 /*
121 121 * Translator functions, defined later but declared here so that
122 122 * the translator table can be defined.
123 123 */
124 124 static write_func_t write_str, write_boolean, write_uint64;
125 125 static read_func_t read_str, read_boolean, read_int64;
126 126
127 127 /*
128 128 * Translator table, indexed by dladm_datatype_t.
129 129 */
130 130 static translator_t translators[] = {
131 131 { "string", write_str, read_str },
132 132 { "boolean", write_boolean, read_boolean },
133 133 { "int", write_uint64, read_int64 }
134 134 };
135 135
136 136 static size_t ntranslators = sizeof (translators) / sizeof (translator_t);
137 137
138 138 #define LINK_PROPERTY_DELIMINATOR ";"
139 139 #define LINK_PROPERTY_TYPE_VALUE_SEP ","
140 140 #define BASE_PROPERTY_LENGTH(t, n) (strlen(translators[(t)].type_name) +\
141 141 strlen(LINK_PROPERTY_TYPE_VALUE_SEP) +\
142 142 strlen(LINK_PROPERTY_DELIMINATOR) +\
143 143 strlen((n)))
144 144 #define GENERATE_PROPERTY_STRING(buf, length, conv, name, type, val) \
145 145 (snprintf((buf), (length), "%s=%s%s" conv "%s", (name), \
146 146 translators[(type)].type_name, \
147 147 LINK_PROPERTY_TYPE_VALUE_SEP, (val), LINK_PROPERTY_DELIMINATOR))
148 148
149 149 /*
150 150 * Name of the cache file to keep the active <link name, linkid> mapping
151 151 */
152 152 char cachefile[MAXPATHLEN];
153 153
154 154 #define DLMGMT_PERSISTENT_DB_PATH "/etc/dladm/datalink.conf"
155 155 #define DLMGMT_MAKE_FILE_DB_PATH(buffer, persistent) \
156 156 (void) snprintf((buffer), MAXPATHLEN, "%s", \
157 157 (persistent) ? DLMGMT_PERSISTENT_DB_PATH : cachefile);
158 158
159 159 typedef struct zopen_arg {
160 160 const char *zopen_modestr;
161 161 int *zopen_pipe;
162 162 int zopen_fd;
163 163 } zopen_arg_t;
164 164
165 165 typedef struct zrename_arg {
166 166 const char *zrename_newname;
167 167 } zrename_arg_t;
168 168
169 169 typedef union zfoparg {
170 170 zopen_arg_t zfop_openarg;
171 171 zrename_arg_t zfop_renamearg;
172 172 } zfoparg_t;
173 173
174 174 typedef struct zfcbarg {
175 175 boolean_t zfarg_inglobalzone; /* is callback in global zone? */
176 176 zoneid_t zfarg_finglobalzone; /* is file in global zone? */
177 177 const char *zfarg_filename;
178 178 zfoparg_t *zfarg_oparg;
179 179 } zfarg_t;
180 180 #define zfarg_openarg zfarg_oparg->zfop_openarg
181 181 #define zfarg_renamearg zfarg_oparg->zfop_renamearg
182 182
183 183 /* zone file callback */
184 184 typedef int zfcb_t(zfarg_t *);
185 185
186 186 /*
187 187 * Execute an operation on filename relative to zoneid's zone root. If the
188 188 * file is in the global zone, then the zfcb() callback will simply be called
189 189 * directly. If the file is in a non-global zone, then zfcb() will be called
190 190 * both from the global zone's context, and from the non-global zone's context
191 191 * (from a fork()'ed child that has entered the non-global zone). This is
192 192 * done to allow the callback to communicate with itself if needed (e.g. to
193 193 * pass back the file descriptor of an opened file).
194 194 */
195 195 static int
196 196 dlmgmt_zfop(const char *filename, zoneid_t zoneid, zfcb_t *zfcb,
197 197 zfoparg_t *zfoparg)
198 198 {
199 199 int ctfd;
200 200 int err;
201 201 pid_t childpid;
202 202 siginfo_t info;
203 203 zfarg_t zfarg;
204 204 ctid_t ct;
205 205
206 206 if (zoneid != GLOBAL_ZONEID) {
207 207 /*
208 208 * We need to access a file that isn't in the global zone.
209 209 * Accessing non-global zone files from the global zone is
210 210 * unsafe (due to symlink attacks), we'll need to fork a child
211 211 * that enters the zone in question and executes the callback
212 212 * that will operate on the file.
213 213 *
214 214 * Before we proceed with this zone tango, we need to create a
215 215 * new process contract for the child, as required by
216 216 * zone_enter().
217 217 */
218 218 errno = 0;
219 219 ctfd = open64("/system/contract/process/template", O_RDWR);
220 220 if (ctfd == -1)
221 221 return (errno);
222 222 if ((err = ct_tmpl_set_critical(ctfd, 0)) != 0 ||
223 223 (err = ct_tmpl_set_informative(ctfd, 0)) != 0 ||
224 224 (err = ct_pr_tmpl_set_fatal(ctfd, CT_PR_EV_HWERR)) != 0 ||
225 225 (err = ct_pr_tmpl_set_param(ctfd, CT_PR_PGRPONLY)) != 0 ||
226 226 (err = ct_tmpl_activate(ctfd)) != 0) {
227 227 (void) close(ctfd);
228 228 return (err);
229 229 }
230 230 childpid = fork();
231 231 switch (childpid) {
232 232 case -1:
233 233 (void) ct_tmpl_clear(ctfd);
234 234 (void) close(ctfd);
235 235 return (err);
236 236 case 0:
237 237 (void) ct_tmpl_clear(ctfd);
238 238 (void) close(ctfd);
239 239 /*
240 240 * Elevate our privileges as zone_enter() requires all
241 241 * privileges.
242 242 */
243 243 if ((err = dlmgmt_elevate_privileges()) != 0)
244 244 _exit(err);
245 245 if (zone_enter(zoneid) == -1)
246 246 _exit(errno);
247 247 if ((err = dlmgmt_drop_privileges()) != 0)
248 248 _exit(err);
249 249 break;
250 250 default:
251 251 if (contract_latest(&ct) == -1)
252 252 ct = -1;
253 253 (void) ct_tmpl_clear(ctfd);
254 254 (void) close(ctfd);
255 255 if (waitid(P_PID, childpid, &info, WEXITED) == -1) {
256 256 (void) contract_abandon_id(ct);
257 257 return (errno);
258 258 }
259 259 (void) contract_abandon_id(ct);
260 260 if (info.si_status != 0)
261 261 return (info.si_status);
262 262 }
263 263 }
264 264
265 265 zfarg.zfarg_inglobalzone = (zoneid == GLOBAL_ZONEID || childpid != 0);
266 266 zfarg.zfarg_finglobalzone = (zoneid == GLOBAL_ZONEID);
267 267 zfarg.zfarg_filename = filename;
268 268 zfarg.zfarg_oparg = zfoparg;
269 269 err = zfcb(&zfarg);
270 270 if (!zfarg.zfarg_inglobalzone)
271 271 _exit(err);
272 272 return (err);
273 273 }
274 274
275 275 static int
276 276 dlmgmt_zopen_cb(zfarg_t *zfarg)
277 277 {
278 278 struct strrecvfd recvfd;
279 279 boolean_t newfile = B_FALSE;
280 280 boolean_t inglobalzone = zfarg->zfarg_inglobalzone;
281 281 zoneid_t finglobalzone = zfarg->zfarg_finglobalzone;
282 282 const char *filename = zfarg->zfarg_filename;
283 283 const char *modestr = zfarg->zfarg_openarg.zopen_modestr;
284 284 int *p = zfarg->zfarg_openarg.zopen_pipe;
285 285 struct stat statbuf;
286 286 int oflags;
287 287 mode_t mode;
288 288 int fd = -1;
289 289 int err;
290 290
291 291 /* We only ever open a file for reading or writing, not both. */
292 292 oflags = (modestr[0] == 'r') ? O_RDONLY : O_WRONLY | O_CREAT | O_TRUNC;
293 293 mode = (modestr[0] == 'r') ? 0 : S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH;
294 294
295 295 /* Open the file if we're in the same zone as the file. */
296 296 if (inglobalzone == finglobalzone) {
297 297 /*
298 298 * First determine if we will be creating the file as part of
299 299 * opening it. If so, then we'll need to ensure that it has
300 300 * the proper ownership after having opened it.
