FreeBSD: Add support for procfs_list

The procfs_list interface is required by several kstats. Implement
this functionality for FreeBSD to provide access to these kstats.
                           
Reviewed-by: Allan Jude <[email protected]>
Reviewed-by: Ryan Moeller <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Matt Macy <[email protected]>
Closes #10890 

11 files changed, 201 insertions, 41 deletions

diff --git a/include/os/freebsd/spl/sys/kstat.h b/include/os/freebsd/spl/sys/kstat.h
index 74c3da8ec..5ceb88b29 100644
--- a/include/os/freebsd/spl/sys/kstat.h
+++ b/include/os/freebsd/spl/sys/kstat.h
@@ -83,6 +83,14 @@ typedef struct kstat_s kstat_t;
 typedef int kid_t;				/* unique kstat id */
 typedef int kstat_update_t(struct kstat_s *, int); /* dynamic update cb */
 
+struct seq_file {
+	char *sf_buf;
+	size_t sf_size;
+};
+
+void seq_printf(struct seq_file *m, const char *fmt, ...);
+
+
 typedef struct kstat_module {
 	char ksm_name[KSTAT_STRLEN+1];		/* module name */
 	struct list_head ksm_module_list;	/* module linkage */
@@ -92,6 +100,7 @@ typedef struct kstat_module {
 
 typedef struct kstat_raw_ops {
 	int (*headers)(char *buf, size_t size);
+	int (*seq_headers)(struct seq_file *);
 	int (*data)(char *buf, size_t size, void *data);
 	void *(*addr)(kstat_t *ksp, loff_t index);
 } kstat_raw_ops_t;
@@ -112,6 +121,7 @@ struct kstat_s {
 	size_t		ks_data_size;		/* size of kstat data section */
 	kstat_update_t	*ks_update;		/* dynamic updates */
 	void		*ks_private;		/* private data */
+	void		*ks_private1;		/* private data */
 	kmutex_t	ks_private_lock;	/* kstat private data lock */
 	kmutex_t	*ks_lock;		/* kstat data lock */
 	struct list_head ks_list;		/* kstat linkage */
@@ -185,6 +195,12 @@ extern void __kstat_set_raw_ops(kstat_t *ksp,
     int (*data)(char *buf, size_t size, void *data),
     void* (*addr)(kstat_t *ksp, loff_t index));
 
+extern void __kstat_set_seq_raw_ops(kstat_t *ksp,
+    int (*headers)(struct seq_file *),
+    int (*data)(char *buf, size_t size, void *data),
+    void* (*addr)(kstat_t *ksp, loff_t index));
+
+
 extern kstat_t *__kstat_create(const char *ks_module, int ks_instance,
     const char *ks_name, const char *ks_class, uchar_t ks_type,
     uint_t ks_ndata, uchar_t ks_flags);
@@ -196,6 +212,8 @@ extern void kstat_waitq_exit(kstat_io_t *);
 extern void kstat_runq_enter(kstat_io_t *);
 extern void kstat_runq_exit(kstat_io_t *);
 
+#define	kstat_set_seq_raw_ops(k, h, d, a) \
+    __kstat_set_seq_raw_ops(k, h, d, a)
 #define	kstat_set_raw_ops(k, h, d, a) \
     __kstat_set_raw_ops(k, h, d, a)
 #define	kstat_create(m, i, n, c, t, s, f) \
diff --git a/include/os/freebsd/spl/sys/procfs_list.h b/include/os/freebsd/spl/sys/procfs_list.h
index 5d623c369..da13f0387 100644
--- a/include/os/freebsd/spl/sys/procfs_list.h
+++ b/include/os/freebsd/spl/sys/procfs_list.h
@@ -33,16 +33,18 @@
  * procfs list manipulation
  */
 
