Print this page
NEX-17845 Remove support for BZIP2 from dump
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
NEX-14185 savecore -f vmdump.1 tries to unpack this to unix.0 and vmcore.0 and other nits
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Reviewed by: Roman Strashkin <roman.strashkin@nexenta.com>
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
NEX-9338 improve the layout of the crash directory (follow-up)
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
NEX-9338 improve the layout of the crash directory
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
Reviewed by: Steve Peng <steve.peng@nexenta.com>
@@ -21,11 +21,11 @@
/*
* Copyright (c) 1983, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright 2016 Joyent, Inc.
*/
/*
- * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2018 Nexenta Systems, Inc. All rights reserved.
*/
#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>
@@ -49,14 +49,15 @@
#include <sys/compress.h>
#include <sys/panic.h>
#include <sys/sysmacros.h>
#include <sys/stat.h>
#include <sys/resource.h>
-#include <bzip2/bzlib.h>
#include <sys/fm/util.h>
#include <fm/libfmevent.h>
#include <sys/int_fmtio.h>
+#include <uuid/uuid.h>
+#include <libgen.h>
/* fread/fwrite buffer size */
#define FBUFSIZE (1ULL << 20)
@@ -66,14 +67,16 @@
/* create this file if metrics collection is enabled in the kernel */
#define METRICSFILE "METRICS.csv"
static char progname[9] = "savecore";
static char *savedir; /* savecore directory */
+static char uuiddir[MAXPATHLEN]; /* UUID directory */
static char *dumpfile; /* source of raw crash dump */
static long bounds = -1; /* numeric suffix */
static long pagesize; /* dump pagesize */
static int dumpfd = -1; /* dumpfile descriptor */
+static boolean_t skip_event = B_FALSE; /* do not raise an event */
static boolean_t have_dumpfile = B_TRUE; /* dumpfile existence */
static dumphdr_t corehdr, dumphdr; /* initial and terminal dumphdrs */
static boolean_t dump_incomplete; /* dumphdr indicates incomplete */
static boolean_t fm_panic; /* dump is the result of fm_panic */
static offset_t endoff; /* offset of end-of-dump header */
@@ -90,10 +93,11 @@
static volatile uint64_t zpages; /* count of zero pages not written */
static dumpdatahdr_t datahdr; /* compression info */
static long coreblksize; /* preferred write size (st_blksize) */
static int cflag; /* run as savecore -c */
static int mflag; /* run as savecore -m */
+static int fflag; /* -f option used */
/*
* Payload information for the events we raise. These are used
* in raise_event to determine what payload to include.
*/
@@ -238,11 +242,12 @@
code = 3;
break;
case SC_EXIT_ERR:
default:
- if (!mflag && logprint_raised++ == 0 && have_dumpfile)
+ if (!mflag && logprint_raised++ == 0 && !skip_event &&
+ have_dumpfile)
raise_event(SC_EVENT_SAVECORE_FAILURE, buf);
code = 1;
break;
}
@@ -299,10 +304,15 @@
static void
Stat(const char *fname, Stat_t *sb)
{
if (stat64(fname, sb) != 0) {
+ /*
+ * If dump/core file doesn't exist, then best
+ * to not go further (raise an event).