301 301 */
302 302 if (oflags & O_CREAT) {
303 303 if (stat(filename, &statbuf) == -1) {
304 304 if (errno == ENOENT)
305 305 newfile = B_TRUE;
306 306 else
307 307 return (errno);
308 308 }
309 309 }
310 310 if ((fd = open(filename, oflags, mode)) == -1)
311 311 return (errno);
312 312 if (newfile) {
313 313 if (chown(filename, UID_DLADM, GID_NETADM) == -1) {
314 314 err = errno;
315 315 (void) close(fd);
316 316 return (err);
317 317 }
318 318 }
319 319 }
320 320
321 321 /*
322 322 * If we're not in the global zone, send the file-descriptor back to
323 323 * our parent in the global zone.
324 324 */
325 325 if (!inglobalzone) {
326 326 assert(!finglobalzone);
327 327 assert(fd != -1);
328 328 return (ioctl(p[1], I_SENDFD, fd) == -1 ? errno : 0);
329 329 }
330 330
331 331 /*
332 332 * At this point, we know we're in the global zone. If the file was
333 333 * in a non-global zone, receive the file-descriptor from our child in
334 334 * the non-global zone.
335 335 */
336 336 if (!finglobalzone) {
337 337 if (ioctl(p[0], I_RECVFD, &recvfd) == -1)
338 338 return (errno);
339 339 fd = recvfd.fd;
340 340 }
341 341
342 342 zfarg->zfarg_openarg.zopen_fd = fd;
343 343 return (0);
344 344 }
345 345
346 346 static int
347 347 dlmgmt_zunlink_cb(zfarg_t *zfarg)
348 348 {
349 349 if (zfarg->zfarg_inglobalzone != zfarg->zfarg_finglobalzone)
350 350 return (0);
351 351 return (unlink(zfarg->zfarg_filename) == 0 ? 0 : errno);
352 352 }
353 353
354 354 static int
355 355 dlmgmt_zrename_cb(zfarg_t *zfarg)
356 356 {
357 357 if (zfarg->zfarg_inglobalzone != zfarg->zfarg_finglobalzone)
358 358 return (0);
359 359 return (rename(zfarg->zfarg_filename,
360 360 zfarg->zfarg_renamearg.zrename_newname) == 0 ? 0 : errno);
361 361 }
362 362
363 363 /*
364 364 * Same as fopen(3C), except that it opens the file relative to zoneid's zone
365 365 * root.
366 366 */
367 367 static FILE *
368 368 dlmgmt_zfopen(const char *filename, const char *modestr, zoneid_t zoneid,
369 369 int *err)
370 370 {
371 371 int p[2];
372 372 zfoparg_t zfoparg;
373 373 FILE *fp = NULL;
374 374
375 375 if (zoneid != GLOBAL_ZONEID && pipe(p) == -1) {
376 376 *err = errno;
377 377 return (NULL);
378 378 }
379 379
380 380 zfoparg.zfop_openarg.zopen_modestr = modestr;
381 381 zfoparg.zfop_openarg.zopen_pipe = p;
382 382 *err = dlmgmt_zfop(filename, zoneid, dlmgmt_zopen_cb, &zfoparg);
383 383 if (zoneid != GLOBAL_ZONEID) {
384 384 (void) close(p[0]);
385 385 (void) close(p[1]);
386 386 }
387 387 if (*err == 0) {
388 388 fp = fdopen(zfoparg.zfop_openarg.zopen_fd, modestr);
389 389 if (fp == NULL) {
390 390 *err = errno;
391 391 (void) close(zfoparg.zfop_openarg.zopen_fd);
392 392 }
393 393 }
394 394 return (fp);
395 395 }
396 396
397 397 /*
398 398 * Same as rename(2), except that old and new are relative to zoneid's zone
399 399 * root.
400 400 */
401 401 static int
402 402 dlmgmt_zrename(const char *old, const char *new, zoneid_t zoneid)
403 403 {
404 404 zfoparg_t zfoparg;
405 405
406 406 zfoparg.zfop_renamearg.zrename_newname = new;
407 407 return (dlmgmt_zfop(old, zoneid, dlmgmt_zrename_cb, &zfoparg));
408 408 }
409 409
410 410 /*
411 411 * Same as unlink(2), except that filename is relative to zoneid's zone root.
412 412 */
413 413 static int
414 414 dlmgmt_zunlink(const char *filename, zoneid_t zoneid)
415 415 {
416 416 return (dlmgmt_zfop(filename, zoneid, dlmgmt_zunlink_cb, NULL));
417 417 }
418 418
419 419 static size_t
420 420 write_str(char *buffer, size_t buffer_length, char *name, void *value)
421 421 {
422 422 char *ptr = value;
423 423 size_t data_length = strnlen(ptr, buffer_length);
424 424
425 425 /*
426 426 * Strings are assumed to be NULL terminated. In order to fit in
427 427 * the buffer, the string's length must be less then buffer_length.
428 428 * If the value is empty, there's no point in writing it, in fact,
429 429 * we shouldn't even see that case.
430 430 */
431 431 if (data_length + BASE_PROPERTY_LENGTH(DLADM_TYPE_STR, name) ==
432 432 buffer_length || data_length == 0)
433 433 return (0);
434 434
435 435 /*
436 436 * Since we know the string will fit in the buffer, snprintf will
437 437 * always return less than buffer_length, so we can just return
438 438 * whatever snprintf returns.
439 439 */
440 440 return (GENERATE_PROPERTY_STRING(buffer, buffer_length, "%s",
441 441 name, DLADM_TYPE_STR, ptr));
442 442 }
443 443
444 444 static size_t
445 445 write_boolean(char *buffer, size_t buffer_length, char *name, void *value)
446 446 {
447 447 boolean_t *ptr = value;
448 448
449 449 /*
450 450 * Booleans are either zero or one, so we only need room for two
451 451 * characters in the buffer.
452 452 */
453 453 if (buffer_length <= 1 + BASE_PROPERTY_LENGTH(DLADM_TYPE_BOOLEAN, name))
454 454 return (0);
455 455
456 456 return (GENERATE_PROPERTY_STRING(buffer, buffer_length, "%d",
457 457 name, DLADM_TYPE_BOOLEAN, *ptr));
458 458 }
459 459
460 460 static size_t
461 461 write_uint64(char *buffer, size_t buffer_length, char *name, void *value)
462 462 {
463 463 uint64_t *ptr = value;
464 464
465 465 /*
466 466 * Limit checking for uint64_t is a little trickier.
