From: Nick Piggin <nickpiggin@yahoo.com.au>
To: Hugh Dickins <hugh@veritas.com>
Cc: Robin Holt <holt@sgi.com>, Roland McGrath <roland@redhat.com>,
linux-mm@kvack.org
Subject: Re: get_user_pages() with write=1 and force=1 gets read-only pages.
Date: Sun, 31 Jul 2005 11:52:22 +1000 [thread overview]
Message-ID: <42EC2ED6.2070700@yahoo.com.au> (raw)
In-Reply-To: <Pine.LNX.4.61.0507302255390.5143@goblin.wat.veritas.com>
[-- Attachment #1: Type: text/plain, Size: 261 bytes --]
Hugh Dickins wrote:
> get_user_pages is hard! I don't know the right answer offhand,
> but thank you for posing a good question.
>
Detect the racing fault perhaps, and retry until we're sure
that a write fault has gone through?
--
SUSE Labs, Novell Inc.
[-- Attachment #2: mm-gup-fix.patch --]
[-- Type: text/plain, Size: 3774 bytes --]
Index: linux-2.6/include/linux/mm.h
===================================================================
--- linux-2.6.orig/include/linux/mm.h 2005-07-28 19:04:34.000000000 +1000
+++ linux-2.6/include/linux/mm.h 2005-07-31 11:40:24.000000000 +1000
@@ -625,6 +625,7 @@ static inline int page_mapped(struct pag
* Used to decide whether a process gets delivered SIGBUS or
* just gets major/minor fault counters bumped up.
*/
+#define VM_FAULT_RACE (-2)
#define VM_FAULT_OOM (-1)
#define VM_FAULT_SIGBUS 0
#define VM_FAULT_MINOR 1
Index: linux-2.6/mm/memory.c
===================================================================
--- linux-2.6.orig/mm/memory.c 2005-07-28 19:04:37.000000000 +1000
+++ linux-2.6/mm/memory.c 2005-07-31 11:49:35.000000000 +1000
@@ -964,6 +964,14 @@ int get_user_pages(struct task_struct *t
return i ? i : -EFAULT;
case VM_FAULT_OOM:
return i ? i : -ENOMEM;
+ case VM_FAULT_RACE:
+ /*
+ * Someone else got there first.
+ * Must retry before we can assume
+ * that we have actually performed
+ * the write fault (below).
+ */
+ continue;
default:
BUG();
}
@@ -1224,6 +1232,7 @@ static int do_wp_page(struct mm_struct *
struct page *old_page, *new_page;
unsigned long pfn = pte_pfn(pte);
pte_t entry;
+ int ret;
if (unlikely(!pfn_valid(pfn))) {
/*
@@ -1280,7 +1289,9 @@ static int do_wp_page(struct mm_struct *
*/
spin_lock(&mm->page_table_lock);
page_table = pte_offset_map(pmd, address);
+ ret = VM_FAULT_RACE;
if (likely(pte_same(*page_table, pte))) {
+ ret = VM_FAULT_MINOR;
if (PageAnon(old_page))
dec_mm_counter(mm, anon_rss);
if (PageReserved(old_page))
@@ -1299,7 +1310,7 @@ static int do_wp_page(struct mm_struct *
page_cache_release(new_page);
page_cache_release(old_page);
spin_unlock(&mm->page_table_lock);
- return VM_FAULT_MINOR;
+ return ret;
no_new_page:
page_cache_release(old_page);
@@ -1654,7 +1665,7 @@ static int do_swap_page(struct mm_struct
if (likely(pte_same(*page_table, orig_pte)))
ret = VM_FAULT_OOM;
else
- ret = VM_FAULT_MINOR;
+ ret = VM_FAULT_RACE;
pte_unmap(page_table);
spin_unlock(&mm->page_table_lock);
goto out;
@@ -1676,7 +1687,7 @@ static int do_swap_page(struct mm_struct
spin_lock(&mm->page_table_lock);
page_table = pte_offset_map(pmd, address);
if (unlikely(!pte_same(*page_table, orig_pte))) {
- ret = VM_FAULT_MINOR;
+ ret = VM_FAULT_RACE;
goto out_nomap;
}
@@ -1737,6 +1748,7 @@ do_anonymous_page(struct mm_struct *mm,
{
pte_t entry;
struct page * page = ZERO_PAGE(addr);
+ int ret = VM_FAULT_MINOR;
/* Read-only mapping of ZERO_PAGE. */
entry = pte_wrprotect(mk_pte(ZERO_PAGE(addr), vma->vm_page_prot));
@@ -1760,6 +1772,7 @@ do_anonymous_page(struct mm_struct *mm,
pte_unmap(page_table);
page_cache_release(page);
spin_unlock(&mm->page_table_lock);
+ ret = VM_FAULT_RACE;
goto out;
}
inc_mm_counter(mm, rss);
@@ -1779,7 +1792,7 @@ do_anonymous_page(struct mm_struct *mm,
lazy_mmu_prot_update(entry);
spin_unlock(&mm->page_table_lock);
out:
- return VM_FAULT_MINOR;
+ return ret;
no_mem:
return VM_FAULT_OOM;
}
@@ -1897,6 +1910,7 @@ retry:
pte_unmap(page_table);
page_cache_release(new_page);
spin_unlock(&mm->page_table_lock);
+ ret = VM_FAULT_RACE;
goto out;
}
Index: linux-2.6/arch/i386/mm/fault.c
===================================================================
--- linux-2.6.orig/arch/i386/mm/fault.c 2005-07-28 19:03:48.000000000 +1000
+++ linux-2.6/arch/i386/mm/fault.c 2005-07-31 11:47:48.000000000 +1000
@@ -351,6 +351,8 @@ good_area:
goto do_sigbus;
case VM_FAULT_OOM:
goto out_of_memory;
+ case VM_FAULT_RACE:
+ break;
default:
BUG();
}
next prev parent reply other threads:[~2005-07-31 1:52 UTC|newest]
Thread overview: 72+ messages / expand[flat|nested] mbox.gz Atom feed top
2005-07-30 20:53 Robin Holt
2005-07-30 22:13 ` Hugh Dickins
2005-07-31 1:52 ` Nick Piggin [this message]
2005-07-31 10:52 ` Robin Holt
2005-07-31 11:07 ` Nick Piggin
2005-07-31 11:30 ` Robin Holt
2005-07-31 11:39 ` Robin Holt
2005-07-31 12:09 ` Robin Holt
2005-07-31 22:27 ` Nick Piggin
2005-08-01 3:22 ` Roland McGrath
2005-08-01 8:21 ` [patch 2.6.13-rc4] fix get_user_pages bug Nick Piggin
2005-08-01 9:19 ` Ingo Molnar
2005-08-01 9:27 ` Nick Piggin
2005-08-01 10:15 ` Ingo Molnar
2005-08-01 10:57 ` Nick Piggin
2005-08-01 19:43 ` Hugh Dickins
2005-08-01 20:08 ` Linus Torvalds
2005-08-01 21:06 ` Hugh Dickins
2005-08-01 21:51 ` Linus Torvalds
2005-08-01 22:01 ` Linus Torvalds
2005-08-02 12:01 ` Martin Schwidefsky
2005-08-02 12:26 ` Hugh Dickins
2005-08-02 12:28 ` Nick Piggin
2005-08-02 15:19 ` Martin Schwidefsky
2005-08-02 15:30 ` Linus Torvalds
2005-08-02 16:03 ` Hugh Dickins
2005-08-02 16:25 ` Linus Torvalds
2005-08-02 17:02 ` Linus Torvalds
2005-08-02 17:27 ` Hugh Dickins
2005-08-02 17:21 ` Hugh Dickins
2005-08-02 18:47 ` Linus Torvalds
2005-08-02 19:20 ` Hugh Dickins
2005-08-02 19:54 ` Linus Torvalds
2005-08-02 20:55 ` Hugh Dickins
2005-08-03 10:24 ` Nick Piggin
2005-08-03 11:47 ` Hugh Dickins
2005-08-03 12:13 ` Nick Piggin
2005-08-03 16:12 ` Linus Torvalds
2005-08-03 16:39 ` Linus Torvalds
2005-08-03 16:42 ` Linus Torvalds
2005-08-03 17:12 ` Hugh Dickins
2005-08-03 23:03 ` Nick Piggin
2005-08-04 14:14 ` Alexander Nyberg
2005-08-04 14:30 ` Nick Piggin
2005-08-04 15:00 ` Alexander Nyberg
2005-08-04 15:35 ` Hugh Dickins
2005-08-04 16:32 ` Russell King
2005-08-04 15:36 ` Linus Torvalds
2005-08-04 16:29 ` Russell King
2005-08-03 10:24 ` Martin Schwidefsky
2005-08-03 11:57 ` Hugh Dickins
2005-08-02 16:44 ` Martin Schwidefsky
2005-08-01 15:42 ` Linus Torvalds
2005-08-01 18:18 ` Linus Torvalds
2005-08-03 8:24 ` Robin Holt
2005-08-03 11:31 ` Hugh Dickins
2005-08-04 11:48 ` Robin Holt
2005-08-04 13:04 ` Hugh Dickins
2005-08-01 19:29 ` Hugh Dickins
2005-08-01 19:48 ` Linus Torvalds
2005-08-02 8:07 ` Martin Schwidefsky
2005-08-01 19:57 ` Andrew Morton
2005-08-01 20:16 ` Linus Torvalds
2005-08-02 0:14 ` Nick Piggin
2005-08-02 1:27 ` Nick Piggin
2005-08-02 3:45 ` Linus Torvalds
2005-08-02 4:25 ` Nick Piggin
2005-08-02 4:35 ` Linus Torvalds
2005-08-01 20:03 ` Hugh Dickins
2005-08-01 20:12 ` Andrew Morton
2005-08-01 20:26 ` Linus Torvalds
2005-08-01 20:51 ` Hugh Dickins
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=42EC2ED6.2070700@yahoo.com.au \
--to=nickpiggin@yahoo.com.au \
--cc=holt@sgi.com \
--cc=hugh@veritas.com \
--cc=linux-mm@kvack.org \
--cc=roland@redhat.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