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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
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 *max_size,
60     uint64_t *ashift);
61 typedef void	vdev_close_func_t(vdev_t *vd);
62 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
63 typedef void	vdev_io_start_func_t(zio_t *zio);
64 typedef void	vdev_io_done_func_t(zio_t *zio);
65 typedef void	vdev_state_change_func_t(vdev_t *vd, int, int);
66 typedef void	vdev_hold_func_t(vdev_t *vd);
67 typedef void	vdev_rele_func_t(vdev_t *vd);
68 
69 typedef struct vdev_ops {
70 	vdev_open_func_t		*vdev_op_open;
71 	vdev_close_func_t		*vdev_op_close;
72 	vdev_asize_func_t		*vdev_op_asize;
73 	vdev_io_start_func_t		*vdev_op_io_start;
74 	vdev_io_done_func_t		*vdev_op_io_done;
75 	vdev_state_change_func_t	*vdev_op_state_change;
76 	vdev_hold_func_t		*vdev_op_hold;
77 	vdev_rele_func_t		*vdev_op_rele;
78 	char				vdev_op_type[16];
79 	boolean_t			vdev_op_leaf;
80 } vdev_ops_t;
81 
82 /*
83  * Virtual device properties
84  */
85 struct vdev_cache_entry {
86 	char		*ve_data;
87 	uint64_t	ve_offset;
88 	uint64_t	ve_lastused;
89 	avl_node_t	ve_offset_node;
90 	avl_node_t	ve_lastused_node;
91 	uint32_t	ve_hits;
92 	uint16_t	ve_missed_update;
93 	zio_t		*ve_fill_io;
94 };
95 
96 struct vdev_cache {
97 	avl_tree_t	vc_offset_tree;
98 	avl_tree_t	vc_lastused_tree;
99 	kmutex_t	vc_lock;
100 };
101 
102 typedef struct vdev_queue_class {
103 	uint32_t	vqc_active;
104 
105 	/*
106 	 * Sorted by offset or timestamp, depending on if the queue is
107 	 * LBA-ordered vs FIFO.
108 	 */
109 	avl_tree_t	vqc_queued_tree;
110 } vdev_queue_class_t;
111 
112 struct vdev_queue {
113 	vdev_t		*vq_vdev;
114 	vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE];
115 	avl_tree_t	vq_active_tree;
116 	avl_tree_t	vq_read_offset_tree;
117 	avl_tree_t	vq_write_offset_tree;
118 	uint64_t	vq_last_offset;
119 	hrtime_t	vq_io_complete_ts; /* time last i/o completed */
120 	kmutex_t	vq_lock;
121 };
122 
123 /*
124  * Virtual device descriptor
125  */
126 struct vdev {
127 	/*
128 	 * Common to all vdev types.
129 	 */
130 	uint64_t	vdev_id;	/* child number in vdev parent	*/
131 	uint64_t	vdev_guid;	/* unique ID for this vdev	*/
132 	uint64_t	vdev_guid_sum;	/* self guid + all child guids	*/
133 	uint64_t	vdev_orig_guid;	/* orig. guid prior to remove	*/
134 	uint64_t	vdev_asize;	/* allocatable device capacity	*/
135 	uint64_t	vdev_min_asize;	/* min acceptable asize		*/
136 	uint64_t	vdev_max_asize;	/* max acceptable asize		*/
137 	uint64_t	vdev_ashift;	/* block alignment shift	*/
138 	uint64_t	vdev_state;	/* see VDEV_STATE_* #defines	*/
139 	uint64_t	vdev_prevstate;	/* used when reopening a vdev	*/
140 	vdev_ops_t	*vdev_ops;	/* vdev operations		*/
141 	spa_t		*vdev_spa;	/* spa for this vdev		*/
142 	void		*vdev_tsd;	/* type-specific data		*/
143 	vnode_t		*vdev_name_vp;	/* vnode for pathname		*/
144 	vnode_t		*vdev_devid_vp;	/* vnode for devid		*/
145 	vdev_t		*vdev_top;	/* top-level vdev		*/
146 	vdev_t		*vdev_parent;	/* parent vdev			*/
147 	vdev_t		**vdev_child;	/* array of children		*/
148 	uint64_t	vdev_children;	/* number of children		*/
149 	vdev_stat_t	vdev_stat;	/* virtual device statistics	*/
150 	boolean_t	vdev_expanding;	/* expand the vdev?		*/
151 	boolean_t	vdev_reopening;	/* reopen in progress?		*/
152 	int		vdev_open_error; /* error on last open		*/
153 	kthread_t	*vdev_open_thread; /* thread opening children	*/
154 	uint64_t	vdev_crtxg;	/* txg when top-level was added */
155 
156 	/*
157 	 * Top-level vdev state.
158 	 */
159 	uint64_t	vdev_ms_array;	/* metaslab array object	*/
160 	uint64_t	vdev_ms_shift;	/* metaslab size shift		*/
161 	uint64_t	vdev_ms_count;	/* number of metaslabs		*/
162 	metaslab_group_t *vdev_mg;	/* metaslab group		*/
163 	metaslab_t	**vdev_ms;	/* metaslab array		*/
164 	txg_list_t	vdev_ms_list;	/* per-txg dirty metaslab lists	*/
165 	txg_list_t	vdev_dtl_list;	/* per-txg dirty DTL lists	*/
166 	txg_node_t	vdev_txg_node;	/* per-txg dirty vdev linkage	*/
167 	boolean_t	vdev_remove_wanted; /* async remove wanted?	*/
168 	boolean_t	vdev_probe_wanted; /* async probe wanted?	*/
169 	list_node_t	vdev_config_dirty_node; /* config dirty list	*/
170 	list_node_t	vdev_state_dirty_node; /* state dirty list	*/
171 	uint64_t	vdev_deflate_ratio; /* deflation ratio (x512)	*/
172 	uint64_t	vdev_islog;	/* is an intent log device	*/
173 	uint64_t	vdev_removing;	/* device is being removed?	*/
174 	boolean_t	vdev_ishole;	/* is a hole in the namespace 	*/
175 
176 	/*
177 	 * Leaf vdev state.
178 	 */
179 	range_tree_t	*vdev_dtl[DTL_TYPES]; /* dirty time logs	*/
180 	space_map_t	*vdev_dtl_sm;	/* dirty time log space map	*/
181 	txg_node_t	vdev_dtl_node;	/* per-txg dirty DTL linkage	*/
182 	uint64_t	vdev_dtl_object; /* DTL object			*/
183 	uint64_t	vdev_psize;	/* physical device capacity	*/
184 	uint64_t	vdev_wholedisk;	/* true if this is a whole disk */
185 	uint64_t	vdev_offline;	/* persistent offline state	*/
186 	uint64_t	vdev_faulted;	/* persistent faulted state	*/
187 	uint64_t	vdev_degraded;	/* persistent degraded state	*/
188 	uint64_t	vdev_removed;	/* persistent removed state	*/
189 	uint64_t	vdev_resilver_txg; /* persistent resilvering state */
190 	uint64_t	vdev_nparity;	/* number of parity devices for raidz */
191 	char		*vdev_path;	/* vdev path (if any)		*/
192 	char		*vdev_devid;	/* vdev devid (if any)		*/
193 	char		*vdev_physpath;	/* vdev device path (if any)	*/
194 	char		*vdev_fru;	/* physical FRU location	*/
195 	uint64_t	vdev_not_present; /* not present during import	*/
196 	uint64_t	vdev_unspare;	/* unspare when resilvering done */
197 	boolean_t	vdev_nowritecache; /* true if flushwritecache failed */
198 	boolean_t	vdev_checkremove; /* temporary online test	*/
199 	boolean_t	vdev_forcefault; /* force online fault		*/
200 	boolean_t	vdev_splitting;	/* split or repair in progress  */
201 	boolean_t	vdev_delayed_close; /* delayed device close?	*/
202 	boolean_t	vdev_tmpoffline; /* device taken offline temporarily? */
203 	boolean_t	vdev_detached;	/* device detached?		*/
204 	boolean_t	vdev_cant_read;	/* vdev is failing all reads	*/
205 	boolean_t	vdev_cant_write; /* vdev is failing all writes	*/
206 	boolean_t	vdev_isspare;	/* was a hot spare		*/
207 	boolean_t	vdev_isl2cache;	/* was a l2cache device		*/
208 	vdev_queue_t	vdev_queue;	/* I/O deadline schedule queue	*/
209 	vdev_cache_t	vdev_cache;	/* physical block cache		*/
210 	spa_aux_vdev_t	*vdev_aux;	/* for l2cache and spares vdevs	*/
211 	zio_t		*vdev_probe_zio; /* root of current probe	*/
212 	vdev_aux_t	vdev_label_aux;	/* on-disk aux state		*/
213 
214 	/*
215 	 * For DTrace to work in userland (libzpool) context, these fields must
216 	 * remain at the end of the structure.  DTrace will use the kernel's
217 	 * CTF definition for 'struct vdev', and since the size of a kmutex_t is
218 	 * larger in userland, the offsets for the rest of the fields would be
219 	 * incorrect.
220 	 */
221 	kmutex_t	vdev_dtl_lock;	/* vdev_dtl_{map,resilver}	*/
222 	kmutex_t	vdev_stat_lock;	/* vdev_stat			*/
223 	kmutex_t	vdev_probe_lock; /* protects vdev_probe_zio	*/
224 };
225 
226 #define	VDEV_RAIDZ_MAXPARITY	3
227 
228 #define	VDEV_PAD_SIZE		(8 << 10)
229 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
230 #define	VDEV_SKIP_SIZE		VDEV_PAD_SIZE * 2
231 #define	VDEV_PHYS_SIZE		(112 << 10)
232 #define	VDEV_UBERBLOCK_RING	(128 << 10)
233 
234 /* The largest uberblock we support is 8k. */
235 #define	MAX_UBERBLOCK_SHIFT (13)
236 #define	VDEV_UBERBLOCK_SHIFT(vd)	\
237 	MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \
238 	    MAX_UBERBLOCK_SHIFT)
239 #define	VDEV_UBERBLOCK_COUNT(vd)	\
240 	(VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
241 #define	VDEV_UBERBLOCK_OFFSET(vd, n)	\
242 	offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
243 #define	VDEV_UBERBLOCK_SIZE(vd)		(1ULL << VDEV_UBERBLOCK_SHIFT(vd))
244 
245 typedef struct vdev_phys {
246 	char		vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
247 	zio_eck_t	vp_zbt;
248 } vdev_phys_t;
249 
250 typedef struct vdev_label {
251 	char		vl_pad1[VDEV_PAD_SIZE];			/*  8K */
252 	char		vl_pad2[VDEV_PAD_SIZE];			/*  8K */
253 	vdev_phys_t	vl_vdev_phys;				/* 112K	*/
254 	char		vl_uberblock[VDEV_UBERBLOCK_RING];	/* 128K	*/
255 } vdev_label_t;							/* 256K total */
256 
257 /*
258  * vdev_dirty() flags
259  */
260 #define	VDD_METASLAB	0x01
261 #define	VDD_DTL		0x02
262 
263 /* Offset of embedded boot loader region on each label */
264 #define	VDEV_BOOT_OFFSET	(2 * sizeof (vdev_label_t))
265 /*
266  * Size of embedded boot loader region on each label.
267  * The total size of the first two labels plus the boot area is 4MB.
268  */
269 #define	VDEV_BOOT_SIZE		(7ULL << 19)			/* 3.5M */
270 
271 /*
272  * Size of label regions at the start and end of each leaf device.
273  */
274 #define	VDEV_LABEL_START_SIZE	(2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
275 #define	VDEV_LABEL_END_SIZE	(2 * sizeof (vdev_label_t))
276 #define	VDEV_LABELS		4
277 #define	VDEV_BEST_LABEL		VDEV_LABELS
278 
279 #define	VDEV_ALLOC_LOAD		0
280 #define	VDEV_ALLOC_ADD		1
281 #define	VDEV_ALLOC_SPARE	2
282 #define	VDEV_ALLOC_L2CACHE	3
283 #define	VDEV_ALLOC_ROOTPOOL	4
284 #define	VDEV_ALLOC_SPLIT	5
285 #define	VDEV_ALLOC_ATTACH	6
286 
287 /*
288  * Allocate or free a vdev
289  */
290 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
291     vdev_ops_t *ops);
292 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
293     vdev_t *parent, uint_t id, int alloctype);
294 extern void vdev_free(vdev_t *vd);
295 
296 /*
297  * Add or remove children and parents
298  */
299 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
300 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
301 extern void vdev_compact_children(vdev_t *pvd);
302 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
303 extern void vdev_remove_parent(vdev_t *cvd);
304 
305 /*
306  * vdev sync load and sync
307  */
308 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd);
309 extern boolean_t vdev_log_state_valid(vdev_t *vd);
310 extern void vdev_load(vdev_t *vd);
311 extern int vdev_dtl_load(vdev_t *vd);
312 extern void vdev_sync(vdev_t *vd, uint64_t txg);
313 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
314 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
315 extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg);
316 
317 /*
318  * Available vdev types.
319  */
320 extern vdev_ops_t vdev_root_ops;
321 extern vdev_ops_t vdev_mirror_ops;
322 extern vdev_ops_t vdev_replacing_ops;
323 extern vdev_ops_t vdev_raidz_ops;
324 extern vdev_ops_t vdev_disk_ops;
325 extern vdev_ops_t vdev_file_ops;
326 extern vdev_ops_t vdev_missing_ops;
327 extern vdev_ops_t vdev_hole_ops;
328 extern vdev_ops_t vdev_spare_ops;
329 
330 /*
331  * Common size functions
332  */
333 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
334 extern uint64_t vdev_get_min_asize(vdev_t *vd);
335 extern void vdev_set_min_asize(vdev_t *vd);
336 
337 /*
338  * Global variables
339  */
340 /* zdb uses this tunable, so it must be declared here to make lint happy. */
341 extern int zfs_vdev_cache_size;
342 
343 /*
344  * The vdev_buf_t is used to translate between zio_t and buf_t, and back again.
345  */
346 typedef struct vdev_buf {
347 	buf_t	vb_buf;		/* buffer that describes the io */
348 	zio_t	*vb_io;		/* pointer back to the original zio_t */
349 } vdev_buf_t;
350 
351 #ifdef	__cplusplus
352 }
353 #endif
354 
355 #endif	/* _SYS_VDEV_IMPL_H */
356