3 # Copyright (c) 2003 Intel Corporation
6 # Copyright (c) 2004-2006 The Trustees of Princeton University
14 import UpdateNodeConfiguration
15 from Exceptions import *
19 import notify_messages
26 Load the kernel off of a node and boot to it.
27 This step assumes the disks are mounted on SYSIMG_PATH.
28 If successful, this function will not return. If it returns, no chain
31 Expect the following variables:
32 SYSIMG_PATH the path where the system image will be mounted
33 (always starts with TEMP_PATH)
34 ROOT_MOUNTED the node root file system is mounted
35 NODE_SESSION the unique session val set when we requested
36 the current boot state
37 PLCONF_DIR The directory to store PL configuration files in
39 Sets the following variables:
40 ROOT_MOUNTED the node root file system is mounted
43 log.write( "\n\nStep: Chain booting node.\n" )
45 # make sure we have the variables we need
47 SYSIMG_PATH= vars["SYSIMG_PATH"]
49 raise ValueError, "SYSIMG_PATH"
51 PLCONF_DIR= vars["PLCONF_DIR"]
53 raise ValueError, "PLCONF_DIR"
55 # its ok if this is blank
56 NODE_SESSION= vars["NODE_SESSION"]
58 NODE_MODEL_OPTIONS= vars["NODE_MODEL_OPTIONS"]
60 PARTITIONS= vars["PARTITIONS"]
61 if PARTITIONS == None:
62 raise ValueError, "PARTITIONS"
65 raise BootManagerException, "Missing variable in vars: %s\n" % var
66 except ValueError, var:
67 raise BootManagerException, "Variable in vars, shouldn't be: %s\n" % var
70 if vars.has_key('ROOT_MOUNTED'):
71 ROOT_MOUNTED= vars['ROOT_MOUNTED']
74 log.write( "Mounting node partitions\n" )
76 # simply creating an instance of this class and listing the system
77 # block devices will make them show up so vgscan can find the planetlab
79 systeminfo.get_block_device_list(vars, log)
81 utils.sysexec( "vgscan", log )
82 utils.sysexec( "vgchange -ay planetlab", log )
84 utils.makedirs( SYSIMG_PATH )
86 cmd = "mount %s %s" % (PARTITIONS["root"],SYSIMG_PATH)
87 utils.sysexec( cmd, log )
88 cmd = "mount %s %s/vservers" % (PARTITIONS["vservers"],SYSIMG_PATH)
89 utils.sysexec( cmd, log )
90 cmd = "mount -t proc none %s/proc" % SYSIMG_PATH
91 utils.sysexec( cmd, log )
94 vars['ROOT_MOUNTED']= 1
97 # write out the session value /etc/planetlab/session
99 session_file_path= "%s/%s/session" % (SYSIMG_PATH,PLCONF_DIR)
100 session_file= file( session_file_path, "w" )
101 session_file.write( str(NODE_SESSION) )
104 log.write( "Updated /etc/planetlab/session\n" )
106 log.write( "Unable to write out /etc/planetlab/session, continuing anyway\n" )
108 # update configuration files
109 log.write( "Updating configuration files.\n" )
111 cmd = "/etc/init.d/conf_files start --noscripts"
112 utils.sysexec_chroot( SYSIMG_PATH, cmd, log )
114 log.write("conf_files failed with \n %s" % e)
116 # update node packages
117 log.write( "Running node update.\n" )
118 if os.path.exists( SYSIMG_PATH + "/usr/bin/NodeUpdate.py" ):
119 cmd = "/usr/bin/NodeUpdate.py start noreboot"
121 # for backwards compatibility
122 cmd = "/usr/local/planetlab/bin/NodeUpdate.py start noreboot"
123 utils.sysexec_chroot( SYSIMG_PATH, cmd, log )
125 # the following step should be done by NM
126 UpdateNodeConfiguration.Run( vars, log )
128 log.write( "Updating ssh public host key with PLC.\n" )
131 ssh_host_key_file= file("%s/etc/ssh/ssh_host_rsa_key.pub"%SYSIMG_PATH,"r")
132 ssh_host_key= ssh_host_key_file.read().strip()
133 ssh_host_key_file.close()
134 ssh_host_key_file= None
139 update_vals['ssh_host_key']= ssh_host_key
140 BootAPI.call_api_function( vars, "BootUpdateNode", (update_vals,) )
142 # get the kernel version
144 if NODE_MODEL_OPTIONS & ModelOptions.SMP:
147 log.write( "Copying kernel and initrd for booting.\n" )
148 utils.sysexec( "cp %s/boot/kernel-boot%s /tmp/kernel" % (SYSIMG_PATH,option), log )
149 utils.sysexec( "cp %s/boot/initrd-boot%s /tmp/initrd" % (SYSIMG_PATH,option), log )
153 log.write( "Unmounting disks.\n" )
155 # backwards compat, though, we should never hit this case post PL 3.2
156 os.stat("%s/rcfs/taskclass"%SYSIMG_PATH)
157 utils.sysexec_chroot_noerr( SYSIMG_PATH, "umount /rcfs", log )
161 utils.sysexec_noerr( "umount %s/proc" % SYSIMG_PATH, log )
162 utils.sysexec_noerr( "umount -r %s/vservers" % SYSIMG_PATH, log )
163 utils.sysexec_noerr( "umount -r %s" % SYSIMG_PATH, log )
164 utils.sysexec_noerr( "vgchange -an", log )
167 vars['ROOT_MOUNTED']= 0
169 log.write( "Unloading modules and chain booting to new kernel.\n" )
171 # further use of log after Upload will only output to screen
174 # regardless of whether kexec works or not, we need to stop trying to
176 cancel_boot_flag= "/tmp/CANCEL_BOOT"
177 utils.sysexec( "touch %s" % cancel_boot_flag, log )
179 # on 2.x cds (2.4 kernel) for sure, we need to shutdown everything
180 # to get kexec to work correctly. Even on 3.x cds (2.6 kernel),
181 # there are a few buggy drivers that don't disable their hardware
182 # correctly unless they are first unloaded.
184 utils.sysexec_noerr( "ifconfig eth0 down", log )
186 utils.sysexec_noerr( "killall dhclient", log )
188 utils.sysexec_noerr( "umount -a -r -t ext2,ext3", log )
189 utils.sysexec_noerr( "modprobe -r lvm-mod", log )
191 # modules that should not get unloaded
192 # unloading cpqphp causes a kernel panic
193 blacklist = [ "floppy", "cpqphp", "i82875p_edac", "mptspi"]
195 modules= file("/tmp/loadedmodules","r")
198 module= string.strip(line)
199 if module in blacklist :
200 log.write("Skipping unload of kernel module '%s'.\n"%module)
202 log.write( "Unloading %s\n" % module )
203 utils.sysexec_noerr( "modprobe -r %s" % module, log )
204 if "e1000" in module:
205 log.write("Unloading e1000 driver; sleeping 4 seconds...\n")
210 log.write( "Couldn't read /tmp/loadedmodules, continuing.\n" )
213 modules= file("/proc/modules", "r")
215 # Get usage count for USB
219 # Module Size UsageCount UsedBy State LoadAddress
220 parts= string.split(line)
222 if parts[0] == "usb_storage":
223 usb_usage += int(parts[2])
224 except IndexError, e:
225 log.write( "Couldn't parse /proc/modules, continuing.\n" )
231 # Module Size UsageCount UsedBy State LoadAddress
232 parts= string.split(line)
234 # While we would like to remove all "unused" modules,
235 # you can't trust usage count, especially for things
236 # like network drivers or RAID array drivers. Just try
237 # and unload a few specific modules that we know cause
238 # problems during chain boot, such as USB host
239 # controller drivers (HCDs) (PL6577).
240 # if int(parts[2]) == 0:
241 if False and re.search('_hcd$', parts[0]):
243 log.write( "NOT unloading %s since USB may be in use\n" % parts[0] )
245 log.write( "Unloading %s\n" % parts[0] )
246 utils.sysexec_noerr( "modprobe -r %s" % parts[0], log )
247 except IndexError, e:
248 log.write( "Couldn't parse /proc/modules, continuing.\n" )
250 log.write( "Couldn't read /proc/modules, continuing.\n" )
253 kargs = "root=%s ramdisk_size=8192" % PARTITIONS["mapper-root"]
254 if NODE_MODEL_OPTIONS & ModelOptions.SMP:
255 kargs = kargs + " " + "acpi=off"
257 kargsfb = open("/kargs.txt","r")
258 moreargs = kargsfb.readline()
260 moreargs = moreargs.strip()
261 log.write( 'Parsed in "%s" kexec args from /kargs.txt\n' % moreargs )
262 kargs = kargs + " " + moreargs
264 # /kargs.txt does not exist, which is fine. Just kexec with default
265 # kargs, which is ramdisk_size=8192
268 utils.sysexec_noerr( 'hwclock --systohc --utc ' )
269 utils.breakpoint ("Before kexec");
271 utils.sysexec( 'kexec --force --initrd=/tmp/initrd ' \
272 '--append="%s" /tmp/kernel' % kargs)
273 except BootManagerException, e:
274 # if kexec fails, we've shut the machine down to a point where nothing
275 # can run usefully anymore (network down, all modules unloaded, file
276 # systems unmounted. write out the error, and cancel the boot process
279 log.write( "-------------------------------------------------------\n" )
280 log.write( "kexec failed with the following error. Please report\n" )
281 log.write( "this problem to support@planet-lab.org.\n\n" )
282 log.write( str(e) + "\n\n" )
283 log.write( "The boot process has been canceled.\n" )
284 log.write( "-------------------------------------------------------\n\n" )