Add linux kernel memory support

Required kmem/vmem changes

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>

lib/libzpool/include/sys/zfs_context.h

@@ -349,10 +349,14 @@ extern void kstat_delete(kstat_t *);
 #define	KM_SLEEP		UMEM_NOFAIL
 #define	KM_PUSHPAGE		KM_SLEEP
 #define	KM_NOSLEEP		UMEM_DEFAULT
+#define	KM_NODEBUG		0x0
 #define	KMC_NODEBUG		UMC_NODEBUG
 #define	kmem_alloc(_s, _f)	umem_alloc(_s, _f)
 #define	kmem_zalloc(_s, _f)	umem_zalloc(_s, _f)
 #define	kmem_free(_b, _s)	umem_free(_b, _s)
+#define	vmem_alloc(_s, _f)	kmem_alloc(_s, _f)
+#define	vmem_zalloc(_s, _f)	kmem_zalloc(_s, _f)
+#define	vmem_free(_b, _s)	kmem_free(_b, _s)
 #define	kmem_cache_create(_a, _b, _c, _d, _e, _f, _g, _h, _i) \
 	umem_cache_create(_a, _b, _c, _d, _e, _f, _g, _h, _i)
 #define	kmem_cache_destroy(_c)	umem_cache_destroy(_c)
@@ -575,6 +579,7 @@ typedef struct callb_cpr {
 #define	zone_dataset_visible(x, y)	(1)
 #define	INGLOBALZONE(z)			(1)
 
+extern char *kmem_vasprintf(const char *fmt, va_list adx);
 extern char *kmem_asprintf(const char *fmt, ...);
 #define	strfree(str) kmem_free((str), strlen(str)+1)
 

lib/libzpool/kernel.c

@@ -1107,25 +1107,27 @@ ksiddomain_rele(ksiddomain_t *ksid)
 	umem_free(ksid, sizeof (ksiddomain_t));
 }
 
-/*
- * Do not change the length of the returned string; it must be freed
- * with strfree().
- */
+char *
+kmem_vasprintf(const char *fmt, va_list adx)
+{
+	char *buf = NULL;
+	va_list adx_copy;
+
+	va_copy(adx_copy, adx);
+	VERIFY(vasprintf(&buf, fmt, adx_copy) != -1);
+	va_end(adx_copy);
+
+	return (buf);
+}
+
 char *
 kmem_asprintf(const char *fmt, ...)
 {
-	int size;
+	char *buf = NULL;
 	va_list adx;
-	char *buf;
 
 	va_start(adx, fmt);
-	size = vsnprintf(NULL, 0, fmt, adx) + 1;
-	va_end(adx);
-
-	buf = kmem_alloc(size, KM_SLEEP);
-
-	va_start(adx, fmt);
-	size = vsnprintf(buf, size, fmt, adx);
+	VERIFY(vasprintf(&buf, fmt, adx) != -1);
 	va_end(adx);
 
 	return (buf);

module/zfs/arc.c

@@ -523,12 +523,13 @@ static boolean_t l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *ab);
  * Hash table routines
  */
 
-#define	HT_LOCK_PAD	64
+#define	HT_LOCK_ALIGN	64
+#define	HT_LOCK_PAD	(P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN)))
 
 struct ht_lock {
 	kmutex_t	ht_lock;
 #ifdef _KERNEL
-	unsigned char	pad[(HT_LOCK_PAD - sizeof (kmutex_t))];
+	unsigned char	pad[HT_LOCK_PAD];
 #endif
 };
 
@@ -772,8 +773,15 @@ buf_fini(void)
 {
 	int i;
 
+#if defined(_KERNEL) && defined(HAVE_SPL)
+	/* Large allocations which do not require contiguous pages
+	 * should be using vmem_free() in the linux kernel */
+	vmem_free(buf_hash_table.ht_table,
+	    (buf_hash_table.ht_mask + 1) * sizeof (void *));
+#else
 	kmem_free(buf_hash_table.ht_table,
 	    (buf_hash_table.ht_mask + 1) * sizeof (void *));
+#endif
 	for (i = 0; i < BUF_LOCKS; i++)
 		mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock);
 	kmem_cache_destroy(hdr_cache);
@@ -875,8 +883,15 @@ buf_init(void)
 		hsize <<= 1;
 retry:
 	buf_hash_table.ht_mask = hsize - 1;
+#if defined(_KERNEL) && defined(HAVE_SPL)
+	/* Large allocations which do not require contiguous pages
+	 * should be using vmem_alloc() in the linux kernel */
+	buf_hash_table.ht_table =
+	    vmem_zalloc(hsize * sizeof (void*), KM_SLEEP);
+#else
 	buf_hash_table.ht_table =
 	    kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP);
+#endif
 	if (buf_hash_table.ht_table == NULL) {
 		ASSERT(hsize > (1ULL << 8));
 		hsize >>= 1;

module/zfs/dbuf.c

@@ -293,7 +293,13 @@ dbuf_init(void)
 
 retry:
 	h->hash_table_mask = hsize - 1;
+#if defined(_KERNEL) && defined(HAVE_SPL)
+	/* Large allocations which do not require contiguous pages
+	 * should be using vmem_alloc() in the linux kernel */
+	h->hash_table = vmem_zalloc(hsize * sizeof (void *), KM_SLEEP);
+#else
 	h->hash_table = kmem_zalloc(hsize * sizeof (void *), KM_NOSLEEP);
+#endif
 	if (h->hash_table == NULL) {
 		/* XXX - we should really return an error instead of assert */
 		ASSERT(hsize > (1ULL << 10));
@@ -317,7 +323,13 @@ dbuf_fini(void)
 
 	for (i = 0; i < DBUF_MUTEXES; i++)
 		mutex_destroy(&h->hash_mutexes[i]);
+#if defined(_KERNEL) && defined(HAVE_SPL)
+	/* Large allocations which do not require contiguous pages
+	 * should be using vmem_free() in the linux kernel */
+	vmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *));
+#else
 	kmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *));
+#endif
 	kmem_cache_destroy(dbuf_cache);
 }
 

module/zfs/ddt.c

@@ -501,6 +501,7 @@ ddt_get_dedup_stats(spa_t *spa, ddt_stat_t *dds_total)
 {
 	ddt_histogram_t *ddh_total;
 
+	/* XXX: Move to a slab */
 	ddh_total = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
 	ddt_get_dedup_histogram(spa, ddh_total);
 	ddt_histogram_stat(dds_total, ddh_total);
@@ -649,6 +650,7 @@ ddt_alloc(const ddt_key_t *ddk)
 {
 	ddt_entry_t *dde;
 
+	/* XXX: Move to a slab */
 	dde = kmem_zalloc(sizeof (ddt_entry_t), KM_SLEEP);
 	cv_init(&dde->dde_cv, NULL, CV_DEFAULT, NULL);
 
@@ -797,7 +799,8 @@ ddt_table_alloc(spa_t *spa, enum zio_checksum c)
 {
 	ddt_t *ddt;
 
-	ddt = kmem_zalloc(sizeof (*ddt), KM_SLEEP);
+	/* XXX: Move to a slab */
+	ddt = kmem_zalloc(sizeof (*ddt), KM_SLEEP | KM_NODEBUG);
 
 	mutex_init(&ddt->ddt_lock, NULL, MUTEX_DEFAULT, NULL);
 	avl_create(&ddt->ddt_tree, ddt_entry_compare,

module/zfs/dmu_send.c

@@ -1337,7 +1337,7 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
 	ra.vp = vp;
 	ra.voff = *voffp;
 	ra.bufsize = 1<<20;
-	ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
+	ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
 
 	/* these were verified in dmu_recv_begin */
 	ASSERT(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo) ==
@@ -1486,7 +1486,7 @@ out:
 		}
 	}
 
-	kmem_free(ra.buf, ra.bufsize);
+	vmem_free(ra.buf, ra.bufsize);
 	*voffp = ra.voff;
 	return (ra.err);
 }

module/zfs/spa.c

@@ -1270,7 +1270,7 @@ load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
 	nvsize = *(uint64_t *)db->db_data;
 	dmu_buf_rele(db, FTAG);
 
-	packed = kmem_alloc(nvsize, KM_SLEEP);
+	packed = kmem_alloc(nvsize, KM_SLEEP | KM_NODEBUG);
 	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
 	    DMU_READ_PREFETCH);
 	if (error == 0)
@@ -5217,7 +5217,7 @@ spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
 	 * saves us a pre-read to get data we don't actually care about.
 	 */
 	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
-	packed = kmem_alloc(bufsize, KM_SLEEP);
+	packed = vmem_alloc(bufsize, KM_SLEEP);
 
 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
 	    KM_SLEEP) == 0);
@@ -5225,7 +5225,7 @@ spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
 
 	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
 
-	kmem_free(packed, bufsize);
+	vmem_free(packed, bufsize);
 
 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
 	dmu_buf_will_dirty(db, tx);

module/zfs/spa_config.c

@@ -96,7 +96,7 @@ spa_config_load(void)
 	if (kobj_get_filesize(file, &fsize) != 0)
 		goto out;
 
-	buf = kmem_alloc(fsize, KM_SLEEP);
+	buf = kmem_alloc(fsize, KM_SLEEP | KM_NODEBUG);
 
 	/*
 	 * Read the nvlist from the file.
@@ -159,7 +159,7 @@ spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 	 */
 	VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
 
-	buf = kmem_alloc(buflen, KM_SLEEP);
+	buf = kmem_alloc(buflen, KM_SLEEP | KM_NODEBUG);
 	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
 
 	VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,

module/zfs/spa_history.c

@@ -428,6 +428,7 @@ log_internal(history_internal_events_t event, spa_t *spa,
     dmu_tx_t *tx, const char *fmt, va_list adx)
 {
 	history_arg_t *ha;
+	va_list adx_copy;
 
 	/*
 	 * If this is part of creating a pool, not everything is
@@ -437,11 +438,9 @@ log_internal(history_internal_events_t event, spa_t *spa,
 		return;
 
 	ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
-	ha->ha_history_str = kmem_alloc(vsnprintf(NULL, 0, fmt, adx) + 1,
-	    KM_SLEEP);
-
-	(void) vsprintf(ha->ha_history_str, fmt, adx);
-
+	va_copy(adx_copy, adx);
+	ha->ha_history_str = kmem_vasprintf(fmt, adx_copy);
+	va_end(adx_copy);
 	ha->ha_log_type = LOG_INTERNAL;
 	ha->ha_event = event;
 	ha->ha_zone = NULL;

module/zfs/spa_misc.c

@@ -438,7 +438,7 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
 
 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
 
-	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
+	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP | KM_NODEBUG);
 
 	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);

module/zfs/txg.c

@@ -49,7 +49,7 @@ txg_init(dsl_pool_t *dp, uint64_t txg)
 	int c;
 	bzero(tx, sizeof (tx_state_t));
 
-	tx->tx_cpu = kmem_zalloc(max_ncpus * sizeof (tx_cpu_t), KM_SLEEP);
+	tx->tx_cpu = vmem_zalloc(max_ncpus * sizeof (tx_cpu_t), KM_SLEEP);
 
 	for (c = 0; c < max_ncpus; c++) {
 		int i;
@@ -107,7 +107,7 @@ txg_fini(dsl_pool_t *dp)
 	if (tx->tx_commit_cb_taskq != NULL)
 		taskq_destroy(tx->tx_commit_cb_taskq);
 
-	kmem_free(tx->tx_cpu, max_ncpus * sizeof (tx_cpu_t));
+	vmem_free(tx->tx_cpu, max_ncpus * sizeof (tx_cpu_t));
 
 	bzero(tx, sizeof (tx_state_t));
 }

module/zfs/zap_micro.c

@@ -532,7 +532,7 @@ mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 
 	sz = zap->zap_dbuf->db_size;
-	mzp = kmem_alloc(sz, KM_SLEEP);
+	mzp = vmem_alloc(sz, KM_SLEEP);
 	bcopy(zap->zap_dbuf->db_data, mzp, sz);
 	nchunks = zap->zap_m.zap_num_chunks;
 
@@ -540,7 +540,7 @@ mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
 		err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
 		    1ULL << fzap_default_block_shift, 0, tx);
 		if (err) {
-			kmem_free(mzp, sz);
+			vmem_free(mzp, sz);
 			return (err);
 		}
 	}
@@ -566,7 +566,7 @@ mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
 		if (err)
 			break;
 	}
-	kmem_free(mzp, sz);
+	vmem_free(mzp, sz);
 	*zapp = zap;
 	return (err);
 }

module/zfs/zfs_ioctl.c

@@ -170,7 +170,7 @@ history_str_get(zfs_cmd_t *zc)
 	if (zc->zc_history == 0)
 		return (NULL);
 
-	buf = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP);
+	buf = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP | KM_NODEBUG);
 	if (copyinstr((void *)(uintptr_t)zc->zc_history,
 	    buf, HIS_MAX_RECORD_LEN, NULL) != 0) {
 		history_str_free(buf);
@@ -1027,7 +1027,7 @@ get_nvlist(uint64_t nvl, uint64_t size, int iflag, nvlist_t **nvp)
 	if (size == 0)
 		return (EINVAL);
 
-	packed = kmem_alloc(size, KM_SLEEP);
+	packed = kmem_alloc(size, KM_SLEEP | KM_NODEBUG);
 
 	if ((error = ddi_copyin((void *)(uintptr_t)nvl, packed, size,
 	    iflag)) != 0) {
@@ -1093,7 +1093,7 @@ put_nvlist(zfs_cmd_t *zc, nvlist_t *nvl)
 	if (size > zc->zc_nvlist_dst_size) {
 		error = ENOMEM;
 	} else {
-		packed = kmem_alloc(size, KM_SLEEP);
+		packed = kmem_alloc(size, KM_SLEEP | KM_NODEBUG);
 		VERIFY(nvlist_pack(nvl, &packed, &size, NV_ENCODE_NATIVE,
 		    KM_SLEEP) == 0);
 		if (ddi_copyout(packed, (void *)(uintptr_t)zc->zc_nvlist_dst,
@@ -5081,7 +5081,7 @@ zfsdev_ioctl(struct file *filp, unsigned cmd, unsigned long arg)
 	if (vec >= sizeof (zfs_ioc_vec) / sizeof (zfs_ioc_vec[0]))
 		return (-EINVAL);
 
-	zc = kmem_zalloc(sizeof (zfs_cmd_t), KM_SLEEP);
+	zc = kmem_zalloc(sizeof (zfs_cmd_t), KM_SLEEP | KM_NODEBUG);
 
 	error = ddi_copyin((void *)arg, zc, sizeof (zfs_cmd_t), flag);
 	if (error != 0)

module/zfs/zil.c

@@ -1075,7 +1075,7 @@ zil_itx_create(uint64_t txtype, size_t lrsize)
 
 	lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t);
 
-	itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP);
+	itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP|KM_NODEBUG);
 	itx->itx_lr.lrc_txtype = txtype;
 	itx->itx_lr.lrc_reclen = lrsize;
 	itx->itx_sod = lrsize; /* if write & WR_NEED_COPY will be increased */
@@ -1939,7 +1939,7 @@ zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE])
 	zr.zr_replay = replay_func;
 	zr.zr_arg = arg;
 	zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log);
-	zr.zr_lr = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP);
+	zr.zr_lr = vmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP);
 
 	/*
 	 * Wait for in-progress removes to sync before starting replay.
@@ -1951,7 +1951,7 @@ zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE])
 	ASSERT(zilog->zl_replay_blks == 0);
 	(void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr,
 	    zh->zh_claim_txg);
-	kmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE);
+	vmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE);
 
 	zil_destroy(zilog, B_FALSE);
 	txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);