+ */
+ skip_event = B_TRUE;
have_dumpfile = B_FALSE;
logprint(SC_SL_ERR | SC_EXIT_ERR, "failed to get status "
"of file %s", fname);
}
}
@@ -376,10 +386,14 @@
if (dumphdr.dump_version != DUMP_VERSION)
logprint(SC_SL_NONE | SC_IF_VERBOSE | SC_EXIT_PEND,
"dump version (%d) != %s version (%d)",
dumphdr.dump_version, progname, DUMP_VERSION);
+ if (datahdr.dump_clevel > DUMP_CLEVEL_LZJB)
+ logprint(SC_SL_NONE | SC_EXIT_PEND,
+ "unsupported compression format (%d)", datahdr.dump_clevel);
+
if (dumphdr.dump_wordsize != DUMP_WORDSIZE)
logprint(SC_SL_NONE | SC_EXIT_PEND,
"dump is from %u-bit kernel - cannot save on %u-bit kernel",
dumphdr.dump_wordsize, DUMP_WORDSIZE);
@@ -418,24 +432,27 @@
static void
check_space(int csave)
{
struct statvfs fsb;
int64_t spacefree, dumpsize, minfree, datasize;
+ char minfreefile[MAXPATHLEN];
if (statvfs(".", &fsb) < 0)
logprint(SC_SL_ERR | SC_EXIT_ERR, "statvfs: %s",
strerror(errno));
+ (void) snprintf(minfreefile, MAXPATHLEN, "%s/minfree", savedir);
+
dumpsize = dumphdr.dump_data - dumphdr.dump_start;
datasize = dumphdr.dump_npages * pagesize;
if (!csave)
dumpsize += datasize;
else
dumpsize += datahdr.dump_data_csize;
spacefree = (int64_t)fsb.f_bavail * fsb.f_frsize;
- minfree = 1024LL * read_number_from_file("minfree", 1024);
+ minfree = 1024LL * read_number_from_file(minfreefile, 1024);
if (spacefree < minfree + dumpsize) {
logprint(SC_SL_ERR | SC_EXIT_ERR,
"not enough space in %s (%lld MB avail, %lld MB needed)",
savedir, spacefree >> 20, (minfree + dumpsize) >> 20);
}
@@ -573,11 +590,11 @@
char *inbuf = Zalloc(bufsz);
offset_t coreoff;
size_t nb;
logprint(SC_SL_ERR | SC_IF_VERBOSE,
- "Copying %s to %s/%s\n", dumpfile, savedir, corefile);
+ "Copying %s to %s/%s\n", dumpfile, uuiddir, corefile);
/*
* This dump file is still compressed
*/
corehdr.dump_flags |= DF_COMPRESSED | DF_VALID;
@@ -698,11 +715,10 @@
blockhdr_t blocks;
pgcnt_t pagenum;
pgcnt_t curpage;
pgcnt_t npages;
pgcnt_t done;
- bz_stream strm;
dumpcsize_t sc;
dumpstreamhdr_t sh;
} stream_t;
static stream_t *streams;
@@ -961,125 +977,10 @@
break;
}
}
}
-/* bzlib library reports errors with this callback */
-void
-bz_internal_error(int errcode)
-{
- logprint(SC_SL_ERR | SC_EXIT_ERR, "bz_internal_error: err %s\n",
- BZ2_bzErrorString(errcode));
-}
-
-/*
- * Return one object in the stream.
- *
- * An object (stream header or page) will likely span an input block
- * of compression data. Return non-zero when an entire object has been
- * retrieved from the stream.
- */
-static int
-bz2decompress(stream_t *s, void *buf, size_t size)
-{
- int rc;
-
- if (s->strm.avail_out == 0) {
- s->strm.next_out = buf;
- s->strm.avail_out = size;
- }
- while (s->strm.avail_in > 0) {
- rc = BZ2_bzDecompress(&s->strm);
- if (rc == BZ_STREAM_END) {
- rc = BZ2_bzDecompressReset(&s->strm);
- if (rc != BZ_OK)
- logprint(SC_SL_ERR | SC_EXIT_ERR,
- "BZ2_bzDecompressReset: %s",
- BZ2_bzErrorString(rc));
- continue;
- }
-
- if (s->strm.avail_out == 0)
- break;
- }
- return (s->strm.avail_out == 0);
-}
-
-/*
- * Process one bzip2 block.
- * The interface is documented here:
- * http://www.bzip.org/1.0.5/bzip2-manual-1.0.5.html
- */
-static void
-bz2block(int corefd, stream_t *s, char *block, size_t blocksz)
-{
- int rc = 0;
- int doflush;
- char *out;
-
- if (!s->init) {
- s->init = 1;
- rc = BZ2_bzDecompressInit(&s->strm, 0, 0);
- if (rc != BZ_OK)
- logprint(SC_SL_ERR | SC_EXIT_ERR,
- "BZ2_bzDecompressInit: %s", BZ2_bzErrorString(rc));
- if (s->blkbuf == NULL)
- s->blkbuf = Zalloc(coreblksize);
- s->strm.avail_out = 0;
- s->state = STREAMSTART;
- }
- s->strm.next_in = block;
- s->strm.avail_in = blocksz;
-
- while (s->strm.avail_in > 0) {
- switch (s->state) {
- case STREAMSTART:
- if (!bz2decompress(s, &s->sh, sizeof (s->sh)))
- return;
- if (strcmp(DUMP_STREAM_MAGIC, s->sh.stream_magic) != 0)
- logprint(SC_SL_ERR | SC_EXIT_ERR,
- "BZ2 STREAMSTART: bad stream header");
- if (s->sh.stream_npages > datahdr.dump_maxrange)
- logprint(SC_SL_ERR | SC_EXIT_ERR,
- "BZ2 STREAMSTART: bad range: %d > %d",
- s->sh.stream_npages, datahdr.dump_maxrange);
- s->pagenum = s->sh.stream_pagenum;
- s->npages = s->sh.stream_npages;
- s->curpage = s->pagenum;
- s->nout = 0;
- s->done = 0;
- s->state = STREAMPAGES;
- break;
- case STREAMPAGES:
- out = s->blkbuf + PTOB(s->nout);
- if (!bz2decompress(s, out, pagesize))
- return;
-
- atomic_inc_64(&saved);
-
- doflush = 0;
- if (s->nout == 0 && iszpage(out)) {
- doflush = 1;
- atomic_inc_64(&zpages);
- } else if (++s->nout >= BTOP(coreblksize) ||
- isblkbnd(s->curpage + s->nout)) {
- doflush = 1;
- }
- if (++s->done >= s->npages) {
- s->state = STREAMSTART;
- doflush = 1;
- }
- if (doflush) {
- putpage(corefd, s->blkbuf, s->curpage, s->nout);
- s->nout = 0;
- s->curpage = s->pagenum + s->done;
- }
- break;
- }
- }
-}
-
/* report progress */
static void
report_progress()
{
int sec, percent;
@@ -1118,16 +1019,12 @@
(void) pthread_cond_signal(&cvwork);
while (s->blocks.head != NULL) {
b = deqh(&s->blocks);
(void) pthread_mutex_unlock(&lock);
- if (datahdr.dump_clevel < DUMP_CLEVEL_BZIP2)
lzjbblock(t->corefd, s, b->block,
b->size);
- else
- bz2block(t->corefd, s, b->block,
- b->size);
(void) pthread_mutex_lock(&lock);
enqt(&freeblocks, b);
(void) pthread_cond_signal(&cvfree);
@@ -1162,20 +1059,18 @@
* savecore creates a thread per stream, but never more threads than
* the number of CPUs running savecore. This is because savecore can
* be processing a crash file from a remote machine, which may have
* more CPUs.
*
- * When the kernel uses parallel lzjb or parallel bzip2, we expect a
- * series of 128KB blocks of compression data. In this case, each
- * block has a "tag", in the range 1-4095. Each block is handed off to
- * to the threads running "runstreams". The dump format is either lzjb
- * or bzip2, never a mixture. These threads, in turn, process the
- * compression data for groups of pages. Groups of pages are delimited
- * by a "stream header", which indicates a starting pfn and number of
- * pages. When a stream block has been read, the condition variable
- * "cvwork" is signalled, which causes one of the avaiable threads to
- * wake up and process the stream.
+ * When the kernel uses parallel compression we expect a series of 128KB
+ * blocks of compression data. In this case, each block has a "tag" in
+ * the range 1-4095. Each block is handed off to the threads running
+ * "runstreams". These threads, in turn, process the compression data
+ * for groups of pages. Groups of pages are delimited by a "stream header",
+ * which indicates a starting pfn and number of pages. When a stream block
+ * has been read, the condition variable "cvwork" is signalled, which causes
+ * one of the available threads to wake up and process the stream.
*
* In the parallel case there will be streams blocks encoding all data
* pages. The stream of blocks is terminated by a zero size
* word. There can be a few lzjb pages tacked on the end, depending on
* the architecture. The sbarrier function ensures that all stream
@@ -1343,11 +1238,11 @@
size_t ksyms_dsize;
Stat_t st;
int corefd = Open(corefile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
int namefd = Open(namelist, O_WRONLY | O_CREAT | O_TRUNC, 0644);
- (void) printf("Constructing namelist %s/%s\n", savedir, namelist);
+ (void) printf("Constructing namelist %s/%s\n", uuiddir, namelist);
/*
* Determine the optimum write size for the core file
*/
Fstat(corefd, &st, corefile);
@@ -1384,11 +1279,11 @@
Pwrite(namefd, ksyms_base, ksyms_size, 0);
(void) close(namefd);
free(ksyms_cbase);
free(ksyms_base);
- (void) printf("Constructing corefile %s/%s\n", savedir, corefile);
+ (void) printf("Constructing corefile %s/%s\n", uuiddir, corefile);
/*
* Read in and write out the pfn table.
*/
pfn_table = Zalloc(pfn_table_size);
@@ -1557,11 +1452,11 @@
raise_event(enum sc_event_type evidx, char *warn_string)
{
uint32_t pl = sc_event[evidx].sce_payload;
char panic_stack[STACK_BUF_SIZE];
nvlist_t *attr = NULL;
- char uuidbuf[36 + 1];
+ char uuidbuf[UUID_PRINTABLE_STRING_LENGTH];
int err = 0;
if (nvlist_alloc(&attr, NV_UNIQUE_NAME, 0) != 0)
goto publish; /* try to send payload-free event */
@@ -1649,13 +1544,20 @@
main(int argc, char *argv[])
{
int i, c, bfd;
Stat_t st;
struct rlimit rl;
+ struct stat sc, sd;
long filebounds = -1;
char namelist[30], corefile[30], boundstr[30];
dumpfile = NULL;
+ char uuidstr[UUID_PRINTABLE_STRING_LENGTH];
+ uuid_t uu;
+ static char boundsfile[MAXPATHLEN];
+ static char boundslink[MAXPATHLEN];
+ static boolean_t fma_layout = B_TRUE;
+ char *slash;
startts = gethrtime();
(void) getrlimit(RLIMIT_NOFILE, &rl);
rl.rlim_cur = rl.rlim_max;
@@ -1686,10 +1588,11 @@
mflag++;
break;
case 'f':
dumpfile = optarg;
filebounds = getbounds(dumpfile);
+ fflag++;
break;
case '?':
usage();
}
}
@@ -1713,28 +1616,51 @@
dumpfd = Open("/dev/dump", O_RDONLY, 0444);
if (dumpfile == NULL) {
dumpfile = Zalloc(MAXPATHLEN);
if (ioctl(dumpfd, DIOCGETDEV, dumpfile) == -1) {
+ skip_event = B_TRUE;
have_dumpfile = B_FALSE;
logprint(SC_SL_NONE | SC_IF_ISATTY | SC_EXIT_ERR,
"no dump device configured");
}
}
if (mflag)
return (message_save());
- if (optind == argc - 1)
+ if (optind == argc - 1) {
+ /*
+ * Use the default layout if directory was specified.
+ * If the directory path matches value configured (by dumpadm),
+ * then revert to fma layout.
+ */
+ fma_layout = B_FALSE;
+ if (savedir != NULL && (stat(savedir, &sc) >= 0 &&
+ stat(argv[optind], &sd) >= 0)) {
+ if (sc.st_ino == sd.st_ino &&
+ sc.st_dev == sd.st_dev) {
+ fma_layout = B_TRUE;
+ }
+ }
savedir = argv[optind];
+ }
if (savedir == NULL || optind < argc - 1)
usage();
- if (livedump && ioctl(dumpfd, DIOCDUMP, NULL) == -1)
+ if (livedump) {
+ /*
+ * For livedump we must update the dump header with
+ * newly generated uuid.
+ */
+ uuid_generate(uu);
+ uuid_unparse(uu, uuidstr);
+ if (ioctl(dumpfd, DIOCDUMP, uuidstr) == -1)
logprint(SC_SL_NONE | SC_EXIT_ERR,
"dedicated dump device required");
+ }
(void) close(dumpfd);
dumpfd = -1;
Stat(dumpfile, &st);
@@ -1789,10 +1715,23 @@
}
if (dumphdr.dump_fm_panic)
fm_panic = B_TRUE;
+ /* remove last slash */
+ slash = strrchr(savedir, '\0');
+ while (--slash > savedir && *slash == '/') {
+ *slash = '\0';
+ }
+
+ if (fma_layout) {
+ (void) snprintf(uuiddir, sizeof (uuiddir), "%s/data/%s",
+ savedir, dumphdr.dump_uuid);
+ } else {
+ (void) strncpy(uuiddir, savedir, sizeof (uuiddir));
+ }
+
/*
* We have a valid dump on a dump device and know as much about
* it as we're going to at this stage. Raise an event for
* logging and so that FMA can open a case for this panic.
* Avoid this step for FMA-initiated panics - FMA will replay
@@ -1826,31 +1765,81 @@
corehdr.dump_uuid,
fm_panic ? "(fault-management initiated)" : "");
/*NOTREACHED*/
}
- if (chdir(savedir) == -1)
+ if (fma_layout && mkdirp(uuiddir, 0755) != 0) {
+ if (errno != EEXIST)
+ logprint(SC_SL_ERR | SC_EXIT_ERR,
+ "mkdirp(\"%s\"): %s",
+ uuiddir, strerror(errno));
+ }
+
+ if (chdir(uuiddir) == -1)
logprint(SC_SL_ERR | SC_EXIT_ERR, "chdir(\"%s\"): %s",
- savedir, strerror(errno));
+ uuiddir, strerror(errno));
check_space(csave);
+ (void) snprintf(boundsfile, MAXPATHLEN, "%s/bounds", savedir);
+
if (filebounds < 0)
- bounds = read_number_from_file("bounds", 0);
+ bounds = read_number_from_file(boundsfile, 0);
else
bounds = filebounds;
+ if (!fflag && disregard_valid_flag && bounds > 0)
+ bounds--;
+
+ (void) snprintf(boundslink, MAXPATHLEN, "%s/%d", savedir, bounds);
+
+ /*
+ * Create a symbolic link to easily maintain the sequential ordering.
+ */
+ if (!fflag && fma_layout && symlink(uuiddir, boundslink) != 0) {
+ if (errno == EEXIST) {
+ char symbuf[MAXPATHLEN] = {'\0'};
+
+ if (readlink(boundslink, symbuf, sizeof (symbuf)) < 0)
+ logprint(SC_SL_ERR | SC_EXIT_ERR,
+ "readlink: %s", strerror(errno));
+ if (strcmp(symbuf, uuiddir) != 0) {
+ logprint(SC_SL_ERR,
+ "Symbolic link %s already exists but "
+ "specifies a wrong UUID directory, "
+ "new symbolic link will be created "
+ "instead", boundslink);
+ (void) unlink(boundslink);
+ if (symlink(uuiddir, boundslink) != 0)
+ logprint(SC_SL_ERR | SC_EXIT_ERR,
+ "symlink: %s", strerror(errno));
+ }
+ } else {
+ logprint(SC_SL_ERR | SC_EXIT_ERR, "symlink: %s",
+ strerror(errno));
+ }
+ }
+
if (csave) {
size_t metrics_size = datahdr.dump_metrics;
(void) sprintf(corefile, "vmdump.%ld", bounds);
+ if (interactive && bounds >= 0 && access(corefile, F_OK)
+ == 0) {
+ skip_event = B_TRUE;
+ logprint(SC_SL_NONE | SC_EXIT_ERR,
+ "%s already exists: remove with "
+ "'rm -f %s/{unix,vmcore}.%ld'",
+ corefile, uuiddir, bounds);
+ }
+
datahdr.dump_metrics = 0;
logprint(SC_SL_ERR,
"Saving compressed system crash dump in %s/%s",
- savedir, corefile);
+ uuiddir, corefile);
copy_crashfile(corefile);
/*
* Raise a fault management event that indicates the system
@@ -1905,26 +1894,28 @@
}
logprint(SC_SL_ERR,
"Decompress the crash dump with "
"\n'savecore -vf %s/%s'",
- savedir, corefile);
+ uuiddir, corefile);
} else {
(void) sprintf(namelist, "unix.%ld", bounds);
(void) sprintf(corefile, "vmcore.%ld", bounds);
- if (interactive && filebounds >= 0 && access(corefile, F_OK)
- == 0)
+ if (interactive && bounds >= 0 && access(corefile, F_OK)
+ == 0) {
+ skip_event = B_TRUE;
logprint(SC_SL_NONE | SC_EXIT_ERR,
"%s already exists: remove with "
"'rm -f %s/{unix,vmcore}.%ld'",
- corefile, savedir, bounds);
+ corefile, uuiddir, bounds);
+ }
logprint(SC_SL_ERR,
"saving system crash dump in %s/{unix,vmcore}.%ld",
- savedir, bounds);
+ uuiddir, bounds);
build_corefile(namelist, corefile);
if (!livedump && !filemode && !fm_panic)
raise_event(SC_EVENT_DUMP_AVAILABLE, NULL);
@@ -1965,11 +1956,11 @@
}
}
if (filebounds < 0) {
(void) sprintf(boundstr, "%ld\n", bounds + 1);
- bfd = Open("bounds", O_WRONLY | O_CREAT | O_TRUNC, 0644);
+ bfd = Open(boundsfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
Pwrite(bfd, boundstr, strlen(boundstr), 0);
(void) close(bfd);
}
if (verbose) {