bytemark/flexnbd-c
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Pull back the changelog generation to the simplest thing that can possibly work

Browse Source
This commit is contained in:
Alex Young
2014-02-25 17:24:25 +00:00
parent 2fa1ce8e6b
commit ffa45879d7
3 changed files with 7 additions and 579 deletions
Download Patch File Download Diff File Expand all files Collapse all files

457
debian/changelog vendored
View File

@@ -1,457 +0,0 @@
flexnbd (0.1.1-435) stable; urgency=low
* proxy: Turn on TCP_CORK
* proxy: Turn on TCP_NODELAY for the proxy->upstream leg
* client: cork/uncork around NBD_REQUEST_READ responses
* sockutil: Add a tcp_cork helper
* Remove a compile-time optional selection of O_DIRECT (was never used)
* Call srand() to make sure request handles are properly randomised
* Merge
* Add some build-depends which make doc needs
* Remove unreachable code to make -Wunreachable-code on clang useful.
* Depend on either libev4 or libev3, whichever is available
* Add asciidoc as a Build-Depends
* Update Rakefile to generate debian/changelog.
-- nick <nick@bytemark.co.uk> Tue, 25 Feb 2014 16:00:48 +0000
flexnbd (0.1.1-422) stable; urgency=low
* Update the Makefile to work with dpkg-buildpackage
* Switch from a rake-based build to a make-based build.
* proxy: Fix a read corruption issue caused by us failing to reset needles on timeout
* client: Add a cork TODO pair
* client: Start checking for exceptions on the client socket
* Some debug and whitespace tweaks
* flexnbd client: Catch a few cases where the killswitch wasn't disarmed
* Increase a timeout on a test to handle slow unlink calls on other filesystems
* Fix stop signal logic broken by the killswitch
* Fix up the check_util test once more
* Fix a few dangling bits in client.h
* Automated merge with ssh://dev.bytemark.co.uk//repos/flexnbd-c
* flexnbd serve: Make the killswitch per-client-thread
* Rework a test
* Another fedora build fix
* Fix compilation on fedora
* Fix a test case
* Fix help_text errors for break and status modes
* tests: Fix an uninitialized memory access
* mirror: lengthen the request timeout to 60 seconds
* acl: Make some compilers happy
* mirror: Couple of tiny cleanups
* mirror: Turn off the 'begin' timer before continuing
* Increase loglevel of some allocation map messages
* mirror: Clean up bps calculation slightly
* mirror: Ensure the bitset is actually disabled on mirror error
* mirror: Don't start migrating until the allocation map is built
* mirror: Don't count bytes transferred for the purposes of keeping the stream empty as part of our bwlimit
* mirror: Don't sleep if our stream is filling up
* bitset: Swap pthread_cond_broadcast for pthread_cond_signal
* bitset: More-efficient implementation of bitset_stream_queued_bytes
* mirror: Reduce the mirror convergence window to 5 seonds, from 60
-- Alex Young <alex@bytemark.co.uk> Fri, 21 Feb 2014 19:39:27 +0000
flexnbd (0.1.0-389) stable; urgency=low
* tests: Add a migration test with many clients connecting in two waves
* Fix a current compiler warning
* Fix a potential compiler warning on 32-bit
* bitset: Rename bitset_stream_on/off as bitset_enable/disable_stream
* serve: Fix calulation of server_mirror_bytes_remaining
* bitset: Rename bitset_mapping to bitset
* serve: Remove an unused (and incorrect, in any case) function
* serve: Fix the sense of allow/forbid_new_clients
* mirror: Remove dead code. We still rely on all_dirty in one place.
* status: Remove useless stats
* status: Expose migration_seconds_left
* serve/mirror: Move some code tracking migration speed into serve
* status: Display number of currently connected clients, and whether new clients are allowed
* serve: Allow number of clients currently being used to be counted
* serve: Add a comment clarifying that a behaviour is safe
* flexnbd: Remove unused ".INCOMPLETE" file code
* flexnbd: Remove the server I/O lock as it no longer has any consumers
* flexnbd: Remove the server I/O lock from around NBD requests
* Fix a warning in a test
* mirror: Fix mirroring, break status
* tests: Fix for the previous commit
* flexnbd: Make a test a bit stricter
* mirror: Remove server I/O lock and dirty map
* mirror: Fix a comment
* serve: Move bitset freeing to after closing the mirror and clients
* flexnbd: Some comments and a minor fix in client.c to do with the event stream
* bitset: add bitset_stream_size and bitset_stream_queued_bytes
* bitset: Add an event stream implementation
* tests: Update check_bitset to use new bitset_free() function
* bitset: Some whitespace changes
* bitset: Add a bitset_free() function
* flexnbd: Disconnect clients at the start of a mirror last pass
* mirror: Ensure the mirror client socket is closed after a fail, and before a retry
* mirror: Remove overly-verbose log line
* mirror: honour max_bytes_per_second - naive scheme
* listen: Turn off CLIENT_MAX_WAIT_SECS
* mirror: Start our timeout watcher from the first, not second, transfer
* flexnbd: add a mirror-speed command to change mirror->max_bytes_per_second
* flexnbd status: display mirror->max_bytes_per_second as mirror_speed_limit
* One more fix
* control/mirror: Use uint64_t and strtoull to get max_Bps into the mirror
* control: Output abandoned mirror state
* mirror: And another abandon fix
* mirror: fix abandon
* mirror: abandon signals are now honoured outside of the remote end being readable / writable
* Branch merge
* Backed out changeset e58ff57b5e2d
* Branch merge
* listen: Allow longer gaps between transfers
* Minor fixes here and there
* mirror: Use libev to provide an event loop inside the mirror thread
* bitset: Add a bitset_run_count_ex that lets you learn the value of the bits in the run
* bitset: Prove that bitset operations with len=0 don't underflow
* status: Add migration_speed ( bytes per second ) and migration_duration( seconds ) to the migration output
* bitset: squash one more bug
* Remove a stray debugging statement
* bitset: Fix large runs
* flexnbd-proxy: ensure upstream cooldown is applied when read init from upstream fails
* flexnbd: Acknowledge new return type of bitset_run_count
* flexnbd: clients should be MADV_RANDOM, rather than MADV_SEQUENTIAL
* non-debug builds get -O2 for impressive bitset speedups
* bitset: A more-efficient bit(set)_run_count
* bitset: Actually enable an optimization in bit_set/clear_range
* bitset: Use uint64_t everywhere to avoid possible integer overflows
* tests: Fix a couple of compile warnings
* test: Add some tests for bitset_run_count
* flexnbd mirror: If the final run would be longer than the file size, truncate to file size
* flexnbd status: Actually output pass statistics
* flexnbd status: Add current pass clean/dirty byte statistics
* flexnbd status: Avoid a possible NULL dereference reading migration status
* flexnbd status: Add the size of the backing file, in bytes
* flexnbd status: Add current migration pass to the status output if we're migrating
* Fix tests for new killswitch argument
* flexnbd: Make the killswitch runtime-selectable
* proxy: Reduce the reconnect cooldown from 15 seconds to 3.
* proxy: Add a 30-second timeout for requests in-flight to upstream
* proxy: Fix a comment
* proxy: Fix the prefetch code
* proxy: Respect the REQUEST_MASK
* ioutil: Add a bit more debug output to iobuf_read/write
* proxy: The minor optimisation bugs if needle is not advanced on iobuf_read()
* proxy: DRY up some code
* proxy: Set proxy->upstream_fd before calling proxy_finish_connect_to_upstream
* proxy: Initial move to event-loop proxy model.
* readwrite: Expose a couple of points of functionality
* tests: Tell us which offset fails
* proxy: log details of a request that fails upstream at the warn level
* serve: Add a killswitch that causes the server to uncleanly exit on hang
* proxy: Some logging cleanups
* docs: Fix the documentation for the loglevel timestamps
* logging: Add a timestamp to the log messages we emit
* mirror: munmap() our range on cleanup
* mirror: Don't undo the MADV_SEQUENTIAL hinting over the course of a migration
* mirror: Only say we're unlinking the file if we actually are
* flexnbd: Add more madvise() hints, both for mirroring out and normal operation.
* Formatting fixup
* Automated merge with ssh://dev/flexnbd-c
* flexnbd: Pass MS_INVALIDATE to our msync calls
* Add a trivial read buffer to flexnbd-proxy.
* flexnbd-proxy: Add UNIX socket support for the listen address
* readwrite.c: Set TCP_NODELAY on our NBD client sockets
* flexnbd: Don't bind() unless a bind address is specified
* sockutil: Make sockaddr_address_string conform to its comment
* proxy: Switch to blocking I/O with signal handlers to exit.
* flexnbd: Split the proxy mode out into its own binary.
* proxy: Deal with close() failures (and EINTR errnos) comprehensively
* Add an explanatory comment in sock_try_connect()
* Turn a couple of FIXME fatals in readwrite.c into warnings
* proxy: Fix the return value of a function to match the comment
* Dummy commit to get past a merge commit
* Automated merge with file:///home/lupine/Development/bigv-repos/flexnbd-c-sockutil
* flexnbd: Add a proxy mode
* serve: Make some error conditions non-fatal, test them.
* Fix some sockutil tests
* serve: Don't die horribly in the event of EINTR being returned by select()
* flexnbd: Fix sock_try_bind so we don't retry on EADDRINUSE
* flexnbd read/write: Switch to a non-blocking connect() to allow us to time these out
* Make acceptance tests verbose by default
* Automated merge with ssh://dev/flexnbd-c
* branch merge
* flexnbd: Add valgrind suppressions for a bug in glibc-2.11
* sockutil: Add some tests for sockaddr_address_string
* serve: Refactor some socket utility code into its own module.
* flexnbd: Remove some obsolete 'rebind' options
* flexnbd: Add some more information to nbdtypes.h
* Introduce socket_nbd_write_hello() and a macro to display errno results nicely
* Add a debug message for cases where sendfile() fails
* flexnbd: Avoid a SIGSEGV when the allocation map fails to build.
* Automated merge with file:///home/zander/00-projects/17-bigv/new-trial/src/incoming/flexnbd-c
* Add an all-debug task to the makefile
* flexnbd: Allocate the right amount of memory for a struct client
* Add a minimal Makefile that lets 'make' and 'make clean' do the Right Thing
* flexnbd: Normalise some variable declarations
* flexnbd: constantize the quiet log level
* madvise after mirroring to control the RSS
* Remove an errant debug output from test_happy_path.rb
* Don't skip parts of a file when calling fiemap
* Fix the error message for flexnbd write --help
* Allocate the bitset in the foreground thread.
* Don't open the control socket until after the server socket is bound
* Merge
* Fixed build_allocation_map in ioutil.c to correctly traverse fiemaps where
* Rename serve->has_control to serve->success.
* Merge
* Nullcheck *before* dereferencing.
* Block allocation map is now built in a separate thread, and does not delay
* Update Rakefile to support locally-installed libcheck, removed efence, pushed
* Backed out changeset c25e7d82e56e
* Merge
* flexnbd: Move building the allocation map to before server socket bind()
* Whitespace
* Quit with an error status on SIGTERM during migration
* Remove an unneeded sanity check from the tests
* Add the write-during-migration test to the acceptance test run
* Fix for deadlocks when writing while migrating
* Add a test case that tickles a deadlock bug when migrating active source discs
* Merge
* Fix some races in the acceptance tests
* Ensure fiemap ioctl calls are synchronous.
* Add a -dbg package to the debian build
* Add default empty LDFLAGS
* Don't free a client which hasn't finished yet.
* Don't close the same file descriptor more than once.
* Further merge fail fix
* Fix a merge fail
* Fix tests broken by entrust removal
* Add the --unlink option to mirror
* Simplify the migration handover protocol
* Update the README for new `listen` behaviour
* Remove listen mode
* Include strerror(errno) in stat debug output
* Add .INCOMPLETE file marker to flexnbd listen
* Force gzipping the man page
* Added the `flexnbd break` command to stop mirroring
* Make an EADDRINUSE on server bind fatal.
* Add the pid to the status output
* Test setting an ACL
* Block a second mirror attempt
* Better error handling for remotes
-- nick <nick@bytemark.co.uk> Tue, 24 Sep 2013 10:11:40 +0100
flexnbd (0.0.2-209) stable; urgency=low
* Add README.txt to the deb task code files
* Flush bad write data from the client socket.
* Switch from expecting a reconnection to *not* doing do
* Move the mirror commit state mbox to struct control
* Don't retry a mirror which failed on the first attempt
* Switch the mirror commit_signal to an mbox
* Delete the MS_FINALISE mirror state
* Close the mirror client socket on rejection
* Simplify acceptance test launching
* Tweak help output for the --bind option
* Add a rake task to build the man page
* Added a README.txt and a man page
* Raise the log level of a write-request-out-of-range
* Only copy constants from C to Ruby once
* Make the --verbose flag universal
* Clobber a dangling fprintf
* Error when reading sent data fails
* Open files with O_DIRECT dependent on a compile-time DIRECT_IO #define.
* Append the CFLAGS environment variable to the build flags.
* Make test sockets less dependent on enviroment
* Unlink the control socket on clean shutdown
* Tweak the fuzz script to work with the new test layout
* Tweak the parse_acl declaration
* Add a newline to log messages at macro expansion
* Switch to use nbd_r2h_reply in read_reply()
* Silence a vfprintf valgrind error
* Free the flexnbd switch mutex
* Add a 'just in case' error case to acl checking
* Remove unused variables
* Enable writing after the 2G boundary
* Rename struct mirror_status to struct mirror
* Add a --quiet command-line option
* Retry failed rebind attempts
* Fix the broken --rebind-port command-line option.
* Minor tidy
* Handle a failed disconnect correctly
* Test a source crashing after an entrust.
* Fix a test broken by stdout/stderr reshuffle
* Free all possibly held mutexes in error handlers
* Test that a blocked entrust causes a retry
* Test that a write reply with the wrong magic will force a retry
* Test that closing the socket immediately after sending write data causes an error
* Test that a disconnect after the write request but before the data is an error
* Test the source handles a dest crash after write correctly
* Refactor FakeSource from a module to a class
* Simplified FlexNBD::FakeDest
* Rename a test class
* Split acceptance tests into separate files
* Moved acceptance tests into tests/acceptance
* Moved unit tests into tests/unit
* Changes to error severity in readwrite.c made a test fail, this patch fixes it
* Test that an invalid write gets an error response
* Check that a mirror write returning an error will cause a reconnect and retry
* Tweak a timeout to prevent an intermittent test failure
* Test that timing out a write causes a disconnect and a reconnect
* Test for acl rejection
* Handle ECONNRESET during a read request
* Factor common code out of fake destinations
* Factor common code out of the test fake sources
* Add a test for a second client connecting during a mirror
* Missing file
* check_acl correctly sets log_level
* make sure that an invalid flexnbd signal fd can't break the serve accept loop
* Add mboxes
* connect failure scenarios
* Make non-fatal errors return properly
* Add listen mode
* Make max_nbd_clients configurable per struct server
* Add the REQUEST_ENTRUST nbd request type
* Squash valgrind errors by making sure client threads get joined on termination
* Make sure filename_incomplete gets freed
* Add a missed free()
* Make sure all the lines we read get freed (including the trailing blank)
* Avoid a use-after-free in serve.c
* Trim the length bitset_run_count looks at not to exceed the bits array
* Squash a valgrind error caused by debug output
* Merge
* Added early-exit on any valgrind error
* Tweak struct initialisation to squash a valgrind error
* Reject mirroring if the remote size doesn't match the local size
* Tidy mirror_runner somewhat
* Fix an argv array reference (root cause of a bug from the last commit)
* Fix two bugs in mirroring.
* Take _GNU_SOURCE out of source and put it in CFLAGS
* Removed proxying completely and fixed the pthread_join bug revealed in the process
* Fix broken error checking around pthread functions
* Simplify a FATAL_IF_NEGATIVE
* Make the error and fatal functions swallow semicolons properly
* Make the error and fatal macros swallow semicolons properly
* Add general-purpose ERROR/FATAL_IF and ERROR/FATAL_UNLESS macros
* Merge
* Automated merge with ssh://dev/flexnbd-c
* test: make check_serve bind() its outgoing socket to a known IP for these tests
* Set default log level to warn to shut the tests up
* Make sure all ifs are braced
* Make the compiler stricter and tidy up code to make the subsequent errors and warnings go away
* Merge
* Whitespace
* Fix handling ACLs where > 1 entry exists
* Explicitly check for which fd is acceptable in server_accept
* tests: Get rid of a warning
* tests: Get check_serve working after the merge of doom
* Merge, just renaming old error macros.
* Rewrote error & log functions to be more general, use longjmp to get out of
* Added hopeful default path to find rake_utils, turned undefined function
* Audit client connections on acl update
* Lock around acl updates
* Move updating the acl object into serve.c
* Fix O_NONBLOCK setting on self_pipes
* Pull ACLs into their own struct
* Tidy the NULLCHECK macro to swallow semicolons properly
* Fix an invalid define symbol
* Merge of doom
* Big bit of debug output in write_not_zeroes (disabled).
* Fixed bug in bitset_run_count which was causing data corruptionn writing
* Added some record-keeping / printing to fuzzer to assist with backtracking.
* Refactored write_not_zeroes to use struct bitset_mapping instead of
* Improved fuzz test to find an actual code bug (previous bug was in the test
* Fixed arguments to debug function.
* Missing break; in switch statement (verbose was setting default deny!)
* Use client stop signals for thread stopping
* Add stop signals to client threads
* Fix the sense of client_serve_request
* Remove the accept lock as being unneeded
* Merge
* Move checking for a closed client out of server_lock_io and into client_serve_request
* Change io lock and unlock to server error on failure
* Better error message
* Use self_pipe for close_signal
* tests: fix the Ruby flexnbd wrapper for mirror
* Automated merge with ssh://dev/flexnbd-c
* mirror: Add --bind to our mirror mode.
* flexnbd: Fix specifying -d as --default-deny on the command line
* Remove a duplicated line due to the last merge
* Automated merge with ssh://dev/flexnbd-c
* flexnbd: Re-add --sock to flexnbd mirror
* flexnbd: Add --bind to flexnbd read and flexnbd write
* Renamed struct client_params to struct client
* Split client-specific code into client.{c,h}
* Fix a \#define symbol
* Moved sockaddr_address_data to serve.c and renamed params.h to serve.h
* Rename control to control_fd and struct mode_serve_params to struct server
* Merge
* Backed out changeset 0cbb8e9cf515 because it breaks deb packaging.
* Merge
* Rename int server to int server_fd in mode_serve_params
* Added fuzzer which currently exposes ugly bug with unaligned writes.
* Fixed FlexNBD#serve parameters, added detection of non-starting server.
* Added msync() call after every write - not sure whether it's necessary yet.
* Isolated missing library code to pkg:deb task - couldn't locate library code
* Rename the 'debug' cli option 'verbose' and switch default-deny from 'D' to 'd'
* Make sure the -d short option is honoured
* Merge
* Add a self_pipe set of convenience functions
* Indentation fix
* Automated merge with ssh://dev/flexnbd-c
* acls: Add a default-deny option, which allows you to specify what an empty ACL means.
* Use nbdtypes to write the nbd hello message
* Null-terminated strings strike again
* Factor out the bulk of client_serve_request, and add convenience converters in src/nbdtypes.c
* Don't check for the INCOMPLETE file on read
* Ditch a couple of unneeded variables to silence gcc warnings
* build and default rake tasks, because I keep trying to type them
* Add .orig merge files to .hgignore
* Merge
* Add a no-op debug() define for non-debug builds and make valgrind optional in nbd_scenarios
* Merge
* Make the Rakefile take note of DEBUG=
* Set TCP_NODELAY on our socket. This decreases average NBD read request RTT from 0.3ms to 0.001ms
* Add a --debug flag for DEBUG builds
* Run the nbd_scenario tests under valgrind
* Merge
* Fixed short copies of struct sockaddr (it's shorter than sockaddr_in6!)
* Make sure we build arch-specific packages
* Basic debian packaging
* Simplify option definition with som handy macros
* Merge
* Added getopt_long command-line handling.
* Added .h files to the Rakefile
* Rearranged the project to have src/ and build/ directories
* Fixed bug where ACL was accidentally deleted when being set from control
* Fix the usage message
* Fixed race in tests.
* Added .INCOMPLETE hack to aid with marking finished transfers.
* Added mirror write barrier / final pass stuff & clean exit afterwards.
* Lots of errors spotted by Alex fixed, added mutexes to accept & I/O, added
* Added another write/read test, fixed bugs in splice() usage and IPv6
* First few external tests with test/unit, some minor tidying of internal data
* Pulled some duplicated code out of control.c into
* Split control-socket functions into separate file.
* Fixed mirroring to work (error reporting suspect though).
* Initial, untested mirror implementation and resolved some type confusion
* More valgrind-found bugs, extracted open_and_mmap from main code.
* Fixed some uninitialised variables courtesy of valgrind.
* Mostly finished bitset tests, fixed test build to include utilities, remove
* Tweaks to bitset.h, established a C test framework.
* Fixed segfaulting access control, allowed change to acl via control socket.
* Added dummy control socket answering / changed serve_accept_loop to use
* Added control socket, doesn't do anything yet.
* Added .hgignore file
* Stopped NBD writes from committing all-zero blocks to disc (tentative, needs
* Split code out into separate compilation units (first pass, anyway).
* Non-functioning commit, half-way through adding sparse bitmap feature.
* Added write mode.
* Added working read via splice syscall.
* Added Rakefile
* Silly bug fixes, added ACL support, added parser for read/write requests.
* Some debugging, got it to serve.
-- Alex Young <alex@bytemark.co.uk> Mon, 16 Jul 2012 10:29:06 +0100
flexnbd (0.0.1-0) stable; urgency=low
* It compiles :)
-- Matthew Bloch <matthew@bytemark.co.uk> Tue, 15 May 2012 02:42:03 +0100