-struct seq_file { };
-void seq_printf(struct seq_file *m, const char *fmt, ...);
-
-typedef struct procfs_list {
+typedef struct procfs_list procfs_list_t;
+struct procfs_list {
 	void		*pl_private;
+	void		*pl_next_data;
 	kmutex_t	pl_lock;
 	list_t		pl_list;
 	uint64_t	pl_next_id;
+	int		(*pl_show)(struct seq_file *f, void *p);
+	int		(*pl_show_header)(struct seq_file *f);
+	int		(*pl_clear)(procfs_list_t *procfs_list);
 	size_t		pl_node_offset;
-} procfs_list_t;
+};
 
 typedef struct procfs_list_node {
 	list_node_t	pln_link;
@@ -50,6 +52,7 @@ typedef struct procfs_list_node {
 } procfs_list_node_t;
 
 void procfs_list_install(const char *module,
+    const char *submodule,
     const char *name,
     mode_t mode,
     procfs_list_t *procfs_list,
diff --git a/include/os/linux/spl/sys/procfs_list.h b/include/os/linux/spl/sys/procfs_list.h
index eb1519c0a..9bb437f55 100644
--- a/include/os/linux/spl/sys/procfs_list.h
+++ b/include/os/linux/spl/sys/procfs_list.h
@@ -57,6 +57,7 @@ typedef struct procfs_list_node {
 } procfs_list_node_t;
 
 void procfs_list_install(const char *module,
+    const char *submodule,
     const char *name,
     mode_t mode,
     procfs_list_t *procfs_list,
diff --git a/include/sys/zfs_context.h b/include/sys/zfs_context.h
index e33f52c17..d9f5ed580 100644
--- a/include/sys/zfs_context.h
+++ b/include/sys/zfs_context.h
@@ -386,6 +386,7 @@ typedef struct procfs_list_node {
 } procfs_list_node_t;
 
 void procfs_list_install(const char *module,
+    const char *submodule,
     const char *name,
     mode_t mode,
     procfs_list_t *procfs_list,
diff --git a/lib/libzpool/kernel.c b/lib/libzpool/kernel.c
index 145b21d40..ca3578993 100644
--- a/lib/libzpool/kernel.c
+++ b/lib/libzpool/kernel.c
@@ -444,6 +444,7 @@ seq_printf(struct seq_file *m, const char *fmt, ...)
 
 void
 procfs_list_install(const char *module,
+    const char *submodule,
     const char *name,
     mode_t mode,
     procfs_list_t *procfs_list,
diff --git a/module/os/freebsd/spl/spl_kstat.c b/module/os/freebsd/spl/spl_kstat.c
index f21575142..4cc77e20a 100644
--- a/module/os/freebsd/spl/spl_kstat.c
+++ b/module/os/freebsd/spl/spl_kstat.c
@@ -55,6 +55,17 @@ __kstat_set_raw_ops(kstat_t *ksp,
 	ksp->ks_raw_ops.addr    = addr;
 }
 
+void
+__kstat_set_seq_raw_ops(kstat_t *ksp,
+    int (*headers)(struct seq_file *f),
+    int (*data)(char *buf, size_t size, void *data),
+    void *(*addr)(kstat_t *ksp, loff_t index))
+{
+	ksp->ks_raw_ops.seq_headers = headers;
+	ksp->ks_raw_ops.data    = data;
+	ksp->ks_raw_ops.addr    = addr;
+}
+
 static int
 kstat_default_update(kstat_t *ksp, int rw)
 {
@@ -160,7 +171,7 @@ kstat_sysctl_raw(SYSCTL_HANDLER_ARGS)
 	void *data;
 	kstat_t *ksp = arg1;
 	void *(*addr_op)(kstat_t *ksp, loff_t index);
-	int n, rc = 0;
+	int n, has_header, rc = 0;
 
 	sb = sbuf_new_auto();
 	if (sb == NULL)
@@ -180,14 +191,25 @@ kstat_sysctl_raw(SYSCTL_HANDLER_ARGS)
 	ksp->ks_raw_buf = malloc(PAGE_SIZE, M_TEMP, M_WAITOK);
 
