d-komarov
6d5a6f05ad
Fix torrent files prioritization ( #3133 )
...
After setting file priority, any subsequent attempt to set different priorities will fail if there is a `file_priority` job running in in disk thread. This happens because `torrent::m_file_priority` is being updated before adding disk thread job. The problem is gone if the file priority vector owned by the torrent object, is modified in the `torrent::on_file_priority` callback, when disk job finishes.
2018-07-04 08:16:33 +02:00
Steven Siloti
b7f230316c
set the minimum number of checking jobs based on the number of hasher threads
...
To effectively mask the latency of generating more hash jobs there need to be
at least two jobs in-flight for each hasher thread.
2018-06-23 23:10:55 +02:00
arvidn
1e3a7cf0f4
fix hash-job fast-path
2018-06-23 23:08:58 +02:00
arvidn
0c147164bc
introduce a fast-path for the hash disk job
2018-06-20 01:00:16 +02:00
Arvid Norberg
794ccf4f79
remove unused disk stats counters
2018-06-05 19:03:04 +02:00
arvidn
c55bc7dd42
improve error handling of failing to change file priority
2018-05-31 00:47:17 +02:00
arvidn
299a545648
add test for suggests- and seed mode combined
2018-05-10 17:40:23 +02:00
Steven Siloti
f5e33932d2
fix use after free in flush_range and flush_iovec
...
Calling blocks_flushed can cause the piece entry to be freed so its
callers need to be aware of that and avoid dereferencing the pointer if
the entry is freed.
2018-04-27 16:15:19 +02:00
Xiyue Deng
21d8e9e26b
Add missing header <cstdarg> for va_list.
...
* Fix building on OpenBSD.
2018-04-24 23:09:25 -04:00
arvidn
aaf9304a3b
update year in copyright header
2018-04-09 09:04:33 +02:00
Steven Siloti
621da10e60
hold an owning reference to storage objects in try_flush_write_blocks
...
It is possible for all other references to a storage object to be
destroyed while try_flush_write_blocks is running. If the storage is
destroyed then find_piece will crash when trying to re-aquire the
shared_ptr. To prevent this, keep the storage alive by holding a
shared_ptr to it in try_flush_write_blocks.
Normally the fence job when stopping a torrent would prevent the storage
object from being destroyed until all flush jobs are complete.
try_flush_write_blocks can be run after every disk job though so it has
the potential to "stradle the fence".
If the associated torrent does get unloaded then it is expected that
find_piece will return NULL thus causing the entry to be ignored.
2017-08-30 12:45:06 +02:00
arvidn
227830e757
fix memory leak in the disk cache. if a cached_piece_entry would stick around in a ghost list (ARC), it would keep the torrent object itself alive. when aborting read jobs, the outstanding_read flag need to be cleared on the piece. When reclaiming a block for a piece that should be deleted (not just evicted) it need to know that. This patch adds an additional bit to cached_piece_entry to communicate this
2017-04-19 22:30:27 -04:00
arvidn
2d6268b580
fix double free in disk cache
2017-04-17 19:54:05 -04:00
arvidn
dec8c853e3
fix default file pool size on windows
2017-03-09 23:26:53 -05:00
arvidn
5cb12318c5
fix bug where settings_pack::file_pool_size setting was not being honored
2017-03-09 08:18:10 -05:00
Arvid Norberg
f5c7a49d61
remove incorrect deleted-storages assert and add some more test coverage in test_metadata_extensions ( #1286 )
2016-11-03 19:39:42 -04:00
Arvid Norberg
d1a59879cb
deprecate mmap_cache ( #1269 )
...
deprecated mmap_cache feature
2016-10-29 13:42:42 -04:00
Steven Siloti
cf7db392b6
assert that offset didn't change while hashing
2016-10-25 07:59:02 -04:00
Steven Siloti
f36b9a805a
avoid concurrent access to hash offset
...
Since partial_hash::offset is not an atomic it is technically UD to access it
concurrently from multiple threads.
2016-10-25 07:59:02 -04:00
Steven Siloti
241e9fd25a
add assert to detect index out-of-range
2016-10-25 07:59:02 -04:00
Arvid Norberg
5f7e2eb928
run undefined sanitier on travis/linux ( #1134 )
...
run undefined sanitizer on travis/linux. back-port ed25519 unit test from master
2016-10-08 12:29:46 -04:00
arvidn
0095aa084f
make coalesce reads and coalesce writes actually work
2016-03-20 16:34:19 -04:00
arvidn
f7c9062019
deprecate more parts of explicit cache as well as fixing typos of TORRENT_NO_DEPRECATE preprocessor tests
2016-03-19 07:54:29 -04:00
arvidn
317eb2fe2a
remove unused disk allocation code
2016-03-16 20:13:28 -04:00
Arvid Norberg
a1d60db706
Merge pull request #547 from arvidn/no-read-cache-1.1
...
fix use_read_cache=false
2016-03-16 08:14:15 -04:00
arvidn
4926081960
fix use_read_cache=false
2016-03-16 02:22:18 -04:00
arvidn
08bac479be
allow each peer have at least 2 allocated disk blocks at any given time, to avoid stalling when cache_size setting is small. also deprecate use_write_cache
2016-03-16 00:28:11 -04:00
Steven Siloti
4331f57021
restrict check_cache_level to only runing on one thread at a time
...
If another thread trys to call check_cache_level then m_cache_check_state is set
to indicate that the thread currently in check_cache_level needs to call it again.
2016-03-14 19:03:12 -07:00
Steven Siloti
d405535a3f
try to free the piece after kicking the hasher and flushing
2016-03-14 19:03:12 -07:00
Steven Siloti
e8f93df700
fix indentation
2016-03-14 19:03:12 -07:00
Steven Siloti
826bbe35ee
always cache writes
2016-03-14 19:03:12 -07:00
Steven Siloti
6e8db41dd2
check_cache_level after performing a job rather than before
...
If dirty blocks are added while the job is running there is a chance of
deadlock if the dirty blocks cause the cache to exceed its size limit and
no further jobs come in to trigger another check.
2016-03-14 19:03:12 -07:00
Steven Siloti
9c5bb25d65
always clear outstanding_flush after attempting to flush a cached write
...
Also clear outstanding_flush before calling kick_hasher because kick_hasher
may unlock the cache mutex which can allow more dirty blocks to be added
which may require another flush_hashed job be added.
2016-03-14 19:03:12 -07:00
arvidn
921cbeebed
add option to delete just the partfile when removing a torrent
2016-03-14 08:26:39 -04:00
arvidn
0d0048d415
try to evict a volatile piece before allocating a new one for a hash check. It may be faster to check files without growing the cache, or reusing the same buffers.
2016-03-13 12:52:44 -04:00
arvidn
67774d660d
always keep at least 2 async hash jobs outstanding when rechecking torrents (to keep the bandwidth delay product full)
2016-03-06 16:31:18 -05:00
arvidn
a02c1d2164
record cache settings in the job instead of checking it twice for more robust behavior
2016-03-06 16:15:13 -05:00
arvidn
8ca76f9d59
fix support for auto disk cache size and disk buffer pool allocator behavior
2016-03-06 16:14:55 -05:00
arvidn
2bd6ad9974
remove invalid assert in the disk thread
2016-03-04 20:20:49 -05:00
arvidn
1f9f588e75
merge copyright year update and changelog from RC_1_0
2016-01-17 18:57:46 -05:00
arvidn
4cc2e879c4
comment out verbose pex logging. fix bug in disabled_storage and an error reporting issue on short reads in default_storage
2016-01-10 11:50:56 -05:00
arvidn
2acb3dcb24
clean up use of getrlimit() by wrapping it and move it to platform_util.cpp. Also take the opportunity to make it simulator friendly (consistent in simulation)
2015-12-02 00:45:34 -05:00
arvidn
b628692223
separate out announce_entry and torrent_status into their own headers
2015-09-18 08:08:08 -04:00
arvidn
17c5759829
disable warning for unused command line argument and fix warnings. remove redundant travis build config. use new preprocessor on clang to avoid warnings
2015-09-06 22:55:41 -04:00
arvidn
904da120d7
fixed some warnings
2015-08-20 22:28:00 +02:00
arvidn
8cf8e65861
turn linked_list into a template to improve type safety and remove some casts
2015-08-20 02:02:46 +02:00
arvidn
ffa870d280
fixed some GCC warnings
2015-08-20 01:33:20 +02:00
arvidn
bd177a857f
make tailequeue a template, introduce type safety, remove old-style casts
2015-08-19 15:22:05 +02:00
arvidn
5c8fc818c1
fix more warnings
2015-08-19 01:39:01 +02:00
arvidn
12da02ae52
fix some warnings
2015-08-18 09:33:00 -04:00