From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: From: Naoya Horiguchi Subject: Re: [PATCH v2 1/1] mm/memory-failure: Poison read receives SIGKILL instead of SIGBUS if mmaped more than once Date: Wed, 24 Jul 2019 23:43:25 +0000 Message-ID: <20190724234318.GA21820@hori.linux.bs1.fc.nec.co.jp> References: <1564007603-9655-1-git-send-email-jane.chu@oracle.com> <1564007603-9655-2-git-send-email-jane.chu@oracle.com> In-Reply-To: <1564007603-9655-2-git-send-email-jane.chu@oracle.com> Content-Language: ja-JP Content-Type: text/plain; charset="iso-2022-jp" Content-ID: <1F5511DB107F864C919C1A6A91C4628E@gisp.nec.co.jp> Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Sender: owner-linux-mm@kvack.org To: Jane Chu Cc: "linux-mm@kvack.org" , "linux-kernel@vger.kernel.org" , "linux-nvdimm@lists.01.org" List-ID: On Wed, Jul 24, 2019 at 04:33:23PM -0600, Jane Chu wrote: > Mmap /dev/dax more than once, then read the poison location using address > from one of the mappings. The other mappings due to not having the page > mapped in will cause SIGKILLs delivered to the process. SIGKILL succeeds > over SIGBUS, so user process looses the opportunity to handle the UE. >=20 > Although one may add MAP_POPULATE to mmap(2) to work around the issue, > MAP_POPULATE makes mapping 128GB of pmem several magnitudes slower, so > isn't always an option. >=20 > Details - >=20 > ndctl inject-error --block=3D10 --count=3D1 namespace6.0 >=20 > ./read_poison -x dax6.0 -o 5120 -m 2 > mmaped address 0x7f5bb6600000 > mmaped address 0x7f3cf3600000 > doing local read at address 0x7f3cf3601400 > Killed >=20 > Console messages in instrumented kernel - >=20 > mce: Uncorrected hardware memory error in user-access at edbe201400 > Memory failure: tk->addr =3D 7f5bb6601000 > Memory failure: address edbe201: call dev_pagemap_mapping_shift > dev_pagemap_mapping_shift: page edbe201: no PUD > Memory failure: tk->size_shift =3D=3D 0 > Memory failure: Unable to find user space address edbe201 in read_poison > Memory failure: tk->addr =3D 7f3cf3601000 > Memory failure: address edbe201: call dev_pagemap_mapping_shift > Memory failure: tk->size_shift =3D 21 > Memory failure: 0xedbe201: forcibly killing read_poison:22434 because of = failure to unmap corrupted page > =3D> to deliver SIGKILL > Memory failure: 0xedbe201: Killing read_poison:22434 due to hardware memo= ry corruption > =3D> to deliver SIGBUS >=20 > Signed-off-by: Jane Chu > Suggested-by: Naoya Horiguchi > --- > mm/memory-failure.c | 62 ++++++++++++++++++++++-------------------------= ------ > 1 file changed, 26 insertions(+), 36 deletions(-) >=20 > diff --git a/mm/memory-failure.c b/mm/memory-failure.c > index d9cc660..bd4db33 100644 > --- a/mm/memory-failure.c > +++ b/mm/memory-failure.c > @@ -199,7 +199,6 @@ struct to_kill { > struct task_struct *tsk; > unsigned long addr; > short size_shift; > - char addr_valid; > }; > =20 > /* > @@ -304,43 +303,43 @@ static unsigned long dev_pagemap_mapping_shift(stru= ct page *page, > /* > * Schedule a process for later kill. > * Uses GFP_ATOMIC allocations to avoid potential recursions in the VM. > - * TBD would GFP_NOIO be enough? > */ > static void add_to_kill(struct task_struct *tsk, struct page *p, > struct vm_area_struct *vma, > - struct list_head *to_kill, > - struct to_kill **tkc) > + struct list_head *to_kill) > { > struct to_kill *tk; > =20 > - if (*tkc) { > - tk =3D *tkc; > - *tkc =3D NULL; > - } else { > - tk =3D kmalloc(sizeof(struct to_kill), GFP_ATOMIC); > - if (!tk) { > - pr_err("Memory failure: Out of memory while machine check handling\n"= ); > - return; > - } > + tk =3D kmalloc(sizeof(struct to_kill), GFP_ATOMIC); > + if (!tk) { > + pr_err("Memory failure: Out of memory while machine check handling\n")= ; > + return; As Dan pointed out, the cleanup part can be delivered as a separate patch. > } > + > tk->addr =3D page_address_in_vma(p, vma); > - tk->addr_valid =3D 1; > if (is_zone_device_page(p)) > tk->size_shift =3D dev_pagemap_mapping_shift(p, vma); > else > tk->size_shift =3D compound_order(compound_head(p)) + PAGE_SHIFT; > =20 > /* > - * In theory we don't have to kill when the page was > - * munmaped. But it could be also a mremap. Since that's > - * likely very rare kill anyways just out of paranoia, but use > - * a SIGKILL because the error is not contained anymore. > + * Send SIGKILL if "tk->addr =3D=3D -EFAULT". Also, as > + * "tk->size_shift" is always non-zero for !is_zone_device_page(), > + * so "tk->size_shift =3D=3D 0" effectively checks no mapping on > + * ZONE_DEVICE. Indeed, when a devdax page is mmapped N times > + * to a process' address space, it's possible not all N VMAs > + * contain mappings for the page, but at least one VMA does. > + * Only deliver SIGBUS with payload derived from the VMA that > + * has a mapping for the page. OK, so SIGBUSs are sent M times (where M is the number of mappings for the page). Then I'm convinced that we need "else if" block below. Thanks, Naoya Horiguchi > */ > - if (tk->addr =3D=3D -EFAULT || tk->size_shift =3D=3D 0) { > + if (tk->addr =3D=3D -EFAULT) { > pr_info("Memory failure: Unable to find user space address %lx in %s\n= ", > page_to_pfn(p), tsk->comm); > - tk->addr_valid =3D 0; > + } else if (tk->size_shift =3D=3D 0) { > + kfree(tk); > + return; > } > + > get_task_struct(tsk); > tk->tsk =3D tsk; > list_add_tail(&tk->nd, to_kill); > @@ -366,7 +365,7 @@ static void kill_procs(struct list_head *to_kill, int= forcekill, bool fail, > * make sure the process doesn't catch the > * signal and then access the memory. Just kill it. > */ > - if (fail || tk->addr_valid =3D=3D 0) { > + if (fail || tk->addr =3D=3D -EFAULT) { > pr_err("Memory failure: %#lx: forcibly killing %s:%d because of fail= ure to unmap corrupted page\n", > pfn, tk->tsk->comm, tk->tsk->pid); > do_send_sig_info(SIGKILL, SEND_SIG_PRIV, > @@ -432,7 +431,7 @@ static struct task_struct *task_early_kill(struct tas= k_struct *tsk, > * Collect processes when the error hit an anonymous page. > */ > static void collect_procs_anon(struct page *page, struct list_head *to_k= ill, > - struct to_kill **tkc, int force_early) > + int force_early) > { > struct vm_area_struct *vma; > struct task_struct *tsk; > @@ -457,7 +456,7 @@ static void collect_procs_anon(struct page *page, str= uct list_head *to_kill, > if (!page_mapped_in_vma(page, vma)) > continue; > if (vma->vm_mm =3D=3D t->mm) > - add_to_kill(t, page, vma, to_kill, tkc); > + add_to_kill(t, page, vma, to_kill); > } > } > read_unlock(&tasklist_lock); > @@ -468,7 +467,7 @@ static void collect_procs_anon(struct page *page, str= uct list_head *to_kill, > * Collect processes when the error hit a file mapped page. > */ > static void collect_procs_file(struct page *page, struct list_head *to_k= ill, > - struct to_kill **tkc, int force_early) > + int force_early) > { > struct vm_area_struct *vma; > struct task_struct *tsk; > @@ -492,7 +491,7 @@ static void collect_procs_file(struct page *page, str= uct list_head *to_kill, > * to be informed of all such data corruptions. > */ > if (vma->vm_mm =3D=3D t->mm) > - add_to_kill(t, page, vma, to_kill, tkc); > + add_to_kill(t, page, vma, to_kill); > } > } > read_unlock(&tasklist_lock); > @@ -501,26 +500,17 @@ static void collect_procs_file(struct page *page, s= truct list_head *to_kill, > =20 > /* > * Collect the processes who have the corrupted page mapped to kill. > - * This is done in two steps for locking reasons. > - * First preallocate one tokill structure outside the spin locks, > - * so that we can kill at least one process reasonably reliable. > */ > static void collect_procs(struct page *page, struct list_head *tokill, > int force_early) > { > - struct to_kill *tk; > - > if (!page->mapping) > return; > =20 > - tk =3D kmalloc(sizeof(struct to_kill), GFP_NOIO); > - if (!tk) > - return; > if (PageAnon(page)) > - collect_procs_anon(page, tokill, &tk, force_early); > + collect_procs_anon(page, tokill, force_early); > else > - collect_procs_file(page, tokill, &tk, force_early); > - kfree(tk); > + collect_procs_file(page, tokill, force_early); > } > =20 > static const char *action_name[] =3D { > --=20 > 1.8.3.1 >=20 > =