fetch: genericize git-svn fetch packing mechanism
[girocco/readme.git] / jobd / update.sh
blobaf0025192ffcf9e0623305501cf1f0be79ebfb47
1 #!/bin/sh
3 . @basedir@/shlib.sh
5 set -e
7 if [ $# -ne 1 ]; then
8 echo "Usage: update.sh projname" >&2
9 exit 1
12 # date -R is linux-only, POSIX equivalent is '+%a, %d %b %Y %T %z'
13 datefmt='+%a, %d %b %Y %T %z'
15 git_fetch_q_progress() {
16 PATH="$var_git_exec_path:$cfg_basedir/bin:$PATH" @basedir@/jobd/git-fetch-q-progress.sh "$@"
19 # freshen_loose_objects full-sha ...
20 # if "$n" is a loose object, set its modification time to now
21 # otherwise silently do nothing with no error. To facilitate conversion
22 # of mirror projects to push projects we also add group write permission.
23 freshen_loose_objects() {
24 _list=
25 for _sha; do
26 _fn="${_sha#??}"
27 _shard="${_sha%$_fn}"
28 _list="$_list objects/$_shard/$_fn"
29 done
30 if [ -n "$_list" ]; then
31 chmod ug+w $_list 2>/dev/null || :
32 touch -c $_list 2>/dev/null || :
36 # darcs fast-export | git fast-import with error handling
37 git_darcs_fetch() (
38 set_utf8_locale
39 _err1=
40 _err2=
41 exec 3>&1
42 { read -r _err1 || :; read -r _err2 || :; } <<-EOT
44 exec 4>&3 3>&1 1>&4 4>&-
46 _e1=0
47 "$cfg_basedir"/bin/darcs-fast-export \
48 --export-marks="$(pwd)/dfe-marks" \
49 --import-marks="$(pwd)/dfe-marks" "$1" 3>&- || _e1=$?
50 echo $_e1 >&3
51 } |
53 _e2=0
54 git fast-import \
55 --export-marks="$(pwd)/gfi-marks" \
56 --export-pack-edges="$(pwd)/gfi-packs" \
57 --import-marks="$(pwd)/gfi-marks" \
58 --force 3>&- || _e2=$?
59 echo $_e2 >&3
62 EOT
63 exec 3>&-
64 [ "$_err1" = 0 ] && [ "$_err2" = 0 ]
65 return $?
68 # bzr fast-export | git fast-import with error handling
69 git_bzr_fetch() (
70 set_utf8_locale
71 BZR_LOG=/dev/null
72 export BZR_LOG
73 _err1=
74 _err2=
75 exec 3>&1
76 { read -r _err1 || :; read -r _err2 || :; } <<-EOT
78 exec 4>&3 3>&1 1>&4 4>&-
80 _e1=0
81 bzr fast-export --plain \
82 --export-marks="$(pwd)/bfe-marks" \
83 --import-marks="$(pwd)/bfe-marks" "$1" 3>&- || _e1=$?
84 echo $_e1 >&3
85 } |
87 _e2=0
88 git fast-import \
89 --export-marks="$(pwd)/gfi-marks" \
90 --export-pack-edges="$(pwd)/gfi-packs" \
91 --import-marks="$(pwd)/gfi-marks" \
92 --force 3>&- || _e2=$?
93 echo $_e2 >&3
96 EOT
97 exec 3>&-
98 [ "$_err1" = 0 ] && [ "$_err2" = 0 ]
99 return $?
102 [ -n "$cfg_mirror" ] || { echo "Mirroring is disabled" >&2; exit 0; }
104 umask 002
105 [ "$cfg_permission_control" != "Hooks" ] || umask 000
106 clean_git_env
108 proj="${1%.git}"
109 cd "$cfg_reporoot/$proj.git"
111 # Activate a mini-gc if needed
112 # We do this here as well as after a successful fetch so that if we're stuck
113 # in a fetch loop where fetches are succeeding in fetching new packs but the
114 # ref update is failing for some reason (perhaps a non-commit under refs/heads)
115 # and a previous invokation therefore had a "bang" exit then we will still
116 # get the .needsgc flag set in a timely fashion to avoid excess pack build up.
117 check_and_set_needsgc
119 trap 'if [ $? != 0 ]; then echo "update failed dir: $PWD" >&2; fi; rm -f "$bang_log"' EXIT
120 trap 'exit 130' INT
121 trap 'exit 143' TERM
123 if [ "${force_update:-0}" = "0" ] && check_interval lastrefresh $cfg_min_mirror_interval; then
124 progress "= [$proj] update skip (last at $(config_get lastrefresh))"
125 exit 0
127 if [ -e .nofetch ]; then
128 progress "x [$proj] update disabled (.nofetch exists)"
129 exit 0
131 progress "+ [$proj] update ($(date))"
133 # Any pre-existing FETCH_HEAD from a previous fetch, failed or not, is garbage
134 rm -f FETCH_HEAD
136 # A previous failed update attempt can leave a huge tmp_pack_XXXXXX file behind.
137 # Since no pushes are allowed to mirrors, we know that any such files that exist
138 # at this point in time are garbage and can be safely deleted, we do not even
139 # need to check how old they are. A tmp_idx_XXXXXX file is also created during
140 # the later stages of the fetch process, so we kill any of those as well.
141 find -L objects/pack -maxdepth 1 -type f -name "tmp_pack_?*" -exec rm -f '{}' + || :
142 find -L objects/pack -maxdepth 1 -type f -name "tmp_idx_?*" -exec rm -f '{}' + || :
144 # Make sure we have a reflogs subdirectory and abort the update if not
145 # This should not count as a normal "bang" failure if unsuccessful
146 [ -d reflogs ] || mkdir -p reflogs >/dev/null 2>&1 || :
147 [ -d reflogs ]
149 keep_bang_log=
150 do_check_after_refs=1
151 bang_setup
152 bang_action="update"
153 bang_trap() {
154 if [ -n "$1" ]; then
155 # Throttle retries
156 # Since gitweb shows the .last_refresh date, it's safe to update
157 # gitweb.lastrefresh to throttle the updates w/o corrupting the
158 # last refresh date display on the gitweb summary page
159 # It's therefore important that we do NOT touch .last_refresh here
160 config_set lastrefresh "$(date "$datefmt")"
164 bang echo "Project: $proj"
165 bang echo " Date: $(TZ=UTC date '+%Y-%m-%d %T UTC')"
166 bang echo ""
167 mail="$(config_get owner)" || :
168 url="$(config_get baseurl)" || :
169 case "$url" in *" "*|*" "*|"")
170 bang_eval 'echo "Bad mirror URL (\"$url\")"; ! :'
171 exit 1
172 esac
173 bang echo "Mirroring from URL \"$url\""
174 bang echo ""
175 statusok="$(git config --bool gitweb.statusupdates 2>/dev/null || echo true)"
176 mailaddrs=
177 [ "$statusok" = "false" ] || [ -z "$mail" ] || mailaddrs="$mail"
178 [ -z "$cfg_admincc" ] || [ "$cfg_admincc" = "0" ] || [ -z "$cfg_admin" ] ||
179 if [ -z "$mailaddrs" ]; then mailaddrs="$cfg_admin"; else mailaddrs="$mailaddrs,$cfg_admin"; fi
181 bang_eval "git for-each-ref --format '%(refname) %(objectname)' >.refs-temp"
182 bang_eval "LC_ALL=C sort -b -k1,1 <.refs-temp >.refs-before"
184 check_after_refs() {
185 [ -n "$do_check_after_refs" ] || return 0
186 bang_eval "git for-each-ref --format '%(refname) %(objectname)' >.refs-temp"
187 bang_eval "LC_ALL=C sort -b -k1,1 <.refs-temp >.refs-after"
188 refschanged=
189 cmp -s .refs-before .refs-after || refschanged=1
190 do_check_after_refs=
193 ! [ -e .delaygc ] || >.allowgc || :
195 # Make sure we don't get any unwanted loose objects
196 # Starting with Git v2.10.0 fast-import can generate loose objects unless we
197 # tweak its configuration to prevent that
198 git_add_config 'fetch.unpackLimit=1'
199 # Note the git config documentation is wrong
200 # transfer.unpackLimit, if set, overrides fetch.unpackLimit
201 git_add_config 'transfer.unpackLimit=1'
202 # But not the Git v2.10.0 and later fastimport.unpackLimit which improperly uses <= instead of <
203 git_add_config 'fastimport.unpackLimit=0'
205 # remember the starting time so we can easily combine fetched loose objects
206 # we sleep for 1 second after creating .needspack to make sure all objects are newer
207 if ! [ -e .needspack ]; then
208 rm -f .needspack
209 >.needspack
210 sleep 1
213 case "$url" in
214 svn://* | svn+http://* | svn+https://* | svn+file://* | svn+ssh://*)
215 [ -n "$cfg_mirror_svn" ] || { echo "Mirroring svn is disabled" >&2; exit 0; }
216 # Use an 'anonsvn' username as is commonly used for anonymous svn
217 # Use an 'anonsvn' password as is commonly used for anonymous svn
218 GIT_ASKPASS_PASSWORD=anonsvn
219 export GIT_ASKPASS_PASSWORD
220 # Update the git svn url to match baseurl but be cognizant of any
221 # needed prefix changes. See the comments in taskd/clone.sh about
222 # why we need to put up with a prefix in the first place.
223 case "$url" in svn+ssh://*) svnurl="$url";; *) svnurl="${url#svn+}";; esac
224 svnurl="${svnurl%/}"
225 svnurlold="$(config_get svnurl)" || :
226 if [ "$svnurl" != "$svnurlold" ]; then
227 # We better already have an svn-remote.svn.fetch setting
228 bang test -n "$(git config --get-all svn-remote.svn.fetch)" || :
229 # the only way to truly know what the proper prefix is
230 # is to attempt a fresh git-svn init -s on the new url
231 rm -rf svn-new-url || :
232 # We require svn info to succeed on the URL otherwise it's
233 # simply not a valid URL and without using -s on the init it
234 # will not otherwise be tested until the fetch
235 bang eval 'svn --non-interactive --username anonsvn --password anonsvn info "$svnurl" >/dev/null'
236 bang mkdir svn-new-url
237 GIT_DIR=svn-new-url bang git init --bare --quiet
238 # We initially use -s for the init which will possibly shorten
239 # the URL. However, the shortening can fail if a password is
240 # not required for the longer version but is for the shorter,
241 # so try again without -s if the -s version fails.
242 cmdstr='git svn init --username=anonsvn --prefix "" -s "$svnurl" </dev/null >/dev/null 2>&1 || '
243 cmdstr="$cmdstr"'git svn init --username=anonsvn --prefix "" "$svnurl" </dev/null >/dev/null 2>&1'
244 GIT_DIR=svn-new-url bang eval "$cmdstr"
245 gitsvnurl="$(GIT_DIR=svn-new-url git config --get svn-remote.svn.url)" || :
246 gitsvnfetch="$(GIT_DIR=svn-new-url git config --get svn-remote.svn.fetch)" || :
247 gitsvnprefixnew="${gitsvnfetch%%:*}"
248 gitsvnsuffixnew="${gitsvnprefixnew##*/}"
249 gitsvnprefixnew="${gitsvnprefixnew%$gitsvnsuffixnew}"
250 rm -rf svn-new-url || :
251 # Using GIT_DIR= with bang leaves it set to svn-new-url, so reset it to .
252 GIT_DIR=.
253 if [ "$gitsvnurl" != "$(git config --get svn-remote.svn.url || :)" ]; then
254 # The url has been changed.
255 # We must update the url and replace the prefix on all config items
256 gitsvnfetch="$(git config --get-all svn-remote.svn.fetch | head -1)" || :
257 gitsvnprefixold="${gitsvnfetch%%:*}"
258 gitsvnsuffixold="${gitsvnprefixold##*/}"
259 gitsvnprefixold="${gitsvnprefixold%$gitsvnsuffixold}"
260 git config --remove-section 'svn-remote.svnnew' 2>/dev/null || :
261 git config 'svn-remote.svnnew.url' "$gitsvnurl"
262 git config --get-regexp '^svn-remote\.svn\.' |
263 while read -r sname sval; do
264 case "$sname" in
265 svn-remote.svn.fetch|svn-remote.svn.branches|svn-remote.svn.tags)
266 sname="${sname#svn-remote.svn.}"
267 sval="${sval#$gitsvnprefixold}"
268 bang git config --add "svn-remote.svnnew.$sname" "${gitsvnprefixnew}$sval"
269 esac
270 done
271 test $? -eq 0
272 bang git config -f svn/.metadata svn-remote.svn.reposRoot "$gitsvnurl"
273 bang git config --remove-section svn-remote.svn
274 bang git config --rename-section svn-remote.svnnew svn-remote.svn
276 bang config_set svnurl "$svnurl"
278 # remove any stale *.lock files greater than 1 hour old in case
279 # git-svn was killed on the last update because it took too long
280 find -L svn -type f -name '*.lock' -mmin +60 -exec rm -f '{}' + 2>/dev/null || :
281 GIT_DIR=. bang git svn fetch --log-window-size=$var_log_window_size --username=anonsvn --quiet </dev/null
282 # git svn does not preserve group permissions in the svn subdirectory
283 chmod -R ug+rw,o+r svn
284 # git svn also leaves behind ref turds that end with @nnn
285 # We get rid of them now
286 git for-each-ref --format='%(refname)' |
287 LC_ALL=C sed '/^..*@[1-9][0-9]*$/!d; s/^/delete /' |
288 git_updateref_stdin
289 unset GIT_ASKPASS_PASSWORD
291 darcs://*)
292 [ -n "$cfg_mirror_darcs" ] || { echo "Mirroring darcs is disabled" >&2; exit 0; }
293 httpurl="http://${url#darcs://}"
294 # remove any stale lock files greater than 1 hour old in case
295 # darcs_fast_export was killed on the last update because it took too long
296 find -L *.darcs -maxdepth 2 -type f -name 'lock' -mmin +60 -exec rm -f '{}' + 2>/dev/null || :
297 bang git_darcs_fetch "$httpurl"
299 bzr://*)
300 [ -n "$cfg_mirror_bzr" ] || { echo "Mirroring bzr is disabled" >&2; exit 0; }
301 bzrurl="${url#bzr://}"
302 bang git_bzr_fetch "$bzrurl"
304 hg+http://* | hg+https://* | hg+file://* | hg+ssh://*)
305 [ -n "$cfg_mirror_hg" ] || { echo "Mirroring hg is disabled" >&2; exit 0; }
306 # We just remove hg+ here, so hg+http://... becomes http://...
307 hgurl="${url#hg+}"
308 # Fetch any new updates
309 bang hg -R "$(pwd)/repo.hg" pull
310 # Do the fast-export | fast-import
311 bang git_hg_fetch
314 [ "$url" = "$(git config --get remote.origin.url || :)" ] || bang config_set_raw remote.origin.url "$url"
315 pruneopt=--prune
316 [ "$(git config --bool fetch.prune 2>/dev/null || :)" != "false" ] || pruneopt=
317 if ! is_gfi_mirror_url "$url"; then
318 lastwasclean=
319 [ "$(git config --bool girocco.lastupdateclean 2>/dev/null || :)" != "true" ] || lastwasclean=1
320 nextisclean=
321 [ "$(git config --bool girocco.cleanmirror 2>/dev/null || :)" != "true" ] || nextisclean=1
322 if [ "$nextisclean" != "$lastwasclean" ]; then
323 if [ -n "$nextisclean" ]; then
324 git config --replace-all remote.origin.fetch "+refs/heads/*:refs/heads/*"
325 git config --add remote.origin.fetch "+refs/tags/*:refs/tags/*"
326 git config --add remote.origin.fetch "+refs/notes/*:refs/notes/*"
327 git config --add remote.origin.fetch "+refs/top-bases/*:refs/top-bases/*"
328 else
329 git config --replace-all remote.origin.fetch "+refs/*:refs/*"
333 # remember the starting time so we can easily detect new packs for fast-import mirrors
334 # we sleep for 1 second after creating .gfipack to make sure all packs are newer
335 if is_gfi_mirror_url "$url" && [ ! -e .gfipack ]; then
336 rm -f .gfipack
337 >.gfipack
338 sleep 1
340 fetcharg="default"
341 git config remotes.default >/dev/null 2>&1 || fetcharg="--all"
342 fetchcmd="git fetch"
343 [ "$show_progress" != "0" ] || fetchcmd="git fetch -q"
344 if [ -n "$var_have_git_171" ] && [ "${show_progress:-0}" != "0" ]; then
345 # git fetch learned --progress in v1.7.1
346 case "$show_progress" in
347 [2-9]*|1[0-9]*)
348 # full volume progress with all the spammy noise
349 fetchcmd="git fetch --progress"
352 # a kinder, gentler progress that doesn't leave one
353 # covered all over in exploded bits of spam afterwards
354 fetchcmd="git_fetch_q_progress"
356 esac
358 # It's possible for a fetch to actually do something while still returning
359 # a non-zero result (perhaps some of the refs were updated but some were
360 # not -- a malicious Git-impersonation trying to set refs/heads/... refs
361 # to non-commit objects for example).
362 GIT_SSL_NO_VERIFY=1 bang_catch eval "$fetchcmd" $pruneopt --multiple "$fetcharg"
363 # If we did fetch anything, don't treat it as an error, but do keep the log;
364 # otherwise invoke bang_failed as for a normal failure
365 if [ "${bang_errcode:-0}" != "0" ]; then
366 save_bang_errcode="$bang_errcode"
367 check_after_refs
368 if [ -n "$refschanged" ]; then
369 keep_bang_log="git fetch${pruneopt:+ $pruneopt} --multiple $fetcharg"
370 else
371 bang_cmd="git fetch${pruneopt:+ $pruneopt} --multiple $fetcharg"
372 bang_errcode="$save_bang_errcode"
373 bang_failed
376 if ! is_gfi_mirror_url "$url" && [ "$nextisclean" != "$lastwasclean" ]; then
377 if [ -n "$nextisclean" ]; then
378 # We must manually purge the unclean refs now as even prune won't do it
379 git for-each-ref --format='%(refname)' |
380 LC_ALL=C sed \
381 -e '/^refs\/heads\//d' \
382 -e '/^refs\/tags\//d' \
383 -e '/^refs\/notes\//d' \
384 -e '/^refs\/top-bases\//d' \
385 -e 's/^/delete /' |
386 git_updateref_stdin
388 git config --bool girocco.lastupdateclean ${nextisclean:-0}
390 if [ -e .gfipack ] && is_gfi_mirror_url "$url"; then
391 find -L objects/pack -type f -newer .gfipack -name "pack-$octet20*.pack" -print >>gfi-packs
392 rm -f .gfipack
395 esac
397 # The objects subdirectories permissions must be updated now.
398 # In the case of a dumb http clone, the permissions will not be correct
399 # (missing group write) despite the core.sharedrepository=1 setting!
400 # The objects themselves seem to have the correct permissions.
401 # This problem appears to have been fixed in the most recent git versions.
402 perms=g+w
403 [ "$cfg_permission_control" != "Hooks" ] || perms=go+w
404 chmod $perms $(find -L objects -maxdepth 1 -type d) 2>/dev/null || :
406 bang git update-server-info
408 # We maintain the last refresh date in two places deliberately
409 # so that it's available as part of the config data and also
410 # as a standalone file timestamp that can be accessed without git.
411 bang config_set lastrefresh "$(date "$datefmt")"
412 { >.last_refresh; } 2>/dev/null || :
414 # Check to see if any refs changed
415 check_after_refs
417 # Force a mini-gc if $Girocco::Config::delay_gfi_redelta is false and there's
418 # at least one gfi pack present now
419 if [ -z "$cfg_delay_gfi_redelta" ] && ! [ -e .needsgc ] &&
420 [ -f gfi-packs ] && [ -s gfi-packs ] && is_gfi_mirror_url "$url"; then
421 >.needsgc
424 # Activate a mini-gc if needed
425 check_and_set_needsgc
427 # Look at which refs changed and trigger ref-change for these
428 sockpath="$cfg_chroot/etc/taskd.socket"
429 if [ -n "$refschanged" ]; then
430 bang config_set lastreceive "$(date '+%a, %d %b %Y %T %z')"
431 # We always use UTC for the log timestamp so that chroot and non-chroot match up.
432 # We don't have to worry about multiple log files since only one update runs
433 lognamets="$(TZ=UTC date '+%Y%m%d_%H%M%S')"
434 loghhmmss="${lognamets##*_}"
435 logname="reflogs/${lognamets%%_*}"
436 # We freshen the mod time to now on any old or new ref that is a loose object
437 # For old refs we do it so we will be able to keep them around for 1 day
438 # For new refs we do it in case we are about to run gc and the new ref
439 # actually points to an oldish loose object that had been unreachable
440 # We probably do not need to do it for new refs as Git tries to do that,
441 # but since we're already doing it for old refs (which Git does not do),
442 # it's almost no extra work for new refs, just in case.
444 echo "ref-changes %$proj% $proj"
445 LC_ALL=C join .refs-before .refs-after |
446 LC_ALL=C sed -e '/^[^ ][^ ]* \([^ ][^ ]*\) \1$/d' |
447 while read ref old new; do
448 echo "$loghhmmss $old $new $ref" >&3
449 freshen_loose_objects "$old" "$new"
450 echo "$old $new $ref"
451 done
452 LC_ALL=C join -v 1 .refs-before .refs-after |
453 while read ref old; do
454 echo "$loghhmmss $old 0000000000000000000000000000000000000000 $ref" >&3
455 freshen_loose_objects "$old"
456 echo "$old 0000000000000000000000000000000000000000 $ref"
457 done
458 LC_ALL=C join -v 2 .refs-before .refs-after |
459 while read ref new; do
460 echo "$loghhmmss 0000000000000000000000000000000000000000 $new $ref" >&3
461 freshen_loose_objects "$new"
462 echo "0000000000000000000000000000000000000000 $new $ref"
463 done
464 git for-each-ref --format='%(objectname) %(objectname) %(refname)' refs/heads
465 echo "done ref-changes %$proj% $proj"
466 } >.refs-temp 3>>"$logname"
467 if [ -S "$sockpath" ]; then
468 trap ':' PIPE
469 nc_openbsd -w 15 -U "$sockpath" <.refs-temp || :
470 trap - PIPE
472 bang config_set lastchange "$(date '+%a, %d %b %Y %T %z')"
473 bang_eval "git for-each-ref --sort=-committerdate --format='%(committerdate:iso8601)' \
474 --count=1 refs/heads >info/lastactivity"
475 ! [ -d htmlcache ] || { >htmlcache/changed; } 2>/dev/null || :
476 rm -f .delaygc .allowgc
478 [ "${cfg_autogchack:-0}" != "0" ] &&
479 [ "$(git config --get --bool girocco.autogchack 2>/dev/null)" != "false" ]
480 then
481 mv -f .refs-after .refs-last
485 # If the repository does not yet have a valid HEAD symref try to set one
486 # If an empty repository was cloned and then later becomes unempty you just
487 # lose out on the fancy "symref=HEAD:" logic and get this version instead
488 check_and_set_head || :
490 rm -f .refs-before .refs-after .refs-temp FETCH_HEAD
492 if is_banged; then
493 [ -z "$mailaddrs" ] || ! was_banged_message_sent ||
495 echo "$proj update succeeded - failure recovery"
496 echo "this status message may be disabled on the project admin page"
497 } | mailref "update@$cfg_gitweburl/$proj.git" -s "[$cfg_name] $proj update succeeded" "$mailaddrs" || :
498 bang_reset
501 if [ -n "$keep_bang_log" ] && [ -s "$bang_log" ]; then
502 cat "$bang_log" >.banglog
503 echo "" >>.banglog
504 echo "$keep_bang_log failed with error code $save_bang_errcode" >>.banglog
507 progress "- [$proj] update ($(date))"