The function stats_tlskeys_list() can meet an undefined behavior when
called with appctx->st2 == STAT_ST_LIST, indeed the ref pointer is used
uninitialized.
However this function was using NULL in appctx->ctx.tlskeys.ref as a
flag to dump every tickets from every references. A real flag
appctx->ctx.tlskeys.dump_all is now used for this behavior.
This patch delete the 'ref' variable and use appctx->ctx.tlskeys.ref
directly.
Valgrind reports that the memory allocated in ssl_get_dh_1024() was leaking. Upon further inspection of openssl code, it seems that SSL_CTX_set_tmp_dh makes a copy of the data, so calling DH_free afterwards makes sense.
If we use the action "http-request add-header" with a Lua sample-fetch or
converter, and the Lua function calls one of the Lua log function, the
header name is corrupted, it contains an extract of the last loggued data.
This is due to an overwrite of the trash buffer, because his scope is not
respected in the "add-header" function. The scope of the trash buffer must
be limited to the function using it. The build_logline() function can
execute a lot of other function which can use the trash buffer.
This patch fix the usage of the trash buffer. It limits the scope of this
global buffer to the local function, we build first the header value using
build_logline, and after we store the header name.
Thanks Michael Ezzell for the repporting.
This patch must be backported in 1.6 version
The header name is copied two time in the buffer. The first copy is a printf-like
function writing the name and the http separators in the buffer, and the second
form is a memcopy. This seems to be inherited from some changes. This patch
removes the printf like, format.
This patch must be backported in 1.6 and 1.5 versions
The number of arguments pushed in the stack are false, so we try to execute a
function out of the stack. This function is always a nil pointer, so the
following message is displayed.
Lua converter 'testconv': runtime error: attempt to call a nil value.
Thanks Michael Ezzell for the repporting.
This patch must be backported in the 1.6 version.
We now instrument the makefile to keep a copy of previous build options.
The goal is to ensure that we'll rebuild everything when build options
change. The options that are watched are TARGET, VERBOSE_CFLAGS, and
BUILD_OPTIONS. These ones are copied into a file ".build_opts" and
compared to the new ones upon each build. This file is referenced in
the DEP variable which all .o files depend on, and it depends on the
code which updates it only upon changes. This ensures that a new file
is regenerated and detected upon change and that everything is rebuilt.
Some users tend to get caught by incorrect builds when they try patches
that modify some include file after they forget to run "make clean".
While we can't blame users who are not developers, forcing developers
to rely on a painful autodepend is not nice either and will cause them
to test their changes less often. Here we propose a reasonable tradeoff.
This patch introduces a new "INCLUDES" variable which enumerates all
the ".h" files and sets them as a build dependency for all ".o" files.
This list is then copied into a "DEP" variable which can safely be
overridden if desired. This way by default all .c files are rebuilt if
any include file changes. This is the safe method for all users. And
developers can simply add "DEP=" to their quick build scripts to keep
the old fast and efficient behaviour.
When a stick table is tracked, and another one is used later on the
configuration, a segfault occurs.
The function "smp_create_src_stkctr" can return a NULL value, and
its value is not tested, so one other function try to dereference
a NULL pointer. This patch just add a verification of the NULL
pointer.
The problem is reproduced with this configuration:
listen www
mode http
bind :12345
tcp-request content track-sc0 src table IPv4
http-request allow if { sc0_inc_gpc0(IPv6) gt 0 }
server dummy 127.0.0.1:80
backend IPv4
stick-table type ip size 10 expire 60s store gpc0
backend IPv6
stick-table type ipv6 size 10 expire 60s store gpc0
Thank to kabefuna@gmail.com for the bug report.
This patch must be backported in the 1.6 and 1.5 version.
The 'set-src' action was not available for tcp actions The action code
has been converted into a function in proto_tcp.c to be used for both
'http-request' and 'tcp-request connection' actions.
Both http and tcp keywords are registered in proto_tcp.c
The reference to the tls_keys_ref was not deleted from the
tlskeys_reference linked list.
