• src/smblib/smblib.c

    From Rob Swindell@VERT to Git commit to main/sbbs/master on Sun May 2 17:14:43 2021
    https://gitlab.synchro.net/main/sbbs/-/commit/ba1f38cc4126d388c6a897ef
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    When adding a msg/file header, remove the DELETE attribute

    When moving or renewing files, they're first removed and then added and they were being added with the DELETE attribute set. The filenames are displayed in red when staged for deletion/removal, so that's how I noticed. Oops.

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Windows)@VERT to Git commit to main/sbbs/master on Mon May 8 18:07:50 2023
    https://gitlab.synchro.net/main/sbbs/-/commit/2ee37e4ee9e91ddb2d790f72
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    Fix the relocation of enlarged message headers in smb_putmsghdr()

    The first commit of this feature would leave the base corrupted for a couple
    of reasons (.sha file wasn't opened, using modified header when re-writing the old header to set the DELETED attribute). Both fixed.

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Debian Linux)@VERT to Git commit to main/sbbs/master on Mon May 8 18:09:07 2023
    https://gitlab.synchro.net/main/sbbs/-/commit/b5c5630c1ccd675a559dd98d
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    More detail in error strings in smb_getmsghdr()

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Windows)@VERT to Git commit to main/sbbs/master on Mon May 8 18:16:05 2023
    https://gitlab.synchro.net/main/sbbs/-/commit/4985797d2b07ca169326ab6f
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    Streamlined code: we need only one check for hyper-alloc'd msg base here

    No functional change.

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Windows 11)@VERT to Git commit to main/sbbs/master on Fri Dec 29 17:10:28 2023
    https://gitlab.synchro.net/main/sbbs/-/commit/47fe4a60247147dc217ad02b
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    Fix stale comments (msg.offset became msg.idx_offset a couple years back).

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Debian Linux)@VERT to Git commit to main/sbbs/master on Fri Sep 13 20:08:11 2024
    https://gitlab.synchro.net/main/sbbs/-/commit/1ab4ef2bfe0d0b1c7598cc53
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    smb_getmsgidx() now reads entire index records from file bases too

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

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Windows 11)@VERT to Git commit to main/sbbs/master on Fri Sep 13 20:30:10 2024
    https://gitlab.synchro.net/main/sbbs/-/commit/ffa2c51ba23c59641e1a6e47
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    MSVC requires a 0 in the struct initializer

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Debian Linux)@VERT to Git commit to main/sbbs/master on Sun Nov 10 16:14:22 2024
    https://gitlab.synchro.net/main/sbbs/-/commit/6224ce6476a079f5e1e7a880
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    Fix printf formatting issue in error message

    GCC and Coverity (x 2) both flagged this one.

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Windows 11)@VERT to Git commit to main/sbbs/master on Fri Jan 16 20:45:33 2026
    https://gitlab.synchro.net/main/sbbs/-/commit/5bd46a2ed2ff6bbc4aa2fb8a
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    smb_lock(): Confirm the lock file exists after closing it and a short delay

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Debian Linux)@VERT to Git commit to main/sbbs/master on Mon May 11 23:57:27 2026
    https://gitlab.synchro.net/main/sbbs/-/commit/931d45a1c1bd411af6c1c16c
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    smb_new_msghdr: auto-repair small .sid/.shd status mismatches; check smb_putstatus return value

    Problem: if the .sid index file length didn't exactly match total_msgs * idxreclen, smb_new_msghdr returned SMB_ERR_FILE_LEN (-206) hard, blocking
    all further message adds until the msgbase was manually repaired. On 2026-05-09 this caused the mail SMB to reject every incoming message for
    ~14 hours (647 logged errors), resulting in SMTP "452 Insufficient system storage" responses.

    Root cause of the corruption: smb_putstatus() was called but its return
    value was silently discarded. If the status write failed after the index record was already appended to .sid, the two files diverged: .sid had one
    more record than total_msgs reflected (or vice-versa).

    Fix 1 — propagate smb_putstatus() failures: the return value is now
    assigned back to i and returned to the caller, so a failed status write is
    no longer silent.

    Fix 2 — auto-repair small discrepancies in smb_new_msghdr (rather than hard-failing) when the mismatch is 1-2 records and the .sid length is an
    exact multiple of idxreclen:
    - LONG by 1 (.sid has one orphan record): truncate the extra record with
    chsize() and continue. This was the May 9 failure mode.
    - SHORT by 1-2 (.sid is missing 1-2 records): reduce total_msgs to match
    the actual index length, persist the correction with smb_putstatus(),
    and continue. The orphaned header/data space is harmless and can be
    reclaimed by smbpack.
    - Mismatch of 3+ records, or non-aligned length: still returns
    SMB_ERR_FILE_LEN as before.

    Test cases (smblib/smbidxtest.c, using CuTest framework):
    Test_NoCorruption - baseline: normal add still works
    Test_ShortIndexByOne - .sid truncated by 1 record: auto-repairs
    Test_ShortIndexByTwo - .sid truncated by 2 records: auto-repairs
    Test_LongIndexByOne - .sid extended by 1 record: truncates and proceeds
    Test_LargeCorruptionFails - .sid truncated by 3 records: still returns -206

    All 5 tests pass with the fix; 3 of 5 fail against the original code.

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Debian Linux)@VERT to Git commit to main/sbbs/master on Sat Jul 4 18:24:37 2026
    https://gitlab.synchro.net/main/sbbs/-/commit/7b48c6a6bb234217b80c2695
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    Prevent votes/polls from corrupting hyper-allocated message bases

    smb_addmsg() has always forced hyper-allocated storage when the base's status.attr has SMB_HYPERALLOC set, but smb_addvote(), smb_addpoll() and smb_addpollclosure() passed the caller's storage mode (typically derived
    from the sub's configured storage setting via smb_storage_mode()) straight through to smb_addmsghdr(). On a message base whose header file carries
    the SMB_HYPERALLOC attribute while the sub is configured for fast or
    self-pack allocation, a vote/poll header was then allocated via
    smb_fallochdr() or smb_allochdr() against allocation files that hyper-allocation deliberately does not maintain, yielding an offset inside existing headers: the new header overwrote a live message header and
    (spanning two blocks) overran the following one, destroying two messages
    per vote and leaving the base unreadable (smb_getmsghdr "corrupt message
    header ID" errors; chksmb "Duplicate Index Offsets" and "Overlapping
    Headers").

    Guard the header-storage branch in smb_new_msghdr() instead: if the base's persisted attr says hyper-allocated, force SMB_HYPERALLOC storage no
    matter what the caller passed. This protects every header-adding path, including any direct smb_addmsghdr() caller.

    Observed in the wild on the DEBATE sub of Vertrauen (its msgs.ini config selects fast allocation, but the 26-year-old base file is branded SMB_HYPERALLOC): the first votes cast on it destroyed messages #16-#19.

    Fixes issue #1181

    Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Debian Linux)@VERT to Git commit to main/sbbs/master on Thu Nov 19 09:20:50 2026
    https://gitlab.synchro.net/main/sbbs/-/commit/ce3494e738760fd80e4c44f0
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    smblib: free text_subtype/text_charset in smb_freemsghdrmem()

    smb_freemsghdrmem() ended with clear_convenience_ptrs(), which NULLs text_subtype and text_charset along with the pointers into header-field
    data. Those two are not such pointers: smb_parse_content_type() strdup()s
    them from a text/* Content-Type field, so they leaked each time a header
    with one was freed this way. MsgBase.put_msg_header() does that on every
    call (it re-reads the header and frees its fields before re-parsing the
    JS header object), leaking both strings per call; valgrind flagged it in
    the jsexec-testsuite-valgrind CI jobs.

    Free them alongside the header fields they are derived from.
    smb_freemsgmem() calls smb_freemsghdrmem(), so its own frees of the two
    are dropped.

    Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net
  • From Rob Swindell (on Debian Linux)@VERT to Git commit to main/sbbs/master on Sun Nov 22 12:19:51 2026
    https://gitlab.synchro.net/main/sbbs/-/commit/3980634758b2757f7645fb67
    Modified Files:
    src/smblib/smblib.c
    Log Message:
    smblib: smb_copymsgmem() leaves nothing allocated when it fails

    On a failed malloc(), smb_copymsgmem() returned with whatever it had
    already allocated for the copy still in it (leaked by callers that do not
    free the copy), and with the copy's dfield, hfield and hfield_dat still pointing at the source message's arrays wherever it had not replaced
    them yet. A caller that frees both after a failure, as MsgBase.copy_msg()
    and move_msg() do, then freed the source's arrays twice, and one that
    freed the copy after a failure part-way through the header fields freed
    the uninitialized entries of hfield_dat.

    The copy now starts with no arrays of its own, allocates hfield_dat
    zeroed, and on any failure frees what it allocated, leaving the copy
    holding no allocations, so smb_freemsgmem() on it is safe and not freeing
    it leaks nothing.

    Coverity CIDs 662423, 662427, 662429 (bulkmail.cpp).

    Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

    ---
    ■ Synchronet ■ Vertrauen ■ Home of Synchronet ■ [vert/cvs/bbs].synchro.net