5 import InstallWriteConfig
6 import UpdateBootStateWithPLC
7 from Exceptions import *
10 from systeminfo import systeminfo
12 import notify_messages
17 Load the kernel off of a node and boot to it.
18 This step assumes the disks are mounted on SYSIMG_PATH.
19 If successful, this function will not return. If it returns, no chain
22 Expect the following variables:
23 BOOT_CD_VERSION A tuple of the current bootcd version
24 SYSIMG_PATH the path where the system image will be mounted
25 (always starts with TEMP_PATH)
26 ROOT_MOUNTED the node root file system is mounted
27 NODE_SESSION the unique session val set when we requested
28 the current boot state
29 PLCONF_DIR The directory to store PL configuration files in
31 Sets the following variables:
32 ROOT_MOUNTED the node root file system is mounted
35 log.write( "\n\nStep: Chain booting node.\n" )
37 # make sure we have the variables we need
39 BOOT_CD_VERSION= vars["BOOT_CD_VERSION"]
40 if BOOT_CD_VERSION == "":
41 raise ValueError, "BOOT_CD_VERSION"
43 SYSIMG_PATH= vars["SYSIMG_PATH"]
45 raise ValueError, "SYSIMG_PATH"
47 PLCONF_DIR= vars["PLCONF_DIR"]
49 raise ValueError, "PLCONF_DIR"
51 # its ok if this is blank
52 NODE_SESSION= vars["NODE_SESSION"]
55 raise BootManagerException, "Missing variable in vars: %s\n" % var
56 except ValueError, var:
57 raise BootManagerException, "Variable in vars, shouldn't be: %s\n" % var
60 if 'ROOT_MOUNTED' in vars.keys():
61 ROOT_MOUNTED= vars['ROOT_MOUNTED']
64 log.write( "Mounting node partitions\n" )
66 # old cds need extra utilities to run lvm
67 if BOOT_CD_VERSION[0] == 2:
68 compatibility.setup_lvm_2x_cd( vars, log )
70 # simply creating an instance of this class and listing the system
71 # block devices will make them show up so vgscan can find the planetlab
73 systeminfo().get_block_device_list()
75 utils.sysexec( "vgscan", log )
76 utils.sysexec( "vgchange -ay planetlab", log )
78 utils.makedirs( SYSIMG_PATH )
80 utils.sysexec( "mount /dev/planetlab/root %s" % SYSIMG_PATH, log )
81 utils.sysexec( "mount /dev/planetlab/vservers %s/vservers" %
85 vars['ROOT_MOUNTED']= 1
88 node_update_cmd= "/usr/local/planetlab/bin/NodeUpdate.py start noreboot"
90 log.write( "Running node update.\n" )
91 utils.sysexec( "chroot %s %s" % (SYSIMG_PATH,node_update_cmd), log )
93 log.write( "Updating ssh public host key with PLC.\n" )
96 ssh_host_key_file= file("%s/etc/ssh/ssh_host_rsa_key.pub"%SYSIMG_PATH,"r")
97 ssh_host_key= ssh_host_key_file.read().strip()
98 ssh_host_key_file.close()
99 ssh_host_key_file= None
103 # write out the session value /etc/planetlab/session
105 session_file_path= "%s/%s/session" % (SYSIMG_PATH,PLCONF_DIR)
106 session_file= file( session_file_path, "w" )
107 session_file.write( str(NODE_SESSION) )
110 log.write( "Updated /etc/planetlab/session\n" )
112 log.write( "Unable to write out /etc/planetlab/session, continuing anyway\n" )
115 update_vals['ssh_host_key']= ssh_host_key
116 BootAPI.call_api_function( vars, "BootUpdateNode", (update_vals,) )
118 # rewrite modprobe.conf in case there were any module changes
119 # from a new kernel installed.
120 log.write( "Rewriting /etc/modprobe.conf\n" )
121 (network_count,storage_count)= \
122 InstallWriteConfig.write_modprobeconf_file( vars, log )
124 log.write( "Copying kernel and initrd for booting.\n" )
125 utils.sysexec( "cp %s/boot/kernel-boot /tmp/kernel" % SYSIMG_PATH, log )
126 utils.sysexec( "cp %s/boot/initrd-boot /tmp/initrd" % SYSIMG_PATH, log )
128 log.write( "Unmounting disks.\n" )
130 # backwards compat, though, we should never hit this case post PL 3.2
131 os.stat("%s/rcfs/taskclass"%SYSIMG_PATH)
132 utils.sysexec_noerr( "chroot %s umount /rcfs" % SYSIMG_PATH, log )
135 utils.sysexec_noerr( "umount -r /dev/planetlab/vservers", log )
136 utils.sysexec_noerr( "umount -r /dev/planetlab/root", log )
137 utils.sysexec_noerr( "vgchange -an", log )
140 vars['ROOT_MOUNTED']= 0
142 # before we do the real kexec, check to see if we had any
143 # network drivers written to modprobe.conf. if not, return -1,
144 # which will cause this node to be switched to a debug state.
145 if network_count == 0:
146 log.write( "\nIt appears we don't have any network drivers. Aborting.\n" )
148 vars['BOOT_STATE']= 'dbg'
149 vars['STATE_CHANGE_NOTIFY']= 1
150 vars['STATE_CHANGE_NOTIFY_MESSAGE']= \
151 notify_messages.MSG_NO_DETECTED_NETWORK
152 UpdateBootStateWithPLC.Run( vars, log )
156 log.write( "Unloading modules and chain booting to new kernel.\n" )
158 # further use of log after Upload will only output to screen
161 # regardless of whether kexec works or not, we need to stop trying to
163 cancel_boot_flag= "/tmp/CANCEL_BOOT"
164 utils.sysexec( "touch %s" % cancel_boot_flag, log )
166 # on 2.x cds (2.4 kernel) for sure, we need to shutdown everything
167 # to get kexec to work correctly. Even on 3.x cds (2.6 kernel),
168 # there are a few buggy drivers that don't disable their hardware
169 # correctly unless they are first unloaded.
171 utils.sysexec_noerr( "ifconfig eth0 down", log )
173 if BOOT_CD_VERSION[0] == 2:
174 utils.sysexec_noerr( "killall dhcpcd", log )
175 elif BOOT_CD_VERSION[0] == 3:
176 utils.sysexec_noerr( "killall dhclient", log )
178 utils.sysexec_noerr( "umount -a -r -t ext2,ext3", log )
179 utils.sysexec_noerr( "modprobe -r lvm-mod", log )
182 modules= file("/tmp/loadedmodules","r")
185 module= string.strip(line)
187 log.write( "Unloading %s\n" % module )
188 utils.sysexec_noerr( "modprobe -r %s" % module, log )
192 log.write( "Couldn't read /tmp/loadedmodules, continuing.\n" )
195 modules= file("/proc/modules", "r")
197 # Get usage count for USB
201 # Module Size UsageCount UsedBy State LoadAddress
202 parts= string.split(line)
204 if parts[0] == "usb_storage":
205 usb_usage += int(parts[2])
206 except IndexError, e:
207 log.write( "Couldn't parse /proc/modules, continuing.\n" )
213 # Module Size UsageCount UsedBy State LoadAddress
214 parts= string.split(line)
216 # While we would like to remove all "unused" modules,
217 # you can't trust usage count, especially for things
218 # like network drivers or RAID array drivers. Just try
219 # and unload a few specific modules that we know cause
220 # problems during chain boot, such as USB host
221 # controller drivers (HCDs) (PL6577).
222 # if int(parts[2]) == 0:
223 if re.search('_hcd$', parts[0]):
225 log.write( "NOT unloading %s since USB may be in use\n" % parts[0] )
227 log.write( "Unloading %s\n" % parts[0] )
228 utils.sysexec_noerr( "modprobe -r %s" % parts[0], log )
229 except IndexError, e:
230 log.write( "Couldn't parse /proc/modules, continuing.\n" )
232 log.write( "Couldn't read /proc/modules, continuing.\n" )
235 kargs = "ramdisk_size=8192"
237 kargsfb = open("/kargs.txt","r")
238 moreargs = kargsfb.readline()
240 moreargs = moreargs.strip()
241 log.write( 'Parsed in "%s" kexec args from /kargs.txt\n' % moreargs )
242 kargs = kargs + " " + moreargs
244 # /kargs.txt does not exist, which is fine. Just kexec with default
245 # kargs, which is ramdisk_size=8192
249 utils.sysexec( 'kexec --force --initrd=/tmp/initrd ' \
250 '--append="%s" /tmp/kernel' % kargs)
251 except BootManagerException, e:
252 # if kexec fails, we've shut the machine down to a point where nothing
253 # can run usefully anymore (network down, all modules unloaded, file
254 # systems unmounted. write out the error, and cancel the boot process
257 log.write( "-------------------------------------------------------\n" )
258 log.write( "kexec failed with the following error. Please report\n" )
259 log.write( "this problem to support@planet-lab.org.\n\n" )
260 log.write( str(e) + "\n\n" )
261 log.write( "The boot process has been canceled.\n" )
262 log.write( "-------------------------------------------------------\n\n" )