3 # Copyright (c) 2003 Intel Corporation
6 # Copyright (c) 2004-2006 The Trustees of Princeton University
17 import notify_messages
20 from Exceptions import BootManagerException
22 import UpdateNodeConfiguration
23 import StopRunlevelAgent
28 Load the kernel off of a node and boot to it.
29 This step assumes the disks are mounted on SYSIMG_PATH.
30 If successful, this function will not return. If it returns, no chain
33 Expect the following variables:
34 SYSIMG_PATH the path where the system image will be mounted
35 (always starts with TEMP_PATH)
36 ROOT_MOUNTED the node root file system is mounted
37 NODE_SESSION the unique session val set when we requested
38 the current boot state
39 PLCONF_DIR The directory to store PL configuration files in
41 Sets the following variables:
42 ROOT_MOUNTED the node root file system is mounted
45 log.write("\n\nStep: Chain booting node.\n")
47 # make sure we have the variables we need
49 SYSIMG_PATH = vars["SYSIMG_PATH"]
51 raise ValueError("SYSIMG_PATH")
53 PLCONF_DIR = vars["PLCONF_DIR"]
55 raise ValueError("PLCONF_DIR")
57 # its ok if this is blank
58 NODE_SESSION = vars["NODE_SESSION"]
60 NODE_MODEL_OPTIONS = vars["NODE_MODEL_OPTIONS"]
62 PARTITIONS = vars["PARTITIONS"]
63 if PARTITIONS == None:
64 raise ValueError("PARTITIONS")
66 except KeyError as var:
67 raise BootManagerException("Missing variable in vars: {}\n".format(var))
68 except ValueError as var:
69 raise BootManagerException("Variable in vars, shouldn't be: {}\n".format(var))
72 if vars.has_key('ROOT_MOUNTED'):
73 ROOT_MOUNTED = vars['ROOT_MOUNTED']
76 log.write("Mounting node partitions\n")
78 # simply creating an instance of this class and listing the system
79 # block devices will make them show up so vgscan can find the planetlab
81 systeminfo.get_block_devices_dict(vars, log)
83 utils.sysexec("vgscan", log)
84 utils.sysexec("vgchange -ay planetlab", log)
86 utils.makedirs(SYSIMG_PATH)
88 cmd = "mount {} {}".format(PARTITIONS["root"], SYSIMG_PATH)
89 utils.sysexec(cmd, log)
90 cmd = "mount -t proc none {}/proc".format(SYSIMG_PATH)
91 utils.sysexec(cmd, log)
92 cmd = "mount {} {}/vservers".format(PARTITIONS["vservers"], SYSIMG_PATH)
93 utils.sysexec(cmd, log)
96 vars['ROOT_MOUNTED'] = 1
98 utils.display_disks_status(PARTITIONS, "In ChainBootNode", log)
100 # write out the session value /etc/planetlab/session
102 session_file_path = "{}/{}/session".format(SYSIMG_PATH, PLCONF_DIR)
103 session_file = file(session_file_path, "w")
104 session_file.write(str(NODE_SESSION))
107 log.write("Updated /etc/planetlab/session\n")
109 log.write("Unable to write out /etc/planetlab/session, continuing anyway\n")
111 # update configuration files
112 log.write("Updating configuration files.\n")
113 # avoid using conf_files initscript as we're moving to systemd on some platforms
115 if (vars['ONE_PARTITION'] != '1'):
117 cmd = "/usr/bin/env python /usr/share/NodeManager/conf_files.py --noscripts"
118 utils.sysexec_chroot(SYSIMG_PATH, cmd, log)
120 log.write("conf_files failed with \n {}".format(e))
122 # update node packages
123 log.write("Running node update.\n")
124 if os.path.exists(SYSIMG_PATH + "/usr/bin/NodeUpdate.py"):
125 cmd = "/usr/bin/NodeUpdate.py start noreboot"
127 # for backwards compatibility
128 cmd = "/usr/local/planetlab/bin/NodeUpdate.py start noreboot"
129 utils.sysexec_chroot(SYSIMG_PATH, cmd, log)
131 # Re-generate initrd right before kexec call
132 # this is not required anymore on recent depls.
133 if vars['virt'] == 'vs':
134 MakeInitrd.Run(vars, log)
136 # the following step should be done by NM
137 UpdateNodeConfiguration.Run(vars, log)
139 log.write("Updating ssh public host key with PLC.\n")
142 ssh_host_key_file = file("{}/etc/ssh/ssh_host_rsa_key.pub".format(SYSIMG_PATH), "r")
143 ssh_host_key = ssh_host_key_file.read().strip()
144 ssh_host_key_file.close()
145 ssh_host_key_file = None
150 update_vals['ssh_rsa_key'] = ssh_host_key
151 BootAPI.call_api_function(vars, "BootUpdateNode", (update_vals,))
154 # get the kernel version
156 if NODE_MODEL_OPTIONS & ModelOptions.SMP:
159 log.write("Copying kernel and initrd for booting.\n")
160 if vars['virt'] == 'vs':
161 utils.sysexec("cp {}/boot/kernel-boot{} /tmp/kernel".format(SYSIMG_PATH, option), log)
162 utils.sysexec("cp {}/boot/initrd-boot{} /tmp/initrd".format(SYSIMG_PATH, option), log)
164 # Use chroot to call rpm, b/c the bootimage&nodeimage rpm-versions may not work together
166 kversion = os.popen("chroot {} rpm -qa kernel | tail -1 | cut -c 8-"\
167 .format(SYSIMG_PATH)).read().rstrip()
168 major_version = int(kversion[0]) # Check if the string looks like a kernel version
170 # Try a different method for non-rpm-based distributions
171 kversion = os.popen("ls -lrt {}/lib/modules | tail -1 | awk '{{print $9;}}'"\
172 .format(SYSIMG_PATH)).read().rstrip()
174 # kernel and initrd come in various locations depending on the distro
176 kernel_candidates = []
177 kernel_candidates.append("/boot/vmlinux-{}*".format(kversion))
178 # on f23 starting dec. 2015 - what a twisted naming scheme
179 kernel_candidates.append("/boot/*/{}*/initrd".format(kversion))
181 initrd_candidates = []
182 # f16/18: expect initramfs image here
183 initrd_candidates.append ("/boot/initramfs-{}.img".format(kversion))
184 # f20: uses a uid of some kind, e.g. /boot/543f88c129de443baaa65800cf3927ce/<kversion>/initrd
185 initrd_candidates.append ("/boot/*/{}/initrd".format(kversion))
187 initrd_candidates.append ("/boot/initrd.img-{}".format(kversion))
189 def find_file_in_sysimg (candidates):
191 for pattern in candidates:
192 matches = glob.glob(SYSIMG_PATH+pattern)
193 log.write("locating initrd: found {} matches in {}\n".format(len(matches), pattern))
197 kernel = find_file_in_sysimg(kernel_candidates)
199 utils.sysexec("cp {} /tmp/kernel".format(kernel), log)
201 raise Exception("Unable to locate kernel - bailing out")
204 initrd = find_file_in_sysimg(initrd_candidates)
206 utils.sysexec("cp {} /tmp/initrd".format(initrd), log)
208 raise Exception("Unable to locate initrd - bailing out")
212 log.write("Unmounting disks.\n")
214 if (vars['ONE_PARTITION'] != '1'):
215 utils.sysexec("umount {}/vservers".format(SYSIMG_PATH), log)
216 utils.sysexec("umount {}/proc".format(SYSIMG_PATH), log)
217 utils.sysexec_noerr("umount {}/dev".format(SYSIMG_PATH), log)
218 utils.sysexec_noerr("umount {}/sys".format(SYSIMG_PATH), log)
219 utils.sysexec("umount {}".format(SYSIMG_PATH), log)
220 utils.sysexec("vgchange -an", log)
223 vars['ROOT_MOUNTED'] = 0
225 # Change runlevel to 'boot' prior to kexec.
226 StopRunlevelAgent.Run(vars, log)
228 log.write("Unloading modules and chain booting to new kernel.\n")
230 # further use of log after Upload will only output to screen
231 log.Upload("/root/.bash_eternal_history")
233 # regardless of whether kexec works or not, we need to stop trying to
235 cancel_boot_flag = "/tmp/CANCEL_BOOT"
236 utils.sysexec("touch {}".format(cancel_boot_flag), log)
238 # on 2.x cds (2.4 kernel) for sure, we need to shutdown everything
239 # to get kexec to work correctly. Even on 3.x cds (2.6 kernel),
240 # there are a few buggy drivers that don't disable their hardware
241 # correctly unless they are first unloaded.
243 utils.sysexec_noerr("ifconfig eth0 down", log)
245 utils.sysexec_noerr("killall dhclient", log)
247 if vars['virt'] == 'vs':
248 utils.sysexec_noerr("umount -a -r -t ext2,ext3", log)
250 utils.sysexec_noerr("umount -a -r -t ext2,ext3,btrfs", log)
251 utils.sysexec_noerr("modprobe -r lvm-mod", log)
253 # modules that should not get unloaded
254 # unloading cpqphp causes a kernel panic
255 blacklist = [ "floppy", "cpqphp", "i82875p_edac", "mptspi"]
257 modules = file("/tmp/loadedmodules","r")
260 module = string.strip(line)
261 if module in blacklist :
262 log.write("Skipping unload of kernel module '{}'.\n".format(module))
264 log.write("Unloading {}\n".format(module))
265 utils.sysexec_noerr("modprobe -r {}".format(module), log)
266 if "e1000" in module:
267 log.write("Unloading e1000 driver; sleeping 4 seconds...\n")
272 log.write("Couldn't read /tmp/loadedmodules, continuing.\n")
275 modules = file("/proc/modules", "r")
277 # Get usage count for USB
281 # Module Size UsageCount UsedBy State LoadAddress
282 parts = string.split(line)
284 if parts[0] == "usb_storage":
285 usb_usage += int(parts[2])
286 except IndexError as e:
287 log.write("Couldn't parse /proc/modules, continuing.\n")
293 # Module Size UsageCount UsedBy State LoadAddress
294 parts = string.split(line)
296 # While we would like to remove all "unused" modules,
297 # you can't trust usage count, especially for things
298 # like network drivers or RAID array drivers. Just try
299 # and unload a few specific modules that we know cause
300 # problems during chain boot, such as USB host
301 # controller drivers (HCDs) (PL6577).
302 # if int(parts[2]) == 0:
303 if False and re.search('_hcd$', parts[0]):
305 log.write("NOT unloading {} since USB may be in use\n".format(parts[0]))
307 log.write("Unloading {}\n".format(parts[0]))
308 utils.sysexec_noerr("modprobe -r {}".format(parts[0]), log)
309 except IndexError as e:
310 log.write("Couldn't parse /proc/modules, continuing.\n")
312 log.write("Couldn't read /proc/modules, continuing.\n")
315 kargs = "root={} ramdisk_size=8192".format(PARTITIONS["root"])
316 if NODE_MODEL_OPTIONS & ModelOptions.SMP:
317 kargs = kargs + " " + "acpi=off"
319 kargsfb = open("/kargs.txt","r")
320 moreargs = kargsfb.readline()
322 moreargs = moreargs.strip()
323 log.write('Parsed in "{}" kexec args from /kargs.txt\n'.format(moreargs))
324 kargs = kargs + " " + moreargs
326 # /kargs.txt does not exist, which is fine. Just kexec with default
327 # kargs, which is ramdisk_size=8192
330 utils.sysexec_noerr('hwclock --systohc --utc ', log)
331 # utils.breakpoint("Before kexec");
333 utils.sysexec('kexec --force --initrd=/tmp/initrd --append="{}" /tmp/kernel'.format(kargs), log)
334 except BootManagerException as e:
335 # if kexec fails, we've shut the machine down to a point where nothing
336 # can run usefully anymore (network down, all modules unloaded, file
337 # systems unmounted. write out the error, and cancel the boot process
340 log.write("-------------------------------------------------------\n")
341 log.write("kexec failed with the following error. Please report\n")
342 log.write("this problem to support@planet-lab.org.\n\n")
343 log.write(str(e) + "\n\n")
344 log.write("The boot process has been canceled.\n")
345 log.write("-------------------------------------------------------\n\n")