扫码分享
验证：
体验盒子|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 |
Source: https://code.google.com/p/google-security-research/issues/detail?id=492 The Samsung Graphics 2D driver (/dev/fimg2d) is accessible by unprivileged users/applications. It was found that the ioctl implementation for this driver contains a locking error which can lead to memory errors (such as use-after-free) due to a race condition. The key observation is in the locking routine definitions in fimg2d.h: #ifdef BLIT_WORKQUE #define g2d_lock(x) do {} while (0) #define g2d_unlock(x) do {} while (0) #define g2d_spin_lock(x, f) spin_lock_irqsave(x, f) #define g2d_spin_unlock(x, f) spin_unlock_irqrestore(x, f) #else #define g2d_lock(x) mutex_lock(x) #define g2d_unlock(x) mutex_unlock(x) #define g2d_spin_lock(x, f) do { f = 0; } while (0) #define g2d_spin_unlock(x, f) do { f = 0; } while (0) #endif This means that the g2d_lock/g2d_unlock routines are no-ops when BLIT_WORKQUE is defined, which appears to be the default configuration. Unfortunately the alternative spin lock routines are not used consistently with this configuration. For example, the FIMG2D_BITBLT_BLIT ioctl command (with notes annotated as "PZ"): ctx = file->private_data; /* PZ: ctx allocated at open(), lives on the heap. */ switch (cmd) { case FIMG2D_BITBLT_BLIT: mm = get_task_mm(current); if (!mm) { fimg2d_err("no mm for ctx\n"); return -ENXIO; } g2d_lock(&ctrl->drvlock); /* PZ: This is a no-op. */ ctx->mm = mm; ret = fimg2d_add_command(ctrl, ctx, (struct fimg2d_blit __user *)arg); if (ret) { ... } ret = fimg2d_request_bitblt(ctrl, ctx); /* PZ: Does stuff with the ctx. */ if (ret) { ... } g2d_unlock(&ctrl->drvlock); /* PZ: Another no-op */ As the lock macros are no-ops, a second process can change ctx->mm when the original process is still using the same ctx->mm (as long as it has access to the same file descriptor). Reproduction steps: Open /dev/fimg2d Fork to get two processes with different mm’s with the access to the fd Concurrently call the FIMG2D_BITBLT_BLIT ioctl from both processes. One ioctl should have valid data, the other should fail At this point ctx->mm will now have invalid or free data (free if the forked process dies). Proof-of-concept code to trigger this condition is attached (fimg2d-lock.c) Proof of Concept: https://gitlab.com/exploit-database/exploitdb-bin-sploits/-/raw/main/bin-sploits/38557.zip |
体验盒子
