| ![[View project]](images/buttons/View%20project.gif) ![[View on GitHub]](images/buttons/View%20on%20GitHub.gif) | Date | Commit | Author | Project | | 20261120 08:11 | 0af6e5a | orgin | sftp-handler | Readme: no email (it goes only into the OS4Depot upload form)
os4depot_submit.py takes the address from OS4DEPOT_EMAIL or
~/openamiga-rules/os4depot.conf and sends it hidden; the readme, the
archive and git never carry it. | | 20261120 08:08 | 2ba98c5 | orgin | sftp-handler | Readme in Markdown per the OpenAmiga readme rules; submit tool sends url and the readme's email (hidden)
The OS4Depot readme follows ~/openamiga-rules/readme_rules.md: Markdown
body (texttype Markdown), title line, OpenAmiga.org paragraph, ## sections,
History as ### x.y (DD.MM.YYYY). tools/update_readme.py is the shared
reference copy; make_lha.sh dates the readme DD.MM.YYYY and looks for
### <version>. os4depot_submit.py: f_url and f_email from the readme, and
the beta check no longer trips over the 0.10 beta History entry. | | 20261120 07:48 | c8145f4 | orgin | sftp-handler | README like FileBrowser's (OpenAmiga.org project link, History); readme: url and OpenAmiga.org project note | | 20261120 07:38 | ddb38f0 | orgin | sftp-handler | tests/os4/run_guest.py: SFH_LIBS puts more libraries first in LIBS: (AmiSSL 6 test) | | 20261120 07:24 | b31613e | orgin | sftp-handler | README: OS4Depot link and changelog (tools/update_readme.py, run by make_lha.sh) | | 20261118 05:56 | 442212d | orgin | sftp-handler | 0.15: faster SFTP: libssh2 patches for whole-packet decryption, short reads and bigger read requests
Two patches to libssh2 1.11.1 (tools/libssh2-patches), reported upstream
as libssh2#2734 and #2735:
- 0001: decrypt the middle of a packet in one crypt() call (OpenSSL
backend, in place) instead of one call per 16-byte block, for every
cipher but chacha20-poly1305. The first block and the end of a packet
(GCM tag) are handled as before. Encryption is unchanged.
- 0002: a short FXP_DATA reply no longer fails with "Read Packet At
Unexpected Offset": the requests after it are dropped (as seek does)
and asked for again from where it ended. New
libssh2_sftp_set_read_size(); the read size is kept per handle and a
server's limit is learned only when the file goes on after a short
reply. The default (30000) is unchanged.
The handler asks for 255 KB per read request and has a new option to
pick the ciphers (DOSDriver CIPHERS=, Linux --ciphers); the negotiated
methods are logged. build_deps_os4.sh unpacks libssh2 fresh and applies
the patches (build_os4.sh rebuilds when a patch changed); the Linux test
build now links the same patched libssh2 (tools/build_deps_linux.sh,
run by the Makefile).
Tests: testserver/cap (OpenSSH behind a proxy that cuts read requests to
a cap, CBC allowed); t_sftp_ciphers (8 ciphers: uploads, whole, piece
and random-offset reads, end first) and t_sftp_short_reads (caps 20000,
65536, 100000; sizes 0 to 3 MB). With the system's unpatched libssh2,
t_sftp_short_reads fails at the first file bigger than the cap. Linux:
65 passed. Guest: functional test ok; tests/os4/ciphers.py 45/45 (9
cipher/cap runs, SHA-256); downloads chacha20 14.2-14.6 MB/s (0.14:
9.6-13.1), aes128-ctr 8.45 MB/s (before: 3.1), aes256-gcm 7.45 MB/s. | | 20261118 04:42 | 7a51c10 | orgin | sftp-handler | 0.14: FTP: ABOR for cut downloads, late replies dropped (ZitaFTP)
The reporter's ZitaFTP log showed the real cause of "No disk present"
with a 1188-file folder (the 425 wait in 0.12 was not it: ZitaFTP frees
its 5 data ports at once). Workbench reads only the start of each file,
so most downloads are cut short, and ZitaFTP
- often sends no reply for a cut download: the handler waited 5 s,
closed the control connection and logged in again (53 times in the
log; 57 logins for 61 downloads on one machine);
- sometimes sends one reply too many: the next EPSV read the stale
reply, PASV read EPSV's answer, and every data connection after that
failed as EPROTO -> EIO -> "No disk present".
stream_stop() now sends ABOR after closing the data connection and reads
replies until the one for ABOR (225/226, or 5xx), skipping 426/451, then
drops what arrives within 300 ms. cmd() drops replies waiting before a
command, and after a cut download skips late 225/226/426.
testserver/zitasim.py: --abort-reply none|two|two-late. New tests
t_ftp_cut_no_reply (35.8 s and a login per file before, now under 8 s
with one login) and t_ftp_cut_two_replies (EIO before); 63 passed.
Guest: Workbench read all 1188 files with one login, 1188 ABORs, 125
late replies dropped, no errors. Docs: History 0.14; ZitaFTP no longer
named as the 425 example. | | 20261118 03:25 | 1f259e5 | orgin | sftp-handler | 0.13: readable messages for refused changes and "file too large"
filesysbox shows "Error -1" for EACCES, EPERM and EFBIG. The glue
already turned "permission denied" on changes into "write protected";
now also on opening a file for writing and on flush (where FTP uploads),
and EFBIG gives "disk is full". Guest: copying, deleting and MakeDir in
a read-only folder say "file is write protected" on SFTP and FTP.
tests/os4/cryptobench.c: measures an Amiga's decryption speed per SSH
cipher (libssh2 chacha20-poly1305, AmiSSL AES/HMAC); from the speed
analysis, kept as a diagnostic tool. | | 20261118 00:24 | f29975c | orgin | sftp-handler | 0.12: FTP waits out "425 cannot open data connection"
A user saw "No disk present in device SFTP" after opening a folder with
1188 files on a ZitaFTP Server (FTP). Workbench reads the start of every
file of a folder it opens (DefIcons), one data connection each, and
relists the folder when the 5 s cache expires. ZitaFTP gives data
connections a port from a fixed range; when it is used up it answers
"425 cannot open data connection". open_data() took that as a protocol
error -> EIO -> the glue's "No disk present" for reads and lookups.
open_data() now treats 425 (EPSV or PASV) as "busy": it waits with
growing pauses (100 ms to 2 s) and asks again, up to the I/O timeout,
then returns -EBUSY ("Object in use"). A 425 to the first EPSV no longer
turns EPSV off. New plat_sleep_ms().
testserver/zitasim.py: an FTP server that behaves like ZitaFTP 1.36
(its reply texts, FEAT without MLSD, a data port range with ports busy
for a while after use -> 425, "421 max clients exceeded (N)"). New
Linux tests t_ftp_data_ports_busy, t_ftp_data_ports_gone,
t_ftp_max_clients; 61 passed. Guest: Workbench opened the 1188-file
folder against it; 0.11 failed 1750 lookups and 636 reads with EIO,
0.12 read all 1188 files with no error.
Guide: Troubleshooting entry for big FTP folders; History 0.12 in
readme and guide. | | 20261117 17:41 | b9803ed | orgin | sftp-handler | Readme: which AmiSSL versions are supported |
|
|
|