xref: /freebsd/sys/cddl/boot/zfs/zfsimpl.h (revision 9a14aa017b21c292740c00ee098195cd46642730)
1 /*-
2  * Copyright (c) 2002 McAfee, Inc.
3  * All rights reserved.
4  *
5  * This software was developed for the FreeBSD Project by Marshall
6  * Kirk McKusick and McAfee Research,, the Security Research Division of
7  * McAfee, Inc. under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as
8  * part of the DARPA CHATS research program
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 /*
32  * CDDL HEADER START
33  *
34  * The contents of this file are subject to the terms of the
35  * Common Development and Distribution License (the "License").
36  * You may not use this file except in compliance with the License.
37  *
38  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
39  * or http://www.opensolaris.org/os/licensing.
40  * See the License for the specific language governing permissions
41  * and limitations under the License.
42  *
43  * When distributing Covered Code, include this CDDL HEADER in each
44  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
45  * If applicable, add the following below this CDDL HEADER, with the
46  * fields enclosed by brackets "[]" replaced with your own identifying
47  * information: Portions Copyright [yyyy] [name of copyright owner]
48  *
49  * CDDL HEADER END
50  */
51 /*
52  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
53  * Use is subject to license terms.
54  */
55 
56 /* CRC64 table */
57 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
58 
59 /*
60  * Macros for various sorts of alignment and rounding when the alignment
61  * is known to be a power of 2.
62  */
63 #define	P2ALIGN(x, align)		((x) & -(align))
64 #define	P2PHASE(x, align)		((x) & ((align) - 1))
65 #define	P2NPHASE(x, align)		(-(x) & ((align) - 1))
66 #define	P2ROUNDUP(x, align)		(-(-(x) & -(align)))
67 #define	P2END(x, align)			(-(~(x) & -(align)))
68 #define	P2PHASEUP(x, align, phase)	((phase) - (((phase) - (x)) & -(align)))
69 #define	P2BOUNDARY(off, len, align)	(((off) ^ ((off) + (len) - 1)) > (align) - 1)
70 
71 /*
72  * General-purpose 32-bit and 64-bit bitfield encodings.
73  */
74 #define	BF32_DECODE(x, low, len)	P2PHASE((x) >> (low), 1U << (len))
75 #define	BF64_DECODE(x, low, len)	P2PHASE((x) >> (low), 1ULL << (len))
76 #define	BF32_ENCODE(x, low, len)	(P2PHASE((x), 1U << (len)) << (low))
77 #define	BF64_ENCODE(x, low, len)	(P2PHASE((x), 1ULL << (len)) << (low))
78 
79 #define	BF32_GET(x, low, len)		BF32_DECODE(x, low, len)
80 #define	BF64_GET(x, low, len)		BF64_DECODE(x, low, len)
81 
82 #define	BF32_SET(x, low, len, val)	\
83 	((x) ^= BF32_ENCODE((x >> low) ^ (val), low, len))
84 #define	BF64_SET(x, low, len, val)	\
85 	((x) ^= BF64_ENCODE((x >> low) ^ (val), low, len))
86 
87 #define	BF32_GET_SB(x, low, len, shift, bias)	\
88 	((BF32_GET(x, low, len) + (bias)) << (shift))
89 #define	BF64_GET_SB(x, low, len, shift, bias)	\
90 	((BF64_GET(x, low, len) + (bias)) << (shift))
91 
92 #define	BF32_SET_SB(x, low, len, shift, bias, val)	\
93 	BF32_SET(x, low, len, ((val) >> (shift)) - (bias))
94 #define	BF64_SET_SB(x, low, len, shift, bias, val)	\
95 	BF64_SET(x, low, len, ((val) >> (shift)) - (bias))
96 
97 /*
98  * Macros to reverse byte order
99  */
100 #define	BSWAP_8(x)	((x) & 0xff)
101 #define	BSWAP_16(x)	((BSWAP_8(x) << 8) | BSWAP_8((x) >> 8))
102 #define	BSWAP_32(x)	((BSWAP_16(x) << 16) | BSWAP_16((x) >> 16))
103 #define	BSWAP_64(x)	((BSWAP_32(x) << 32) | BSWAP_32((x) >> 32))
104 
105 /*
106  * We currently support nine block sizes, from 512 bytes to 128K.
107  * We could go higher, but the benefits are near-zero and the cost
108  * of COWing a giant block to modify one byte would become excessive.
109  */
110 #define	SPA_MINBLOCKSHIFT	9
111 #define	SPA_MAXBLOCKSHIFT	17
112 #define	SPA_MINBLOCKSIZE	(1ULL << SPA_MINBLOCKSHIFT)
113 #define	SPA_MAXBLOCKSIZE	(1ULL << SPA_MAXBLOCKSHIFT)
114 
115 #define	SPA_BLOCKSIZES		(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)
116 
117 /*
118  * The DVA size encodings for LSIZE and PSIZE support blocks up to 32MB.
119  * The ASIZE encoding should be at least 64 times larger (6 more bits)
120  * to support up to 4-way RAID-Z mirror mode with worst-case gang block
121  * overhead, three DVAs per bp, plus one more bit in case we do anything
122  * else that expands the ASIZE.
