From: Elliot Berman <quic_eberman@quicinc.com>
To: Andrew Morton <akpm@linux-foundation.org>,
Sean Christopherson <seanjc@google.com>,
Paolo Bonzini <pbonzini@redhat.com>,
Thomas Gleixner <tglx@linutronix.de>,
Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
Dave Hansen <dave.hansen@linux.intel.com>,
Fuad Tabba <tabba@google.com>,
David Hildenbrand <david@redhat.com>,
Patrick Roy <roypat@amazon.co.uk>, <qperret@google.com>,
Ackerley Tng <ackerleytng@google.com>,
Mike Rapoport <rppt@kernel.org>, <x86@kernel.org>,
"H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-kernel@vger.kernel.org>, <linux-mm@kvack.org>,
<kvm@vger.kernel.org>, <linux-coco@lists.linux.dev>,
<linux-arm-msm@vger.kernel.org>,
Elliot Berman <quic_eberman@quicinc.com>
Subject: [PATCH RFC v2 5/5] mm: guest_memfd: Add option to remove inaccessible memory from direct map
Date: Thu, 29 Aug 2024 15:24:13 -0700 [thread overview]
Message-ID: <20240829-guest-memfd-lib-v2-5-b9afc1ff3656@quicinc.com> (raw)
In-Reply-To: <20240829-guest-memfd-lib-v2-0-b9afc1ff3656@quicinc.com>
When memory is made inaccessible to the host, Linux may still
speculatively access the folio if a load_unaligned_zeropad is performed
at the end of the prior page. To ensure Linux itself catches such errors
without hypervisor crashing Linux, unmap the guest-inaccessible pages
from the direct map.
This feature is made optional because arm64 pKVM can provide a special,
detectable fault which can be fixed up directly.
Signed-off-by: Elliot Berman <quic_eberman@quicinc.com>
---
include/linux/guest_memfd.h | 1 +
mm/guest_memfd.c | 79 +++++++++++++++++++++++++++++++++++++++++++--
2 files changed, 78 insertions(+), 2 deletions(-)
diff --git a/include/linux/guest_memfd.h b/include/linux/guest_memfd.h
index 66e5d3ab42613..de53bce15db99 100644
--- a/include/linux/guest_memfd.h
+++ b/include/linux/guest_memfd.h
@@ -33,6 +33,7 @@ enum guest_memfd_grab_flags {
enum guest_memfd_create_flags {
GUEST_MEMFD_FLAG_CLEAR_INACCESSIBLE = (1UL << 0),
+ GUEST_MEMFD_FLAG_REMOVE_DIRECT_MAP = (1UL << 1),
};
struct folio *guest_memfd_grab_folio(struct file *file, pgoff_t index, u32 flags);
diff --git a/mm/guest_memfd.c b/mm/guest_memfd.c
index 194b2c3ea1525..d4232739d4c5b 100644
--- a/mm/guest_memfd.c
+++ b/mm/guest_memfd.c
@@ -8,6 +8,7 @@
#include <linux/falloc.h>
#include <linux/guest_memfd.h>
#include <linux/pagemap.h>
+#include <linux/set_memory.h>
#include <linux/wait.h>
#include "internal.h"
@@ -26,6 +27,45 @@ struct guest_memfd_private {
atomic_t safe;
};
+static inline int folio_set_direct_map_invalid_noflush(struct folio *folio)
+{
+ unsigned long i, nr = folio_nr_pages(folio);
+ int r;
+
+ for (i = 0; i < nr; i++) {
+ struct page *page = folio_page(folio, i);
+
+ r = set_direct_map_invalid_noflush(page);
+ if (r)
+ goto out_remap;
+ }
+ /**
+ * Currently no need to flush as hypervisor will also be flushing
+ * tlb when giving the folio to guest.
+ */
+
+ return 0;
+out_remap:
+ for (; i > 0; i--) {
+ struct page *page = folio_page(folio, i - 1);
+
+ BUG_ON(set_direct_map_default_noflush(page));
+ }
+
+ return r;
+}
+
+static inline void folio_set_direct_map_default_noflush(struct folio *folio)
+{
+ unsigned long i, nr = folio_nr_pages(folio);
+
+ for (i = 0; i < nr; i++) {
+ struct page *page = folio_page(folio, i);
+
+ BUG_ON(set_direct_map_default_noflush(page));
+ }
+}
+
static inline int base_safe_refs(struct folio *folio)
{
/* 1 for filemap */
@@ -131,6 +171,12 @@ struct folio *guest_memfd_grab_folio(struct file *file, pgoff_t index, u32 flags
goto out_free;
}
} else {
+ if (gmem_flags & GUEST_MEMFD_FLAG_REMOVE_DIRECT_MAP) {
+ r = folio_set_direct_map_invalid_noflush(folio);
+ if (r < 0)
+ goto out_free;
+ }
+
if (ops->prepare_inaccessible) {
r = ops->prepare_inaccessible(inode, folio);
if (r < 0)
@@ -203,6 +249,7 @@ int guest_memfd_make_accessible(struct folio *folio)
struct guest_memfd_private *private = folio_get_private(folio);
struct inode *inode = folio_inode(folio);
struct guest_memfd_operations *ops = inode->i_private;
+ unsigned long gmem_flags;
int r;
/*
@@ -218,6 +265,10 @@ int guest_memfd_make_accessible(struct folio *folio)
if (!r)
return -EBUSY;
+ gmem_flags = (unsigned long)inode->i_mapping->i_private_data;
