OS-7058 need zfs trim support

Requires SPL: 4692ce3f158ebf17707f4462b8a39e005be00a10

Porting notes:

VOP_SPACE's porting equivalent VNOP_IOCTL(F_PUNCHHOLE), was only added
at High Sierra - no hole support before then. But this is on pool file
vdevs, so it is mostly for debugging use.

Currently the unusual way to create kstats in spa_misc.c fails, and
is commented out.

cmd/zpool/zpool_main.c

@@ -26,7 +26,7 @@
  * Copyright (c) 2012 by Frederik Wessels. All rights reserved.
  * Copyright (c) 2012 by Cyril Plisko. All rights reserved.
  * Copyright (c) 2013 by Prasad Joshi (sTec). All rights reserved.
- * Copyright 2016 Nexenta Systems, Inc.
+ * Copyright 2017 Nexenta Systems, Inc.
  * Copyright (c) 2017 Datto Inc.
  */
 
@@ -99,6 +99,7 @@ static int zpool_do_split(int, char **);
 
 static int zpool_do_initialize(int, char **);
 static int zpool_do_scrub(int, char **);
+static int zpool_do_trim(int, char **);
 
 static int zpool_do_import(int, char **);
 static int zpool_do_export(int, char **);
@@ -150,6 +151,7 @@ typedef enum {
 	HELP_REMOVE,
 	HELP_INITIALIZE,
 	HELP_SCRUB,
+	HELP_TRIM,
 	HELP_STATUS,
 	HELP_UPGRADE,
 	HELP_EVENTS,
@@ -278,6 +280,8 @@ static zpool_command_t command_table[] = {
 	{ "initialize",	zpool_do_initialize,	HELP_INITIALIZE		},
 	{ "scrub",	zpool_do_scrub,		HELP_SCRUB		},
 	{ NULL },
+	{ "trim",	zpool_do_trim,		HELP_TRIM		},
+	{ NULL },
 	{ "import",	zpool_do_import,	HELP_IMPORT		},
 	{ "export",	zpool_do_export,	HELP_EXPORT		},
 	{ "upgrade",	zpool_do_upgrade,	HELP_UPGRADE		},
@@ -358,6 +362,8 @@ get_usage(zpool_help_t idx) {
 		return (gettext("\tinitialize [-cs] <pool> [<device> ...]\n"));
 	case HELP_SCRUB:
 		return (gettext("\tscrub [-s | -p] <pool> ...\n"));
+	case HELP_TRIM:
+		return (gettext("\ttrim [-s|-r <rate>] <pool> ...\n"));
 	case HELP_STATUS:
 		return (gettext("\tstatus [-gLPvxD] [-T d|u] [pool] ... "
 		    "[interval [count]]\n"));
@@ -5728,6 +5734,31 @@ scrub_callback(zpool_handle_t *zhp, void *data)
 	return (err != 0);
 }
 
+typedef struct trim_cbdata {
+	boolean_t	cb_start;
+	uint64_t	cb_rate;
+} trim_cbdata_t;
+
+int
+trim_callback(zpool_handle_t *zhp, void *data)
+{
+	trim_cbdata_t *cb = data;
+	int err;
+
+	/*
+	 * Ignore faulted pools.
+	 */
+	if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
+		(void) fprintf(stderr, gettext("cannot trim '%s': pool is "
+		    "currently unavailable\n"), zpool_get_name(zhp));
+		return (1);
+	}
+
+	err = zpool_trim(zhp, cb->cb_start, cb->cb_rate);
+
+	return (err != 0);
+}
+
 /*
  * zpool scrub [-s | -p] <pool> ...
  *
@@ -5892,6 +5923,52 @@ zpool_do_initialize(int argc, char **argv)
 	return (err);
 }
 
+/*
+ * zpool trim [-s|-r <rate>] <pool> ...
+ *
+ *	-s		Stop. Stops any in-progress trim.
+ *	-r <rate>	Sets the TRIM rate.
+ */
+int
+zpool_do_trim(int argc, char **argv)
+{
+	int c;
+	trim_cbdata_t cb;
+
+	cb.cb_start = B_TRUE;
+	cb.cb_rate = 0;
+
+	/* check options */
+	while ((c = getopt(argc, argv, "sr:")) != -1) {
+		switch (c) {
+		case 's':
+			cb.cb_start = B_FALSE;
+			break;
+		case 'r':
+			if (zfs_nicestrtonum(NULL, optarg, &cb.cb_rate) == -1) {
+				(void) fprintf(stderr,
+				    gettext("invalid value for rate\n"));
+				usage(B_FALSE);
+			}
+			break;
+		case '?':
+			(void) fprintf(stderr, gettext("invalid option '%c'\n"),
+			    optopt);
+			usage(B_FALSE);
+		}
+	}
+
+	argc -= optind;
+	argv += optind;
+
+	if (argc < 1) {
+		(void) fprintf(stderr, gettext("missing pool name argument\n"));
+		usage(B_FALSE);
+	}
+
+	return (for_each_pool(argc, argv, B_TRUE, NULL, trim_callback, &cb));
+}
+
 typedef struct status_cbdata {
 	int		cb_count;
 	int		cb_name_flags;
@@ -6224,6 +6301,59 @@ print_checkpoint_status(pool_checkpoint_stat_t *pcs)
 	    space_buf);
 }
 
+static void
+print_trim_status(uint64_t trim_prog, uint64_t total_size, uint64_t rate,
+    uint64_t start_time_u64, uint64_t end_time_u64)
+{
+	time_t start_time = start_time_u64, end_time = end_time_u64;
+	char *buf;
+
+	assert(trim_prog <= total_size);
+	if (trim_prog != 0 && trim_prog != total_size) {
+		buf = ctime(&start_time);
+		buf[strlen(buf) - 1] = '\0';	/* strip trailing newline */
+		if (rate != 0) {
+			char rate_str[32];
+			zfs_nicenum(rate, rate_str, sizeof (rate_str));
+			(void) printf("  trim: %.02f%%\tstarted: %s\t"
+			    "(rate: %s/s)\n", (((double)trim_prog) /
+			    total_size) * 100, buf, rate_str);
+		} else {
+			(void) printf("  trim: %.02f%%\tstarted: %s\t"
+			    "(rate: max)\n", (((double)trim_prog) /
+			    total_size) * 100, buf);
+		}
+	} else {
+		if (start_time != 0) {
+			/*
+			 * Non-zero start time means we were run at some point
+			 * in the past.
+			 */
+			if (end_time != 0) {
+				/* Non-zero end time means we completed */
+				time_t diff = end_time - start_time;
+				int hrs, mins;
+
+				buf = ctime(&end_time);
+				buf[strlen(buf) - 1] = '\0';
+				hrs = diff / 3600;
+				mins = (diff % 3600) / 60;
+				(void) printf(gettext("  trim: completed on %s "
+				    "(after %dh%dm)\n"), buf, hrs, mins);
+			} else {
+				buf = ctime(&start_time);
+				buf[strlen(buf) - 1] = '\0';
+				/* Zero end time means we were interrupted */
+				(void) printf(gettext("  trim: interrupted\t"
+				    "(started %s)\n"), buf);
+			}
+		} else {
+			/* trim was never run */
+			(void) printf(gettext("  trim: none requested\n"));
+		}
+	}
+}
+
 static void
 print_error_log(zpool_handle_t *zhp)
 {
@@ -6335,6 +6465,43 @@ print_dedup_stats(nvlist_t *config)
 	zpool_dump_ddt(dds, ddh);
 }
 
+/*
+ * Calculates the total space available on log devices on the pool.
+ * For whatever reason, this is not counted in the root vdev's space stats.
+ */
+static uint64_t
+zpool_slog_space(nvlist_t *nvroot)
+{
+	nvlist_t **newchild;
+	uint_t c, children;
+	uint64_t space = 0;
+
+	verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
+	    &newchild, &children) == 0);
+
+	for (c = 0; c < children; c++) {
+		uint64_t islog = B_FALSE;
+		vdev_stat_t *vs;
+		uint_t n;
+		uint_t n_subchildren = 1;
+		nvlist_t **subchild;
+
+		(void) nvlist_lookup_uint64(newchild[c], ZPOOL_CONFIG_IS_LOG,
+		    &islog);
+		if (!islog)
+			continue;
+		verify(nvlist_lookup_uint64_array(newchild[c],
+		    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &n) == 0);
+
+		/* vdev can be non-leaf, so multiply by number of children */
+		(void) nvlist_lookup_nvlist_array(newchild[c],
+		    ZPOOL_CONFIG_CHILDREN, &subchild, &n_subchildren);
+		space += n_subchildren * vs->vs_space;
+	}
+
+	return (space);
+}
+
 /*
  * Display a summary of pool status.  Displays a summary such as:
  *
@@ -6638,6 +6805,7 @@ status_callback(zpool_handle_t *zhp, void *data)
 		pool_checkpoint_stat_t *pcs = NULL;
 		pool_scan_stat_t *ps = NULL;
 		pool_removal_stat_t *prs = NULL;
+		uint64_t trim_prog, trim_rate, trim_start_time, trim_stop_time;
 
 		(void) nvlist_lookup_uint64_array(nvroot,
 		    ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t **)&pcs, &c);
@@ -6651,6 +6819,24 @@ status_callback(zpool_handle_t *zhp, void *data)
 		print_removal_status(zhp, prs);
 		print_checkpoint_status(pcs);
 
+		/* Grab trim stats if the pool supports it */
+		if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_TRIM_PROG,
+				&trim_prog) == 0 &&
+		    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TRIM_RATE,
+				&trim_rate) == 0 &&
+		    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TRIM_START_TIME,
+				&trim_start_time) == 0 &&
+		    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TRIM_STOP_TIME,
+				&trim_stop_time) == 0) {
+			/*
+			 * For whatever reason, root vdev_stats_t don't
+			 * include log devices.
+			 */
+			print_trim_status(trim_prog, vs->vs_space +
+			    zpool_slog_space(nvroot), trim_rate,
+			    trim_start_time, trim_stop_time);
+		}
+
 		namewidth = max_width(zhp, nvroot, 0, 0, 0);
 		if (namewidth < 10)
 			namewidth = 10;

include/libzfs.h

@@ -290,6 +290,7 @@ typedef struct splitflags {
 extern int zpool_scan(zpool_handle_t *, pool_scan_func_t, pool_scrub_cmd_t);
 extern int zpool_initialize(zpool_handle_t *, pool_initialize_func_t,
     nvlist_t *);
+extern int zpool_trim(zpool_handle_t *, boolean_t start, uint64_t rate);
 extern int zpool_clear(zpool_handle_t *, const char *, nvlist_t *);
 extern int zpool_reguid(zpool_handle_t *);
 extern int zpool_reopen(zpool_handle_t *);

include/sys/dmu.h

@@ -21,7 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
- * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2017 Nexenta Systems, Inc. All rights reserved.
  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  * Copyright 2014 HybridCluster. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
@@ -370,6 +370,8 @@ typedef struct dmu_buf {
 #define	DMU_POOL_OBSOLETE_BPOBJ		"com.delphix:obsolete_bpobj"
 #define	DMU_POOL_CONDENSING_INDIRECT	"com.delphix:condensing_indirect"
 #define	DMU_POOL_ZPOOL_CHECKPOINT	"com.delphix:zpool_checkpoint"
+#define	DMU_POOL_TRIM_START_TIME	"trim_start_time"
+#define	DMU_POOL_TRIM_STOP_TIME		"trim_stop_time"
 
 /*
  * Allocate an object from this objset.  The range of object numbers

include/sys/fs/zfs.h

@@ -239,6 +239,8 @@ typedef enum {
 	ZPOOL_PROP_TNAME,
 	ZPOOL_PROP_BOOTSIZE,
 	ZPOOL_PROP_CHECKPOINT,
+	ZPOOL_PROP_FORCETRIM,
+	ZPOOL_PROP_AUTOTRIM,
 	ZPOOL_NUM_PROPS
 } zpool_prop_t;
 
@@ -697,6 +699,10 @@ typedef struct zpool_load_policy {
 #define	ZPOOL_CONFIG_REMOVED		"removed"
 #define	ZPOOL_CONFIG_FRU		"fru"
 #define	ZPOOL_CONFIG_AUX_STATE		"aux_state"
+#define	ZPOOL_CONFIG_TRIM_PROG		"trim_prog"
+#define	ZPOOL_CONFIG_TRIM_RATE		"trim_rate"
+#define	ZPOOL_CONFIG_TRIM_START_TIME	"trim_start_time"
+#define	ZPOOL_CONFIG_TRIM_STOP_TIME	"trim_stop_time"
 
 /* Pool load policy parameters */
 #define	ZPOOL_LOAD_POLICY		"load-policy"
@@ -826,6 +832,14 @@ typedef enum pool_scan_func {
 	POOL_SCAN_FUNCS
 } pool_scan_func_t;
 
+/*
+ * TRIM command configuration info.
+ */
+typedef struct trim_cmd_info_s {
+	uint64_t	tci_start;	/* B_TRUE = start; B_FALSE = stop */
+	uint64_t	tci_rate;	/* requested TRIM rate in bytes/sec */
+} trim_cmd_info_t;
+
 /*
  * Used to control scrub pause and resume.
  */

include/sys/kstat_osx.h

@@ -136,6 +136,10 @@ typedef struct osx_kstat {
 
 	kstat_named_t zfs_lua_max_instrlimit;
 	kstat_named_t zfs_lua_max_memlimit;
+
+	kstat_named_t zfs_trim;
+	kstat_named_t zfs_trim_min_ext_sz;
+
 } osx_kstat_t;
 
 
@@ -232,6 +236,9 @@ extern uint64_t zfs_vdev_file_size_mismatch_cnt;
 extern uint64_t zfs_lua_max_instrlimit;
 extern uint64_t zfs_lua_max_memlimit;
 
+extern uint64_t zfs_trim;
+extern uint64_t zfs_trim_min_ext_sz;
+
 int        kstat_osx_init(void);
 void       kstat_osx_fini(void);
 

include/sys/ldi_impl_osx.h

@@ -41,6 +41,8 @@
 #define	_SYS_LDI_IMPL_OSX_H
 
 #include <sys/ldi_osx.h>
+#include <sys/disk.h>
+#include <sys/dkio.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -139,6 +141,8 @@ int ldi_open_media_by_dev(dev_t, int, ldi_handle_t *);
 int ldi_open_media_by_path(char *, int, ldi_handle_t *);
 int handle_get_bootinfo_iokit(struct ldi_handle *,
     struct io_bootinfo *);
+int handle_unmap_iokit(struct ldi_handle *,
+    dkioc_free_list_t *);
 
 /* Handle vnode functions */
 dev_t dev_from_path(char *);
@@ -159,6 +163,8 @@ int buf_strategy_vnode(ldi_buf_t *, struct ldi_handle *);
 int ldi_open_vnode_by_path(char *, dev_t, int, ldi_handle_t *);
 int handle_get_bootinfo_vnode(struct ldi_handle *,
     struct io_bootinfo *);
+int handle_unmap_vnode(struct ldi_handle *,
+    dkioc_free_list_t *);
 
 /*
  * LDI event information