 	n = 0;
+	has_header = (ksp->ks_raw_ops.headers ||
+	    ksp->ks_raw_ops.seq_headers);
+
 restart_headers:
 	if (ksp->ks_raw_ops.headers) {
 		rc = ksp->ks_raw_ops.headers(
 		    ksp->ks_raw_buf, ksp->ks_raw_bufsize);
+	} else if (ksp->ks_raw_ops.seq_headers) {
+		struct seq_file f;
+
+		f.sf_buf = ksp->ks_raw_buf;
+		f.sf_size = ksp->ks_raw_bufsize;
+		rc = ksp->ks_raw_ops.seq_headers(&f);
+	}
+	if (has_header) {
 		if (rc == ENOMEM && !kstat_resize_raw(ksp))
 			goto restart_headers;
 		if (rc == 0)
-			sbuf_printf(sb, "%s", ksp->ks_raw_buf);
+			sbuf_printf(sb, "\n%s", ksp->ks_raw_buf);
 	}
 
 	while ((data = addr_op(ksp, n)) != NULL) {
@@ -220,16 +242,21 @@ kstat_t *
 __kstat_create(const char *module, int instance, const char *name,
     const char *class, uchar_t ks_type, uint_t ks_ndata, uchar_t flags)
 {
+	char buf[KSTAT_STRLEN];
 	struct sysctl_oid *root;
 	kstat_t *ksp;
+	char *pool;
 
 	KASSERT(instance == 0, ("instance=%d", instance));
 	if ((ks_type == KSTAT_TYPE_INTR) || (ks_type == KSTAT_TYPE_IO))
 		ASSERT(ks_ndata == 1);
 
+	if (class == NULL)
+		class = "misc";
+
 	/*
-	 * Allocate the main structure. We don't need to copy module/class/name
-	 * stuff in here, because it is only used for sysctl node creation
+	 * Allocate the main structure. We don't need to keep a copy of
+	 * module in here, because it is only used for sysctl node creation
 	 * done in this function.
 	 */
 	ksp = malloc(sizeof (*ksp), M_KSTAT, M_WAITOK|M_ZERO);
@@ -237,8 +264,8 @@ __kstat_create(const char *module, int instance, const char *name,
 	ksp->ks_crtime = gethrtime();
 	ksp->ks_snaptime = ksp->ks_crtime;
 	ksp->ks_instance = instance;
-	strncpy(ksp->ks_name, name, KSTAT_STRLEN);
-	strncpy(ksp->ks_class, class, KSTAT_STRLEN);
+	(void) strlcpy(ksp->ks_name, name, KSTAT_STRLEN);
+	(void) strlcpy(ksp->ks_class, class, KSTAT_STRLEN);
 	ksp->ks_type = ks_type;
 	ksp->ks_flags = flags;
 	ksp->ks_update = kstat_default_update;
@@ -280,10 +307,22 @@ __kstat_create(const char *module, int instance, const char *name,
 			ksp = NULL;
 		}
 	}
+
+	/*
+	 * Some kstats use a module name like "zfs/poolname" to distinguish a
+	 * set of kstats belonging to a specific pool.  Split on '/' to add an
+	 * extra node for the pool name if needed.
+	 */
+	(void) strlcpy(buf, module, KSTAT_STRLEN);
+	module = buf;
+	pool = strchr(module, '/');
+	if (pool != NULL)
+		*pool++ = '\0';
+
 	/*
 	 * Create sysctl tree for those statistics:
 	 *
-	 *	kstat.<module>.<class>.<name>.
+	 *	kstat.<module>[.<pool>].<class>.<name>
 	 */
 	sysctl_ctx_init(&ksp->ks_sysctl_ctx);
 	root = SYSCTL_ADD_NODE(&ksp->ks_sysctl_ctx,
@@ -295,11 +334,26 @@ __kstat_create(const char *module, int instance, const char *name,
 		free(ksp, M_KSTAT);
 		return (NULL);
 	}
+	if (pool != NULL) {
+		root = SYSCTL_ADD_NODE(&ksp->ks_sysctl_ctx,
+		    SYSCTL_CHILDREN(root), OID_AUTO, pool, CTLFLAG_RW, 0, "");
+		if (root == NULL) {
+			printf("%s: Cannot create kstat.%s.%s tree!\n",
+			    __func__, module, pool);
+			sysctl_ctx_free(&ksp->ks_sysctl_ctx);
+			free(ksp, M_KSTAT);
+			return (NULL);
+		}
+	}
 	root = SYSCTL_ADD_NODE(&ksp->ks_sysctl_ctx, SYSCTL_CHILDREN(root),
 	    OID_AUTO, class, CTLFLAG_RW, 0, "");
 	if (root == NULL) {
-		printf("%s: Cannot create kstat.%s.%s tree!\n", __func__,
-		    module, class);
+		if (pool != NULL)
+			printf("%s: Cannot create kstat.%s.%s.%s tree!\n",
+			    __func__, module, pool, class);
+		else
+			printf("%s: Cannot create kstat.%s.%s tree!\n",
+			    __func__, module, class);
 		sysctl_ctx_free(&ksp->ks_sysctl_ctx);
 		free(ksp, M_KSTAT);
 		return (NULL);
@@ -309,8 +363,13 @@ __kstat_create(const char *module, int instance, const char *name,
 		    SYSCTL_CHILDREN(root),
 		    OID_AUTO, name, CTLFLAG_RW, 0, "");
 		if (root == NULL) {
-			printf("%s: Cannot create kstat.%s.%s.%s tree!\n",
-			    __func__, module, class, name);
+			if (pool != NULL)
+				printf("%s: Cannot create kstat.%s.%s.%s.%s "
+				    "tree!\n", __func__, module, pool, class,
+				    name);
+			else
+				printf("%s: Cannot create kstat.%s.%s.%s "
+				    "tree!\n", __func__, module, class, name);
 			sysctl_ctx_free(&ksp->ks_sysctl_ctx);
 			free(ksp, M_KSTAT);
 			return (NULL);
@@ -411,7 +470,6 @@ kstat_install(kstat_t *ksp)
 	switch (ksp->ks_type) {
 	case KSTAT_TYPE_NAMED:
 		return (kstat_install_named(ksp));
-		break;
 	case KSTAT_TYPE_RAW:
 		if (ksp->ks_raw_ops.data) {
 			root = SYSCTL_ADD_PROC(&ksp->ks_sysctl_ctx,
@@ -426,7 +484,6 @@ kstat_install(kstat_t *ksp)
 			    CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
 			    ksp, 0, kstat_sysctl_raw, "", ksp->ks_name);
 		}
-		VERIFY(root != NULL);
 		break;
 	case KSTAT_TYPE_IO:
 		root = SYSCTL_ADD_PROC(&ksp->ks_sysctl_ctx,
@@ -440,6 +497,7 @@ kstat_install(kstat_t *ksp)
 	default:
 		panic("unsupported kstat type %d\n", ksp->ks_type);
 	}
+	VERIFY(root != NULL);
 	ksp->ks_sysctl_root = root;
 }
 
diff --git a/module/os/freebsd/spl/spl_procfs_list.c b/module/os/freebsd/spl/spl_procfs_list.c
index 7b4ae9d0e..e8448ce00 100644
--- a/module/os/freebsd/spl/spl_procfs_list.c
+++ b/module/os/freebsd/spl/spl_procfs_list.c
@@ -32,12 +32,74 @@ __FBSDID("$FreeBSD$");
 #include <sys/mutex.h>
 #include <sys/procfs_list.h>
 
+typedef struct procfs_list_iter {
+	procfs_list_t *pli_pl;
+	void *pli_elt;
+} pli_t;
+
 void
-seq_printf(struct seq_file *m, const char *fmt, ...)
-{}
+seq_printf(struct seq_file *f, const char *fmt, ...)
+{
+	va_list adx;
+
+	va_start(adx, fmt);
+	(void) vsnprintf(f->sf_buf, f->sf_size, fmt, adx);
+	va_end(adx);
+}
+
+static int
+procfs_list_update(kstat_t *ksp, int rw)
+{
+	procfs_list_t *pl = ksp->ks_private;
+
+	if (rw == KSTAT_WRITE)
+		pl->pl_clear(pl);
+
+	return (0);
+}
+
+static int
+procfs_list_data(char *buf, size_t size, void *data)
+{
+	pli_t *p;
+	void *elt;
+	procfs_list_t *pl;
+	struct seq_file f;
+
+	p = data;
+	pl = p->pli_pl;
+	elt = p->pli_elt;
+	free(p, M_TEMP);
+	f.sf_buf = buf;
+	f.sf_size = size;
+	return (pl->pl_show(&f, elt));
+}
+
+static void *
+procfs_list_addr(kstat_t *ksp, loff_t n)
+{
+	procfs_list_t *pl = ksp->ks_private;
+	void *elt = ksp->ks_private1;
+	pli_t *p = NULL;
+
+
+	if (n == 0)
+		ksp->ks_private1 = list_head(&pl->pl_list);
+	else if (elt)
+		ksp->ks_private1 = list_next(&pl->pl_list, elt);
+
+	if (ksp->ks_private1) {
+		p = malloc(sizeof (*p), M_TEMP, M_WAITOK);
+		p->pli_pl = pl;
+		p->pli_elt = ksp->ks_private1;
+	}
+
+	return (p);
+}
 
 void
 procfs_list_install(const char *module,
+    const char *submodule,
     const char *name,
     mode_t mode,
     procfs_list_t *procfs_list,
@@ -46,12 +108,31 @@ procfs_list_install(const char *module,
     int (*clear)(procfs_list_t *procfs_list),
     size_t procfs_list_node_off)
 {
+	kstat_t *procfs_kstat;
+
 	mutex_init(&procfs_list->pl_lock, NULL, MUTEX_DEFAULT, NULL);
 	list_create(&procfs_list->pl_list,
 	    procfs_list_node_off + sizeof (procfs_list_node_t),
 	    procfs_list_node_off + offsetof(procfs_list_node_t, pln_link));
+	procfs_list->pl_show = show;
+	procfs_list->pl_show_header = show_header;
+	procfs_list->pl_clear = clear;
 	procfs_list->pl_next_id = 1;
 	procfs_list->pl_node_offset = procfs_list_node_off;
+
+	procfs_kstat =  kstat_create(module, 0, name, submodule,
+	    KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VIRTUAL);
+
+	if (procfs_kstat) {
+		procfs_kstat->ks_lock = &procfs_list->pl_lock;
+		procfs_kstat->ks_ndata = UINT32_MAX;
+		procfs_kstat->ks_private = procfs_list;
+		procfs_kstat->ks_update = procfs_list_update;
+		kstat_set_seq_raw_ops(procfs_kstat, show_header,
+		    procfs_list_data, procfs_list_addr);
+		kstat_install(procfs_kstat);
+		procfs_list->pl_private = procfs_kstat;
+	}
 }
 
 void
@@ -62,6 +143,7 @@ void
 procfs_list_destroy(procfs_list_t *procfs_list)
 {
 	ASSERT(list_is_empty(&procfs_list->pl_list));
+	kstat_delete(procfs_list->pl_private);
 	list_destroy(&procfs_list->pl_list);
 	mutex_destroy(&procfs_list->pl_lock);
 }
diff --git a/module/os/linux/spl/spl-procfs-list.c b/module/os/linux/spl/spl-procfs-list.c
index 189d6a7c6..778da7bc4 100644
--- a/module/os/linux/spl/spl-procfs-list.c
+++ b/module/os/linux/spl/spl-procfs-list.c
@@ -207,6 +207,7 @@ static const kstat_proc_op_t procfs_list_operations = {
  */
 void
 procfs_list_install(const char *module,
+    const char *submodule,
     const char *name,
     mode_t mode,
     procfs_list_t *procfs_list,
@@ -215,6 +216,12 @@ procfs_list_install(const char *module,
     int (*clear)(procfs_list_t *procfs_list),
     size_t procfs_list_node_off)
 {
+	char *modulestr;
+
+	if (submodule != NULL)
+		modulestr = kmem_asprintf("%s/%s", module, submodule);
+	else
+		modulestr = kmem_asprintf("%s", module);
 	mutex_init(&procfs_list->pl_lock, NULL, MUTEX_DEFAULT, NULL);
 	list_create(&procfs_list->pl_list,
 	    procfs_list_node_off + sizeof (procfs_list_node_t),
@@ -225,9 +232,10 @@ procfs_list_install(const char *module,
 	procfs_list->pl_clear = clear;
 	procfs_list->pl_node_offset = procfs_list_node_off;
 
-	kstat_proc_entry_init(&procfs_list->pl_kstat_entry, module, name);
+	kstat_proc_entry_init(&procfs_list->pl_kstat_entry, modulestr, name);
 	kstat_proc_entry_install(&procfs_list->pl_kstat_entry, mode,
 	    &procfs_list_operations, procfs_list);
+	kmem_strfree(modulestr);
 }
 EXPORT_SYMBOL(procfs_list_install);
 
diff --git a/module/os/linux/zfs/zfs_debug.c b/module/os/linux/zfs/zfs_debug.c
index d98463f1b..8d7f04097 100644
--- a/module/os/linux/zfs/zfs_debug.c
+++ b/module/os/linux/zfs/zfs_debug.c
@@ -94,6 +94,7 @@ void
 zfs_dbgmsg_init(void)
 {
 	procfs_list_install("zfs",
+	    NULL,
 	    "dbgmsg",
 	    0600,
 	    &zfs_dbgmsgs,
diff --git a/module/zfs/spa_misc.c b/module/zfs/spa_misc.c
index 41f0ddbde..042104728 100644
--- a/module/zfs/spa_misc.c
+++ b/module/zfs/spa_misc.c
@@ -2169,6 +2169,7 @@ spa_import_progress_init(void)
 	    spa_import_progress_list;
 
 	procfs_list_install("zfs",
+	    NULL,
 	    "import_progress",
 	    0644,
 	    &spa_import_progress_list->procfs_list,
diff --git a/module/zfs/spa_stats.c b/module/zfs/spa_stats.c
index cf0be3c45..c3eacc142 100644
--- a/module/zfs/spa_stats.c
+++ b/module/zfs/spa_stats.c
@@ -122,14 +122,11 @@ static void
 spa_read_history_init(spa_t *spa)
 {
 	spa_history_list_t *shl = &spa->spa_stats.read_history;
-	char *module;
 
 	shl->size = 0;
-
-	module = kmem_asprintf("zfs/%s", spa_name(spa));
-
 	shl->procfs_list.pl_private = shl;
-	procfs_list_install(module,
+	procfs_list_install("zfs",
+	    spa_name(spa),
 	    "reads",
 	    0600,
 	    &shl->procfs_list,
@@ -137,8 +134,6 @@ spa_read_history_init(spa_t *spa)
 	    spa_read_history_show_header,
 	    spa_read_history_clear,
 	    offsetof(spa_read_history_t, srh_node));
-
-	kmem_strfree(module);
 }
 
 static void
@@ -293,14 +288,11 @@ static void
 spa_txg_history_init(spa_t *spa)
 {
 	spa_history_list_t *shl = &spa->spa_stats.txg_history;
-	char *module;
 
 	shl->size = 0;
-
-	module = kmem_asprintf("zfs/%s", spa_name(spa));
-
 	shl->procfs_list.pl_private = shl;
-	procfs_list_install(module,
+	procfs_list_install("zfs",
+	    spa_name(spa),
 	    "txgs",
 	    0644,
 	    &shl->procfs_list,
@@ -308,8 +300,6 @@ spa_txg_history_init(spa_t *spa)
 	    spa_txg_history_show_header,
 	    spa_txg_history_clear,
 	    offsetof(spa_txg_history_t, sth_node));
-
-	kmem_strfree(module);
 }
 
 static void
@@ -699,14 +689,12 @@ static void
 spa_mmp_history_init(spa_t *spa)
 {
 	spa_history_list_t *shl = &spa->spa_stats.mmp_history;
-	char *module;
 
 	shl->size = 0;
 
-	module = kmem_asprintf("zfs/%s", spa_name(spa));
-
 	shl->procfs_list.pl_private = shl;
-	procfs_list_install(module,
+	procfs_list_install("zfs",
+	    spa_name(spa),
 	    "multihost",
 	    0644,
 	    &shl->procfs_list,
@@ -714,8 +702,6 @@ spa_mmp_history_init(spa_t *spa)
 	    spa_mmp_history_show_header,
 	    spa_mmp_history_clear,
 	    offsetof(spa_mmp_history_t, smh_node));
-
-	kmem_strfree(module);
 }
 
 static void