5 import InstallWriteConfig
6 import UpdateBootStateWithPLC
7 from Exceptions import *
10 from systeminfo import systeminfo
12 import notify_messages
14 from GetAndUpdateNodeDetails import SMP_OPT
19 Load the kernel off of a node and boot to it.
20 This step assumes the disks are mounted on SYSIMG_PATH.
21 If successful, this function will not return. If it returns, no chain
24 Expect the following variables:
25 BOOT_CD_VERSION A tuple of the current bootcd version
26 SYSIMG_PATH the path where the system image will be mounted
27 (always starts with TEMP_PATH)
28 ROOT_MOUNTED the node root file system is mounted
29 NODE_SESSION the unique session val set when we requested
30 the current boot state
31 PLCONF_DIR The directory to store PL configuration files in
33 Sets the following variables:
34 ROOT_MOUNTED the node root file system is mounted
37 log.write( "\n\nStep: Chain booting node.\n" )
39 # make sure we have the variables we need
41 BOOT_CD_VERSION= vars["BOOT_CD_VERSION"]
42 if BOOT_CD_VERSION == "":
43 raise ValueError, "BOOT_CD_VERSION"
45 SYSIMG_PATH= vars["SYSIMG_PATH"]
47 raise ValueError, "SYSIMG_PATH"
49 PLCONF_DIR= vars["PLCONF_DIR"]
51 raise ValueError, "PLCONF_DIR"
53 # its ok if this is blank
54 NODE_SESSION= vars["NODE_SESSION"]
56 NODE_MODEL_OPTIONS= vars["NODE_MODEL_OPTIONS"]
59 raise BootManagerException, "Missing variable in vars: %s\n" % var
60 except ValueError, var:
61 raise BootManagerException, "Variable in vars, shouldn't be: %s\n" % var
64 if 'ROOT_MOUNTED' in vars.keys():
65 ROOT_MOUNTED= vars['ROOT_MOUNTED']
68 log.write( "Mounting node partitions\n" )
70 # old cds need extra utilities to run lvm
71 if BOOT_CD_VERSION[0] == 2:
72 compatibility.setup_lvm_2x_cd( vars, log )
74 # simply creating an instance of this class and listing the system
75 # block devices will make them show up so vgscan can find the planetlab
77 systeminfo().get_block_device_list()
79 utils.sysexec( "vgscan", log )
80 utils.sysexec( "vgchange -ay planetlab", log )
82 utils.makedirs( SYSIMG_PATH )
84 utils.sysexec( "mount /dev/planetlab/root %s" % SYSIMG_PATH, log )
85 utils.sysexec( "mount /dev/planetlab/vservers %s/vservers" %
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 & SMP_OPT:
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 -r /dev/planetlab/vservers", log )
145 utils.sysexec_noerr( "umount -r /dev/planetlab/root", log )
146 utils.sysexec_noerr( "vgchange -an", log )
149 vars['ROOT_MOUNTED']= 0
151 # before we do the real kexec, check to see if we had any
152 # network drivers written to modprobe.conf. if not, return -1,
153 # which will cause this node to be switched to a debug state.
154 if network_count == 0:
155 log.write( "\nIt appears we don't have any network drivers. Aborting.\n" )
157 vars['BOOT_STATE']= 'dbg'
158 vars['STATE_CHANGE_NOTIFY']= 1
159 vars['STATE_CHANGE_NOTIFY_MESSAGE']= \
160 notify_messages.MSG_NO_DETECTED_NETWORK
161 UpdateBootStateWithPLC.Run( vars, log )
165 log.write( "Unloading modules and chain booting to new kernel.\n" )
167 # further use of log after Upload will only output to screen
170 # regardless of whether kexec works or not, we need to stop trying to
172 cancel_boot_flag= "/tmp/CANCEL_BOOT"
173 utils.sysexec( "touch %s" % cancel_boot_flag, log )
175 # on 2.x cds (2.4 kernel) for sure, we need to shutdown everything
176 # to get kexec to work correctly. Even on 3.x cds (2.6 kernel),
177 # there are a few buggy drivers that don't disable their hardware
178 # correctly unless they are first unloaded.
180 utils.sysexec_noerr( "ifconfig eth0 down", log )
182 if BOOT_CD_VERSION[0] == 2:
183 utils.sysexec_noerr( "killall dhcpcd", log )
184 elif BOOT_CD_VERSION[0] == 3:
185 utils.sysexec_noerr( "killall dhclient", log )
187 utils.sysexec_noerr( "umount -a -r -t ext2,ext3", log )
188 utils.sysexec_noerr( "modprobe -r lvm-mod", log )
191 modules= file("/tmp/loadedmodules","r")
194 module= string.strip(line)
196 log.write( "Unloading %s\n" % module )
197 utils.sysexec_noerr( "modprobe -r %s" % module, log )
201 log.write( "Couldn't read /tmp/loadedmodules, continuing.\n" )
204 modules= file("/proc/modules", "r")
206 # Get usage count for USB
210 # Module Size UsageCount UsedBy State LoadAddress
211 parts= string.split(line)
213 if parts[0] == "usb_storage":
214 usb_usage += int(parts[2])
215 except IndexError, e:
216 log.write( "Couldn't parse /proc/modules, continuing.\n" )
222 # Module Size UsageCount UsedBy State LoadAddress
223 parts= string.split(line)
225 # While we would like to remove all "unused" modules,
226 # you can't trust usage count, especially for things
227 # like network drivers or RAID array drivers. Just try
228 # and unload a few specific modules that we know cause
229 # problems during chain boot, such as USB host
230 # controller drivers (HCDs) (PL6577).
231 # if int(parts[2]) == 0:
232 if re.search('_hcd$', parts[0]):
234 log.write( "NOT unloading %s since USB may be in use\n" % parts[0] )
236 log.write( "Unloading %s\n" % parts[0] )
237 utils.sysexec_noerr( "modprobe -r %s" % parts[0], log )
238 except IndexError, e:
239 log.write( "Couldn't parse /proc/modules, continuing.\n" )
241 log.write( "Couldn't read /proc/modules, continuing.\n" )
244 kargs = "root=/dev/mapper/planetlab-root ramdisk_size=8192"
246 kargsfb = open("/kargs.txt","r")
247 moreargs = kargsfb.readline()
249 moreargs = moreargs.strip()
250 log.write( 'Parsed in "%s" kexec args from /kargs.txt\n' % moreargs )
251 kargs = kargs + " " + moreargs
253 # /kargs.txt does not exist, which is fine. Just kexec with default
254 # kargs, which is ramdisk_size=8192
258 utils.sysexec( 'kexec --force --initrd=/tmp/initrd ' \
259 '--append="%s" /tmp/kernel' % kargs)
260 except BootManagerException, e:
261 # if kexec fails, we've shut the machine down to a point where nothing
262 # can run usefully anymore (network down, all modules unloaded, file
263 # systems unmounted. write out the error, and cancel the boot process
266 log.write( "-------------------------------------------------------\n" )
267 log.write( "kexec failed with the following error. Please report\n" )
268 log.write( "this problem to support@planet-lab.org.\n\n" )
269 log.write( str(e) + "\n\n" )
270 log.write( "The boot process has been canceled.\n" )
271 log.write( "-------------------------------------------------------\n\n" )