Hi David,
> > There are cases when we try to pin a folio but discover that it has
> > not been faulted-in. So, we try to allocate it in memfd_alloc_folio()
> > but there is a chance that we might encounter a crash/failure
> > (VM_BUG_ON(!h->resv_huge_pages)) if there are no active reservations
> > at that instant. This issue was reported by syzbot:
> >
> > kernel BUG at mm/hugetlb.c:2403!
> > Oops: invalid opcode: 0000 [#1] PREEMPT SMP KASAN NOPTI
> > CPU: 0 UID: 0 PID: 5315 Comm: syz.0.0 Not tainted
> > 6.13.0-rc5-syzkaller-00161-g63676eefb7a0 #0
> > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
> > 1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014
> > RIP: 0010:alloc_hugetlb_folio_reserve+0xbc/0xc0 mm/hugetlb.c:2403
> > Code: 1f eb 05 e8 56 18 a0 ff 48 c7 c7 40 56 61 8e e8 ba 21 cc 09 4c 89
> > f0 5b 41 5c 41 5e 41 5f 5d c3 cc cc cc cc e8 35 18 a0 ff 90 <0f> 0b 66
> > 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 f3 0f
> > RSP: 0018:ffffc9000d3d77f8 EFLAGS: 00010087
> > RAX: ffffffff81ff6beb RBX: 0000000000000000 RCX: 0000000000100000
> > RDX: ffffc9000e51a000 RSI: 00000000000003ec RDI: 00000000000003ed
> > RBP: 1ffffffff34810d9 R08: ffffffff81ff6ba3 R09: 1ffffd4000093005
> > R10: dffffc0000000000 R11: fffff94000093006 R12: dffffc0000000000
> > R13: dffffc0000000000 R14: ffffea0000498000 R15: ffffffff9a4086c8
> > FS: 00007f77ac12e6c0(0000) GS:ffff88801fc00000(0000)
> > knlGS:0000000000000000
> > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > CR2: 00007f77ab54b170 CR3: 0000000040b70000 CR4: 0000000000352ef0
> > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> > Call Trace:
> > <TASK>
> > memfd_alloc_folio+0x1bd/0x370 mm/memfd.c:88
> > memfd_pin_folios+0xf10/0x1570 mm/gup.c:3750
> > udmabuf_pin_folios drivers/dma-buf/udmabuf.c:346 [inline]
> > udmabuf_create+0x70e/0x10c0 drivers/dma-buf/udmabuf.c:443
> > udmabuf_ioctl_create drivers/dma-buf/udmabuf.c:495 [inline]
> > udmabuf_ioctl+0x301/0x4e0 drivers/dma-buf/udmabuf.c:526
> > vfs_ioctl fs/ioctl.c:51 [inline]
> > __do_sys_ioctl fs/ioctl.c:906 [inline]
> > __se_sys_ioctl+0xf5/0x170 fs/ioctl.c:892
> > do_syscall_x64 arch/x86/entry/common.c:52 [inline]
> > do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83
> > entry_SYSCALL_64_after_hwframe+0x77/0x7f
> >
> > Therefore, to avoid this situation and fix this issue, we just need
> > to make a reservation before we try to allocate the folio. While at
> > it, also remove the VM_BUG_ON() as there is no need to crash the
> > system in this scenario and instead we could just fail the allocation.
> >
> > Fixes: 26a8ea80929c ("mm/hugetlb: fix memfd_pin_folios resv_huge_pages
> leak")
> > Reported-by: syzbot+a504cb5bae4fe117ba94@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Signed-off-by: Vivek Kasireddy <vivek.kasireddy@xxxxxxxxx>
> > Cc: Steve Sistare <steven.sistare@xxxxxxxxxx>
> > Cc: Muchun Song <muchun.song@xxxxxxxxx>
> > Cc: David Hildenbrand <david@xxxxxxxxxx>
> > Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > ---
> > mm/hugetlb.c | 9 ++++++---
> > mm/memfd.c | 5 +++++
> > 2 files changed, 11 insertions(+), 3 deletions(-)
> >
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index c498874a7170..e46c461210a4 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -2397,12 +2397,15 @@ struct folio *alloc_hugetlb_folio_reserve(struct
> hstate *h, int preferred_nid,
> > struct folio *folio;
> >
> > spin_lock_irq(&hugetlb_lock);
> > + if (!h->resv_huge_pages) {
>
> Should this be a "if (WARN_ON_ONCE(!h->resv_huge_pages)) {", because the
> "_reserve" in the function indicates that this is indeed something that
> must never happen?
Yes, adding a WARN_ON_ONCE makes sense in this case. I'll add it in v2.
>
> > + spin_unlock_irq(&hugetlb_lock);
> > + return NULL;
> > + }
> > +
> > folio = dequeue_hugetlb_folio_nodemask(h, gfp_mask,
> preferred_nid,
> > nmask);
> > - if (folio) {
> > - VM_BUG_ON(!h->resv_huge_pages);
> > + if (folio)
> > h->resv_huge_pages--;
> > - }
> >
> > spin_unlock_irq(&hugetlb_lock);
> > return folio;
> > diff --git a/mm/memfd.c b/mm/memfd.c
> > index 35a370d75c9a..a3012c444285 100644
> > --- a/mm/memfd.c
> > +++ b/mm/memfd.c
> > @@ -85,6 +85,10 @@ struct folio *memfd_alloc_folio(struct file *memfd,
> pgoff_t idx)
> > gfp_mask &= ~(__GFP_HIGHMEM | __GFP_MOVABLE);
> > idx >>= huge_page_order(h);
> >
> > + if (!hugetlb_reserve_pages(file_inode(memfd),
> > + idx, idx + 1, NULL, 0))
> > + return ERR_PTR(-ENOMEM);
> > +
> > folio = alloc_hugetlb_folio_reserve(h,
> > numa_node_id(),
> > NULL,
> > @@ -100,6 +104,7 @@ struct folio *memfd_alloc_folio(struct file *memfd,
> pgoff_t idx)
> > folio_unlock(folio);
> > return folio;
> > }
> > + hugetlb_unreserve_pages(file_inode(memfd), idx, idx + 1, 1);
> > return ERR_PTR(-ENOMEM);
>
> Staring at hugetlb_reserve_pages() I assume this will also work as
> expected if already reserved.
IIUC, if a range is already reserved, and when hugetlb_reserve_pages()
gets called for the same range, it would mostly become a no-op as
region_chg() and region_add() return 0 (based on my light testing).
Thanks,
Vivek
>
>
> --
> Cheers,
>
> David / dhildenb
