Re: [Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Okay, IC. So we have to take care of all errors for ocfs2_write_begin_nolock. On 2016/9/14 16:43, Eric Ren wrote: > Hi Joseph, > > On 09/14/2016 04:25 PM, Joseph Qi wrote: >> Hi Eric, >> Sorry for the delayed response. >> I have got your explanation. So we have to unlock the page only in case >> of retry, right? >> If so, I think the unlock should be right before "goto try_again". > No, the mmapped page should be unlocked as long as we cannot return > VM_FAULT_LOCKED > to do_page_mkpage(). Otherwise, the deadlock will happen in do_page_mkpage(). > Please > see the recent 2 mails;-) > > Eric >> >> Thanks, >> Joseph >> >> On 2016/9/14 16:04, Eric Ren wrote: >>> Hi Joseph, >> In ocfs2_write_begin_nolock(), we first grab the pages and then >> allocate disk space for this write; ocfs2_try_to_free_truncate_log() >> will be called if ENOSPC is turned; if we're lucky to get enough >> clusters, >> which is usually the case, we start over again. But in >> ocfs2_free_write_ctxt() >> the target page isn't unlocked, so we will deadlock when trying to grab >> the target page again. > IMO, in ocfs2_grab_pages_for_write, mmap_page is mapping to w_pages and > w_target_locked is set to true, and then will be unlocked by > ocfs2_unlock_pages in ocfs2_free_write_ctxt. > So I'm not getting the case "page isn't unlock". Could you please explain > it in more detail? Thanks for review;-) Follow up the calling chain: ocfs2_free_write_ctxt() ->ocfs2_unlock_pages() in ocfs2_unlock_pages (https://github.com/torvalds/linux/blob/master/fs/ocfs2/aops.c#L793), we can see the code just put_page(target_page), but not unlock it. >>> Did this answer your question? >>> >>> Thanks, >>> Eric Yeah, I will think this a bit more like: why not unlock the target_page there? Is there other potential problems if the "ret" is not "-ENOSPC" but other possible error code? Thanks, Eric > Thanks, > Joseph > >> Fix this issue by unlocking the target page after we fail to allocate >> enough space at the first time. >> >> Jan Kara helps me clear out the JBD2 part, and suggest the hint for root >> cause. >> >> Signed-off-by: Eric Ren >> --- >>fs/ocfs2/aops.c | 7 +++ >>1 file changed, 7 insertions(+) >> >> diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c >> index 98d3654..78d1d67 100644 >> --- a/fs/ocfs2/aops.c >> +++ b/fs/ocfs2/aops.c >> @@ -1860,6 +1860,13 @@ out: >> */ >>try_free = 0; >>+/* >> + * Unlock mmap_page because the page has been locked when we >> + * are here. >> + */ >> +if (mmap_page) >> +unlock_page(mmap_page); >> + >>ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); >>if (ret1 == 1) >>goto try_again; >> > ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel >>> >> >> > > > . > ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel
Re: [Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Hi Joseph, On 09/14/2016 04:25 PM, Joseph Qi wrote: > Hi Eric, > Sorry for the delayed response. > I have got your explanation. So we have to unlock the page only in case > of retry, right? > If so, I think the unlock should be right before "goto try_again". No, the mmapped page should be unlocked as long as we cannot return VM_FAULT_LOCKED to do_page_mkpage(). Otherwise, the deadlock will happen in do_page_mkpage(). Please see the recent 2 mails;-) Eric > > Thanks, > Joseph > > On 2016/9/14 16:04, Eric Ren wrote: >> Hi Joseph, > In ocfs2_write_begin_nolock(), we first grab the pages and then > allocate disk space for this write; ocfs2_try_to_free_truncate_log() > will be called if ENOSPC is turned; if we're lucky to get enough clusters, > which is usually the case, we start over again. But in > ocfs2_free_write_ctxt() > the target page isn't unlocked, so we will deadlock when trying to grab > the target page again. IMO, in ocfs2_grab_pages_for_write, mmap_page is mapping to w_pages and w_target_locked is set to true, and then will be unlocked by ocfs2_unlock_pages in ocfs2_free_write_ctxt. So I'm not getting the case "page isn't unlock". Could you please explain it in more detail? >>> Thanks for review;-) Follow up the calling chain: >>> >>> ocfs2_free_write_ctxt() >>> ->ocfs2_unlock_pages() >>> >>> in ocfs2_unlock_pages >>> (https://github.com/torvalds/linux/blob/master/fs/ocfs2/aops.c#L793), we >>> can see the code just put_page(target_page), but not unlock it. >> Did this answer your question? >> >> Thanks, >> Eric >>> Yeah, I will think this a bit more like: >>> why not unlock the target_page there? Is there other potential problems if >>> the "ret" is not "-ENOSPC" but >>> other possible error code? >>> >>> Thanks, >>> Eric >>> Thanks, Joseph > Fix this issue by unlocking the target page after we fail to allocate > enough space at the first time. > > Jan Kara helps me clear out the JBD2 part, and suggest the hint for root > cause. > > Signed-off-by: Eric Ren > --- >fs/ocfs2/aops.c | 7 +++ >1 file changed, 7 insertions(+) > > diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c > index 98d3654..78d1d67 100644 > --- a/fs/ocfs2/aops.c > +++ b/fs/ocfs2/aops.c > @@ -1860,6 +1860,13 @@ out: > */ >try_free = 0; >+/* > + * Unlock mmap_page because the page has been locked when we > + * are here. > + */ > +if (mmap_page) > +unlock_page(mmap_page); > + >ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); >if (ret1 == 1) >goto try_again; > >>> >>> >>> >>> ___ >>> Ocfs2-devel mailing list >>> Ocfs2-devel@oss.oracle.com >>> https://oss.oracle.com/mailman/listinfo/ocfs2-devel >> > > ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel
Re: [Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Hi Eric, Sorry for the delayed response. I have got your explanation. So we have to unlock the page only in case of retry, right? If so, I think the unlock should be right before "goto try_again". Thanks, Joseph On 2016/9/14 16:04, Eric Ren wrote: > Hi Joseph, In ocfs2_write_begin_nolock(), we first grab the pages and then allocate disk space for this write; ocfs2_try_to_free_truncate_log() will be called if ENOSPC is turned; if we're lucky to get enough clusters, which is usually the case, we start over again. But in ocfs2_free_write_ctxt() the target page isn't unlocked, so we will deadlock when trying to grab the target page again. >>> IMO, in ocfs2_grab_pages_for_write, mmap_page is mapping to w_pages and >>> w_target_locked is set to true, and then will be unlocked by >>> ocfs2_unlock_pages in ocfs2_free_write_ctxt. >>> So I'm not getting the case "page isn't unlock". Could you please explain >>> it in more detail? >> Thanks for review;-) Follow up the calling chain: >> >> ocfs2_free_write_ctxt() >> ->ocfs2_unlock_pages() >> >> in ocfs2_unlock_pages >> (https://github.com/torvalds/linux/blob/master/fs/ocfs2/aops.c#L793), we >> can see the code just put_page(target_page), but not unlock it. > Did this answer your question? > > Thanks, > Eric >> >> Yeah, I will think this a bit more like: >> why not unlock the target_page there? Is there other potential problems if >> the "ret" is not "-ENOSPC" but >> other possible error code? >> >> Thanks, >> Eric >> >>> >>> Thanks, >>> Joseph >>> Fix this issue by unlocking the target page after we fail to allocate enough space at the first time. Jan Kara helps me clear out the JBD2 part, and suggest the hint for root cause. Signed-off-by: Eric Ren --- fs/ocfs2/aops.c | 7 +++ 1 file changed, 7 insertions(+) diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c index 98d3654..78d1d67 100644 --- a/fs/ocfs2/aops.c +++ b/fs/ocfs2/aops.c @@ -1860,6 +1860,13 @@ out: */ try_free = 0; +/* + * Unlock mmap_page because the page has been locked when we + * are here. + */ +if (mmap_page) +unlock_page(mmap_page); + ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); if (ret1 == 1) goto try_again; >>> >>> >> >> >> >> >> ___ >> Ocfs2-devel mailing list >> Ocfs2-devel@oss.oracle.com >> https://oss.oracle.com/mailman/listinfo/ocfs2-devel > > ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel
Re: [Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Hi, On 09/12/2016 11:06 AM, Eric Ren wrote: > Hi, >>> IMO, in ocfs2_grab_pages_for_write, mmap_page is mapping to w_pages and >>> w_target_locked is set to true, and then will be unlocked by >>> ocfs2_unlock_pages in ocfs2_free_write_ctxt. >>> So I'm not getting the case "page isn't unlock". Could you please explain >>> it in more detail? >> Thanks for review;-) Follow up the calling chain: >> >> ocfs2_free_write_ctxt() >> ->ocfs2_unlock_pages() >> >> in ocfs2_unlock_pages >> (https://github.com/torvalds/linux/blob/master/fs/ocfs2/aops.c#L793), we >> can see the code just put_page(target_page), but not unlock it. >> >> Yeah, I will think this a bit more like: >> why not unlock the target_page there? Is there other potential problems if >> the "ret" is >> not "-ENOSPC" but >> other possible error code? > 1. ocfs2_unlock_pages() will be called in ocfs2_write_end_nolock(), in this > case, we > definitely want to return a locked mmaped page > to VM code (do_page_mkwrite) when VM_FAULT_LOCKED is set. > > 2. But there's indeed a potential existing deadlock situation: > ocfs2_grab_pages_for_write() ==> return > -ENOMEM and with > the mmaped page locked > ocfs2_free_write_ctxt() ==> leave the > mmapped page locked >ocfs2_write_begin_nolock() ==> return -ENOMEM > __ocfs2_page_mkwrite() ==> return VM_FAULT_OMM >__do_page_mkwrite() ==> deadlock here > (https://github.com/torvalds/linux/blob/master/mm/memory.c#L2054) > This is another corner case, right? > > Anyway, I think this patch is good for the -ENOSPC case. And another patch > should be > proposed for -ENOMEM case? Yes, I think we can catch both -ENOSPC and -ENOMEM cases in the failure path by unlocking the mmaped page after ocfs2_free_write_ctx(), right? Eric > > Thanks, > Eric > >> Thanks, >> Eric >> >>> Thanks, >>> Joseph >>> Fix this issue by unlocking the target page after we fail to allocate enough space at the first time. Jan Kara helps me clear out the JBD2 part, and suggest the hint for root cause. Signed-off-by: Eric Ren --- fs/ocfs2/aops.c | 7 +++ 1 file changed, 7 insertions(+) diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c index 98d3654..78d1d67 100644 --- a/fs/ocfs2/aops.c +++ b/fs/ocfs2/aops.c @@ -1860,6 +1860,13 @@ out: */ try_free = 0; +/* + * Unlock mmap_page because the page has been locked when we + * are here. + */ +if (mmap_page) +unlock_page(mmap_page); + ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); if (ret1 == 1) goto try_again; >>> >> > > ___ > Ocfs2-devel mailing list > Ocfs2-devel@oss.oracle.com > https://oss.oracle.com/mailman/listinfo/ocfs2-devel > ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel
Re: [Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Hi Joseph, In ocfs2_write_begin_nolock(), we first grab the pages and then allocate disk space for this write; ocfs2_try_to_free_truncate_log() will be called if ENOSPC is turned; if we're lucky to get enough clusters, which is usually the case, we start over again. But in ocfs2_free_write_ctxt() the target page isn't unlocked, so we will deadlock when trying to grab the target page again. IMO, in ocfs2_grab_pages_for_write, mmap_page is mapping to w_pages and w_target_locked is set to true, and then will be unlocked by ocfs2_unlock_pages in ocfs2_free_write_ctxt. So I'm not getting the case "page isn't unlock". Could you please explain it in more detail? Thanks for review;-) Follow up the calling chain: ocfs2_free_write_ctxt() ->ocfs2_unlock_pages() in ocfs2_unlock_pages (https://github.com/torvalds/linux/blob/master/fs/ocfs2/aops.c#L793), we can see the code just put_page(target_page), but not unlock it. Did this answer your question? Thanks, Eric Yeah, I will think this a bit more like: why not unlock the target_page there? Is there other potential problems if the "ret" is not "-ENOSPC" but other possible error code? Thanks, Eric Thanks, Joseph Fix this issue by unlocking the target page after we fail to allocate enough space at the first time. Jan Kara helps me clear out the JBD2 part, and suggest the hint for root cause. Signed-off-by: Eric Ren --- fs/ocfs2/aops.c | 7 +++ 1 file changed, 7 insertions(+) diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c index 98d3654..78d1d67 100644 --- a/fs/ocfs2/aops.c +++ b/fs/ocfs2/aops.c @@ -1860,6 +1860,13 @@ out: */ try_free = 0; +/* + * Unlock mmap_page because the page has been locked when we + * are here. + */ +if (mmap_page) +unlock_page(mmap_page); + ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); if (ret1 == 1) goto try_again; ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel
Re: [Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Hi, >> IMO, in ocfs2_grab_pages_for_write, mmap_page is mapping to w_pages and >> w_target_locked is set to true, and then will be unlocked by >> ocfs2_unlock_pages in ocfs2_free_write_ctxt. >> So I'm not getting the case "page isn't unlock". Could you please explain >> it in more detail? > Thanks for review;-) Follow up the calling chain: > > ocfs2_free_write_ctxt() > ->ocfs2_unlock_pages() > > in ocfs2_unlock_pages > (https://github.com/torvalds/linux/blob/master/fs/ocfs2/aops.c#L793), we > can see the code just put_page(target_page), but not unlock it. > > Yeah, I will think this a bit more like: > why not unlock the target_page there? Is there other potential problems if > the "ret" is > not "-ENOSPC" but > other possible error code? 1. ocfs2_unlock_pages() will be called in ocfs2_write_end_nolock(), in this case, we definitely want to return a locked mmaped page to VM code (do_page_mkwrite) when VM_FAULT_LOCKED is set. 2. But there's indeed a potential existing deadlock situation: ocfs2_grab_pages_for_write() ==> return -ENOMEM and with the mmaped page locked ocfs2_free_write_ctxt() ==> leave the mmapped page locked ocfs2_write_begin_nolock() ==> return -ENOMEM __ocfs2_page_mkwrite() ==> return VM_FAULT_OMM __do_page_mkwrite() ==> deadlock here (https://github.com/torvalds/linux/blob/master/mm/memory.c#L2054) This is another corner case, right? Anyway, I think this patch is good for the -ENOSPC case. And another patch should be proposed for -ENOMEM case? Thanks, Eric > > Thanks, > Eric > >> >> Thanks, >> Joseph >> >>> Fix this issue by unlocking the target page after we fail to allocate >>> enough space at the first time. >>> >>> Jan Kara helps me clear out the JBD2 part, and suggest the hint for root >>> cause. >>> >>> Signed-off-by: Eric Ren >>> --- >>> fs/ocfs2/aops.c | 7 +++ >>> 1 file changed, 7 insertions(+) >>> >>> diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c >>> index 98d3654..78d1d67 100644 >>> --- a/fs/ocfs2/aops.c >>> +++ b/fs/ocfs2/aops.c >>> @@ -1860,6 +1860,13 @@ out: >>>*/ >>> try_free = 0; >>> +/* >>> + * Unlock mmap_page because the page has been locked when we >>> + * are here. >>> + */ >>> +if (mmap_page) >>> +unlock_page(mmap_page); >>> + >>> ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); >>> if (ret1 == 1) >>> goto try_again; >>> >> >> > > ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel
Re: [Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Hi Joseph, On 09/12/2016 09:37 AM, Joseph Qi wrote: Hi Eric, On 2016/9/10 17:55, Eric Ren wrote: The testcase "mmaptruncate" of ocfs2-test deadlocked occasionally. In this testcase, we create a 2*CLUSTER_SIZE file and mmap() on it; there are 2 process repeatedly performing the following operations respectively: one is doing memset(mmaped_addr + 2*CLUSTER_SIZE - 1, 'a', 1), while the another is playing ftruncate(fd, 2*CLUSTER_SIZE) and then ftruncate(fd, CLUSTER_SIZE) again and again. This is the backtrace when the deadlock happens: [] __wait_on_bit_lock+0x50/0xa0 [] __lock_page+0xb7/0xc0 [] ? autoremove_wake_function+0x40/0x40 [] ocfs2_write_begin_nolock+0x163f/0x1790 [ocfs2] [] ? ocfs2_allocate_extend_trans+0x180/0x180 [ocfs2] [] ocfs2_page_mkwrite+0x1c7/0x2a0 [ocfs2] [] do_page_mkwrite+0x66/0xc0 [] handle_mm_fault+0x685/0x1350 [] ? __fpu__restore_sig+0x70/0x530 [] __do_page_fault+0x1d8/0x4d0 [] trace_do_page_fault+0x37/0xf0 [] do_async_page_fault+0x19/0x70 [] async_page_fault+0x28/0x30 In ocfs2_write_begin_nolock(), we first grab the pages and then allocate disk space for this write; ocfs2_try_to_free_truncate_log() will be called if ENOSPC is turned; if we're lucky to get enough clusters, which is usually the case, we start over again. But in ocfs2_free_write_ctxt() the target page isn't unlocked, so we will deadlock when trying to grab the target page again. IMO, in ocfs2_grab_pages_for_write, mmap_page is mapping to w_pages and w_target_locked is set to true, and then will be unlocked by ocfs2_unlock_pages in ocfs2_free_write_ctxt. So I'm not getting the case "page isn't unlock". Could you please explain it in more detail? Thanks for review;-) Follow up the calling chain: ocfs2_free_write_ctxt() ->ocfs2_unlock_pages() in ocfs2_unlock_pages (https://github.com/torvalds/linux/blob/master/fs/ocfs2/aops.c#L793), we can see the code just put_page(target_page), but not unlock it. Yeah, I will think this a bit more like: why not unlock the target_page there? Is there other potential problems if the "ret" is not "-ENOSPC" but other possible error code? Thanks, Eric Thanks, Joseph Fix this issue by unlocking the target page after we fail to allocate enough space at the first time. Jan Kara helps me clear out the JBD2 part, and suggest the hint for root cause. Signed-off-by: Eric Ren --- fs/ocfs2/aops.c | 7 +++ 1 file changed, 7 insertions(+) diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c index 98d3654..78d1d67 100644 --- a/fs/ocfs2/aops.c +++ b/fs/ocfs2/aops.c @@ -1860,6 +1860,13 @@ out: */ try_free = 0; + /* +* Unlock mmap_page because the page has been locked when we +* are here. +*/ + if (mmap_page) + unlock_page(mmap_page); + ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); if (ret1 == 1) goto try_again; ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel
Re: [Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Reviewed-by: Gang He Thanks Gang >>> > The testcase "mmaptruncate" of ocfs2-test deadlocked occasionally. > > In this testcase, we create a 2*CLUSTER_SIZE file and mmap() on it; > there are 2 process repeatedly performing the following operations > respectively: one is doing memset(mmaped_addr + 2*CLUSTER_SIZE - 1, > 'a', 1), while the another is playing ftruncate(fd, 2*CLUSTER_SIZE) > and then ftruncate(fd, CLUSTER_SIZE) again and again. > > This is the backtrace when the deadlock happens: > [] __wait_on_bit_lock+0x50/0xa0 > [] __lock_page+0xb7/0xc0 > [] ? autoremove_wake_function+0x40/0x40 > [] ocfs2_write_begin_nolock+0x163f/0x1790 [ocfs2] > [] ? ocfs2_allocate_extend_trans+0x180/0x180 [ocfs2] > [] ocfs2_page_mkwrite+0x1c7/0x2a0 [ocfs2] > [] do_page_mkwrite+0x66/0xc0 > [] handle_mm_fault+0x685/0x1350 > [] ? __fpu__restore_sig+0x70/0x530 > [] __do_page_fault+0x1d8/0x4d0 > [] trace_do_page_fault+0x37/0xf0 > [] do_async_page_fault+0x19/0x70 > [] async_page_fault+0x28/0x30 > > In ocfs2_write_begin_nolock(), we first grab the pages and then > allocate disk space for this write; ocfs2_try_to_free_truncate_log() > will be called if ENOSPC is turned; if we're lucky to get enough clusters, > which is usually the case, we start over again. But in > ocfs2_free_write_ctxt() > the target page isn't unlocked, so we will deadlock when trying to grab > the target page again. > > Fix this issue by unlocking the target page after we fail to allocate > enough space at the first time. > > Jan Kara helps me clear out the JBD2 part, and suggest the hint for root > cause. > > Signed-off-by: Eric Ren > --- > fs/ocfs2/aops.c | 7 +++ > 1 file changed, 7 insertions(+) > > diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c > index 98d3654..78d1d67 100644 > --- a/fs/ocfs2/aops.c > +++ b/fs/ocfs2/aops.c > @@ -1860,6 +1860,13 @@ out: >*/ > try_free = 0; > > + /* > + * Unlock mmap_page because the page has been locked when we > + * are here. > + */ > + if (mmap_page) > + unlock_page(mmap_page); > + > ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); > if (ret1 == 1) > goto try_again; > -- > 2.6.6 > > > ___ > Ocfs2-devel mailing list > Ocfs2-devel@oss.oracle.com > https://oss.oracle.com/mailman/listinfo/ocfs2-devel ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel
Re: [Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
Hi Eric, On 2016/9/10 17:55, Eric Ren wrote: > The testcase "mmaptruncate" of ocfs2-test deadlocked occasionally. > > In this testcase, we create a 2*CLUSTER_SIZE file and mmap() on it; > there are 2 process repeatedly performing the following operations > respectively: one is doing memset(mmaped_addr + 2*CLUSTER_SIZE - 1, > 'a', 1), while the another is playing ftruncate(fd, 2*CLUSTER_SIZE) > and then ftruncate(fd, CLUSTER_SIZE) again and again. > > This is the backtrace when the deadlock happens: > [] __wait_on_bit_lock+0x50/0xa0 > [] __lock_page+0xb7/0xc0 > [] ? autoremove_wake_function+0x40/0x40 > [] ocfs2_write_begin_nolock+0x163f/0x1790 [ocfs2] > [] ? ocfs2_allocate_extend_trans+0x180/0x180 [ocfs2] > [] ocfs2_page_mkwrite+0x1c7/0x2a0 [ocfs2] > [] do_page_mkwrite+0x66/0xc0 > [] handle_mm_fault+0x685/0x1350 > [] ? __fpu__restore_sig+0x70/0x530 > [] __do_page_fault+0x1d8/0x4d0 > [] trace_do_page_fault+0x37/0xf0 > [] do_async_page_fault+0x19/0x70 > [] async_page_fault+0x28/0x30 > > In ocfs2_write_begin_nolock(), we first grab the pages and then > allocate disk space for this write; ocfs2_try_to_free_truncate_log() > will be called if ENOSPC is turned; if we're lucky to get enough clusters, > which is usually the case, we start over again. But in ocfs2_free_write_ctxt() > the target page isn't unlocked, so we will deadlock when trying to grab > the target page again. IMO, in ocfs2_grab_pages_for_write, mmap_page is mapping to w_pages and w_target_locked is set to true, and then will be unlocked by ocfs2_unlock_pages in ocfs2_free_write_ctxt. So I'm not getting the case "page isn't unlock". Could you please explain it in more detail? Thanks, Joseph > > Fix this issue by unlocking the target page after we fail to allocate > enough space at the first time. > > Jan Kara helps me clear out the JBD2 part, and suggest the hint for root > cause. > > Signed-off-by: Eric Ren > --- > fs/ocfs2/aops.c | 7 +++ > 1 file changed, 7 insertions(+) > > diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c > index 98d3654..78d1d67 100644 > --- a/fs/ocfs2/aops.c > +++ b/fs/ocfs2/aops.c > @@ -1860,6 +1860,13 @@ out: >*/ > try_free = 0; > > + /* > + * Unlock mmap_page because the page has been locked when we > + * are here. > + */ > + if (mmap_page) > + unlock_page(mmap_page); > + > ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); > if (ret1 == 1) > goto try_again; > ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel
[Ocfs2-devel] [PATCH] ocfs2: fix deadlock on mmapped page in ocfs2_write_begin_nolock()
The testcase "mmaptruncate" of ocfs2-test deadlocked occasionally. In this testcase, we create a 2*CLUSTER_SIZE file and mmap() on it; there are 2 process repeatedly performing the following operations respectively: one is doing memset(mmaped_addr + 2*CLUSTER_SIZE - 1, 'a', 1), while the another is playing ftruncate(fd, 2*CLUSTER_SIZE) and then ftruncate(fd, CLUSTER_SIZE) again and again. This is the backtrace when the deadlock happens: [] __wait_on_bit_lock+0x50/0xa0 [] __lock_page+0xb7/0xc0 [] ? autoremove_wake_function+0x40/0x40 [] ocfs2_write_begin_nolock+0x163f/0x1790 [ocfs2] [] ? ocfs2_allocate_extend_trans+0x180/0x180 [ocfs2] [] ocfs2_page_mkwrite+0x1c7/0x2a0 [ocfs2] [] do_page_mkwrite+0x66/0xc0 [] handle_mm_fault+0x685/0x1350 [] ? __fpu__restore_sig+0x70/0x530 [] __do_page_fault+0x1d8/0x4d0 [] trace_do_page_fault+0x37/0xf0 [] do_async_page_fault+0x19/0x70 [] async_page_fault+0x28/0x30 In ocfs2_write_begin_nolock(), we first grab the pages and then allocate disk space for this write; ocfs2_try_to_free_truncate_log() will be called if ENOSPC is turned; if we're lucky to get enough clusters, which is usually the case, we start over again. But in ocfs2_free_write_ctxt() the target page isn't unlocked, so we will deadlock when trying to grab the target page again. Fix this issue by unlocking the target page after we fail to allocate enough space at the first time. Jan Kara helps me clear out the JBD2 part, and suggest the hint for root cause. Signed-off-by: Eric Ren --- fs/ocfs2/aops.c | 7 +++ 1 file changed, 7 insertions(+) diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c index 98d3654..78d1d67 100644 --- a/fs/ocfs2/aops.c +++ b/fs/ocfs2/aops.c @@ -1860,6 +1860,13 @@ out: */ try_free = 0; + /* +* Unlock mmap_page because the page has been locked when we +* are here. +*/ + if (mmap_page) + unlock_page(mmap_page); + ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need); if (ret1 == 1) goto try_again; -- 2.6.6 ___ Ocfs2-devel mailing list Ocfs2-devel@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-devel