xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/vdev_impl.h (revision 48edc7cf07b5dccc3ad84bf2dafe4150bd666d60)
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 extern int zfs_vdev_queue_depth_pct;
57 extern uint32_t zfs_vdev_async_write_max_active;
58 
59 /*
60  * Virtual device operations
61  */
62 typedef int	vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size,
63     uint64_t *ashift);
64 typedef void	vdev_close_func_t(vdev_t *vd);
65 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
66 typedef void	vdev_io_start_func_t(zio_t *zio);
67 typedef void	vdev_io_done_func_t(zio_t *zio);
68 typedef void	vdev_state_change_func_t(vdev_t *vd, int, int);
69 typedef void	vdev_hold_func_t(vdev_t *vd);
70 typedef void	vdev_rele_func_t(vdev_t *vd);
71 
72 typedef struct vdev_ops {
73 	vdev_open_func_t		*vdev_op_open;
74 	vdev_close_func_t		*vdev_op_close;
75 	vdev_asize_func_t		*vdev_op_asize;
76 	vdev_io_start_func_t		*vdev_op_io_start;
77 	vdev_io_done_func_t		*vdev_op_io_done;
78 	vdev_state_change_func_t	*vdev_op_state_change;
79 	vdev_hold_func_t		*vdev_op_hold;
80 	vdev_rele_func_t		*vdev_op_rele;
81 	char				vdev_op_type[16];
82 	boolean_t			vdev_op_leaf;
83 } vdev_ops_t;
84 
85 /*
86  * Virtual device properties
87  */
88 struct vdev_cache_entry {
89 	char		*ve_data;
90 	uint64_t	ve_offset;
91 	uint64_t	ve_lastused;
92 	avl_node_t	ve_offset_node;
93 	avl_node_t	ve_lastused_node;
94 	uint32_t	ve_hits;
95 	uint16_t	ve_missed_update;
96 	zio_t		*ve_fill_io;
97 };
98 
99 struct vdev_cache {
100 	avl_tree_t	vc_offset_tree;
101 	avl_tree_t	vc_lastused_tree;
102 	kmutex_t	vc_lock;
103 };
104 
105 typedef struct vdev_queue_class {
106 	uint32_t	vqc_active;
107 
108 	/*
109 	 * Sorted by offset or timestamp, depending on if the queue is
110 	 * LBA-ordered vs FIFO.
111 	 */
112 	avl_tree_t	vqc_queued_tree;
113 } vdev_queue_class_t;
114 
115 struct vdev_queue {
116 	vdev_t		*vq_vdev;
117 	vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE];
118 	avl_tree_t	vq_active_tree;
119 	avl_tree_t	vq_read_offset_tree;
120 	avl_tree_t	vq_write_offset_tree;
121 	uint64_t	vq_last_offset;
122 	hrtime_t	vq_io_complete_ts; /* time last i/o completed */
123 	kmutex_t	vq_lock;
124 };
125 
126 /*
127  * Virtual device descriptor
128  */
129 struct vdev {
130 	/*
131 	 * Common to all vdev types.
132 	 */
133 	uint64_t	vdev_id;	/* child number in vdev parent	*/
134 	uint64_t	vdev_guid;	/* unique ID for this vdev	*/
135 	uint64_t	vdev_guid_sum;	/* self guid + all child guids	*/
136 	uint64_t	vdev_orig_guid;	/* orig. guid prior to remove	*/
137 	uint64_t	vdev_asize;	/* allocatable device capacity	*/
138 	uint64_t	vdev_min_asize;	/* min acceptable asize		*/
139 	uint64_t	vdev_max_asize;	/* max acceptable asize		*/
140 	uint64_t	vdev_ashift;	/* block alignment shift	*/
141 	uint64_t	vdev_state;	/* see VDEV_STATE_* #defines	*/
142 	uint64_t	vdev_prevstate;	/* used when reopening a vdev	*/
143 	vdev_ops_t	*vdev_ops;	/* vdev operations		*/
144 	spa_t		*vdev_spa;	/* spa for this vdev		*/
145 	void		*vdev_tsd;	/* type-specific data		*/
146 	vnode_t		*vdev_name_vp;	/* vnode for pathname		*/
147 	vnode_t		*vdev_devid_vp;	/* vnode for devid		*/
148 	vdev_t		*vdev_top;	/* top-level vdev		*/
149 	vdev_t		*vdev_parent;	/* parent vdev			*/
150 	vdev_t		**vdev_child;	/* array of children		*/
151 	uint64_t	vdev_children;	/* number of children		*/
152 	vdev_stat_t	vdev_stat;	/* virtual device statistics	*/
153 	boolean_t	vdev_expanding;	/* expand the vdev?		*/
154 	boolean_t	vdev_reopening;	/* reopen in progress?		*/
155 	int		vdev_open_error; /* error on last open		*/
156 	kthread_t	*vdev_open_thread; /* thread opening children	*/
157 	uint64_t	vdev_crtxg;	/* txg when top-level was added */
158 
159 	/*
160 	 * Top-level vdev state.
161 	 */
162 	uint64_t	vdev_ms_array;	/* metaslab array object	*/
163 	uint64_t	vdev_ms_shift;	/* metaslab size shift		*/
164 	uint64_t	vdev_ms_count;	/* number of metaslabs		*/
165 	metaslab_group_t *vdev_mg;	/* metaslab group		*/
166 	metaslab_t	**vdev_ms;	/* metaslab array		*/
167 	txg_list_t	vdev_ms_list;	/* per-txg dirty metaslab lists	*/
168 	txg_list_t	vdev_dtl_list;	/* per-txg dirty DTL lists	*/
169 	txg_node_t	vdev_txg_node;	/* per-txg dirty vdev linkage	*/
170 	boolean_t	vdev_remove_wanted; /* async remove wanted?	*/
171 	boolean_t	vdev_probe_wanted; /* async probe wanted?	*/
172 	list_node_t	vdev_config_dirty_node; /* config dirty list	*/
173 	list_node_t	vdev_state_dirty_node; /* state dirty list	*/
174 	uint64_t	vdev_deflate_ratio; /* deflation ratio (x512)	*/
175 	uint64_t	vdev_islog;	/* is an intent log device	*/
176 	uint64_t	vdev_removing;	/* device is being removed?	*/
177 	boolean_t	vdev_ishole;	/* is a hole in the namespace	*/
178 	kmutex_t	vdev_queue_lock; /* protects vdev_queue_depth	*/
179 	uint64_t	vdev_top_zap;
180 
181 	/*
182 	 * The queue depth parameters determine how many async writes are
183 	 * still pending (i.e. allocated by net yet issued to disk) per
184 	 * top-level (vdev_async_write_queue_depth) and the maximum allowed
185 	 * (vdev_max_async_write_queue_depth). These values only apply to
186 	 * top-level vdevs.
187 	 */
188 	uint64_t	vdev_async_write_queue_depth;
189 	uint64_t	vdev_max_async_write_queue_depth;
190 
191 	/*
192 	 * Leaf vdev state.
193 	 */
194 	range_tree_t	*vdev_dtl[DTL_TYPES]; /* dirty time logs	*/
195 	space_map_t	*vdev_dtl_sm;	/* dirty time log space map	*/
196 	txg_node_t	vdev_dtl_node;	/* per-txg dirty DTL linkage	*/
197 	uint64_t	vdev_dtl_object; /* DTL object			*/
198 	uint64_t	vdev_psize;	/* physical device capacity	*/
199 	uint64_t	vdev_wholedisk;	/* true if this is a whole disk */
200 	uint64_t	vdev_offline;	/* persistent offline state	*/
201 	uint64_t	vdev_faulted;	/* persistent faulted state	*/
202 	uint64_t	vdev_degraded;	/* persistent degraded state	*/
203 	uint64_t	vdev_removed;	/* persistent removed state	*/
204 	uint64_t	vdev_resilver_txg; /* persistent resilvering state */
205 	uint64_t	vdev_nparity;	/* number of parity devices for raidz */
206 	char		*vdev_path;	/* vdev path (if any)		*/
207 	char		*vdev_devid;	/* vdev devid (if any)		*/
208 	char		*vdev_physpath;	/* vdev device path (if any)	*/
209 	char		*vdev_fru;	/* physical FRU location	*/
210 	uint64_t	vdev_not_present; /* not present during import	*/
211 	uint64_t	vdev_unspare;	/* unspare when resilvering done */
212 	boolean_t	vdev_nowritecache; /* true if flushwritecache failed */
213 	boolean_t	vdev_checkremove; /* temporary online test	*/
214 	boolean_t	vdev_forcefault; /* force online fault		*/
215 	boolean_t	vdev_splitting;	/* split or repair in progress  */
216 	boolean_t	vdev_delayed_close; /* delayed device close?	*/
217 	boolean_t	vdev_tmpoffline; /* device taken offline temporarily? */
218 	boolean_t	vdev_detached;	/* device detached?		*/
219 	boolean_t	vdev_cant_read;	/* vdev is failing all reads	*/
220 	boolean_t	vdev_cant_write; /* vdev is failing all writes	*/
221 	boolean_t	vdev_isspare;	/* was a hot spare		*/
222 	boolean_t	vdev_isl2cache;	/* was a l2cache device		*/
223 	vdev_queue_t	vdev_queue;	/* I/O deadline schedule queue	*/
224 	vdev_cache_t	vdev_cache;	/* physical block cache		*/
225 	spa_aux_vdev_t	*vdev_aux;	/* for l2cache and spares vdevs	*/
226 	zio_t		*vdev_probe_zio; /* root of current probe	*/
227 	vdev_aux_t	vdev_label_aux;	/* on-disk aux state		*/
228 	uint64_t	vdev_leaf_zap;
229 
230 	/*
231 	 * For DTrace to work in userland (libzpool) context, these fields must
232 	 * remain at the end of the structure.  DTrace will use the kernel's
233 	 * CTF definition for 'struct vdev', and since the size of a kmutex_t is
234 	 * larger in userland, the offsets for the rest of the fields would be
235 	 * incorrect.
236 	 */
237 	kmutex_t	vdev_dtl_lock;	/* vdev_dtl_{map,resilver}	*/
238 	kmutex_t	vdev_stat_lock;	/* vdev_stat			*/
239 	kmutex_t	vdev_probe_lock; /* protects vdev_probe_zio	*/
240 };
241 
242 #define	VDEV_RAIDZ_MAXPARITY	3
243 
244 #define	VDEV_PAD_SIZE		(8 << 10)
245 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
246 #define	VDEV_SKIP_SIZE		VDEV_PAD_SIZE * 2
247 #define	VDEV_PHYS_SIZE		(112 << 10)
248 #define	VDEV_UBERBLOCK_RING	(128 << 10)
249 
250 /* The largest uberblock we support is 8k. */
251 #define	MAX_UBERBLOCK_SHIFT (13)
252 #define	VDEV_UBERBLOCK_SHIFT(vd)	\
253 	MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \
254 	    MAX_UBERBLOCK_SHIFT)
255 #define	VDEV_UBERBLOCK_COUNT(vd)	\
256 	(VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
257 #define	VDEV_UBERBLOCK_OFFSET(vd, n)	\
258 	offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
259 #define	VDEV_UBERBLOCK_SIZE(vd)		(1ULL << VDEV_UBERBLOCK_SHIFT(vd))
260 
261 typedef struct vdev_phys {
262 	char		vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
263 	zio_eck_t	vp_zbt;
264 } vdev_phys_t;
265 
266 typedef struct vdev_label {
267 	char		vl_pad1[VDEV_PAD_SIZE];			/*  8K */
268 	char		vl_pad2[VDEV_PAD_SIZE];			/*  8K */
269 	vdev_phys_t	vl_vdev_phys;				/* 112K	*/
270 	char		vl_uberblock[VDEV_UBERBLOCK_RING];	/* 128K	*/
271 } vdev_label_t;							/* 256K total */
272 
273 /*
274  * vdev_dirty() flags
275  */
276 #define	VDD_METASLAB	0x01
277 #define	VDD_DTL		0x02
278 
279 /* Offset of embedded boot loader region on each label */
280 #define	VDEV_BOOT_OFFSET	(2 * sizeof (vdev_label_t))
281 /*
282  * Size of embedded boot loader region on each label.
283  * The total size of the first two labels plus the boot area is 4MB.
284  */
285 #define	VDEV_BOOT_SIZE		(7ULL << 19)			/* 3.5M */
286 
287 /*
288  * Size of label regions at the start and end of each leaf device.
289  */
290 #define	VDEV_LABEL_START_SIZE	(2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
291 #define	VDEV_LABEL_END_SIZE	(2 * sizeof (vdev_label_t))
292 #define	VDEV_LABELS		4
293 #define	VDEV_BEST_LABEL		VDEV_LABELS
294 
295 #define	VDEV_ALLOC_LOAD		0
296 #define	VDEV_ALLOC_ADD		1
297 #define	VDEV_ALLOC_SPARE	2
298 #define	VDEV_ALLOC_L2CACHE	3
299 #define	VDEV_ALLOC_ROOTPOOL	4
300 #define	VDEV_ALLOC_SPLIT	5
301 #define	VDEV_ALLOC_ATTACH	6
302 
303 /*
304  * Allocate or free a vdev
305  */
306 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
307     vdev_ops_t *ops);
308 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
309     vdev_t *parent, uint_t id, int alloctype);
310 extern void vdev_free(vdev_t *vd);
311 
312 /*
313  * Add or remove children and parents
314  */
315 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
316 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
317 extern void vdev_compact_children(vdev_t *pvd);
318 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
319 extern void vdev_remove_parent(vdev_t *cvd);
320 
321 /*
322  * vdev sync load and sync
323  */
324 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd);
325 extern boolean_t vdev_log_state_valid(vdev_t *vd);
326 extern void vdev_load(vdev_t *vd);
327 extern int vdev_dtl_load(vdev_t *vd);
328 extern void vdev_sync(vdev_t *vd, uint64_t txg);
329 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
330 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
331 extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg);
332 
333 /*
334  * Available vdev types.
335  */
336 extern vdev_ops_t vdev_root_ops;
337 extern vdev_ops_t vdev_mirror_ops;
338 extern vdev_ops_t vdev_replacing_ops;
339 extern vdev_ops_t vdev_raidz_ops;
340 extern vdev_ops_t vdev_disk_ops;
341 extern vdev_ops_t vdev_file_ops;
342 extern vdev_ops_t vdev_missing_ops;
343 extern vdev_ops_t vdev_hole_ops;
344 extern vdev_ops_t vdev_spare_ops;
345 
346 /*
347  * Common size functions
348  */
349 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
350 extern uint64_t vdev_get_min_asize(vdev_t *vd);
351 extern void vdev_set_min_asize(vdev_t *vd);
352 
353 /*
354  * Global variables
355  */
356 /* zdb uses this tunable, so it must be declared here to make lint happy. */
357 extern int zfs_vdev_cache_size;
358 
359 /*
360  * The vdev_buf_t is used to translate between zio_t and buf_t, and back again.
361  */
362 typedef struct vdev_buf {
363 	buf_t	vb_buf;		/* buffer that describes the io */
364 	zio_t	*vb_io;		/* pointer back to the original zio_t */
365 } vdev_buf_t;
366 
367 #ifdef	__cplusplus
368 }
369 #endif
370 
371 #endif	/* _SYS_VDEV_IMPL_H */
372