Print this page
4334 Improve ZFS N-way mirror read performance
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/sys/vdev_impl.h
+++ new/usr/src/uts/common/fs/zfs/sys/vdev_impl.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
|
↓ open down ↓ |
13 lines elided |
↑ open up ↑ |
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 + * Copyright (c) 2013 Steven Hartland. All rights reserved.
24 25 */
25 26
26 27 #ifndef _SYS_VDEV_IMPL_H
27 28 #define _SYS_VDEV_IMPL_H
28 29
29 30 #include <sys/avl.h>
30 31 #include <sys/dmu.h>
31 32 #include <sys/metaslab.h>
32 33 #include <sys/nvpair.h>
33 34 #include <sys/space_map.h>
34 35 #include <sys/vdev.h>
35 36 #include <sys/dkio.h>
36 37 #include <sys/uberblock_impl.h>
37 38
38 39 #ifdef __cplusplus
39 40 extern "C" {
40 41 #endif
41 42
42 43 /*
43 44 * Virtual device descriptors.
44 45 *
45 46 * All storage pool operations go through the virtual device framework,
46 47 * which provides data replication and I/O scheduling.
47 48 */
48 49
49 50 /*
50 51 * Forward declarations that lots of things need.
51 52 */
52 53 typedef struct vdev_queue vdev_queue_t;
53 54 typedef struct vdev_cache vdev_cache_t;
54 55 typedef struct vdev_cache_entry vdev_cache_entry_t;
55 56
56 57 /*
57 58 * Virtual device operations
58 59 */
59 60 typedef int vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size,
60 61 uint64_t *ashift);
61 62 typedef void vdev_close_func_t(vdev_t *vd);
62 63 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
63 64 typedef int vdev_io_start_func_t(zio_t *zio);
64 65 typedef void vdev_io_done_func_t(zio_t *zio);
65 66 typedef void vdev_state_change_func_t(vdev_t *vd, int, int);
66 67 typedef void vdev_hold_func_t(vdev_t *vd);
67 68 typedef void vdev_rele_func_t(vdev_t *vd);
68 69
69 70 typedef struct vdev_ops {
70 71 vdev_open_func_t *vdev_op_open;
71 72 vdev_close_func_t *vdev_op_close;
72 73 vdev_asize_func_t *vdev_op_asize;
73 74 vdev_io_start_func_t *vdev_op_io_start;
74 75 vdev_io_done_func_t *vdev_op_io_done;
75 76 vdev_state_change_func_t *vdev_op_state_change;
76 77 vdev_hold_func_t *vdev_op_hold;
77 78 vdev_rele_func_t *vdev_op_rele;
78 79 char vdev_op_type[16];
79 80 boolean_t vdev_op_leaf;
80 81 } vdev_ops_t;
81 82
82 83 /*
83 84 * Virtual device properties
84 85 */
85 86 struct vdev_cache_entry {
86 87 char *ve_data;
87 88 uint64_t ve_offset;
88 89 uint64_t ve_lastused;
89 90 avl_node_t ve_offset_node;
90 91 avl_node_t ve_lastused_node;
91 92 uint32_t ve_hits;
92 93 uint16_t ve_missed_update;
93 94 zio_t *ve_fill_io;
94 95 };
95 96
96 97 struct vdev_cache {
97 98 avl_tree_t vc_offset_tree;
98 99 avl_tree_t vc_lastused_tree;
99 100 kmutex_t vc_lock;
100 101 };
101 102
102 103 typedef struct vdev_queue_class {
103 104 uint32_t vqc_active;
104 105
105 106 /*
106 107 * Sorted by offset or timestamp, depending on if the queue is
107 108 * LBA-ordered vs FIFO.
108 109 */
|
↓ open down ↓ |
75 lines elided |
↑ open up ↑ |
109 110 avl_tree_t vqc_queued_tree;
110 111 } vdev_queue_class_t;
111 112
112 113 struct vdev_queue {
113 114 vdev_t *vq_vdev;
114 115 vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE];
115 116 avl_tree_t vq_active_tree;
116 117 uint64_t vq_last_offset;
117 118 hrtime_t vq_io_complete_ts; /* time last i/o completed */
118 119 kmutex_t vq_lock;
120 + uint64_t vq_lastoffset;
119 121 };
120 122
121 123 /*
122 124 * Virtual device descriptor
123 125 */
124 126 struct vdev {
125 127 /*
126 128 * Common to all vdev types.
127 129 */
128 130 uint64_t vdev_id; /* child number in vdev parent */
129 131 uint64_t vdev_guid; /* unique ID for this vdev */
130 132 uint64_t vdev_guid_sum; /* self guid + all child guids */
131 133 uint64_t vdev_orig_guid; /* orig. guid prior to remove */
132 134 uint64_t vdev_asize; /* allocatable device capacity */
133 135 uint64_t vdev_min_asize; /* min acceptable asize */
134 136 uint64_t vdev_max_asize; /* max acceptable asize */
135 137 uint64_t vdev_ashift; /* block alignment shift */
136 138 uint64_t vdev_state; /* see VDEV_STATE_* #defines */
137 139 uint64_t vdev_prevstate; /* used when reopening a vdev */
138 140 vdev_ops_t *vdev_ops; /* vdev operations */
139 141 spa_t *vdev_spa; /* spa for this vdev */
140 142 void *vdev_tsd; /* type-specific data */
141 143 vnode_t *vdev_name_vp; /* vnode for pathname */
142 144 vnode_t *vdev_devid_vp; /* vnode for devid */
143 145 vdev_t *vdev_top; /* top-level vdev */
144 146 vdev_t *vdev_parent; /* parent vdev */
145 147 vdev_t **vdev_child; /* array of children */
146 148 uint64_t vdev_children; /* number of children */
147 149 vdev_stat_t vdev_stat; /* virtual device statistics */
148 150 boolean_t vdev_expanding; /* expand the vdev? */
149 151 boolean_t vdev_reopening; /* reopen in progress? */
150 152 int vdev_open_error; /* error on last open */
151 153 kthread_t *vdev_open_thread; /* thread opening children */
152 154 uint64_t vdev_crtxg; /* txg when top-level was added */
153 155
154 156 /*
155 157 * Top-level vdev state.
156 158 */
157 159 uint64_t vdev_ms_array; /* metaslab array object */
158 160 uint64_t vdev_ms_shift; /* metaslab size shift */
159 161 uint64_t vdev_ms_count; /* number of metaslabs */
160 162 metaslab_group_t *vdev_mg; /* metaslab group */
161 163 metaslab_t **vdev_ms; /* metaslab array */
162 164 txg_list_t vdev_ms_list; /* per-txg dirty metaslab lists */
163 165 txg_list_t vdev_dtl_list; /* per-txg dirty DTL lists */
164 166 txg_node_t vdev_txg_node; /* per-txg dirty vdev linkage */
165 167 boolean_t vdev_remove_wanted; /* async remove wanted? */
166 168 boolean_t vdev_probe_wanted; /* async probe wanted? */
167 169 list_node_t vdev_config_dirty_node; /* config dirty list */
168 170 list_node_t vdev_state_dirty_node; /* state dirty list */
169 171 uint64_t vdev_deflate_ratio; /* deflation ratio (x512) */
170 172 uint64_t vdev_islog; /* is an intent log device */
171 173 uint64_t vdev_removing; /* device is being removed? */
172 174 boolean_t vdev_ishole; /* is a hole in the namespace */
173 175
174 176 /*
175 177 * Leaf vdev state.
