Fix missing entries in simplefs_lookup

[ibat10clw]

simplefs_lookup must scan all entries in a directory before it can determine whether a file exists.

The current implementation stops early when it encounters a hole in the directory, which can cause later files to be missed. In addition, fi is used as the index into dblock, so the loop should be bounded by SIMPLEFS_FILES_PER_BLOCK rather than dblock->nr_files. If a file is located near the end of the directory block, the loop may terminate early when fi >= dblock->nr_files, resulting in an incomplete scan.

Fix this by skipping holes during filename comparison and continuing to advance fi until all live directory entries have been examined.

This issue can be reproduced with a freshly formatted simplefs image using
the following steps.

Before this patch

$ sudo touch test/{1..10}.txt
$ sudo rm test/1.txt
$ echo 3 | sudo tee /proc/sys/vm/drop_caches
$ sudo touch test/9.txt
$ ls test/
10.txt  2.txt  3.txt  4.txt  5.txt  6.txt  7.txt  8.txt  9.txt  9.txt

After this patch

$ sudo touch test/{1..10}.txt
$ sudo rm test/1.txt
$ echo 3 | sudo tee /proc/sys/vm/drop_caches
$ sudo touch test/9.txt
$ ls test/
10.txt  2.txt  3.txt  4.txt  5.txt  6.txt  7.txt  8.txt  9.txt 

---

Summary by cubic

Ensure simplefs_lookup scans the whole directory block and skips holes so files aren’t missed, including entries near the end. This fixes false “not found” results and eliminates duplicate-looking listings after deletions.

• Bug Fixes
  • Continue scanning across holes; compare names only when f->inode is set.
  • Bound the loop by SIMPLEFS_FILES_PER_BLOCK and track live entries with a local nr_files.
  • Advance with fi += f->nr_blk to correctly traverse sparse directory blocks.

^{Written for commit f85250a. Summary will update on new commits.}

[jserv]

I defer to @RoyWFHuang for confirmation.

[RoyWFHuang]

I agree. The early exit on holes is exactly what causes files to be missed in the scan.