xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/vdev_impl.h (revision d7c57852794a9ac50d4b6a9c4161cb10919658aa)
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 struct abd;
56 
57 extern int zfs_vdev_queue_depth_pct;
58 extern uint32_t zfs_vdev_async_write_max_active;
59 
60 /*
61  * Virtual device operations
62  */
63 typedef int	vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size,
64     uint64_t *ashift);
65 typedef void	vdev_close_func_t(vdev_t *vd);
66 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
67 typedef void	vdev_io_start_func_t(zio_t *zio);
68 typedef void	vdev_io_done_func_t(zio_t *zio);
69 typedef void	vdev_state_change_func_t(vdev_t *vd, int, int);
70 typedef void	vdev_hold_func_t(vdev_t *vd);
71 typedef void	vdev_rele_func_t(vdev_t *vd);
72 
73 typedef struct vdev_ops {
74 	vdev_open_func_t		*vdev_op_open;
75 	vdev_close_func_t		*vdev_op_close;
76 	vdev_asize_func_t		*vdev_op_asize;
77 	vdev_io_start_func_t		*vdev_op_io_start;
78 	vdev_io_done_func_t		*vdev_op_io_done;
79 	vdev_state_change_func_t	*vdev_op_state_change;
80 	vdev_hold_func_t		*vdev_op_hold;
81 	vdev_rele_func_t		*vdev_op_rele;
82 	char				vdev_op_type[16];
83 	boolean_t			vdev_op_leaf;
84 } vdev_ops_t;
85 
86 /*
87  * Virtual device properties
88  */
89 struct vdev_cache_entry {
90 	struct abd	*ve_abd;
91 	uint64_t	ve_offset;
92 	uint64_t	ve_lastused;
93 	avl_node_t	ve_offset_node;
94 	avl_node_t	ve_lastused_node;
95 	uint32_t	ve_hits;
96 	uint16_t	ve_missed_update;
97 	zio_t		*ve_fill_io;
98 };
99 
100 struct vdev_cache {
101 	avl_tree_t	vc_offset_tree;
102 	avl_tree_t	vc_lastused_tree;
103 	kmutex_t	vc_lock;
104 };
105 
106 typedef struct vdev_queue_class {
107 	uint32_t	vqc_active;
108 
109 	/*
110 	 * Sorted by offset or timestamp, depending on if the queue is
111 	 * LBA-ordered vs FIFO.
112 	 */
113 	avl_tree_t	vqc_queued_tree;
114 } vdev_queue_class_t;
115 
116 struct vdev_queue {
117 	vdev_t		*vq_vdev;
118 	vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE];
119 	avl_tree_t	vq_active_tree;
120 	avl_tree_t	vq_read_offset_tree;
121 	avl_tree_t	vq_write_offset_tree;
122 	uint64_t	vq_last_offset;
123 	hrtime_t	vq_io_complete_ts; /* time last i/o completed */
124 	kmutex_t	vq_lock;
125 };
126 
127 /*
128  * Virtual device descriptor
129  */
130 struct vdev {
131 	/*
132 	 * Common to all vdev types.
133 	 */
134 	uint64_t	vdev_id;	/* child number in vdev parent	*/
135 	uint64_t	vdev_guid;	/* unique ID for this vdev	*/
136 	uint64_t	vdev_guid_sum;	/* self guid + all child guids	*/
137 	uint64_t	vdev_orig_guid;	/* orig. guid prior to remove	*/
138 	uint64_t	vdev_asize;	/* allocatable device capacity	*/
139 	uint64_t	vdev_min_asize;	/* min acceptable asize		*/
140 	uint64_t	vdev_max_asize;	/* max acceptable asize		*/
141 	uint64_t	vdev_ashift;	/* block alignment shift	*/
142 	uint64_t	vdev_state;	/* see VDEV_STATE_* #defines	*/
143 	uint64_t	vdev_prevstate;	/* used when reopening a vdev	*/
144 	vdev_ops_t	*vdev_ops;	/* vdev operations		*/
145 	spa_t		*vdev_spa;	/* spa for this vdev		*/
146 	void		*vdev_tsd;	/* type-specific data		*/
147 	vnode_t		*vdev_name_vp;	/* vnode for pathname		*/
148 	vnode_t		*vdev_devid_vp;	/* vnode for devid		*/
149 	vdev_t		*vdev_top;	/* top-level vdev		*/
150 	vdev_t		*vdev_parent;	/* parent vdev			*/
151 	vdev_t		**vdev_child;	/* array of children		*/
152 	uint64_t	vdev_children;	/* number of children		*/
153 	vdev_stat_t	vdev_stat;	/* virtual device statistics	*/
154 	boolean_t	vdev_expanding;	/* expand the vdev?		*/
155 	boolean_t	vdev_reopening;	/* reopen in progress?		*/
156 	int		vdev_open_error; /* error on last open		*/
157 	kthread_t	*vdev_open_thread; /* thread opening children	*/
158 	uint64_t	vdev_crtxg;	/* txg when top-level was added */
159 
160 	/*
161 	 * Top-level vdev state.
162 	 */
163 	uint64_t	vdev_ms_array;	/* metaslab array object	*/
164 	uint64_t	vdev_ms_shift;	/* metaslab size shift		*/
165 	uint64_t	vdev_ms_count;	/* number of metaslabs		*/
166 	metaslab_group_t *vdev_mg;	/* metaslab group		*/
167 	metaslab_t	**vdev_ms;	/* metaslab array		*/
168 	txg_list_t	vdev_ms_list;	/* per-txg dirty metaslab lists	*/
169 	txg_list_t	vdev_dtl_list;	/* per-txg dirty DTL lists	*/
170 	txg_node_t	vdev_txg_node;	/* per-txg dirty vdev linkage	*/
171 	boolean_t	vdev_remove_wanted; /* async remove wanted?	*/
172 	boolean_t	vdev_probe_wanted; /* async probe wanted?	*/
173 	list_node_t	vdev_config_dirty_node; /* config dirty list	*/
174 	list_node_t	vdev_state_dirty_node; /* state dirty list	*/
175 	uint64_t	vdev_deflate_ratio; /* deflation ratio (x512)	*/
176 	uint64_t	vdev_islog;	/* is an intent log device	*/
177 	uint64_t	vdev_removing;	/* device is being removed?	*/
178 	boolean_t	vdev_ishole;	/* is a hole in the namespace	*/
179 	kmutex_t	vdev_queue_lock; /* protects vdev_queue_depth	*/
180 	uint64_t	vdev_top_zap;
181 
182 	/*
183 	 * The queue depth parameters determine how many async writes are
184 	 * still pending (i.e. allocated by net yet issued to disk) per
185 	 * top-level (vdev_async_write_queue_depth) and the maximum allowed
186 	 * (vdev_max_async_write_queue_depth). These values only apply to
187 	 * top-level vdevs.
188 	 */
189 	uint64_t	vdev_async_write_queue_depth;
190 	uint64_t	vdev_max_async_write_queue_depth;
191 
192 	/*
193 	 * Leaf vdev state.
194 	 */
195 	range_tree_t	*vdev_dtl[DTL_TYPES]; /* dirty time logs	*/
196 	space_map_t	*vdev_dtl_sm;	/* dirty time log space map	*/
197 	txg_node_t	vdev_dtl_node;	/* per-txg dirty DTL linkage	*/
198 	uint64_t	vdev_dtl_object; /* DTL object			*/
199 	uint64_t	vdev_psize;	/* physical device capacity	*/
200 	uint64_t	vdev_wholedisk;	/* true if this is a whole disk */
201 	uint64_t	vdev_offline;	/* persistent offline state	*/
202 	uint64_t	vdev_faulted;	/* persistent faulted state	*/
203 	uint64_t	vdev_degraded;	/* persistent degraded state	*/
204 	uint64_t	vdev_removed;	/* persistent removed state	*/
205 	uint64_t	vdev_resilver_txg; /* persistent resilvering state */
206 	uint64_t	vdev_nparity;	/* number of parity devices for raidz */
207 	char		*vdev_path;	/* vdev path (if any)		*/
208 	char		*vdev_devid;	/* vdev devid (if any)		*/
209 	char		*vdev_physpath;	/* vdev device path (if any)	*/
210 	char		*vdev_fru;	/* physical FRU location	*/
211 	uint64_t	vdev_not_present; /* not present during import	*/
212 	uint64_t	vdev_unspare;	/* unspare when resilvering done */
213 	boolean_t	vdev_nowritecache; /* true if flushwritecache failed */
214 	boolean_t	vdev_checkremove; /* temporary online test	*/
215 	boolean_t	vdev_forcefault; /* force online fault		*/
216 	boolean_t	vdev_splitting;	/* split or repair in progress  */
217 	boolean_t	vdev_delayed_close; /* delayed device close?	*/
218 	boolean_t	vdev_tmpoffline; /* device taken offline temporarily? */
219 	boolean_t	vdev_detached;	/* device detached?		*/
220 	boolean_t	vdev_cant_read;	/* vdev is failing all reads	*/
221 	boolean_t	vdev_cant_write; /* vdev is failing all writes	*/
222 	boolean_t	vdev_isspare;	/* was a hot spare		*/
223 	boolean_t	vdev_isl2cache;	/* was a l2cache device		*/
224 	vdev_queue_t	vdev_queue;	/* I/O deadline schedule queue	*/
225 	vdev_cache_t	vdev_cache;	/* physical block cache		*/
226 	spa_aux_vdev_t	*vdev_aux;	/* for l2cache and spares vdevs	*/
227 	zio_t		*vdev_probe_zio; /* root of current probe	*/
228 	vdev_aux_t	vdev_label_aux;	/* on-disk aux state		*/
229 	uint64_t	vdev_leaf_zap;
230 
231 	/*
232 	 * For DTrace to work in userland (libzpool) context, these fields must
233 	 * remain at the end of the structure.  DTrace will use the kernel's
234 	 * CTF definition for 'struct vdev', and since the size of a kmutex_t is
235 	 * larger in userland, the offsets for the rest of the fields would be
236 	 * incorrect.
237 	 */
238 	kmutex_t	vdev_dtl_lock;	/* vdev_dtl_{map,resilver}	*/
239 	kmutex_t	vdev_stat_lock;	/* vdev_stat			*/
240 	kmutex_t	vdev_probe_lock; /* protects vdev_probe_zio	*/
241 };
242 
243 #define	VDEV_RAIDZ_MAXPARITY	3
244 
245 #define	VDEV_PAD_SIZE		(8 << 10)
246 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
247 #define	VDEV_SKIP_SIZE		VDEV_PAD_SIZE * 2
248 #define	VDEV_PHYS_SIZE		(112 << 10)
249 #define	VDEV_UBERBLOCK_RING	(128 << 10)
250 
251 /* The largest uberblock we support is 8k. */
252 #define	MAX_UBERBLOCK_SHIFT (13)
253 #define	VDEV_UBERBLOCK_SHIFT(vd)	\
254 	MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \
255 	    MAX_UBERBLOCK_SHIFT)
256 #define	VDEV_UBERBLOCK_COUNT(vd)	\
257 	(VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
258 #define	VDEV_UBERBLOCK_OFFSET(vd, n)	\
259 	offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
260 #define	VDEV_UBERBLOCK_SIZE(vd)		(1ULL << VDEV_UBERBLOCK_SHIFT(vd))
261 
262 typedef struct vdev_phys {
263 	char		vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
264 	zio_eck_t	vp_zbt;
265 } vdev_phys_t;
266 
267 typedef struct vdev_label {
268 	char		vl_pad1[VDEV_PAD_SIZE];			/*  8K */
269 	char		vl_pad2[VDEV_PAD_SIZE];			/*  8K */
270 	vdev_phys_t	vl_vdev_phys;				/* 112K	*/
271 	char		vl_uberblock[VDEV_UBERBLOCK_RING];	/* 128K	*/
272 } vdev_label_t;							/* 256K total */
273 
274 /*
275  * vdev_dirty() flags
276  */
277 #define	VDD_METASLAB	0x01
278 #define	VDD_DTL		0x02
279 
280 /* Offset of embedded boot loader region on each label */
281 #define	VDEV_BOOT_OFFSET	(2 * sizeof (vdev_label_t))
282 /*
283  * Size of embedded boot loader region on each label.
284  * The total size of the first two labels plus the boot area is 4MB.
285  */
286 #define	VDEV_BOOT_SIZE		(7ULL << 19)			/* 3.5M */
287 
288 /*
289  * Size of label regions at the start and end of each leaf device.
290  */
291 #define	VDEV_LABEL_START_SIZE	(2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
292 #define	VDEV_LABEL_END_SIZE	(2 * sizeof (vdev_label_t))
293 #define	VDEV_LABELS		4
294 #define	VDEV_BEST_LABEL		VDEV_LABELS
295 
296 #define	VDEV_ALLOC_LOAD		0
297 #define	VDEV_ALLOC_ADD		1
298 #define	VDEV_ALLOC_SPARE	2
299 #define	VDEV_ALLOC_L2CACHE	3
300 #define	VDEV_ALLOC_ROOTPOOL	4
301 #define	VDEV_ALLOC_SPLIT	5
302 #define	VDEV_ALLOC_ATTACH	6
303 
304 /*
305  * Allocate or free a vdev
306  */
307 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
308     vdev_ops_t *ops);
309 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
310     vdev_t *parent, uint_t id, int alloctype);
311 extern void vdev_free(vdev_t *vd);
312 
313 /*
314  * Add or remove children and parents
315  */
316 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
317 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
318 extern void vdev_compact_children(vdev_t *pvd);
319 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
320 extern void vdev_remove_parent(vdev_t *cvd);
321 
322 /*
323  * vdev sync load and sync
324  */
325 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd);
326 extern boolean_t vdev_log_state_valid(vdev_t *vd);
327 extern void vdev_load(vdev_t *vd);
328 extern int vdev_dtl_load(vdev_t *vd);
329 extern void vdev_sync(vdev_t *vd, uint64_t txg);
330 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
331 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
332 extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg);
333 
334 /*
335  * Available vdev types.
336  */
337 extern vdev_ops_t vdev_root_ops;
338 extern vdev_ops_t vdev_mirror_ops;
339 extern vdev_ops_t vdev_replacing_ops;
340 extern vdev_ops_t vdev_raidz_ops;
341 extern vdev_ops_t vdev_disk_ops;
342 extern vdev_ops_t vdev_file_ops;
343 extern vdev_ops_t vdev_missing_ops;
344 extern vdev_ops_t vdev_hole_ops;
345 extern vdev_ops_t vdev_spare_ops;
346 
347 /*
348  * Common size functions
349  */
350 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
351 extern uint64_t vdev_get_min_asize(vdev_t *vd);
352 extern void vdev_set_min_asize(vdev_t *vd);
353 
354 /*
355  * Global variables
356  */
357 /* zdb uses this tunable, so it must be declared here to make lint happy. */
358 extern int zfs_vdev_cache_size;
359 
360 /*
361  * The vdev_buf_t is used to translate between zio_t and buf_t, and back again.
362  */
363 typedef struct vdev_buf {
364 	buf_t	vb_buf;		/* buffer that describes the io */
365 	zio_t	*vb_io;		/* pointer back to the original zio_t */
366 } vdev_buf_t;
367 
368 #ifdef	__cplusplus
369 }
370 #endif
371 
372 #endif	/* _SYS_VDEV_IMPL_H */
373