When the SSL is malconfigured, it can lead to an access to freed memory
during a "show tls-keys" on the admin socked.
Olivier Doucet reported that "show servers state" was producing an invalid
output with some configurations where nbproc > 1.
Indeed, commit 76a99784f4 fixed some issues but unfortunately introduced a
regression when a backend bound to the same process as the stats socket and a
previous backend is bound to another one.
For example :
global
daemon
nbproc 2
stats socket /var/run/haproxy-1.sock process 1
stats socket /var/run/haproxy-2.sock process 2
listen proc1
bind 127.0.0.1:9001
bind-process 1
server WRONG 127.0.0.1:80
listen proc2
bind 127.0.0.1:9002
bind-process 2
server RIGHT 127.0.0.1:80
Requesting "show servers state" on /var/run/haproxy-2.sock was producing a line
like :
3 proc2 1 WRONG 127.0.0.1 2 0 1 1 4 1 0 2 0 0 0 0
whereas the line below was awaited :
3 proc2 1 RIGHT 127.0.0.1 2 0 1 1 5 1 0 2 0 0 0 0
This was caused by the initialization of the server loop too early, before the
bind_proc filtering whereas it should be done after.
This fix should be backported to 1.6, where the regression has unfortunately
been backported.
Ben Cabot reported that after commit 5e4261b ("CLEANUP: config:
detect double registration of a config section") recently introduced
in 1.7-dev, it's not possible anymore to load multiple configuration
files. Bryan Talbot provided a simple reproducer to exhibit the issue.
It turns out that function readcfgfile() registers new parsers for
section keywords for each new file. In addition to being useless, this
has the negative effect of wasting memory and slowing down the config
parser as the number of configuration files increases.
This fix only needs to be backported if/where the commit above is
backported.
htonll()/ntohll() already exist on Solaris 11 with a different declaration,
causing a build error as reported by Jonathan Fisher. They used to exist on
OSX with a #define which allowed us to detect them. It was a bad idea to give
these functions a name subject to conflicts like this. Simply rename them
my_htonll()/my_ntohll() to definitely get rid of the conflict.
This patch must be backported to 1.6.
DNS requests (using the internal resolver) are corrupted since commit
e2f8497716 ("BUG/MINOR: dns: fix DNS header definition").
Fix it by defining the struct in network byte order, while complying
with RFC 2535, section 6.1.
First reported by Eduard Vopicka on discourse.
This must be backported to 1.6 (1.6.5 is affected).
The stick-table converters used to take a string on input because it
was the only type that could be casted to from any other type. This is
inefficient and possibly inaccurate sometimes. For example in order to
look up an IP address, it must first be converted to a string then
converted back to an IP address.
We've had SMP_T_ANY introduced long ago in 1.6, but unfortunately it
was not propagated to these converters, so let's do it now.
It's important to note that a few direct type conversions which already
would not make any sense are not possible (for example, converting a
boolean to an IP address or an HTTP method to an integer). While this
would have caused the lookup to be performed on the wrong key, now the
lookup will fail and the converter will return no data. While there
should not be any case where this happens, it's probably best to avoid
backporting this change before a longer observation period.
Baptiste reported that the table_conn_rate() converter would always
return zero in 1.6.5. In fact, commit bc8c404 ("MAJOR: stick-tables:
use sample types in place of dedicated types") broke all stick-table
converters because smp_to_stkey() now returns a pointer to the sample
instead of holding a copy of the key, and the converters used to
reinitialize the sample prior to performing the lookup. Only
"in_table()" continued to work.
The construct is still fragile, so some comments were added to a few
function to clarify their impacts. It's also worth noting that there
is no point anymore in forcing these converters to take a string on
input, but that will be changed in another commit.
The bug was introduced in 1.6-dev4, this fix must be backported to 1.6.
Commit 108b1dd ("MEDIUM: http: configurable http result codes for
http-request deny") introduced in 1.6-dev2 was incomplete. It introduced
a new field "rule_deny_status" into struct http_txn, which is filled only
by actions "http-request deny" and "http-request tarpit". It's then used
in the deny code path to emit the proper error message, but is used
uninitialized when the deny comes from a "reqdeny" rule, causing random
behaviours ranging from returning a 200, an empty response, or crashing
the process. Often upon startup only 200 was returned but after the fields
are used the crash happens. This can be sped up using -dM.
There's no need at all for storing this status in the http_txn struct
anyway since it's used immediately after being set. Let's store it in
a temporary variable instead which is passed as an argument to function
http_req_get_intercept_rule().
As an extra benefit, removing it from struct http_txn reduced the size
of this struct by 8 bytes.
This fix must be backported to 1.6 where the bug was detected. Special
thanks to Falco Schmutz for his detailed report including an exploitable
core and a reproducer.
When compiled with GCC 6, the IP address specified for a frontend was
ignored and HAProxy was listening on all addresses instead. This is
caused by an incomplete copy of a "struct sockaddr_storage".
With the GNU Libc, "struct sockaddr_storage" is defined as this:
struct sockaddr_storage
{
sa_family_t ss_family;
unsigned long int __ss_align;
char __ss_padding[(128 - (2 * sizeof (unsigned long int)))];
};
Doing an aggregate copy (ss1 = ss2) is different than using memcpy():
only members of the aggregate have to be copied. Notably, padding can be
or not be copied. In GCC 6, some optimizations use this fact and if a
"struct sockaddr_storage" contains a "struct sockaddr_in", the port and
the address are part of the padding (between sa_family and __ss_align)
and can be not copied over.
Therefore, we replace any aggregate copy by a memcpy(). There is another
place using the same pattern. We also fix a function receiving a "struct
sockaddr_storage" by copy instead of by reference. Since it only needs a
read-only copy, the function is converted to request a reference.
When comparing very different branches, it can take a very long time
to scan all commits from the very old common ancestor (eg: haproxy
1.4 to 1.7). Now it is possible to specify a range of commits instead
of a specific branch, and the analysis will be limited to this range
for all commits. The user is responsible for ensuring that the range
covers all possible backports from base to ref, otherwise some of them
may be reported missing while they are not.
This also works with linux kernels, for example :
git-show-backports -u -q -m -r v3.14.69 -b v3.14.65 v3.10.101..HEAD
When git-show-backports is called with -u, instead of reporting the
commit IDs of the original branch on the left, it will display the
upstream commit IDs when such IDs are known, and will also display
them in the suggested "git cherry-pick" command line. This is useful
when the new branch is more recent than the one being checked and/or
it is desired to get rid of intermediary changes.
This patch removes setlocale from the main function. It was introduced
by commit 379d9c7 ("MEDIUM: init: allow directory as argument of -f")
in 1.7-dev a few commits ago after a discussion on the mailing list.
Some regex may have different behaviours depending on the
locale. Some LUA scripts may change their behaviour too
(http://lua-users.org/wiki/LuaLocales).
Without this patch (haproxy is using setlocale) :
$ cat locale.cfg
defaults
mode http
frontend test
bind :9000
mode http
use_backend testbk if { hdr_reg(X-Test) ^\w+$ }
backend testbk
mode http
server s 127.0.0.1:80
$ LANG=fr_FR.UTF-8 ./haproxy -f locale.cfg
$ curl -i -H "X-Test: chec" localhost:9000
HTTP/1.1 200 OK
...
$ LANG=C ./haproxy -f locale.cfg
$ curl -i -H "X-Test: chec" localhost:9000
HTTP/1.0 503 Service Unavailable
...
'channel_analyze' callback has been removed. Now, there are 2 callbacks to
surround calls to analyzers:
* channel_pre_analyze: Called BEFORE all filterable analyzers. it can be
called many times for the same analyzer, once at each loop until the
analyzer finishes its processing. This callback is resumable, it returns a
negative value if an error occurs, 0 if it needs to wait, any other value
otherwise.
