]>
Commit | Line | Data |
---|---|---|
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
1bd201e7 | 23 | * Copyright (c) 2012 by Delphix. All rights reserved. |
34dc7c2f BB |
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 | |
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; | |
86dd0fd9 | 53 | typedef struct vdev_io vdev_io_t; |
34dc7c2f BB |
54 | typedef struct vdev_cache vdev_cache_t; |
55 | typedef struct vdev_cache_entry vdev_cache_entry_t; | |
56 | ||
57 | /* | |
58 | * Virtual device operations | |
59 | */ | |
1bd201e7 CS |
60 | typedef int vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size, |
61 | uint64_t *ashift); | |
34dc7c2f | 62 | typedef void vdev_close_func_t(vdev_t *vd); |
34dc7c2f BB |
63 | typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize); |
64 | typedef int vdev_io_start_func_t(zio_t *zio); | |
b128c09f | 65 | typedef void vdev_io_done_func_t(zio_t *zio); |
34dc7c2f | 66 | typedef void vdev_state_change_func_t(vdev_t *vd, int, int); |
428870ff BB |
67 | typedef void vdev_hold_func_t(vdev_t *vd); |
68 | typedef void vdev_rele_func_t(vdev_t *vd); | |
34dc7c2f BB |
69 | |
70 | typedef struct vdev_ops { | |
71 | vdev_open_func_t *vdev_op_open; | |
72 | vdev_close_func_t *vdev_op_close; | |
34dc7c2f BB |
73 | vdev_asize_func_t *vdev_op_asize; |
74 | vdev_io_start_func_t *vdev_op_io_start; | |
75 | vdev_io_done_func_t *vdev_op_io_done; | |
76 | vdev_state_change_func_t *vdev_op_state_change; | |
428870ff BB |
77 | vdev_hold_func_t *vdev_op_hold; |
78 | vdev_rele_func_t *vdev_op_rele; | |
34dc7c2f BB |
79 | char vdev_op_type[16]; |
80 | boolean_t vdev_op_leaf; | |
81 | } vdev_ops_t; | |
82 | ||
83 | /* | |
84 | * Virtual device properties | |
85 | */ | |
86 | struct vdev_cache_entry { | |
87 | char *ve_data; | |
88 | uint64_t ve_offset; | |
89 | uint64_t ve_lastused; | |
90 | avl_node_t ve_offset_node; | |
91 | avl_node_t ve_lastused_node; | |
92 | uint32_t ve_hits; | |
93 | uint16_t ve_missed_update; | |
94 | zio_t *ve_fill_io; | |
95 | }; | |
96 | ||
97 | struct vdev_cache { | |
98 | avl_tree_t vc_offset_tree; | |
99 | avl_tree_t vc_lastused_tree; | |
100 | kmutex_t vc_lock; | |
101 | }; | |
102 | ||
103 | struct vdev_queue { | |
104 | avl_tree_t vq_deadline_tree; | |
105 | avl_tree_t vq_read_tree; | |
106 | avl_tree_t vq_write_tree; | |
107 | avl_tree_t vq_pending_tree; | |
86dd0fd9 | 108 | list_t vq_io_list; |
34dc7c2f BB |
109 | kmutex_t vq_lock; |
110 | }; | |
111 | ||
86dd0fd9 BB |
112 | struct vdev_io { |
113 | char vi_buffer[SPA_MAXBLOCKSIZE]; /* Must be first */ | |
114 | list_node_t vi_node; | |
115 | }; | |
116 | ||
34dc7c2f BB |
117 | /* |
118 | * Virtual device descriptor | |
119 | */ | |
120 | struct vdev { | |
121 | /* | |
122 | * Common to all vdev types. | |
123 | */ | |
124 | uint64_t vdev_id; /* child number in vdev parent */ | |
125 | uint64_t vdev_guid; /* unique ID for this vdev */ | |
126 | uint64_t vdev_guid_sum; /* self guid + all child guids */ | |
428870ff | 127 | uint64_t vdev_orig_guid; /* orig. guid prior to remove */ |
34dc7c2f | 128 | uint64_t vdev_asize; /* allocatable device capacity */ |
9babb374 | 129 | uint64_t vdev_min_asize; /* min acceptable asize */ |
1bd201e7 | 130 | uint64_t vdev_max_asize; /* max acceptable asize */ |
34dc7c2f BB |
