On lDOS (lMS-DOS) LH leads to "MCB chain corrupt, or MS-DOS incompatible system."

[ecm-pushbx]

The message "MCB chain corrupt, or MS-DOS incompatible system." is displayed on attempts to use LH on a recent lDOS kernel.

With this patch it works:

diff --git a/shell/loadhigh.c b/shell/loadhigh.c
index 81340ec..49d4c37 100644
--- a/shell/loadhigh.c
+++ b/shell/loadhigh.c
@@ -422,7 +422,7 @@ static int loadhigh_prepare(void)
 
   /* Set the UMB link and malloc strategy */
   dosSetUMBLinkState(1);
-  dosSetAllocStrategy(0);
+  dosSetAllocStrategy(0x80);
 
   if ((availBlock = malloc(256 * sizeof(*availBlock))) == 0)
     return err_out_of_memory;
@@ -462,8 +462,10 @@ static int loadhigh_prepare(void)
         /* Found a free memory block: allocate it */
         word bl = DosAlloc(mcb->mcb_size);
 
-        if (bl != FP_SEG(mcb) + 1)  /* Did we get the block we wanted? */
+        if (bl != FP_SEG(mcb) + 1)  /* Did we get the block we wanted? */ {
+          printf("bl = %04X, mcb = %04X\n", bl, FP_SEG(mcb));
           return err_mcb_chain;
+        }
 
         if (region->access)		/* /L option allows access to this region */
         {
@@ -540,6 +542,7 @@ static int loadfix_prepare(void)
 
       if (bl != FP_SEG(mcb) + 1)  /* Did we get the block we wanted? */
       	{
+          printf("bl = %04X, mcb = %04X\n", bl, FP_SEG(mcb));
         return err_mcb_chain;
         }
 

Without the line 422 hunk the debug message that I added to loadhigh_prepare (line 462 hunk) is displayed so that seems to be where the error code is generated.

I don't see how with a strategy of 0 ("LMA then UMA, as one area") the allocation should occur in the UMA first as this code expects. I will have to study the FreeDOS kernel to see whether it behaves differently.

[ecm-pushbx]

With this patch:

diff --git a/shell/loadhigh.c b/shell/loadhigh.c
index 81340ec..65223e1 100644
--- a/shell/loadhigh.c
+++ b/shell/loadhigh.c
@@ -422,6 +422,7 @@ static int loadhigh_prepare(void)
 
   /* Set the UMB link and malloc strategy */
   dosSetUMBLinkState(1);
+  /* dosSetAllocStrategy(0x80); */
   dosSetAllocStrategy(0);
 
   if ((availBlock = malloc(256 * sizeof(*availBlock))) == 0)
@@ -462,8 +463,12 @@ static int loadhigh_prepare(void)
         /* Found a free memory block: allocate it */
         word bl = DosAlloc(mcb->mcb_size);
 
-        if (bl != FP_SEG(mcb) + 1)  /* Did we get the block we wanted? */
+        printf("bl = %04X, mcb = %04X, strat = %04X\n",
+          bl, FP_SEG(mcb), (int)dosGetAllocStrategy());
+        if (bl != FP_SEG(mcb) + 1)  /* Did we get the block we wanted? */ {
+          printf("Error unexpected allocation!\n", bl, FP_SEG(mcb));
           return err_mcb_chain;
+        }
 
         if (region->access)		/* /L option allows access to this region */
         {
@@ -540,6 +545,7 @@ static int loadfix_prepare(void)
 
       if (bl != FP_SEG(mcb) + 1)  /* Did we get the block we wanted? */
       	{
+          printf("bl = %04X, mcb = %04X\n", bl, FP_SEG(mcb));
         return err_mcb_chain;
         }
 

On FreeDOS we get:

C:\>lh tsr.com
bl = 02C3, mcb = 02C2, strat = 0000
bl = B11C, mcb = B11B, strat = 0000
bl = C1FB, mcb = C1FA, strat = 0000
bl = F001, mcb = F000, strat = 0000

On lDOS we get:

C:\>lh tsr.com
bl = 0283, mcb = B230, strat = 0000
Error unexpected allocation!
MCB chain corrupt, or MS-DOS incompatible system.

Also, the memory allocated (at segment 283h) seems to leak.

[ecm-pushbx]

The && mcb->mcb_size > 0 condition in this line is the culprit:

freecom/shell/loadhigh.c

Lines 456 to 458 in 20e61b5

	for (mcbAssign(mcb, region->start)
	; FP_SEG(mcb) < region->end && mcb->mcb_type == 'M' && mcb->mcb_size > 0
	; mcbNext(mcb))

In the current lDOS, with things loaded into the HMA or UMA, the only parts remaining in the LMA are:

• IVT, BDA, DOSENTRY, DOSDATA (all before the first MCB)
• UPBs, DPBs (both in an lDOS style S MCB rather than an SD MCB)
• The empty SD / devmark MCB (size of memory block = 0)

This last one causes FreeCOM to skip the region in the condition I referenced. MEM also has a problem with this: https://gitlab.com/FreeDOS/base/mem/-/merge_requests/1

[ecm-pushbx]

Another problem: allocatedBlocks isn't checked for an array overflow of array blocks. It seems that blocks is allocated for 256 entries:

freecom/shell/loadhigh.c

Lines 156 to 158 in 20e61b5

	/* Allocate dynamic memory for some arrays */
	if ((block = malloc(256 * sizeof(*block))) == 0)
	return err_out_of_memory;

After that if there are more than 256 blocks to reserve, the array will overflow.