xas_for_each_marked() is using entry == NULL as a termination condition of the iteration. When xas_for_each_marked() is used protected only by RCU, this can however race with xas_store(xas, NULL) in the following way:
TASK1 TASK2 page_cache_delete() find_get_pages_range_tag() xas_for_each_marked() xas_find_marked() off = xas_find_chunk()
xas_store(&xas, NULL) xas_init_marks(&xas); ... rcu_assign_pointer(*slot, NULL); entry = xa_entry(off);
And thus xas_for_each_marked() terminates prematurely possibly leading to missed entries in the iteration (translating to missing writeback of some pages or a similar problem).
Fix the problem by creating a special version of xas_find_marked() - xas_find_valid_marked() - that does not return NULL marked entries and changing xas_next_marked() in the same way.
CC: stable@vger.kernel.org Fixes: ef8e5717db01 "page cache: Convert delete_batch to XArray" Signed-off-by: Jan Kara jack@suse.cz --- include/linux/xarray.h | 64 ++++++++++++++++++++++++++++++++++++-------------- 1 file changed, 47 insertions(+), 17 deletions(-)
diff --git a/include/linux/xarray.h b/include/linux/xarray.h index f73e1775ded0..5370716d7010 100644 --- a/include/linux/xarray.h +++ b/include/linux/xarray.h @@ -1633,33 +1633,63 @@ static inline unsigned int xas_find_chunk(struct xa_state *xas, bool advance, }
/** - * xas_next_marked() - Advance iterator to next marked entry. + * xas_find_valid_marked() - Find the next marked valid entry in the XArray. + * @xas: XArray operation state. + * @max: Highest index to return. + * @mark: Mark number to search for. + * + * This is like xas_find_marked() except that we also skip over all %NULL + * marked entries. + * + * Return: The entry, if found, otherwise %NULL. + */ +static inline void *xas_find_valid_marked(struct xa_state *xas, + unsigned long max, xa_mark_t mark) +{ + void *entry; + + do { + entry = xas_find_marked(xas, max, mark); + } while (unlikely(entry == NULL) && xas_valid(xas)); + + return entry; +} + +/** + * xas_next_valid_marked() - Advance iterator to next valid marked entry. * @xas: XArray operation state. * @max: Highest index to return. * @mark: Mark to search for. * - * xas_next_marked() is an inline function to optimise xarray traversal for - * speed. It is equivalent to calling xas_find_marked(), and will call - * xas_find_marked() for all the hard cases. + * xas_next_valid_marked() is an inline function to optimise xarray traversal + * for speed. It is equivalent to calling xas_find_valid_marked(), and will + * call xas_find_marked() for all the hard cases. The function skips over %NULL + * marked entries. * * Return: The next marked entry after the one currently referred to by @xas. */ -static inline void *xas_next_marked(struct xa_state *xas, unsigned long max, - xa_mark_t mark) +static inline void *xas_next_valid_marked(struct xa_state *xas, + unsigned long max, xa_mark_t mark) { struct xa_node *node = xas->xa_node; unsigned int offset; + void *entry;
if (unlikely(xas_not_node(node) || node->shift)) - return xas_find_marked(xas, max, mark); - offset = xas_find_chunk(xas, true, mark); - xas->xa_offset = offset; - xas->xa_index = (xas->xa_index & ~XA_CHUNK_MASK) + offset; - if (xas->xa_index > max) - return NULL; - if (offset == XA_CHUNK_SIZE) - return xas_find_marked(xas, max, mark); - return xa_entry(xas->xa, node, offset); + return xas_find_valid_marked(xas, max, mark); + + do { + offset = xas_find_chunk(xas, true, mark); + xas->xa_offset = offset; + xas->xa_index = (xas->xa_index & ~XA_CHUNK_MASK) + offset; + if (xas->xa_index > max) + return NULL; + if (offset == XA_CHUNK_SIZE) + return xas_find_valid_marked(xas, max, mark); + entry = xa_entry(xas->xa, node, offset); + } while (unlikely(!entry)); + + return entry; }
/* @@ -1702,8 +1732,8 @@ enum { * xas_pause() first. */ #define xas_for_each_marked(xas, entry, max, mark) \ - for (entry = xas_find_marked(xas, max, mark); entry; \ - entry = xas_next_marked(xas, max, mark)) + for (entry = xas_find_valid_marked(xas, max, mark); entry; \ + entry = xas_next_valid_marked(xas, max, mark))
/** * xas_for_each_conflict() - Iterate over a range of an XArray.
Hi Jan,
I fixed this in a different way ... also, I added a test to the test-suite so this shouldn't regress. Thanks for writing the commit message ;-)
From 7e934cf5ace1dceeb804f7493fa28bb697ed3c52 Mon Sep 17 00:00:00 2001
From: "Matthew Wilcox (Oracle)" willy@infradead.org Date: Thu, 12 Mar 2020 17:29:11 -0400 Subject: [PATCH] xarray: Fix early termination of xas_for_each_marked
xas_for_each_marked() is using entry == NULL as a termination condition of the iteration. When xas_for_each_marked() is used protected only by RCU, this can however race with xas_store(xas, NULL) in the following way:
TASK1 TASK2 page_cache_delete() find_get_pages_range_tag() xas_for_each_marked() xas_find_marked() off = xas_find_chunk()
xas_store(&xas, NULL) xas_init_marks(&xas); ... rcu_assign_pointer(*slot, NULL); entry = xa_entry(off);
And thus xas_for_each_marked() terminates prematurely possibly leading to missed entries in the iteration (translating to missing writeback of some pages or a similar problem).
If we find a NULL entry that has been marked, skip it (unless we're trying to allocate an entry).
Reported-by: Jan Kara jack@suse.cz CC: stable@vger.kernel.org Fixes: ef8e5717db01 ("page cache: Convert delete_batch to XArray") Signed-off-by: Matthew Wilcox (Oracle) willy@infradead.org --- include/linux/xarray.h | 6 +- lib/xarray.c | 2 + tools/testing/radix-tree/Makefile | 4 +- tools/testing/radix-tree/iteration_check_2.c | 87 ++++++++++++++++++++ tools/testing/radix-tree/main.c | 1 + tools/testing/radix-tree/test.h | 1 + 6 files changed, 98 insertions(+), 3 deletions(-) create mode 100644 tools/testing/radix-tree/iteration_check_2.c
diff --git a/include/linux/xarray.h b/include/linux/xarray.h index a491653d8882..14c893433139 100644 --- a/include/linux/xarray.h +++ b/include/linux/xarray.h @@ -1648,6 +1648,7 @@ static inline void *xas_next_marked(struct xa_state *xas, unsigned long max, xa_mark_t mark) { struct xa_node *node = xas->xa_node; + void *entry; unsigned int offset;
if (unlikely(xas_not_node(node) || node->shift)) @@ -1659,7 +1660,10 @@ static inline void *xas_next_marked(struct xa_state *xas, unsigned long max, return NULL; if (offset == XA_CHUNK_SIZE) return xas_find_marked(xas, max, mark); - return xa_entry(xas->xa, node, offset); + entry = xa_entry(xas->xa, node, offset); + if (!entry) + return xas_find_marked(xas, max, mark); + return entry; }
/* diff --git a/lib/xarray.c b/lib/xarray.c index f448bcd263ac..e9e641d3c0c3 100644 --- a/lib/xarray.c +++ b/lib/xarray.c @@ -1208,6 +1208,8 @@ void *xas_find_marked(struct xa_state *xas, unsigned long max, xa_mark_t mark) }
entry = xa_entry(xas->xa, xas->xa_node, xas->xa_offset); + if (!entry && !(xa_track_free(xas->xa) && mark == XA_FREE_MARK)) + continue; if (!xa_is_node(entry)) return entry; xas->xa_node = xa_to_node(entry); diff --git a/tools/testing/radix-tree/Makefile b/tools/testing/radix-tree/Makefile index 397d6b612502..aa6abfe0749c 100644 --- a/tools/testing/radix-tree/Makefile +++ b/tools/testing/radix-tree/Makefile @@ -7,8 +7,8 @@ LDLIBS+= -lpthread -lurcu TARGETS = main idr-test multiorder xarray CORE_OFILES := xarray.o radix-tree.o idr.o linux.o test.o find_bit.o bitmap.o OFILES = main.o $(CORE_OFILES) regression1.o regression2.o regression3.o \ - regression4.o \ - tag_check.o multiorder.o idr-test.o iteration_check.o benchmark.o + regression4.o tag_check.o multiorder.o idr-test.o iteration_check.o \ + iteration_check_2.o benchmark.o
ifndef SHIFT SHIFT=3 diff --git a/tools/testing/radix-tree/iteration_check_2.c b/tools/testing/radix-tree/iteration_check_2.c new file mode 100644 index 000000000000..aac5c50a3674 --- /dev/null +++ b/tools/testing/radix-tree/iteration_check_2.c @@ -0,0 +1,87 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * iteration_check_2.c: Check that deleting a tagged entry doesn't cause + * an RCU walker to finish early. + * Copyright (c) 2020 Oracle + * Author: Matthew Wilcox willy@infradead.org + */ +#include <pthread.h> +#include "test.h" + +static volatile bool test_complete; + +static void *iterator(void *arg) +{ + XA_STATE(xas, arg, 0); + void *entry; + + rcu_register_thread(); + + while (!test_complete) { + xas_set(&xas, 0); + rcu_read_lock(); + xas_for_each_marked(&xas, entry, ULONG_MAX, XA_MARK_0) + ; + rcu_read_unlock(); + assert(xas.xa_index >= 100); + } + + rcu_unregister_thread(); + return NULL; +} + +static void *throbber(void *arg) +{ + struct xarray *xa = arg; + + rcu_register_thread(); + + while (!test_complete) { + int i; + + for (i = 0; i < 100; i++) { + xa_store(xa, i, xa_mk_value(i), GFP_KERNEL); + xa_set_mark(xa, i, XA_MARK_0); + } + for (i = 0; i < 100; i++) + xa_erase(xa, i); + } + + rcu_unregister_thread(); + return NULL; +} + +void iteration_test2(unsigned test_duration) +{ + pthread_t threads[2]; + DEFINE_XARRAY(array); + int i; + + printv(1, "Running iteration test 2 for %d seconds\n", test_duration); + + test_complete = false; + + xa_store(&array, 100, xa_mk_value(100), GFP_KERNEL); + xa_set_mark(&array, 100, XA_MARK_0); + + if (pthread_create(&threads[0], NULL, iterator, &array)) { + perror("create iterator thread"); + exit(1); + } + if (pthread_create(&threads[1], NULL, throbber, &array)) { + perror("create throbber thread"); + exit(1); + } + + sleep(test_duration); + test_complete = true; + + for (i = 0; i < 2; i++) { + if (pthread_join(threads[i], NULL)) { + perror("pthread_join"); + exit(1); + } + } + + xa_destroy(&array); +} diff --git a/tools/testing/radix-tree/main.c b/tools/testing/radix-tree/main.c index 7a22d6e3732e..f2cbc8e5b97c 100644 --- a/tools/testing/radix-tree/main.c +++ b/tools/testing/radix-tree/main.c @@ -311,6 +311,7 @@ int main(int argc, char **argv) regression4_test(); iteration_test(0, 10 + 90 * long_run); iteration_test(7, 10 + 90 * long_run); + iteration_test2(10 + 90 * long_run); single_thread_tests(long_run);
/* Free any remaining preallocated nodes */ diff --git a/tools/testing/radix-tree/test.h b/tools/testing/radix-tree/test.h index 1ee4b2c0ad10..34dab4d18744 100644 --- a/tools/testing/radix-tree/test.h +++ b/tools/testing/radix-tree/test.h @@ -34,6 +34,7 @@ void xarray_tests(void); void tag_check(void); void multiorder_checks(void); void iteration_test(unsigned order, unsigned duration); +void iteration_test2(unsigned duration); void benchmark(void); void idr_checks(void); void ida_tests(void);
On Thu 12-03-20 14:45:48, Matthew Wilcox wrote:
Hi Jan,
I fixed this in a different way ... also, I added a test to the test-suite so this shouldn't regress. Thanks for writing the commit message ;-)
From 7e934cf5ace1dceeb804f7493fa28bb697ed3c52 Mon Sep 17 00:00:00 2001 From: "Matthew Wilcox (Oracle)" willy@infradead.org Date: Thu, 12 Mar 2020 17:29:11 -0400 Subject: [PATCH] xarray: Fix early termination of xas_for_each_marked
xas_for_each_marked() is using entry == NULL as a termination condition of the iteration. When xas_for_each_marked() is used protected only by RCU, this can however race with xas_store(xas, NULL) in the following way:
TASK1 TASK2 page_cache_delete() find_get_pages_range_tag() xas_for_each_marked() xas_find_marked() off = xas_find_chunk()
xas_store(&xas, NULL) xas_init_marks(&xas); ... rcu_assign_pointer(*slot, NULL); entry = xa_entry(off);
And thus xas_for_each_marked() terminates prematurely possibly leading to missed entries in the iteration (translating to missing writeback of some pages or a similar problem).
If we find a NULL entry that has been marked, skip it (unless we're trying to allocate an entry).
Reported-by: Jan Kara jack@suse.cz CC: stable@vger.kernel.org Fixes: ef8e5717db01 ("page cache: Convert delete_batch to XArray") Signed-off-by: Matthew Wilcox (Oracle) willy@infradead.org
The patch looks good to me. Thanks for looking into this. You can add:
Reviewed-by: Jan Kara jack@suse.cz
Honza
include/linux/xarray.h | 6 +- lib/xarray.c | 2 + tools/testing/radix-tree/Makefile | 4 +- tools/testing/radix-tree/iteration_check_2.c | 87 ++++++++++++++++++++ tools/testing/radix-tree/main.c | 1 + tools/testing/radix-tree/test.h | 1 + 6 files changed, 98 insertions(+), 3 deletions(-) create mode 100644 tools/testing/radix-tree/iteration_check_2.c
diff --git a/include/linux/xarray.h b/include/linux/xarray.h index a491653d8882..14c893433139 100644 --- a/include/linux/xarray.h +++ b/include/linux/xarray.h @@ -1648,6 +1648,7 @@ static inline void *xas_next_marked(struct xa_state *xas, unsigned long max, xa_mark_t mark) { struct xa_node *node = xas->xa_node;
- void *entry; unsigned int offset;
if (unlikely(xas_not_node(node) || node->shift)) @@ -1659,7 +1660,10 @@ static inline void *xas_next_marked(struct xa_state *xas, unsigned long max, return NULL; if (offset == XA_CHUNK_SIZE) return xas_find_marked(xas, max, mark);
- return xa_entry(xas->xa, node, offset);
- entry = xa_entry(xas->xa, node, offset);
- if (!entry)
return xas_find_marked(xas, max, mark);- return entry;
} /* diff --git a/lib/xarray.c b/lib/xarray.c index f448bcd263ac..e9e641d3c0c3 100644 --- a/lib/xarray.c +++ b/lib/xarray.c @@ -1208,6 +1208,8 @@ void *xas_find_marked(struct xa_state *xas, unsigned long max, xa_mark_t mark) } entry = xa_entry(xas->xa, xas->xa_node, xas->xa_offset);
if (!entry && !(xa_track_free(xas->xa) && mark == XA_FREE_MARK))continue;diff --git a/tools/testing/radix-tree/Makefile b/tools/testing/radix-tree/Makefile index 397d6b612502..aa6abfe0749c 100644 --- a/tools/testing/radix-tree/Makefile +++ b/tools/testing/radix-tree/Makefile @@ -7,8 +7,8 @@ LDLIBS+= -lpthread -lurcu TARGETS = main idr-test multiorder xarray CORE_OFILES := xarray.o radix-tree.o idr.o linux.o test.o find_bit.o bitmap.o OFILES = main.o $(CORE_OFILES) regression1.o regression2.o regression3.o \
regression4.o \tag_check.o multiorder.o idr-test.o iteration_check.o benchmark.o
regression4.o tag_check.o multiorder.o idr-test.o iteration_check.o \iteration_check_2.o benchmark.oifndef SHIFT SHIFT=3 diff --git a/tools/testing/radix-tree/iteration_check_2.c b/tools/testing/radix-tree/iteration_check_2.c new file mode 100644 index 000000000000..aac5c50a3674 --- /dev/null +++ b/tools/testing/radix-tree/iteration_check_2.c @@ -0,0 +1,87 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/*
- iteration_check_2.c: Check that deleting a tagged entry doesn't cause
- an RCU walker to finish early.
- Copyright (c) 2020 Oracle
- Author: Matthew Wilcox willy@infradead.org
- */
+#include <pthread.h> +#include "test.h"
+static volatile bool test_complete;
+static void *iterator(void *arg) +{
- XA_STATE(xas, arg, 0);
- void *entry;
- rcu_register_thread();
- while (!test_complete) {
xas_set(&xas, 0);rcu_read_lock();xas_for_each_marked(&xas, entry, ULONG_MAX, XA_MARK_0);rcu_read_unlock();assert(xas.xa_index >= 100);- }
- rcu_unregister_thread();
- return NULL;
+}
+static void *throbber(void *arg) +{
- struct xarray *xa = arg;
- rcu_register_thread();
- while (!test_complete) {
int i;for (i = 0; i < 100; i++) {xa_store(xa, i, xa_mk_value(i), GFP_KERNEL);xa_set_mark(xa, i, XA_MARK_0);}for (i = 0; i < 100; i++)xa_erase(xa, i);- }
- rcu_unregister_thread();
- return NULL;
+}
+void iteration_test2(unsigned test_duration) +{
- pthread_t threads[2];
- DEFINE_XARRAY(array);
- int i;
- printv(1, "Running iteration test 2 for %d seconds\n", test_duration);
- test_complete = false;
- xa_store(&array, 100, xa_mk_value(100), GFP_KERNEL);
- xa_set_mark(&array, 100, XA_MARK_0);
- if (pthread_create(&threads[0], NULL, iterator, &array)) {
perror("create iterator thread");exit(1);- }
- if (pthread_create(&threads[1], NULL, throbber, &array)) {
perror("create throbber thread");exit(1);- }
- sleep(test_duration);
- test_complete = true;
- for (i = 0; i < 2; i++) {
if (pthread_join(threads[i], NULL)) {perror("pthread_join");exit(1);}- }
- xa_destroy(&array);
+} diff --git a/tools/testing/radix-tree/main.c b/tools/testing/radix-tree/main.c index 7a22d6e3732e..f2cbc8e5b97c 100644 --- a/tools/testing/radix-tree/main.c +++ b/tools/testing/radix-tree/main.c @@ -311,6 +311,7 @@ int main(int argc, char **argv) regression4_test(); iteration_test(0, 10 + 90 * long_run); iteration_test(7, 10 + 90 * long_run);
- iteration_test2(10 + 90 * long_run); single_thread_tests(long_run);
/* Free any remaining preallocated nodes */ diff --git a/tools/testing/radix-tree/test.h b/tools/testing/radix-tree/test.h index 1ee4b2c0ad10..34dab4d18744 100644 --- a/tools/testing/radix-tree/test.h +++ b/tools/testing/radix-tree/test.h @@ -34,6 +34,7 @@ void xarray_tests(void); void tag_check(void); void multiorder_checks(void); void iteration_test(unsigned order, unsigned duration); +void iteration_test2(unsigned duration); void benchmark(void); void idr_checks(void); void ida_tests(void); -- 2.25.1
linux-stable-mirror@lists.linaro.org
-
Jan Kara -
Matthew Wilcox