123  */
124 #define	SPA_LSIZEBITS		16	/* LSIZE up to 32M (2^16 * 512)	*/
125 #define	SPA_PSIZEBITS		16	/* PSIZE up to 32M (2^16 * 512)	*/
126 #define	SPA_ASIZEBITS		24	/* ASIZE up to 64 times larger	*/
127 
128 /*
129  * All SPA data is represented by 128-bit data virtual addresses (DVAs).
130  * The members of the dva_t should be considered opaque outside the SPA.
131  */
132 typedef struct dva {
133 	uint64_t	dva_word[2];
134 } dva_t;
135 
136 /*
137  * Each block has a 256-bit checksum -- strong enough for cryptographic hashes.
138  */
139 typedef struct zio_cksum {
140 	uint64_t	zc_word[4];
141 } zio_cksum_t;
142 
143 /*
144  * Each block is described by its DVAs, time of birth, checksum, etc.
145  * The word-by-word, bit-by-bit layout of the blkptr is as follows:
146  *
147  *	64	56	48	40	32	24	16	8	0
148  *	+-------+-------+-------+-------+-------+-------+-------+-------+
149  * 0	|		vdev1		| GRID  |	  ASIZE		|
150  *	+-------+-------+-------+-------+-------+-------+-------+-------+
151  * 1	|G|			 offset1				|
152  *	+-------+-------+-------+-------+-------+-------+-------+-------+
153  * 2	|		vdev2		| GRID  |	  ASIZE		|
154  *	+-------+-------+-------+-------+-------+-------+-------+-------+
155  * 3	|G|			 offset2				|
156  *	+-------+-------+-------+-------+-------+-------+-------+-------+
157  * 4	|		vdev3		| GRID  |	  ASIZE		|
158  *	+-------+-------+-------+-------+-------+-------+-------+-------+
159  * 5	|G|			 offset3				|
160  *	+-------+-------+-------+-------+-------+-------+-------+-------+
161  * 6	|BDX|lvl| type	| cksum | comp	|     PSIZE	|     LSIZE	|
162  *	+-------+-------+-------+-------+-------+-------+-------+-------+
163  * 7	|			padding					|
164  *	+-------+-------+-------+-------+-------+-------+-------+-------+
165  * 8	|			padding					|
166  *	+-------+-------+-------+-------+-------+-------+-------+-------+
167  * 9	|			physical birth txg			|
168  *	+-------+-------+-------+-------+-------+-------+-------+-------+
169  * a	|			logical birth txg			|
170  *	+-------+-------+-------+-------+-------+-------+-------+-------+
171  * b	|			fill count				|
172  *	+-------+-------+-------+-------+-------+-------+-------+-------+
173  * c	|			checksum[0]				|
174  *	+-------+-------+-------+-------+-------+-------+-------+-------+
175  * d	|			checksum[1]				|
176  *	+-------+-------+-------+-------+-------+-------+-------+-------+
177  * e	|			checksum[2]				|
178  *	+-------+-------+-------+-------+-------+-------+-------+-------+
179  * f	|			checksum[3]				|
180  *	+-------+-------+-------+-------+-------+-------+-------+-------+
181  *
182  * Legend:
183  *
184  * vdev		virtual device ID
185  * offset	offset into virtual device
186  * LSIZE	logical size
187  * PSIZE	physical size (after compression)
188  * ASIZE	allocated size (including RAID-Z parity and gang block headers)
189  * GRID		RAID-Z layout information (reserved for future use)
190  * cksum	checksum function
191  * comp		compression function
192  * G		gang block indicator
193  * B		byteorder (endianness)
194  * D		dedup
195  * X		unused
196  * lvl		level of indirection
197  * type		DMU object type
198  * phys birth	txg of block allocation; zero if same as logical birth txg
199  * log. birth	transaction group in which the block was logically born
200  * fill count	number of non-zero blocks under this bp
201  * checksum[4]	256-bit checksum of the data this bp describes
202  */
203 #define	SPA_BLKPTRSHIFT	7		/* blkptr_t is 128 bytes	*/
204 #define	SPA_DVAS_PER_BP	3		/* Number of DVAs in a bp	*/
205 
206 typedef struct blkptr {
207 	dva_t		blk_dva[SPA_DVAS_PER_BP]; /* Data Virtual Addresses */
208 	uint64_t	blk_prop;	/* size, compression, type, etc	    */
209 	uint64_t	blk_pad[2];	/* Extra space for the future	    */
210 	uint64_t	blk_phys_birth;	/* txg when block was allocated	    */
211 	uint64_t	blk_birth;	/* transaction group at birth	    */
212 	uint64_t	blk_fill;	/* fill count			    */
213 	zio_cksum_t	blk_cksum;	/* 256-bit checksum		    */
214 } blkptr_t;
215 
216 /*
217  * Macros to get and set fields in a bp or DVA.