* channel_post_analyze: Called AFTER all filterable analyzers. Here, AFTER
means when an analyzer finishes its processing. This callback is NOT
resumable, it returns a negative value if an error occurs, any other value
otherwise.
Pre and post analyzer callbacks are not automatically called. 'pre_analyzers'
and 'post_analyzers' bit fields in the filter structure must be set to the right
value using AN_* flags (see include/types/channel.h).
The flag AN_RES_ALL has been added (AN_REQ_ALL already exists) to ease the life
of filter developers. AN_REQ_ALL and AN_RES_ALL include all filterable
analyzers.
Now, to call an analyzer in 'process_stream' function, we should use
FLT_ANALAYZE or ANALYZE macros, depending if this is a filterable analyzer or
not.
Instead of calling 'channel_analyze' callback with the flag AN_FLT_HTTP_HDRS,
now we use the new callback 'http_headers'. This change is done because
'channel_analyze' callback will be removed in a next commit.
As suggested by Pavlos, it's too bad that we didn't have a %Td log
format tag given that there are a few mentions of Td corresponding
to the data transmission time already in the doc, so this is now done.
Just like the other specifiers, we report -1 if the connection failed
before reaching the data transmission state.
In an effort to make the config parser more robust, we should validate
that everything we register is not already registered. Most cfg_register_*
functions unfortunately return void and just perform a LIST_ADDQ(), so they
will have to change for this. At least cfg_register_section() does perform
a bit of checks and is easy to check for such errors, so let's start with
this one. Future patches will definitely have to focus on the remaining
functions and ensure unicity of all config parsers.
If -f argument is a directory add all the files (and only files) it
containes to the config files list.
These files are added in lexical order (respecting LC_COLLATE).
Only files with ".cfg" extension are added.
Only non hidden files (not prefixed with ".") are added.
Symlink are followed.
The -f order is still respected:
$ tree -a rootdir
rootdir
|-- dir1
||-- .6.cfg
||-- 1.cfg
||-- 2
||-- 3.cfg
||-- 4.cfg -> 1.cfg
||-- 5 -> 1.cfg
||-- 7.cfg -> .
|`-- dir4
|`-- 8.cfg
|-- dir2
||-- 10.cfg
|`-- 9.cfg
|-- dir3
|`-- 11.cfg
|-- link -> dir3/
|-- root1
|-- root2
`-- root3
$ ./haproxy -C rootdir -f root2 -f dir2 -f root3 -f dir1 \
-f link -f root1
root2
dir2/10.cfg
dir2/9.cfg
root3
dir1/1.cfg
dir1/3.cfg
dir1/4.cfg
link/11.cfg
root1
This can be useful on systemd where you can't change the haproxy
commande line options on service reload.
int list_append_word(struct list *li, const char *str, char **err)
Append a copy of string <str> (inside a wordlist) at the end of
the list <li>.
The caller is responsible for freeing the <err> and <str> copy memory
area using free().
On failure : return 0 and <err> filled with an error message.
Released version 1.7-dev3 with the following main changes :
- MINOR: sample: Moves ARGS underlying type from 32 to 64 bits.
- BUG/MINOR: log: Don't use strftime() which can clobber timezone if chrooted
- BUILD: namespaces: fix a potential build warning in namespaces.c
- MINOR: da: Using ARG12 macro for the sample fetch and the convertor.
- DOC: add encoding to json converter example
- BUG/MINOR: conf: "listener id" expects integer, but its not checked
- DOC: Clarify tunes.vars.xxx-max-size settings
- CLEANUP: chunk: adding NULL check to chunk_dup allocation.
- CLEANUP: connection: fix double negation on memcmp()
- BUG/MEDIUM: peers: fix incorrect age in frequency counters
- BUG/MEDIUM: Fix RFC5077 resumption when more than TLS_TICKETS_NO are present
- BUG/MAJOR: Fix crash in http_get_fhdr with exactly MAX_HDR_HISTORY headers
- BUG/MINOR: lua: can't load external libraries
- BUG/MINOR: prevent the dump of uninitialized vars
- CLEANUP: map: it seems that the map were planed to be chained
- MINOR: lua: move class registration facilities
- MINOR: lua: remove some useless checks
- CLEANUP: lua: Remove two same functions
- MINOR: lua: refactor the Lua object registration
- MINOR: lua: precise message when a critical error is catched
- MINOR: lua: post initialization
- MINOR: lua: Add internal function which strip spaces
- MINOR: lua: convert field to lua type
- DOC: "addr" parameter applies to both health and agent checks
- DOC: timeout client: pointers to timeout http-request
- DOC: typo on stick-store response
- DOC: stick-table: amend paragraph blaming the loss of table upon reload
- DOC: typo: ACL subdir match
- DOC: typo: maxconn paragraph is wrong due to a wrong buffer size
- DOC: regsub: parser limitation about the inability to use closing square brackets
- DOC: typo: req.uri is now replaced by capture.req.uri
- DOC: name set-gpt0 mismatch with the expected keyword
- MINOR: http: sample fetch which returns unique-id
- MINOR: dumpstats: extract stats fields enum and names
- MINOR: dumpstats: split stats_dump_info_to_buffer() in two parts
- MINOR: dumpstats: split stats_dump_fe_stats() in two parts
- MINOR: dumpstats: split stats_dump_li_stats() in two parts
- MINOR: dumpstats: split stats_dump_sv_stats() in two parts
- MINOR: dumpstats: split stats_dump_be_stats() in two parts
- MINOR: lua: dump general info
- MINOR: lua: add class proxy
- MINOR: lua: add class server
- MINOR: lua: add class listener
- BUG/MEDIUM: stick-tables: some sample-fetch doesn't work in the connection state.
- MEDIUM: proxy: use dynamic allocation for error dumps
- CLEANUP: remove unneeded casts
- CLEANUP: uniformize last argument of malloc/calloc
- DOC: fix "needed" typo
- BUG/MINOR: dumpstats: fix write to global chunk
- BUG/MINOR: dns: inapropriate way out after a resolution timeout
- BUG/MINOR: dns: trigger a DNS query type change on resolution timeout
- CLEANUP: proto_http: few corrections for gcc warnings.
- BUG/MINOR: DNS: resolution structure change
- BUG/MINOR : allow to log cookie for tarpit and denied request
- BUG/MEDIUM: ssl: rewind the BIO when reading certificates
- OPTIM/MINOR: session: abort if possible before connecting to the backend
- DOC: http: rename the unique-id sample and add the documentation
- BUG/MEDIUM: trace.c: rdtsc() is defined in two files
- BUG/MEDIUM: channel: fix miscalculation of available buffer space (2nd try)
- BUG/MINOR: server: risk of over reading the pref_net array.
- BUG/MINOR: cfgparse: couple of small memory leaks.
- BUG/MEDIUM: sample: initialize the pointer before parse_binary call.
- DOC: fix discrepancy in the example for http-request redirect
- MINOR: acl: Add predefined METH_DELETE, METH_PUT
- CLEANUP: .gitignore cleanup
- DOC: Clarify IPv4 address / mask notation rules
- CLEANUP: fix inconsistency between fd->iocb, proto->accept and accept()
- BUG/MEDIUM: fix maxaccept computation on per-process listeners
- BUG/MINOR: listener: stop unbound listeners on startup
- BUG/MINOR: fix maxaccept computation according to the frontend process range
- TESTS: add blocksig.c to run tests with all signals blocked
- MEDIUM: unblock signals on startup.
- MINOR: filters: Print the list of existing filters during HA startup
- MINOR: filters: Typo in an error message
- MINOR: filters: Filters must define the callbacks struct during config parsing
- DOC: filters: Add filters documentation
- BUG/MEDIUM: channel: don't allow to overwrite the reserve until connected
- BUG/MEDIUM: channel: incorrect polling condition may delay event delivery
- BUG/MEDIUM: channel: fix miscalculation of available buffer space (3rd try)
- BUG/MEDIUM: log: fix risk of segfault when logging HTTP fields in TCP mode
- MINOR: Add ability for agent-check to set server maxconn
- CLEANUP: Use server_parse_maxconn_change_request for maxconn CLI updates
- MINOR: filters: add opaque data
- BUG/MEDIUM: lua: protects the upper boundary of the argument list for converters/fetches.
- MINOR: lua: migrate the argument mask to 64 bits type.
- BUG/MINOR: dumpstats: Fix the "Total bytes saved" counter in backends stats
- BUG/MINOR: log: fix a typo that would cause %HP to log <BADREQ>
- BUG/MEDIUM: http: fix incorrect reporting of server errors
- MINOR: channel: add new function channel_congested()
- BUG/MEDIUM: http: fix risk of CPU spikes with pipelined requests from dead client
- BUG/MAJOR: channel: fix miscalculation of available buffer space (4th try)
- BUG/MEDIUM: stream: ensure the SI_FL_DONT_WAKE flag is properly cleared
- BUG/MEDIUM: channel: fix inconsistent handling of 4GB-1 transfers
- BUG/MEDIUM: stats: show servers state may show an empty or incomplete result
- BUG/MEDIUM: stats: show backend may show an empty or incomplete result
- MINOR: stats: fix typo in help messages
- MINOR: stats: show stat resolvers missing in the help message
- BUG/MINOR: dns: fix DNS header definition
- BUG/MEDIUM: dns: fix alignment issue when building DNS queries
- CLEANUP: don't ignore scripts in .gitignore
- BUILD: add a few release and backport scripts in scripts/
These ones have been used for several months already and are quite
convenient to emit new releases and backport fixes. I'm fed up with
having different versions on different machines, let's commit them
now.
On some architectures, unaligned access is not authorized. On most
architectures, it is just slower. Therefore, we have to use memcpy()
when an unaligned access is needed, specifically when writing the qinfo.
Also remove the unaligned access when reading answer count when reading
the answer. It's likely that this instruction was optimized away by the
compiler since it is unneeded. Add a comment to explain why we use 7 as
an offset instead of 6. Not an unaligned offset since "resp" is
"unsigned char", then promoted to int.
Conforming to RFC 2535, section 6.1. This is not an important bug as
those fields don't seem to be set to something else than 0 and to be
checked on answers.
The help message provided by the "help" command on the unix socket didn't
provide "show stat resolvers" in the list of supported commands.
This patch should be backported to 1.6
This is the same issue as "show servers state", where the result is incorrect
it the data can't fit in one buffer. The similar fix is applied, to restart
the data processing where it stopped as buffers are sent to the client.
This fix should be backported to haproxy 1.6
It was reported that the unix socket command "show servers state" returned an
empty response while "show servers state <backend>" worked.
In fact, both cases can reproduce the issue. It happens when the response can't
fit in one buffer.
The fix consists in processing the response in several steps, as it is done in
some others commands, by restarting where it was stopped after the buffer is
sent to the client.
This fix should be backported to haproxy 1.6
In 1.4-dev3, commit 31971e5 ("[MEDIUM] add support for infinite forwarding")
made it possible to configure the lower layer to forward data indefinitely
by setting the forward size to CHN_INFINITE_FORWARD (4GB-1). By then larger
chunk sizes were not supported so there was no confusion in the usage of the
function.
Since 1.5 we support 64-bit content-lengths and chunk sizes and the function
has grown to support 64-bit arguments, though it still limits a single pass
to 32-bit quantities (what fit in the channel's to_forward field). The issue
now becomes that a 4GB-1 content-length can be confused with infinite
forwarding (in fact it's 4GB-1+what was already in the buffer). It causes a
visible effect when transferring this exact size because the transfer rate
is lower than with other sizes due in part to the disabling of the Nagle
algorithm on the sendto() call.
In theory with keep-alive it should prevent a second request from being
processed after such a transfer, but since the analysers are still present,
the forwarding analyser properly counts down the remaining size to transfer
and ultimately the transaction gets correctly reset so there is no visible
effect.
Since the root cause of the issue is an API problem (lack of distinction
between a real valid length and a magic value), this patch modifies the API
to have a new dedicated function called channel_forward_forever() to program
a permanent forwarding. The existing function __channel_forward() was modified
to properly take care of the requested sizes and ensure it 1) never overflows
and 2) never reaches CHN_INFINITE_FORWARD by accident.
It is worth noting that the function used to have a bug causing a 2GB
forward to be scheduled if it was called with less data than what is present
in buf->i. Fortunately this bug couldn't be triggered with existing code.
This fix should be backported to 1.6 and 1.5. While it also theorically
affects 1.4, it's better not to backport it there, as the risk of breaking
large object transfers due to significant API differences is high, compared
to the fact that the largest supported objects (4GB-1) are just slower to
transfer.
The previous buffer space bug has revealed an issue causing some stalled
connections to remain orphaned forever, preventing an old process from
dying. The issue is that once in a while a task may be woken up because
a disabled expiration timer has been reached despite no timeout being
reached. In this case we exit very early but the SI_FL_DONT_WAKE flag
wasn't cleared, resulting in new events not waking the task up. It may
be one of the reasons why a few people have already observed some peers
connections stuck in CLOSE_WAIT state.
This bug was introduced in 1.5-dev13 by commit 798f432 ("OPTIM: session:
don't process the whole session when only timers need a refresh"), so
the fix must be backported to 1.6 and 1.5.
Unfortunately, commit 169c470 ("BUG/MEDIUM: channel: fix miscalculation of
available buffer space (3rd try)") was still not enough to completely
address the issue. It fell into an integer comparison trap. Contrary to
expectations, chn->to_forward may also have the sign bit set when
forwarding regular data having a large content-length, resulting in
an incomplete check of the result and of the reserve because the with
to_forward very large, to_forward+o could become very small and also
the reserve could become positive again and make channel_recv_limit()
return a negative value.
One way to reproduce this situation is to transfer a large file (> 2GB)
with http-keep-alive or http-server-close, without splicing, and ensure
that the server uses content-length instead of chunks. The transfer
should stall very early after the first buffer has been transferred
to the client.
This fix now properly checks 1) for an overflow caused by summing o and
to_forward, and 2) for o+to_forward being smaller or larger than maxrw
before performing the subtract, so that all sensitive operations are
properly performed on 33-bit arithmetics.
The code was subjected again to a series of tests using inject+httpterm
scanning a wide range of object sizes (+10MB after each new request) :
$ printf "new page 1\nget 127.0.0.1:8002 / s=%%s0m\n" | \
inject64 -o 1 -u 1 -f /dev/stdin
With previous fix, the transfer would suddenly stop when reaching 2GB :
hits ^hits hits/s ^h/s bytes kB/s last errs tout htime sdht ptime
203 1 2 1 216816173354 2710202 3144892 0 0 685.0 0.0 685.0
205 2 2 2 219257283186 2706880 2441109 0 0 679.5 6.5 679.5
205 0 2 0 219257283186 2673836 0 0 0 0.0 0.0 0.0
205 0 2 0 219257283186 2641622 0 0 0 0.0 0.0 0.0
205 0 2 0 219257283186 2610174 0 0 0 0.0 0.0 0.0
Now it's fine even past 4 GB.
Many thanks to Vedran Furac for reporting this issue early with a common
access pattern helping to troubleshoot this.
This fix must be backported to 1.6 and 1.5 where the commit above was
already backported.
William Lallemand
Dragan Dosen
Roberto Guimaraes
Thierry Fournier
Willy Tarreau
William Lallemand
Cyril Bont
Lukas Tribus
Nenad Merdanovic
Vincent Bernat
Maxime de Roucy
Christopher Faulet