The terminating NUL is actually part of the index record, but with file
corruption, it's technically possible the NUL could be missing.
Fixes CID 509552

Use strnicmp() insted of stricmp() in smb_removefile()
Fixes CID 509551

The terminating NUL is actually part of the index record, but with file
corruption, it's technically possible the NUL could be missing.

Fixes CID 509554

Multiple directory searching functions wouldn't terminate on Ctrl-C

If the user didn't supply a wildcard, only a single file (the first file
matching the liberal file pattern) would ever be listed in the file
listing/searches, even when multiple files in the directory matched the
liberal file matching pattern. This is not a new bug.

Commit 3a3c889b (2 years ago now) changed loadfiles() to use liberal file
matching (e.g. "syncterm.exe" matched both "syncterm.exe" and
"syncterm_v1.2b.exe").

This could produce surprising results when doing file list querieis/operations
with the FileBase methods via JS (e.g. jsexec utils) and (now that I look at
it), the FTP server too.

So we should not have been doing liberal file matching *everywhere* loadfiles
is used, just where it was a usability issue (due to displayed filenames being
truncated to 12 chars for <=80 column terminals).

Now solved by add/use of new liberal_filepattern() function only in the
built-in file listing methods: sbbs_t::listfiles() and sbbs_t::listfileinfo().

Note: Custom JS file searching/listing scripts may now need their own
work-arounds for this usability issue, if they have it.

Clarify that this is the method to use to rename a file.

Since sorting works on the file index record date-time (date/time added)
it was confusing to display dates that didn't always match the sort order.

Should fix issue #790

If you want to "touch" a file in the filebase (so it appears a new again)
this is the script to use to do that.

DDFileLister: When displaying a file description, remove/replace cursor movement characters, which can corrupt the display

See merge request !460

Eric Oulashin authored 6 months ago and Rob Swindell committed 6 months ago

DDFileLister: When displaying a file description, remove/replace cursor movement characters, which can corrupt the display

emailfiles.js puts files here that the user requested to download "via email"
and might not have been successfully delivered and deleted. So clean-up.

to force SBBSecho to create a .req file (in FLO-mode), rather than send as
netmail.

For Keyop

... otherwise, we'll erase the file's extended description or auxdata, if it
has any.

Nelgin reported a weird error with a failed very large allocation for the
base/code argument to the FileBase constructor. There's no good reason
these strings were heap-allocated in the first place, so just change to
use a stack allocated variable instead. I don't know why this would fix
anything, but at least there's one less heap allocation and potential
for memory leak here.

Fix 2 bugs in js_update_file():
1. missing parenthesis (really?!? Caught by Coverity - sigh) in last
commmit caused attempt/failure/error to remove file after making any
updates. The removing is only supposed to happen when its necessary to
remove and re-add the file to the filebase (e.g. updating extended
description or auxdata).
2. Wrong filename used in 'removing' exception string.

This fixes a filebase corruption issue that could be triggered by using
FileBase.update() to rename a file while also changing the file's auxdata or
extended description: you can't remove a file header that has the new filename
(in the file_t.name field), cause it doesn't exist yet. So we needed an SMBLIB
API function to pass the (original) filename instead. Much like the filedat.c
removefile() function, but without the SMB opening/closing feature.

FileBase.update() will now throw an exception if attempting to rename a file
to a filename that already exists in the filebase index.

Prior to this change, one could (via JS methods) rename a file in the base to
create a duplicate filename which would corrupt the base (one index entry and
two header records for the same filename).

This will help to determine cause of any file update (e.g. rename)
failures, as reported by Nelgin.

I was able to corrupt a filebase using fileman.js, renaming a file
(the filename in the index as viewed with 'smbutil x' did not match
the filename listed with 'smbutil l') - yet, chksmb reported no errors
with the filebase.

Now chksmb will detect that type of corruption.

chksmb needed this to perform filename checks (index against headers).

This served exactly no purpose (since the idxrec_t existed at the same
offset in both parts of the union). So this was just some weird artifact
from the new filebase development.

Increasing size of mode[] element by 2 bytes eliminated these GCC warnings
that seem like false-positives to me:

sbbs.h:194:48: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
  194 |                                 (ret)[JSSTSpos]=(char)JSSTSstrval[JSSTSpos]; \
      |                                 ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~
js_file.c:225:25: note: in expansion of macro ‘JSSTRING_TO_STRBUF’
  225 |                         JSSTRING_TO_STRBUF(cx, str, p->mode, sizeof(p->mode), NULL);
      |                         ^~~~~~~~~~~~~~~~~~
js_file.c:42:17: note: at offset 5 into destination object ‘mode’ of size 5
   42 |         char    mode[5];
      |                 ^~~~

Similar use of JSSTRING_TO_STRBUF in other files (js_console.cpp, js_archive.c)
(with larger target buffers) does not trigger the same warnings.

DD file lister: Check for null when getting extended metadata from the DB (possibly caused by corruption). Also, allow changing the filename when editing file info/metadata.

See merge request !459

Eric Oulashin authored 6 months ago and Rob Swindell committed 6 months ago

DD file lister: Check for null when getting extended metadata from the DB (possibly caused by corruption). Also, allow changing the filename when editing file info/metadata.

Ctrl-A - Select All
Ctrl-F - Filter IP address
DEL - Close Socket

Also, use Begin/EndUpdate() calls to batch updates to the ListView in Timer
tick (no visible difference, but it's the right thing to do).

I don't know why C++Builder increased the TextHeight value of this form from
13 to 15, but doesn't seem to really matter (or at least, I couldn't tell).

I don't know why this was set so wide (10 pixels), but that did help to easily
see that it was being placed in the wrong spot in the form (between the tool
button bar and the top panel). But we don't need it to be that thick. 1 pixel
is still enough to be able to grab it with the mouse and resize the top/bottom
panels.

For who knows how long now, the horizontal splitter that allows the sysop
to control the size of the top 2 panels (server forms), was displayed above
the top panel, not between the top and bottom panel, thus making the panel
heights not controllable. Weird. Change the order of making the controls
visible to insure that the splitter is displayed below the top panel and
on top of the bottom panel.

Also (non-functional changes):
Remove redunant mutex from client_on() - makes no difference.
Use try/catch in client_on() to try to catch VCL exceptions - doesn't catch
the "invalid index" error popups that we see on occasion. <shrug>

... bug introduced in commit 29a35642.

strListMerge() doesn't realloc the strings in the list, so we don't want to
free the strings in this list here.

This is likely the cause of the crash Keyop eluded to in #synchronet.

... and log a debug-level log message when doing so.

The current input/receive timeout duration (5 minutes) might be a bit long to
leave a message base open (though, there's no known issue with doing so). For
example, when users read message bases using the terminal server, they can sit
for much longer than 5 minutes reading an open message base.

As Nelgin pointed out, the nntpservice can leave a message bases open while
a client is idle. Normally, an inactive client would be disconnected after 5
minutes of inactivity, so that wouldn't be much of an issue. However, if
authenticated as a user with the H-exemption, no auto-disconnection after
input timeout would happen. So let's close any open message base for good
measure.

The input/receive timeout probably should be configurable and lowered to
a smaller duration (one minute?). I'll leave that for another commit.

Reported by Keyop (upon being auto-disconnected overnight):
!ERROR 9 (Bad file descriptor) in data_ovl.cpp line 47 (putmsgptrs) writing
"message pointers" access=0

Unfortunately, the way putmsgptrs() was written, we couldn't tell if this was
an open (data/user/*.subs file) issue or a writing issue.

So create putmsgptrs_fp() and use that from sbbs_t::putmsgptrs(), so we can
log a different error (with more accurate details) if it's a file-open failure
versues a file-write failure.

Fix #782 - websocketservice.js garbles input on Banana Pi running Armbian

Closes #782

See merge request !458

The "from_ip" header field of QWK messages is checked against blocked IP
addresses, but was only checking against IP addresses from ip.can, not any
addresses listed in ip-silent.can. These 2 list files are now merged together
for the purposes of filtering during QWK/REP packet import.

Add '-v' (increase verbosity) option, used to display msg dates and timezones
... when using the the 'l' (list messages) command (to view post date/time).
Use '-vv' or '-v -v' to see timezones of messages.

The -v option is now also applicable to the 'v' (view) messages command (now
redundant with the 'V' command).

Features as requested by Nelgin as part of issue #786.

Removed day-of-week from date/times displayed. We don't need that level of
user-friendliness with this tool.
However, we are also displaying 12h/am/pm times. Some sysops probably would
prefer 24hour time, so that should be considered at some point.

... related to issue #786.

Also, don't check for a message-ID if the message-type does not match the
expected message type ("type mismatch").

1. When a 0-length bundle file was encountered or an unpacking error occurred,
   any remaining bundles for the current search day-of-week (e.g. *.SU*) would
   be skipped/ignored. This bug (issue #764, regarding the 0-length file part),
   is fixed by not incrementing the day-of-week index in the main loop, but
   rather only incremeting the index when all bundles for the current
   day-of-week have been processed.
2. The age calculation for 0-byte/length bundle files was incorrect, so all
   0-length  bundle files would always be considered "less than 24-hours old"
   (and thus, never auto-deleted).
   This exacerbated the problem of issue #764 since it would persist until the
   0-length files were manually deleted. Fixed the file age calculation and
   now logging the date/timestamp of the 0-length file as well.
3. Don't do the switch/case/sprintf dance when we're not re-running a glob()
   search.
4. Replace the switch/case statement with an array of week day names/patterns.
5. Ignore (with a warning log message) any sub-directories of the inbound
   directory that happen to match the bundle file search pattern.
6. Use better variable naming.
7. Refer to files with a length of 0 as "0-length" instead of "0-byte" in log
   messages.