Add DDT prune command

Requires the new 'flat' physical data which has the start
time for a class entry.

The amount to prune can be based on a target percentage of
the unique entries or based on the age (i.e., every entry
older than N days).

Sponsored-by: Klara, Inc.
Sponsored-by: iXsystems, Inc.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Don Brady <don.brady@klarasystems.com>
Closes #16277

include/libzfs.h

@@ -305,6 +305,9 @@ _LIBZFS_H int zpool_reopen_one(zpool_handle_t *, void *);
 
 _LIBZFS_H int zpool_sync_one(zpool_handle_t *, void *);
 
+_LIBZFS_H int zpool_ddt_prune(zpool_handle_t *, zpool_ddt_prune_unit_t,
+    uint64_t);
+
 _LIBZFS_H int zpool_vdev_online(zpool_handle_t *, const char *, int,
     vdev_state_t *);
 _LIBZFS_H int zpool_vdev_offline(zpool_handle_t *, const char *, boolean_t);

include/libzfs_core.h

@@ -161,6 +161,9 @@ _LIBZFS_CORE_H int lzc_set_vdev_prop(const char *, nvlist_t *, nvlist_t **);
 
 _LIBZFS_CORE_H int lzc_scrub(zfs_ioc_t, const char *, nvlist_t *, nvlist_t **);
 
+_LIBZFS_CORE_H int lzc_ddt_prune(const char *, zpool_ddt_prune_unit_t,
+    uint64_t);
+
 #ifdef	__cplusplus
 }
 #endif

include/sys/ddt.h

@@ -405,6 +405,9 @@ extern int ddt_walk(spa_t *spa, ddt_bookmark_t *ddb,
 
 extern boolean_t ddt_addref(spa_t *spa, const blkptr_t *bp);
 
+extern int ddt_prune_unique_entries(spa_t *spa, zpool_ddt_prune_unit_t unit,
+    uint64_t amount);
+
 #ifdef	__cplusplus
 }
 #endif

include/sys/ddt_impl.h

@@ -35,8 +35,11 @@ extern "C" {
 #endif
 
 /* DDT version numbers */
-#define	DDT_VERSION_LEGACY	(0)
-#define	DDT_VERSION_FDT		(1)
+#define	DDT_VERSION_LEGACY		(0)
+#define	DDT_VERSION_FDT			(1)
+
+/* Dummy version to signal that configure is still necessary */
+#define	DDT_VERSION_UNCONFIGURED	(UINT64_MAX)
 
 /* Names of interesting objects in the DDT root dir */
 #define	DDT_DIR_VERSION		"version"
@@ -187,8 +190,11 @@ extern void ddt_log_commit(ddt_t *ddt, ddt_log_update_t *dlu);
 
 extern boolean_t ddt_log_take_first(ddt_t *ddt, ddt_log_t *ddl,
     ddt_lightweight_entry_t *ddlwe);
-extern boolean_t ddt_log_take_key(ddt_t *ddt, ddt_log_t *ddl,
-    const ddt_key_t *ddk, ddt_lightweight_entry_t *ddlwe);
+
+extern boolean_t ddt_log_find_key(ddt_t *ddt, const ddt_key_t *ddk,
+    ddt_lightweight_entry_t *ddlwe);
+extern boolean_t ddt_log_remove_key(ddt_t *ddt, ddt_log_t *ddl,
+    const ddt_key_t *ddk);
 
 extern void ddt_log_checkpoint(ddt_t *ddt, ddt_lightweight_entry_t *ddlwe,
     dmu_tx_t *tx);
@@ -211,6 +217,44 @@ extern void ddt_log_fini(void);
  * them up.
  */
 
+/*
+ * We use a histogram to convert a percentage request into a
+ * cutoff value where entries older than the cutoff get pruned.
+ *
+ * The histogram bins represent hours in power-of-two increments.
+ * 16 bins covers up to four years.
+ */
+#define	HIST_BINS 16
+
+typedef struct ddt_age_histo {
+	uint64_t dah_entries;
+	uint64_t dah_age_histo[HIST_BINS];
+} ddt_age_histo_t;
+
+void ddt_prune_walk(spa_t *spa, uint64_t cutoff, ddt_age_histo_t *histogram);
+
+#if defined(_KERNEL) || !defined(ZFS_DEBUG)
+#define	ddt_dump_age_histogram(histo, cutoff)	((void)0)
+#else
+static inline void
+ddt_dump_age_histogram(ddt_age_histo_t *histogram, uint64_t cutoff)
+{
+	if (histogram->dah_entries == 0)
+		return;
+
+	(void) printf("DDT prune unique class age, %llu hour cutoff\n",
+	    (u_longlong_t)(gethrestime_sec() - cutoff)/3600);
+	(void) printf("%5s  %9s  %4s\n", "age", "blocks", "amnt");
+	(void) printf("%5s  %9s  %4s\n", "-----", "---------", "----");
+	for (int i = 0; i < HIST_BINS; i++) {
+		(void) printf("%5d  %9llu %4d%%\n", 1<<i,
+		    (u_longlong_t)histogram->dah_age_histo[i],
+		    (int)((histogram->dah_age_histo[i] * 100) /
+		    histogram->dah_entries));
+	}
+}
+#endif
+
 /*
  * Enough room to expand DMU_POOL_DDT format for all possible DDT
  * checksum/class/type combinations.

include/sys/fs/zfs.h

@@ -1422,7 +1422,7 @@ typedef enum {
  */
 typedef enum zfs_ioc {
 	/*
-	 * Core features - 88/128 numbers reserved.
+	 * Core features - 89/128 numbers reserved.
 	 */
 #ifdef __FreeBSD__
 	ZFS_IOC_FIRST =	0,
@@ -1519,6 +1519,7 @@ typedef enum zfs_ioc {
 	ZFS_IOC_VDEV_SET_PROPS,			/* 0x5a56 */
 	ZFS_IOC_POOL_SCRUB,			/* 0x5a57 */
 	ZFS_IOC_POOL_PREFETCH,			/* 0x5a58 */
+	ZFS_IOC_DDT_PRUNE,			/* 0x5a59 */
 
 	/*
 	 * Per-platform (Optional) - 8/128 numbers reserved.
@@ -1655,6 +1656,12 @@ typedef enum {
 	ZPOOL_PREFETCH_DDT
 } zpool_prefetch_type_t;
 
+typedef enum {
+	ZPOOL_DDT_PRUNE_NONE,
+	ZPOOL_DDT_PRUNE_AGE,		/* in seconds */
+	ZPOOL_DDT_PRUNE_PERCENTAGE,	/* 1 - 100 */
+} zpool_ddt_prune_unit_t;
+
 /*
  * Bookmark name values.
  */
@@ -1753,6 +1760,12 @@ typedef enum {
  */
 #define	ZPOOL_PREFETCH_TYPE		"prefetch_type"
 
+/*
+ * The following are names used when invoking ZFS_IOC_DDT_PRUNE.
+ */
+#define	DDT_PRUNE_UNIT		"ddt_prune_unit"
+#define	DDT_PRUNE_AMOUNT	"ddt_prune_amount"
+
 /*
  * Flags for ZFS_IOC_VDEV_SET_STATE
  */

include/sys/spa_impl.h

@@ -412,6 +412,7 @@ struct spa {
 	uint64_t	spa_dedup_dspace;	/* Cache get_dedup_dspace() */
 	uint64_t	spa_dedup_checksum;	/* default dedup checksum */
 	uint64_t	spa_dspace;		/* dspace in normal class */
+	boolean_t	spa_active_ddt_prune;	/* ddt prune process active */
 	struct brt	*spa_brt;		/* in-core BRT */
 	kmutex_t	spa_vdev_top_lock;	/* dueling offline/remove */
 	kmutex_t	spa_proc_lock;		/* protects spa_proc* */