1 files changed, 147 insertions, 0 deletions
diff --git a/src/ceph-crash.in b/src/ceph-crash.in
new file mode 100755
index 000000000..0e02837fa
--- /dev/null
+++ b/src/ceph-crash.in
@@ -0,0 +1,147 @@
+#!@Python3_EXECUTABLE@
+# -*- mode:python -*-
+# vim: ts=4 sw=4 smarttab expandtab
+
+import argparse
+import grp
+import logging
+import os
+import pwd
+import signal
+import socket
+import subprocess
+import sys
+import time
+
+logging.basicConfig(level=logging.INFO)
+log = logging.getLogger('ceph-crash')
+
+auth_names = ['client.crash.%s' % socket.gethostname(),
+              'client.crash',
+              'client.admin']
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        '-p', '--path', default='/var/lib/ceph/crash',
+        help='base path to monitor for crash dumps')
+    parser.add_argument(
+        '-d', '--delay', default=10.0, type=float,
+        help='minutes to delay between scans (0 to exit after one)',
+    )
+    parser.add_argument(
+        '--name', '-n',
+        help='ceph name to authenticate as '
+             '(default: try client.crash, client.admin)')
+    parser.add_argument(
+        '--log-level', '-l',
+        help='log level output (default: INFO), support INFO or DEBUG')
+
+    return parser.parse_args()
+
+
+def post_crash(path):
+    rc = 0
+    for n in auth_names:
+        pr = subprocess.Popen(
+            args=['timeout', '30', 'ceph',
+                  '-n', n,
+                  'crash', 'post', '-i', '-'],
+            stdin=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+        )
+        f = open(os.path.join(path, 'meta'), 'rb')
+        (_, stderr) = pr.communicate(input=f.read())
+        stderr = stderr.decode()
+        rc = pr.wait()
+        f.close()
+        if rc != 0 or stderr != "":
+            log.warning('post %s as %s failed: %s' % (path, n, stderr))
+        if rc == 0:
+            break
+    return rc
+
+
+def scrape_path(path):
+    for p in os.listdir(path):
+        crashpath = os.path.join(path, p)
+        if not os.access(crashpath, os.R_OK):
+            log.warning('unable to read crash path %s' % (crashpath))
+            continue
+        metapath = os.path.join(crashpath, 'meta')
+        donepath = os.path.join(crashpath, 'done')
+        if os.path.isfile(metapath):
+            if not os.path.isfile(donepath):
+                # hang out just for a bit; either we interrupted the dump
+                # or the daemon crashed before finishing it
+                time.sleep(1)
+                if not os.path.isfile(donepath):
+                    return
+            # ok, we can process this one
+            rc = post_crash(crashpath)
+            if rc == 0:
+                os.rename(crashpath, os.path.join(path, 'posted/', p))
+                log.debug(
+                    "posted %s and renamed %s -> %s " %
+                    (metapath, p, os.path.join('posted/', p))
+                )
+
+
+def handler(signum, frame):
+    print('*** Interrupted with signal %d ***' % signum)
+    sys.exit(0)
+
+
+def drop_privs():
+    if os.getuid() == 0:
+        try:
+            ceph_uid = pwd.getpwnam("ceph").pw_uid
+            ceph_gid = grp.getgrnam("ceph").gr_gid
+            os.setgroups([])
+            os.setgid(ceph_gid)
+            os.setuid(ceph_uid)
+        except Exception as e:
+            log.error(f"Unable to drop privileges: {e}")
+            sys.exit(1)
+
+
+def main():
+    global auth_names
+
+    # run as unprivileged ceph user
+    drop_privs()
+
+    # exit code 0 on SIGINT, SIGTERM
+    signal.signal(signal.SIGINT, handler)
+    signal.signal(signal.SIGTERM, handler)
+
+    args = parse_args()
+    if args.log_level == 'DEBUG':
+        log.setLevel(logging.DEBUG)
+
+    postdir = os.path.join(args.path, 'posted')
+    if args.name:
+        auth_names = [args.name]
+
+    while not os.path.isdir(postdir):
+        log.error("directory %s does not exist; please create" % postdir)
+        time.sleep(30)
+
+    log.info("pinging cluster to exercise our key")
+    pr = subprocess.Popen(args=['timeout', '30', 'ceph', '-s'])
+    pr.wait()
+
+    log.info("monitoring path %s, delay %ds" % (args.path, args.delay * 60.0))
+    while True:
+        try:
+            scrape_path(args.path)
+        except Exception as e:
+            log.error(f"Error scraping {args.path}: {e}")
+        if args.delay == 0:
+            sys.exit(0)
+        time.sleep(args.delay * 60)
+
+
+if __name__ == "__main__":
+    main()