Skip to contentSkip to conversation

net_dma: fix memory leak in dma_pin_iocvec_pages

PATCH Applied Roman Gushchin 3 participants 7 messages Sep 3, 2014 – Sep 28, 2014 dmaengine

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net_dma: fix memory leak in dma_pin_iocvec_pages

dma_pin_iovec_pages() calls get_user_pages() for each iovec. If get_user_pages() returns a number smaller than the requested number, dma_pin_iovec_pages() calls dma_unpin_iovec_pages(). It releases previously allocated iovecs, but pages pinned by last get_user_pages() call remain unreleased.
Fix this by calling put_page() for each such page.

---
 drivers/dma/iovlock.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/dma/iovlock.c +4 -1
index bb48a57..c393cf9 100644
@@ -107,8 +107,11 @@ struct dma_pinned_list *dma_pin_iovec_pages(struct iovec *iov, size_t len)
 			NULL);
 		up_read(&current->mm->mmap_sem);
 
-		if (ret != page_list->nr_pages)
+		if (ret != page_list->nr_pages) {
+			for (i = 0; i < ret; i++)
+				put_page(page_list->pages[i]);
 			goto unpin;
+		}
 
 		local_list->nr_iovecs = i + 1;
 	}
Signature · 1 line
1.9.3
#1 lore raw

Ping?

03.09.2014, 13:38, "Roman Gushchin" <[email protected]>:

Quoted 37 lines

 dma_pin_iovec_pages() calls get_user_pages() for each iovec. If get_user_pages() returns a number smaller than the requested number, dma_pin_iovec_pages() calls dma_unpin_iovec_pages(). It releases previously allocated iovecs, but pages pinned by last get_user_pages() call remain unreleased.
 Fix this by calling put_page() for each such page.

 ---
  drivers/dma/iovlock.c | 5 ++++-

  1 file changed, 4 insertions(+), 1 deletion(-)

 diff --git a/drivers/dma/iovlock.c b/drivers/dma/iovlock.c
 index bb48a57..c393cf9 100644
 --- a/drivers/dma/iovlock.c
 +++ b/drivers/dma/iovlock.c
 @@ -107,8 +107,11 @@ struct dma_pinned_list *dma_pin_iovec_pages(struct iovec *iov, size_t len) NULL);
                  up_read(&current->mm->mmap_sem);

 - if (ret != page_list->nr_pages)
 + if (ret != page_list->nr_pages) {
 + for (i = 0; i < ret; i++)
 + put_page(page_list->pages[i]);
                          goto unpin;
 + }

                  local_list->nr_iovecs = i + 1;
          }
 --
 1.9.3

 --
 To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
 More majordomo info at  http://vger.kernel.org/majordomo-info.html
 Please read the FAQ at  http://www.tux.org/lkml/

Applied
#2 lore raw
In reply to Roman Gushchin
Quoted 7 lines · Roman Gushchin

dma_pin_iovec_pages() calls get_user_pages() for each iovec. If get_user_pages() returns a number smaller than the requested number, dma_pin_iovec_pages() calls dma_unpin_iovec_pages(). It releases previously allocated iovecs, but pages pinned by last get_user_pages() call remain unreleased.
Fix this by calling put_page() for each such page.

Applied, thanks

Signature · 1 line
~Vinod
#3 lore raw
Quoted 9 lines · Vinod Koul
Quoted 7 lines · Roman Gushchin

dma_pin_iovec_pages() calls get_user_pages() for each iovec. If get_user_pages() returns a number smaller than the requested number, dma_pin_iovec_pages() calls dma_unpin_iovec_pages(). It releases previously allocated iovecs, but pages pinned by last get_user_pages() call remain unreleased.
Fix this by calling put_page() for each such page.

Applied, thanks

Vinod, I have a patch in my queue that completely removes the remnants of NET_DMA. It's been deprecated for more than a few cycles now without any reports to reinstate it. Time to push the final removal patch. I'll rebase it on top of this so it doesn't collide.

#4 lore raw
Quoted 9 lines · Dan Williams
Quoted 8 lines · Vinod Koul
Quoted 7 lines · Roman Gushchin

dma_pin_iovec_pages() calls get_user_pages() for each iovec. If get_user_pages() returns a number smaller than the requested number, dma_pin_iovec_pages() calls dma_unpin_iovec_pages(). It releases previously allocated iovecs, but pages pinned by last get_user_pages() call remain unreleased.
Fix this by calling put_page() for each such page.

Quoted 6 lines
Quoted 1 line

Applied, thanks

Vinod, I have a patch in my queue that completely removes the remnants of NET_DMA. It's been deprecated for more than a few cycles now without any reports to reinstate it. Time to push the final removal patch. I'll rebase it on top of this so it doesn't collide.

Okay, thanks for update. If you wnat I can remove it, so you don't need to rebase.

Signature · 1 line
~Vinod
#5 lore raw
Quoted 12 lines · Vinod Koul
Quoted 11 lines · Dan Williams
Quoted 9 lines · Vinod Koul
Quoted 7 lines · Roman Gushchin

dma_pin_iovec_pages() calls get_user_pages() for each iovec. If get_user_pages() returns a number smaller than the requested number, dma_pin_iovec_pages() calls dma_unpin_iovec_pages(). It releases previously allocated iovecs, but pages pinned by last get_user_pages() call remain unreleased.
Fix this by calling put_page() for each such page.

Applied, thanks

Quoted 8 lines
Quoted 4 lines

Vinod, I have a patch in my queue that completely removes the remnants of NET_DMA. It's been deprecated for more than a few cycles now without any reports to reinstate it. Time to push the final removal patch. I'll rebase it on top of this so it doesn't collide.

Okay, thanks for update. If you wnat I can remove it, so you don't need to rebase.

Yes, please drop it. I'm sending the full removal to Linus. I added Roman to the Reported-by and cc'd stable.

#6 lore raw
Quoted 18 lines · Dan Williams
Quoted 17 lines · Vinod Koul
Quoted 15 lines · Dan Williams
Quoted 9 lines · Vinod Koul
Quoted 7 lines · Roman Gushchin

dma_pin_iovec_pages() calls get_user_pages() for each iovec. If get_user_pages() returns a number smaller than the requested number, dma_pin_iovec_pages() calls dma_unpin_iovec_pages(). It releases previously allocated iovecs, but pages pinned by last get_user_pages() call remain unreleased.
Fix this by calling put_page() for each such page.

Applied, thanks

Vinod, I have a patch in my queue that completely removes the remnants of NET_DMA. It's been deprecated for more than a few cycles now without any reports to reinstate it. Time to push the final removal patch. I'll rebase it on top of this so it doesn't collide.

Quoted 6 lines
Quoted 3 lines

Okay, thanks for update. If you wnat I can remove it, so you don't need to rebase.

Yes, please drop it. I'm sending the full removal to Linus. I added Roman to the Reported-by and cc'd stable.

Done, i have pushed the updated tree.

Signature · 1 line
~Vinod