83ca755d1b
simplify setup_listener
2017-07-26 12:02:58 -07:00
cd343c91bc
make base32encode flags type safe
2017-07-26 10:38:14 -07:00
b38efb6c65
make alert category flags be a strong type
2017-07-26 07:20:47 -07:00
c167e28788
move stack allocator implementation into its own cpp file
2017-07-23 16:49:24 -07:00
330904f460
convert picker_flags_t to type safe flags
2017-07-23 09:29:25 -07:00
78bbd298a5
convert file_open_mode to type safe flags
2017-07-22 23:58:24 -07:00
140b8ace8d
remove use of deprecated function readdir_r
2017-07-22 22:40:17 -07:00
c4eb4c8b5f
use span in torrent_info constructors and deprecate the redundant flags argument
2017-07-22 00:00:44 -07:00
5eba09fce7
changes based on review feedback
2017-07-21 19:18:44 -07:00
1a64490ef0
fix
2017-07-21 19:18:44 -07:00
f0ae2145ea
use a handle class for listen sockets instead of raw pointers
...
Listen socket handles need to be week_ptr so that they can detect when a listen
socket is removed in the middle of a tracker announce on it. Use a wrapper class
to enforce that no one outside of session_impl can hold ownership of the socket
across event boundaries.
To support unit tests, listen_socket_t is also split into a base class which
holds the members needed to support the public listen socket API.
2017-07-21 19:18:44 -07:00
e0dcd34352
convert m_listen_sockets to a vector of shared_ptr
2017-07-21 19:18:43 -07:00
0f30815d8f
Revert "don't move listen_socket_t when deleting sockets"
...
This reverts commit 718c5f0dcb
2017-07-21 19:18:43 -07:00
87ca4bb26f
Fix bug where the resume data would fail to load the piece bitmask for seeds when suggest_cache was enabled
2017-07-21 19:06:02 -07:00
33a2abebfe
submit disk jobs in read_piece()
2017-07-18 07:21:07 -07:00
f9c45db942
make bandwidth state flags use type-safe flags
2017-07-17 19:03:42 -07:00
06070ea499
convert block request flags to type-safe flags
2017-07-17 17:56:07 -07:00
ea8c5b32da
add reserve entry::to_string() ( #2148 )
2017-07-17 15:05:58 -07:00
c4659bc345
convert peer_flags and peer_source_flags to type-safe flags
2017-07-17 10:22:32 -07:00
4947602a2f
make add_torrent_params::flags_t an enum class and move it out into its own header
2017-07-17 01:05:31 -07:00
00655d562b
Refactor several flags to torrent_handle::get_flags/set_flags.
2017-07-17 01:05:31 -07:00
73942de97a
carve out peer_list_entry into also being deprecated (it was only used by deprecated functions and some tests)
2017-07-16 19:20:29 -07:00
b1c3e12cd0
added block_uploaded_alert to allow client to track upload activity
2017-07-16 11:04:03 -07:00
c4afb5e949
use a dedicated class template for flags types instead of enum class. It's more backwards compatible and improves ergonomics
2017-07-15 01:15:57 -07:00
c692147c94
merged RC_1_1 into master
2017-07-13 19:55:29 -07:00
cc30434c52
fix inconsistency in file_priorities and override_resume_data behavior. file_priorities are not subject to the override_resume_data flag
2017-07-12 15:36:55 -07:00
a422af1a7b
remove call _strchr
...
_TEXT SEGMENT
_c$ = 8 ; size = 1
?is_space@@YA_ND@Z PROC ; is_space, COMDAT
; 8 : static const char* ws = " \t\n\r\f\v";
; 9 : return strchr(ws, c) != nullptr;
movsx eax, BYTE PTR _c$[esp-4]
push eax
push DWORD PTR ?ws@?1??is_space@@YA_ND@Z@4PBDB
call _strchr
add esp, 8
neg eax
sbb eax, eax
neg eax
; 10 : }
2017-07-12 15:34:45 -07:00
85cf521145
fix bandwith rate limit calculation ( #2134 )
...
back-port of fc0cbfb789
2017-07-11 21:35:12 -07:00
2e79c5e648
fix handling of SSL listen sockets
2017-07-10 21:41:39 -07:00
11d6a00bec
fix storage destruction order issue ( #2138 )
...
back-ported from b553cb32f7
2017-07-10 21:35:34 -07:00
389b4eb3f2
make listen_socket_t non-copyable and non-movable ( #2137 )
...
The DHT stores a pointer to this struct so it must not be copied or moved once
it is stored in m_listen_sockets.
2017-07-10 10:14:36 -07:00
718c5f0dcb
don't move listen_socket_t when deleting sockets
...
Pointers to listen_socket_t are stored in dht_tracker, so they
cannot be copied or moved once they are added to m_listen_sockets.
listen_socket_t should really be made non-copyable and non-moveable, but the
socket creation path needs to be refactored before that can be done.
2017-07-08 08:42:04 -07:00
80036f0377
avoid executing timed async task if the dht node is already removed ( #2133 )
2017-07-08 08:32:24 -07:00
10bfe18ebe
added consts and minor code formatting
2017-07-03 16:58:49 -04:00
285daa21a5
fixed another shorten-64-to-32 warning in netlink code
2017-07-03 13:53:15 -04:00
ec2fb91aa5
reject put messages with incorrect bencoding ( #2118 )
...
add function to check for soft bdecode errors. reject put messages with incorrect bencoding
2017-07-02 18:30:32 -04:00
5188c190db
bump version number
2017-07-02 15:03:46 -04:00
5c37eb174f
drop support for windows compilers without std::string and wchar_t (believed to be old cygwin)
2017-07-02 08:24:55 -04:00
7af38ba05a
add const and static
2017-06-30 16:07:23 -04:00
92a923134c
fix generate_token
2017-06-30 13:22:30 -04:00
db09332a02
merged RC_1_1 into master
2017-06-30 08:58:21 -04:00
20c1407612
corrected missing const qualifiers on bdecode_node
2017-06-30 08:19:10 -04:00
a78ec94184
fixed compilation warnings in enum_net netlink code
2017-06-29 15:48:28 -04:00
3c8450d47c
fix backwards compatibility issue when loading the torrent info dict from resume data
2017-06-29 15:03:49 -04:00
506950001d
fix regression where paused torrents could not have their queue position changed
2017-06-27 12:37:08 -04:00
20e05a12ac
broadcast_socket: remove obsolete special case
...
enum_net_interfaces now fills in the scope_id correctly so there's no need for
a special case to set it here.
2017-06-27 08:09:20 -04:00
bbf361336b
Use netlink to enumerate network interfaces on linux ( #2105 )
...
* move getting a table dump over netlink to a separate function
* use netlink to enumerate network interfaces on linux
2017-06-27 00:34:24 -04:00
0a9e2c965d
invert logic in read_nl_sock for better clarity and reliability ( #2101 )
...
The seq and pid parameter are apparently used as a safety check in case
a stream of netlink messages is not properly terminated. I initially thought
that they represented the expected values of the incoming messages.
Instead seq is set to one-past the expected value and the loop aborts when
seq and pid match the message.
Inverting the logic so that reading continues as long as the seq and pid match
the incoming messages is, to me at least, more intuitive. It is also more
reliable since it does not rely on the next seq being strictly sequential. It
also catches unexpected messages preceding or interleaved with the expected
messages.
There seems to be a lot of confusion all across the internet over the
nlmsg_pid field. The changes here were made based on examination of the
kernel source and iproute2 (a common user of netlink).
The kernel does not read nlmsg_pid at all for routing messages.
Iproute2 sets it to zero unconditionally so I changed this code to do the same.
2017-06-25 10:29:32 -04:00
2b91b1070d
merged RC_1_1 into master
2017-06-24 12:37:13 -04:00
ec30a5e9ec
fix out-of-bounds read in bdecode
...
Fixes #2099
2017-06-24 10:52:49 -04:00
arvidn