]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/vdev_impl.h
MFV r329793, r329795:
[FreeBSD/FreeBSD.git] / sys / cddl / contrib / opensolaris / uts / common / fs / zfs / sys / vdev_impl.h
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 *logical_ashift, uint64_t *physical_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         uint64_t        vq_lastoffset;
136 };
137
138 /*
139  * On-disk indirect vdev state.
140  *
141  * An indirect vdev is described exclusively in the MOS config of a pool.
142  * The config for an indirect vdev includes several fields, which are
143  * accessed in memory by a vdev_indirect_config_t.
144  */
145 typedef struct vdev_indirect_config {
146         /*
147          * Object (in MOS) which contains the indirect mapping. This object
148          * contains an array of vdev_indirect_mapping_entry_phys_t ordered by
149          * vimep_src. The bonus buffer for this object is a
150          * vdev_indirect_mapping_phys_t. This object is allocated when a vdev
151          * removal is initiated.
152          *
153          * Note that this object can be empty if none of the data on the vdev
154          * has been copied yet.
155          */
156         uint64_t        vic_mapping_object;
157
158         /*
159          * Object (in MOS) which contains the birth times for the mapping
160          * entries. This object contains an array of
161          * vdev_indirect_birth_entry_phys_t sorted by vibe_offset. The bonus
162          * buffer for this object is a vdev_indirect_birth_phys_t. This object
163          * is allocated when a vdev removal is initiated.
164          *
165          * Note that this object can be empty if none of the vdev has yet been
166          * copied.
167          */
168         uint64_t        vic_births_object;
169
170         /*
171          * This is the vdev ID which was removed previous to this vdev, or
172          * UINT64_MAX if there are no previously removed vdevs.
173          */
174         uint64_t        vic_prev_indirect_vdev;
175 } vdev_indirect_config_t;
176
177 /*
178  * Virtual device descriptor
179  */
180 struct vdev {
181         /*
182          * Common to all vdev types.
183          */
184         uint64_t        vdev_id;        /* child number in vdev parent  */
185         uint64_t        vdev_guid;      /* unique ID for this vdev      */
186         uint64_t        vdev_guid_sum;  /* self guid + all child guids  */
187         uint64_t        vdev_orig_guid; /* orig. guid prior to remove   */
188         uint64_t        vdev_asize;     /* allocatable device capacity  */
189         uint64_t        vdev_min_asize; /* min acceptable asize         */
190         uint64_t        vdev_max_asize; /* max acceptable asize         */
191         uint64_t        vdev_ashift;    /* block alignment shift        */
192         /*
193          * Logical block alignment shift
194          *
195          * The smallest sized/aligned I/O supported by the device.
196          */
197         uint64_t        vdev_logical_ashift;
198         /*
199          * Physical block alignment shift
200          *
201          * The device supports logical I/Os with vdev_logical_ashift
202          * size/alignment, but optimum performance will be achieved by
203          * aligning/sizing requests to vdev_physical_ashift.  Smaller
204          * requests may be inflated or incur device level read-modify-write
205          * operations.
206          *
207          * May be 0 to indicate no preference (i.e. use vdev_logical_ashift).
208          */
209         uint64_t        vdev_physical_ashift;
210         uint64_t        vdev_state;     /* see VDEV_STATE_* #defines    */
211         uint64_t        vdev_prevstate; /* used when reopening a vdev   */
212         vdev_ops_t      *vdev_ops;      /* vdev operations              */
213         spa_t           *vdev_spa;      /* spa for this vdev            */
214         void            *vdev_tsd;      /* type-specific data           */
215         vnode_t         *vdev_name_vp;  /* vnode for pathname           */
216         vnode_t         *vdev_devid_vp; /* vnode for devid              */
217         vdev_t          *vdev_top;      /* top-level vdev               */
218         vdev_t          *vdev_parent;   /* parent vdev                  */
219         vdev_t          **vdev_child;   /* array of children            */
220         uint64_t        vdev_children;  /* number of children           */
221         vdev_stat_t     vdev_stat;      /* virtual device statistics    */
222         boolean_t       vdev_expanding; /* expand the vdev?             */
223         boolean_t       vdev_reopening; /* reopen in progress?          */
224         int             vdev_open_error; /* error on last open          */
225         kthread_t       *vdev_open_thread; /* thread opening children   */
226         uint64_t        vdev_crtxg;     /* txg when top-level was added */
227
228         /*
229          * Top-level vdev state.
230          */
231         uint64_t        vdev_ms_array;  /* metaslab array object        */
232         uint64_t        vdev_ms_shift;  /* metaslab size shift          */
233         uint64_t        vdev_ms_count;  /* number of metaslabs          */
234         metaslab_group_t *vdev_mg;      /* metaslab group               */
235         metaslab_t      **vdev_ms;      /* metaslab array               */
236         txg_list_t      vdev_ms_list;   /* per-txg dirty metaslab lists */
237         txg_list_t      vdev_dtl_list;  /* per-txg dirty DTL lists      */
238         txg_node_t      vdev_txg_node;  /* per-txg dirty vdev linkage   */
239         boolean_t       vdev_remove_wanted; /* async remove wanted?     */
240         boolean_t       vdev_probe_wanted; /* async probe wanted?       */
241         list_node_t     vdev_config_dirty_node; /* config dirty list    */
242         list_node_t     vdev_state_dirty_node; /* state dirty list      */
243         uint64_t        vdev_deflate_ratio; /* deflation ratio (x512)   */
244         uint64_t        vdev_islog;     /* is an intent log device      */
245         uint64_t        vdev_removing;  /* device is being removed?     */
246         boolean_t       vdev_ishole;    /* is a hole in the namespace   */
247         kmutex_t        vdev_queue_lock; /* protects vdev_queue_depth   */
248         uint64_t        vdev_top_zap;
249
250         /*
251          * Values stored in the config for an indirect or removing vdev.
252          */
253         vdev_indirect_config_t  vdev_indirect_config;
254
255         /*
256          * The vdev_indirect_rwlock protects the vdev_indirect_mapping
257          * pointer from changing on indirect vdevs (when it is condensed).
258          * Note that removing (not yet indirect) vdevs have different
259          * access patterns (the mapping is not accessed from open context,
260          * e.g. from zio_read) and locking strategy (e.g. svr_lock).
261          */
262         krwlock_t vdev_indirect_rwlock;
263         vdev_indirect_mapping_t *vdev_indirect_mapping;
264         vdev_indirect_births_t *vdev_indirect_births;
265
266         /*
267          * In memory data structures used to manage the obsolete sm, for
268          * indirect or removing vdevs.
269          *
270          * The vdev_obsolete_segments is the in-core record of the segments
271          * that are no longer referenced anywhere in the pool (due to
272          * being freed or remapped and not referenced by any snapshots).
273          * During a sync, segments are added to vdev_obsolete_segments
274          * via vdev_indirect_mark_obsolete(); at the end of each sync
275          * pass, this is appended to vdev_obsolete_sm via
276          * vdev_indirect_sync_obsolete().  The vdev_obsolete_lock
277          * protects against concurrent modifications of vdev_obsolete_segments
278          * from multiple zio threads.
279          */
280         kmutex_t        vdev_obsolete_lock;
281         range_tree_t    *vdev_obsolete_segments;
282         space_map_t     *vdev_obsolete_sm;
283
284         /*
285          * The queue depth parameters determine how many async writes are
286          * still pending (i.e. allocated by net yet issued to disk) per
287          * top-level (vdev_async_write_queue_depth) and the maximum allowed
288          * (vdev_max_async_write_queue_depth). These values only apply to
289          * top-level vdevs.
290          */
291         uint64_t        vdev_async_write_queue_depth;
292         uint64_t        vdev_max_async_write_queue_depth;
293
294         /*
295          * Leaf vdev state.
296          */
297         range_tree_t    *vdev_dtl[DTL_TYPES]; /* dirty time logs        */
298         space_map_t     *vdev_dtl_sm;   /* dirty time log space map     */
299         txg_node_t      vdev_dtl_node;  /* per-txg dirty DTL linkage    */
300         uint64_t        vdev_dtl_object; /* DTL object                  */
301         uint64_t        vdev_psize;     /* physical device capacity     */
302         uint64_t        vdev_wholedisk; /* true if this is a whole disk */
303         uint64_t        vdev_offline;   /* persistent offline state     */
304         uint64_t        vdev_faulted;   /* persistent faulted state     */
305         uint64_t        vdev_degraded;  /* persistent degraded state    */
306         uint64_t        vdev_removed;   /* persistent removed state     */
307         uint64_t        vdev_resilver_txg; /* persistent resilvering state */
308         uint64_t        vdev_nparity;   /* number of parity devices for raidz */
309         char            *vdev_path;     /* vdev path (if any)           */
310         char            *vdev_devid;    /* vdev devid (if any)          */
311         char            *vdev_physpath; /* vdev device path (if any)    */
312         char            *vdev_fru;      /* physical FRU location        */
313         uint64_t        vdev_not_present; /* not present during import  */
314         uint64_t        vdev_unspare;   /* unspare when resilvering done */
315         boolean_t       vdev_nowritecache; /* true if flushwritecache failed */
316         boolean_t       vdev_notrim;    /* true if trim failed */
317         boolean_t       vdev_checkremove; /* temporary online test      */
318         boolean_t       vdev_forcefault; /* force online fault          */
319         boolean_t       vdev_splitting; /* split or repair in progress  */
320         boolean_t       vdev_delayed_close; /* delayed device close?    */
321         boolean_t       vdev_tmpoffline; /* device taken offline temporarily? */
322         boolean_t       vdev_detached;  /* device detached?             */
323         boolean_t       vdev_cant_read; /* vdev is failing all reads    */
324         boolean_t       vdev_cant_write; /* vdev is failing all writes  */
325         boolean_t       vdev_isspare;   /* was a hot spare              */
326         boolean_t       vdev_isl2cache; /* was a l2cache device         */
327         vdev_queue_t    vdev_queue;     /* I/O deadline schedule queue  */
328         vdev_cache_t    vdev_cache;     /* physical block cache         */
329         spa_aux_vdev_t  *vdev_aux;      /* for l2cache and spares vdevs */
330         zio_t           *vdev_probe_zio; /* root of current probe       */
331         vdev_aux_t      vdev_label_aux; /* on-disk aux state            */
332         struct trim_map *vdev_trimmap;  /* map on outstanding trims     */ 
333         uint16_t        vdev_rotation_rate; /* rotational rate of the media */
334 #define VDEV_RATE_UNKNOWN       0
335 #define VDEV_RATE_NON_ROTATING  1
336         uint64_t        vdev_leaf_zap;
337
338         /*
339          * For DTrace to work in userland (libzpool) context, these fields must
340          * remain at the end of the structure.  DTrace will use the kernel's
341          * CTF definition for 'struct vdev', and since the size of a kmutex_t is
342          * larger in userland, the offsets for the rest of the fields would be
343          * incorrect.
344          */
345         kmutex_t        vdev_dtl_lock;  /* vdev_dtl_{map,resilver}      */
346         kmutex_t        vdev_stat_lock; /* vdev_stat                    */
347         kmutex_t        vdev_probe_lock; /* protects vdev_probe_zio     */
348 };
349
350 #define VDEV_RAIDZ_MAXPARITY    3
351
352 #define VDEV_PAD_SIZE           (8 << 10)
353 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
354 #define VDEV_SKIP_SIZE          VDEV_PAD_SIZE * 2
355 #define VDEV_PHYS_SIZE          (112 << 10)
356 #define VDEV_UBERBLOCK_RING     (128 << 10)
357
358 /* The largest uberblock we support is 8k. */
359 #define MAX_UBERBLOCK_SHIFT (13)
360 #define VDEV_UBERBLOCK_SHIFT(vd)        \
361         MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \
362             MAX_UBERBLOCK_SHIFT)
363 #define VDEV_UBERBLOCK_COUNT(vd)        \
364         (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
365 #define VDEV_UBERBLOCK_OFFSET(vd, n)    \
366         offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
367 #define VDEV_UBERBLOCK_SIZE(vd)         (1ULL << VDEV_UBERBLOCK_SHIFT(vd))
368
369 typedef struct vdev_phys {
370         char            vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
371         zio_eck_t       vp_zbt;
372 } vdev_phys_t;
373
374 typedef struct vdev_label {
375         char            vl_pad1[VDEV_PAD_SIZE];                 /*  8K */
376         char            vl_pad2[VDEV_PAD_SIZE];                 /*  8K */
377         vdev_phys_t     vl_vdev_phys;                           /* 112K */
378         char            vl_uberblock[VDEV_UBERBLOCK_RING];      /* 128K */
379 } vdev_label_t;                                                 /* 256K total */
380
381 /*
382  * vdev_dirty() flags
383  */
384 #define VDD_METASLAB    0x01
385 #define VDD_DTL         0x02
386
387 /* Offset of embedded boot loader region on each label */
388 #define VDEV_BOOT_OFFSET        (2 * sizeof (vdev_label_t))
389 /*
390  * Size of embedded boot loader region on each label.
391  * The total size of the first two labels plus the boot area is 4MB.
392  */
393 #define VDEV_BOOT_SIZE          (7ULL << 19)                    /* 3.5M */
394
395 /*
396  * Size of label regions at the start and end of each leaf device.
397  */
398 #define VDEV_LABEL_START_SIZE   (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
399 #define VDEV_LABEL_END_SIZE     (2 * sizeof (vdev_label_t))
400 #define VDEV_LABELS             4
401 #define VDEV_BEST_LABEL         VDEV_LABELS
402
403 #define VDEV_ALLOC_LOAD         0
404 #define VDEV_ALLOC_ADD          1
405 #define VDEV_ALLOC_SPARE        2
406 #define VDEV_ALLOC_L2CACHE      3
407 #define VDEV_ALLOC_ROOTPOOL     4
408 #define VDEV_ALLOC_SPLIT        5
409 #define VDEV_ALLOC_ATTACH       6
410
411 /*
412  * Allocate or free a vdev
413  */
414 extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid,
415     vdev_ops_t *ops);
416 extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config,
417     vdev_t *parent, uint_t id, int alloctype);
418 extern void vdev_free(vdev_t *vd);
419
420 /*
421  * Add or remove children and parents
422  */
423 extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd);
424 extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd);
425 extern void vdev_compact_children(vdev_t *pvd);
426 extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops);
427 extern void vdev_remove_parent(vdev_t *cvd);
428
429 /*
430  * vdev sync load and sync
431  */
432 extern boolean_t vdev_log_state_valid(vdev_t *vd);
433 extern int vdev_load(vdev_t *vd);
434 extern int vdev_dtl_load(vdev_t *vd);
435 extern void vdev_sync(vdev_t *vd, uint64_t txg);
436 extern void vdev_sync_done(vdev_t *vd, uint64_t txg);
437 extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg);
438 extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg);
439
440 /*
441  * Available vdev types.
442  */
443 extern vdev_ops_t vdev_root_ops;
444 extern vdev_ops_t vdev_mirror_ops;
445 extern vdev_ops_t vdev_replacing_ops;
446 extern vdev_ops_t vdev_raidz_ops;
447 #ifdef _KERNEL
448 extern vdev_ops_t vdev_geom_ops;
449 #else
450 extern vdev_ops_t vdev_disk_ops;
451 #endif
452 extern vdev_ops_t vdev_file_ops;
453 extern vdev_ops_t vdev_missing_ops;
454 extern vdev_ops_t vdev_hole_ops;
455 extern vdev_ops_t vdev_spare_ops;
456 extern vdev_ops_t vdev_indirect_ops;
457
458 /*
459  * Common size functions
460  */
461 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize);
462 extern uint64_t vdev_get_min_asize(vdev_t *vd);
463 extern void vdev_set_min_asize(vdev_t *vd);
464
465 /*
466  * Global variables
467  */
468 /* zdb uses this tunable, so it must be declared here to make lint happy. */
469 extern int zfs_vdev_cache_size;
470 extern uint_t zfs_geom_probe_vdev_key;
471
472 /*
473  * Functions from vdev_indirect.c
474  */
475 extern void vdev_indirect_sync_obsolete(vdev_t *vd, dmu_tx_t *tx);
476 extern boolean_t vdev_indirect_should_condense(vdev_t *vd);
477 extern void spa_condense_indirect_start_sync(vdev_t *vd, dmu_tx_t *tx);
478 extern int vdev_obsolete_sm_object(vdev_t *vd);
479 extern boolean_t vdev_obsolete_counts_are_precise(vdev_t *vd);
480
481 #ifdef illumos
482 /*
483  * The vdev_buf_t is used to translate between zio_t and buf_t, and back again.
484  */
485 typedef struct vdev_buf {
486         buf_t   vb_buf;         /* buffer that describes the io */
487         zio_t   *vb_io;         /* pointer back to the original zio_t */
488 } vdev_buf_t;
489 #endif
490
491 #ifdef  __cplusplus
492 }
493 #endif
494
495 #endif  /* _SYS_VDEV_IMPL_H */