467 467 */
468 468 if (snprintf(NULL, 0, "%lld", *ptr) +
469 469 BASE_PROPERTY_LENGTH(DLADM_TYPE_UINT64, name) >= buffer_length)
470 470 return (0);
471 471
472 472 return (GENERATE_PROPERTY_STRING(buffer, buffer_length, "%lld",
473 473 name, DLADM_TYPE_UINT64, *ptr));
474 474 }
475 475
476 476 static size_t
477 477 read_str(char *buffer, void **value)
478 478 {
479 479 char *ptr = calloc(MAXLINKATTRVALLEN, sizeof (char));
480 480 ssize_t len;
481 481
482 482 if (ptr == NULL || (len = strlcpy(ptr, buffer, MAXLINKATTRVALLEN))
483 483 >= MAXLINKATTRVALLEN) {
484 484 free(ptr);
485 485 return (0);
486 486 }
487 487
488 488 *(char **)value = ptr;
489 489
490 490 /* Account for NULL terminator */
491 491 return (len + 1);
492 492 }
493 493
494 494 static size_t
495 495 read_boolean(char *buffer, void **value)
496 496 {
497 497 boolean_t *ptr = calloc(1, sizeof (boolean_t));
498 498
499 499 if (ptr == NULL)
500 500 return (0);
501 501
502 502 *ptr = atoi(buffer);
503 503 *(boolean_t **)value = ptr;
504 504
505 505 return (sizeof (boolean_t));
506 506 }
507 507
508 508 static size_t
509 509 read_int64(char *buffer, void **value)
510 510 {
511 511 int64_t *ptr = calloc(1, sizeof (int64_t));
512 512
513 513 if (ptr == NULL)
514 514 return (0);
515 515
516 516 *ptr = (int64_t)atoll(buffer);
517 517 *(int64_t **)value = ptr;
518 518
519 519 return (sizeof (int64_t));
520 520 }
521 521
522 522 static dlmgmt_db_req_t *
523 523 dlmgmt_db_req_alloc(dlmgmt_db_op_t op, const char *linkname,
524 524 datalink_id_t linkid, zoneid_t zoneid, uint32_t flags, int *err)
525 525 {
526 526 dlmgmt_db_req_t *req;
527 527
528 528 if ((req = calloc(1, sizeof (dlmgmt_db_req_t))) == NULL) {
529 529 *err = errno;
530 530 } else {
531 531 req->ls_op = op;
532 532 if (linkname != NULL)
533 533 (void) strlcpy(req->ls_link, linkname, MAXLINKNAMELEN);
534 534 req->ls_linkid = linkid;
535 535 req->ls_zoneid = zoneid;
536 536 req->ls_flags = flags;
537 537 }
538 538 return (req);
539 539 }
540 540
541 541 /*
542 542 * Update the db entry with name "entryname" using information from "linkp".
543 543 */
544 544 static int
545 545 dlmgmt_db_update(dlmgmt_db_op_t op, const char *entryname, dlmgmt_link_t *linkp,
546 546 uint32_t flags)
547 547 {
|
↓ open down ↓ |
547 lines elided |
↑ open up ↑ |
548 548 dlmgmt_db_req_t *req;
549 549 int err;
550 550
551 551 /* It is either a persistent request or an active request, not both. */
552 552 assert((flags == DLMGMT_PERSIST) || (flags == DLMGMT_ACTIVE));
553 553
554 554 if ((req = dlmgmt_db_req_alloc(op, entryname, linkp->ll_linkid,
555 555 linkp->ll_zoneid, flags, &err)) == NULL)
556 556 return (err);
557 557
558 - /* If transient op and onloan, use the global zone cache file. */
559 - if (flags == DLMGMT_ACTIVE && linkp->ll_onloan)
560 - req->ls_zoneid = GLOBAL_ZONEID;
561 -
562 558 /*
563 559 * If the return error is EINPROGRESS, this request is handled
564 560 * asynchronously; return success.
565 561 */
566 562 err = dlmgmt_process_db_req(req);
567 563 if (err != EINPROGRESS)
568 564 free(req);
569 565 else
570 566 err = 0;
571 567 return (err);
572 568 }
573 569
574 570 #define DLMGMT_DB_OP_STR(op) \
575 571 (((op) == DLMGMT_DB_OP_READ) ? "read" : \
576 572 (((op) == DLMGMT_DB_OP_WRITE) ? "write" : "delete"))
577 573
578 574 #define DLMGMT_DB_CONF_STR(flag) \
579 575 (((flag) == DLMGMT_ACTIVE) ? "active" : \
580 576 (((flag) == DLMGMT_PERSIST) ? "persistent" : ""))
581 577
582 578 static int
583 579 dlmgmt_process_db_req(dlmgmt_db_req_t *req)
584 580 {
585 581 pthread_t tid;
586 582 boolean_t writeop;
587 583 int err;
588 584
589 585 /*
590 586 * If there are already pending "write" requests, queue this request in
591 587 * the pending list. Note that this function is called while the
592 588 * dlmgmt_rw_lock is held, so it is safe to access the global variables.
593 589 */
594 590 writeop = (req->ls_op != DLMGMT_DB_OP_READ);
595 591 if (writeop && (req->ls_flags == DLMGMT_PERSIST) &&
596 592 (dlmgmt_db_req_head != NULL)) {
597 593 dlmgmt_db_req_tail->ls_next = req;
598 594 dlmgmt_db_req_tail = req;
599 595 return (EINPROGRESS);
600 596 }
601 597
602 598 err = dlmgmt_process_db_onereq(req, writeop);
603 599 if (err != EINPROGRESS && err != 0 && err != ENOENT) {
604 600 /*
605 601 * Log the error unless the request processing is still in
606 602 * progress or if the configuration file hasn't been created
607 603 * yet (ENOENT).
608 604 */
609 605 dlmgmt_log(LOG_WARNING, "dlmgmt_process_db_onereq() %s "
610 606 "operation on %s configuration failed: %s",
611 607 DLMGMT_DB_OP_STR(req->ls_op),
612 608 DLMGMT_DB_CONF_STR(req->ls_flags), strerror(err));
613 609 }
614 610
615 611 if (err == EINPROGRESS) {
616 612 assert(req->ls_flags == DLMGMT_PERSIST);
617 613 assert(writeop && dlmgmt_db_req_head == NULL);
618 614 dlmgmt_db_req_tail = dlmgmt_db_req_head = req;
619 615 err = pthread_create(&tid, NULL, dlmgmt_db_update_thread, NULL);
620 616 if (err == 0)
621 617 return (EINPROGRESS);
622 618 }
623 619 return (err);
624 620 }
625 621
626 622 static int
627 623 dlmgmt_process_db_onereq(dlmgmt_db_req_t *req, boolean_t writeop)
628 624 {
629 625 int err = 0;
630 626 FILE *fp, *nfp = NULL;
631 627 char file[MAXPATHLEN];
632 628 char newfile[MAXPATHLEN];
633 629
634 630 DLMGMT_MAKE_FILE_DB_PATH(file, (req->ls_flags == DLMGMT_PERSIST));
635 631 fp = dlmgmt_zfopen(file, "r", req->ls_zoneid, &err);
636 632 /*
637 633 * Note that it is not an error if the file doesn't exist. If we're
638 634 * reading, we treat this case the same way as an empty file. If
639 635 * we're writing, the file will be created when we open the file for
640 636 * writing below.
641 637 */
642 638 if (fp == NULL && !writeop)
643 639 return (err);
644 640
645 641 if (writeop) {
646 642 (void) snprintf(newfile, MAXPATHLEN, "%s.new", file);
647 643 nfp = dlmgmt_zfopen(newfile, "w", req->ls_zoneid, &err);
648 644 if (nfp == NULL) {
649 645 /*
650 646 * EROFS can happen at boot when the file system is
651 647 * read-only. Return EINPROGRESS so that the caller
652 648 * can add this request to the pending request list
653 649 * and start a retry thread.
654 650 */
655 651 err = (errno == EROFS ? EINPROGRESS : errno);
656 652 goto done;
657 653 }
658 654 }
659 655 if (writeop) {
660 656 if ((err = process_db_write(req, fp, nfp)) == 0)
661 657 err = dlmgmt_zrename(newfile, file, req->ls_zoneid);
662 658 } else {
663 659 err = process_db_read(req, fp);
664 660 }
665 661
666 662 done:
667 663 if (nfp != NULL) {
668 664 (void) fclose(nfp);
669 665 if (err != 0)
670 666 (void) dlmgmt_zunlink(newfile, req->ls_zoneid);
671 667 }
672 668 (void) fclose(fp);
673 669 return (err);
674 670 }
675 671
676 672 /*ARGSUSED*/
677 673 static void *
678 674 dlmgmt_db_update_thread(void *arg)
679 675 {
680 676 dlmgmt_db_req_t *req;
681 677
682 678 dlmgmt_table_lock(B_TRUE);
683 679
684 680 assert(dlmgmt_db_req_head != NULL);
685 681 while ((req = dlmgmt_db_req_head) != NULL) {
686 682 assert(req->ls_flags == DLMGMT_PERSIST);
687 683 if (dlmgmt_process_db_onereq(req, B_TRUE) == EINPROGRESS) {
688 684 /*
689 685 * The filesystem is still read only. Go to sleep and
690 686 * try again.
