BBlack has uploaded a new change for review.
https://gerrit.wikimedia.org/r/274083
Change subject: normalize_path: move to own include file
......................................................................
normalize_path: move to own include file
Bug: T127387
Change-Id: I8813059c5f1fe55d0899b03b83ee7bb7e9257832
---
M modules/role/manifests/cache/text.pp
A templates/varnish/normalize_path.inc.vcl.erb
M templates/varnish/text-frontend.inc.vcl.erb
3 files changed, 84 insertions(+), 84 deletions(-)
git pull ssh://gerrit.wikimedia.org:29418/operations/puppet
refs/changes/83/274083/1
diff --git a/modules/role/manifests/cache/text.pp
b/modules/role/manifests/cache/text.pp
index 66d37fd..dba9522 100644
--- a/modules/role/manifests/cache/text.pp
+++ b/modules/role/manifests/cache/text.pp
@@ -181,7 +181,7 @@
varnish::instance { 'text-frontend':
name => 'frontend',
vcl => 'text-frontend',
- extra_vcl => ['text-common', 'zero'],
+ extra_vcl => ['text-common', 'zero', 'normalize_path'],
ports => [ 80 ],
admin_port => 6082,
runtime_parameters => ['default_ttl=2592000'],
diff --git a/templates/varnish/normalize_path.inc.vcl.erb
b/templates/varnish/normalize_path.inc.vcl.erb
new file mode 100644
index 0000000..b314042
--- /dev/null
+++ b/templates/varnish/normalize_path.inc.vcl.erb
@@ -0,0 +1,82 @@
+C{
+ #include <string.h>
+}C
+sub normalize_path {
+ /* Rewrite the path part of the URL, replacing unnecessarily escaped
+ * punctuation with the actual characters. The character list is from
+ * MediaWiki's wfUrlencode(), so the URLs produced here will be the
same as
+ * the ones produced by MediaWiki in href attributes. Doing this reduces
+ * cache fragmentation and fixes T29935, i.e. stale cache entries due to
+ * MediaWiki purging only the wfUrlencode'd version of the URL.
+ */
+ C{
+ /* DIY hexadecimal conversion, since it is simple enough for a
fixed
+ * width, and all the relevant standard C library functions
promise to
+ * malfunction if the locale is set to anything other than "C"
+ */
+ #define NP_HEX_DIGIT(c) ( \
+ (c) >= '0' && (c) <= '9' ? (c) - '0' : ( \
+ (c) >= 'A' && (c) <= 'F' ? (c) - 'A' + 0x0a : (
\
+ (c) >= 'a' && (c) <= 'f' ? (c) - 'a' +
0x0a : -1 ) ) )
+ #define NP_IS_HEX(c) (NP_HEX_DIGIT(c) != -1)
+ #define NP_HEXCHAR(c1, c2) (char)( (NP_HEX_DIGIT(c1) << 4) |
NP_HEX_DIGIT(c2) )
+ const char * url = VRT_r_req_url(sp);
+ size_t i, outPos;
+ const size_t urlLength = strlen(url);
+ // index for the last position %XX can start at:
+ const size_t lastConvertIdx = urlLength > 2 ? urlLength - 3 : 0;
+ char c;
+ int dirty = 0;
+
+ /* Allocate destination memory from the stack using the C99
+ * variable-length automatic feature. We know the length in
advance
+ * because this function can only shorten the input string.
+ */
+ char destBuffer[urlLength + 1];
+ if (url) {
+ for (i = 0, outPos = 0; i < urlLength; i++) {
+ if (i <= lastConvertIdx && url[i] == '%' &&
NP_IS_HEX(url[i+1]) && NP_IS_HEX(url[i+2])) {
+ c = NP_HEXCHAR(url[i+1], url[i+2]);
+ if (c == ';'
+ || c == '@'
+ || c == '$'
+ || c == '!'
+ || c == '*'
+ || c == '('
+ || c == ')'
+ || c == ','
+ || c == '/'
+ || c == ':')
+ {
+ destBuffer[outPos++] = c;
+ dirty = 1;
+ i += 2;
+ } else {
+ destBuffer[outPos++] = url[i];
+ }
+ } else if (url[i] == '?') {
+ /* Reached the query part. Just copy
the rest of the URL
+ * to the destination.
+ */
+ memcpy(destBuffer + outPos, url + i,
sizeof(char) * (urlLength - i));
+ outPos += urlLength - i;
+ i = urlLength;
+ } else {
+ destBuffer[outPos++] = url[i];
+ }
+ }
+ destBuffer[outPos] = '\0';
+
+ /* Set req.url. This will copy our stack buffer into
the workspace.
+ * VRT_l_req_url() is varadic, and concatenates its
arguments. The
+ * vrt_magic_string_end marks the end of the list.
+ */
+ if (dirty) {
+ VRT_l_req_url(sp, destBuffer,
vrt_magic_string_end);
+ }
+ }
+ #undef NP_IS_HEX
+ #undef NP_HEX_DIGIT
+ #undef NP_HEXCHAR
+ }C
+}
diff --git a/templates/varnish/text-frontend.inc.vcl.erb
b/templates/varnish/text-frontend.inc.vcl.erb
index 5ae41f0..cfdc299 100644
--- a/templates/varnish/text-frontend.inc.vcl.erb
+++ b/templates/varnish/text-frontend.inc.vcl.erb
@@ -3,6 +3,7 @@
include "text-common.inc.vcl";
include "zero.inc.vcl";
include "geoip.inc.vcl";
+include "normalize_path.inc.vcl";
// Note that analytics.inc.vcl will set an X-Analytics value of proxy=IORG
// without inspecting whether there's an existing proxy=<proxy> key-
@@ -44,89 +45,6 @@
}
unset req.http.MobileHost;
}
-}
-
-C{
- #include <string.h>
-}C
-sub normalize_path {
- /* Rewrite the path part of the URL, replacing unnecessarily escaped
- * punctuation with the actual characters. The character list is from
- * MediaWiki's wfUrlencode(), so the URLs produced here will be the
same as
- * the ones produced by MediaWiki in href attributes. Doing this reduces
- * cache fragmentation and fixes T29935, i.e. stale cache entries due to
- * MediaWiki purging only the wfUrlencode'd version of the URL.
- */
- C{
- /* DIY hexadecimal conversion, since it is simple enough for a
fixed
- * width, and all the relevant standard C library functions
promise to
- * malfunction if the locale is set to anything other than "C"
- */
- #define NP_HEX_DIGIT(c) ( \
- (c) >= '0' && (c) <= '9' ? (c) - '0' : ( \
- (c) >= 'A' && (c) <= 'F' ? (c) - 'A' + 0x0a : (
\
- (c) >= 'a' && (c) <= 'f' ? (c) - 'a' +
0x0a : -1 ) ) )
- #define NP_IS_HEX(c) (NP_HEX_DIGIT(c) != -1)
- #define NP_HEXCHAR(c1, c2) (char)( (NP_HEX_DIGIT(c1) << 4) |
NP_HEX_DIGIT(c2) )
- const char * url = VRT_r_req_url(sp);
- size_t i, outPos;
- const size_t urlLength = strlen(url);
- // index for the last position %XX can start at:
- const size_t lastConvertIdx = urlLength > 2 ? urlLength - 3 : 0;
- char c;
- int dirty = 0;
-
- /* Allocate destination memory from the stack using the C99
- * variable-length automatic feature. We know the length in
advance
- * because this function can only shorten the input string.
- */
- char destBuffer[urlLength + 1];
- if (url) {
- for (i = 0, outPos = 0; i < urlLength; i++) {
- if (i <= lastConvertIdx && url[i] == '%' &&
NP_IS_HEX(url[i+1]) && NP_IS_HEX(url[i+2])) {
- c = NP_HEXCHAR(url[i+1], url[i+2]);
- if (c == ';'
- || c == '@'
- || c == '$'
- || c == '!'
- || c == '*'
- || c == '('
- || c == ')'
- || c == ','
- || c == '/'
- || c == ':')
- {
- destBuffer[outPos++] = c;
- dirty = 1;
- i += 2;
- } else {
- destBuffer[outPos++] = url[i];
- }
- } else if (url[i] == '?') {
- /* Reached the query part. Just copy
the rest of the URL
- * to the destination.
- */
- memcpy(destBuffer + outPos, url + i,
sizeof(char) * (urlLength - i));
- outPos += urlLength - i;
- i = urlLength;
- } else {
- destBuffer[outPos++] = url[i];
- }
- }
- destBuffer[outPos] = '\0';
-
- /* Set req.url. This will copy our stack buffer into
the workspace.
- * VRT_l_req_url() is varadic, and concatenates its
arguments. The
- * vrt_magic_string_end marks the end of the list.
- */
- if (dirty) {
- VRT_l_req_url(sp, destBuffer,
vrt_magic_string_end);
- }
- }
- #undef NP_IS_HEX
- #undef NP_HEX_DIGIT
- #undef NP_HEXCHAR
- }C
}
sub cluster_fe_recv_pre_purge {
--
To view, visit https://gerrit.wikimedia.org/r/274083
To unsubscribe, visit https://gerrit.wikimedia.org/r/settings
Gerrit-MessageType: newchange
Gerrit-Change-Id: I8813059c5f1fe55d0899b03b83ee7bb7e9257832
Gerrit-PatchSet: 1
Gerrit-Project: operations/puppet
Gerrit-Branch: production
Gerrit-Owner: BBlack <[email protected]>
_______________________________________________
MediaWiki-commits mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/mediawiki-commits
