xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/vdev_impl.h (revision 5cabbc6b49070407fb9610cfe73d4c0e0dea3e77)
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/bpobj.h>
31 #include <sys/dmu.h>
32 #include <sys/metaslab.h>
33 #include <sys/nvpair.h>
34 #include <sys/space_map.h>
35 #include <sys/vdev.h>
36 #include <sys/dkio.h>
37 #include <sys/uberblock_impl.h>
38 #include <sys/vdev_indirect_mapping.h>
39 #include <sys/vdev_indirect_births.h>
40 #include <sys/vdev_removal.h>
41 
42 #ifdef	__cplusplus
43 extern "C" {
44 #endif
45 
46 /*
47  * Virtual device descriptors.
48  *
49  * All storage pool operations go through the virtual device framework,
50  * which provides data replication and I/O scheduling.
51  */
52 
53 /*
54  * Forward declarations that lots of things need.
55  */
56 typedef struct vdev_queue vdev_queue_t;
57 typedef struct vdev_cache vdev_cache_t;
58 typedef struct vdev_cache_entry vdev_cache_entry_t;
59 struct abd;
60 
61 extern int zfs_vdev_queue_depth_pct;
62 extern uint32_t zfs_vdev_async_write_max_active;
63 
64 /*
65  * Virtual device operations
66  */
67 typedef int	vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size,
68     uint64_t *ashift);
69 typedef void	vdev_close_func_t(vdev_t *vd);
70 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize);
71 typedef void	vdev_io_start_func_t(zio_t *zio);
72 typedef void	vdev_io_done_func_t(zio_t *zio);
73 typedef void	vdev_state_change_func_t(vdev_t *vd, int, int);
74 typedef void	vdev_hold_func_t(vdev_t *vd);
75 typedef void	vdev_rele_func_t(vdev_t *vd);
76 
77 typedef void	vdev_remap_cb_t(uint64_t inner_offset, vdev_t *vd,
78     uint64_t offset, uint64_t size, void *arg);
79 typedef void	vdev_remap_func_t(vdev_t *vd, uint64_t offset, uint64_t size,
80     vdev_remap_cb_t callback, void *arg);
81 
82 typedef struct vdev_ops {
83 	vdev_open_func_t		*vdev_op_open;
84 	vdev_close_func_t		*vdev_op_close;
85 	vdev_asize_func_t		*vdev_op_asize;
86 	vdev_io_start_func_t		*vdev_op_io_start;
87 	vdev_io_done_func_t		*vdev_op_io_done;
88 	vdev_state_change_func_t	*vdev_op_state_change;
89 	vdev_hold_func_t		*vdev_op_hold;
90 	vdev_rele_func_t		*vdev_op_rele;
91 	vdev_remap_func_t		*vdev_op_remap;
92 	char				vdev_op_type[16];
93 	boolean_t			vdev_op_leaf;
94 } vdev_ops_t;
95 
96 /*
97  * Virtual device properties
98  */
99 struct vdev_cache_entry {
100 	struct abd	*ve_abd;
101 	uint64_t	ve_offset;
102 	uint64_t	ve_lastused;
103 	avl_node_t	ve_offset_node;
104 	avl_node_t	ve_lastused_node;
105 	uint32_t	ve_hits;
106 	uint16_t	ve_missed_update;
107 	zio_t		*ve_fill_io;
108 };
109 
110 struct vdev_cache {
111 	avl_tree_t	vc_offset_tree;
112 	avl_tree_t	vc_lastused_tree;
113 	kmutex_t	vc_lock;
114 };
115 
116 typedef struct vdev_queue_class {
117 	uint32_t	vqc_active;
118 
119 	/*
120 	 * Sorted by offset or timestamp, depending on if the queue is
121 	 * LBA-ordered vs FIFO.
122 	 */
123 	avl_tree_t	vqc_queued_tree;
124 } vdev_queue_class_t;
125 
126 struct vdev_queue {
127 	vdev_t		*vq_vdev;
128 	vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE];
129 	avl_tree_t	vq_active_tree;
130 	avl_tree_t	vq_read_offset_tree;
131 	avl_tree_t	vq_write_offset_tree;
132 	uint64_t	vq_last_offset;
133 	hrtime_t	vq_io_complete_ts; /* time last i/o completed */
134 	kmutex_t	vq_lock;
135 };
136 
137 /*
138  * On-disk indirect vdev state.
139  *
140  * An indirect vdev is described exclusively in the MOS config of a pool.
141  * The config for an indirect vdev includes several fields, which are
142  * accessed in memory by a vdev_indirect_config_t.
143  */
144 typedef struct vdev_indirect_config {
145 	/*
146 	 * Object (in MOS) which contains the indirect mapping. This object
147 	 * contains an array of vdev_indirect_mapping_entry_phys_t ordered by
148 	 * vimep_src. The bonus buffer for this object is a
149 	 * vdev_indirect_mapping_phys_t. This object is allocated when a vdev
150 	 * removal is initiated.
151 	 *
152 	 * Note that this object can be empty if none of the data on the vdev
153 	 * has been copied yet.
154 	 */
155 	uint64_t	vic_mapping_object;
156 
157 	/*
158 	 * Object (in MOS) which contains the birth times for the mapping
159 	 * entries. This object contains an array of
160 	 * vdev_indirect_birth_entry_phys_t sorted by vibe_offset. The bonus
161 	 * buffer for this object is a vdev_indirect_birth_phys_t. This object
162 	 * is allocated when a vdev removal is initiated.
163 	 *
164 	 * Note that this object can be empty if none of the vdev has yet been
165 	 * copied.
166 	 */
167 	uint64_t	vic_births_object;
168 
169 	/*
170 	 * This is the vdev ID which was removed previous to this vdev, or
171 	 * UINT64_MAX if there are no previously removed vdevs.
172 	 */
173 	uint64_t	vic_prev_indirect_vdev;
174 } vdev_indirect_config_t;
175 
176 /*
177  * Virtual device descriptor
178  */
179 struct vdev {
180 	/*
181 	 * Common to all vdev types.
182 	 */
183 	uint64_t	vdev_id;	/* child number in vdev parent	*/
184 	uint64_t	vdev_guid;	/* unique ID for this vdev	*/
185 	uint64_t	vdev_guid_sum;	/* self guid + all child guids	*/
186 	uint64_t	vdev_orig_guid;	/* orig. guid prior to remove	*/
187 	uint64_t	vdev_asize;	/* allocatable device capacity	*/
188 	uint64_t	vdev_min_asize;	/* min acceptable asize		*/
189 	uint64_t	vdev_max_asize;	/* max acceptable asize		*/
190 	uint64_t	vdev_ashift;	/* block alignment shift	*/
191 	uint64_t	vdev_state;	/* see VDEV_STATE_* #defines	*/
192 	uint64_t	vdev_prevstate;	/* used when reopening a vdev	*/
193 	vdev_ops_t	*vdev_ops;	/* vdev operations		*/
194 	spa_t		*vdev_spa;	/* spa for this vdev		*/
195 	void		*vdev_tsd;	/* type-specific data		*/
196 	vnode_t		*vdev_name_vp;	/* vnode for pathname		*/
197 	vnode_t		*vdev_devid_vp;	/* vnode for devid		*/
198 	vdev_t		*vdev_top;	/* top-level vdev		*/
199 	vdev_t		*vdev_parent;	/* parent vdev			*/
200 	vdev_t		**vdev_child;	/* array of children		*/
201 	uint64_t	vdev_children;	/* number of children		*/
202 	vdev_stat_t	vdev_stat;	/* virtual device statistics	*/
203 	boolean_t	vdev_expanding;	/* expand the vdev?		*/
204 	boolean_t	vdev_reopening;	/* reopen in progress?		*/
205 	int		vdev_open_error; /* error on last open		*/
206 	kthread_t	*vdev_open_thread; /* thread opening children	*/
207 	uint64_t	vdev_crtxg;	/* txg when top-level was added */
208 
209 	/*
210 	 * Top-level vdev state.
211 	 */
212 	uint64_t	vdev_ms_array;	/* metaslab array object	*/
213 	uint64_t	vdev_ms_shift;	/* metaslab size shift		*/
214 	uint64_t	vdev_ms_count;	/* number of metaslabs		*/
215 	metaslab_group_t *vdev_mg;	/* metaslab group		*/
216 	metaslab_t	**vdev_ms;	/* metaslab array		*/
217 	txg_list_t	vdev_ms_list;	/* per-txg dirty metaslab lists	*/
218 	txg_list_t	vdev_dtl_list;	/* per-txg dirty DTL lists	*/
219 	txg_node_t	vdev_txg_node;	/* per-txg dirty vdev linkage	*/
220 	boolean_t	vdev_remove_wanted; /* async remove wanted?	*/
221 	boolean_t	vdev_probe_wanted; /* async probe wanted?	*/
222 	list_node_t	vdev_config_dirty_node; /* config dirty list	*/
223 	list_node_t	vdev_state_dirty_node; /* state dirty list	*/
224 	uint64_t	vdev_deflate_ratio; /* deflation ratio (x512)	*/
225 	uint64_t	vdev_islog;	/* is an intent log device	*/
226 	uint64_t	vdev_removing;	/* device is being removed?	*/
227 	boolean_t	vdev_ishole;	/* is a hole in the namespace	*/
228 	kmutex_t	vdev_queue_lock; /* protects vdev_queue_depth	*/
229 	uint64_t	vdev_top_zap;
230 
231 	/*
232 	 * Values stored in the config for an indirect or removing vdev.
233 	 */
234 	vdev_indirect_config_t	vdev_indirect_config;
235 
236 	/*
237 	 * The vdev_indirect_rwlock protects the vdev_indirect_mapping
238 	 * pointer from changing on indirect vdevs (when it is condensed).
239 	 * Note that removing (not yet indirect) vdevs have different
240 	 * access patterns (the mapping is not accessed from open context,
241 	 * e.g. from zio_read) and locking strategy (e.g. svr_lock).
242 	 */
243 	krwlock_t vdev_indirect_rwlock;
244 	vdev_indirect_mapping_t *vdev_indirect_mapping;
245 	vdev_indirect_births_t *vdev_indirect_births;
246 
247 	/*
248 	 * In memory data structures used to manage the obsolete sm, for
249 	 * indirect or removing vdevs.
250 	 *
251 	 * The vdev_obsolete_segments is the in-core record of the segments
252 	 * that are no longer referenced anywhere in the pool (due to
253 	 * being freed or remapped and not referenced by any snapshots).
254 	 * During a sync, segments are added to vdev_obsolete_segments
255 	 * via vdev_indirect_mark_obsolete(); at the end of each sync
256 	 * pass, this is appended to vdev_obsolete_sm via
257 	 * vdev_indirect_sync_obsolete().  The vdev_obsolete_lock
258 	 * protects against concurrent modifications of vdev_obsolete_segments
259 	 * from multiple zio threads.
260 	 */
261 	kmutex_t	vdev_obsolete_lock;
262 	range_tree_t	*vdev_obsolete_segments;
263 	space_map_t	*vdev_obsolete_sm;
264 
265 	/*
266 	 * The queue depth parameters determine how many async writes are
267 	 * still pending (i.e. allocated by net yet issued to disk) per
268 	 * top-level (vdev_async_write_queue_depth) and the maximum allowed
269 	 * (vdev_max_async_write_queue_depth). These values only apply to
270 	 * top-level vdevs.
271 	 */
272 	uint64_t	vdev_async_write_queue_depth;
273 	uint64_t	vdev_max_async_write_queue_depth;
274 
275 	/*
276 	 * Leaf vdev state.
277 	 */
278 	range_tree_t	*vdev_dtl[DTL_TYPES]; /* dirty time logs	*/
279 	space_map_t	*vdev_dtl_sm;	/* dirty time log space map	*/
280 	txg_node_t	vdev_dtl_node;	/* per-txg dirty DTL linkage	*/
281 	uint64_t	vdev_dtl_object; /* DTL object			*/
282 	uint64_t	vdev_psize;	/* physical device capacity	*/
283 	uint64_t	vdev_wholedisk;	/* true if this is a whole disk */
284 	uint64_t	vdev_offline;	/* persistent offline state	*/
285 	uint64_t	vdev_faulted;	/* persistent faulted state	*/
286 	uint64_t	vdev_degraded;	/* persistent degraded state	*/
287 	uint64_t	vdev_removed;	/* persistent removed state	*/
288 	uint64_t	vdev_resilver_txg; /* persistent resilvering state */
289 	uint64_t	vdev_nparity;	/* number of parity devices for raidz */
290 	char		*vdev_path;	/* vdev path (if any)		*/
291 	char		*vdev_devid;	/* vdev devid (if any)		*/
292 	char		*vdev_physpath;	/* vdev device path (if any)	*/
293 	char		*vdev_fru;	/* physical FRU location	*/
294 	uint64_t	vdev_not_present; /* not present during import	*/
295 	uint64_t	vdev_unspare;	/* unspare when resilvering done */
296 	boolean_t	vdev_nowritecache; /* true if flushwritecache failed */
297 	boolean_t	vdev_checkremove; /* temporary online test	*/
298 	boolean_t	vdev_forcefault; /* force online fault		*/
299 	boolean_t	vdev_splitting;	/* split or repair in progress  */
300 	boolean_t	vdev_delayed_close; /* delayed device close?	*/
301 	boolean_t	vdev_tmpoffline; /* device taken offline temporarily? */
302 	boolean_t	vdev_detached;	/* device detached?		*/
303 	boolean_t	vdev_cant_read;	/* vdev is failing all reads	*/
304 	boolean_t	vdev_cant_write; /* vdev is failing all writes	*/
305 	boolean_t	vdev_isspare;	/* was a hot spare		*/
306 	boolean_t	vdev_isl2cache;	/* was a l2cache device		*/
307 	vdev_queue_t	vdev_queue;	/* I/O deadline schedule queue	*/
308 	vdev_cache_t	vdev_cache;	/* physical block cache		*/
309 	spa_aux_vdev_t	*vdev_aux;	/* for l2cache and spares vdevs	*/
310 	zio_t		*vdev_probe_zio; /* root of current probe	*/
311 	vdev_aux_t	vdev_label_aux;	/* on-disk aux state		*/
312 	uint64_t	vdev_leaf_zap;
313 
314 	/*
315 	 * For DTrace to work in userland (libzpool) context, these fields must
316 	 * remain at the end of the structure.  DTrace will use the kernel's
317 	 * CTF definition for 'struct vdev', and since the size of a kmutex_t is
318 	 * larger in userland, the offsets for the rest of the fields would be
319 	 * incorrect.
320 	 */
321 	kmutex_t	vdev_dtl_lock;	/* vdev_dtl_{map,resilver}	*/
322 	kmutex_t	vdev_stat_lock;	/* vdev_stat			*/
323 	kmutex_t	vdev_probe_lock; /* protects vdev_probe_zio	*/
324 };
325 
326 #define	VDEV_RAIDZ_MAXPARITY	3
327 
328 #define	VDEV_PAD_SIZE		(8 << 10)
329 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
330 #define	VDEV_SKIP_SIZE		VDEV_PAD_SIZE * 2
331 #define	VDEV_PHYS_SIZE		(112 << 10)
332 #define	VDEV_UBERBLOCK_RING	(128 << 10)
333 
334 /* The largest uberblock we support is 8k. */
335 #define	MAX_UBERBLOCK_SHIFT (13)
336 #define	VDEV_UBERBLOCK_SHIFT(vd)	\
337 	MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \
338 	    MAX_UBERBLOCK_SHIFT)
339 #define	VDEV_UBERBLOCK_COUNT(vd)	\
340 	(VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
341 #define	VDEV_UBERBLOCK_OFFSET(vd, n)	\
342 	offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
343 #define	VDEV_UBERBLOCK_SIZE(vd)		(1ULL << VDEV_UBERBLOCK_SHIFT(vd))
344 
345 typedef struct vdev_phys {
346 	char		vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
347 	zio_eck_t	vp_zbt;
348 } vdev_phys_t;
349 
350 typedef struct vdev_label {
351 	char		vl_pad1[VDEV_PAD_SIZE];			/*  8K */
352 	char		vl_pad2[VDEV_PAD_SIZE];			/*  8K */
353 	vdev_phys_t	vl_vdev_phys;				/* 112K	*/
354 	char		vl_uberblock[VDEV_UBERBLOCK_RING];	/* 128K	*/
355 } vdev_label_t;							/* 256K total */
356 
357 /*
358  * vdev_dirty() flags
359  */
360 #define	VDD_METASLAB	0x01
361 #define	VDD_DTL		0x02
362 
363 /* Offset of embedded boot loader region on each label */
364 #define	VDEV_BOOT_OFFSET	(2 * sizeof (vdev_label_t))
365 /*
366  * Size of embedded boot loader region on each label.
