From: Fengnan Chang <changfengnan@bytedance.com>
To: Olivier Matz <olivier.matz@6wind.com>
Cc: david.marchand@redhat.com, mb@smartsharesystems.com, dev@dpdk.org
Subject: Re: [External] Re: [PATCH v2] mempool: fix rte_mempool_avail_count may segment fault when used in multiprocess
Date: Tue, 29 Nov 2022 17:57:05 +0800 [thread overview]
Message-ID: <CAPFOzZu9QfCmWFrQd1ByA-oHA6btsYxbfDAGnp-+Jfz+hood9Q@mail.gmail.com> (raw)
In-Reply-To: <Y3zpy+WYve3l3d88@platinum>
Olivier Matz <olivier.matz@6wind.com> 于2022年11月22日周二 23:25写道：
>
> Hi,
>
> On Tue, Nov 15, 2022 at 08:35:02PM +0800, Fengnan Chang wrote:
> > rte_mempool_create put tailq entry into rte_mempool_tailq list before
> > populate, and pool_data set when populate. So in multi process, if
> > process A create mempool, and process B can get mempool through
> > rte_mempool_lookup before pool_data set, if B call rte_mempool_avail_count,
> > it will cause segment fault.
> >
> > Fix this by put tailq entry into rte_mempool_tailq after populate.
> >
> > Signed-off-by: Fengnan Chang <changfengnan@bytedance.com>
> > ---
> > lib/mempool/rte_mempool.c | 43 ++++++++++++++++++++++-----------------
> > 1 file changed, 24 insertions(+), 19 deletions(-)
> >
> > diff --git a/lib/mempool/rte_mempool.c b/lib/mempool/rte_mempool.c
> > index 4c78071a34..b3a6572fc8 100644
> > --- a/lib/mempool/rte_mempool.c
> > +++ b/lib/mempool/rte_mempool.c
> > @@ -155,6 +155,27 @@ get_min_page_size(int socket_id)
> > return wa.min == SIZE_MAX ? (size_t) rte_mem_page_size() : wa.min;
> > }
> >
> > +static int
> > +add_mempool_to_list(struct rte_mempool *mp)
> > +{
> > + struct rte_mempool_list *mempool_list;
> > + struct rte_tailq_entry *te = NULL;
> > +
> > + /* try to allocate tailq entry */
> > + te = rte_zmalloc("MEMPOOL_TAILQ_ENTRY", sizeof(*te), 0);
> > + if (te == NULL) {
> > + RTE_LOG(ERR, MEMPOOL, "Cannot allocate tailq entry!\n");
> > + return -ENOMEM;
> > + }
> > +
> > + te->data = mp;
> > + mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
> > + rte_mcfg_tailq_write_lock();
> > + TAILQ_INSERT_TAIL(mempool_list, te, next);
> > + rte_mcfg_tailq_write_unlock();
> > +
> > + return 0;
> > +}
> >
> > static void
> > mempool_add_elem(struct rte_mempool *mp, __rte_unused void *opaque,
> > @@ -304,6 +325,9 @@ mempool_ops_alloc_once(struct rte_mempool *mp)
> > if (ret != 0)
> > return ret;
> > mp->flags |= RTE_MEMPOOL_F_POOL_CREATED;
> > + ret = add_mempool_to_list(mp);
> > + if (ret != 0)
> > + return ret;
>
> One issue here is that if the rte_zmalloc("MEMPOOL_TAILQ_ENTRY") fails,
> the function will fail, but rte_mempool_ops_alloc() may already be
> successful.
>
> I agree it's theorical, because an allocation failure would cause more
> issues at the end. But, to be rigorous, I think we should do something
> like this instead (not tested, just for the idea):
>
> static int
> mempool_ops_alloc_once(struct rte_mempool *mp)
> {
> struct rte_mempool_list *mempool_list;
> struct rte_tailq_entry *te = NULL;
> int ret;
>
> /* only create the driver ops and add in tailq in if not already done */
> if ((mp->flags & RTE_MEMPOOL_F_POOL_CREATED))
> return 0;
>
> te = rte_zmalloc("MEMPOOL_TAILQ_ENTRY", sizeof(*te), 0);
> if (te == NULL) {
> RTE_LOG(ERR, MEMPOOL, "Cannot allocate tailq entry!\n");
> ret = -rte_errno;
> goto fail;
> }
> te->data = mp;
> mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
>
> ret = rte_mempool_ops_alloc(mp);
> if (ret != 0)
> goto fail;
>
> mp->flags |= RTE_MEMPOOL_F_POOL_CREATED;
> rte_mcfg_tailq_write_lock();
> TAILQ_INSERT_TAIL(mempool_list, te, next);
> rte_mcfg_tailq_write_unlock();
>
> return 0;
>
> fail:
> rte_free(te);
> return ret;
> }
>
>
> Thinking a bit more about the problem itself: the segfault that you
> describe could also happen in a primary, without multi-process:
> - create an empty mempool
> - call rte_mempool_avail_count() before it is populated
>
> This simply means that an empty mempool is not ready for use, until
> rte_mempool_set_ops_byname() or rte_mempool_populate*() is called. This
> is something that we should document above the declaration of
> rte_mempool_create_empty(). We could also say there that the mempool
> will become visible to the secondary processes as soon as the driver ops
> are set.
>
> However I still believe that a better synchronization point is required
> in the application. After all, the presence in the TAILQ does not give
> any hint on the status of the object. Can we imagine a case where a
> mempool is created empty in a primary, and populated in a secondary? If
> such use-case exist, we may not want to take this patch.
Maybe there is a case like you said, do you think adding check mempool flags in
rte_mempool_avail_count is acceptable ?
If RTE_MEMPOOL_F_POOL_CREATED not set, just return 0.
>
> > }
> > return 0;
> > }
> > @@ -798,9 +822,7 @@ rte_mempool_create_empty(const char *name, unsigned n, unsigned elt_size,
> > int socket_id, unsigned flags)
> > {
> > char mz_name[RTE_MEMZONE_NAMESIZE];
> > - struct rte_mempool_list *mempool_list;
> > struct rte_mempool *mp = NULL;
> > - struct rte_tailq_entry *te = NULL;
> > const struct rte_memzone *mz = NULL;
> > size_t mempool_size;
> > unsigned int mz_flags = RTE_MEMZONE_1GB|RTE_MEMZONE_SIZE_HINT_ONLY;
> > @@ -820,8 +842,6 @@ rte_mempool_create_empty(const char *name, unsigned n, unsigned elt_size,
> > RTE_CACHE_LINE_MASK) != 0);
> > #endif
> >
> > - mempool_list = RTE_TAILQ_CAST(rte_mempool_tailq.head, rte_mempool_list);
> > -
> > /* asked for zero items */
> > if (n == 0) {
> > rte_errno = EINVAL;
> > @@ -866,14 +886,6 @@ rte_mempool_create_empty(const char *name, unsigned n, unsigned elt_size,
> > private_data_size = (private_data_size +
> > RTE_MEMPOOL_ALIGN_MASK) & (~RTE_MEMPOOL_ALIGN_MASK);
> >
> > -
> > - /* try to allocate tailq entry */
> > - te = rte_zmalloc("MEMPOOL_TAILQ_ENTRY", sizeof(*te), 0);
> > - if (te == NULL) {
> > - RTE_LOG(ERR, MEMPOOL, "Cannot allocate tailq entry!\n");
> > - goto exit_unlock;
> > - }
> > -
> > mempool_size = RTE_MEMPOOL_HEADER_SIZE(mp, cache_size);
> > mempool_size += private_data_size;
> > mempool_size = RTE_ALIGN_CEIL(mempool_size, RTE_MEMPOOL_ALIGN);
> > @@ -923,20 +935,13 @@ rte_mempool_create_empty(const char *name, unsigned n, unsigned elt_size,
> > cache_size);
> > }
> >
> > - te->data = mp;
> > -
> > - rte_mcfg_tailq_write_lock();
> > - TAILQ_INSERT_TAIL(mempool_list, te, next);
> > - rte_mcfg_tailq_write_unlock();
> > rte_mcfg_mempool_write_unlock();
> > -
> > rte_mempool_trace_create_empty(name, n, elt_size, cache_size,
> > private_data_size, flags, mp);
> > return mp;
> >
> > exit_unlock:
> > rte_mcfg_mempool_write_unlock();
> > - rte_free(te);
> > rte_mempool_free(mp);
> > return NULL;
> > }
> > --
> > 2.37.0 (Apple Git-136)
> >
next prev parent reply other threads:[~2022-11-29 16:59 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-11-15 12:35 Fengnan Chang
2022-11-22 15:24 ` Olivier Matz
2022-11-29 9:57 ` Fengnan Chang [this message]
2023-07-17 16:43 ` [External] " Stephen Hemminger
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=CAPFOzZu9QfCmWFrQd1ByA-oHA6btsYxbfDAGnp-+Jfz+hood9Q@mail.gmail.com \
--to=changfengnan@bytedance.com \
--cc=david.marchand@redhat.com \
--cc=dev@dpdk.org \
--cc=mb@smartsharesystems.com \
--cc=olivier.matz@6wind.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).