xref: /titanic_52/usr/src/cmd/fm/fmd/common/fmd_log.c (revision b6c3f7863936abeae522e48a13887dddeb691a45)
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 2006 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * FMD Log File Subsystem
31  *
32  * Events are written to one of two log files as they are received or created;
33  * the error log tracks all ereport.* events received on the inbound event
34  * transport, and the fault log tracks all list.* events generated by fmd or
35  * its client modules.  In addition, we use the same log file format to cache
36  * state and events associated with ASRUs that are named in a diagnosis.
37  *
38  * The log files use the exacct format manipulated by libexacct(3LIB) and
39  * originally defined in PSARC 1999/119.  However, the exacct library was
40  * designed primarily for read-only clients and without the synchronous i/o
41  * considerations and seeking required for fmd, so we use libexacct here only
42  * to read and write the file headers and to pack data from memory into a file
43  * bytestream.  All of the i/o and file offset manipulations are performed by
44  * the fmd code below.  Our exacct file management uses the following grammar:
45  *
46  * file := hdr toc event*
47  * hdr := EXD_FMA_LABEL EXD_FMA_VERSION EXD_FMA_OSREL EXD_FMA_OSVER
48  * EXD_FMA_PLAT EXD_FMA_UUID
49  * toc := EXD_FMA_OFFSET
50  * event := EXD_FMA_TODSEC EXD_FMA_TODNSEC EXD_FMA_NVLIST evref* or legacy evref
51  * evref := EXD_FMA_UUID EXD_FMA_OFFSET
52  * legacy evref := EXD_FMA_MAJOR EXD_FMA_MINOR EXD_FMA_INODE EXD_FMA_OFFSET
53  *
54  * Any event can be uniquely identified by the tuple (file, offset) where file
55  * is encoded as (uuid) when we are cross-linking files.  For legacy file
56  * formats we still support encoding the reference as (major, minor, inode).
57  * Note that we break out of the file's dev_t into its two 32-bit components to
58  * permit development of either 32-bit or 64-bit log readers and writers; the
59  * LFS APIs do not yet export a 64-bit dev_t to fstat64(), so there is no way
60  * for a 32-bit application to retrieve and store a 64-bit dev_t.
61  *
62  * In order to replay events in the event of an fmd crash, events are initially
63  * written to the error log using the group catalog tag EXD_GROUP_RFMA by the
64  * fmd_log_append() function.  Later, once an event transitions from the
65  * received state to one of its other states (see fmd_event.c for details),
66  * fmd_log_commit() is used to overwrite the tag with EXD_GROUP_FMA, indicating
67  * that the event is fully processed and no longer needs to be replayed.
68  */
69 
70 #include <sys/types.h>
71 #include <sys/mkdev.h>
72 #include <sys/statvfs.h>
73 #include <sys/fm/protocol.h>
74 #include <sys/exacct_impl.h>
75 #include <uuid/uuid.h>
76 
77 #include <unistd.h>
78 #include <limits.h>
79 #include <fcntl.h>
80 #include <ctype.h>
81 
82 #include <fmd_alloc.h>
83 #include <fmd_error.h>
84 #include <fmd_string.h>
85 #include <fmd_event.h>
86 #include <fmd_conf.h>
87 #include <fmd_subr.h>
88 #include <fmd_case.h>
89 #include <fmd_log.h>
90 
91 #include <fmd.h>
92 
93 #define	CAT_FMA_RGROUP	(EXT_GROUP | EXC_DEFAULT | EXD_GROUP_RFMA)
94 #define	CAT_FMA_GROUP	(EXT_GROUP | EXC_DEFAULT | EXD_GROUP_FMA)
95 
96 #define	CAT_FMA_LABEL	(EXT_STRING | EXC_DEFAULT | EXD_FMA_LABEL)
97 #define	CAT_FMA_VERSION	(EXT_STRING | EXC_DEFAULT | EXD_FMA_VERSION)
98 #define	CAT_FMA_OSREL	(EXT_STRING | EXC_DEFAULT | EXD_FMA_OSREL)
99 #define	CAT_FMA_OSVER	(EXT_STRING | EXC_DEFAULT | EXD_FMA_OSVER)
100 #define	CAT_FMA_PLAT	(EXT_STRING | EXC_DEFAULT | EXD_FMA_PLAT)
101 #define	CAT_FMA_UUID	(EXT_STRING | EXC_DEFAULT | EXD_FMA_UUID)
102 #define	CAT_FMA_TODSEC	(EXT_UINT64 | EXC_DEFAULT | EXD_FMA_TODSEC)
103 #define	CAT_FMA_TODNSEC	(EXT_UINT64 | EXC_DEFAULT | EXD_FMA_TODNSEC)
104 #define	CAT_FMA_NVLIST	(EXT_RAW | EXC_DEFAULT | EXD_FMA_NVLIST)
105 #define	CAT_FMA_MAJOR	(EXT_UINT32 | EXC_DEFAULT | EXD_FMA_MAJOR)
106 #define	CAT_FMA_MINOR	(EXT_UINT32 | EXC_DEFAULT | EXD_FMA_MINOR)
107 #define	CAT_FMA_INODE	(EXT_UINT64 | EXC_DEFAULT | EXD_FMA_INODE)
108 #define	CAT_FMA_OFFSET	(EXT_UINT64 | EXC_DEFAULT | EXD_FMA_OFFSET)
109 