367  * The total size of the first two labels plus the boot area is 4MB.
368  */
369 #define	VDEV_BOOT_SIZE		(7ULL << 19)			/* 3.5M */
370 
371 /*
372  * Size of label regions at the start and end of each leaf device.
373  */
374 #define	VDEV_LABEL_START_SIZE	(2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
375 #define	VDEV_LABEL_END_SIZE	(2 * sizeof (vdev_label_t))
376 #define	VDEV_LABELS		4
377 #define	VDEV_BEST_LABEL		VDEV_LABELS
378 
379 #define	VDEV_ALLOC_LOAD		0
380 #define	VDEV_ALLOC_ADD		1
381 #define	VDEV_ALLOC_SPARE	2
382 #define	VDEV_ALLOC_L2CACHE	3
383 #define	VDEV_ALLOC_ROOTPOOL	4
384 #define	VDEV_ALLOC_SPLIT	5
385 #define	VDEV_ALLOC_ATTACH	6
386 
387 /*
388  * Allocate or free a vdev
389  */
390 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
391     vdev_ops_t *ops);
392 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
393     vdev_t *parent, uint_t id, int alloctype);
394 extern void vdev_free(vdev_t *vd);
395 
396 /*
397  * Add or remove children and parents
398  */
399 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
400 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
401 extern void vdev_compact_children(vdev_t *pvd);
402 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
403 extern void vdev_remove_parent(vdev_t *cvd);
404 
405 /*
406  * vdev sync load and sync
407  */
408 extern void vdev_load_log_state(vdev_t *nvd, vdev_t *ovd);
409 extern boolean_t vdev_log_state_valid(vdev_t *vd);
410 extern int vdev_load(vdev_t *vd);
411 extern int vdev_dtl_load(vdev_t *vd);
412 extern void vdev_sync(vdev_t *vd, uint64_t txg);
413 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
414 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
415 extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg);
416 
417 /*
418  * Available vdev types.
419  */
420 extern vdev_ops_t vdev_root_ops;
421 extern vdev_ops_t vdev_mirror_ops;
422 extern vdev_ops_t vdev_replacing_ops;
423 extern vdev_ops_t vdev_raidz_ops;
424 extern vdev_ops_t vdev_disk_ops;
425 extern vdev_ops_t vdev_file_ops;
426 extern vdev_ops_t vdev_missing_ops;
427 extern vdev_ops_t vdev_hole_ops;
428 extern vdev_ops_t vdev_spare_ops;
429 extern vdev_ops_t vdev_indirect_ops;
430 
431 /*
432  * Common size functions
433  */
434 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
435 extern uint64_t vdev_get_min_asize(vdev_t *vd);
436 extern void vdev_set_min_asize(vdev_t *vd);
437 
438 /*
439  * Global variables
440  */
441 /* zdb uses this tunable, so it must be declared here to make lint happy. */
442 extern int zfs_vdev_cache_size;
443 
444 /*
445  * Functions from vdev_indirect.c
446  */
447 extern void vdev_indirect_sync_obsolete(vdev_t *vd, dmu_tx_t *tx);
448 extern boolean_t vdev_indirect_should_condense(vdev_t *vd);
449 extern void spa_condense_indirect_start_sync(vdev_t *vd, dmu_tx_t *tx);
450 extern int vdev_obsolete_sm_object(vdev_t *vd);
451 extern boolean_t vdev_obsolete_counts_are_precise(vdev_t *vd);
452 
453 /*
454  * The vdev_buf_t is used to translate between zio_t and buf_t, and back again.
455  */
456 typedef struct vdev_buf {
457 	buf_t	vb_buf;		/* buffer that describes the io */
458 	zio_t	*vb_io;		/* pointer back to the original zio_t */
459 } vdev_buf_t;
460 
461 #ifdef	__cplusplus
462 }
463 #endif
464 
465 #endif	/* _SYS_VDEV_IMPL_H */
466