X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=conf_files%2Fpl_boot;h=1f5fa5a8d71a2e5a5a3e2f0770037e14ecafcf79;hb=refs%2Fheads%2Fplanetlab-3_3-branch;hp=960995e695e274934b75342fcb77c5558569606c;hpb=9ff8e209681347ea06573858dd25449238a5735d;p=bootcd.git

diff --git a/conf_files/pl_boot b/conf_files/pl_boot
index 960995e..1f5fa5a 100644
--- a/conf_files/pl_boot
+++ b/conf_files/pl_boot
@@ -5,6 +5,14 @@ GNUPGHOME=/root
 export GNUPGHOME
 /usr/bin/gpg --yes 2>/dev/null </dev/null
 
+# if this file is present, cancel the boot (exit this script)
+CANCEL_BOOT_FLAG=/tmp/CANCEL_BOOT
+
+# how many times to fail in attempting to contact primary server
+# before falling back to original. if the backup fails this many times
+# too, then the process is repeated started with the primary server
+ATTEMPTS_BEFORE_BACKUP=3
+
 # where all the configuration files for contacting
 # the boot server are stored
 BOOT_DIR=/usr/boot/
@@ -14,13 +22,26 @@ BOOT_SERVER=`cat $BOOT_DIR/boot_server`
 BOOT_SERVER_PORT=`cat $BOOT_DIR/boot_server_port`
 
 # the file to request from the boot server
-SERVER_BOOT_DIR=`cat $BOOT_DIR/boot_server_path`
+BOOT_SERVER_PATH=`cat $BOOT_DIR/boot_server_path`
 
 # location of the cacert for this boot server
-BOOT_CACERT=$BOOT_DIR/cacert.pem
+BOOT_SERVER_CACERT=$BOOT_DIR/cacert.pem
 
 # location of the gpg key ring to verify scripts
-BOOT_GPG_KEYRING=$BOOT_DIR/pubring.gpg
+BOOT_SERVER_GPG_KEYRING=$BOOT_DIR/pubring.gpg
+
+# get the backup server we are going to be contacting
+BACKUP_BOOT_SERVER=`cat $BOOT_DIR/backup/boot_server`
+BACKUP_BOOT_SERVER_PORT=`cat $BOOT_DIR/backup/boot_server_port`
+
+# the file to request from the backup boot server
+BACKUP_BOOT_SERVER_PATH=`cat $BOOT_DIR/backup/boot_server_path`
+
+# location of the cacert for the backup boot server
+BACKUP_BOOT_SERVER_CACERT=$BOOT_DIR/backup/cacert.pem
+
+# location of the gpg key ring for backup server to verify scripts
+BACKUP_BOOT_SERVER_GPG_KEYRING=$BOOT_DIR/backup/pubring.gpg
 
 # location of a file containing this boot cd version
 BOOT_VERSION_FILE=/pl_version
@@ -29,53 +50,101 @@ BOOT_VERSION_FILE=/pl_version
 UNVERIFIED_SCRIPT=/tmp/bootscript.gpg
 VERIFIED_SCRIPT=/tmp/bootscript
 
-# asemble the curl transaction
-CURL_CMD="/usr/bin/curl \
-   --connect-timeout 60 \
-   --max-time 600 \
-   --form version=<$BOOT_VERSION_FILE \
-   --form cmdline=</proc/cmdline \
-   --form uptime=</proc/uptime \
-   --form ifconfig=</tmp/ifconfig \
-   --form cpuinfo=</proc/cpuinfo \
-   --form meminfo=</proc/meminfo \
-   --form nonce=</tmp/nonce \
-   --location \
-   --output $UNVERIFIED_SCRIPT \
-   --sslv3  \
-   --silent \
-   --show-error \
-   --fail \
-   --stderr /tmp/curl_errors \
-   --cacert $BOOT_CACERT \
-   https://$BOOT_SERVER:$BOOT_SERVER_PORT/$SERVER_BOOT_DIR"
-
-
-# assemble the gpg command line
-GPG_CMD="/usr/bin/gpg \
-    --no-default-keyring \
-    --keyring $BOOT_GPG_KEYRING \
-    --output $VERIFIED_SCRIPT \
-    --always-trust \
-    --decrypt $UNVERIFIED_SCRIPT"
+
+# --------------------------
 
 
 # now, contact the boot server, run the script, and do it over again.
-first=1
-while true; do
+contact_count=0
+
+# set to one when we are trying to contact backup server
+on_backup_server=0
+
+# start out contacting the primary servers
+CONNECT_BOOT_SERVER=$BOOT_SERVER
+CONNECT_BOOT_SERVER_PORT=$BOOT_SERVER_PORT
+CONNECT_BOOT_SERVER_PATH=$BOOT_SERVER_PATH
+CONNECT_BOOT_SERVER_GPG_KEYRING=$BOOT_SERVER_GPG_KEYRING
+CONNECT_BOOT_SERVER_CACERT=$BOOT_SERVER_CACERT
+
+while : ; do
+
+    if [[ -f $CANCEL_BOOT_FLAG ]]; then
+	echo "pl_boot: got request to cancel boot, exiting"
+	exit 0
+    fi
+    
+    if [[ $contact_count -ge $ATTEMPTS_BEFORE_BACKUP ]]; then
+
+	contact_count=0
+
+	if [[ $on_backup_server == 1 ]]; then
+	    echo "pl_boot: failed to contact backup server, trying primary."
+
+	    on_backup_server=0
+
+	    CONNECT_BOOT_SERVER=$BOOT_SERVER
+	    CONNECT_BOOT_SERVER_PORT=$BOOT_SERVER_PORT
+	    CONNECT_BOOT_SERVER_PATH=$BOOT_SERVER_PATH
+	    CONNECT_BOOT_SERVER_GPG_KEYRING=$BOOT_SERVER_GPG_KEYRING
+	    CONNECT_BOOT_SERVER_CACERT=$BOOT_SERVER_CACERT
+	else
+	    echo "pl_boot: failed to contact primary server, trying backup."
 
-    if [[ $first -eq 0 ]]; then
-	echo "pl_boot: fetching new script in 30 seconds"
+	    on_backup_server=1
+
+	    CONNECT_BOOT_SERVER=$BACKUP_BOOT_SERVER
+	    CONNECT_BOOT_SERVER_PORT=$BACKUP_BOOT_SERVER_PORT
+	    CONNECT_BOOT_SERVER_PATH=$BACKUP_BOOT_SERVER_PATH
+	    CONNECT_BOOT_SERVER_GPG_KEYRING=$BACKUP_BOOT_SERVER_GPG_KEYRING
+	    CONNECT_BOOT_SERVER_CACERT=$BACKUP_BOOT_SERVER_CACERT
+	fi
+    fi
+
+    if [[ $contact_count != 0 ]]; then
+
+	if [[ $on_backup_server == 1 ]]; then
+	    echo "pl_boot: attempting to fetch script from backup server in 30s"
+	else
+	    echo "pl_boot: attempting to fetch script from primary server in 30s"
+	fi
 	/bin/sleep 30
     fi
-    first=0
+
+    # assemble the curl transaction
+    CURL_CMD="/usr/bin/curl \
+        --connect-timeout 60 \
+        --max-time 600 \
+        --form version=<$BOOT_VERSION_FILE \
+        --form cmdline=</proc/cmdline \
+        --form uptime=</proc/uptime \
+        --form ifconfig=</tmp/ifconfig \
+        --form nonce=</tmp/nonce \
+        --location \
+        --output $UNVERIFIED_SCRIPT \
+        --sslv3  \
+        --silent \
+        --show-error \
+        --fail \
+        --stderr /tmp/curl_errors \
+        --cacert $CONNECT_BOOT_SERVER_CACERT \
+   https://$CONNECT_BOOT_SERVER:$CONNECT_BOOT_SERVER_PORT/$CONNECT_BOOT_SERVER_PATH"
+
+    # assemble the gpg command line
+    GPG_CMD="/usr/bin/gpg \
+        --no-default-keyring \
+        --keyring $CONNECT_BOOT_SERVER_GPG_KEYRING \
+        --output $VERIFIED_SCRIPT \
+        --always-trust \
+        --decrypt $UNVERIFIED_SCRIPT"
 
     echo "pl_boot: generating new nonce"
     /usr/bin/head --bytes=32 /dev/urandom | \
 	/usr/bin/od -tx1 -An --width=32 | \
 	/bin/sed 's/ //g' > /tmp/nonce
 
-    echo "pl_boot: fetching script from boot server $BOOT_SERVER"
+    echo "pl_boot: fetching script from boot server $CONNECT_BOOT_SERVER"
+    ((contact_count++))
     rm -f $UNVERIFIED_SCRIPT
     $CURL_CMD
     curl_err=$?
@@ -86,7 +155,7 @@ while true; do
 	continue
     fi 
 
-    echo "pl_boot: verifing downloaded script"
+    echo "pl_boot: verifying downloaded script"
     rm -f $VERIFIED_SCRIPT
     $GPG_CMD 2> /tmp/gpg_errors
     if [ $? -ne 0 ]; then
@@ -105,4 +174,4 @@ while true; do
     echo "pl_boot: downloaded script has returned"
 done
 
-echo "pl_boot: automatic boot process canceled"
+echo "pl_boot: automatic boot process canceled by user"