691 687 */
692 688 dlmgmt_table_unlock();
693 689 (void) sleep(5);
694 690 dlmgmt_table_lock(B_TRUE);
695 691 continue;
696 692 }
697 693
698 694 /*
699 695 * The filesystem is no longer read only. Continue processing
700 696 * and remove the request from the pending list.
701 697 */
702 698 dlmgmt_db_req_head = req->ls_next;
703 699 if (dlmgmt_db_req_tail == req) {
704 700 assert(dlmgmt_db_req_head == NULL);
705 701 dlmgmt_db_req_tail = NULL;
706 702 }
707 703 free(req);
708 704 }
709 705
710 706 dlmgmt_table_unlock();
711 707 return (NULL);
712 708 }
713 709
714 710 static int
715 711 parse_linkprops(char *buf, dlmgmt_link_t *linkp)
716 712 {
717 713 boolean_t found_type = B_FALSE;
718 714 dladm_datatype_t type = DLADM_TYPE_STR;
719 715 int i, len;
|
↓ open down ↓ |
148 lines elided |
↑ open up ↑ |
720 716 char *curr;
721 717 char attr_name[MAXLINKATTRLEN];
722 718 size_t attr_buf_len = 0;
723 719 void *attr_buf = NULL;
724 720 boolean_t rename;
725 721
726 722 curr = buf;
727 723 len = strlen(buf);
728 724 attr_name[0] = '\0';
729 725 for (i = 0; i < len; i++) {
730 - rename = B_FALSE;
731 726 char c = buf[i];
732 727 boolean_t match = (c == '=' ||
733 728 (c == ',' && !found_type) || c == ';');
734 729
730 + rename = B_FALSE;
735 731 /*
736 732 * Move to the next character if there is no match and
737 733 * if we have not reached the last character.
738 734 */
739 735 if (!match && i != len - 1)
740 736 continue;
741 737
742 738 if (match) {
743 739 /*
744 740 * NUL-terminate the string pointed to by 'curr'.
745 741 */
746 742 buf[i] = '\0';
747 743 if (*curr == '\0')
748 744 goto parse_fail;
749 745 }
750 746
751 747 if (attr_name[0] != '\0' && found_type) {
752 748 /*
753 749 * We get here after we have processed the "<prop>="
754 750 * pattern. The pattern we are now interested in is
755 751 * "<val>;".
756 752 */
757 753 if (c == '=')
758 754 goto parse_fail;
759 755
760 756 if (strcmp(attr_name, "linkid") == 0) {
761 757 if (read_int64(curr, &attr_buf) == 0)
762 758 goto parse_fail;
763 759 linkp->ll_linkid =
764 760 (datalink_class_t)*(int64_t *)attr_buf;
765 761 } else if (strcmp(attr_name, "name") == 0) {
766 762 if (read_str(curr, &attr_buf) == 0)
767 763 goto parse_fail;
768 764 (void) snprintf(linkp->ll_link,
769 765 MAXLINKNAMELEN, "%s", attr_buf);
770 766 } else if (strcmp(attr_name, "class") == 0) {
771 767 if (read_int64(curr, &attr_buf) == 0)
772 768 goto parse_fail;
773 769 linkp->ll_class =
774 770 (datalink_class_t)*(int64_t *)attr_buf;
775 771 } else if (strcmp(attr_name, "media") == 0) {
776 772 if (read_int64(curr, &attr_buf) == 0)
777 773 goto parse_fail;
778 774 linkp->ll_media =
779 775 (uint32_t)*(int64_t *)attr_buf;
780 776 } else if (strcmp(attr_name, "zone") == 0) {
781 777 if (read_str(curr, &attr_buf) == 0)
782 778 goto parse_fail;
783 779 linkp->ll_zoneid = getzoneidbyname(attr_buf);
784 780 if (linkp->ll_zoneid == -1) {
785 781 if (errno == EFAULT)
786 782 abort();
787 783 /*
788 784 * If we can't find the zone, assign the
789 785 * link to the GZ and mark it for being
790 786 * renamed.
791 787 */
792 788 linkp->ll_zoneid = 0;
793 789 rename = B_TRUE;
794 790 }
795 791 } else {
796 792 attr_buf_len = translators[type].read_func(curr,
797 793 &attr_buf);
798 794 if (attr_buf_len == 0)
799 795 goto parse_fail;
800 796
801 797 if (linkattr_set(&(linkp->ll_head), attr_name,
802 798 attr_buf, attr_buf_len, type) != 0) {
803 799 free(attr_buf);
804 800 goto parse_fail;
805 801 }
806 802 }
807 803
808 804 free(attr_buf);
809 805 attr_name[0] = '\0';
810 806 found_type = B_FALSE;
811 807 } else if (attr_name[0] != '\0') {
812 808 /*
813 809 * Non-zero length attr_name and found_type of false
814 810 * indicates that we have not found the type for this
815 811 * attribute. The pattern now is "<type>,<val>;", we
816 812 * want the <type> part of the pattern.
817 813 */
818 814 for (type = 0; type < ntranslators; type++) {
819 815 if (strcmp(curr,
820 816 translators[type].type_name) == 0) {
821 817 found_type = B_TRUE;
822 818 break;
823 819 }
824 820 }
825 821
826 822 if (!found_type)
827 823 goto parse_fail;
828 824 } else {
829 825 /*
830 826 * A zero length attr_name indicates we are looking
831 827 * at the beginning of a link attribute.
832 828 */
833 829 if (c != '=')
834 830 goto parse_fail;
835 831
836 832 (void) snprintf(attr_name, MAXLINKATTRLEN, "%s", curr);
837 833 }
838 834
839 835 /*
840 836 * The zone that this link belongs to has died, we are
841 837 * reparenting it to the GZ and renaming it to avoid name
842 838 * collisions.
843 839 */
844 840 if (rename == B_TRUE) {
845 841 (void) snprintf(linkp->ll_link, MAXLINKNAMELEN,
846 842 "SUNWorphan%u", (uint16_t)(gethrtime() / 1000));
847 843 }
848 844 curr = buf + i + 1;
849 845 }
850 846
851 847 /* Correct any erroneous IPTUN datalink class constant in the file */
852 848 if (linkp->ll_class == 0x60) {
853 849 linkp->ll_class = DATALINK_CLASS_IPTUN;
854 850 rewrite_needed = B_TRUE;
855 851 }
856 852
857 853 return (0);
858 854
859 855 parse_fail:
860 856 /*
861 857 * Free linkp->ll_head (link attribute list)
862 858 */
863 859 linkattr_destroy(linkp);
864 860 return (-1);
865 861 }
866 862
867 863 static boolean_t
868 864 process_link_line(char *buf, dlmgmt_link_t *linkp)
869 865 {
870 866 int i, len, llen;
871 867 char *str, *lasts;
872 868 char tmpbuf[MAXLINELEN];
873 869
874 870 bzero(linkp, sizeof (*linkp));
875 871 linkp->ll_linkid = DATALINK_INVALID_LINKID;
876 872
877 873 /*
878 874 * Use a copy of buf for parsing so that we can do whatever we want.
879 875 */
880 876 (void) strlcpy(tmpbuf, buf, MAXLINELEN);
881 877
882 878 /*
883 879 * Skip leading spaces, blank lines, and comments.
884 880 */
885 881 len = strlen(tmpbuf);
886 882 for (i = 0; i < len; i++) {
887 883 if (!isspace(tmpbuf[i]))
888 884 break;
889 885 }
890 886 if (i == len || tmpbuf[i] == '#')
891 887 return (B_TRUE);
892 888
893 889 str = tmpbuf + i;
894 890 /*
895 891 * Find the link name and assign it to the link structure.
896 892 */
897 893 if (strtok_r(str, " \n\t", &lasts) == NULL)
898 894 goto fail;
899 895
900 896 llen = strlen(str);
901 897 /*
902 898 * Note that a previous version of the persistent datalink.conf file
903 899 * stored the linkid as the first field. In that case, the name will
904 900 * be obtained through parse_linkprops from a property with the format
905 901 * "name=<linkname>". If we encounter such a format, we set
906 902 * rewrite_needed so that dlmgmt_db_init() can rewrite the file with
907 903 * the new format after it's done reading in the data.
908 904 */
909 905 if (isdigit(str[0])) {
910 906 linkp->ll_linkid = atoi(str);
911 907 rewrite_needed = B_TRUE;
912 908 } else {
913 909 if (strlcpy(linkp->ll_link, str, sizeof (linkp->ll_link)) >=
914 910 sizeof (linkp->ll_link))
915 911 goto fail;
916 912 }
917 913
918 914 str += llen + 1;
919 915 if (str >= tmpbuf + len)
920 916 goto fail;
921 917
922 918 /*
923 919 * Now find the list of link properties.
924 920 */
925 921 if ((str = strtok_r(str, " \n\t", &lasts)) == NULL)
926 922 goto fail;
927 923
928 924 if (parse_linkprops(str, linkp) < 0)
929 925 goto fail;
930 926
931 927 return (B_TRUE);
932 928
933 929 fail:
934 930 /*
935 931 * Delete corrupted line.
936 932 */
937 933 buf[0] = '\0';
938 934 return (B_FALSE);
939 935 }
940 936
941 937 /*
942 938 * Find any properties in linkp that refer to "old", and rename to "new".
943 939 * Return B_TRUE if any renaming occurred.
944 940 */
945 941 static int
946 942 dlmgmt_attr_rename(dlmgmt_link_t *linkp, const char *old, const char *new,
947 943 boolean_t *renamed)
948 944 {
949 945 dlmgmt_linkattr_t *attrp;
950 946 char *newval = NULL, *pname;
951 947 char valcp[MAXLINKATTRVALLEN];
952 948 size_t newsize;
953 949
954 950 *renamed = B_FALSE;
955 951
956 952 if ((attrp = linkattr_find(linkp->ll_head, "linkover")) != NULL ||
957 953 (attrp = linkattr_find(linkp->ll_head, "simnetpeer")) != NULL) {
958 954 if (strcmp(old, (char *)attrp->lp_val) == 0) {
959 955 newsize = strlen(new) + 1;
960 956 if ((newval = malloc(newsize)) == NULL)
961 957 return (errno);
962 958 (void) strcpy(newval, new);
963 959 free(attrp->lp_val);
964 960 attrp->lp_val = newval;
965 961 attrp->lp_sz = newsize;
966 962 *renamed = B_TRUE;
967 963 }
968 964 return (0);
969 965 }
970 966
971 967 if ((attrp = linkattr_find(linkp->ll_head, "portnames")) == NULL)
972 968 return (0);
973 969
974 970 /* <linkname>:[<linkname>:]... */
975 971 if ((newval = calloc(MAXLINKATTRVALLEN, sizeof (char))) == NULL)
976 972 return (errno);
977 973
978 974 bcopy(attrp->lp_val, valcp, sizeof (valcp));
979 975 pname = strtok(valcp, ":");
980 976 while (pname != NULL) {
981 977 if (strcmp(pname, old) == 0) {
982 978 (void) strcat(newval, new);
983 979 *renamed = B_TRUE;
984 980 } else {
985 981 (void) strcat(newval, pname);
986 982 }
987 983 (void) strcat(newval, ":");
988 984 pname = strtok(NULL, ":");
989 985 }
990 986 if (*renamed) {
991 987 free(attrp->lp_val);
992 988 attrp->lp_val = newval;
993 989 attrp->lp_sz = strlen(newval) + 1;
994 990 } else {
995 991 free(newval);
996 992 }
997 993 return (0);
998 994 }
999 995
1000 996 static int
1001 997 process_db_write(dlmgmt_db_req_t *req, FILE *fp, FILE *nfp)
1002 998 {
1003 999 boolean_t done = B_FALSE;
1004 1000 int err = 0;
1005 1001 dlmgmt_link_t link_in_file, *linkp = NULL, *dblinkp;
1006 1002 boolean_t persist = (req->ls_flags == DLMGMT_PERSIST);
1007 1003 boolean_t writeall, rename, attr_renamed;
1008 1004 char buf[MAXLINELEN];
1009 1005
1010 1006 writeall = (req->ls_linkid == DATALINK_ALL_LINKID);
1011 1007
1012 1008 if (req->ls_op == DLMGMT_DB_OP_WRITE && !writeall) {
1013 1009 /*
1014 1010 * find the link in the avl tree with the given linkid.
1015 1011 */
1016 1012 linkp = link_by_id(req->ls_linkid, req->ls_zoneid);
1017 1013 if (linkp == NULL || (linkp->ll_flags & req->ls_flags) == 0) {
1018 1014 /*
1019 1015 * This link has already been changed. This could
1020 1016 * happen if the request is pending because of
1021 1017 * read-only file-system. If so, we are done.
1022 1018 */
1023 1019 return (0);
1024 1020 }
1025 1021 /*
1026 1022 * In the case of a rename, linkp's name has been updated to
1027 1023 * the new name, and req->ls_link is the old link name.
1028 1024 */
1029 1025 rename = (strcmp(req->ls_link, linkp->ll_link) != 0);
1030 1026 }
1031 1027
1032 1028 /*
1033 1029 * fp can be NULL if the file didn't initially exist and we're
1034 1030 * creating it as part of this write operation.
1035 1031 */
1036 1032 if (fp == NULL)
1037 1033 goto write;
1038 1034
1039 1035 while (err == 0 && fgets(buf, sizeof (buf), fp) != NULL &&
1040 1036 process_link_line(buf, &link_in_file)) {
1041 1037 /*
1042 1038 * Only the link name is needed. Free the memory allocated for
1043 1039 * the link attributes list of link_in_file.
1044 1040 */
1045 1041 linkattr_destroy(&link_in_file);
1046 1042
1047 1043 if (link_in_file.ll_link[0] == '\0' || done) {
1048 1044 /*
1049 1045 * this is a comment line or we are done updating the
1050 1046 * line for the specified link, write the rest of
1051 1047 * lines out.
1052 1048 */
1053 1049 if (fputs(buf, nfp) == EOF)
1054 1050 err = errno;
1055 1051 continue;
1056 1052 }
1057 1053
1058 1054 switch (req->ls_op) {
1059 1055 case DLMGMT_DB_OP_WRITE:
1060 1056 /*
1061 1057 * For write operations, we generate a new output line
1062 1058 * if we're either writing all links (writeall) or if
1063 1059 * the name of the link in the file matches the one
1064 1060 * we're looking for. Otherwise, we write out the
1065 1061 * buffer as-is.
1066 1062 *
1067 1063 * If we're doing a rename operation, ensure that any
1068 1064 * references to the link being renamed in link
1069 1065 * properties are also updated before we write
1070 1066 * anything.
1071 1067 */
1072 1068 if (writeall) {
1073 1069 linkp = link_by_name(link_in_file.ll_link,
1074 1070 req->ls_zoneid);
1075 1071 }
1076 1072 if (writeall || strcmp(req->ls_link,
1077 1073 link_in_file.ll_link) == 0) {
1078 1074 generate_link_line(linkp, persist, buf);
1079 1075 if (!writeall && !rename)
1080 1076 done = B_TRUE;
1081 1077 } else if (rename && persist) {
1082 1078 dblinkp = link_by_name(link_in_file.ll_link,
1083 1079 req->ls_zoneid);
1084 1080 err = dlmgmt_attr_rename(dblinkp, req->ls_link,
1085 1081 linkp->ll_link, &attr_renamed);
1086 1082 if (err != 0)
1087 1083 break;
1088 1084 if (attr_renamed) {
1089 1085 generate_link_line(dblinkp, persist,
1090 1086 buf);
1091 1087 }
1092 1088 }
1093 1089 if (fputs(buf, nfp) == EOF)
1094 1090 err = errno;
1095 1091 break;
1096 1092 case DLMGMT_DB_OP_DELETE:
1097 1093 /*
1098 1094 * Delete is simple. If buf does not represent the
1099 1095 * link we're deleting, write it out.
1100 1096 */
1101 1097 if (strcmp(req->ls_link, link_in_file.ll_link) != 0) {
1102 1098 if (fputs(buf, nfp) == EOF)
1103 1099 err = errno;
1104 1100 } else {
1105 1101 done = B_TRUE;
1106 1102 }
1107 1103 break;
1108 1104 case DLMGMT_DB_OP_READ:
1109 1105 default:
1110 1106 err = EINVAL;
1111 1107 break;
1112 1108 }
1113 1109 }
1114 1110
1115 1111 write:
1116 1112 /*
1117 1113 * If we get to the end of the file and have not seen what linkid
1118 1114 * points to, write it out then.
1119 1115 */
1120 1116 if (req->ls_op == DLMGMT_DB_OP_WRITE && !writeall && !rename && !done) {
1121 1117 generate_link_line(linkp, persist, buf);
1122 1118 done = B_TRUE;
1123 1119 if (fputs(buf, nfp) == EOF)
1124 1120 err = errno;
1125 1121 }
1126 1122
1127 1123 return (err);
1128 1124 }
1129 1125
1130 1126 static int
1131 1127 process_db_read(dlmgmt_db_req_t *req, FILE *fp)
1132 1128 {
1133 1129 avl_index_t name_where, id_where;
1134 1130 dlmgmt_link_t link_in_file, *newlink, *link_in_db;
1135 1131 char buf[MAXLINELEN];
1136 1132 int err = 0;
1137 1133
1138 1134 /*
1139 1135 * This loop processes each line of the configuration file.
1140 1136 */
1141 1137 while (fgets(buf, MAXLINELEN, fp) != NULL) {
1142 1138 if (!process_link_line(buf, &link_in_file)) {
1143 1139 err = EINVAL;
1144 1140 break;
1145 1141 }
1146 1142
1147 1143 /*
1148 1144 * Skip the comment line.
1149 1145 */
1150 1146 if (link_in_file.ll_link[0] == '\0') {
1151 1147 linkattr_destroy(&link_in_file);
1152 1148 continue;
1153 1149 }
1154 1150
1155 1151 if ((req->ls_flags & DLMGMT_ACTIVE) &&
1156 1152 link_in_file.ll_linkid == DATALINK_INVALID_LINKID) {
1157 1153 linkattr_destroy(&link_in_file);
1158 1154 continue;
1159 1155 }
1160 1156
1161 1157 link_in_file.ll_zoneid = req->ls_zoneid;
1162 1158 link_in_db = link_by_name(link_in_file.ll_link,
1163 1159 link_in_file.ll_zoneid);
1164 1160 if (link_in_db != NULL) {
1165 1161 /*
1166 1162 * If the link in the database already has the flag
1167 1163 * for this request set, then the entry is a
1168 1164 * duplicate. If it's not a duplicate, then simply
1169 1165 * turn on the appropriate flag on the existing link.
1170 1166 */
1171 1167 if (link_in_db->ll_flags & req->ls_flags) {
1172 1168 dlmgmt_log(LOG_WARNING, "Duplicate links "
1173 1169 "in the repository: %s",
1174 1170 link_in_file.ll_link);
1175 1171 linkattr_destroy(&link_in_file);
1176 1172 } else {
1177 1173 if (req->ls_flags & DLMGMT_PERSIST) {
1178 1174 /*
1179 1175 * Save the newly read properties into
1180 1176 * the existing link.
1181 1177 */
1182 1178 assert(link_in_db->ll_head == NULL);
1183 1179 link_in_db->ll_head =
1184 1180 link_in_file.ll_head;
1185 1181 } else {
1186 1182 linkattr_destroy(&link_in_file);
1187 1183 }
1188 1184 link_in_db->ll_flags |= req->ls_flags;
1189 1185 }
1190 1186 } else {
1191 1187 /*
1192 1188 * This is a new link. Allocate a new dlmgmt_link_t
1193 1189 * and add it to the trees.
1194 1190 */
1195 1191 newlink = calloc(1, sizeof (*newlink));
1196 1192 if (newlink == NULL) {
1197 1193 dlmgmt_log(LOG_WARNING, "Unable to allocate "
1198 1194 "memory to create new link %s",
1199 1195 link_in_file.ll_link);
1200 1196 linkattr_destroy(&link_in_file);
1201 1197 continue;
1202 1198 }
1203 1199 bcopy(&link_in_file, newlink, sizeof (*newlink));
1204 1200
1205 1201 if (newlink->ll_linkid == DATALINK_INVALID_LINKID)
1206 1202 newlink->ll_linkid = dlmgmt_nextlinkid;
1207 1203 if (avl_find(&dlmgmt_id_avl, newlink, &id_where) !=
1208 1204 NULL) {
1209 1205 dlmgmt_log(LOG_WARNING, "Link ID %d is already"
1210 1206 " in use, destroying link %s",
1211 1207 newlink->ll_linkid, newlink->ll_link);
1212 1208 link_destroy(newlink);
1213 1209 continue;
1214 1210 }
1215 1211
1216 1212 if ((req->ls_flags & DLMGMT_ACTIVE) &&
1217 1213 link_activate(newlink) != 0) {
1218 1214 dlmgmt_log(LOG_WARNING, "Unable to activate %s",
1219 1215 newlink->ll_link);
1220 1216 link_destroy(newlink);
1221 1217 continue;
1222 1218 }
1223 1219
1224 1220 avl_insert(&dlmgmt_id_avl, newlink, id_where);
1225 1221 /*
1226 1222 * link_activate call above can insert newlink in
1227 1223 * dlmgmt_name_avl tree when activating a link that is
1228 1224 * assigned to a NGZ.
1229 1225 */
1230 1226 if (avl_find(&dlmgmt_name_avl, newlink,
1231 1227 &name_where) == NULL)
1232 1228 avl_insert(&dlmgmt_name_avl, newlink,
1233 1229 name_where);
1234 1230
1235 1231 dlmgmt_advance(newlink);
1236 1232 newlink->ll_flags |= req->ls_flags;
1237 1233 }
1238 1234 }
1239 1235
1240 1236 return (err);
1241 1237 }
1242 1238
1243 1239 /*
1244 1240 * Generate an entry in the link database.
1245 1241 * Each entry has this format:
1246 1242 * <link name> <prop0>=<type>,<val>;...;<propn>=<type>,<val>;
1247 1243 */
1248 1244 static void
1249 1245 generate_link_line(dlmgmt_link_t *linkp, boolean_t persist, char *buf)
1250 1246 {
1251 1247 char tmpbuf[MAXLINELEN];
1252 1248 char *ptr = tmpbuf;
1253 1249 char *lim = tmpbuf + MAXLINELEN;
1254 1250 dlmgmt_linkattr_t *cur_p = NULL;
1255 1251 uint64_t u64;
1256 1252
1257 1253 ptr += snprintf(ptr, BUFLEN(lim, ptr), "%s\t", linkp->ll_link);
1258 1254 if (!persist) {
1259 1255 char zname[ZONENAME_MAX];
1260 1256 /*
1261 1257 * We store the linkid and the zone name in the active database
1262 1258 * so that dlmgmtd can recover in the event that it is
1263 1259 * restarted.
1264 1260 */
1265 1261 u64 = linkp->ll_linkid;
1266 1262 ptr += write_uint64(ptr, BUFLEN(lim, ptr), "linkid", &u64);
1267 1263
1268 1264 if (getzonenamebyid(linkp->ll_zoneid, zname,
1269 1265 sizeof (zname)) != -1) {
1270 1266 ptr += write_str(ptr, BUFLEN(lim, ptr), "zone", zname);
1271 1267 }
1272 1268 }
1273 1269 u64 = linkp->ll_class;
1274 1270 ptr += write_uint64(ptr, BUFLEN(lim, ptr), "class", &u64);
1275 1271 u64 = linkp->ll_media;
1276 1272 ptr += write_uint64(ptr, BUFLEN(lim, ptr), "media", &u64);
1277 1273
1278 1274 /*
1279 1275 * The daemon does not keep any active link attribute. Only store the
1280 1276 * attributes if this request is for persistent configuration,
1281 1277 */
1282 1278 if (persist) {
1283 1279 for (cur_p = linkp->ll_head; cur_p != NULL;
1284 1280 cur_p = cur_p->lp_next) {
1285 1281 ptr += translators[cur_p->lp_type].write_func(ptr,
1286 1282 BUFLEN(lim, ptr), cur_p->lp_name, cur_p->lp_val);
1287 1283 }
1288 1284 }
1289 1285
1290 1286 if (ptr <= lim)
1291 1287 (void) snprintf(buf, MAXLINELEN, "%s\n", tmpbuf);
1292 1288 }
1293 1289
1294 1290 int
1295 1291 dlmgmt_delete_db_entry(dlmgmt_link_t *linkp, uint32_t flags)
1296 1292 {
1297 1293 return (dlmgmt_db_update(DLMGMT_DB_OP_DELETE, linkp->ll_link, linkp,
1298 1294 flags));
1299 1295 }
1300 1296
1301 1297 int
1302 1298 dlmgmt_write_db_entry(const char *entryname, dlmgmt_link_t *linkp,
1303 1299 uint32_t flags)
1304 1300 {
1305 1301 int err;
1306 1302
1307 1303 if (flags & DLMGMT_PERSIST) {
1308 1304 if ((err = dlmgmt_db_update(DLMGMT_DB_OP_WRITE, entryname,
1309 1305 linkp, DLMGMT_PERSIST)) != 0) {
1310 1306 return (err);
1311 1307 }
1312 1308 }
1313 1309
1314 1310 if (flags & DLMGMT_ACTIVE) {
1315 1311 if (((err = dlmgmt_db_update(DLMGMT_DB_OP_WRITE, entryname,
1316 1312 linkp, DLMGMT_ACTIVE)) != 0) && (flags & DLMGMT_PERSIST)) {
1317 1313 (void) dlmgmt_db_update(DLMGMT_DB_OP_DELETE, entryname,
1318 1314 linkp, DLMGMT_PERSIST);
1319 1315 return (err);
1320 1316 }
1321 1317 }
1322 1318
1323 1319 return (0);
1324 1320 }
1325 1321
1326 1322 /*
1327 1323 * Upgrade properties that have link IDs as values to link names. Because '.'
1328 1324 * is a valid linkname character, the port separater for link aggregations
1329 1325 * must be changed to ':'.
1330 1326 */
1331 1327 static void
1332 1328 linkattr_upgrade(dlmgmt_linkattr_t *attrp)
1333 1329 {
1334 1330 datalink_id_t linkid;
1335 1331 char *portidstr;
1336 1332 char portname[MAXLINKNAMELEN + 1];
1337 1333 dlmgmt_link_t *linkp;
1338 1334 char *new_attr_val;
1339 1335 size_t new_attr_sz;
1340 1336 boolean_t upgraded = B_FALSE;
1341 1337
1342 1338 if (strcmp(attrp->lp_name, "linkover") == 0 ||
1343 1339 strcmp(attrp->lp_name, "simnetpeer") == 0) {
1344 1340 if (attrp->lp_type == DLADM_TYPE_UINT64) {
1345 1341 linkid = (datalink_id_t)*(uint64_t *)attrp->lp_val;
1346 1342 if ((linkp = link_by_id(linkid, GLOBAL_ZONEID)) == NULL)
1347 1343 return;
1348 1344 new_attr_sz = strlen(linkp->ll_link) + 1;
1349 1345 if ((new_attr_val = malloc(new_attr_sz)) == NULL)
1350 1346 return;
1351 1347 (void) strcpy(new_attr_val, linkp->ll_link);
1352 1348 upgraded = B_TRUE;
1353 1349 }
1354 1350 } else if (strcmp(attrp->lp_name, "portnames") == 0) {
1355 1351 /*
1356 1352 * The old format for "portnames" was
1357 1353 * "<linkid>.[<linkid>.]...". The new format is
1358 1354 * "<linkname>:[<linkname>:]...".
1359 1355 */
1360 1356 if (!isdigit(((char *)attrp->lp_val)[0]))
1361 1357 return;
1362 1358 new_attr_val = calloc(MAXLINKATTRVALLEN, sizeof (char));
1363 1359 if (new_attr_val == NULL)
1364 1360 return;
1365 1361 portidstr = (char *)attrp->lp_val;
1366 1362 while (*portidstr != '\0') {
1367 1363 errno = 0;
1368 1364 linkid = strtol(portidstr, &portidstr, 10);
1369 1365 if (linkid == 0 || *portidstr != '.' ||
1370 1366 (linkp = link_by_id(linkid, GLOBAL_ZONEID)) ==
1371 1367 NULL) {
1372 1368 free(new_attr_val);
1373 1369 return;
1374 1370 }
1375 1371 (void) snprintf(portname, sizeof (portname), "%s:",
1376 1372 linkp->ll_link);
1377 1373 if (strlcat(new_attr_val, portname,
1378 1374 MAXLINKATTRVALLEN) >= MAXLINKATTRVALLEN) {
1379 1375 free(new_attr_val);
1380 1376 return;
1381 1377 }
1382 1378 /* skip the '.' delimiter */
1383 1379 portidstr++;
1384 1380 }
1385 1381 new_attr_sz = strlen(new_attr_val) + 1;
1386 1382 upgraded = B_TRUE;
1387 1383 }
1388 1384
1389 1385 if (upgraded) {
1390 1386 attrp->lp_type = DLADM_TYPE_STR;
1391 1387 attrp->lp_sz = new_attr_sz;
1392 1388 free(attrp->lp_val);
1393 1389 attrp->lp_val = new_attr_val;
1394 1390 }
1395 1391 }
1396 1392
1397 1393 static void
1398 1394 dlmgmt_db_upgrade(dlmgmt_link_t *linkp)
1399 1395 {
1400 1396 dlmgmt_linkattr_t *attrp;
1401 1397
1402 1398 for (attrp = linkp->ll_head; attrp != NULL; attrp = attrp->lp_next)
1403 1399 linkattr_upgrade(attrp);
1404 1400 }
1405 1401
1406 1402 static void
1407 1403 dlmgmt_db_phys_activate(dlmgmt_link_t *linkp)
1408 1404 {
1409 1405 linkp->ll_flags |= DLMGMT_ACTIVE;
1410 1406 (void) dlmgmt_write_db_entry(linkp->ll_link, linkp, DLMGMT_ACTIVE);
1411 1407 }
1412 1408
1413 1409 static void
1414 1410 dlmgmt_db_walk(zoneid_t zoneid, datalink_class_t class, db_walk_func_t *func)
1415 1411 {
1416 1412 dlmgmt_link_t *linkp;
1417 1413
1418 1414 for (linkp = avl_first(&dlmgmt_id_avl); linkp != NULL;
1419 1415 linkp = AVL_NEXT(&dlmgmt_id_avl, linkp)) {
1420 1416 if (linkp->ll_zoneid == zoneid && (linkp->ll_class & class))
1421 1417 func(linkp);
1422 1418 }
1423 1419 }
1424 1420
1425 1421 /*
1426 1422 * Attempt to mitigate one of the deadlocks in the dlmgmtd architecture.
1427 1423 *
1428 1424 * dlmgmt_db_init() calls dlmgmt_process_db_req() which eventually gets to
1429 1425 * dlmgmt_zfop() which tries to fork, enter the zone and read the file.
1430 1426 * Because of the upcall architecture of dlmgmtd this can lead to deadlock
1431 1427 * with the following scenario:
1432 1428 * a) the thread preparing to fork will have acquired the malloc locks
1433 1429 * then attempt to suspend every thread in preparation to fork.
1434 1430 * b) all of the upcalls will be blocked in door_ucred() trying to malloc()
1435 1431 * and get the credentials of their caller.
1436 1432 * c) we can't suspend the in-kernel thread making the upcall.
1437 1433 *
1438 1434 * Thus, we cannot serve door requests because we're blocked in malloc()
1439 1435 * which fork() owns, but fork() is in turn blocked on the in-kernel thread
1440 1436 * making the door upcall. This is a fundamental architectural problem with
1441 1437 * any server handling upcalls and also trying to fork().
1442 1438 *
1443 1439 * To minimize the chance of this deadlock occuring, we check ahead of time to
1444 1440 * see if the file we want to read actually exists in the zone (which it almost
1445 1441 * never does), so we don't need fork in that case (i.e. rarely to never).
1446 1442 */
1447 1443 static boolean_t
1448 1444 zone_file_exists(char *zoneroot, char *filename)
1449 1445 {
1450 1446 struct stat sb;
1451 1447 char fname[MAXPATHLEN];
1452 1448
1453 1449 (void) snprintf(fname, sizeof (fname), "%s/%s", zoneroot, filename);
1454 1450
1455 1451 if (stat(fname, &sb) == -1)
1456 1452 return (B_FALSE);
1457 1453
1458 1454 return (B_TRUE);
1459 1455 }
1460 1456
1461 1457 /*
1462 1458 * Initialize the datalink <link name, linkid> mapping and the link's
1463 1459 * attributes list based on the configuration file /etc/dladm/datalink.conf
1464 1460 * and the active configuration cache file
1465 1461 * /etc/svc/volatile/dladm/datalink-management:default.cache.
1466 1462 */
1467 1463 int
1468 1464 dlmgmt_db_init(zoneid_t zoneid, char *zoneroot)
1469 1465 {
1470 1466 dlmgmt_db_req_t *req;
1471 1467 int err;
1472 1468 boolean_t boot = B_FALSE;
1473 1469 char tdir[MAXPATHLEN];
1474 1470 char *path = cachefile;
1475 1471
1476 1472 if ((req = dlmgmt_db_req_alloc(DLMGMT_DB_OP_READ, NULL,
1477 1473 DATALINK_INVALID_LINKID, zoneid, DLMGMT_ACTIVE, &err)) == NULL)
1478 1474 return (err);
1479 1475
1480 1476 /* Handle running in a non-native branded zone (i.e. has /native) */
1481 1477 if (zone_file_exists(zoneroot, "/native" DLMGMT_TMPFS_DIR)) {
1482 1478 (void) snprintf(tdir, sizeof (tdir), "/native%s", cachefile);
1483 1479 path = tdir;
1484 1480 }
1485 1481
1486 1482 if (zone_file_exists(zoneroot, path)) {
1487 1483 if ((err = dlmgmt_process_db_req(req)) != 0) {
1488 1484 /*
1489 1485 * If we get back ENOENT, that means that the active
1490 1486 * configuration file doesn't exist yet, and is not an
1491 1487 * error. We'll create it down below after we've
|
↓ open down ↓ |
747 lines elided |
↑ open up ↑ |
1492 1488 * loaded the persistent configuration.
1493 1489 */
1494 1490 if (err != ENOENT)
1495 1491 goto done;
1496 1492 boot = B_TRUE;
1497 1493 }
1498 1494 } else {
1499 1495 boot = B_TRUE;
1500 1496 }
1501 1497
1502 - if (zone_file_exists(zoneroot, DLMGMT_PERSISTENT_DB_PATH)) {
1503 - req->ls_flags = DLMGMT_PERSIST;
1504 - err = dlmgmt_process_db_req(req);
1505 - if (err != 0 && err != ENOENT)
1506 - goto done;
1507 - }
1498 + req->ls_flags = DLMGMT_PERSIST;
1499 + err = dlmgmt_process_db_req(req);
1500 + if (err != 0 && err != ENOENT)
1501 + goto done;
1508 1502 err = 0;
1509 1503 if (rewrite_needed) {
1510 1504 /*
1511 1505 * First update links in memory, then dump the entire db to
1512 1506 * disk.
1513 1507 */
1514 1508 dlmgmt_db_walk(zoneid, DATALINK_CLASS_ALL, dlmgmt_db_upgrade);
1515 1509 req->ls_op = DLMGMT_DB_OP_WRITE;
1516 1510 req->ls_linkid = DATALINK_ALL_LINKID;
1517 1511 if ((err = dlmgmt_process_db_req(req)) != 0 &&
1518 1512 err != EINPROGRESS)
1519 1513 goto done;
1520 1514 }
1521 1515 if (boot) {
1522 1516 dlmgmt_db_walk(zoneid, DATALINK_CLASS_PHYS,
1523 1517 dlmgmt_db_phys_activate);
1524 1518 }
1525 1519
1526 1520 done:
1527 1521 if (err == EINPROGRESS)
1528 1522 err = 0;
1529 1523 else
1530 1524 free(req);
1531 1525 return (err);
1532 1526 }
1533 1527
1534 1528 /*
1535 1529 * Remove all links in the given zoneid.
1536 1530 *
1537 1531 * We do this work in two different passes. In the first pass, we remove any
1538 1532 * entry that hasn't been loaned and mark every entry that has been loaned as
1539 1533 * something that is going to be tombstomed. In the second pass, we drop the
1540 1534 * table lock for every entry and remove the tombstombed entry for our zone.
1541 1535 */
1542 1536 void
1543 1537 dlmgmt_db_fini(zoneid_t zoneid)
1544 1538 {
1545 1539 dlmgmt_link_t *linkp = avl_first(&dlmgmt_name_avl), *next_linkp;
1546 1540
1547 1541 while (linkp != NULL) {
1548 1542 next_linkp = AVL_NEXT(&dlmgmt_name_avl, linkp);
1549 1543 if (linkp->ll_zoneid == zoneid) {
1550 1544 boolean_t onloan = linkp->ll_onloan;
1551 1545
1552 1546 /*
1553 1547 * Cleanup any VNICs that were loaned to the zone
1554 1548 * before the zone goes away and we can no longer
|
↓ open down ↓ |
37 lines elided |
↑ open up ↑ |
1555 1549 * refer to the VNIC by the name/zoneid.
1556 1550 */
1557 1551 if (onloan) {
1558 1552 (void) dlmgmt_delete_db_entry(linkp,
1559 1553 DLMGMT_ACTIVE);
1560 1554 linkp->ll_tomb = B_TRUE;
1561 1555 } else {
1562 1556 (void) dlmgmt_destroy_common(linkp,
1563 1557 DLMGMT_ACTIVE | DLMGMT_PERSIST);
1564 1558 }
1565 -
1566 1559 }
1567 1560 linkp = next_linkp;
1568 1561 }
1569 1562
1570 1563 again:
1571 1564 linkp = avl_first(&dlmgmt_name_avl);
1572 1565 while (linkp != NULL) {
1573 1566 vnic_ioc_delete_t ioc;
1574 1567
1575 1568 next_linkp = AVL_NEXT(&dlmgmt_name_avl, linkp);
1576 1569
1577 1570 if (linkp->ll_zoneid != zoneid) {
1578 1571 linkp = next_linkp;
1579 1572 continue;
1580 1573 }
1581 1574 ioc.vd_vnic_id = linkp->ll_linkid;
1582 1575 if (linkp->ll_tomb != B_TRUE)
1583 1576 abort();
1584 1577
1585 1578 /*
1586 1579 * We have to drop the table lock while going up into the
1587 1580 * kernel. If we hold the table lock while deleting a vnic, we
1588 1581 * may get blocked on the mac perimeter and the holder of it may
1589 1582 * want something from dlmgmtd.
1590 1583 */
1591 1584 dlmgmt_table_unlock();
1592 1585
1593 1586 if (ioctl(dladm_dld_fd(dld_handle),
1594 1587 VNIC_IOC_DELETE, &ioc) < 0)
1595 1588 dlmgmt_log(LOG_WARNING, "dlmgmt_db_fini "
1596 1589 "delete VNIC ioctl failed %d %d",
1597 1590 ioc.vd_vnic_id, errno);
1598 1591
1599 1592 /*
1600 1593 * Even though we've dropped the lock, we know that nothing else
1601 1594 * could have removed us. Therefore, it should be safe to go
1602 1595 * through and delete ourselves, but do nothing else. We'll have
1603 1596 * to restart iteration from the beginning. This can be painful.
1604 1597 */
1605 1598 dlmgmt_table_lock(B_TRUE);
1606 1599
1607 1600 (void) dlmgmt_destroy_common(linkp,
1608 1601 DLMGMT_ACTIVE | DLMGMT_PERSIST);
1609 1602 goto again;
1610 1603 }
1611 1604
1612 1605 }
|
↓ open down ↓ |
37 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX