* hasher thread could die if a client would rapidly
upload and delete files (so very unlikely)
* two unprotected calls to register_vpath which was
almost-definitely safe because the volumes
already existed in the registry
adds options `--bauth-last` to lower the preference for
taking the basic-auth password in case of conflict,
and `--no-bauth` to entirely disable basic-authentication
if a client is providing multiple passwords, for example when
"logged in" with one password (the `cppwd` cookie) and switching
to another account by also sending a PW header/url-param, then
the default evaluation order to determine which password to use is:
url-param `pw`, header `pw`, basic-auth header, cookie (cppwd/cppws)
so if a client supplies a basic-auth header, it will ignore the cookie
and use the basic-auth password instead, which usually makes sense
but this can become a problem if you have other webservers running
on the same domain which also support basic-authentication
--bauth-last is a good choice for cooperating with such services, as
--no-bauth currently breaks support for the android app...
plus misc similar technically-incorrect addq usages;
most of these don't matter in practice since they'll
never get a url with a hash, but makes the intent clear
and make sure hashes never get passed around
like they're part of a dirkey, harmless as it is
counterpart of `--s-wr-sz` which existed already
the default (256 KiB) appears optimal in the most popular scenario
(linux host with storage on local physical disk, usually NVMe)
was previously 32 KiB, so large uploads should now use 17% less CPU
also adds sanchecks for values of `--iobuf`, `--s-rd-sz`, `--s-wr-sz`
also adds file-overwrite feature for multipart posts
the default (256 KiB) appears optimal in the most popular scenario
(linux host with storage on local physical disk, usually NVMe)
was previously a mix of 64 and 512 KiB;
now the same value is enforced everywhere
download-as-tar is now 20% faster with the default value
it is now possible to grant access to users other than `${u}`
(the user which the volume belongs to)
previously, permissions did not apply correctly to IdP volumes due to
the way `${u}` and `${g}` was expanded, which was a funky iteration
over all known users/groups instead of... just expanding them?
also adds another sanchk that a volume's URL must contain a
`${u}` to be allowed to mention `${u}` in the accs list, and
similarly for `${g}` / `@${g}` since users can be in multiple groups
the volflags of `/` were used to determine if e2d was enabled,
which is wrong in two ways:
* if there is no `/` volume, it would be globally disabled
* if `/` has e2d, but another volume doesn't, it would
erroneously think unpost was available, which is not an
issue unless that volume used to have e2d enabled AND
there is stale data matching the client's IP
3f05b665 (v1.11.0) had an incomplete fix for the stale-data part of
the above, which also introduced the other issue
this commit partially fixes the following issue:
if a client manages to escape real-ip detection, copyparty will
try to ban the reverse-proxy instead, effectively banning all clients
this can happen if the configuration says to obtain client real-ip
from a cloudflare header, but the server is not configured to reject
connections from non-cloudflare IPs, so a scanner will eventually
hit the server IP with malicious-looking requests and trigger a ban
copyparty will now continue to process requests from banned IPs until
the header has been parsed and the real-ip has been obtained (or not),
causing an increased server load from malicious clients
assuming the `--xff-src` and `--xff-hdr` config is correct,
this issue should no longer be hitting innocent clients
the old behavior of immediately rejecting a banned IP address
can be re-enabled with the new option `--early-ban`
too fraught with subtle dangers, such as other copyparty instances
ending up sharing knowledge of volumes unintentionally, and
configuration becoming mysteriously sticky (not to mention
this would all become hella difficult to reason about)
instead, rely entirely on users seeing the big red warning
added in 2ebfdc25 if their configuration is dangerous
this decision has the drawback that there will be server stuttering
whenever a new user makes themselves known since the last restart,
as it realizes the volumes exist and does the usual e2ds indexing,
instead of doing it early during startup
but it's probably good enough
when switching to another folder with identical filenames, the
mediaplayer would get confused and think it was the same files,
messing up the playback order
to abort an upload, refresh the page and access the unpost tab,
which now includes unfinished uploads (sorted before completed ones)
can be configured through u2abort (global or volflag);
by default it requires both the IP and account to match
https://a.ocv.me/pub/g/nerd-stuff/2024-0310-stoltzekleiven.jpg
running behind cloudflare doesn't necessarily
mean being accessible ONLY through cloudflare
also include a general warning about optimal
configuration for non-cloudflare intermediates
as this option is very rarely useful, add global-option `--k304` to
unhide the button and/or set it default-enabled
the toggle will still appear when the feature was previously enabled by
a client, and the feature is still default-enabled for all IE clients
if a reverse-proxy starts hijacking requests and replying with HTML,
don't panic when it fails to decode as a handshake json
fix this for most other json-expecting gizmos too,
and take the opportunity to cleanup some text formatting
this improves performance on s3-backed volumes
noktuas reported on discord that the upload performance was
unexpectedly poor when writing to an s3 bucket through a JuiceFS
fuse-mount, only getting 1.5 MiB/s with copyparty, meanwhile a
regular filecopy averaged 30 MiB/s plus
the issue was that s3 does not support sparse files, so copyparty
would fall back to sequential uploading, and also disable fpool,
causing JuiceFS to repeatedly commit the same 5 MiB range to
the storage provider as each chunk arrived from the client
by forcing use of sparse files, s3 adapters such as JuiceFS and
geesefs will "only" write the entire file to s3 *twice*, initially
it writes the full filesize of zerobytes (depending on adapter,
hopefully using gzip compression to reduce the bandwidth necessary)
and then the actual file data in an adapter-specific chunksize
with this volflag, copyparty appears to reach the full expected speed
* docker: warn if there are config-files in ~/.config/copyparty
because somebody copied their config into
/cfg/copyparty instead of /cfg as intended
* docker: warn if there are no config-files in an included directory
* make misconfigured reverse-proxies more obvious
* explain cors rejections in server log
* indicate cors rejection in error toast
nothing dangerous, just confusing log messages if an
admin hammers the reload button 100+ times per second,
or another linux process rapidly sends SIGUSR1
`@import url(https://...)` would get rewritten to baseURL + https://...
also reorder the generated csstext so that @imports appear first;
necessary for stuff like googlefonts to take effect
some reverse-proxies expect plaintext replies, and
we don't have a brotli decompressor to satisfy this
additionally, because brotli is https-gated (thx google),
it was already an impractical mess anyways
the sfx is now 7 KiB larger
chrome crashes if there's more than 2000 unique SVGs on one page, so
there was serverside useragent-sniffing to determine if the icon should
be an svg or a raster
however since the useragent is not in our vary, cloudflare wouldn't see
the difference and cache everything equally, meaning most folders would
display a random mix of png and svg thumbnails
move browser detection to the clientside to ensure unique URLs
* if a nic was restarted mid-transfer, the server could crash
* this workaround will probably fix a bunch of similar issues too
* fix resource leak if dualstack fails the ipv4 bind
on phones especially, hitting the end of a folder while playing music
could permanently stop audio playback, because the browser will
revoke playback privileges unless we have a song ready to go...
there's no time to navigate through folders looking for the next file
the preloader will now start jumping through folders ahead of time
some cifs servers cause sqlite to fail in interesting ways; any attempt
to create a table can instantly throw an exception, which results in a
zerobyte database being created. During the next startup, the db would
be determined to be corrupted, and up2k would invoke _backup_db before
deleting and recreating it -- except that sqlite's connection.backup()
will hang indefinitely and deadlock up2k
add a watchdog which fires if it takes longer than 1 minute to open the
database, printing a big warning that the filesystem probably does not
support locking or is otherwise sqlite-incompatible, then writing a
stacktrace of all threads to a textfile in the config directory
(in case this deadlock is due to something completely different),
before finally crashing spectacularly
additionally, delete the database if the creation fails, which should
prevents the deadlock on the next startup, so combine that with a
message hinting at the filesystem incompatibility
the 1-minute limit may sound excessively gracious, but considering what
some of the copyparty instances out there is running on, really isn't
this was reported when connecting to a cifs server running alpine
thx to abex on discord for the detailed bug report!
as each chunk is written to the file, httpcli calls
up2k.confirm_chunk to register the chunk as completed, and the reply
indicates whether that was the final outstanding chunk, in which case
httpcli closes the file descriptors since there's nothing more to write
the issue is that the final chunk is registered as completed before the
file descriptors are closed, meaning there could be writes that haven't
finished flushing to disk yet
if the client decides to issue another handshake during this window,
up2k sees that all chunks are complete and calls up2k.finish_upload
even as some threads might still be flushing the final writes to disk
so the conditions to hit this bug were as follows (all must be true):
* multiprocessing is disabled
* there is a reverse-proxy
* a client has several idle connections and reuses one of those
* the server's filesystem is EXTREMELY slow, to the point where
closing a file takes over 30 seconds
the fix is to stop handshakes from being processed while a file is
being closed, which is unfortunately a small bottleneck in that it
prohibits initiating another upload while one is being finalized, but
the required complexity to handle this better is probably not worth it
(a separate mutex for each upload session or something like that)
this issue is mostly harmless, partially because it is super tricky to
hit (only aware of it happening synthetically), and because there is
usually no harmful consequences; the worst-case is if this were to
happen exactly as the server OS decides to crash, which would make the
file appear to be fully uploaded even though it's missing some data
(all extremely unlikely, but not impossible)
there is no performance impact; if anything it should now accept
new tcp connections slightly faster thanks to more granular locking
