Distributed Network Block Device 3 --rewrite for Kernel 3.x https://git.openslx.org/dnbd3.git/atom/inc?h=HEAD 2025-12-09T14:33:20+00:00 2025-12-09T14:33:20+00:00 Simon Rettberg 2025-10-29T17:05:00+00:00 urn:sha1:062df52ab9f3464ce79ca91d54fa3d2ad209a31b - Fold header/data handling into one function This uncovered a few redundant checks and makes it easier to reason about control flow - Make all iscsi_pdu stack-allocated This greatly reduces the number of malloc and free calls during normal operation, lowers the risk of memory management bugs, and potentially increases performance in high concurrency scenarios. 2025-12-09T14:33:20+00:00 Simon Rettberg 2025-10-24T06:52:25+00:00 urn:sha1:9e2e94ecb8140b159e1ba4d148d2e6dc57b5fc92 2025-12-09T14:33:20+00:00 Simon Rettberg 2025-10-16T12:50:30+00:00 urn:sha1:619502eea60436f50b63300ab3d4e73077f5194c 2025-12-09T14:33:20+00:00 Simon Rettberg 2025-10-10T16:13:16+00:00 urn:sha1:531ba156de326210e4807b701183eaf2f506cf2a Work towards simplifying the iscsi implementation has begun. Goals are: - Simpler and easier to understand resource/lifecycle management of allocations - Single-threaded architecture, making locking unnecessary - Moving as many allocations as possible to the stack - Making the call-stack more shallow for easier tracking of code flow 2024-05-13T16:26:42+00:00 Simon Rettberg 2024-04-11T14:14:59+00:00 urn:sha1:ef86f10d71bf2081b17025e0ea2ad020845233bc - Add support for aborting session and further uploads via SIGQUIT - Make API versioned - Change it to "uuid" everywhere instead of a mix of uuid and guid - Server can now tell us to wait with further uploads - merge request is now urlencoded POST instead of mime data 2024-05-13T16:26:42+00:00 Simon Rettberg 2024-04-08T14:07:28+00:00 urn:sha1:c4a26bb3930dfee339461322719f7f9a94c8de52 Add a version to the URL, rest can be configured dynamically by -C as it's a simple prefix. 2024-05-13T16:26:42+00:00 Simon Rettberg 2023-04-14T15:10:02+00:00 urn:sha1:9bf6fbce6dfccd16bb54a9801ca287bac6950a09 Don't allocate a data cluster in data file for empty l2 entries when resizing the image file. Calculating l2 offset in metadata file was broken and overlapping l1. Delete unneeded entries from cow struct. Rename a few more variables. Fix a few possible race conditions. Only upload modified blocks from cluster. Simplify cow_write() function by handling misaligned start/end first. Try to also simplify cow_read() a bit. TODO: Documentation, update the cow merger service. 2024-05-13T16:26:42+00:00 Simon Rettberg 2023-03-29T10:03:54+00:00 urn:sha1:ec0b9b5f72607335ac19d4ea00c24919943b7970 - Use the term "cluster" for a group of dnbd3-blocks instead of also calling them blocks. - Use term "table" instead of "array" for the L1 and L2 tables. - Use term "index" instead of "offset" when addressing those tables - Fix a few logic bugs, use-after-free - Add TODOs for parts that need better comments 2024-05-13T16:25:14+00:00 Michael Scherle 2022-02-17T17:58:25+00:00 urn:sha1:d92d1d3faf7cde517d8c3fe914301eb5aa872f62 2022-02-18T20:34:55+00:00 Simon Rettberg 2022-02-12T22:56:35+00:00 urn:sha1:eb2876f6542af2bfa47c7a6905ecc4f81f1d2ad3 Using workqueues frees us from having to manage the lifecycle of three dedicated threads. Discovery (alt server checks) and sending keepalive packets is now done using work on the power efficient system queue. Sending and receiving happens via dedicated work queues with higher priority. blk-mq has also been around for quite a while in the kernel, so switching to it doesn't hurt backwards compatibility. As the code is now refactored to work more as blk-mq is designed, backwards compatibility even improved while at the same time freeing us from an arsenal of macros that were required to make the blk-mq port look and feel like the old implementation. For example, the code now compiles on CentOS 7 with kernel 3.10 without requiring special macros to detect the heavily modified RedHat kernel with all its backported features. A few other design limitations have been rectified along the way, e.g. switching to another server now doesn't internally disconnect from the current one first, which theoretically could lead to a non-working setup, if the new server isn't reachable and then - because of some transient network error - switching back also fails. As the discover-thread was torn down from the disconnect call, the connection would also not repair itself eventually. we now establish the new connection in parallel to the old one, and only if that succeeds do we replace the old one with it, similar to how the automatic alt-server switch already does it.