Segments signedness issue

[fabianfreyer]

Version and Platform (required):

• Binary Ninja Version: 3.0.3222-dev (2022-01-26 06:24:12)
• OS: macOS
• OS Version: 11

Bug Description: Segment addresses and lengths seem to have a signedness issue.

Steps To Reproduce:

1. Create a small test binary:

   truncate -s 4 empty

2. open with binja, select any arch.

3. Run the following commands in the python shell:

   # can change these lengths arbitrarily within size
   len1 = 1
   len2 = 2
   bv.add_user_segment(0, len1, 0, len1, 5)
   bv.add_user_segment(len1, len2, len1, len2, 7)

   Segments now look like this:

   Type: Mapped
   Platform: x86_64
   Architecture: x86_64
   
   Segments:
   r-x  0x00000000-0x00000001 
   rwx  0x00000001-0x00000003 
   r-x  0x00000003-0x00000004 

4. Save bndb

5. Open bndb Segments now look like this:

   Type: Mapped
   Platform: x86_64
   Architecture: x86_64
   
   Segments:
   r-x  0x00000000-0x00000004 
   r-x  0xfffffffffffffffd-0xfffffffffffffffe 
   rwx  0xfffffffffffffffe-0x00000000 

Expected Behavior: Segments should be the same before and after saving.

Screenshots: Before saving:

After saving:

[fabianfreyer]

Closing in favor of #2882

[psifertex]

2882 has been renamed to describe a more broad feature that would include this fix as a super set but has been pushed out further to 3.2 so we're opening this to track just the existing bug and potentially a quicker temporary fix in the core during creation of user segments when no auto-segments exist such that this issue no longer occurs.

[psifertex]

Also, note that we believe that using the open-with-options JSON specification for segments is a temporary work-around in the meantime.

[bpotchik]

Previously, normalized offsets were based on bv.start. Now normalized offsets are based on bv.image_base. This allows us to support adding 'user' segments and/or memory regions before the image base. Now, rebasing and save/load operations preserve the memory map correctly.

Fixed in 4.2.5758.