# # urlinstall.py - URL based install source method # # Erik Troan # # Copyright 1999-2002 Red Hat, Inc. # # This software may be freely redistributed under the terms of the GNU # library public license. # # You should have received a copy of the GNU Library Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. # from hdrlist import groupSetFromCompsFile, HeaderList from installmethod import InstallMethod, FileCopyException import os import rpm import time import urllib2 import string import struct import socket from snack import * from constants import * from rhpl.translate import _ # we import these explicitly because urllib loads them dynamically, which # stinks -- and we need to have them imported for the --traceonly option import ftplib import httplib import StringIO from rhpl.log import log FILENAME = 1000000 DISCNUM = 1000002 def urlretrieve(location, file, callback=None): """Downloads from location and saves to file.""" if callback is not None: callback(_("Connecting..."), 0) try: url = urllib2.urlopen(location) except urllib2.HTTPError, e: raise IOError(e.code, e.msg) except urllib2.URLError, e: raise IOError(-1, e.reason) # see if there is a size try: filesize = int(url.info()["Content-Length"]) except: filesize = None # handle zero length case if filesize == 0: filesize = None # create output file f = open(file, 'w+') # if they dont want a status callback just do it in one big swoop if callback is None: f.write(url.read()) else: buf = url.read(65535) tot = len(buf) while len(buf) > 0: if filesize is not None: callback("downloading", "%3d%%" % ((100*tot)/filesize,)) else: callback("downloading", "%dKB" % (tot/1024,)) f.write(buf) buf = url.read(65535) tot += len(buf) f.close() url.close() class UrlInstallMethod(InstallMethod): def readCompsViaMethod(self, hdlist): fname = self.findBestFileMatch(None, 'comps.xml') # if not local then assume its on host if fname is None: fname = '%s/%s/base/comps.xml' % (self.baseUrl, productPath) log("Comps not in update dirs, using %s",fname) return groupSetFromCompsFile(fname, hdlist) def getFilename(self, filename, callback=None, destdir=None, retry=1, disc = 1): if destdir is None: tmppath = self.getTempPath() else: tmppath = destdir if self.multiDiscs: base = "%s/disc%d" %(self.pkgUrl, disc) else: base = self.pkgUrl fullPath = base + "/" + filename file = tmppath + "/" + os.path.basename(fullPath) tries = 0 while tries < 5: try: rc=urlretrieve(fullPath, file, callback=callback) except IOError, (errnum, msg): log("IOError %s occurred getting %s: %s" %(errnum, fullPath, str(msg))) if not retry: raise FileCopyException time.sleep(5) else: break tries = tries + 1 if tries >= 5: raise FileCopyException return file def getRPMFilename(self, h, timer, callback=None): fullPath = "/%s/RPMS/%s" % (productPath, h[FILENAME]) return self.getFilename(fullPath, callback=callback, disc = h[DISCNUM]) def copyFileToTemp(self, filename): tmppath = self.getTempPath() if self.multiDiscs: base = "%s/disc1" % (self.pkgUrl,) else: base = self.pkgUrl fullPath = base + "/" + filename file = tmppath + "/" + os.path.basename(fullPath) tries = 0 while tries < 5: try: urlretrieve(fullPath, file) except IOError, (errnum, msg): log("IOError %s occurred getting %s: %s", errnum, fullPath, str(msg)) time.sleep(5) else: break tries = tries + 1 if tries >= 5: raise FileCopyException return file def unlinkFilename(self, fullName): os.remove(fullName) def readHeaders(self): tries = 0 while tries < 5: hdurl = "%s/%s/base/hdlist" % (self.baseUrl, productPath) try: url = urllib2.urlopen(hdurl) except urllib2.HTTPError, e: log("HTTPError: %s occurred getting %s", hdurl, e) except urllib2.URLError, e: log("URLError: %s occurred getting %s", hdurl, e) except IOError, (errnum, msg): log("IOError %s occurred getting %s: %s", errnum, hdurl, msg) else: break time.sleep(5) tries = tries + 1 if tries >= 5: raise FileCopyException raw = url.read(16) if raw is None or len(raw) < 1: raise TypeError, "header list is empty!" hl = [] while (raw and len(raw)>0): info = struct.unpack("iiii", raw) magic1 = socket.ntohl(info[0]) & 0xffffffffL if (magic1 != 0x8eade801L or info[1]): raise TypeError, "bad magic in header" il = socket.ntohl(info[2]) dl = socket.ntohl(info[3]) totalSize = il * 16 + dl; hdrString = raw[8:] + url.read(totalSize) hdr = rpm.headerLoad(hdrString) hl.append(hdr) raw = url.read(16) return HeaderList(hl) def mergeFullHeaders(self, hdlist): fn = self.getFilename("%s/base/hdlist2" % (productPath,), callback=None) hdlist.mergeFullHeaders(fn) os.unlink(fn) def setIntf(self, intf): self.intf = intf def __init__(self, url, rootPath): InstallMethod.__init__(self, rootPath) if url.startswith("ftp"): isFtp = 1 else: isFtp = 0 # build up the url. this is tricky so that we can replace # the first instance of // with /%3F to do absolute URLs right i = string.index(url, '://') + 3 self.baseUrl = url[:i] rem = url[i:] i = string.index(rem, '/') + 1 self.baseUrl = self.baseUrl + rem[:i] rem = rem[i:] # encoding fun so that we can handle absolute paths if rem.startswith("/") and isFtp: rem = "%2F" + rem[1:] self.baseUrl = self.baseUrl + rem if self.baseUrl[-1] == "/": self.baseUrl = self.baseUrl[:-1] # self.baseUrl points at the path which contains the 'RedHat' # directory with the hdlist. if self.baseUrl[-6:] == "/disc1": self.multiDiscs = 1 self.pkgUrl = self.baseUrl[:-6] else: self.multiDiscs = 0 self.pkgUrl = self.baseUrl self.intf = None