Print this page
4045 zfs write throttle & i/o scheduler performance work
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
| 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
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 24 */
25 25
26 26 #ifndef _SYS_VDEV_IMPL_H
27 27 #define _SYS_VDEV_IMPL_H
28 28
29 29 #include <sys/avl.h>
30 30 #include <sys/dmu.h>
31 31 #include <sys/metaslab.h>
32 32 #include <sys/nvpair.h>
33 33 #include <sys/space_map.h>
34 34 #include <sys/vdev.h>
35 35 #include <sys/dkio.h>
36 36 #include <sys/uberblock_impl.h>
37 37
38 38 #ifdef __cplusplus
39 39 extern "C" {
40 40 #endif
41 41
42 42 /*
43 43 * Virtual device descriptors.
44 44 *
45 45 * All storage pool operations go through the virtual device framework,
46 46 * which provides data replication and I/O scheduling.
47 47 */
48 48
49 49 /*
50 50 * Forward declarations that lots of things need.
51 51 */
52 52 typedef struct vdev_queue vdev_queue_t;
53 53 typedef struct vdev_cache vdev_cache_t;
54 54 typedef struct vdev_cache_entry vdev_cache_entry_t;
55 55
56 56 /*
57 57 * Virtual device operations
58 58 */
59 59 typedef int vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size,
60 60 uint64_t *ashift);
61 61 typedef void vdev_close_func_t(vdev_t *vd);
62 62 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
63 63 typedef int vdev_io_start_func_t(zio_t *zio);
64 64 typedef void vdev_io_done_func_t(zio_t *zio);
65 65 typedef void vdev_state_change_func_t(vdev_t *vd, int, int);
66 66 typedef void vdev_hold_func_t(vdev_t *vd);
67 67 typedef void vdev_rele_func_t(vdev_t *vd);
68 68
69 69 typedef struct vdev_ops {
70 70 vdev_open_func_t *vdev_op_open;
71 71 vdev_close_func_t *vdev_op_close;
72 72 vdev_asize_func_t *vdev_op_asize;
73 73 vdev_io_start_func_t *vdev_op_io_start;
74 74 vdev_io_done_func_t *vdev_op_io_done;
75 75 vdev_state_change_func_t *vdev_op_state_change;
76 76 vdev_hold_func_t *vdev_op_hold;
77 77 vdev_rele_func_t *vdev_op_rele;
78 78 char vdev_op_type[16];
79 79 boolean_t vdev_op_leaf;
80 80 } vdev_ops_t;
81 81
82 82 /*
83 83 * Virtual device properties
84 84 */
85 85 struct vdev_cache_entry {
86 86 char *ve_data;
87 87 uint64_t ve_offset;
88 88 uint64_t ve_lastused;
89 89 avl_node_t ve_offset_node;
90 90 avl_node_t ve_lastused_node;
91 91 uint32_t ve_hits;
|
↓ open down ↓ |
91 lines elided |
↑ open up ↑ |
92 92 uint16_t ve_missed_update;
93 93 zio_t *ve_fill_io;
94 94 };
95 95
96 96 struct vdev_cache {
97 97 avl_tree_t vc_offset_tree;
98 98 avl_tree_t vc_lastused_tree;
99 99 kmutex_t vc_lock;
100 100 };
101 101
102 +typedef struct vdev_queue_class {
103 + uint32_t vqc_active;
104 +
105 + /*
106 + * Sorted by offset or timestamp, depending on if the queue is
107 + * LBA-ordered vs FIFO.
108 + */
109 + avl_tree_t vqc_queued_tree;
110 +} vdev_queue_class_t;
111 +
102 112 struct vdev_queue {
103 - avl_tree_t vq_deadline_tree;
104 - avl_tree_t vq_read_tree;
105 - avl_tree_t vq_write_tree;
106 - avl_tree_t vq_pending_tree;
107 - hrtime_t vq_io_complete_ts;
113 + vdev_t *vq_vdev;
114 + vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE];
115 + avl_tree_t vq_active_tree;
116 + uint64_t vq_last_offset;
117 + hrtime_t vq_io_complete_ts; /* time last i/o completed */
108 118 kmutex_t vq_lock;
109 119 };
110 120
111 121 /*
112 122 * Virtual device descriptor
113 123 */
114 124 struct vdev {
115 125 /*
116 126 * Common to all vdev types.