110 static ssize_t
111 fmd_log_write(fmd_log_t *lp, const void *buf, size_t n)
112 {
113 	ssize_t resid = n;
114 	ssize_t len;
115 
116 	ASSERT(MUTEX_HELD(&lp->log_lock));
117 
118 	while (resid != 0) {
119 		if ((len = write(lp->log_fd, buf, resid)) <= 0)
120 			break;
121 
122 		resid -= len;
123 		buf = (char *)buf + len;
124 	}
125 
126 	if (resid == n && n != 0)
127 		return (-1);
128 
129 	return (n - resid);
130 }
131 
132 static int
133 fmd_log_write_hdr(fmd_log_t *lp, const char *tag)
134 {
135 	ea_object_t hdr, toc, i0, i1, i2, i3, i4, i5, i6;
136 	const char *osrel, *osver, *plat;
137 	off64_t off = 0;
138 	int err = 0;
139 	uuid_t uuid;
140 
141 	(void) fmd_conf_getprop(fmd.d_conf, "osrelease", &osrel);
142 	(void) fmd_conf_getprop(fmd.d_conf, "osversion", &osver);
143 	(void) fmd_conf_getprop(fmd.d_conf, "platform", &plat);
144 	(void) fmd_conf_getprop(fmd.d_conf, "uuidlen", &lp->log_uuidlen);
145 
146 	lp->log_uuid = fmd_zalloc(lp->log_uuidlen + 1, FMD_SLEEP);
147 	uuid_generate(uuid);
148 	uuid_unparse(uuid, lp->log_uuid);
149 
150 	err |= ea_set_group(&hdr, CAT_FMA_GROUP);
151 	err |= ea_set_group(&toc, CAT_FMA_GROUP);
152 
153 	err |= ea_set_item(&i0, CAT_FMA_LABEL, tag, 0);
154 	err |= ea_set_item(&i1, CAT_FMA_VERSION, fmd.d_version, 0);
155 	err |= ea_set_item(&i2, CAT_FMA_OSREL, osrel, 0);
156 	err |= ea_set_item(&i3, CAT_FMA_OSVER, osver, 0);
157 	err |= ea_set_item(&i4, CAT_FMA_PLAT, plat, 0);
158 	err |= ea_set_item(&i5, CAT_FMA_UUID, lp->log_uuid, 0);
159 	err |= ea_set_item(&i6, CAT_FMA_OFFSET, &off, 0);
160 
161 	(void) ea_attach_to_group(&hdr, &i0);
162 	(void) ea_attach_to_group(&hdr, &i1);
163 	(void) ea_attach_to_group(&hdr, &i2);
164 	(void) ea_attach_to_group(&hdr, &i3);
165 	(void) ea_attach_to_group(&hdr, &i4);
166 	(void) ea_attach_to_group(&hdr, &i5);
167 	(void) ea_attach_to_group(&toc, &i6);
168 
169 	if (err == 0) {
170 		size_t hdr_size = ea_pack_object(&hdr, NULL, 0);
171 		size_t toc_size = ea_pack_object(&toc, NULL, 0);
172 
173 		size_t size = hdr_size + toc_size;
174 		void *buf = fmd_alloc(size, FMD_SLEEP);
175 
176 		(void) ea_pack_object(&hdr, buf, hdr_size);
177 		(void) ea_pack_object(&toc, (char *)buf + hdr_size, toc_size);
178 
179 		if ((lp->log_off = lseek64(lp->log_fd, 0, SEEK_END)) == -1L)
180 			fmd_panic("failed to seek log %s", lp->log_name);
181 
182 		if (fmd_log_write(lp, buf, size) != size)
183 			err = errno; /* save errno for fmd_set_errno() below */
184 
185 		fmd_free(buf, size);
186 
187 		lp->log_toc = lp->log_off + hdr_size;
188 		lp->log_beg = lp->log_off + hdr_size + toc_size;
189 		lp->log_off = lp->log_off + hdr_size + toc_size;
190 
191 		if (lp->log_off != lseek64(lp->log_fd, 0, SEEK_END))
192 			fmd_panic("eof off != log_off 0x%llx\n", lp->log_off);
193 	} else
194 		err = EFMD_LOG_EXACCT;
195 
196 	(void) ea_free_item(&i0, EUP_ALLOC);
197 	(void) ea_free_item(&i1, EUP_ALLOC);
198 	(void) ea_free_item(&i2, EUP_ALLOC);
199 	(void) ea_free_item(&i3, EUP_ALLOC);
200 	(void) ea_free_item(&i4, EUP_ALLOC);
201 	(void) ea_free_item(&i5, EUP_ALLOC);
202 	(void) ea_free_item(&i6, EUP_ALLOC);
203 
204 	return (err ? fmd_set_errno(err) : 0);
205 }
206 
207 static int
208 fmd_log_check_err(fmd_log_t *lp, int err, const char *msg)
209 {
210 	int eaerr = ea_error();
211 	char buf[BUFSIZ];
212 
213 	(void) snprintf(buf, sizeof (buf), "%s: %s: %s\n",
214 	    lp->log_name, msg, eaerr != EXR_OK ?
215 	    fmd_ea_strerror(eaerr) : "catalog tag mismatch");
216 
217 	fmd_error(err, buf);
218 	return (fmd_set_errno(err));
219 }
220 
221 static int
222 fmd_log_check_hdr(fmd_log_t *lp, const char *tag)
223 {
224 	int got_version = 0, got_label = 0;
225 	ea_object_t *grp, *obj;
226 	off64_t hdr_off, hdr_size;
227 	int dvers, fvers;
228 	const char *p;
229 
230 	ea_clear(&lp->log_ea); /* resync exacct file */
231 
232 	if ((hdr_off = lseek64(lp->log_fd, 0, SEEK_CUR)) == -1L)
233 		fmd_panic("failed to seek log %s", lp->log_name);
234 
235 	/*
236 	 * Read the first group of log meta-data: the write-once read-only
237 	 * file header.  We read all records in this group, ignoring all but
238 	 * the VERSION and LABEL, which are required and must be verified.
239 	 */
240 	if ((grp = ea_get_object_tree(&lp->log_ea, 1)) == NULL ||
241 	    grp->eo_catalog != CAT_FMA_GROUP) {
242 		ea_free_object(grp, EUP_ALLOC);
243 		return (fmd_log_check_err(lp, EFMD_LOG_INVAL,
244 		    "invalid fma hdr record group"));
245 	}
246 
247 	for (obj = grp->eo_group.eg_objs; obj != NULL; obj = obj->eo_next) {
248 		switch (obj->eo_catalog) {
249 		case CAT_FMA_VERSION:
250 			for (dvers = 0, p = fmd.d_version;
251 			    *p != '\0'; p++) {
252 				if (isdigit(*p))
253 					dvers = dvers * 10 + (*p - '0');
254 				else
255 					break;
256 			}
257 
258 			for (fvers = 0, p = obj->eo_item.ei_string;
259 			    *p != '\0'; p++) {
260 				if (isdigit(*p))
261 					fvers = fvers * 10 + (*p - '0');
262 				else
263 					break;
264 			}
265 
266 			if (fvers > dvers) {
267 				fmd_error(EFMD_LOG_INVAL, "%s: log version "
268 				    "%s is not supported by this daemon\n",
269 				    lp->log_name, obj->eo_item.ei_string);
270 				ea_free_object(grp, EUP_ALLOC);
271 				return (fmd_set_errno(EFMD_LOG_VERSION));
272 			}
273 
274 			got_version++;
275 			break;
276 
277 		case CAT_FMA_LABEL:
278 			if (strcmp(obj->eo_item.ei_string, tag) != 0) {
279 				fmd_error(EFMD_LOG_INVAL, "%s: log tag '%s' "
280 				    "does not matched expected tag '%s'\n",
281 				    lp->log_name, obj->eo_item.ei_string, tag);
282 				ea_free_object(grp, EUP_ALLOC);
283 				return (fmd_set_errno(EFMD_LOG_INVAL));
284 			}
285 			got_label++;
286 			break;
287 		case CAT_FMA_UUID:
288 			lp->log_uuid = fmd_strdup(obj->eo_item.ei_string,
289 			    FMD_SLEEP);
290 			lp->log_uuidlen = strlen(lp->log_uuid);
291 			break;
292 		}
293 	}
294 
295 	hdr_size = ea_pack_object(grp, NULL, 0);
296 	ea_free_object(grp, EUP_ALLOC);
297 
298 	if (!got_version || !got_label) {
299 		fmd_error(EFMD_LOG_INVAL, "%s: fmd hdr record group did not "
300 		    "include mandatory version and/or label\n", lp->log_name);
301 		return (fmd_set_errno(EFMD_LOG_INVAL));
302 	}
303 
304 	/*
305 	 * Read the second group of log meta-data: the table of contents.  We
306 	 * expect this group to contain an OFFSET object indicating the current
307 	 * value of log_skip.  We save this in our fmd_log_t and then return.
308 	 */
309 	if ((grp = ea_get_object_tree(&lp->log_ea, 1)) == NULL ||
310 	    grp->eo_catalog != CAT_FMA_GROUP || grp->eo_group.eg_nobjs < 1 ||
311 	    grp->eo_group.eg_objs->eo_catalog != CAT_FMA_OFFSET) {
312 		ea_free_object(grp, EUP_ALLOC);
313 		return (fmd_log_check_err(lp, EFMD_LOG_INVAL,
314 		    "invalid fma toc record group"));
315 	}
316 
317 	lp->log_toc = hdr_off + hdr_size;
318 	lp->log_beg = hdr_off + hdr_size + ea_pack_object(grp, NULL, 0);
319 	lp->log_off = lseek64(lp->log_fd, 0, SEEK_END);
320 	lp->log_skip = grp->eo_group.eg_objs->eo_item.ei_uint64;
321 
322 	if (lp->log_skip > lp->log_off) {
323 		fmd_error(EFMD_LOG_INVAL, "%s: skip %llx exceeds file size; "
324 		    "resetting to zero\n", lp->log_name, lp->log_skip);
325 		lp->log_skip = 0;
326 	}
327 
328 	ea_free_object(grp, EUP_ALLOC);
329 	return (0);
330 }
331 
332 static int
333 fmd_log_open_exacct(fmd_log_t *lp, int aflags, int oflags)
334 {
335 	int fd = dup(lp->log_fd);
336 	const char *creator;
337 
338 	(void) fmd_conf_getprop(fmd.d_conf, "log.creator", &creator);
339 
340 	if (ea_fdopen(&lp->log_ea, fd, creator, aflags, oflags) != 0) {
341 		fmd_error(EFMD_LOG_EXACCT, "%s: failed to open log file: %s\n",
342 		    lp->log_name, fmd_ea_strerror(ea_error()));
343 		(void) close(fd);
344 		return (fmd_set_errno(EFMD_LOG_EXACCT));
345 	}
346 
347 	lp->log_flags |= FMD_LF_EAOPEN;
348 	return (0);
349 }
350 
351 static fmd_log_t *
352 fmd_log_xopen(const char *root, const char *name, const char *tag, int oflags)
353 {
354 	fmd_log_t *lp = fmd_zalloc(sizeof (fmd_log_t), FMD_SLEEP);
355 
356 	char buf[PATH_MAX];
357 	char *slash = "/";
358 	size_t len;
359 	int err;
360 
361 	(void) pthread_mutex_init(&lp->log_lock, NULL);
362 	(void) pthread_cond_init(&lp->log_cv, NULL);
363 	(void) pthread_mutex_lock(&lp->log_lock);
364 
365 	if (strcmp(root, "") == 0)
366 		slash = "";
367 	len = strlen(root) + strlen(name) + strlen(slash) + 1; /* for "\0" */
368 	lp->log_name = fmd_alloc(len, FMD_SLEEP);
369 	(void) snprintf(lp->log_name, len, "%s%s%s", root, slash, name);
370 	lp->log_tag = fmd_strdup(tag, FMD_SLEEP);
371 	(void) fmd_conf_getprop(fmd.d_conf, "log.minfree", &lp->log_minfree);
372 
373 	if (strcmp(lp->log_tag, FMD_LOG_ERROR) == 0)
374 		lp->log_flags |= FMD_LF_REPLAY;
375 
376 	if (strcmp(lp->log_tag, FMD_LOG_XPRT) == 0)
377 		oflags &= ~O_SYNC;
378 
379 top:
380 	if ((lp->log_fd = open64(lp->log_name, oflags, 0644)) == -1 ||
381 	    fstat64(lp->log_fd, &lp->log_stat) == -1) {
382 		fmd_error(EFMD_LOG_OPEN, "failed to open log %s", lp->log_name);
383 		fmd_log_close(lp);
384 		return (NULL);
385 	}
386 
387 	/*
388 	 * If our open() created the log file, use libexacct to write a header
389 	 * and position the file just after the header (EO_TAIL).  If the log
390 	 * file already existed, use libexacct to validate the header and again
391 	 * position the file just after the header (EO_HEAD).  Note that we lie
392 	 * to libexacct about 'oflags' in order to achieve the desired result.
393 	 */
394 	if (lp->log_stat.st_size == 0) {
395 		err = fmd_log_open_exacct(lp, EO_VALID_HDR | EO_TAIL,
396 		    O_CREAT | O_WRONLY) || fmd_log_write_hdr(lp, tag);
397 	} else {
398 		err = fmd_log_open_exacct(lp, EO_VALID_HDR | EO_HEAD,
399 		    O_RDONLY) || fmd_log_check_hdr(lp, tag);
400 	}
401 
402 	/*
403 	 * If ea_fdopen() failed and the log was pre-existing, attempt to move
404 	 * it aside and start a new one.  If we created the log but failed to
405 	 * initialize it, then we have no choice but to give up (e.g. EROFS).
406 	 */
407 	if (err) {
408 		fmd_error(EFMD_LOG_OPEN,
409 		    "failed to initialize log %s", lp->log_name);
410 
411 		if (lp->log_flags & FMD_LF_EAOPEN) {
412 			lp->log_flags &= ~FMD_LF_EAOPEN;
413 			(void) ea_close(&lp->log_ea);
414 		}
415 
416 		(void) close(lp->log_fd);
417 		lp->log_fd = -1;
418 
419 		if (lp->log_stat.st_size != 0 && snprintf(buf,
420 		    sizeof (buf), "%s-", lp->log_name) < PATH_MAX &&
421 		    rename(lp->log_name, buf) == 0) {
422 			TRACE((FMD_DBG_LOG, "mv %s to %s", lp->log_name, buf));
423 			if (oflags & O_CREAT)
424 				goto top;
425 		}
426 
427 		fmd_log_close(lp);
428 		return (NULL);
429 	}
430 
431 	lp->log_refs++;
432 	(void) pthread_mutex_unlock(&lp->log_lock);
433 
434 	return (lp);
435 }
436 
437 fmd_log_t *
438 fmd_log_tryopen(const char *root, const char *name, const char *tag)
439 {
440 	return (fmd_log_xopen(root, name, tag, O_RDWR | O_SYNC));
441 }
442 
443 fmd_log_t *
444 fmd_log_open(const char *root, const char *name, const char *tag)
445 {
446 	return (fmd_log_xopen(root, name, tag, O_RDWR | O_CREAT | O_SYNC));
447 }
448 
449 void
450 fmd_log_close(fmd_log_t *lp)
451 {
452 	ASSERT(MUTEX_HELD(&lp->log_lock));
453 	ASSERT(lp->log_refs == 0);
454 
455 	if ((lp->log_flags & FMD_LF_EAOPEN) && ea_close(&lp->log_ea) != 0) {
456 		fmd_error(EFMD_LOG_CLOSE, "failed to close log %s: %s\n",
457 		    lp->log_name, fmd_ea_strerror(ea_error()));
458 	}
459 
460 	if (lp->log_fd >= 0 && close(lp->log_fd) != 0) {
461 		fmd_error(EFMD_LOG_CLOSE,
462 		    "failed to close log %s", lp->log_name);
463 	}
464 
465 	fmd_strfree(lp->log_name);
466 	fmd_strfree(lp->log_tag);
467 	if (lp->log_uuid != NULL)
468 		fmd_free(lp->log_uuid, lp->log_uuidlen + 1);
469 
470 	fmd_free(lp, sizeof (fmd_log_t));
471 }
472 
473 void
474 fmd_log_hold_pending(fmd_log_t *lp)
475 {
476 	(void) pthread_mutex_lock(&lp->log_lock);
477 
478 	lp->log_refs++;
479 	ASSERT(lp->log_refs != 0);
480 
481 	if (lp->log_flags & FMD_LF_REPLAY) {
482 		lp->log_pending++;
483 		ASSERT(lp->log_pending != 0);
484 	}
485 
486 	(void) pthread_mutex_unlock(&lp->log_lock);
487 }
488 
489 void
490 fmd_log_hold(fmd_log_t *lp)
491 {
492 	(void) pthread_mutex_lock(&lp->log_lock);
493 	lp->log_refs++;
494 	ASSERT(lp->log_refs != 0);
495 	(void) pthread_mutex_unlock(&lp->log_lock);
496 }
497 
498 void
499 fmd_log_rele(fmd_log_t *lp)
500 {
501 	(void) pthread_mutex_lock(&lp->log_lock);
502 	ASSERT(lp->log_refs != 0);
503 
504 	if (--lp->log_refs == 0)
505 		fmd_log_close(lp);
506 	else
507 		(void) pthread_mutex_unlock(&lp->log_lock);
508 }
509 
510 void
511 fmd_log_append(fmd_log_t *lp, fmd_event_t *e, fmd_case_t *cp)
512 {
513 	fmd_event_impl_t *ep = (fmd_event_impl_t *)e;
514 	fmd_case_impl_t *cip = (fmd_case_impl_t *)cp;
515 	int err = 0;
516 
517 	ea_object_t grp0, grp1, i0, i1, i2, *items;
518 	ea_object_t **fe = NULL;
519 	size_t nvsize, easize, itsize, frsize;
520 	char *nvbuf, *eabuf;
521 	statvfs64_t stv;
522 
523 	(void) pthread_mutex_lock(&ep->ev_lock);
524 
525 	ASSERT(ep->ev_flags & FMD_EVF_VOLATILE);
526 	ASSERT(ep->ev_log == NULL);
527 
528 	(void) nvlist_size(ep->ev_nvl, &nvsize, NV_ENCODE_XDR);
529 	nvbuf = fmd_alloc(nvsize, FMD_SLEEP);
530 	(void) nvlist_pack(ep->ev_nvl, &nvbuf, &nvsize, NV_ENCODE_XDR, 0);
531 
532 	if (lp->log_flags & FMD_LF_REPLAY)
533 		err |= ea_set_group(&grp0, CAT_FMA_RGROUP);
534 	else
535 		err |= ea_set_group(&grp0, CAT_FMA_GROUP);
536 
537 	err |= ea_set_item(&i0, CAT_FMA_TODSEC, &ep->ev_time.ftv_sec, 0);
538 	err |= ea_set_item(&i1, CAT_FMA_TODNSEC, &ep->ev_time.ftv_nsec, 0);
539 	err |= ea_set_item(&i2, CAT_FMA_NVLIST, nvbuf, nvsize);
540 
541 	if (err != 0) {
542 		(void) pthread_mutex_unlock(&ep->ev_lock);
543 		err = EFMD_LOG_EXACCT;
544 		goto exerr;
545 	}
546 
547 	(void) ea_attach_to_group(&grp0, &i0);
548 	(void) ea_attach_to_group(&grp0, &i1);
549 	(void) ea_attach_to_group(&grp0, &i2);
550 
551 	/*
552 	 * If this event has a case associated with it (i.e. it is a list),
553 	 * then allocate a block of ea_object_t's and fill in a group for
554 	 * each event saved in the case's item list.  For each such group,
555 	 * we attach it to grp1, which in turn will be attached to grp0.
556 	 */
557 	if (cp != NULL) {
558 		ea_object_t *egrp, *ip, **fp;
559 		fmd_event_impl_t *eip;
560 		fmd_case_item_t *cit;
561 
562 		(void) ea_set_group(&grp1, CAT_FMA_GROUP);
563 		frsize = sizeof (ea_object_t *) * cip->ci_nitems;
564 		itsize = sizeof (ea_object_t) * cip->ci_nitems * 5;
565 		items = ip = fmd_alloc(itsize, FMD_SLEEP);
566 
567 		for (cit = cip->ci_items; cit != NULL; cit = cit->cit_next) {
568 			major_t maj;
569 			minor_t min;
570 
571 			eip = (fmd_event_impl_t *)cit->cit_event;
572 
573 			if (eip->ev_log == NULL)
574 				continue; /* event was never logged */
575 
576 			maj = major(eip->ev_log->log_stat.st_dev);
577 			min = minor(eip->ev_log->log_stat.st_dev);
578 
579 			(void) ea_set_group(ip, CAT_FMA_GROUP);
580 			egrp = ip++; /* first obj is group */
581 
582 			/*
583 			 * If the event log file is in legacy format,
584 			 * then write the xref to the file in the legacy
585 			 * maj/min/inode method else write it using the
586 			 * file uuid.
587 			 */
588 			if (eip->ev_log->log_uuid == NULL) {
589 				(void) ea_set_item(ip, CAT_FMA_MAJOR, &maj, 0);
590 				(void) ea_attach_to_group(egrp, ip++);
591 				(void) ea_set_item(ip, CAT_FMA_MINOR, &min, 0);
592 				(void) ea_attach_to_group(egrp, ip++);
593 				(void) ea_set_item(ip, CAT_FMA_INODE,
594 				    &eip->ev_log->log_stat.st_ino, 0);
595 				(void) ea_attach_to_group(egrp, ip++);
596 			} else {
597 				if (ea_set_item(ip, CAT_FMA_UUID,
598 				    eip->ev_log->log_uuid, 0) == -1) {
599 					err = EFMD_LOG_EXACCT;
600 					goto exerrcp;
601 				}
602 				if (fe == NULL)
603 					fe = fp = fmd_zalloc(frsize, FMD_SLEEP);
604 				*fp++ = ip;
605 				(void) ea_attach_to_group(egrp, ip++);
606 			}
607 			(void) ea_set_item(ip, CAT_FMA_OFFSET, &eip->ev_off, 0);
608 			(void) ea_attach_to_group(egrp, ip++);
609 			(void) ea_attach_to_group(&grp1, egrp);
610 		}
611 		(void) ea_attach_to_group(&grp0, &grp1);
612 	}
613 
614 	easize = ea_pack_object(&grp0, NULL, 0);
615 	eabuf = fmd_alloc(easize, FMD_SLEEP);
616 	(void) ea_pack_object(&grp0, eabuf, easize);
617 
618 	/*
619 	 * Before writing the record, check to see if this would cause the free
620 	 * space in the filesystem to drop below our minfree threshold.  If so,
621 	 * don't bother attempting the write and instead pretend it failed.  As
622 	 * fmd(1M) runs as root, it will be able to access the space "reserved"
623 	 * for root, and therefore can run the system of out of disk space in a
624 	 * heavy error load situation, violating the basic design principle of
625 	 * fmd(1M) that we don't want to make a bad situation even worse.
626 	 */
627 	(void) pthread_mutex_lock(&lp->log_lock);
628 
629 	if (lp->log_minfree != 0 && fstatvfs64(lp->log_fd, &stv) == 0 &&
630 	    stv.f_bavail * stv.f_frsize < lp->log_minfree + easize) {
631 
632 		TRACE((FMD_DBG_LOG, "append %s crosses minfree", lp->log_tag));
633 		err = EFMD_LOG_MINFREE;
634 
635 	} else if (fmd_log_write(lp, eabuf, easize) == easize) {
636 		TRACE((FMD_DBG_LOG, "append %s %p off=0x%llx",
637 		    lp->log_tag, (void *)ep, (u_longlong_t)lp->log_off));
638 
639 		ep->ev_flags &= ~FMD_EVF_VOLATILE;
640 		ep->ev_log = lp;
641 		ep->ev_off = lp->log_off;
642 		ep->ev_len = easize;
643 
644 		if (lp->log_flags & FMD_LF_REPLAY) {
645 			lp->log_pending++;
646 			ASSERT(lp->log_pending != 0);
647 		}
648 
649 		lp->log_refs++;
650 		ASSERT(lp->log_refs != 0);
651 		lp->log_off += easize;
652 	} else {
653 		err = errno; /* save errno for fmd_error() call below */
654 
655 		/*
656 		 * If we can't write append the record, seek the file back to
657 		 * the original location and truncate it there in order to make
658 		 * sure the file is always in a sane state w.r.t. libexacct.
659 		 */
660 		(void) lseek64(lp->log_fd, lp->log_off, SEEK_SET);
661 		(void) ftruncate64(lp->log_fd, lp->log_off);
662 	}
663 
664 	(void) pthread_mutex_unlock(&lp->log_lock);
665 	(void) pthread_mutex_unlock(&ep->ev_lock);
666 
667 	fmd_free(eabuf, easize);
668 
669 exerrcp:
670 	if (cp != NULL) {
671 		if (fe != NULL) {
672 			ea_object_t **fp = fe;
673 			int i = 0;
674 
675 			for (; *fp != NULL && i < cip->ci_nitems; i++)
676 				(void) ea_free_item(*fp++, EUP_ALLOC);
677 			fmd_free(fe, frsize);
678 		}
679 
680 		fmd_free(items, itsize);
681 	}
682 
683 exerr:
684 	fmd_free(nvbuf, nvsize);
685 
686 	(void) ea_free_item(&i0, EUP_ALLOC);
687 	(void) ea_free_item(&i1, EUP_ALLOC);
688 	(void) ea_free_item(&i2, EUP_ALLOC);
689 
690 	/*
691 	 * Keep track of out-of-space errors using global statistics.  As we're
692 	 * out of disk space, it's unlikely the EFMD_LOG_APPEND will be logged.
693 	 */
694 	if (err == ENOSPC || err == EFMD_LOG_MINFREE) {
695 		fmd_stat_t *sp;
696 
697 		if (lp == fmd.d_errlog)
698 			sp = &fmd.d_stats->ds_err_enospc;
699 		else if (lp == fmd.d_fltlog)
700 			sp = &fmd.d_stats->ds_flt_enospc;
701 		else
702 			sp = &fmd.d_stats->ds_oth_enospc;
703 
704 		(void) pthread_mutex_lock(&fmd.d_stats_lock);
705 		sp->fmds_value.ui64++;
706 		(void) pthread_mutex_unlock(&fmd.d_stats_lock);
707 	}
708 
709 	if (err != 0) {
710 		fmd_error(EFMD_LOG_APPEND, "failed to log_append %s %p: %s\n",
711 		    lp->log_tag, (void *)ep, fmd_strerror(err));
712 	}
713 }
714 
715 /*
716  * Commit an event to the log permanently, indicating that it should not be
717  * replayed on restart.  This is done by overwriting the event group's catalog
718  * code with EXD_GROUP_FMA (from EXD_GROUP_RFMA used in fmd_log_append()).  We
719  * use pwrite64() to update the existing word directly, using somewhat guilty
720  * knowledge that exacct stores the 32-bit catalog word first for each object.
721  * Since we are overwriting an existing log location using pwrite64() and hold
722  * the event lock, we do not need to hold the log_lock during the i/o.
723  */
724 void
725 fmd_log_commit(fmd_log_t *lp, fmd_event_t *e)
726 {
727 	fmd_event_impl_t *ep = (fmd_event_impl_t *)e;
728 	ea_catalog_t c;
729 	int err = 0;
730 
731 	if (!(lp->log_flags & FMD_LF_REPLAY))
732 		return; /* log does not require replay tagging */
733 
734 	ASSERT(MUTEX_HELD(&ep->ev_lock));
735 	ASSERT(ep->ev_log == lp && ep->ev_off != 0);
736 
737 	c = CAT_FMA_GROUP;
738 	exacct_order32(&c);
739 
740 	if (pwrite64(lp->log_fd, &c, sizeof (c), ep->ev_off) == sizeof (c)) {
741 		TRACE((FMD_DBG_LOG, "commit %s %p", lp->log_tag, (void *)ep));
742 		ep->ev_flags &= ~FMD_EVF_REPLAY;
743 
744 		/*
745 		 * If we have committed the event, check to see if the TOC skip
746 		 * offset needs to be updated, and decrement the pending count.
747 		 */
748 		(void) pthread_mutex_lock(&lp->log_lock);
749 
750 		if (lp->log_skip == ep->ev_off) {
751 			lp->log_flags |= FMD_LF_DIRTY;
752 			lp->log_skip += ep->ev_len;
753 		}
754 
755 		ASSERT(lp->log_pending != 0);
756 		lp->log_pending--;
757 
758 		(void) pthread_cond_broadcast(&lp->log_cv);
759 		(void) pthread_mutex_unlock(&lp->log_lock);
760 
761 	} else {
762 		fmd_error(EFMD_LOG_COMMIT, "failed to log_commit %s %p: %s\n",
763 		    lp->log_tag, (void *)ep, fmd_strerror(err));
764 	}
765 }
766 
767 /*
768  * If we need to destroy an event and it wasn't able to be committed, we permit
769  * the owner to decommit from ever trying again.  This operation decrements the
770  * pending count on the log and broadcasts to anyone waiting on log_cv.
771  */
772 void
773 fmd_log_decommit(fmd_log_t *lp, fmd_event_t *e)
774 {
775 	fmd_event_impl_t *ep = (fmd_event_impl_t *)e;
776 
777 	if (!(lp->log_flags & FMD_LF_REPLAY))
778 		return; /* log does not require replay tagging */
779 
780 	ASSERT(MUTEX_HELD(&ep->ev_lock));
781 	ASSERT(ep->ev_log == lp);
782 
783 	(void) pthread_mutex_lock(&lp->log_lock);
784 
785 	TRACE((FMD_DBG_LOG, "decommit %s %p", lp->log_tag, (void *)ep));
786 	ep->ev_flags &= ~FMD_EVF_REPLAY;
787 
788 	ASSERT(lp->log_pending != 0);
789 	lp->log_pending--;
790 
791 	(void) pthread_cond_broadcast(&lp->log_cv);
792 	(void) pthread_mutex_unlock(&lp->log_lock);
793 }
794 
795 static fmd_event_t *
796 fmd_log_unpack(fmd_log_t *lp, ea_object_t *grp, off64_t off)
797 {
798 	fmd_timeval_t ftv = { -1ULL, -1ULL };
799 	nvlist_t *nvl = NULL;
800 
801 	ea_object_t *obj;
802 	char *class;
803 	int err;
804 
805 	for (obj = grp->eo_group.eg_objs; obj != NULL; obj = obj->eo_next) {
806 		switch (obj->eo_catalog) {
807 		case CAT_FMA_NVLIST:
808 			if ((err = nvlist_xunpack(obj->eo_item.ei_raw,
809 			    obj->eo_item.ei_size, &nvl, &fmd.d_nva)) != 0) {
810 				fmd_error(EFMD_LOG_UNPACK, "failed to unpack "
811 				    "log nvpair: %s\n", fmd_strerror(err));
812 				return (NULL);
813 			}
814 			break;
815 
816 		case CAT_FMA_TODSEC:
817 			ftv.ftv_sec = obj->eo_item.ei_uint64;
818 			break;
819 
820 		case CAT_FMA_TODNSEC:
821 			ftv.ftv_nsec = obj->eo_item.ei_uint64;
822 			break;
823 		}
824 	}
825 
826 	if (nvl == NULL || ftv.ftv_sec == -1ULL || ftv.ftv_nsec == -1ULL) {
827 		fmd_error(EFMD_LOG_UNPACK, "failed to unpack log event: "
828 		    "required object(s) missing from record group\n");
829 		nvlist_free(nvl);
830 		return (NULL);
831 	}
832 
833 	if (nvlist_lookup_string(nvl, FM_CLASS, &class) != 0) {
834 		fmd_error(EFMD_LOG_UNPACK, "failed to unpack log event: "
835 		    "record is missing required '%s' nvpair\n", FM_CLASS);
836 		nvlist_free(nvl);
837 		return (NULL);
838 	}
839 
840 	return (fmd_event_recreate(FMD_EVT_PROTOCOL,
841 	    &ftv, nvl, class, lp, off, ea_pack_object(grp, NULL, 0)));
842 }
843 
844 /*
845  * Replay event(s) from the specified log by invoking the specified callback
846  * function 'func' for each event.  If the log has the FMD_LF_REPLAY flag set,
847  * we replay all events after log_skip that have the FMA_RGROUP group tag.
848  * This mode is used for the error telemetry log.  If the log does not have
849  * this flag set (used for ASRU logs), only the most recent event is replayed.
850  */
851 void
852 fmd_log_replay(fmd_log_t *lp, fmd_log_f *func, void *data)
853 {
854 	ea_object_t obj, *grp;
855 	ea_object_type_t type;
856 	ea_catalog_t c;
857 	fmd_event_t *ep;
858 	off64_t off, skp;
859 	uint_t n = 0;
860 
861 	(void) pthread_mutex_lock(&lp->log_lock);
862 
863 	if (lp->log_stat.st_size == 0 && (lp->log_flags & FMD_LF_REPLAY)) {
864 		(void) pthread_mutex_unlock(&lp->log_lock);
865 		return; /* we just created this log: never replay events */
866 	}
867 
868 	while (lp->log_flags & FMD_LF_BUSY)
869 		(void) pthread_cond_wait(&lp->log_cv, &lp->log_lock);
870 
871 	if (lp->log_off == lp->log_beg) {
872 		(void) pthread_mutex_unlock(&lp->log_lock);
873 		return; /* no records appended yet */
874 	}
875 
876 	lp->log_flags |= FMD_LF_BUSY;
877 	skp = lp->log_skip;
878 	ea_clear(&lp->log_ea); /* resync exacct file */
879 
880 	/*
881 	 * If FMD_LF_REPLAY is set, begin our replay at either log_skip (if it
882 	 * is non-zero) or at log_beg.  Otherwise replay from the end (log_off)
883 	 */
884 	if (lp->log_flags & FMD_LF_REPLAY) {
885 		off = MAX(lp->log_beg, lp->log_skip);
886 		c = CAT_FMA_RGROUP;
887 	} else {
888 		off = lp->log_off;
889 		c = CAT_FMA_GROUP;
890 	}
891 
892 	if (lseek64(lp->log_fd, off, SEEK_SET) != off) {
893 		fmd_panic("failed to seek %s to 0x%llx\n",
894 		    lp->log_name, (u_longlong_t)off);
895 	}
896 
897 	/*
898 	 * If FMD_LF_REPLAY is not set, back up to the start of the previous
899 	 * object and make sure this object is an EO_GROUP; otherwise return.
900 	 */
901 	if (!(lp->log_flags & FMD_LF_REPLAY) &&
902 	    (type = ea_previous_object(&lp->log_ea, &obj)) != EO_GROUP) {
903 		fmd_error(EFMD_LOG_REPLAY, "last log object is of unexpected "
904 		    "type %d (log may be truncated or corrupt)\n", type);
905 		goto out;
906 	}
907 
908 	while ((grp = ea_get_object_tree(&lp->log_ea, 1)) != NULL) {
909 		if (!(lp->log_flags & FMD_LF_REPLAY))
910 			off -= ea_pack_object(grp, NULL, 0);
911 		else if (n == 0 && grp->eo_catalog == CAT_FMA_GROUP)
912 			skp = off; /* update skip */
913 
914 		/*
915 		 * We temporarily drop log_lock around the call to unpack the
916 		 * event, hold it, and perform the callback, because these
917 		 * operations may try to acquire log_lock to bump log_refs.
918 		 * We cannot lose control because the FMD_LF_BUSY flag is set.
919 		 */
920 		(void) pthread_mutex_unlock(&lp->log_lock);
921 
922 		if (grp->eo_catalog == c &&
923 		    (ep = fmd_log_unpack(lp, grp, off)) != NULL) {
924 
925 			TRACE((FMD_DBG_LOG, "replay %s %p off %llx",
926 			    lp->log_tag, (void *)ep, (u_longlong_t)off));
927 
928 			fmd_event_hold(ep);
929 			func(lp, ep, data);
930 			fmd_event_rele(ep);
931 			n++;
932 		}
933 
934 		(void) pthread_mutex_lock(&lp->log_lock);
935 		off += ea_pack_object(grp, NULL, 0);
936 		ea_free_object(grp, EUP_ALLOC);
937 	}
938 
939 	if (ea_error() != EXR_EOF) {
940 		fmd_error(EFMD_LOG_REPLAY, "failed to replay %s event at "
941 		    "offset 0x%llx: %s\n", lp->log_name, (u_longlong_t)off,
942 		    fmd_ea_strerror(ea_error()));
943 	}
944 
945 	if (n == 0)
946 		skp = off; /* if no replays, move skip to where we ended up */
947 
948 out:
949 	if (lseek64(lp->log_fd, lp->log_off, SEEK_SET) != lp->log_off) {
950 		fmd_panic("failed to seek %s to 0x%llx\n",
951 		    lp->log_name, (u_longlong_t)lp->log_off);
952 	}
953 
954 	if (skp != lp->log_skip) {
955 		lp->log_flags |= FMD_LF_DIRTY;
956 		lp->log_skip = skp;
957 	}
958 
959 	lp->log_flags &= ~FMD_LF_BUSY;
960 	(void) pthread_cond_broadcast(&lp->log_cv);
961 	(void) pthread_mutex_unlock(&lp->log_lock);
962 }
963 
964 void
965 fmd_log_update(fmd_log_t *lp)
966 {
967 	ea_object_t toc, item;
968 	off64_t skip = 0;
969 	size_t size;
970 	void *buf;
971 
972 	(void) pthread_mutex_lock(&lp->log_lock);
973 
974 	if (lp->log_flags & FMD_LF_DIRTY) {
975 		lp->log_flags &= ~FMD_LF_DIRTY;
976 		skip = lp->log_skip;
977 	}
978 
979 	(void) pthread_mutex_unlock(&lp->log_lock);
980 
981 	/*
982 	 * If the skip needs to be updated, construct a TOC record group
983 	 * containing the skip offset and overwrite the TOC in-place.
984 	 */
985 	if (skip != 0 && ea_set_group(&toc, CAT_FMA_GROUP) == 0 &&
986 	    ea_set_item(&item, CAT_FMA_OFFSET, &skip, 0) == 0) {
987 
988 		(void) ea_attach_to_group(&toc, &item);
989 		size = ea_pack_object(&toc, NULL, 0);
990 		buf = fmd_alloc(size, FMD_SLEEP);
991 
992 		(void) ea_pack_object(&toc, buf, size);
993 		ASSERT(lp->log_toc + size == lp->log_beg);
994 
995 		if (pwrite64(lp->log_fd, buf, size, lp->log_toc) == size) {
996 			TRACE((FMD_DBG_LOG, "updated skip to %llx", skip));
997 		} else {
998 			fmd_error(EFMD_LOG_UPDATE,
999 			    "failed to log_update %s", lp->log_tag);
1000 		}
1001 
1002 		fmd_free(buf, size);
1003 		(void) ea_free_item(&item, EUP_ALLOC);
1004 	}
1005 }
1006 
1007 /*
1008  * Rotate the specified log by renaming its underlying file to a staging file
1009  * that can be handed off to logadm(1M) or an administrator script.  If the
1010  * rename succeeds, open a new log file using the old path and return it.
1011  * Note that we are relying our caller to use some higher-level mechanism to
1012  * ensure that fmd_log_rotate() cannot be called while other threads are
1013  * attempting fmd_log_append() using the same log (fmd's d_log_lock is used
1014  * for the global errlog and fltlog).
1015  */
1016 fmd_log_t *
1017 fmd_log_rotate(fmd_log_t *lp)
1018 {
1019 	char npath[PATH_MAX];
1020 	fmd_log_t *nlp;
1021 
1022 	(void) snprintf(npath, sizeof (npath), "%s+", lp->log_name);
1023 
1024 	/*
1025 	 * Open new log file.
1026 	 */
1027 	if ((nlp = fmd_log_open("", npath, lp->log_tag)) == NULL) {
1028 		fmd_error(EFMD_LOG_ROTATE, "failed to open %s", npath);
1029 		(void) fmd_set_errno(EFMD_LOG_ROTATE);
1030 		return (NULL);
1031 	}
1032 
1033 	(void) snprintf(npath, sizeof (npath), "%s.0-", lp->log_name);
1034 	(void) pthread_mutex_lock(&lp->log_lock);
1035 
1036 	/*
1037 	 * Check for any pending commits to drain before proceeding.  We can't
1038 	 * rotate the log out if commits are pending because if we die after
1039 	 * the log is moved aside, we won't be able to replay them on restart.
1040 	 */
1041 	if (lp->log_pending != 0) {
1042 		(void) pthread_mutex_unlock(&lp->log_lock);
1043 		(void) unlink(nlp->log_name);
1044 		fmd_log_rele(nlp);
1045 		(void) fmd_set_errno(EFMD_LOG_ROTBUSY);
1046 		return (NULL);
1047 	}
1048 
1049 	if (rename(lp->log_name, npath) != 0) {
1050 		(void) pthread_mutex_unlock(&lp->log_lock);
1051 		fmd_error(EFMD_LOG_ROTATE, "failed to rename %s", lp->log_name);
1052 		(void) unlink(nlp->log_name);
1053 		fmd_log_rele(nlp);
1054 		(void) fmd_set_errno(EFMD_LOG_ROTATE);
1055 		return (NULL);
1056 	}
1057 
1058 	if (rename(nlp->log_name, lp->log_name) != 0) {
1059 		(void) pthread_mutex_unlock(&lp->log_lock);
1060 		fmd_error(EFMD_LOG_ROTATE, "failed to rename %s",
1061 		    nlp->log_name);
1062 		(void) unlink(nlp->log_name);
1063 		fmd_log_rele(nlp);
1064 		(void) fmd_set_errno(EFMD_LOG_ROTATE);
1065 		return (NULL);
1066 	}
1067 
1068 	/*
1069 	 * Change name of new log file
1070 	 */
1071 	fmd_strfree(nlp->log_name);
1072 	nlp->log_name = fmd_strdup(lp->log_name, FMD_SLEEP);
1073 
1074 	/*
1075 	 * If we've rotated the log, no pending events exist so we don't have
1076 	 * any more commits coming, and our caller should have arranged for
1077 	 * no more calls to append.  As such, we can close log_fd for good.
1078 	 */
1079 	if (lp->log_flags & FMD_LF_EAOPEN) {
1080 		(void) ea_close(&lp->log_ea);
1081 		lp->log_flags &= ~FMD_LF_EAOPEN;
1082 	}
1083 
1084 	(void) close(lp->log_fd);
1085 	lp->log_fd = -1;
1086 
1087 	(void) pthread_mutex_unlock(&lp->log_lock);
1088 	return (nlp);
1089 }
1090