4 from Exceptions import *
7 from systeminfo import systeminfo
13 Load the kernel off of a node and boot to it.
14 This step assumes the disks are mounted on SYSIMG_PATH.
16 Expect the following variables:
17 BOOT_CD_VERSION A tuple of the current bootcd version
18 SYSIMG_PATH the path where the system image will be mounted
19 (always starts with TEMP_PATH)
20 ROOT_MOUNTED the node root file system is mounted
21 NODE_SESSION the unique session val set when we requested
22 the current boot state
23 PLCONF_DIR The directory to store PL configuration files in
25 Sets the following variables:
26 ROOT_MOUNTED the node root file system is mounted
29 log.write( "\n\nStep: Chain booting node.\n" )
31 # make sure we have the variables we need
33 BOOT_CD_VERSION= vars["BOOT_CD_VERSION"]
34 if BOOT_CD_VERSION == "":
35 raise ValueError, "BOOT_CD_VERSION"
37 SYSIMG_PATH= vars["SYSIMG_PATH"]
39 raise ValueError, "SYSIMG_PATH"
41 PLCONF_DIR= vars["PLCONF_DIR"]
43 raise ValueError, "PLCONF_DIR"
45 # its ok if this is blank
46 NODE_SESSION= vars["NODE_SESSION"]
49 raise BootManagerException, "Missing variable in vars: %s\n" % var
50 except ValueError, var:
51 raise BootManagerException, "Variable in vars, shouldn't be: %s\n" % var
54 if 'ROOT_MOUNTED' in vars.keys():
55 ROOT_MOUNTED= vars['ROOT_MOUNTED']
58 log.write( "Mounting node partitions\n" )
60 # old cds need extra utilities to run lvm
61 if BOOT_CD_VERSION[0] == 2:
62 compatibility.setup_lvm_2x_cd( vars, log )
64 # simply creating an instance of this class and listing the system
65 # block devices will make them show up so vgscan can find the planetlab
67 systeminfo().get_block_device_list()
69 utils.sysexec( "vgscan", log )
70 utils.sysexec( "vgchange -ay planetlab", log )
72 utils.makedirs( SYSIMG_PATH )
74 utils.sysexec( "mount /dev/planetlab/root %s" % SYSIMG_PATH, log )
75 utils.sysexec( "mount /dev/planetlab/vservers %s/vservers" %
79 vars['ROOT_MOUNTED']= 1
82 node_update_cmd= "/usr/local/planetlab/bin/NodeUpdate.py start noreboot"
84 log.write( "Running node update.\n" )
85 utils.sysexec( "chroot %s %s" % (SYSIMG_PATH,node_update_cmd), log )
87 log.write( "Updating ssh public host key with PLC.\n" )
90 ssh_host_key_file= file("%s/etc/ssh/ssh_host_rsa_key.pub"%SYSIMG_PATH,"r")
91 ssh_host_key= ssh_host_key_file.read().strip()
92 ssh_host_key_file.close()
93 ssh_host_key_file= None
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" )
109 update_vals['ssh_host_key']= ssh_host_key
110 BootAPI.call_api_function( vars, "BootUpdateNode", (update_vals,) )
113 log.write( "Copying kernel and initrd for booting.\n" )
114 utils.sysexec( "cp %s/boot/kernel-boot /tmp/kernel" % SYSIMG_PATH, log )
115 utils.sysexec( "cp %s/boot/initrd-boot /tmp/initrd" % SYSIMG_PATH, log )
117 log.write( "Unmounting disks.\n" )
118 utils.sysexec_noerr( "chroot %s umount /rcfs" % SYSIMG_PATH, log )
119 utils.sysexec_noerr( "umount -r /dev/planetlab/vservers", log )
120 utils.sysexec_noerr( "umount -r /dev/planetlab/root", log )
121 utils.sysexec_noerr( "vgchange -an", log )
124 vars['ROOT_MOUNTED']= 0
126 log.write( "Unloading modules and chain booting to new kernel.\n" )
128 # further use of log after Upload will only output to screen
131 # regardless of whether kexec works or not, we need to stop trying to
133 cancel_boot_flag= "/tmp/CANCEL_BOOT"
134 utils.sysexec( "touch %s" % cancel_boot_flag, log )
136 # on 2.x cds (2.4 kernel) for sure, we need to shutdown everything
137 # to get kexec to work correctly. Even on 3.x cds (2.6 kernel),
138 # there are a few buggy drivers that don't disable their hardware
139 # correctly unless they are first unloaded.
141 utils.sysexec_noerr( "ifconfig eth0 down", log )
143 if BOOT_CD_VERSION[0] == 2:
144 utils.sysexec_noerr( "killall dhcpcd", log )
145 elif BOOT_CD_VERSION[0] == 3:
146 utils.sysexec_noerr( "killall dhclient", log )
148 utils.sysexec_noerr( "umount -a -r -t ext2,ext3", log )
149 utils.sysexec_noerr( "modprobe -r lvm-mod", log )
152 modules= file("/tmp/loadedmodules","r")
155 module= string.strip(line)
157 log.write( "Unloading %s\n" % module )
158 utils.sysexec_noerr( "modprobe -r %s" % module, log )
162 log.write( "Couldn't read /tmp/loadedmodules, continuing.\n" )
165 modules= file("/proc/modules", "r")
167 # Get usage count for USB
171 # Module Size UsageCount UsedBy State LoadAddress
172 parts= string.split(line)
174 if parts[0] == "usb_storage":
175 usb_usage += int(parts[2])
176 except IndexError, e:
177 log.write( "Couldn't parse /proc/modules, continuing.\n" )
183 # Module Size UsageCount UsedBy State LoadAddress
184 parts= string.split(line)
186 # While we would like to remove all "unused" modules,
187 # you can't trust usage count, especially for things
188 # like network drivers or RAID array drivers. Just try
189 # and unload a few specific modules that we know cause
190 # problems during chain boot, such as USB host
191 # controller drivers (HCDs) (PL6577).
192 # if int(parts[2]) == 0:
193 if re.search('_hcd$', parts[0]):
195 log.write( "NOT unloading %s since USB may be in use\n" % parts[0] )
197 log.write( "Unloading %s\n" % parts[0] )
198 utils.sysexec_noerr( "modprobe -r %s" % parts[0], log )
199 except IndexError, e:
200 log.write( "Couldn't parse /proc/modules, continuing.\n" )
202 log.write( "Couldn't read /proc/modules, continuing.\n" )
205 kargs = "ramdisk_size=8192"
207 kargsfb = open("/kargs.txt","r")
208 moreargs = kargsfb.readline()
210 moreargs = moreargs.strip()
211 log.write( 'Parsed in "%s" kexec args from /kargs.txt\n' % moreargs )
212 kargs = kargs + " " + moreargs
214 # /kargs.txt does not exist, which is fine. Just kexec with default
215 # kargs, which is ramdisk_size=8192
219 utils.sysexec( 'kexec --force --initrd=/tmp/initrd ' \
220 '--append="%s" /tmp/kernel' % kargs)
221 except BootManagerException, e:
222 # if kexec fails, we've shut the machine down to a point where nothing
223 # can run usefully anymore (network down, all modules unloaded, file
224 # systems unmounted. write out the error, and cancel the boot process
227 log.write( "-------------------------------------------------------\n" )
228 log.write( "kexec failed with the following error. Please report\n" )
229 log.write( "this problem to support@planet-lab.org.\n\n" )
230 log.write( str(e) + "\n\n" )
231 log.write( "The boot process has been canceled.\n" )
232 log.write( "-------------------------------------------------------\n\n" )