xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/vdev_impl.h (revision 095bcd6622e3b3520eb3b71039a3be5cfab25b74)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #ifndef _SYS_VDEV_IMPL_H
27 #define	_SYS_VDEV_IMPL_H
28 
29 #include <sys/avl.h>
30 #include <sys/dmu.h>
31 #include <sys/metaslab.h>
32 #include <sys/nvpair.h>
33 #include <sys/space_map.h>
34 #include <sys/vdev.h>
35 #include <sys/dkio.h>
36 #include <sys/uberblock_impl.h>
37 
38 #ifdef	__cplusplus
39 extern "C" {
40 #endif
41 
42 /*
43  * Virtual device descriptors.
44  *
45  * All storage pool operations go through the virtual device framework,
46  * which provides data replication and I/O scheduling.
47  */
48 
49 /*
50  * Forward declarations that lots of things need.
51  */
52 typedef struct vdev_queue vdev_queue_t;
53 typedef struct vdev_cache vdev_cache_t;
54 typedef struct vdev_cache_entry vdev_cache_entry_t;
55 
56 /*
57  * Virtual device operations
58  */
59 typedef int	vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *ashift);
60 typedef void	vdev_close_func_t(vdev_t *vd);
61 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
62 typedef int	vdev_io_start_func_t(zio_t *zio);
63 typedef void	vdev_io_done_func_t(zio_t *zio);
64 typedef void	vdev_state_change_func_t(vdev_t *vd, int, int);
65 
66 typedef struct vdev_ops {
67 	vdev_open_func_t		*vdev_op_open;
68 	vdev_close_func_t		*vdev_op_close;
69 	vdev_asize_func_t		*vdev_op_asize;
70 	vdev_io_start_func_t		*vdev_op_io_start;
71 	vdev_io_done_func_t		*vdev_op_io_done;
72 	vdev_state_change_func_t	*vdev_op_state_change;
73 	char				vdev_op_type[16];
74 	boolean_t			vdev_op_leaf;
75 } vdev_ops_t;
76 
77 /*
78  * Virtual device properties
79  */
80 struct vdev_cache_entry {
81 	char		*ve_data;
82 	uint64_t	ve_offset;
83 	uint64_t	ve_lastused;
84 	avl_node_t	ve_offset_node;
85 	avl_node_t	ve_lastused_node;
86 	uint32_t	ve_hits;
87 	uint16_t	ve_missed_update;
88 	zio_t		*ve_fill_io;
89 };
90 
91 struct vdev_cache {
92 	avl_tree_t	vc_offset_tree;
93 	avl_tree_t	vc_lastused_tree;
94 	kmutex_t	vc_lock;
95 };
96 
97 struct vdev_queue {
98 	avl_tree_t	vq_deadline_tree;
99 	avl_tree_t	vq_read_tree;
100 	avl_tree_t	vq_write_tree;
101 	avl_tree_t	vq_pending_tree;
102 	kmutex_t	vq_lock;
103 };
104 
105 /*
106  * Virtual device descriptor
107  */
108 struct vdev {
109 	/*
110 	 * Common to all vdev types.
111 	 */
112 	uint64_t	vdev_id;	/* child number in vdev parent	*/
113 	uint64_t	vdev_guid;	/* unique ID for this vdev	*/
114 	uint64_t	vdev_guid_sum;	/* self guid + all child guids	*/
115 	uint64_t	vdev_asize;	/* allocatable device capacity	*/
116 	uint64_t	vdev_min_asize;	/* min acceptable asize		*/
117 	uint64_t	vdev_ashift;	/* block alignment shift	*/
118 	uint64_t	vdev_state;	/* see VDEV_STATE_* #defines	*/
119 	uint64_t	vdev_prevstate;	/* used when reopening a vdev	*/
120 	vdev_ops_t	*vdev_ops;	/* vdev operations		*/
121 	spa_t		*vdev_spa;	/* spa for this vdev		*/
122 	void		*vdev_tsd;	/* type-specific data		*/
123 	vdev_t		*vdev_top;	/* top-level vdev		*/
124 	vdev_t		*vdev_parent;	/* parent vdev			*/
125 	vdev_t		**vdev_child;	/* array of children		*/
126 	uint64_t	vdev_children;	/* number of children		*/
127 	space_map_t	vdev_dtl[DTL_TYPES]; /* in-core dirty time logs	*/
128 	vdev_stat_t	vdev_stat;	/* virtual device statistics	*/
129 	boolean_t	vdev_expanding;	/* expand the vdev?		*/
130 	boolean_t	vdev_reopening;	/* reopen in progress?		*/
131 	int		vdev_open_error; /* error on last open		*/
132 	kthread_t	*vdev_open_thread; /* thread opening children	*/
133 	uint64_t	vdev_crtxg;	/* txg when top-level was added */
134 
135 	/*
136 	 * Top-level vdev state.
137 	 */
138 	uint64_t	vdev_ms_array;	/* metaslab array object	*/
139 	uint64_t	vdev_ms_shift;	/* metaslab size shift		*/
140 	uint64_t	vdev_ms_count;	/* number of metaslabs		*/
141 	metaslab_group_t *vdev_mg;	/* metaslab group		*/
142 	metaslab_t	**vdev_ms;	/* metaslab array		*/
143 	txg_list_t	vdev_ms_list;	/* per-txg dirty metaslab lists	*/
144 	txg_list_t	vdev_dtl_list;	/* per-txg dirty DTL lists	*/
145 	txg_node_t	vdev_txg_node;	/* per-txg dirty vdev linkage	*/
146 	boolean_t	vdev_remove_wanted; /* async remove wanted?	*/
147 	boolean_t	vdev_probe_wanted; /* async probe wanted?	*/
148 	boolean_t	vdev_removing;	/* device is being removed?	*/
149 	list_node_t	vdev_config_dirty_node; /* config dirty list	*/
150 	list_node_t	vdev_state_dirty_node; /* state dirty list	*/
151 	uint64_t	vdev_deflate_ratio; /* deflation ratio (x512)	*/
152 	uint64_t	vdev_islog;	/* is an intent log device	*/
153 	uint64_t	vdev_ishole;	/* is a hole in the namespace 	*/
154 
155 	/*
156 	 * Leaf vdev state.
157 	 */
158 	uint64_t	vdev_psize;	/* physical device capacity	*/
159 	space_map_obj_t	vdev_dtl_smo;	/* dirty time log space map obj	*/
160 	txg_node_t	vdev_dtl_node;	/* per-txg dirty DTL linkage	*/
161 	uint64_t	vdev_wholedisk;	/* true if this is a whole disk */
162 	uint64_t	vdev_offline;	/* persistent offline state	*/
163 	uint64_t	vdev_faulted;	/* persistent faulted state	*/
164 	uint64_t	vdev_degraded;	/* persistent degraded state	*/
165 	uint64_t	vdev_removed;	/* persistent removed state	*/
166 	uint64_t	vdev_nparity;	/* number of parity devices for raidz */
167 	char		*vdev_path;	/* vdev path (if any)		*/
168 	char		*vdev_devid;	/* vdev devid (if any)		*/
169 	char		*vdev_physpath;	/* vdev device path (if any)	*/
170 	char		*vdev_fru;	/* physical FRU location	*/
171 	uint64_t	vdev_not_present; /* not present during import	*/
172 	uint64_t	vdev_unspare;	/* unspare when resilvering done */
173 	hrtime_t	vdev_last_try;	/* last reopen time		*/
174 	boolean_t	vdev_nowritecache; /* true if flushwritecache failed */
175 	boolean_t	vdev_checkremove; /* temporary online test	*/
176 	boolean_t	vdev_forcefault; /* force online fault		*/
177 	uint8_t		vdev_tmpoffline; /* device taken offline temporarily? */
178 	uint8_t		vdev_detached;	/* device detached?		*/
179 	uint8_t		vdev_cant_read;	/* vdev is failing all reads	*/
180 	uint8_t		vdev_cant_write; /* vdev is failing all writes	*/
181 	uint64_t	vdev_isspare;	/* was a hot spare		*/
182 	uint64_t	vdev_isl2cache;	/* was a l2cache device		*/
183 	vdev_queue_t	vdev_queue;	/* I/O deadline schedule queue	*/
184 	vdev_cache_t	vdev_cache;	/* physical block cache		*/
185 	spa_aux_vdev_t	*vdev_aux;	/* for l2cache vdevs		*/
186 	zio_t		*vdev_probe_zio; /* root of current probe	*/
187 	vdev_aux_t	vdev_label_aux;	/* on-disk aux state		*/
188 
189 	/*
190 	 * For DTrace to work in userland (libzpool) context, these fields must
191 	 * remain at the end of the structure.  DTrace will use the kernel's
192 	 * CTF definition for 'struct vdev', and since the size of a kmutex_t is
193 	 * larger in userland, the offsets for the rest fields would be
194 	 * incorrect.
195 	 */
196 	kmutex_t	vdev_dtl_lock;	/* vdev_dtl_{map,resilver}	*/
197 	kmutex_t	vdev_stat_lock;	/* vdev_stat			*/
198 	kmutex_t	vdev_probe_lock; /* protects vdev_probe_zio	*/
199 };
200 
201 #define	VDEV_PAD_SIZE		(8 << 10)
202 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
203 #define	VDEV_SKIP_SIZE		VDEV_PAD_SIZE * 2
204 #define	VDEV_PHYS_SIZE		(112 << 10)
205 #define	VDEV_UBERBLOCK_RING	(128 << 10)
206 
207 #define	VDEV_UBERBLOCK_SHIFT(vd)	\
208 	MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT)
209 #define	VDEV_UBERBLOCK_COUNT(vd)	\
210 	(VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
211 #define	VDEV_UBERBLOCK_OFFSET(vd, n)	\
212 	offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
213 #define	VDEV_UBERBLOCK_SIZE(vd)		(1ULL << VDEV_UBERBLOCK_SHIFT(vd))
214 
215 typedef struct vdev_phys {
216 	char		vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_block_tail_t)];
217 	zio_block_tail_t vp_zbt;
218 } vdev_phys_t;
219 
220 typedef struct vdev_label {
221 	char		vl_pad1[VDEV_PAD_SIZE];			/*  8K */
222 	char		vl_pad2[VDEV_PAD_SIZE];			/*  8K */
223 	vdev_phys_t	vl_vdev_phys;				/* 112K	*/
224 	char		vl_uberblock[VDEV_UBERBLOCK_RING];	/* 128K	*/
225 } vdev_label_t;							/* 256K total */
226 
227 /*
228  * vdev_dirty() flags
229  */
230 #define	VDD_METASLAB	0x01
231 #define	VDD_DTL		0x02
232 
233 /*
234  * Size and offset of embedded boot loader region on each label.
235  * The total size of the first two labels plus the boot area is 4MB.
236  */
237 #define	VDEV_BOOT_OFFSET	(2 * sizeof (vdev_label_t))
238 #define	VDEV_BOOT_SIZE		(7ULL << 19)			/* 3.5M	*/
239 
240 /*
241  * Size of label regions at the start and end of each leaf device.
242  */
243 #define	VDEV_LABEL_START_SIZE	(2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
244 #define	VDEV_LABEL_END_SIZE	(2 * sizeof (vdev_label_t))
245 #define	VDEV_LABELS		4
246 
247 #define	VDEV_ALLOC_LOAD		0
248 #define	VDEV_ALLOC_ADD		1
249 #define	VDEV_ALLOC_SPARE	2
250 #define	VDEV_ALLOC_L2CACHE	3
251 #define	VDEV_ALLOC_ROOTPOOL	4
252 
253 /*
254  * Allocate or free a vdev
255  */
256 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
257     vdev_ops_t *ops);
258 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
259     vdev_t *parent, uint_t id, int alloctype);
260 extern void vdev_free(vdev_t *vd);
261 
262 /*
263  * Add or remove children and parents
264  */
265 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
266 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
267 extern void vdev_compact_children(vdev_t *pvd);
268 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
269 extern void vdev_remove_parent(vdev_t *cvd);
270 
271 /*
272  * vdev sync load and sync
273  */
274 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd);
275 extern void vdev_load(vdev_t *vd);
276 extern void vdev_sync(vdev_t *vd, uint64_t txg);
277 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
278 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
279 
280 /*
281  * Available vdev types.
282  */
283 extern vdev_ops_t vdev_root_ops;
284 extern vdev_ops_t vdev_mirror_ops;
285 extern vdev_ops_t vdev_replacing_ops;
286 extern vdev_ops_t vdev_raidz_ops;
287 extern vdev_ops_t vdev_disk_ops;
288 extern vdev_ops_t vdev_file_ops;
289 extern vdev_ops_t vdev_missing_ops;
290 extern vdev_ops_t vdev_hole_ops;
291 extern vdev_ops_t vdev_spare_ops;
292 
293 /*
294  * Common size functions
295  */
296 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
297 extern uint64_t vdev_get_min_asize(vdev_t *vd);
298 extern void vdev_set_min_asize(vdev_t *vd);
299 
300 /*
301  * zdb uses this tunable, so it must be declared here to make lint happy.
302  */
303 extern int zfs_vdev_cache_size;
304 
305 #ifdef	__cplusplus
306 }
307 #endif
308 
309 #endif	/* _SYS_VDEV_IMPL_H */
310