117 127 */
118 128 uint64_t vdev_id; /* child number in vdev parent */
119 129 uint64_t vdev_guid; /* unique ID for this vdev */
120 130 uint64_t vdev_guid_sum; /* self guid + all child guids */
121 131 uint64_t vdev_orig_guid; /* orig. guid prior to remove */
122 132 uint64_t vdev_asize; /* allocatable device capacity */
123 133 uint64_t vdev_min_asize; /* min acceptable asize */
124 134 uint64_t vdev_max_asize; /* max acceptable asize */
125 135 uint64_t vdev_ashift; /* block alignment shift */
126 136 uint64_t vdev_state; /* see VDEV_STATE_* #defines */
127 137 uint64_t vdev_prevstate; /* used when reopening a vdev */
128 138 vdev_ops_t *vdev_ops; /* vdev operations */
129 139 spa_t *vdev_spa; /* spa for this vdev */
130 140 void *vdev_tsd; /* type-specific data */
131 141 vnode_t *vdev_name_vp; /* vnode for pathname */
132 142 vnode_t *vdev_devid_vp; /* vnode for devid */
133 143 vdev_t *vdev_top; /* top-level vdev */
134 144 vdev_t *vdev_parent; /* parent vdev */
135 145 vdev_t **vdev_child; /* array of children */
136 146 uint64_t vdev_children; /* number of children */
137 147 space_map_t vdev_dtl[DTL_TYPES]; /* in-core dirty time logs */
138 148 vdev_stat_t vdev_stat; /* virtual device statistics */
139 149 boolean_t vdev_expanding; /* expand the vdev? */
140 150 boolean_t vdev_reopening; /* reopen in progress? */
141 151 int vdev_open_error; /* error on last open */
142 152 kthread_t *vdev_open_thread; /* thread opening children */
143 153 uint64_t vdev_crtxg; /* txg when top-level was added */
144 154
145 155 /*
146 156 * Top-level vdev state.
147 157 */
148 158 uint64_t vdev_ms_array; /* metaslab array object */
149 159 uint64_t vdev_ms_shift; /* metaslab size shift */
150 160 uint64_t vdev_ms_count; /* number of metaslabs */
151 161 metaslab_group_t *vdev_mg; /* metaslab group */
152 162 metaslab_t **vdev_ms; /* metaslab array */
153 163 txg_list_t vdev_ms_list; /* per-txg dirty metaslab lists */
154 164 txg_list_t vdev_dtl_list; /* per-txg dirty DTL lists */
155 165 txg_node_t vdev_txg_node; /* per-txg dirty vdev linkage */
156 166 boolean_t vdev_remove_wanted; /* async remove wanted? */
157 167 boolean_t vdev_probe_wanted; /* async probe wanted? */
158 168 uint64_t vdev_removing; /* device is being removed? */
159 169 list_node_t vdev_config_dirty_node; /* config dirty list */
160 170 list_node_t vdev_state_dirty_node; /* state dirty list */
161 171 uint64_t vdev_deflate_ratio; /* deflation ratio (x512) */
162 172 uint64_t vdev_islog; /* is an intent log device */
163 173 uint64_t vdev_ishole; /* is a hole in the namespace */
164 174
165 175 /*
166 176 * Leaf vdev state.
167 177 */
168 178 uint64_t vdev_psize; /* physical device capacity */
169 179 space_map_obj_t vdev_dtl_smo; /* dirty time log space map obj */
170 180 txg_node_t vdev_dtl_node; /* per-txg dirty DTL linkage */
171 181 uint64_t vdev_wholedisk; /* true if this is a whole disk */
172 182 uint64_t vdev_offline; /* persistent offline state */
173 183 uint64_t vdev_faulted; /* persistent faulted state */
174 184 uint64_t vdev_degraded; /* persistent degraded state */
175 185 uint64_t vdev_removed; /* persistent removed state */
176 186 uint64_t vdev_resilver_txg; /* persistent resilvering state */
177 187 uint64_t vdev_nparity; /* number of parity devices for raidz */
178 188 char *vdev_path; /* vdev path (if any) */
179 189 char *vdev_devid; /* vdev devid (if any) */
180 190 char *vdev_physpath; /* vdev device path (if any) */
181 191 char *vdev_fru; /* physical FRU location */
182 192 uint64_t vdev_not_present; /* not present during import */
183 193 uint64_t vdev_unspare; /* unspare when resilvering done */
184 194 hrtime_t vdev_last_try; /* last reopen time */
185 195 boolean_t vdev_nowritecache; /* true if flushwritecache failed */
186 196 boolean_t vdev_checkremove; /* temporary online test */
187 197 boolean_t vdev_forcefault; /* force online fault */
188 198 boolean_t vdev_splitting; /* split or repair in progress */
189 199 boolean_t vdev_delayed_close; /* delayed device close? */
190 200 uint8_t vdev_tmpoffline; /* device taken offline temporarily? */
191 201 uint8_t vdev_detached; /* device detached? */
192 202 uint8_t vdev_cant_read; /* vdev is failing all reads */
193 203 uint8_t vdev_cant_write; /* vdev is failing all writes */
194 204 uint64_t vdev_isspare; /* was a hot spare */
195 205 uint64_t vdev_isl2cache; /* was a l2cache device */
196 206 vdev_queue_t vdev_queue; /* I/O deadline schedule queue */
197 207 vdev_cache_t vdev_cache; /* physical block cache */
198 208 spa_aux_vdev_t *vdev_aux; /* for l2cache vdevs */
199 209 zio_t *vdev_probe_zio; /* root of current probe */
200 210 vdev_aux_t vdev_label_aux; /* on-disk aux state */
201 211
202 212 /*
203 213 * For DTrace to work in userland (libzpool) context, these fields must
204 214 * remain at the end of the structure. DTrace will use the kernel's
205 215 * CTF definition for 'struct vdev', and since the size of a kmutex_t is
206 216 * larger in userland, the offsets for the rest of the fields would be
207 217 * incorrect.
208 218 */
209 219 kmutex_t vdev_dtl_lock; /* vdev_dtl_{map,resilver} */
210 220 kmutex_t vdev_stat_lock; /* vdev_stat */
211 221 kmutex_t vdev_probe_lock; /* protects vdev_probe_zio */
212 222 };
213 223
214 224 #define VDEV_RAIDZ_MAXPARITY 3
215 225
216 226 #define VDEV_PAD_SIZE (8 << 10)
217 227 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
218 228 #define VDEV_SKIP_SIZE VDEV_PAD_SIZE * 2
219 229 #define VDEV_PHYS_SIZE (112 << 10)
220 230 #define VDEV_UBERBLOCK_RING (128 << 10)
221 231
222 232 #define VDEV_UBERBLOCK_SHIFT(vd) \
223 233 MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT)
224 234 #define VDEV_UBERBLOCK_COUNT(vd) \
225 235 (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
226 236 #define VDEV_UBERBLOCK_OFFSET(vd, n) \
227 237 offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
228 238 #define VDEV_UBERBLOCK_SIZE(vd) (1ULL << VDEV_UBERBLOCK_SHIFT(vd))
229 239
230 240 typedef struct vdev_phys {
231 241 char vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
232 242 zio_eck_t vp_zbt;
233 243 } vdev_phys_t;
234 244
235 245 typedef struct vdev_label {
236 246 char vl_pad1[VDEV_PAD_SIZE]; /* 8K */
237 247 char vl_pad2[VDEV_PAD_SIZE]; /* 8K */
238 248 vdev_phys_t vl_vdev_phys; /* 112K */
239 249 char vl_uberblock[VDEV_UBERBLOCK_RING]; /* 128K */
240 250 } vdev_label_t; /* 256K total */
241 251
242 252 /*
243 253 * vdev_dirty() flags
244 254 */
245 255 #define VDD_METASLAB 0x01
246 256 #define VDD_DTL 0x02
247 257
248 258 /* Offset of embedded boot loader region on each label */
249 259 #define VDEV_BOOT_OFFSET (2 * sizeof (vdev_label_t))
250 260 /*
251 261 * Size of embedded boot loader region on each label.
252 262 * The total size of the first two labels plus the boot area is 4MB.
253 263 */
254 264 #define VDEV_BOOT_SIZE (7ULL << 19) /* 3.5M */
255 265
256 266 /*
257 267 * Size of label regions at the start and end of each leaf device.
258 268 */
259 269 #define VDEV_LABEL_START_SIZE (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
260 270 #define VDEV_LABEL_END_SIZE (2 * sizeof (vdev_label_t))
261 271 #define VDEV_LABELS 4
262 272 #define VDEV_BEST_LABEL VDEV_LABELS
263 273
264 274 #define VDEV_ALLOC_LOAD 0
265 275 #define VDEV_ALLOC_ADD 1
266 276 #define VDEV_ALLOC_SPARE 2
267 277 #define VDEV_ALLOC_L2CACHE 3
268 278 #define VDEV_ALLOC_ROOTPOOL 4
269 279 #define VDEV_ALLOC_SPLIT 5
270 280 #define VDEV_ALLOC_ATTACH 6
271 281
272 282 /*
273 283 * Allocate or free a vdev
274 284 */
275 285 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
276 286 vdev_ops_t *ops);
277 287 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
278 288 vdev_t *parent, uint_t id, int alloctype);
279 289 extern void vdev_free(vdev_t *vd);
280 290
281 291 /*
282 292 * Add or remove children and parents
283 293 */
284 294 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
285 295 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
286 296 extern void vdev_compact_children(vdev_t *pvd);
287 297 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
288 298 extern void vdev_remove_parent(vdev_t *cvd);
289 299
290 300 /*
291 301 * vdev sync load and sync
292 302 */
293 303 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd);
294 304 extern boolean_t vdev_log_state_valid(vdev_t *vd);
295 305 extern void vdev_load(vdev_t *vd);
296 306 extern void vdev_sync(vdev_t *vd, uint64_t txg);
297 307 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
298 308 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
299 309
300 310 /*
301 311 * Available vdev types.
302 312 */
303 313 extern vdev_ops_t vdev_root_ops;
304 314 extern vdev_ops_t vdev_mirror_ops;
305 315 extern vdev_ops_t vdev_replacing_ops;
306 316 extern vdev_ops_t vdev_raidz_ops;
307 317 extern vdev_ops_t vdev_disk_ops;
308 318 extern vdev_ops_t vdev_file_ops;
309 319 extern vdev_ops_t vdev_missing_ops;
310 320 extern vdev_ops_t vdev_hole_ops;
311 321 extern vdev_ops_t vdev_spare_ops;
312 322
313 323 /*
314 324 * Common size functions
315 325 */
316 326 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
317 327 extern uint64_t vdev_get_min_asize(vdev_t *vd);
318 328 extern void vdev_set_min_asize(vdev_t *vd);
319 329
320 330 /*
321 331 * Global variables
322 332 */
323 333 /* zdb uses this tunable, so it must be declared here to make lint happy. */
324 334 extern int zfs_vdev_cache_size;
325 335
326 336 /*
327 337 * The vdev_buf_t is used to translate between zio_t and buf_t, and back again.
328 338 */
329 339 typedef struct vdev_buf {
330 340 buf_t vb_buf; /* buffer that describes the io */
331 341 zio_t *vb_io; /* pointer back to the original zio_t */
332 342 } vdev_buf_t;
333 343
334 344 #ifdef __cplusplus
335 345 }
336 346 #endif
337 347
338 348 #endif /* _SYS_VDEV_IMPL_H */
|
↓ open down ↓ |
221 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX