5 import InstallWriteConfig
6 import UpdateBootStateWithPLC
7 from Exceptions import *
10 from systeminfo import systeminfo
12 import notify_messages
18 Load the kernel off of a node and boot to it.
19 This step assumes the disks are mounted on SYSIMG_PATH.
20 If successful, this function will not return. If it returns, no chain
23 Expect the following variables:
24 BOOT_CD_VERSION A tuple of the current bootcd version
25 SYSIMG_PATH the path where the system image will be mounted
26 (always starts with TEMP_PATH)
27 ROOT_MOUNTED the node root file system is mounted
28 NODE_SESSION the unique session val set when we requested
29 the current boot state
30 PLCONF_DIR The directory to store PL configuration files in
32 Sets the following variables:
33 ROOT_MOUNTED the node root file system is mounted
36 log.write( "\n\nStep: Chain booting node.\n" )
38 # make sure we have the variables we need
40 BOOT_CD_VERSION= vars["BOOT_CD_VERSION"]
41 if BOOT_CD_VERSION == "":
42 raise ValueError, "BOOT_CD_VERSION"
44 SYSIMG_PATH= vars["SYSIMG_PATH"]
46 raise ValueError, "SYSIMG_PATH"
48 PLCONF_DIR= vars["PLCONF_DIR"]
50 raise ValueError, "PLCONF_DIR"
52 # its ok if this is blank
53 NODE_SESSION= vars["NODE_SESSION"]
55 NODE_MODEL_OPTIONS= vars["NODE_MODEL_OPTIONS"]
58 raise BootManagerException, "Missing variable in vars: %s\n" % var
59 except ValueError, var:
60 raise BootManagerException, "Variable in vars, shouldn't be: %s\n" % var
63 if 'ROOT_MOUNTED' in vars.keys():
64 ROOT_MOUNTED= vars['ROOT_MOUNTED']
67 log.write( "Mounting node partitions\n" )
69 # old cds need extra utilities to run lvm
70 if BOOT_CD_VERSION[0] == 2:
71 compatibility.setup_lvm_2x_cd( vars, log )
73 # simply creating an instance of this class and listing the system
74 # block devices will make them show up so vgscan can find the planetlab
76 systeminfo().get_block_device_list()
78 utils.sysexec( "vgscan", log )
79 utils.sysexec( "vgchange -ay planetlab", log )
81 utils.makedirs( SYSIMG_PATH )
83 utils.sysexec( "mount /dev/planetlab/root %s" % SYSIMG_PATH, log )
84 utils.sysexec( "mount /dev/planetlab/vservers %s/vservers" %
86 utils.sysexec( "mount -t proc none %s/proc" % SYSIMG_PATH, log )
89 vars['ROOT_MOUNTED']= 1
92 node_update_cmd= "/usr/local/planetlab/bin/NodeUpdate.py start noreboot"
94 log.write( "Running node update.\n" )
95 utils.sysexec( "chroot %s %s" % (SYSIMG_PATH,node_update_cmd), log )
97 log.write( "Updating ssh public host key with PLC.\n" )
100 ssh_host_key_file= file("%s/etc/ssh/ssh_host_rsa_key.pub"%SYSIMG_PATH,"r")
101 ssh_host_key= ssh_host_key_file.read().strip()
102 ssh_host_key_file.close()
103 ssh_host_key_file= None
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" )
119 update_vals['ssh_host_key']= ssh_host_key
120 BootAPI.call_api_function( vars, "BootUpdateNode", (update_vals,) )
122 # rewrite modprobe.conf in case there were any module changes
123 # from a new kernel installed.
124 log.write( "Rewriting /etc/modprobe.conf\n" )
125 (network_count,storage_count)= \
126 InstallWriteConfig.write_modprobeconf_file( vars, log )
128 # get the kernel version
130 if NODE_MODEL_OPTIONS & ModelOptions.SMP:
133 log.write( "Copying kernel and initrd for booting.\n" )
134 utils.sysexec( "cp %s/boot/kernel-boot%s /tmp/kernel" % (SYSIMG_PATH,option), log )
135 utils.sysexec( "cp %s/boot/initrd-boot%s /tmp/initrd" % (SYSIMG_PATH,option), log )
137 log.write( "Unmounting disks.\n" )
139 # backwards compat, though, we should never hit this case post PL 3.2
140 os.stat("%s/rcfs/taskclass"%SYSIMG_PATH)
141 utils.sysexec_noerr( "chroot %s umount /rcfs" % SYSIMG_PATH, log )
144 utils.sysexec_noerr( "umount %s/proc" % SYSIMG_PATH, log )
145 utils.sysexec_noerr( "umount -r /dev/planetlab/vservers", log )
146 utils.sysexec_noerr( "umount -r /dev/planetlab/root", log )
147 utils.sysexec_noerr( "vgchange -an", log )
150 vars['ROOT_MOUNTED']= 0
152 # before we do the real kexec, check to see if we had any
153 # network drivers written to modprobe.conf. if not, return -1,
154 # which will cause this node to be switched to a debug state.
155 if network_count == 0:
156 log.write( "\nIt appears we don't have any network drivers. Aborting.\n" )
158 vars['BOOT_STATE']= 'dbg'
159 vars['STATE_CHANGE_NOTIFY']= 1
160 vars['STATE_CHANGE_NOTIFY_MESSAGE']= \
161 notify_messages.MSG_NO_DETECTED_NETWORK
162 UpdateBootStateWithPLC.Run( vars, log )
166 log.write( "Unloading modules and chain booting to new kernel.\n" )
168 # further use of log after Upload will only output to screen
171 # regardless of whether kexec works or not, we need to stop trying to
173 cancel_boot_flag= "/tmp/CANCEL_BOOT"
174 utils.sysexec( "touch %s" % cancel_boot_flag, log )
176 # on 2.x cds (2.4 kernel) for sure, we need to shutdown everything
177 # to get kexec to work correctly. Even on 3.x cds (2.6 kernel),
178 # there are a few buggy drivers that don't disable their hardware
179 # correctly unless they are first unloaded.
181 utils.sysexec_noerr( "ifconfig eth0 down", log )
183 if BOOT_CD_VERSION[0] == 2:
184 utils.sysexec_noerr( "killall dhcpcd", log )
185 elif BOOT_CD_VERSION[0] == 3:
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 )
192 modules= file("/tmp/loadedmodules","r")
195 module= string.strip(line)
197 log.write( "Unloading %s\n" % module )
198 utils.sysexec_noerr( "modprobe -r %s" % module, log )
202 log.write( "Couldn't read /tmp/loadedmodules, continuing.\n" )
205 modules= file("/proc/modules", "r")
207 # Get usage count for USB
211 # Module Size UsageCount UsedBy State LoadAddress
212 parts= string.split(line)
214 if parts[0] == "usb_storage":
215 usb_usage += int(parts[2])
216 except IndexError, e:
217 log.write( "Couldn't parse /proc/modules, continuing.\n" )
223 # Module Size UsageCount UsedBy State LoadAddress
224 parts= string.split(line)
226 # While we would like to remove all "unused" modules,
227 # you can't trust usage count, especially for things
228 # like network drivers or RAID array drivers. Just try
229 # and unload a few specific modules that we know cause
230 # problems during chain boot, such as USB host
231 # controller drivers (HCDs) (PL6577).
232 # if int(parts[2]) == 0:
233 if re.search('_hcd$', parts[0]):
235 log.write( "NOT unloading %s since USB may be in use\n" % parts[0] )
237 log.write( "Unloading %s\n" % parts[0] )
238 utils.sysexec_noerr( "modprobe -r %s" % parts[0], log )
239 except IndexError, e:
240 log.write( "Couldn't parse /proc/modules, continuing.\n" )
242 log.write( "Couldn't read /proc/modules, continuing.\n" )
245 kargs = "root=/dev/mapper/planetlab-root ramdisk_size=8192"
246 if NODE_MODEL_OPTIONS & ModelOptions.SMP:
247 kargs = kargs + " " + "acpi=off"
249 kargsfb = open("/kargs.txt","r")
250 moreargs = kargsfb.readline()
252 moreargs = moreargs.strip()
253 log.write( 'Parsed in "%s" kexec args from /kargs.txt\n' % moreargs )
254 kargs = kargs + " " + moreargs
256 # /kargs.txt does not exist, which is fine. Just kexec with default
257 # kargs, which is ramdisk_size=8192
261 utils.sysexec( 'kexec --force --initrd=/tmp/initrd ' \
262 '--append="%s" /tmp/kernel' % kargs)
263 except BootManagerException, e:
264 # if kexec fails, we've shut the machine down to a point where nothing
265 # can run usefully anymore (network down, all modules unloaded, file
266 # systems unmounted. write out the error, and cancel the boot process
269 log.write( "-------------------------------------------------------\n" )
270 log.write( "kexec failed with the following error. Please report\n" )
271 log.write( "this problem to support@planet-lab.org.\n\n" )
272 log.write( str(e) + "\n\n" )
273 log.write( "The boot process has been canceled.\n" )
274 log.write( "-------------------------------------------------------\n\n" )