mm: allow !GFP_KERNEL allocations for kvmalloc

Support for GFP_NO{FS,IO} and __GFP_NOFAIL has been implemented by previous patches so we can allow the support for kvmalloc. This will allow some external users to simplify or completely remove their helpers. GFP_NOWAIT semantic hasn't been supported so far but it hasn't been explicitly documented so let's add a note about that. ceph_kvmalloc is the first helper to be dropped and changed to kvmalloc.

Michal Hocko committed 4 years ago

Commit a421ef31 parent 30d3f018 files · +15 −50Browse files

Unified · Split

8 files

include/linux/ceph/libceph.h +0 −1
mm/util.c +4 −11
net/ceph/buffer.c +2 −2
net/ceph/ceph_common.c +0 −27
net/ceph/crypto.c +1 −1
net/ceph/messenger.c +1 −1
net/ceph/messenger_v2.c +1 −1
net/ceph/osdmap.c +6 −6

include/linux/ceph/libceph.h+0−1

@@ -295,7 +295,6 @@ extern bool libceph_compatible(void *data); · 294 unmodified lines
295	295
296	296	extern const char *ceph_msg_type_name(int type);
297	297	extern int ceph_check_fsid(struct ceph_client client, struct ceph_fsid fsid);
298		−extern void *ceph_kvmalloc(size_t size, gfp_t flags);
299	298
300	299	struct fs_parameter;
301	300	struct fc_log;

mm/util.c+4−11

@@ -549,13 +549,10 @@ EXPORT_SYMBOL(vm_mmap); · 548 unmodified lines
549	549	* Uses kmalloc to get the memory but if the allocation fails then falls back
550	550	* to the vmalloc allocator. Use kvfree for freeing the memory.
551	551	*
552		− * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
	552	+ * GFP_NOWAIT and GFP_ATOMIC are not supported, neither is the __GFP_NORETRY modifier.
553	553	* __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
554	554	* preferable to the vmalloc fallback, due to visible performance drawbacks.
555	555	*
556		− * Please note that any use of gfp flags outside of GFP_KERNEL is careful to not
557		− * fall back to vmalloc.
558		− *
559	556	* Return: pointer to the allocated memory of %NULL in case of failure
560	557	*/
561	558	void *kvmalloc_node(size_t size, gfp_t flags, int node)
@@ -563,13 +560,6 @@ void *kvmalloc_node(size_t size, gfp_t flags, int node) · 1 unmodified line
563	560	gfp_t kmalloc_flags = flags;
564	561	void *ret;
565	562
566		− /*
567		− * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
568		− * so the given set of flags has to be compatible.
569		− */
570		− if ((flags & GFP_KERNEL) != GFP_KERNEL)
571		− return kmalloc_node(size, flags, node);
572		−
573	563	/*
574	564	* We want to attempt a large physically contiguous block first because
575	565	* it is less likely to fragment multiple larger blocks and therefore
@@ -582,6 +572,9 @@ void *kvmalloc_node(size_t size, gfp_t flags, int node) · 6 unmodified lines
582	572
583	573	if (!(kmalloc_flags & __GFP_RETRY_MAYFAIL))
584	574	kmalloc_flags \|= __GFP_NORETRY;
	575	+
	576	+ /* nofail semantic is implemented by the vmalloc fallback */
	577	+ kmalloc_flags &= ~__GFP_NOFAIL;
585	578	}
586	579
587	580	ret = kmalloc_node(size, kmalloc_flags, node);

net/ceph/buffer.c+2−2

@@ -7,7 +7,7 @@ · 6 unmodified lines
7	7
8	8	#include <linux/ceph/buffer.h>
9	9	#include <linux/ceph/decode.h>
10		−#include <linux/ceph/libceph.h> /* for ceph_kvmalloc */
	10	+#include <linux/ceph/libceph.h> /* for kvmalloc */
11	11
12	12	struct ceph_buffer *ceph_buffer_new(size_t len, gfp_t gfp)
13	13	{
@@ -17,7 +17,7 @@ struct ceph_buffer *ceph_buffer_new(size_t len, gfp_t gfp) · 3 unmodified lines
17	17	if (!b)
18	18	return NULL;
19	19
20		− b->vec.iov_base = ceph_kvmalloc(len, gfp);
	20	+ b->vec.iov_base = kvmalloc(len, gfp);
21	21	if (!b->vec.iov_base) {
22	22	kfree(b);
23	23	return NULL;

net/ceph/ceph_common.c+0−27

@@ -190,33 +190,6 @@ int ceph_compare_options(struct ceph_options *new_opt, · 189 unmodified lines
190	190	}
191	191	EXPORT_SYMBOL(ceph_compare_options);
192	192
193		−/*
194		− * kvmalloc() doesn't fall back to the vmalloc allocator unless flags are
195		− * compatible with (a superset of) GFP_KERNEL. This is because while the
196		− * actual pages are allocated with the specified flags, the page table pages
197		− * are always allocated with GFP_KERNEL.
198		− *
199		− * ceph_kvmalloc() may be called with GFP_KERNEL, GFP_NOFS or GFP_NOIO.
200		− */
201		−void *ceph_kvmalloc(size_t size, gfp_t flags)
202		−{
203		− void *p;
204		−
205		− if ((flags & (__GFP_IO \| __GFP_FS)) == (__GFP_IO \| __GFP_FS)) {
206		− p = kvmalloc(size, flags);
207		− } else if ((flags & (__GFP_IO \| __GFP_FS)) == __GFP_IO) {
208		− unsigned int nofs_flag = memalloc_nofs_save();
209		− p = kvmalloc(size, GFP_KERNEL);
210		− memalloc_nofs_restore(nofs_flag);
211		− } else {
212		− unsigned int noio_flag = memalloc_noio_save();
213		− p = kvmalloc(size, GFP_KERNEL);
214		− memalloc_noio_restore(noio_flag);
215		− }
216		−
217		− return p;
218		−}
219		−
220	193	static int parse_fsid(const char str, struct ceph_fsid fsid)
221	194	{
222	195	int i = 0;

net/ceph/crypto.c+1−1

@@ -147,7 +147,7 @@ void ceph_crypto_key_destroy(struct ceph_crypto_key *key) · 146 unmodified lines
147	147	static const u8 aes_iv = (u8 )CEPH_AES_IV;
148	148
149	149	/*
150		− * Should be used for buffers allocated with ceph_kvmalloc().
	150	+ * Should be used for buffers allocated with kvmalloc().
151	151	* Currently these are encrypt out-buffer (ceph_buffer) and decrypt
152	152	* in-buffer (msg front).
153	153	*

net/ceph/messenger.c+1−1

@@ -1920,7 +1920,7 @@ struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items, · 1919 unmodified lines
1920	1920
1921	1921	/* front */
1922	1922	if (front_len) {
1923		− m->front.iov_base = ceph_kvmalloc(front_len, flags);
	1923	+ m->front.iov_base = kvmalloc(front_len, flags);
1924	1924	if (m->front.iov_base == NULL) {
1925	1925	dout("ceph_msg_new can't allocate %d bytes\n",
1926	1926	front_len);

net/ceph/messenger_v2.c+1−1

@@ -308,7 +308,7 @@ static void alloc_conn_buf(struct ceph_connection con, int len) · 307 unmodified lines
308	308	if (WARN_ON(con->v2.conn_buf_cnt >= ARRAY_SIZE(con->v2.conn_bufs)))
309	309	return NULL;
310	310
311		− buf = ceph_kvmalloc(len, GFP_NOIO);
	311	+ buf = kvmalloc(len, GFP_NOIO);
312	312	if (!buf)
313	313	return NULL;
314	314

net/ceph/osdmap.c+6−6

@@ -980,7 +980,7 @@ static struct crush_work alloc_workspace(const struct crush_map c) · 979 unmodified lines
980	980	work_size = crush_work_size(c, CEPH_PG_MAX_SIZE);
981	981	dout("%s work_size %zu bytes\n", __func__, work_size);
982	982
983		− work = ceph_kvmalloc(work_size, GFP_NOIO);
	983	+ work = kvmalloc(work_size, GFP_NOIO);
984	984	if (!work)
985	985	return NULL;
986	986
@@ -1190,9 +1190,9 @@ static int osdmap_set_max_osd(struct ceph_osdmap *map, u32 max) · 203 unmodified lines
1190	1190	if (max == map->max_osd)
1191	1191	return 0;
1192	1192
1193		− state = ceph_kvmalloc(array_size(max, sizeof(*state)), GFP_NOFS);
1194		− weight = ceph_kvmalloc(array_size(max, sizeof(*weight)), GFP_NOFS);
1195		− addr = ceph_kvmalloc(array_size(max, sizeof(*addr)), GFP_NOFS);
	1193	+ state = kvmalloc(array_size(max, sizeof(*state)), GFP_NOFS);
	1194	+ weight = kvmalloc(array_size(max, sizeof(*weight)), GFP_NOFS);
	1195	+ addr = kvmalloc(array_size(max, sizeof(*addr)), GFP_NOFS);
1196	1196	if (!state \|\| !weight \|\| !addr) {
1197	1197	kvfree(state);
1198	1198	kvfree(weight);
@@ -1222,7 +1222,7 @@ static int osdmap_set_max_osd(struct ceph_osdmap *map, u32 max) · 23 unmodified lines
1222	1222	if (map->osd_primary_affinity) {
1223	1223	u32 *affinity;
1224	1224
1225		− affinity = ceph_kvmalloc(array_size(max, sizeof(*affinity)),
	1225	+ affinity = kvmalloc(array_size(max, sizeof(*affinity)),
1226	1226	GFP_NOFS);
1227	1227	if (!affinity)
1228	1228	return -ENOMEM;
@@ -1503,7 +1503,7 @@ static int set_primary_affinity(struct ceph_osdmap *map, int osd, u32 aff) · 274 unmodified lines
1503	1503	if (!map->osd_primary_affinity) {
1504	1504	int i;
1505	1505
1506		− map->osd_primary_affinity = ceph_kvmalloc(
	1506	+ map->osd_primary_affinity = kvmalloc(
1507	1507	array_size(map->max_osd, sizeof(*map->osd_primary_affinity)),
1508	1508	GFP_NOFS);
1509	1509	if (!map->osd_primary_affinity)

mm: allow !GFP_KERNEL allocations for kvmalloc

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code