+ if (gmem_flags & GUEST_MEMFD_FLAG_REMOVE_DIRECT_MAP)
+ folio_set_direct_map_default_noflush(folio);
+
if (ops->prepare_accessible) {
r = ops->prepare_accessible(inode, folio);
if (r)
@@ -248,6 +299,7 @@ int guest_memfd_make_inaccessible(struct folio *folio)
struct guest_memfd_private *private = folio_get_private(folio);
struct inode *inode = folio_inode(folio);
struct guest_memfd_operations *ops = inode->i_private;
+ unsigned long gmem_flags;
int r;
r = atomic_dec_if_positive(&private->accessible);
@@ -266,6 +318,13 @@ int guest_memfd_make_inaccessible(struct folio *folio)
goto err;
}
+ gmem_flags = (unsigned long)inode->i_mapping->i_private_data;
+ if (gmem_flags & GUEST_MEMFD_FLAG_REMOVE_DIRECT_MAP) {
+ r = folio_set_direct_map_invalid_noflush(folio);
+ if (r)
+ goto err;
+ }
+
if (ops->prepare_inaccessible) {
r = ops->prepare_inaccessible(inode, folio);
if (r)
@@ -454,6 +513,7 @@ static int gmem_error_folio(struct address_space *mapping, struct folio *folio)
struct guest_memfd_operations *ops = inode->i_private;
off_t offset = folio->index;
size_t nr = folio_nr_pages(folio);
+ unsigned long gmem_flags;
int ret;
filemap_invalidate_lock_shared(mapping);
@@ -464,6 +524,10 @@ static int gmem_error_folio(struct address_space *mapping, struct folio *folio)
filemap_invalidate_unlock_shared(mapping);
+ gmem_flags = (unsigned long)inode->i_mapping->i_private_data;
+ if (gmem_flags & GUEST_MEMFD_FLAG_REMOVE_DIRECT_MAP)
+ folio_set_direct_map_default_noflush(folio);
+
return ret;
}
@@ -474,7 +538,7 @@ static bool gmem_release_folio(struct folio *folio, gfp_t gfp)
struct guest_memfd_operations *ops = inode->i_private;
off_t offset = folio->index;
size_t nr = folio_nr_pages(folio);
- unsigned long val, expected;
+ unsigned long val, expected, gmem_flags;
int ret;
ret = ops->invalidate_begin(inode, offset, nr);
@@ -483,6 +547,10 @@ static bool gmem_release_folio(struct folio *folio, gfp_t gfp)
if (ops->invalidate_end)
ops->invalidate_end(inode, offset, nr);
+ gmem_flags = (unsigned long)inode->i_mapping->i_private_data;
+ if (gmem_flags & GUEST_MEMFD_FLAG_REMOVE_DIRECT_MAP)
+ folio_set_direct_map_default_noflush(folio);
+
expected = base_safe_refs(folio);
val = atomic_read(&private->safe);
WARN_ONCE(val != expected, "folio[%x] safe ref: %d != expected %d\n",
@@ -518,7 +586,14 @@ static inline bool guest_memfd_check_ops(const struct guest_memfd_operations *op
static inline unsigned long guest_memfd_valid_flags(void)
{
- return GUEST_MEMFD_FLAG_CLEAR_INACCESSIBLE;
+ unsigned long flags = GUEST_MEMFD_FLAG_CLEAR_INACCESSIBLE;
+
+#ifdef CONFIG_ARCH_HAS_SET_DIRECT_MAP
+ if (can_set_direct_map())
+ flags |= GUEST_MEMFD_FLAG_REMOVE_DIRECT_MAP;
+#endif
+
+ return flags;
}
/**
--
2.34.1
next prev parent reply other threads:[~2024-08-29 22:24 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-08-29 22:24 [PATCH RFC v2 0/5] mm: Introduce guest_memfd library Elliot Berman
2024-08-29 22:24 ` [PATCH RFC v2 1/5] mm: Introduce guest_memfd Elliot Berman
2024-08-29 22:24 ` [PATCH RFC v2 2/5] mm: guest_memfd: Allow folios to be accessible to host Elliot Berman
2024-10-09 20:14 ` Vishal Annapurve
2024-08-29 22:24 ` [PATCH RFC v2 3/5] kvm: Convert to use guest_memfd library Elliot Berman
2024-10-23 1:18 ` Mike Day
2024-08-29 22:24 ` [PATCH RFC v2 4/5] mm: guest_memfd: Add ability for userspace to mmap pages Elliot Berman
2024-08-29 22:24 ` Elliot Berman [this message]
2024-10-10 13:04 ` [PATCH RFC v2 0/5] mm: Introduce guest_memfd library Paolo Bonzini
2024-10-10 21:20 ` Elliot Berman
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20240829-guest-memfd-lib-v2-5-b9afc1ff3656@quicinc.com \
--to=quic_eberman@quicinc.com \
--cc=ackerleytng@google.com \
--cc=akpm@linux-foundation.org \
--cc=bp@alien8.de \
--cc=dave.hansen@linux.intel.com \
--cc=david@redhat.com \
--cc=hpa@zytor.com \
--cc=kvm@vger.kernel.org \
--cc=linux-arm-msm@vger.kernel.org \
--cc=linux-coco@lists.linux.dev \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mingo@redhat.com \
--cc=pbonzini@redhat.com \
--cc=qperret@google.com \
--cc=roypat@amazon.co.uk \
--cc=rppt@kernel.org \
--cc=seanjc@google.com \
--cc=tabba@google.com \
--cc=tglx@linutronix.de \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox