Skip to content

DEBUG: ENTER-DIRECTORY ,,DISC-SCSI-1,0 fails with "APPROACHING END OF ACCOUNTING FILE"

Symptom: the SCSI disk boots, but @ENTER-DIRECTORY ,,DISC-SCSI-1,0 always fails, and the console shows APPROACHING END OF ACCOUNTING FILE.

Bottom line (VERIFIED from SINTRAN L source + carved kernel): that message is a red herring. It is SINTRAN error 243B (163 decimal), a benign accounting warning that is emitted from one place only - the RT-accounting collector at logout - and has no code path from ENTER-DIRECTORY, CHDSI, the mass-storage layer, or the SCSI driver. The real failure is a SCSI page-0 read/geometry problem whose genuine error code is almost certainly 232B "Device error" (or 141B "Transfer error"), being mis-rendered to the accounting string in the emulator's error-message lookup, or the accounting warning is a pre-existing, coincidental console state. Instrument the SCSI READ of page 0 first.

All numbers octal unless marked. Grading: VERIFIED = kernel/source/disk bytes prove it; INFERRED = strong reasoning, not byte-proven; OPEN = unsettled.

See also - concrete fix plan: SINTRAN/Devices/SCSI/SCSI-MOUNT-FIX-PLAN.md ranks the root causes against the actual RetroCore C# controller lines and gives ordered, manually-testable fixes plus a [SCSI-TRACE] instrumentation plan. It also re-grounds the SCSI-side claims below in the carved SINTRAN L bytes: SCDTS (62107B) and SCSID (62217B) are confirmed present in the carved 006-S3FS, but the exact internal-code -> user-code mapping (232B/141B/252B/224B) is currently only readable from NPL (a different revision than L) - treat that table as INFERRED until the 62107B region is hand-decoded.

1. What "APPROACHING END OF ACCOUNTING FILE" actually is

VERIFIED. It is SINTRAN error code 243B = 163 decimal.

Error-message table (Reference-Manuals/ND-60.128.5 EN SINTRAN III Reference Manual.md, and Developer/MON/Monitor Calls.md):

| Octal | Decimal | Message                            |
| 242   | 162     | Segment not fixed                  |
| 243   | 163     | Approaching end of accounting file |   <-- our message
| 244   | 164     | Accounting file full               |
| 245   | 165     | No more unused spooling files ...  |

Where 243B is set - the only place in the entire source tree:

SINTRAN/NPL-SOURCE/NPL/RP-P2-ACCRT.NPL (the RT-accounting collector, RP-P2-ACCRT = Read/Process accounting RT-program):

123125  %   CHECK THAT THERE IS ROOM IN THE ACCOUNTS:DATA FILE ...
123125          IF NMBR > RMAX THEN            % ACCOUNTS:DATA FILE FULL
123131             244=:ERFL; GO FAR SUIT6     % 244B = "accounting file full"
123142          ELSE
123143             IF A > DMAX THEN            % passed the WARNING threshold
123146                243=:ERFL                % 243B = "approaching end ..."
123150             FI
123150          FI
...
123341          *MON 2RTEX                     % RT EXIT  (this is an RT program)

Facts that make it a red herring (all VERIFIED):

  • It is set only inside RP-P2-ACCRT, an RT program that runs at logout to append a record to ACCOUNTS:DATA. It is gated on A > DMAX - the number of accounting records already in the file exceeding the warning limit (default 1000, per Reference-Manuals/ND-10315B SINTRAN III ACCOUNTING SYSTEM.md and Operations/SINTRAN/ND-30.003.007 EN SINTRAN III System Supervisor.md).
  • The manuals state the message is printed "on the terminal at every log out whilst accounting is running" (ND-10315B line 511; ND-60.128.5 line 5310) - not at mount time, and only if the ACCOUNTING system was started (@INIT-ACCOUNTING / @START-ACCOUNTING).
  • ENTER-DIRECTORY never enters the accounting collector, never opens ACCOUNTS:DATA, and never touches ERFL. There is no call edge from the mount path to RP-P2-ACCRT.

Conclusion (Hypothesis #4 - the "honest accounting" angle): RULED OUT. Mounting a directory cannot genuinely raise 243B. If the accounting message is truly being generated, it is an independent, coincidental logout event on a system where accounting happens to be running and the file happens to be near full - unrelated to the SCSI mount.

2. Hypothesis #1 - the string is wrong / mis-indexed

INFERRED (strong). The genuine failure code from a SCSI ENTER-DIRECTORY read error is not 243B - it is one of the mass-storage codes below. If the emulator's error-to-string table is offset or truncated, the real code renders as the wrong string, and 243B "approaching end of accounting file" is what the user sees.

