T2_Dircache_attack.c File Reference

Test the following: More...

#include "afpcmd.h"
#include "afphelper.h"
#include "testhelper.h"

Functions
STATIC void	test500 ()
STATIC void	test501 ()
STATIC void	test502 ()
STATIC void	test503 ()
STATIC void	test504 ()
STATIC void	test505 ()
STATIC void	test506 ()
STATIC void	test508 ()
	test508: Dircache stress test with cache filling and evictions
static int	access_working_set (uint16_t vol, int start_idx, int end_idx, int passes, const char *prefix)
	Helper: Access a working set of directories multiple times.
static int	sequential_scan (uint16_t vol, int start_idx, int end_idx, const char *prefix)
	Helper: Perform sequential scan of directories.
STATIC void	test509 ()
	test509: Scan resistance test with realistic workload pattern
void	Dircache_attack_test ()

Detailed Description

Test the following:

test500()
=========
 
client 1:
   mkdir dir1
   mkdir dir2
 
client 2:
   mv dir1 dir2/renamed
 
client 1:
   ls dir2
   stat dir2/renamed
 
Check: CNID must not change
Targets: enumerate()
 
test501()
=========
 
client 1:
   mkdir dir1
   mkdir dir2
 
client 2:
   mv dir1 dir2/renamed
 
client 1:
   stat renamed
 
Check: CNID must not change
Targets: getfildirparms()
 
 
test502()
=========
 
client 1:
   mkdir dir1
   mkdir dir2
 
client 2:
   mv dir1 dir2/renamed
 
client 1:
   ls renamed
 
Check: CNID must not change
Targets: enumerate()
 
 
test503()
=========
 
client 1:
   mkdir dir1
   mkdir dir2
 
client 2:
   mv dir1 dir2/renamed
 
client 1:
   stat renamed
 
Check: CNID must not change
Targets: getfildirparms()
 
 
test504()
=========
 
client 1:
   mkdir -p dir1/dir2
   touch dir1/dir2/file
 
client 2:
   mv dir1 renamed1
 
client 1:
   stat file
 
Check: CNID must not change
Targets: getfildirparms()
 
 
test505()
=========
 
client 1:
   mkdir -p dir1/dir2
 
client 2:
   mv dir1 renamed1
 
client 1:
   stat dir2
 
Check: CNID must not change
Targets: getfildirparms()
 
 
test506()
=========
 
client 1:
   mkdir -p dir1/dir2
 
client 2:
   mv dir1 renamed1
   mkdir dir1
 
client 1:
   stat dir2

Function Documentation

access_working_set()

int access_working_set ( uint16_t vol, int start_idx, int end_idx, int passes, const char * prefix )

static

Helper: Access a working set of directories multiple times.

Accesses directories in the specified range multiple times to establish them as "frequently accessed" which should promote them to T2 in ARC.

Parameters

[in]	vol	Volume ID
[in]	start_idx	Starting directory index
[in]	end_idx	Ending directory index (exclusive)
[in]	passes	Number of times to access each directory
[in]	prefix	Directory name prefix

Returns

Number of successful accesses

Dircache_attack_test()

void Dircache_attack_test

(

)

sequential_scan()

int sequential_scan ( uint16_t vol, int start_idx, int end_idx, const char * prefix )

static

Helper: Perform sequential scan of directories.

Sequentially accesses a range of directories once, simulating a scan operation that should pass through ARC's T1 without evicting T2 entries, but would evict LRU cache entries.

Parameters

[in]	vol	Volume ID
[in]	start_idx	Starting directory index
[in]	end_idx	Ending directory index (exclusive)
[in]	prefix	Directory name prefix

Returns

Number of successful accesses

test500()

STATIC void test500

(

)

move and rename dir, enumerate new parent, stat renamed dir

test501()

STATIC void test501

(

)

move and rename dir, then stat it

test502()

STATIC void test502

(

)

move and rename dir, enumerate renamed dir

test503()

STATIC void test503

(

)

move and rename dir, stat renamed dir

test504()

STATIC void test504

(

)

rename topdir, stat file in subdir of renamed topdir

test505()

STATIC void test505

(

)

rename dir, stat subdir in renamed dir

test506()

STATIC void test506

(

)

stat subdir in poisened path

test508()

STATIC void test508

(

)

test508: Dircache stress test with cache filling and evictions

Test Requirements:

• Configure AFP_DIRCACHESIZE=1024 (small cache for reasonable test time)
• Works with both LRU and ARC modes

This test exercises core dircache behaviors by creating more directories than can fit in cache, forcing evictions and re-access patterns to test cache management under stress.

Test Pattern:

1. Create 2100 directories (2x cache size)
   • During creation: periodically re-access early dirs to establish frequency patterns
2. Multiple re-access passes over different ranges
   • Pass 1-3: Repeatedly access dirs 500-900 (tests frequent access handling)
   • Pass 4: Access dirs 800-1100 (overlapping range)
   • Pass 5: Access dirs 1000-1200 (different range for pattern adaptation)

Expected Behavior:

• LRU: Simple eviction of least recently used entries, all re-accesses are cache hits if not evicted
• ARC: Evictions to ghost lists (B1/B2), ghost hits on re-access, parameter adaptation

Verification: Check logs for evictions, cache hits, and (in ARC) ghost hits and adaptations

test509()

STATIC void test509

(

)

test509: Scan resistance test with realistic workload pattern

Test Requirements:

• Configure AFP_DIRCACHESIZE=1024 (small cache, working set fits)
• Works with both LRU and ARC modes for performance comparison

This test provides a realistic workload pattern that simulates common AFP usage: frequently accessed directories (hot working set) mixed with sequential scans (one-time directory traversals). This pattern is optimized for ARC's frequency detection, but also exposes LRU's limitations for comparison purposes.

Realistic Workload Pattern:

• Hot working set (300 dirs): User's active projects/folders (accessed 5x)
• Warm working set (200 dirs): Recently used folders (accessed 2x)
• Sequential scans (1500 dirs): Spotlight indexing, backups, virus scans

Test Pattern (5 cycles of working set + scan):

1. Create 2100 directories (cache=1024)
2. Build hot working set: Access 300 directories 5x (very frequent)
3. Build warm working set: Access 200 directories 2x (moderately frequent) 4-13. Five cycles of:
   • Sequential scan: Access 1500 directories once (1.5x cache size)
   • Verify hot set: Access 300 hot directories 2x
   • Verify warm set: Access 200 warm directories 1x

Expected Behavior:

• LRU: Each scan evicts all working sets → Re-access = misses every cycle
• ARC: Hot/warm sets stay in T2 → Re-access = hits every cycle

Key Differences from test508:

• Separate hot (5x) and warm (2x) working sets to test frequency detection
• Multiple scan/verify cycles (5×) to show sustained advantage
• Larger scans (1500 dirs = 1.5x cache) to stress scan resistance
• Realistic workload pattern amplifies LRU vs ARC performance differences