]> git.proxmox.com Git - mirror_zfs-debian.git/blame - module/zfs/include/sys/vdev_impl.h
Rebase master to b117
[mirror_zfs-debian.git] / module / zfs / include / sys / vdev_impl.h
CommitLineData
34dc7c2f
BB
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/*
9babb374 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
34dc7c2f
BB
23 * Use is subject to license terms.
24 */
25
26#ifndef _SYS_VDEV_IMPL_H
27#define _SYS_VDEV_IMPL_H
28
34dc7c2f
BB
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
39extern "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 */
52typedef struct vdev_queue vdev_queue_t;
53typedef struct vdev_cache vdev_cache_t;
54typedef struct vdev_cache_entry vdev_cache_entry_t;
55
56/*
57 * Virtual device operations
58 */
59typedef int vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *ashift);
60typedef void vdev_close_func_t(vdev_t *vd);
34dc7c2f
BB
61typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
62typedef int vdev_io_start_func_t(zio_t *zio);
b128c09f 63typedef void vdev_io_done_func_t(zio_t *zio);
34dc7c2f
BB
64typedef void vdev_state_change_func_t(vdev_t *vd, int, int);
65
66typedef struct vdev_ops {
67 vdev_open_func_t *vdev_op_open;
68 vdev_close_func_t *vdev_op_close;
34dc7c2f
BB
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 */
80struct 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
91struct vdev_cache {
92 avl_tree_t vc_offset_tree;
93 avl_tree_t vc_lastused_tree;
94 kmutex_t vc_lock;
95};
96
97struct 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 */
108struct 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 */
9babb374 116 uint64_t vdev_min_asize; /* min acceptable asize */
34dc7c2f
BB
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 */
fb5f0bc8 127 space_map_t vdev_dtl[DTL_TYPES]; /* in-core dirty time logs */
34dc7c2f 128 vdev_stat_t vdev_stat; /* virtual device statistics */
9babb374 129 boolean_t vdev_expanding; /* expand the vdev? */
34dc7c2f
BB
130
131 /*
132 * Top-level vdev state.
133 */
134 uint64_t vdev_ms_array; /* metaslab array object */
135 uint64_t vdev_ms_shift; /* metaslab size shift */
136 uint64_t vdev_ms_count; /* number of metaslabs */
137 metaslab_group_t *vdev_mg; /* metaslab group */
138 metaslab_t **vdev_ms; /* metaslab array */
139 txg_list_t vdev_ms_list; /* per-txg dirty metaslab lists */
140 txg_list_t vdev_dtl_list; /* per-txg dirty DTL lists */
141 txg_node_t vdev_txg_node; /* per-txg dirty vdev linkage */
142 boolean_t vdev_remove_wanted; /* async remove wanted? */
b128c09f
BB
143 boolean_t vdev_probe_wanted; /* async probe wanted? */
144 list_node_t vdev_config_dirty_node; /* config dirty list */
145 list_node_t vdev_state_dirty_node; /* state dirty list */
34dc7c2f
BB
146 uint64_t vdev_deflate_ratio; /* deflation ratio (x512) */
147 uint64_t vdev_islog; /* is an intent log device */
148
149 /*
150 * Leaf vdev state.
151 */
152 uint64_t vdev_psize; /* physical device capacity */
fb5f0bc8 153 space_map_obj_t vdev_dtl_smo; /* dirty time log space map obj */
34dc7c2f
BB
154 txg_node_t vdev_dtl_node; /* per-txg dirty DTL linkage */
155 uint64_t vdev_wholedisk; /* true if this is a whole disk */
156 uint64_t vdev_offline; /* persistent offline state */
157 uint64_t vdev_faulted; /* persistent faulted state */
158 uint64_t vdev_degraded; /* persistent degraded state */
159 uint64_t vdev_removed; /* persistent removed state */
160 uint64_t vdev_nparity; /* number of parity devices for raidz */
161 char *vdev_path; /* vdev path (if any) */
162 char *vdev_devid; /* vdev devid (if any) */
163 char *vdev_physpath; /* vdev device path (if any) */
9babb374 164 char *vdev_fru; /* physical FRU location */
b128c09f
BB
165 uint64_t vdev_not_present; /* not present during import */
166 uint64_t vdev_unspare; /* unspare when resilvering done */
167 hrtime_t vdev_last_try; /* last reopen time */
168 boolean_t vdev_nowritecache; /* true if flushwritecache failed */
169 boolean_t vdev_checkremove; /* temporary online test */
170 boolean_t vdev_forcefault; /* force online fault */
34dc7c2f
BB
171 uint8_t vdev_tmpoffline; /* device taken offline temporarily? */
172 uint8_t vdev_detached; /* device detached? */
b128c09f
BB
173 uint8_t vdev_cant_read; /* vdev is failing all reads */
174 uint8_t vdev_cant_write; /* vdev is failing all writes */
34dc7c2f
BB
175 uint64_t vdev_isspare; /* was a hot spare */
176 uint64_t vdev_isl2cache; /* was a l2cache device */
177 vdev_queue_t vdev_queue; /* I/O deadline schedule queue */
178 vdev_cache_t vdev_cache; /* physical block cache */
b128c09f
BB
179 spa_aux_vdev_t *vdev_aux; /* for l2cache vdevs */
180 zio_t *vdev_probe_zio; /* root of current probe */
34dc7c2f
BB
181
182 /*
183 * For DTrace to work in userland (libzpool) context, these fields must
184 * remain at the end of the structure. DTrace will use the kernel's
185 * CTF definition for 'struct vdev', and since the size of a kmutex_t is
186 * larger in userland, the offsets for the rest fields would be
187 * incorrect.
188 */
189 kmutex_t vdev_dtl_lock; /* vdev_dtl_{map,resilver} */
190 kmutex_t vdev_stat_lock; /* vdev_stat */
b128c09f 191 kmutex_t vdev_probe_lock; /* protects vdev_probe_zio */
34dc7c2f
BB
192};
193
9babb374
BB
194#define VDEV_PAD_SIZE (8 << 10)
195/* 2 padding areas (vl_pad1 and vl_pad2) to skip */
196#define VDEV_SKIP_SIZE VDEV_PAD_SIZE * 2
34dc7c2f
BB
197#define VDEV_PHYS_SIZE (112 << 10)
198#define VDEV_UBERBLOCK_RING (128 << 10)
199
200#define VDEV_UBERBLOCK_SHIFT(vd) \
201 MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT)
202#define VDEV_UBERBLOCK_COUNT(vd) \
203 (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
204#define VDEV_UBERBLOCK_OFFSET(vd, n) \
205 offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
206#define VDEV_UBERBLOCK_SIZE(vd) (1ULL << VDEV_UBERBLOCK_SHIFT(vd))
207
34dc7c2f
BB
208typedef struct vdev_phys {
209 char vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_block_tail_t)];
210 zio_block_tail_t vp_zbt;
211} vdev_phys_t;
212
213typedef struct vdev_label {
9babb374
BB
214 char vl_pad1[VDEV_PAD_SIZE]; /* 8K */
215 char vl_pad2[VDEV_PAD_SIZE]; /* 8K */
34dc7c2f
BB
216 vdev_phys_t vl_vdev_phys; /* 112K */
217 char vl_uberblock[VDEV_UBERBLOCK_RING]; /* 128K */
218} vdev_label_t; /* 256K total */
219
220/*
221 * vdev_dirty() flags
222 */
223#define VDD_METASLAB 0x01
224#define VDD_DTL 0x02
225
226/*
227 * Size and offset of embedded boot loader region on each label.
228 * The total size of the first two labels plus the boot area is 4MB.
229 */
230#define VDEV_BOOT_OFFSET (2 * sizeof (vdev_label_t))
231#define VDEV_BOOT_SIZE (7ULL << 19) /* 3.5M */
232
233/*
234 * Size of label regions at the start and end of each leaf device.
235 */
236#define VDEV_LABEL_START_SIZE (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
237#define VDEV_LABEL_END_SIZE (2 * sizeof (vdev_label_t))
238#define VDEV_LABELS 4
239
240#define VDEV_ALLOC_LOAD 0
241#define VDEV_ALLOC_ADD 1
242#define VDEV_ALLOC_SPARE 2
243#define VDEV_ALLOC_L2CACHE 3
9babb374 244#define VDEV_ALLOC_ROOTPOOL 4
34dc7c2f
BB
245
246/*
247 * Allocate or free a vdev
248 */
249extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
250 vdev_t *parent, uint_t id, int alloctype);
251extern void vdev_free(vdev_t *vd);
252
253/*
254 * Add or remove children and parents
255 */
256extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
257extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
258extern void vdev_compact_children(vdev_t *pvd);
259extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
260extern void vdev_remove_parent(vdev_t *cvd);
261
262/*
263 * vdev sync load and sync
264 */
9babb374 265extern void vdev_load_log_state(vdev_t *vd, nvlist_t *nv);
34dc7c2f
BB
266extern void vdev_load(vdev_t *vd);
267extern void vdev_sync(vdev_t *vd, uint64_t txg);
268extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
269extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
270
271/*
272 * Available vdev types.
273 */
274extern vdev_ops_t vdev_root_ops;
275extern vdev_ops_t vdev_mirror_ops;
276extern vdev_ops_t vdev_replacing_ops;
277extern vdev_ops_t vdev_raidz_ops;
278extern vdev_ops_t vdev_disk_ops;
279extern vdev_ops_t vdev_file_ops;
280extern vdev_ops_t vdev_missing_ops;
281extern vdev_ops_t vdev_spare_ops;
282
283/*
284 * Common size functions
285 */
286extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
9babb374
BB
287extern uint64_t vdev_get_min_asize(vdev_t *vd);
288extern void vdev_set_min_asize(vdev_t *vd);
34dc7c2f
BB
289
290/*
291 * zdb uses this tunable, so it must be declared here to make lint happy.
292 */
293extern int zfs_vdev_cache_size;
294
295#ifdef __cplusplus
296}
297#endif
298
299#endif /* _SYS_VDEV_IMPL_H */