3 # Copyright (c) 2003 Intel Corporation
6 # Copyright (c) 2004-2006 The Trustees of Princeton University
14 import UpdateBootStateWithPLC
15 from Exceptions import *
20 import notify_messages
27 Load the kernel off of a node and boot to it.
28 This step assumes the disks are mounted on SYSIMG_PATH.
29 If successful, this function will not return. If it returns, no chain
32 Expect the following variables:
33 BOOT_CD_VERSION A tuple of the current bootcd version
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 BOOT_CD_VERSION= vars["BOOT_CD_VERSION"]
50 if BOOT_CD_VERSION == "":
51 raise ValueError, "BOOT_CD_VERSION"
53 SYSIMG_PATH= vars["SYSIMG_PATH"]
55 raise ValueError, "SYSIMG_PATH"
57 PLCONF_DIR= vars["PLCONF_DIR"]
59 raise ValueError, "PLCONF_DIR"
61 # its ok if this is blank
62 NODE_SESSION= vars["NODE_SESSION"]
64 NODE_MODEL_OPTIONS= vars["NODE_MODEL_OPTIONS"]
66 PARTITIONS= vars["PARTITIONS"]
67 if PARTITIONS == None:
68 raise ValueError, "PARTITIONS"
71 raise BootManagerException, "Missing variable in vars: %s\n" % var
72 except ValueError, var:
73 raise BootManagerException, "Variable in vars, shouldn't be: %s\n" % var
76 if 'ROOT_MOUNTED' in vars.keys():
77 ROOT_MOUNTED= vars['ROOT_MOUNTED']
80 log.write( "Mounting node partitions\n" )
82 # old cds need extra utilities to run lvm
83 if BOOT_CD_VERSION[0] == 2:
84 compatibility.setup_lvm_2x_cd( vars, log )
86 # simply creating an instance of this class and listing the system
87 # block devices will make them show up so vgscan can find the planetlab
89 systeminfo.get_block_device_list(vars, log)
91 utils.sysexec( "vgscan", log )
92 utils.sysexec( "vgchange -ay planetlab", log )
94 utils.makedirs( SYSIMG_PATH )
96 cmd = "mount %s %s" % (PARTITIONS["root"],SYSIMG_PATH)
97 utils.sysexec( cmd, log )
98 cmd = "mount %s %s/vservers" % (PARTITIONS["vservers"],SYSIMG_PATH)
99 utils.sysexec( cmd, log )
100 cmd = "mount -t proc none %s/proc" % SYSIMG_PATH
101 utils.sysexec( cmd, log )
104 vars['ROOT_MOUNTED']= 1
107 # write out the session value /etc/planetlab/session
109 session_file_path= "%s/%s/session" % (SYSIMG_PATH,PLCONF_DIR)
110 session_file= file( session_file_path, "w" )
111 session_file.write( str(NODE_SESSION) )
114 log.write( "Updated /etc/planetlab/session\n" )
116 log.write( "Unable to write out /etc/planetlab/session, continuing anyway\n" )
118 # update configuration files
119 log.write( "Updating configuration files.\n" )
121 cmd = "/etc/init.d/conf_files start --noscripts"
122 utils.sysexec( "chroot %s %s" % (SYSIMG_PATH, cmd), log )
124 log.write("conf_files failed with \n %s" % e)
126 # update node packages
127 log.write( "Running node update.\n" )
128 if os.path.exists( SYSIMG_PATH + "/usr/bin/NodeUpdate.py" ):
129 cmd = "chroot %s /usr/bin/NodeUpdate.py start noreboot" % SYSIMG_PATH
131 # for backwards compatibility
132 cmd = "chroot %s /usr/local/planetlab/bin/NodeUpdate.py start noreboot" % SYSIMG_PATH
133 utils.sysexec( cmd, log )
135 log.write( "Updating ssh public host key with PLC.\n" )
138 ssh_host_key_file= file("%s/etc/ssh/ssh_host_rsa_key.pub"%SYSIMG_PATH,"r")
139 ssh_host_key= ssh_host_key_file.read().strip()
140 ssh_host_key_file.close()
141 ssh_host_key_file= None
146 update_vals['ssh_host_key']= ssh_host_key
147 BootAPI.call_api_function( vars, "BootUpdateNode", (update_vals,) )
149 # get the kernel version
151 if NODE_MODEL_OPTIONS & ModelOptions.SMP:
154 log.write( "Copying kernel and initrd for booting.\n" )
155 utils.sysexec( "cp %s/boot/kernel-boot%s /tmp/kernel" % (SYSIMG_PATH,option), log )
156 utils.sysexec( "cp %s/boot/initrd-boot%s /tmp/initrd" % (SYSIMG_PATH,option), log )
160 log.write( "Unmounting disks.\n" )
162 # backwards compat, though, we should never hit this case post PL 3.2
163 os.stat("%s/rcfs/taskclass"%SYSIMG_PATH)
164 utils.sysexec_noerr( "chroot %s umount /rcfs" % SYSIMG_PATH, log )
168 utils.sysexec_noerr( "umount %s/proc" % SYSIMG_PATH, log )
169 utils.sysexec_noerr( "umount -r %s/vservers" % SYSIMG_PATH, log )
170 utils.sysexec_noerr( "umount -r %s" % SYSIMG_PATH, log )
171 utils.sysexec_noerr( "vgchange -an", log )
174 vars['ROOT_MOUNTED']= 0
176 log.write( "Unloading modules and chain booting to new kernel.\n" )
178 # further use of log after Upload will only output to screen
181 # regardless of whether kexec works or not, we need to stop trying to
183 cancel_boot_flag= "/tmp/CANCEL_BOOT"
184 utils.sysexec( "touch %s" % cancel_boot_flag, log )
186 # on 2.x cds (2.4 kernel) for sure, we need to shutdown everything
187 # to get kexec to work correctly. Even on 3.x cds (2.6 kernel),
188 # there are a few buggy drivers that don't disable their hardware
189 # correctly unless they are first unloaded.
191 utils.sysexec_noerr( "ifconfig eth0 down", log )
193 if BOOT_CD_VERSION[0] == 2:
194 utils.sysexec_noerr( "killall dhcpcd", log )
195 elif BOOT_CD_VERSION[0] >= 3:
196 utils.sysexec_noerr( "killall dhclient", log )
198 utils.sysexec_noerr( "umount -a -r -t ext2,ext3", log )
199 utils.sysexec_noerr( "modprobe -r lvm-mod", log )
202 modules= file("/tmp/loadedmodules","r")
205 module= string.strip(line)
206 if module == "floppy":
207 log.write("Skipping unload of floppy module for dc7800.\n")
208 log.write("--!! Unknown effect on other platforms !!--\n")
210 log.write( "Unloading %s\n" % module )
211 utils.sysexec_noerr( "modprobe -r %s" % module, log )
212 if module == "e1000":
213 log.write("Unloading e1000 driver; sleeping 4 seconds...\n")
218 log.write( "Couldn't read /tmp/loadedmodules, continuing.\n" )
221 modules= file("/proc/modules", "r")
223 # Get usage count for USB
227 # Module Size UsageCount UsedBy State LoadAddress
228 parts= string.split(line)
230 if parts[0] == "usb_storage":
231 usb_usage += int(parts[2])
232 except IndexError, e:
233 log.write( "Couldn't parse /proc/modules, continuing.\n" )
239 # Module Size UsageCount UsedBy State LoadAddress
240 parts= string.split(line)
242 # While we would like to remove all "unused" modules,
243 # you can't trust usage count, especially for things
244 # like network drivers or RAID array drivers. Just try
245 # and unload a few specific modules that we know cause
246 # problems during chain boot, such as USB host
247 # controller drivers (HCDs) (PL6577).
248 # if int(parts[2]) == 0:
249 if False and re.search('_hcd$', parts[0]):
251 log.write( "NOT unloading %s since USB may be in use\n" % parts[0] )
253 log.write( "Unloading %s\n" % parts[0] )
254 utils.sysexec_noerr( "modprobe -r %s" % parts[0], log )
255 except IndexError, e:
256 log.write( "Couldn't parse /proc/modules, continuing.\n" )
258 log.write( "Couldn't read /proc/modules, continuing.\n" )
261 kargs = "root=%s ramdisk_size=8192" % PARTITIONS["mapper-root"]
262 if NODE_MODEL_OPTIONS & ModelOptions.SMP:
263 kargs = kargs + " " + "acpi=off"
265 kargsfb = open("/kargs.txt","r")
266 moreargs = kargsfb.readline()
268 moreargs = moreargs.strip()
269 log.write( 'Parsed in "%s" kexec args from /kargs.txt\n' % moreargs )
270 kargs = kargs + " " + moreargs
272 # /kargs.txt does not exist, which is fine. Just kexec with default
273 # kargs, which is ramdisk_size=8192
276 utils.breakpoint ("Before kexec");
278 utils.sysexec( 'kexec --force --initrd=/tmp/initrd ' \
279 '--append="%s" /tmp/kernel' % kargs)
280 except BootManagerException, e:
281 # if kexec fails, we've shut the machine down to a point where nothing
282 # can run usefully anymore (network down, all modules unloaded, file
283 # systems unmounted. write out the error, and cancel the boot process
286 log.write( "-------------------------------------------------------\n" )
287 log.write( "kexec failed with the following error. Please report\n" )
288 log.write( "this problem to support@planet-lab.org.\n\n" )
289 log.write( str(e) + "\n\n" )
290 log.write( "The boot process has been canceled.\n" )
291 log.write( "-------------------------------------------------------\n\n" )