176 178 */
177 179 range_tree_t *vdev_dtl[DTL_TYPES]; /* dirty time logs */
178 180 space_map_t *vdev_dtl_sm; /* dirty time log space map */
179 181 txg_node_t vdev_dtl_node; /* per-txg dirty DTL linkage */
180 182 uint64_t vdev_dtl_object; /* DTL object */
181 183 uint64_t vdev_psize; /* physical device capacity */
182 184 uint64_t vdev_wholedisk; /* true if this is a whole disk */
183 185 uint64_t vdev_offline; /* persistent offline state */
184 186 uint64_t vdev_faulted; /* persistent faulted state */
185 187 uint64_t vdev_degraded; /* persistent degraded state */
186 188 uint64_t vdev_removed; /* persistent removed state */
187 189 uint64_t vdev_resilver_txg; /* persistent resilvering state */
188 190 uint64_t vdev_nparity; /* number of parity devices for raidz */
189 191 char *vdev_path; /* vdev path (if any) */
190 192 char *vdev_devid; /* vdev devid (if any) */
191 193 char *vdev_physpath; /* vdev device path (if any) */
192 194 char *vdev_fru; /* physical FRU location */
193 195 uint64_t vdev_not_present; /* not present during import */
194 196 uint64_t vdev_unspare; /* unspare when resilvering done */
195 197 boolean_t vdev_nowritecache; /* true if flushwritecache failed */
196 198 boolean_t vdev_checkremove; /* temporary online test */
197 199 boolean_t vdev_forcefault; /* force online fault */
198 200 boolean_t vdev_splitting; /* split or repair in progress */
199 201 boolean_t vdev_delayed_close; /* delayed device close? */
200 202 boolean_t vdev_tmpoffline; /* device taken offline temporarily? */
|
↓ open down ↓ |
72 lines elided |
↑ open up ↑ |
201 203 boolean_t vdev_detached; /* device detached? */
202 204 boolean_t vdev_cant_read; /* vdev is failing all reads */
203 205 boolean_t vdev_cant_write; /* vdev is failing all writes */
204 206 boolean_t vdev_isspare; /* was a hot spare */
205 207 boolean_t vdev_isl2cache; /* was a l2cache device */
206 208 vdev_queue_t vdev_queue; /* I/O deadline schedule queue */
207 209 vdev_cache_t vdev_cache; /* physical block cache */
208 210 spa_aux_vdev_t *vdev_aux; /* for l2cache vdevs */
209 211 zio_t *vdev_probe_zio; /* root of current probe */
210 212 vdev_aux_t vdev_label_aux; /* on-disk aux state */
213 + uint16_t vdev_rotation_rate; /* rotational rate of the media */
214 +#define VDEV_RATE_UNKNOWN 0
215 +#define VDEV_RATE_NON_ROTATING 1
211 216
212 217 /*
213 218 * For DTrace to work in userland (libzpool) context, these fields must
214 219 * remain at the end of the structure. DTrace will use the kernel's
215 220 * CTF definition for 'struct vdev', and since the size of a kmutex_t is
216 221 * larger in userland, the offsets for the rest of the fields would be
217 222 * incorrect.
218 223 */
219 224 kmutex_t vdev_dtl_lock; /* vdev_dtl_{map,resilver} */
220 225 kmutex_t vdev_stat_lock; /* vdev_stat */
221 226 kmutex_t vdev_probe_lock; /* protects vdev_probe_zio */
222 227 };
223 228
224 229 #define VDEV_RAIDZ_MAXPARITY 3
225 230
226 231 #define VDEV_PAD_SIZE (8 << 10)
227 232 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
228 233 #define VDEV_SKIP_SIZE VDEV_PAD_SIZE * 2
229 234 #define VDEV_PHYS_SIZE (112 << 10)
230 235 #define VDEV_UBERBLOCK_RING (128 << 10)
231 236
232 237 #define VDEV_UBERBLOCK_SHIFT(vd) \
233 238 MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT)
234 239 #define VDEV_UBERBLOCK_COUNT(vd) \
235 240 (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
236 241 #define VDEV_UBERBLOCK_OFFSET(vd, n) \
237 242 offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
238 243 #define VDEV_UBERBLOCK_SIZE(vd) (1ULL << VDEV_UBERBLOCK_SHIFT(vd))
239 244
240 245 typedef struct vdev_phys {
241 246 char vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
242 247 zio_eck_t vp_zbt;
243 248 } vdev_phys_t;
244 249
245 250 typedef struct vdev_label {
246 251 char vl_pad1[VDEV_PAD_SIZE]; /* 8K */
247 252 char vl_pad2[VDEV_PAD_SIZE]; /* 8K */
248 253 vdev_phys_t vl_vdev_phys; /* 112K */
249 254 char vl_uberblock[VDEV_UBERBLOCK_RING]; /* 128K */
250 255 } vdev_label_t; /* 256K total */
251 256
252 257 /*
253 258 * vdev_dirty() flags
254 259 */
255 260 #define VDD_METASLAB 0x01
256 261 #define VDD_DTL 0x02
257 262
258 263 /* Offset of embedded boot loader region on each label */
259 264 #define VDEV_BOOT_OFFSET (2 * sizeof (vdev_label_t))
260 265 /*
261 266 * Size of embedded boot loader region on each label.
262 267 * The total size of the first two labels plus the boot area is 4MB.
263 268 */
264 269 #define VDEV_BOOT_SIZE (7ULL << 19) /* 3.5M */
265 270
266 271 /*
267 272 * Size of label regions at the start and end of each leaf device.
268 273 */
269 274 #define VDEV_LABEL_START_SIZE (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
270 275 #define VDEV_LABEL_END_SIZE (2 * sizeof (vdev_label_t))
271 276 #define VDEV_LABELS 4
272 277 #define VDEV_BEST_LABEL VDEV_LABELS
273 278
274 279 #define VDEV_ALLOC_LOAD 0
275 280 #define VDEV_ALLOC_ADD 1
276 281 #define VDEV_ALLOC_SPARE 2
277 282 #define VDEV_ALLOC_L2CACHE 3
278 283 #define VDEV_ALLOC_ROOTPOOL 4
279 284 #define VDEV_ALLOC_SPLIT 5
280 285 #define VDEV_ALLOC_ATTACH 6
281 286
282 287 /*
283 288 * Allocate or free a vdev
284 289 */
285 290 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
286 291 vdev_ops_t *ops);
287 292 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
288 293 vdev_t *parent, uint_t id, int alloctype);
289 294 extern void vdev_free(vdev_t *vd);
290 295
291 296 /*
292 297 * Add or remove children and parents
293 298 */
294 299 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
295 300 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
296 301 extern void vdev_compact_children(vdev_t *pvd);
297 302 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
298 303 extern void vdev_remove_parent(vdev_t *cvd);
299 304
300 305 /*
301 306 * vdev sync load and sync
302 307 */
303 308 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd);
304 309 extern boolean_t vdev_log_state_valid(vdev_t *vd);
305 310 extern void vdev_load(vdev_t *vd);
306 311 extern int vdev_dtl_load(vdev_t *vd);
307 312 extern void vdev_sync(vdev_t *vd, uint64_t txg);
308 313 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
309 314 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
310 315 extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg);
311 316
312 317 /*
313 318 * Available vdev types.
314 319 */
315 320 extern vdev_ops_t vdev_root_ops;
316 321 extern vdev_ops_t vdev_mirror_ops;
317 322 extern vdev_ops_t vdev_replacing_ops;
318 323 extern vdev_ops_t vdev_raidz_ops;
319 324 extern vdev_ops_t vdev_disk_ops;
320 325 extern vdev_ops_t vdev_file_ops;
321 326 extern vdev_ops_t vdev_missing_ops;
322 327 extern vdev_ops_t vdev_hole_ops;
323 328 extern vdev_ops_t vdev_spare_ops;
324 329
325 330 /*
326 331 * Common size functions
327 332 */
328 333 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
329 334 extern uint64_t vdev_get_min_asize(vdev_t *vd);
330 335 extern void vdev_set_min_asize(vdev_t *vd);
331 336
332 337 /*
333 338 * Global variables
334 339 */
335 340 /* zdb uses this tunable, so it must be declared here to make lint happy. */
336 341 extern int zfs_vdev_cache_size;
337 342
338 343 /*
339 344 * The vdev_buf_t is used to translate between zio_t and buf_t, and back again.
340 345 */
341 346 typedef struct vdev_buf {
342 347 buf_t vb_buf; /* buffer that describes the io */
343 348 zio_t *vb_io; /* pointer back to the original zio_t */
344 349 } vdev_buf_t;
345 350
346 351 #ifdef __cplusplus
347 352 }
348 353 #endif
349 354
350 355 #endif /* _SYS_VDEV_IMPL_H */
|
↓ open down ↓ |
130 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX