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
110static ssize_t
111fmd_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
132static int
133fmd_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
207static int
208fmd_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
221static int
222fmd_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
332static int
333fmd_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
351static fmd_log_t *
352fmd_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
379top:
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
437fmd_log_t *
438fmd_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
443fmd_log_t *
444fmd_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
449void
450fmd_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
473void
474fmd_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
489void
490fmd_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
498void
499fmd_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
510void
511fmd_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
669exerrcp:
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
683exerr:
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 */
724void
725fmd_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 */
772void
773fmd_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
795static fmd_event_t *
796fmd_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 */
851void
852fmd_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
948out:
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
964void
965fmd_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 */
1016fmd_log_t *
1017fmd_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