[LU-12580] usercopy exposure attempt detected in LL_IOC_LOV_GETSTRIPE ioctl

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.14.0, Lustre 2.12.5
• Affects Version/s: Upstream
• Environment: CentOS 7.6.1810, starting with kernel 3.10.0-957
• Severity: 3
• Assignee: Marco Grossi (Inactive)
• Reporter: Marco Grossi (Inactive)

Description

Software like darshan are still using the old `ioctl` way to gather striping info for a file.

The kernel BUG is easily triggered on a not-PFL striped file, using:

lum->lmm_magic = LOV_USER_MAGIC;
lum->lmm_stripe_count = LOV_MAX_STRIPE_COUNT;

Any `lmm_stripe_count` greater than the actual file's stripe count will trigger the bug.

Kernel side the issue appears to be in `lov_getstripe`: with a positive `lum_size`(line 409), `lmm_size` is set as `lum_size`(line 442) even if `lmm_magic != LOV_MAGIC_COMP_V1`(line 414), while instead the structure is just as big as `lmmk_size`:

404         if (lum.lmm_magic == LOV_USER_MAGIC_V1 ||
405             lum.lmm_magic == LOV_USER_MAGIC_V3)
406                 lum_size = lov_user_md_size(lum.lmm_stripe_count,
407                                             lum.lmm_magic);
408 
409         if (lum_size != 0) {
410                 struct lov_mds_md *comp_md = lmmk;
411 
412                 /* Legacy app (ADIO for instance) treats the layout as V1/V3
413                  * blindly, we'd return a reasonable V1/V3 for them. */
414                 if (lmmk->lmm_magic == LOV_MAGIC_COMP_V1) {
[...]
439                 }
440 
441                 lmm = comp_md;
442                 lmm_size = lum_size;
443         } else {
444                 lmm = lmmk;
445                 lmm_size = lmmk_size;
446         }
447         /**
448          * User specified limited buffer size, usually the buffer is
449          * from ll_lov_setstripe(), and the buffer can only hold basic
450          * layout template info.
451          */
452         if (size == 0 || size > lmm_size)
453                 size = lmm_size;
454         if (copy_to_user(lump, lmm, size))
455                 GOTO(out_free, rc = -EFAULT);

Please find as attachment the kernel trace and a reproducer, to be invoked as:

$ gcc reproducer.c -o reproducer -W -Wall --pedantic
$ ./reproducer <file_path>

Issue Links

• is duplicated by: LU-13596 usercopy: kernel memory exposure attempt detected from ffff98c06ba17d80 (kmalloc-128) (48032 bytes)
• is related to: LU-14316 lfs: using old ioctl(LL_IOC_LOV_GETSTRIPE), use llapi_layout_get_by_path()