From: Bob Pearson <rpearsonhpe@gmail.com>
To: Zhu Yanjun <zyjzyj2000@gmail.com>, linux-rdma@vger.kernel.org
Cc: Bob Pearson <rpearson@hpe.com>
Subject: Re: [PATCH v3 11/17] rdma_rxe: Address an issue with hardened user copy
Date: Fri, 21 Aug 2020 23:16:59 -0500 [thread overview]
Message-ID: <f69f8a27-e4e6-88ae-77d8-358fde60d72e@gmail.com> (raw)
In-Reply-To: <4fd91289-7cd7-a62c-54ee-4ace9eb45a14@gmail.com>
On 8/21/20 10:32 PM, Zhu Yanjun wrote:
> On 8/21/2020 6:46 AM, Bob Pearson wrote:
>> Added a new feature to pools to let driver white list a region of
>> a pool object. This removes a kernel oops caused when create qp
>> returns the qp number so the next patch will work without errors.
>>
>> Signed-off-by: Bob Pearson <rpearson@hpe.com>
>> ---
>> drivers/infiniband/sw/rxe/rxe_pool.c | 20 +++++++++++++++++---
>> drivers/infiniband/sw/rxe/rxe_pool.h | 4 ++++
>> 2 files changed, 21 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/infiniband/sw/rxe/rxe_pool.c b/drivers/infiniband/sw/rxe/rxe_pool.c
>> index 5679714827ec..374e56689d30 100644
>> --- a/drivers/infiniband/sw/rxe/rxe_pool.c
>> +++ b/drivers/infiniband/sw/rxe/rxe_pool.c
>> @@ -40,9 +40,12 @@ struct rxe_type_info rxe_type_info[RXE_NUM_TYPES] = {
>> .name = "rxe-qp",
>> .size = sizeof(struct rxe_qp),
>> .cleanup = rxe_qp_cleanup,
>> - .flags = RXE_POOL_INDEX,
>> + .flags = RXE_POOL_INDEX
>> + | RXE_POOL_WHITELIST,
>> .min_index = RXE_MIN_QP_INDEX,
>> .max_index = RXE_MAX_QP_INDEX,
>> + .user_offset = offsetof(struct rxe_qp, ibqp.qp_num),
>> + .user_size = sizeof(u32),
>> },
>> [RXE_TYPE_CQ] = {
>> .name = "rxe-cq",
>> @@ -116,10 +119,21 @@ int rxe_cache_init(void)
>> type = &rxe_type_info[i];
>> size = ALIGN(type->size, RXE_POOL_ALIGN);
>> if (!(type->flags & RXE_POOL_NO_ALLOC)) {
>> - type->cache =
>> - kmem_cache_create(type->name, size,
>> + if (type->flags & RXE_POOL_WHITELIST) {
>> + type->cache =
>> + kmem_cache_create_usercopy(
>> + type->name, size,
>> + RXE_POOL_ALIGN,
>> + RXE_POOL_CACHE_FLAGS,
>> + type->user_offset,
>> + type->user_size, NULL);
>> + } else {
>> + type->cache =
>> + kmem_cache_create(type->name, size,
>> RXE_POOL_ALIGN,
>> RXE_POOL_CACHE_FLAGS, NULL);
>> + }
>> +
>> if (!type->cache) {
>> pr_err("Unable to init kmem cache for %s\n",
>> type->name);
>> diff --git a/drivers/infiniband/sw/rxe/rxe_pool.h b/drivers/infiniband/sw/rxe/rxe_pool.h
>> index 664153bf9392..fc5b584a8137 100644
>> --- a/drivers/infiniband/sw/rxe/rxe_pool.h
>> +++ b/drivers/infiniband/sw/rxe/rxe_pool.h
>> @@ -17,6 +17,7 @@ enum rxe_pool_flags {
>> RXE_POOL_INDEX = BIT(1),
>> RXE_POOL_KEY = BIT(2),
>> RXE_POOL_NO_ALLOC = BIT(4),
>> + RXE_POOL_WHITELIST = BIT(5),
>> };
>> enum rxe_elem_type {
>> @@ -44,6 +45,9 @@ struct rxe_type_info {
>> u32 min_index;
>> size_t key_offset;
>> size_t key_size;
>> + /* for white listing where necessary */
>
> s/where/when
>
>
>> + unsigned int user_offset;
>> + unsigned int user_size;
>> struct kmem_cache *cache;
>> };
>>
>
>
The reason for this change is that every time I do anything with rdma_rxe on current head of tree I get a kernel oops with a warning that there is a bad or missing white list. I traced this back to the user_copy routine which (recently) decided that when you copy just a part of a kernel memory object stored in a kmem cache that this represented a risk of leaking information from the kernel to user space. For the QP object the qp_num is copied back to user space in the user API. They also provided a new kmem_ccache_create_usercopy call that allows you to specify a 'whitelisted' portion of each object with an offset and length. So I just made it a feature of pools since it may come up again instead of treating QPs differently that all the other objects. This is part of a general program to harden the Linux kernel.
You can see the change to rxe_cache_init in the same file. Perhaps just dropping the comment would address the concern. See an earlier post I made with a pointer to an article in lwn describing the changes to the kernel.
next prev parent reply other threads:[~2020-08-22 4:17 UTC|newest]
Thread overview: 34+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-08-20 22:46 [PATCH v3 00/17] Memory window support for rdma_rxe Bob Pearson
2020-08-20 22:46 ` [PATCH v3 01/17] rdma_rxe: Added SPDX headers to rxe source files Bob Pearson
2020-08-24 10:03 ` Leon Romanovsky
2020-08-20 22:46 ` [PATCH v3 02/17] rdma_rxe: Fixed style warnings Bob Pearson
2020-08-27 12:52 ` Jason Gunthorpe
2020-08-20 22:46 ` [PATCH v3 03/17] ib_user_verbs.h: Added ib_uverbs_wc_opcode Bob Pearson
2020-08-27 12:53 ` Jason Gunthorpe
2020-08-20 22:46 ` [PATCH v3 04/17] ib_verbs.h: Added missing IB_WR_BIND_MW opcode Bob Pearson
2020-08-27 12:54 ` Jason Gunthorpe
2020-08-20 22:46 ` [PATCH v3 05/17] rdma_rxe: Added bind_mw parameters to rxe_send_wr Bob Pearson
2020-08-27 12:56 ` Jason Gunthorpe
2020-08-20 22:46 ` [PATCH v3 06/17] rdma_rxe: Added stubs for alloc_mw and dealloc_mw verbs Bob Pearson
2020-08-27 12:56 ` Jason Gunthorpe
2020-08-20 22:46 ` [PATCH v3 07/17] rdma_rxe: Separated MR and MW objects Bob Pearson
2020-08-20 22:46 ` [PATCH v3 08/17] rdma_rxe: Added mw object Bob Pearson
2020-08-22 3:39 ` Zhu Yanjun
2020-08-20 22:46 ` [PATCH v3 09/17] rdma_rxe: Extended pools to support both keys and indices Bob Pearson
2020-08-20 22:46 ` [PATCH v3 10/17] rdma_rxe: Implemented functional alloc_mw and dealloc_mw APIs Bob Pearson
2020-08-20 22:46 ` [PATCH v3 11/17] rdma_rxe: Address an issue with hardened user copy Bob Pearson
2020-08-22 3:32 ` Zhu Yanjun
2020-08-22 4:16 ` Bob Pearson [this message]
2020-08-24 8:47 ` Leon Romanovsky
2020-08-24 8:52 ` Leon Romanovsky
2020-08-24 23:52 ` Bob Pearson
2020-08-25 5:04 ` Leon Romanovsky
2020-08-20 22:46 ` [PATCH v3 12/17] rdma_rxe: Added bind mw API stub Bob Pearson
2020-08-20 22:46 ` [PATCH v3 13/17] rdma_rxe: Give MR and MW objects indices and keys Bob Pearson
2020-08-20 22:46 ` [PATCH v3 14/17] rdma_rxe: Added stub for invalidate mw Bob Pearson
2020-08-20 22:46 ` [PATCH v3 15/17] rdma_rxe: Added functional bind and invalidate MW ops Bob Pearson
2020-08-20 22:46 ` [PATCH v3 16/17] rdma_rxe: Implemented read/write/atomic access to MW Bob Pearson
2020-08-20 22:46 ` [PATCH v3 17/17] rdma_rxe: minor cleanups Bob Pearson
[not found] ` <a153a775-9b53-3ccc-4c2a-ec76f863d1a1@gmail.com>
2020-08-22 4:05 ` Bob Pearson
2020-08-24 9:02 ` Leon Romanovsky
2020-08-27 13:00 ` Jason Gunthorpe
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=f69f8a27-e4e6-88ae-77d8-358fde60d72e@gmail.com \
--to=rpearsonhpe@gmail.com \
--cc=linux-rdma@vger.kernel.org \
--cc=rpearson@hpe.com \
--cc=zyjzyj2000@gmail.com \
/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;
as well as URLs for NNTP newsgroup(s).