Re: [PATCH v2] vhost-vdpa: fix page pinning leakage in error path (rework)

From: Jason Wang
Date: Sun Nov 08 2020 - 22:22:20 EST



On 2020/11/6 上午6:57, si-wei liu wrote:

On 11/4/2020 7:26 PM, Jason Wang wrote:

On 2020/11/5 上午7:33, Si-Wei Liu wrote:
Pinned pages are not properly accounted particularly when
mapping error occurs on IOTLB update. Clean up dangling
pinned pages for the error path.

The memory usage for bookkeeping pinned pages is reverted
to what it was before: only one single free page is needed.
This helps reduce the host memory demand for VM with a large
amount of memory, or in the situation where host is running
short of free memory.

Fixes: 4c8cf31885f6 ("vhost: introduce vDPA-based backend")
Signed-off-by: Si-Wei Liu <si-wei.liu@xxxxxxxxxx>
---
Changes in v2:
- Drop the reversion patch
- Fix unhandled page leak towards the end of page_list

  drivers/vhost/vdpa.c | 79 ++++++++++++++++++++++++++++++++++++++++------------
  1 file changed, 61 insertions(+), 18 deletions(-)

diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index b6d9016..e112854 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c
@@ -560,6 +560,8 @@ static int vhost_vdpa_map(struct vhost_vdpa *v,
        if (r)
          vhost_iotlb_del_range(dev->iotlb, iova, iova + size - 1);
+    else
+        atomic64_add(size >> PAGE_SHIFT, &dev->mm->pinned_vm);
        return r;
  }
@@ -591,14 +593,16 @@ static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v,
      unsigned long list_size = PAGE_SIZE / sizeof(struct page *);
      unsigned int gup_flags = FOLL_LONGTERM;
      unsigned long npages, cur_base, map_pfn, last_pfn = 0;
-    unsigned long locked, lock_limit, pinned, i;
+    unsigned long lock_limit, sz2pin, nchunks, i;
      u64 iova = msg->iova;
+    long pinned;
      int ret = 0;
        if (vhost_iotlb_itree_first(iotlb, msg->iova,
                      msg->iova + msg->size - 1))
          return -EEXIST;
  +    /* Limit the use of memory for bookkeeping */
      page_list = (struct page **) __get_free_page(GFP_KERNEL);
      if (!page_list)
          return -ENOMEM;
@@ -607,52 +611,75 @@ static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v,
          gup_flags |= FOLL_WRITE;
        npages = PAGE_ALIGN(msg->size + (iova & ~PAGE_MASK)) >> PAGE_SHIFT;
-    if (!npages)
-        return -EINVAL;
+    if (!npages) {
+        ret = -EINVAL;
+        goto free;
+    }
        mmap_read_lock(dev->mm);
  -    locked = atomic64_add_return(npages, &dev->mm->pinned_vm);
      lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
-
-    if (locked > lock_limit) {
+    if (npages + atomic64_read(&dev->mm->pinned_vm) > lock_limit) {
          ret = -ENOMEM;
-        goto out;
+        goto unlock;
      }
        cur_base = msg->uaddr & PAGE_MASK;
      iova &= PAGE_MASK;
+    nchunks = 0;
        while (npages) {
-        pinned = min_t(unsigned long, npages, list_size);
-        ret = pin_user_pages(cur_base, pinned,
-                     gup_flags, page_list, NULL);
-        if (ret != pinned)
+        sz2pin = min_t(unsigned long, npages, list_size);
+        pinned = pin_user_pages(cur_base, sz2pin,
+                    gup_flags, page_list, NULL);
+        if (sz2pin != pinned) {
+            if (pinned < 0) {
+                ret = pinned;
+            } else {
+                unpin_user_pages(page_list, pinned);
+                ret = -ENOMEM;
+            }
              goto out;
+        }
+        nchunks++;
            if (!last_pfn)
              map_pfn = page_to_pfn(page_list[0]);
  -        for (i = 0; i < ret; i++) {
+        for (i = 0; i < pinned; i++) {
              unsigned long this_pfn = page_to_pfn(page_list[i]);
              u64 csize;
                if (last_pfn && (this_pfn != last_pfn + 1)) {
                  /* Pin a contiguous chunk of memory */
                  csize = (last_pfn - map_pfn + 1) << PAGE_SHIFT;
-                if (vhost_vdpa_map(v, iova, csize,
-                           map_pfn << PAGE_SHIFT,
-                           msg->perm))
+                ret = vhost_vdpa_map(v, iova, csize,
+                             map_pfn << PAGE_SHIFT,
+                             msg->perm);
+                if (ret) {
+                    /*
+                     * Unpin the pages that are left unmapped
+                     * from this point on in the current
+                     * page_list. The remaining outstanding
+                     * ones which may stride across several
+                     * chunks will be covered in the common
+                     * error path subsequently.
+                     */
+                    unpin_user_pages(&page_list[i],
+                             pinned - i);


Can we simply do last_pfn = this_pfn here?
Nope. They are not contiguous segments of memory. Noted the conditional (this_pfn != last_pfn + 1) being held here.


Right.





                      goto out;
+                }
+
                  map_pfn = this_pfn;
                  iova += csize;
+                nchunks = 0;
              }
                last_pfn = this_pfn;
          }
  -        cur_base += ret << PAGE_SHIFT;
-        npages -= ret;
+        cur_base += pinned << PAGE_SHIFT;
+        npages -= pinned;
      }
        /* Pin the rest chunk */
@@ -660,10 +687,26 @@ static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v,
                   map_pfn << PAGE_SHIFT, msg->perm);
  out:
      if (ret) {
+        if (nchunks && last_pfn) {


Any reason for checking last_pfn here?

Note that we did:

+        nchunks++;

         if (!last_pfn)
             map_pfn = page_to_pfn(page_list[0]);
It's for explicit coding to make sure this common error path can be reused no matter if last_pfn has a sane value assigned or not. I can change it to an implicit WARN_ON() if need be.


Just to make sure I understand. A question, when will we get nchunks != 0 but last_pfn == 0?

Thanks


Thanks,
-Siwei



Thanks


+            unsigned long pfn;
+
+            /*
+             * Unpin the outstanding pages which are yet to be
+             * mapped but haven't due to vdpa_map() or
+             * pin_user_pages() failure.
+             *
+             * Mapped pages are accounted in vdpa_map(), hence
+             * the corresponding unpinning will be handled by
+             * vdpa_unmap().
+             */
+            for (pfn = map_pfn; pfn <= last_pfn; pfn++)
+                unpin_user_page(pfn_to_page(pfn));
+        }
          vhost_vdpa_unmap(v, msg->iova, msg->size);
-        atomic64_sub(npages, &dev->mm->pinned_vm);
      }
+unlock:
      mmap_read_unlock(dev->mm);
+free:
      free_page((unsigned long)page_list);
      return ret;
  }