131 | uint64_t vdev_ashift; /* block alignment shift */ |
132 | uint64_t vdev_state; /* see VDEV_STATE_* #defines */ | |
133 | uint64_t vdev_prevstate; /* used when reopening a vdev */ | |
134 | vdev_ops_t *vdev_ops; /* vdev operations */ | |
135 | spa_t *vdev_spa; /* spa for this vdev */ | |
136 | void *vdev_tsd; /* type-specific data */ | |
428870ff BB |
137 | vnode_t *vdev_name_vp; /* vnode for pathname */ |
138 | vnode_t *vdev_devid_vp; /* vnode for devid */ | |
34dc7c2f BB |
139 | vdev_t *vdev_top; /* top-level vdev */ |
140 | vdev_t *vdev_parent; /* parent vdev */ | |
141 | vdev_t **vdev_child; /* array of children */ | |
142 | uint64_t vdev_children; /* number of children */ | |
fb5f0bc8 | 143 | space_map_t vdev_dtl[DTL_TYPES]; /* in-core dirty time logs */ |
34dc7c2f | 144 | vdev_stat_t vdev_stat; /* virtual device statistics */ |
9babb374 | 145 | boolean_t vdev_expanding; /* expand the vdev? */ |
428870ff | 146 | boolean_t vdev_reopening; /* reopen in progress? */ |
45d1cae3 BB |
147 | int vdev_open_error; /* error on last open */ |
148 | kthread_t *vdev_open_thread; /* thread opening children */ | |
428870ff | 149 | uint64_t vdev_crtxg; /* txg when top-level was added */ |
34dc7c2f BB |
150 | |
151 | /* | |
152 | * Top-level vdev state. | |
153 | */ | |
154 | uint64_t vdev_ms_array; /* metaslab array object */ | |
155 | uint64_t vdev_ms_shift; /* metaslab size shift */ | |
156 | uint64_t vdev_ms_count; /* number of metaslabs */ | |
157 | metaslab_group_t *vdev_mg; /* metaslab group */ | |
158 | metaslab_t **vdev_ms; /* metaslab array */ | |
159 | txg_list_t vdev_ms_list; /* per-txg dirty metaslab lists */ | |
160 | txg_list_t vdev_dtl_list; /* per-txg dirty DTL lists */ | |
161 | txg_node_t vdev_txg_node; /* per-txg dirty vdev linkage */ | |
162 | boolean_t vdev_remove_wanted; /* async remove wanted? */ | |
b128c09f | 163 | boolean_t vdev_probe_wanted; /* async probe wanted? */ |
428870ff | 164 | uint64_t vdev_removing; /* device is being removed? */ |
b128c09f BB |
165 | list_node_t vdev_config_dirty_node; /* config dirty list */ |
166 | list_node_t vdev_state_dirty_node; /* state dirty list */ | |
34dc7c2f BB |
167 | uint64_t vdev_deflate_ratio; /* deflation ratio (x512) */ |
168 | uint64_t vdev_islog; /* is an intent log device */ | |
428870ff | 169 | uint64_t vdev_ishole; /* is a hole in the namespace */ |
34dc7c2f BB |
170 | |
171 | /* | |
172 | * Leaf vdev state. | |
173 | */ | |
174 | uint64_t vdev_psize; /* physical device capacity */ | |
fb5f0bc8 | 175 | space_map_obj_t vdev_dtl_smo; /* dirty time log space map obj */ |
34dc7c2f BB |
176 | txg_node_t vdev_dtl_node; /* per-txg dirty DTL linkage */ |
177 | uint64_t vdev_wholedisk; /* true if this is a whole disk */ | |
178 | uint64_t vdev_offline; /* persistent offline state */ | |
179 | uint64_t vdev_faulted; /* persistent faulted state */ | |
180 | uint64_t vdev_degraded; /* persistent degraded state */ | |
181 | uint64_t vdev_removed; /* persistent removed state */ | |
572e2857 | 182 | uint64_t vdev_resilvering; /* persistent resilvering state */ |
34dc7c2f BB |
183 | uint64_t vdev_nparity; /* number of parity devices for raidz */ |
184 | char *vdev_path; /* vdev path (if any) */ | |
185 | char *vdev_devid; /* vdev devid (if any) */ | |
186 | char *vdev_physpath; /* vdev device path (if any) */ | |
9babb374 | 187 | char *vdev_fru; /* physical FRU location */ |
b128c09f BB |
188 | uint64_t vdev_not_present; /* not present during import */ |
189 | uint64_t vdev_unspare; /* unspare when resilvering done */ | |
190 | hrtime_t vdev_last_try; /* last reopen time */ | |
191 | boolean_t vdev_nowritecache; /* true if flushwritecache failed */ | |
192 | boolean_t vdev_checkremove; /* temporary online test */ | |
193 | boolean_t vdev_forcefault; /* force online fault */ | |
428870ff BB |
194 | boolean_t vdev_splitting; /* split or repair in progress */ |
195 | boolean_t vdev_delayed_close; /* delayed device close? */ | |
34dc7c2f BB |
196 | uint8_t vdev_tmpoffline; /* device taken offline temporarily? */ |
197 | uint8_t vdev_detached; /* device detached? */ | |
b128c09f BB |
198 | uint8_t vdev_cant_read; /* vdev is failing all reads */ |
199 | uint8_t vdev_cant_write; /* vdev is failing all writes */ | |
34dc7c2f BB |
200 | uint64_t vdev_isspare; /* was a hot spare */ |
201 | uint64_t vdev_isl2cache; /* was a l2cache device */ | |
202 | vdev_queue_t vdev_queue; /* I/O deadline schedule queue */ | |
203 | vdev_cache_t vdev_cache; /* physical block cache */ | |
b128c09f BB |
204 | spa_aux_vdev_t *vdev_aux; /* for l2cache vdevs */ |
205 | zio_t *vdev_probe_zio; /* root of current probe */ | |
428870ff | 206 | vdev_aux_t vdev_label_aux; /* on-disk aux state */ |
34dc7c2f BB |
207 | |
208 | /* | |
209 | * For DTrace to work in userland (libzpool) context, these fields must | |
210 | * remain at the end of the structure. DTrace will use the kernel's | |
211 | * CTF definition for 'struct vdev', and since the size of a kmutex_t is | |
212 | * larger in userland, the offsets for the rest fields would be | |
213 | * incorrect. | |
214 | */ | |
215 | kmutex_t vdev_dtl_lock; /* vdev_dtl_{map,resilver} */ | |
216 | kmutex_t vdev_stat_lock; /* vdev_stat */ | |
b128c09f | 217 | kmutex_t vdev_probe_lock; /* protects vdev_probe_zio */ |
34dc7c2f BB |
218 | }; |
219 | ||
428870ff BB |
220 | #define VDEV_RAIDZ_MAXPARITY 3 |
221 | ||
9babb374 BB |
222 | #define VDEV_PAD_SIZE (8 << 10) |
223 | /* 2 padding areas (vl_pad1 and vl_pad2) to skip */ | |
224 | #define VDEV_SKIP_SIZE VDEV_PAD_SIZE * 2 | |
34dc7c2f BB |
225 | #define VDEV_PHYS_SIZE (112 << 10) |
226 | #define VDEV_UBERBLOCK_RING (128 << 10) | |
227 | ||
228 | #define VDEV_UBERBLOCK_SHIFT(vd) \ | |
229 | MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT) | |
230 | #define VDEV_UBERBLOCK_COUNT(vd) \ | |
231 | (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd)) | |
232 | #define VDEV_UBERBLOCK_OFFSET(vd, n) \ | |
233 | offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)]) | |
234 | #define VDEV_UBERBLOCK_SIZE(vd) (1ULL << VDEV_UBERBLOCK_SHIFT(vd)) | |
235 | ||
34dc7c2f | 236 | typedef struct vdev_phys { |
428870ff BB |
237 | char vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)]; |
238 | zio_eck_t vp_zbt; | |
34dc7c2f BB |
239 | } vdev_phys_t; |
240 | ||
241 | typedef struct vdev_label { | |
9babb374 BB |
242 | char vl_pad1[VDEV_PAD_SIZE]; /* 8K */ |
243 | char vl_pad2[VDEV_PAD_SIZE]; /* 8K */ | |
34dc7c2f BB |
244 | vdev_phys_t vl_vdev_phys; /* 112K */ |
245 | char vl_uberblock[VDEV_UBERBLOCK_RING]; /* 128K */ | |
246 | } vdev_label_t; /* 256K total */ | |
247 | ||
248 | /* | |
249 | * vdev_dirty() flags | |
250 | */ | |
251 | #define VDD_METASLAB 0x01 | |
252 | #define VDD_DTL 0x02 | |
253 | ||
254 | /* | |
255 | * Size and offset of embedded boot loader region on each label. | |
256 | * The total size of the first two labels plus the boot area is 4MB. | |
257 | */ | |
258 | #define VDEV_BOOT_OFFSET (2 * sizeof (vdev_label_t)) | |
259 | #define VDEV_BOOT_SIZE (7ULL << 19) /* 3.5M */ | |
260 | ||
261 | /* | |
262 | * Size of label regions at the start and end of each leaf device. | |
263 | */ | |
264 | #define VDEV_LABEL_START_SIZE (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE) | |
265 | #define VDEV_LABEL_END_SIZE (2 * sizeof (vdev_label_t)) | |
266 | #define VDEV_LABELS 4 | |
267 | ||
268 | #define VDEV_ALLOC_LOAD 0 | |
269 | #define VDEV_ALLOC_ADD 1 | |
270 | #define VDEV_ALLOC_SPARE 2 | |
271 | #define VDEV_ALLOC_L2CACHE 3 | |
9babb374 | 272 | #define VDEV_ALLOC_ROOTPOOL 4 |
428870ff | 273 | #define VDEV_ALLOC_SPLIT 5 |
5ffb9d1d | 274 | #define VDEV_ALLOC_ATTACH 6 |
34dc7c2f BB |
275 | |
276 | /* | |
277 | * Allocate or free a vdev | |
278 | */ | |
428870ff BB |
279 | extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, |
280 | vdev_ops_t *ops); | |
34dc7c2f BB |
281 | extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config, |
282 | vdev_t *parent, uint_t id, int alloctype); | |
283 | extern void vdev_free(vdev_t *vd); | |
284 | ||
285 | /* | |
286 | * Add or remove children and parents | |
287 | */ | |
288 | extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd); | |
289 | extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd); | |
290 | extern void vdev_compact_children(vdev_t *pvd); | |
291 | extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops); | |
292 | extern void vdev_remove_parent(vdev_t *cvd); | |
293 | ||
294 | /* | |
295 | * vdev sync load and sync | |
296 | */ | |
428870ff | 297 | extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd); |
572e2857 | 298 | extern boolean_t vdev_log_state_valid(vdev_t *vd); |
34dc7c2f BB |
299 | extern void vdev_load(vdev_t *vd); |
300 | extern void vdev_sync(vdev_t *vd, uint64_t txg); | |
301 | extern void vdev_sync_done(vdev_t *vd, uint64_t txg); | |
302 | extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg); | |
303 | ||
304 | /* | |
305 | * Available vdev types. | |
306 | */ | |
307 | extern vdev_ops_t vdev_root_ops; | |
308 | extern vdev_ops_t vdev_mirror_ops; | |
309 | extern vdev_ops_t vdev_replacing_ops; | |
310 | extern vdev_ops_t vdev_raidz_ops; | |
311 | extern vdev_ops_t vdev_disk_ops; | |
312 | extern vdev_ops_t vdev_file_ops; | |
313 | extern vdev_ops_t vdev_missing_ops; | |
428870ff | 314 | extern vdev_ops_t vdev_hole_ops; |
34dc7c2f BB |
315 | extern vdev_ops_t vdev_spare_ops; |
316 | ||
317 | /* | |
318 | * Common size functions | |
319 | */ | |
320 | extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize); | |
9babb374 BB |
321 | extern uint64_t vdev_get_min_asize(vdev_t *vd); |
322 | extern void vdev_set_min_asize(vdev_t *vd); | |
34dc7c2f BB |
323 | |
324 | /* | |
325 | * zdb uses this tunable, so it must be declared here to make lint happy. | |
326 | */ | |
327 | extern int zfs_vdev_cache_size; | |
328 | ||
329 | #ifdef __cplusplus | |
330 | } | |
331 | #endif | |
332 | ||
333 | #endif /* _SYS_VDEV_IMPL_H */ |