What codes the SCSI / mass-storage read path can actually return

The SCSI driver's internal error codes are tiny T-register values (they are not the 2xxB user codes):

SINTRAN/NPL-SOURCE/SYMBOLS/L07/FILSYS-SYMBOLS.SYMB.TXT:

ILRCS=000002   NOLUN=000007   BADPA=000005   ILNOD=000006

They are translated to the user-facing SINTRAN error numbers by routine SCDTS ("determine status", SINTRAN/NPL-SOURCE/NPL/IP-P2-SCSI-DISK.NPL line 1010 / s3vs-4.symb 061511):

061514   X:=17/\A; X:=SCSTA(X)        % low 4 status bits -> code via SCSTA table
061526   IF D=TYPER THEN X:=240       % Illegal device type
061533   IF D=ILAOP THEN X:=201       % Illegal function code
061540   IF D=BADPA THEN X:=174       % Illegal parameter
061545   IF D=ILNOD OR D=NOLUN THEN X:=33    % No such logical unit
061555   IF D=COPNP THEN X:=3206      % Illegal request
061562   IF D=TRANE THEN X:=141       % Transfer error
061567   IF D=SBUSY THEN X:=3207      % Device busy
061574   IF D=RCONF THEN X:=3210      % Reservation conflict
061601   IF D=NESER THEN X:=3211      % Device do not answer
061606   ELSE          X:=232        % DEVICE ERROR  <-- default catch-all

None of these is 243B. A hard SCSI read failure of page 0 surfaces as 232B "Device error" (the default), or 141B "Transfer error", 33B "No such logical unit", 240B "Illegal device type", 174B "Illegal parameter", or the 32xxB SCSI-status range. Two other geometry-specific codes from the mass-storage layer are relevant: 252B "Not a multiple of hardware block size" and 224B "Incompatible device sizes."

So: the code the mount path really returns is one of {232, 141, 33, 252, 224, 240, 174}B. If the user's console prints 243B, either the emulator mis-maps one of those into 243B's slot, or (per section 1) the accounting message is unrelated. Test: capture the numeric error the mount returns (see checklist), do not trust the rendered string.

3. The ENTER-DIRECTORY -> page-0 read path (CHDSI)

Segment 006-S3FS, load address 26000B (tools/sintran-segment-carver/versions/L-VSX-500/segments/006-S3FS.bin; annotated dis at .../re/006-S3FS.annotated.dis). Anchors (VERIFIED, carved SINTRAN L bytes):

  • CHDSI = 37763B - enter/validate a directory.
  • RXDIR = 37643B - read the 8-word extended-info block from page 0.
  • WXDIR = 37702B - recompute checksum + write the block back.
  • REENB = 40162B - clear the "entered" flag on release.

CHDSI reads page-0 word 1750B (byte 2000 / 0x07D0) - the 8-word extended-info block - then the 32-byte master block at word 1760B (byte 2016 / 0x07E0). Page 0 = 1024 words = 2048 bytes, big-endian (SINTRAN/Filesystem/on-disk-format/extended-info-block.md, .../directory-label.md).

Critical VERIFIED behaviour (from SINTRAN/Filesystem/NDFS-VALIDATION.md Q1/Q7): on a bad or zero checksum, CHDSI does NOT reject the mount - it zeroes and REBUILDS the extended-info block (writes the geometry-derived capacity into words 1756-1757B, stamps the owner system number + "entered" flag bit 15, recomputes the checksum via WXDIR, and writes it back). Consequences:

  • A page-0 read that returns garbage does not produce a checksum-mount error. It triggers a write-back. So the surfaced failure is either (a) the write of the rebuilt block failing -> 232B device error, or (b) a silent "success" onto corrupt geometry that fails again on the next enter - matching the user's "always the same" symptom.
  • CHDSI also compares the stored page capacity (words 1756-1757B) against a device-geometry figure it derives from the unit datafield (which is populated from SCSI READ CAPACITY). A wrong READ CAPACITY value feeds a wrong geometry into this compare/rebuild. (Capacity-compare is VERIFIED in NDFS-VALIDATION Q3; the exact error code on a capacity mismatch is OPEN - see the live-trace note in section 6.)

The read that CHDSI/RXDIR issues goes through the mass-storage layer to the SCSI driver as an ordinary page read (transfer unit = one filesystem page = 1024 words = 2048 bytes at page 0 / LBA-equivalent 0). Any driver-level failure of that read returns through SCDTS (section 2).

For the full carved-bytes trace of how the mount reaches this read - the ENDIR 140176B command worker, the RXDIR/RCBLO page-cache lookup that dispatches the transfer through the device datafield pointer, and the exact disk-controller read contract - see code-logic/enter-directory.md.

4. SCSI page <-> sector geometry the driver requires

VERIFIED from SINTRAN/NPL-SOURCE/NPL/IP-P2-SCSI-DISK.NPL. The driver is self-configuring from READ CAPACITY, but it is built around a 1024-byte logical block:

  • MODE SELECT sets the drive to 1024-byte blocks (line 556): 
    057574   2000; *STATX 50        % BLOCK SIZE 1024 BYTES   (2000B = 1024 dec)
  • The direct-access "record size" base is 1024 bytes (lines 1267/1270, X:=12 octal = 10 decimal = log2(1024)): 
    062425   X:=12                  % RECORD SIZE 1024 BYTES  (direct-access disk)
  • It issues READ CAPACITY (opcode 22400B) to learn the actual block size (lines 1234-1237, INQUI/geometry init), stores it in SURSZ, and builds three shift instructions SUSI1/SUSI2/SUSI3 (*EXR SA/*EXR ST) that scale between ND page/word/byte quantities and SCSI LBA + block count: 
    062434   IF X=:D>1 THEN         % X starts = 10 (log2 of the 1024-byte base)
062444   DO WHILE A NBIT 0      % shift the reported block size right ...
062446      X-1; A SHZ -1       % ... decrementing X per bit
062450   OD
062451   IF A><1 GO FAR RSZER   % block size NOT a power of 2 -> ILLEGAL RECORD SIZE
062454   D-X; IF X<0 THEN X:=0 FI   % byte-and-record shift
062460   SUSI3; X\/A=:SUSI1; A\/D=:SUSI3   % build the shift instructions
  • The READ command builder (lines 1480-1533) uses those shifts to turn the ND amount/address into the SCSI byte count / LBA, and chooses READ(6) vs READ(10) ("10 byte format necessary" when the LBA or count overflows the 6-byte fields, lines 1504/1533).

Required geometry the emulated drive must present, therefore:

  1. READ CAPACITY must return a power-of-two logical block size (the driver was configured for 1024 bytes; 512 also works arithmetically because the shift adapts, but the block size must be a power of two or the driver aborts with ILRCS = 2 -> ILLEGAL RECORD SIZE, line 062522 RSZER: T:=ILRCS).
  2. READ CAPACITY's block size and the READ data payload must agree. If READ CAPACITY says 1024 but the READ actually delivers 512-byte sectors (or vice versa), the shift-derived LBA/count is wrong and page 0 reads the wrong bytes.
  3. READ CAPACITY's max-LBA (capacity) must be right. It feeds CHDSI's capacity compare/rebuild (section 3) and the pages-available field.
  4. One ND filesystem page = 2048 bytes = 2 x 1024-byte SCSI blocks = 4 x 512-byte sectors. Any off-by-one or wrong sectors-per-page makes page 0 (master block + extended info at byte 2000/2016) read from the wrong LBA.

RSZER->ILRCS(2) is translated by the caller; note SCDTS does not list ILRCS explicitly, so it falls to the 232B default at the user level.

5. Why boot works but ENTER-DIRECTORY fails (INFERRED)

Boot uses the ROM/MASB bootstrap path with fixed, hard-wired transfer parameters (a raw load of the boot area, low LBAs, small fixed count) and does not depend on READ CAPACITY, the SUSI shift instructions, the master/ extended-info block, or CHDSI. ENTER-DIRECTORY is the first operation that (a) runs the full READ-CAPACITY-driven geometry, (b) reads page-0 metadata at byte 2000/2016, and (c) validates/rebuilds it. So a geometry or status-byte bug that boot tolerates will first bite at mount. This is consistent with "boots but won't mount, always the same."

6. Ranked checklist - what to instrument in the SCSI emulator

Ranked by likelihood given the evidence above.

  1. Capture the real numeric error, not the string. Break in the emulator where the mount returns its error and read the code. If it is 232B / 141B / 33B / 252B / 224B, the "accounting" string is a rendering bug in the error-message table - fix the table mapping and chase the real code below. If the code genuinely is 243B, accounting is running and the message is an unrelated logout event (section 1) - separate problem.

  2. READ CAPACITY correctness (highest-value data bug). Verify the emulated drive returns: a power-of-two block size (ideally 1024), the correct last-LBA/capacity, in the exact byte layout the driver's LDDTX expects (IP-P2-SCSI-DISK.NPL 062357-062365, big-endian 32-bit). A non-power-of-two size aborts with ILRCS; a wrong capacity corrupts CHDSI's capacity compare.

  3. Block size vs actual READ payload agreement. Confirm the READ data for LBA 0..n delivers exactly the sector size READ CAPACITY advertised. A 1024-vs-512 mismatch (READ CAPACITY says one, transfer uses the other) makes the SUSI1/2/3 shift math read page 0 from the wrong LBA -> garbage master/ extended-info block. This is the prime suspect the task flagged.

  4. LBA / block-count mapping and READ(6) vs READ(10). Instrument the exact opcode + LBA + transfer length the driver issues for the first page-0 read. Expect a read of the first 2048 bytes (= 2 blocks @1024 / 4 sectors @512) at LBA 0. Watch for off-by-one LBA and wrong block count (the 6-byte vs 10-byte format switch at IP-P2-SCSI-DISK.NPL 1504/1533).

  5. Byte order of the returned page. Page 0 is big-endian words; the extended-info checksum is a 16-bit additive sum of words 1751-1757B, master block at 1760B. If the emulator byte-swaps the payload, the checksum mismatches -> CHDSI rebuilds and writes back (section 3); if the write-back path also mis-orders bytes it never converges -> persistent failure.

  6. SCSI status / phase bytes. The low 4 bits of the status word index SCSTA (IP-P2-SCSI-DISK.NPL 1010) and feed SCDTS. A bogus non-zero status (CHECK CONDITION not cleared, wrong message/phase sequence, missing REQUEST SENSE handling) forces the 232B/141B device-error path even when the data would have been fine. Verify status = 0 (GOOD) on a clean read and that sense data is well-formed.

  7. Write-back path. Because CHDSI rebuilds and writes page 0 on any checksum miss, verify WRITE(6)/WRITE(10) to LBA 0 works with the same geometry as READ. A read that is merely garbled but a write that fails yields a device error at exactly mount time.

If you can run a live nd100x DAP trace

Break at CHDSI = 37763B (segment 006-S3FS, loaded at 26000B; absolute break address depends on where the resident FS segment is mapped in your run) and:

  • Single-step into RXDIR = 37643B and dump the 2048 bytes it read for page
    1. Compare against the on-disk SMD0.IMG-equivalent bytes at 0x07D0 (extended info) and 0x07E0 (master block). If they differ, it is a read/geometry bug (items 2-5); if they match but the enter still fails, it is the capacity compare or the write-back (items 2, 7).
  • Watch whether execution reaches the rebuild branch (zero-fill loop + WXDIR = 37702B). Reaching rebuild every time = the read data never validates = geometry/byte-order bug.
  • At the point the mount returns to the command interpreter, read the T/A error code and confirm whether it is 232B (real) vs whether the console still prints 243B (rendering bug).

7. Evidence / provenance

  • Error table: Reference-Manuals/ND-60.128.5 EN SINTRAN III Reference Manual.md (243B/244B), Developer/MON/Monitor Calls.md.
  • Accounting-only source of 243B: SINTRAN/NPL-SOURCE/NPL/RP-P2-ACCRT.NPL (123146 243=:ERFL), gating + "at logout" semantics in Reference-Manuals/ND-10315B SINTRAN III ACCOUNTING SYSTEM.md and Operations/SINTRAN/ND-30.003.007 EN SINTRAN III System Supervisor.md.
  • SCSI error translation SCDTS: SINTRAN/NPL-SOURCE/NPL/IP-P2-SCSI-DISK.NPL (~line 1010) / SINTRAN/NPL-SOURCE/s3vs-4.symb 061511-061607; internal codes in SINTRAN/NPL-SOURCE/SYMBOLS/L07/FILSYS-SYMBOLS.SYMB.TXT.
  • SCSI geometry / block size / READ CAPACITY / shift instructions / READ(6/10): SINTRAN/NPL-SOURCE/NPL/IP-P2-SCSI-DISK.NPL lines 556, 1234-1345, 1480-1533.
  • CHDSI/RXDIR/WXDIR behaviour + rebuild-on-bad-checksum + capacity compare: SINTRAN/Filesystem/NDFS-VALIDATION.md, SINTRAN/Filesystem/on-disk-format/extended-info-block.md, SINTRAN/Filesystem/on-disk-format/directory-label.md; carved segment tools/sintran-segment-carver/versions/L-VSX-500/segments/006-S3FS.bin (load 26000B) and .../re/006-S3FS.annotated.dis.