218  */
219 #define	DVA_GET_ASIZE(dva)	\
220 	BF64_GET_SB((dva)->dva_word[0], 0, 24, SPA_MINBLOCKSHIFT, 0)
221 #define	DVA_SET_ASIZE(dva, x)	\
222 	BF64_SET_SB((dva)->dva_word[0], 0, 24, SPA_MINBLOCKSHIFT, 0, x)
223 
224 #define	DVA_GET_GRID(dva)	BF64_GET((dva)->dva_word[0], 24, 8)
225 #define	DVA_SET_GRID(dva, x)	BF64_SET((dva)->dva_word[0], 24, 8, x)
226 
227 #define	DVA_GET_VDEV(dva)	BF64_GET((dva)->dva_word[0], 32, 32)
228 #define	DVA_SET_VDEV(dva, x)	BF64_SET((dva)->dva_word[0], 32, 32, x)
229 
230 #define	DVA_GET_OFFSET(dva)	\
231 	BF64_GET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0)
232 #define	DVA_SET_OFFSET(dva, x)	\
233 	BF64_SET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0, x)
234 
235 #define	DVA_GET_GANG(dva)	BF64_GET((dva)->dva_word[1], 63, 1)
236 #define	DVA_SET_GANG(dva, x)	BF64_SET((dva)->dva_word[1], 63, 1, x)
237 
238 #define	BP_GET_LSIZE(bp)	\
239 	(BP_IS_HOLE(bp) ? 0 : \
240 	BF64_GET_SB((bp)->blk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1))
241 #define	BP_SET_LSIZE(bp, x)	\
242 	BF64_SET_SB((bp)->blk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1, x)
243 
244 #define	BP_GET_PSIZE(bp)	\
245 	BF64_GET_SB((bp)->blk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1)
246 #define	BP_SET_PSIZE(bp, x)	\
247 	BF64_SET_SB((bp)->blk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1, x)
248 
249 #define	BP_GET_COMPRESS(bp)	BF64_GET((bp)->blk_prop, 32, 8)
250 #define	BP_SET_COMPRESS(bp, x)	BF64_SET((bp)->blk_prop, 32, 8, x)
251 
252 #define	BP_GET_CHECKSUM(bp)	BF64_GET((bp)->blk_prop, 40, 8)
253 #define	BP_SET_CHECKSUM(bp, x)	BF64_SET((bp)->blk_prop, 40, 8, x)
254 
255 #define	BP_GET_TYPE(bp)		BF64_GET((bp)->blk_prop, 48, 8)
256 #define	BP_SET_TYPE(bp, x)	BF64_SET((bp)->blk_prop, 48, 8, x)
257 
258 #define	BP_GET_LEVEL(bp)	BF64_GET((bp)->blk_prop, 56, 5)
259 #define	BP_SET_LEVEL(bp, x)	BF64_SET((bp)->blk_prop, 56, 5, x)
260 
261 #define	BP_GET_DEDUP(bp)	BF64_GET((bp)->blk_prop, 62, 1)
262 #define	BP_SET_DEDUP(bp, x)	BF64_SET((bp)->blk_prop, 62, 1, x)
263 
264 #define	BP_GET_BYTEORDER(bp)	(0 - BF64_GET((bp)->blk_prop, 63, 1))
265 #define	BP_SET_BYTEORDER(bp, x)	BF64_SET((bp)->blk_prop, 63, 1, x)
266 
267 #define	BP_PHYSICAL_BIRTH(bp)		\
268 	((bp)->blk_phys_birth ? (bp)->blk_phys_birth : (bp)->blk_birth)
269 
270 #define	BP_GET_ASIZE(bp)	\
271 	(DVA_GET_ASIZE(&(bp)->blk_dva[0]) + DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
272 		DVA_GET_ASIZE(&(bp)->blk_dva[2]))
273 
274 #define	BP_GET_UCSIZE(bp) \
275 	((BP_GET_LEVEL(bp) > 0 || dmu_ot[BP_GET_TYPE(bp)].ot_metadata) ? \
276 	BP_GET_PSIZE(bp) : BP_GET_LSIZE(bp));
277 
278 #define	BP_GET_NDVAS(bp)	\
279 	(!!DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \
280 	!!DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
281 	!!DVA_GET_ASIZE(&(bp)->blk_dva[2]))
282 
283 #define	BP_COUNT_GANG(bp)	\
284 	(DVA_GET_GANG(&(bp)->blk_dva[0]) + \
285 	DVA_GET_GANG(&(bp)->blk_dva[1]) + \
286 	DVA_GET_GANG(&(bp)->blk_dva[2]))
287 
288 #define	DVA_EQUAL(dva1, dva2)	\
289 	((dva1)->dva_word[1] == (dva2)->dva_word[1] && \
290 	(dva1)->dva_word[0] == (dva2)->dva_word[0])
291 
292 #define	ZIO_CHECKSUM_EQUAL(zc1, zc2) \
293 	(0 == (((zc1).zc_word[0] - (zc2).zc_word[0]) | \
294 	((zc1).zc_word[1] - (zc2).zc_word[1]) | \
295 	((zc1).zc_word[2] - (zc2).zc_word[2]) | \
296 	((zc1).zc_word[3] - (zc2).zc_word[3])))
297 
298 
299 #define	DVA_IS_VALID(dva)	(DVA_GET_ASIZE(dva) != 0)
300 
301 #define	ZIO_SET_CHECKSUM(zcp, w0, w1, w2, w3)	\
302 {						\
303 	(zcp)->zc_word[0] = w0;			\
304 	(zcp)->zc_word[1] = w1;			\
305 	(zcp)->zc_word[2] = w2;			\
306 	(zcp)->zc_word[3] = w3;			\
307 }
308 
309 #define	BP_IDENTITY(bp)		(&(bp)->blk_dva[0])
310 #define	BP_IS_GANG(bp)		DVA_GET_GANG(BP_IDENTITY(bp))
311 #define	BP_IS_HOLE(bp)		((bp)->blk_birth == 0)
312 #define	BP_IS_OLDER(bp, txg)	(!BP_IS_HOLE(bp) && (bp)->blk_birth < (txg))
313 
314 #define	BP_ZERO(bp)				\
315 {						\
316 	(bp)->blk_dva[0].dva_word[0] = 0;	\
317 	(bp)->blk_dva[0].dva_word[1] = 0;	\
318 	(bp)->blk_dva[1].dva_word[0] = 0;	\
319 	(bp)->blk_dva[1].dva_word[1] = 0;	\
320 	(bp)->blk_dva[2].dva_word[0] = 0;	\
321 	(bp)->blk_dva[2].dva_word[1] = 0;	\
322 	(bp)->blk_prop = 0;			\
323 	(bp)->blk_pad[0] = 0;			\
324 	(bp)->blk_pad[1] = 0;			\
325 	(bp)->blk_phys_birth = 0;		\
326 	(bp)->blk_birth = 0;			\
327 	(bp)->blk_fill = 0;			\
328 	ZIO_SET_CHECKSUM(&(bp)->blk_cksum, 0, 0, 0, 0);	\
329 }
330 
331 /*
332  * Embedded checksum
333  */
334 #define	ZEC_MAGIC	0x210da7ab10c7a11ULL
335 
336 typedef struct zio_eck {
337 	uint64_t	zec_magic;	/* for validation, endianness	*/
338 	zio_cksum_t	zec_cksum;	/* 256-bit checksum		*/
339 } zio_eck_t;
340 
341 /*
342  * Gang block headers are self-checksumming and contain an array
343  * of block pointers.
344  */
345 #define	SPA_GANGBLOCKSIZE	SPA_MINBLOCKSIZE
346 #define	SPA_GBH_NBLKPTRS	((SPA_GANGBLOCKSIZE - \
347 	sizeof (zio_eck_t)) / sizeof (blkptr_t))
348 #define	SPA_GBH_FILLER		((SPA_GANGBLOCKSIZE - \
349 	sizeof (zio_eck_t) - \
350 	(SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
351 	sizeof (uint64_t))
352 
353 typedef struct zio_gbh {
354 	blkptr_t		zg_blkptr[SPA_GBH_NBLKPTRS];
355 	uint64_t		zg_filler[SPA_GBH_FILLER];
356 	zio_eck_t		zg_tail;
357 } zio_gbh_phys_t;
358 
359 #define	VDEV_RAIDZ_MAXPARITY	3
360 
361 #define	VDEV_PAD_SIZE		(8 << 10)
362 /* 2 padding areas (vl_pad1 and vl_pad2) to skip */
363 #define	VDEV_SKIP_SIZE		VDEV_PAD_SIZE * 2
364 #define	VDEV_PHYS_SIZE		(112 << 10)
365 #define	VDEV_UBERBLOCK_RING	(128 << 10)
366 
367 #define	VDEV_UBERBLOCK_SHIFT(vd)	\
368 	MAX((vd)->v_top->v_ashift, UBERBLOCK_SHIFT)
369 #define	VDEV_UBERBLOCK_COUNT(vd)	\
370 	(VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
371 #define	VDEV_UBERBLOCK_OFFSET(vd, n)	\
372 	offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)])
373 #define	VDEV_UBERBLOCK_SIZE(vd)		(1ULL << VDEV_UBERBLOCK_SHIFT(vd))
374 
375 typedef struct vdev_phys {
376 	char		vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)];
377 	zio_eck_t	vp_zbt;
378 } vdev_phys_t;
379 
380 typedef struct vdev_label {
381 	char		vl_pad1[VDEV_PAD_SIZE];			/*  8K  */
382 	char		vl_pad2[VDEV_PAD_SIZE];			/*  8K  */
383 	vdev_phys_t	vl_vdev_phys;				/* 112K	*/
384 	char		vl_uberblock[VDEV_UBERBLOCK_RING];	/* 128K	*/
385 } vdev_label_t;							/* 256K total */
386 
387 /*
388  * vdev_dirty() flags
389  */
390 #define	VDD_METASLAB	0x01
391 #define	VDD_DTL		0x02
392 
393 /*
394  * Size and offset of embedded boot loader region on each label.
395  * The total size of the first two labels plus the boot area is 4MB.
396  */
397 #define	VDEV_BOOT_OFFSET	(2 * sizeof (vdev_label_t))
398 #define	VDEV_BOOT_SIZE		(7ULL << 19)			/* 3.5M	*/
399 
400 /*
401  * Size of label regions at the start and end of each leaf device.
402  */
403 #define	VDEV_LABEL_START_SIZE	(2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE)
404 #define	VDEV_LABEL_END_SIZE	(2 * sizeof (vdev_label_t))
405 #define	VDEV_LABELS		4
406 
407 enum zio_checksum {
408 	ZIO_CHECKSUM_INHERIT = 0,
409 	ZIO_CHECKSUM_ON,
410 	ZIO_CHECKSUM_OFF,
411 	ZIO_CHECKSUM_LABEL,
412 	ZIO_CHECKSUM_GANG_HEADER,
413 	ZIO_CHECKSUM_ZILOG,
414 	ZIO_CHECKSUM_FLETCHER_2,
415 	ZIO_CHECKSUM_FLETCHER_4,
416 	ZIO_CHECKSUM_SHA256,
417 	ZIO_CHECKSUM_ZILOG2,
418 	ZIO_CHECKSUM_FUNCTIONS
419 };
420 
421 #define	ZIO_CHECKSUM_ON_VALUE	ZIO_CHECKSUM_FLETCHER_4
422 #define	ZIO_CHECKSUM_DEFAULT	ZIO_CHECKSUM_ON
423 
424 enum zio_compress {
425 	ZIO_COMPRESS_INHERIT = 0,
426 	ZIO_COMPRESS_ON,
427 	ZIO_COMPRESS_OFF,
428 	ZIO_COMPRESS_LZJB,
429 	ZIO_COMPRESS_EMPTY,
430 	ZIO_COMPRESS_GZIP_1,
431 	ZIO_COMPRESS_GZIP_2,
432 	ZIO_COMPRESS_GZIP_3,
433 	ZIO_COMPRESS_GZIP_4,
434 	ZIO_COMPRESS_GZIP_5,
435 	ZIO_COMPRESS_GZIP_6,
436 	ZIO_COMPRESS_GZIP_7,
437 	ZIO_COMPRESS_GZIP_8,
438 	ZIO_COMPRESS_GZIP_9,
439 	ZIO_COMPRESS_ZLE,
440 	ZIO_COMPRESS_FUNCTIONS
441 };
442 
443 #define	ZIO_COMPRESS_ON_VALUE	ZIO_COMPRESS_LZJB
444 #define	ZIO_COMPRESS_DEFAULT	ZIO_COMPRESS_OFF
445 
446 /* nvlist pack encoding */
447 #define	NV_ENCODE_NATIVE	0
448 #define	NV_ENCODE_XDR		1
449 
450 typedef enum {
451 	DATA_TYPE_UNKNOWN = 0,
452 	DATA_TYPE_BOOLEAN,
453 	DATA_TYPE_BYTE,
454 	DATA_TYPE_INT16,
455 	DATA_TYPE_UINT16,
456 	DATA_TYPE_INT32,
457 	DATA_TYPE_UINT32,
458 	DATA_TYPE_INT64,
459 	DATA_TYPE_UINT64,
460 	DATA_TYPE_STRING,
461 	DATA_TYPE_BYTE_ARRAY,
462 	DATA_TYPE_INT16_ARRAY,
463 	DATA_TYPE_UINT16_ARRAY,
464 	DATA_TYPE_INT32_ARRAY,
465 	DATA_TYPE_UINT32_ARRAY,
466 	DATA_TYPE_INT64_ARRAY,
467 	DATA_TYPE_UINT64_ARRAY,
468 	DATA_TYPE_STRING_ARRAY,
469 	DATA_TYPE_HRTIME,
470 	DATA_TYPE_NVLIST,
471 	DATA_TYPE_NVLIST_ARRAY,
472 	DATA_TYPE_BOOLEAN_VALUE,
473 	DATA_TYPE_INT8,
474 	DATA_TYPE_UINT8,
475 	DATA_TYPE_BOOLEAN_ARRAY,
476 	DATA_TYPE_INT8_ARRAY,
477 	DATA_TYPE_UINT8_ARRAY
478 } data_type_t;
479 
480 /*
481  * On-disk version number.
482  */
483 #define	SPA_VERSION_1			1ULL
484 #define	SPA_VERSION_2			2ULL
485 #define	SPA_VERSION_3			3ULL
486 #define	SPA_VERSION_4			4ULL
487 #define	SPA_VERSION_5			5ULL
488 #define	SPA_VERSION_6			6ULL
489 #define	SPA_VERSION_7			7ULL
490 #define	SPA_VERSION_8			8ULL
491 #define	SPA_VERSION_9			9ULL
492 #define	SPA_VERSION_10			10ULL
493 #define	SPA_VERSION_11			11ULL
494 #define	SPA_VERSION_12			12ULL
495 #define	SPA_VERSION_13			13ULL
496 #define	SPA_VERSION_14			14ULL
497 #define	SPA_VERSION_15			15ULL
498 #define	SPA_VERSION_16			16ULL
499 #define	SPA_VERSION_17			17ULL
500 #define	SPA_VERSION_18			18ULL
501 #define	SPA_VERSION_19			19ULL
502 #define	SPA_VERSION_20			20ULL
503 #define	SPA_VERSION_21			21ULL
504 #define	SPA_VERSION_22			22ULL
505 #define	SPA_VERSION_23			23ULL
506 #define	SPA_VERSION_24			24ULL
507 #define	SPA_VERSION_25			25ULL
508 #define	SPA_VERSION_26			26ULL
509 #define	SPA_VERSION_27			27ULL
510 #define	SPA_VERSION_28			28ULL
511 
512 /*
513  * When bumping up SPA_VERSION, make sure GRUB ZFS understands the on-disk
514  * format change. Go to usr/src/grub/grub-0.97/stage2/{zfs-include/, fsys_zfs*},
515  * and do the appropriate changes.  Also bump the version number in
516  * usr/src/grub/capability.
517  */
518 #define	SPA_VERSION			SPA_VERSION_28
519 #define	SPA_VERSION_STRING		"28"
520 
521 /*
522  * Symbolic names for the changes that caused a SPA_VERSION switch.
523  * Used in the code when checking for presence or absence of a feature.
524  * Feel free to define multiple symbolic names for each version if there
525  * were multiple changes to on-disk structures during that version.
526  *
527  * NOTE: When checking the current SPA_VERSION in your code, be sure
528  *       to use spa_version() since it reports the version of the
529  *       last synced uberblock.  Checking the in-flight version can
530  *       be dangerous in some cases.
531  */
532 #define	SPA_VERSION_INITIAL		SPA_VERSION_1
533 #define	SPA_VERSION_DITTO_BLOCKS	SPA_VERSION_2
534 #define	SPA_VERSION_SPARES		SPA_VERSION_3
535 #define	SPA_VERSION_RAID6		SPA_VERSION_3
536 #define	SPA_VERSION_BPLIST_ACCOUNT	SPA_VERSION_3
537 #define	SPA_VERSION_RAIDZ_DEFLATE	SPA_VERSION_3
538 #define	SPA_VERSION_DNODE_BYTES		SPA_VERSION_3
539 #define	SPA_VERSION_ZPOOL_HISTORY	SPA_VERSION_4
540 #define	SPA_VERSION_GZIP_COMPRESSION	SPA_VERSION_5
541 #define	SPA_VERSION_BOOTFS		SPA_VERSION_6
542 #define	SPA_VERSION_SLOGS		SPA_VERSION_7
543 #define	SPA_VERSION_DELEGATED_PERMS	SPA_VERSION_8
544 #define	SPA_VERSION_FUID		SPA_VERSION_9
545 #define	SPA_VERSION_REFRESERVATION	SPA_VERSION_9
546 #define	SPA_VERSION_REFQUOTA		SPA_VERSION_9
547 #define	SPA_VERSION_UNIQUE_ACCURATE	SPA_VERSION_9
548 #define	SPA_VERSION_L2CACHE		SPA_VERSION_10
549 #define	SPA_VERSION_NEXT_CLONES		SPA_VERSION_11
550 #define	SPA_VERSION_ORIGIN		SPA_VERSION_11
551 #define	SPA_VERSION_DSL_SCRUB		SPA_VERSION_11
552 #define	SPA_VERSION_SNAP_PROPS		SPA_VERSION_12
553 #define	SPA_VERSION_USED_BREAKDOWN	SPA_VERSION_13
554 #define	SPA_VERSION_PASSTHROUGH_X	SPA_VERSION_14
555 #define SPA_VERSION_USERSPACE		SPA_VERSION_15
556 #define	SPA_VERSION_STMF_PROP		SPA_VERSION_16
557 #define	SPA_VERSION_RAIDZ3		SPA_VERSION_17
558 #define	SPA_VERSION_USERREFS		SPA_VERSION_18
559 #define	SPA_VERSION_HOLES		SPA_VERSION_19
560 #define	SPA_VERSION_ZLE_COMPRESSION	SPA_VERSION_20
561 #define	SPA_VERSION_DEDUP		SPA_VERSION_21
562 #define	SPA_VERSION_RECVD_PROPS		SPA_VERSION_22
563 #define	SPA_VERSION_SLIM_ZIL		SPA_VERSION_23
564 #define	SPA_VERSION_SA			SPA_VERSION_24
565 #define	SPA_VERSION_SCAN		SPA_VERSION_25
566 #define	SPA_VERSION_DIR_CLONES		SPA_VERSION_26
567 #define	SPA_VERSION_DEADLISTS		SPA_VERSION_26
568 #define	SPA_VERSION_FAST_SNAP		SPA_VERSION_27
569 #define	SPA_VERSION_MULTI_REPLACE	SPA_VERSION_28
570 
571 /*
572  * The following are configuration names used in the nvlist describing a pool's
573  * configuration.
574  */
575 #define	ZPOOL_CONFIG_VERSION		"version"
576 #define	ZPOOL_CONFIG_POOL_NAME		"name"
577 #define	ZPOOL_CONFIG_POOL_STATE		"state"
578 #define	ZPOOL_CONFIG_POOL_TXG		"txg"
579 #define	ZPOOL_CONFIG_POOL_GUID		"pool_guid"
580 #define	ZPOOL_CONFIG_CREATE_TXG		"create_txg"
581 #define	ZPOOL_CONFIG_TOP_GUID		"top_guid"
582 #define	ZPOOL_CONFIG_VDEV_TREE		"vdev_tree"
583 #define	ZPOOL_CONFIG_TYPE		"type"
584 #define	ZPOOL_CONFIG_CHILDREN		"children"
585 #define	ZPOOL_CONFIG_ID			"id"
586 #define	ZPOOL_CONFIG_GUID		"guid"
587 #define	ZPOOL_CONFIG_PATH		"path"
588 #define	ZPOOL_CONFIG_DEVID		"devid"
589 #define	ZPOOL_CONFIG_METASLAB_ARRAY	"metaslab_array"
590 #define	ZPOOL_CONFIG_METASLAB_SHIFT	"metaslab_shift"
591 #define	ZPOOL_CONFIG_ASHIFT		"ashift"
592 #define	ZPOOL_CONFIG_ASIZE		"asize"
593 #define	ZPOOL_CONFIG_DTL		"DTL"
594 #define	ZPOOL_CONFIG_STATS		"stats"
595 #define	ZPOOL_CONFIG_WHOLE_DISK		"whole_disk"
596 #define	ZPOOL_CONFIG_ERRCOUNT		"error_count"
597 #define	ZPOOL_CONFIG_NOT_PRESENT	"not_present"
598 #define	ZPOOL_CONFIG_SPARES		"spares"
599 #define	ZPOOL_CONFIG_IS_SPARE		"is_spare"
600 #define	ZPOOL_CONFIG_NPARITY		"nparity"
601 #define	ZPOOL_CONFIG_HOSTID		"hostid"
602 #define	ZPOOL_CONFIG_HOSTNAME		"hostname"
603 #define	ZPOOL_CONFIG_IS_LOG		"is_log"
604 #define	ZPOOL_CONFIG_TIMESTAMP		"timestamp" /* not stored on disk */
605 
606 /*
607  * The persistent vdev state is stored as separate values rather than a single
608  * 'vdev_state' entry.  This is because a device can be in multiple states, such
609  * as offline and degraded.
610  */
611 #define	ZPOOL_CONFIG_OFFLINE            "offline"
612 #define	ZPOOL_CONFIG_FAULTED            "faulted"
613 #define	ZPOOL_CONFIG_DEGRADED           "degraded"
614 #define	ZPOOL_CONFIG_REMOVED            "removed"
615 #define	ZPOOL_CONFIG_FRU		"fru"
616 #define	ZPOOL_CONFIG_AUX_STATE		"aux_state"
617 
618 #define	VDEV_TYPE_ROOT			"root"
619 #define	VDEV_TYPE_MIRROR		"mirror"
620 #define	VDEV_TYPE_REPLACING		"replacing"
621 #define	VDEV_TYPE_RAIDZ			"raidz"
622 #define	VDEV_TYPE_DISK			"disk"
623 #define	VDEV_TYPE_FILE			"file"
624 #define	VDEV_TYPE_MISSING		"missing"
625 #define	VDEV_TYPE_HOLE			"hole"
626 #define	VDEV_TYPE_SPARE			"spare"
627 #define	VDEV_TYPE_LOG			"log"
628 #define	VDEV_TYPE_L2CACHE		"l2cache"
629 
630 /*
631  * This is needed in userland to report the minimum necessary device size.
632  */
633 #define	SPA_MINDEVSIZE		(64ULL << 20)
634 
635 /*
636  * The location of the pool configuration repository, shared between kernel and
637  * userland.
638  */
639 #define	ZPOOL_CACHE		"/boot/zfs/zpool.cache"
640 
641 /*
642  * vdev states are ordered from least to most healthy.
643  * A vdev that's CANT_OPEN or below is considered unusable.
644  */
645 typedef enum vdev_state {
646 	VDEV_STATE_UNKNOWN = 0,	/* Uninitialized vdev			*/
647 	VDEV_STATE_CLOSED,	/* Not currently open			*/
648 	VDEV_STATE_OFFLINE,	/* Not allowed to open			*/
649 	VDEV_STATE_REMOVED,	/* Explicitly removed from system	*/
650 	VDEV_STATE_CANT_OPEN,	/* Tried to open, but failed		*/
651 	VDEV_STATE_FAULTED,	/* External request to fault device	*/
652 	VDEV_STATE_DEGRADED,	/* Replicated vdev with unhealthy kids	*/
653 	VDEV_STATE_HEALTHY	/* Presumed good			*/
654 } vdev_state_t;
655 
656 /*
657  * vdev aux states.  When a vdev is in the CANT_OPEN state, the aux field
658  * of the vdev stats structure uses these constants to distinguish why.
659  */
660 typedef enum vdev_aux {
661 	VDEV_AUX_NONE,		/* no error				*/
662 	VDEV_AUX_OPEN_FAILED,	/* ldi_open_*() or vn_open() failed	*/
663 	VDEV_AUX_CORRUPT_DATA,	/* bad label or disk contents		*/
664 	VDEV_AUX_NO_REPLICAS,	/* insufficient number of replicas	*/
665 	VDEV_AUX_BAD_GUID_SUM,	/* vdev guid sum doesn't match		*/
666 	VDEV_AUX_TOO_SMALL,	/* vdev size is too small		*/
667 	VDEV_AUX_BAD_LABEL,	/* the label is OK but invalid		*/
668 	VDEV_AUX_VERSION_NEWER,	/* on-disk version is too new		*/
669 	VDEV_AUX_VERSION_OLDER,	/* on-disk version is too old		*/
670 	VDEV_AUX_SPARED		/* hot spare used in another pool	*/
671 } vdev_aux_t;
672 
673 /*
674  * pool state.  The following states are written to disk as part of the normal
675  * SPA lifecycle: ACTIVE, EXPORTED, DESTROYED, SPARE.  The remaining states are
676  * software abstractions used at various levels to communicate pool state.
677  */
678 typedef enum pool_state {
679 	POOL_STATE_ACTIVE = 0,		/* In active use		*/
680 	POOL_STATE_EXPORTED,		/* Explicitly exported		*/
681 	POOL_STATE_DESTROYED,		/* Explicitly destroyed		*/
682 	POOL_STATE_SPARE,		/* Reserved for hot spare use	*/
683 	POOL_STATE_UNINITIALIZED,	/* Internal spa_t state		*/
684 	POOL_STATE_UNAVAIL,		/* Internal libzfs state	*/
685 	POOL_STATE_POTENTIALLY_ACTIVE	/* Internal libzfs state	*/
686 } pool_state_t;
687 
688 /*
689  * The uberblock version is incremented whenever an incompatible on-disk
690  * format change is made to the SPA, DMU, or ZAP.
691  *
692  * Note: the first two fields should never be moved.  When a storage pool
693  * is opened, the uberblock must be read off the disk before the version
694  * can be checked.  If the ub_version field is moved, we may not detect
695  * version mismatch.  If the ub_magic field is moved, applications that
696  * expect the magic number in the first word won't work.
697  */
698 #define	UBERBLOCK_MAGIC		0x00bab10c		/* oo-ba-bloc!	*/
699 #define	UBERBLOCK_SHIFT		10			/* up to 1K	*/
700 
701 struct uberblock {
702 	uint64_t	ub_magic;	/* UBERBLOCK_MAGIC		*/
703 	uint64_t	ub_version;	/* SPA_VERSION			*/
704 	uint64_t	ub_txg;		/* txg of last sync		*/
705 	uint64_t	ub_guid_sum;	/* sum of all vdev guids	*/
706 	uint64_t	ub_timestamp;	/* UTC time of last sync	*/
707 	blkptr_t	ub_rootbp;	/* MOS objset_phys_t		*/
708 };
709 
710 /*
711  * Flags.
712  */
713 #define	DNODE_MUST_BE_ALLOCATED	1
714 #define	DNODE_MUST_BE_FREE	2
715 
716 /*
717  * Fixed constants.
718  */
719 #define	DNODE_SHIFT		9	/* 512 bytes */
720 #define	DN_MIN_INDBLKSHIFT	10	/* 1k */
721 #define	DN_MAX_INDBLKSHIFT	14	/* 16k */
722 #define	DNODE_BLOCK_SHIFT	14	/* 16k */
723 #define	DNODE_CORE_SIZE		64	/* 64 bytes for dnode sans blkptrs */
724 #define	DN_MAX_OBJECT_SHIFT	48	/* 256 trillion (zfs_fid_t limit) */
725 #define	DN_MAX_OFFSET_SHIFT	64	/* 2^64 bytes in a dnode */
726 
727 /*
728  * Derived constants.
729  */
730 #define	DNODE_SIZE	(1 << DNODE_SHIFT)
731 #define	DN_MAX_NBLKPTR	((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
732 #define	DN_MAX_BONUSLEN	(DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT))
733 #define	DN_MAX_OBJECT	(1ULL << DN_MAX_OBJECT_SHIFT)
734 
735 #define	DNODES_PER_BLOCK_SHIFT	(DNODE_BLOCK_SHIFT - DNODE_SHIFT)
736 #define	DNODES_PER_BLOCK	(1ULL << DNODES_PER_BLOCK_SHIFT)
737 #define	DNODES_PER_LEVEL_SHIFT	(DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
738 
739 /* The +2 here is a cheesy way to round up */
740 #define	DN_MAX_LEVELS	(2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \
741 	(DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT)))
742 
743 #define	DN_BONUS(dnp)	((void*)((dnp)->dn_bonus + \
744 	(((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
745 
746 #define	DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
747 	(dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
748 
749 #define	EPB(blkshift, typeshift)	(1 << (blkshift - typeshift))
750 
751 /* Is dn_used in bytes?  if not, it's in multiples of SPA_MINBLOCKSIZE */
752 #define	DNODE_FLAG_USED_BYTES		(1<<0)
753 #define	DNODE_FLAG_USERUSED_ACCOUNTED	(1<<1)
754 
755 /* Does dnode have a SA spill blkptr in bonus? */
756 #define	DNODE_FLAG_SPILL_BLKPTR	(1<<2)
757 
758 typedef struct dnode_phys {
759 	uint8_t dn_type;		/* dmu_object_type_t */
760 	uint8_t dn_indblkshift;		/* ln2(indirect block size) */
761 	uint8_t dn_nlevels;		/* 1=dn_blkptr->data blocks */
762 	uint8_t dn_nblkptr;		/* length of dn_blkptr */
763 	uint8_t dn_bonustype;		/* type of data in bonus buffer */
764 	uint8_t	dn_checksum;		/* ZIO_CHECKSUM type */
765 	uint8_t	dn_compress;		/* ZIO_COMPRESS type */
766 	uint8_t dn_flags;		/* DNODE_FLAG_* */
767 	uint16_t dn_datablkszsec;	/* data block size in 512b sectors */
768 	uint16_t dn_bonuslen;		/* length of dn_bonus */
769 	uint8_t dn_pad2[4];
770 
771 	/* accounting is protected by dn_dirty_mtx */
772 	uint64_t dn_maxblkid;		/* largest allocated block ID */
773 	uint64_t dn_used;		/* bytes (or sectors) of disk space */
774 
775 	uint64_t dn_pad3[4];
776 
777 	blkptr_t dn_blkptr[1];
778 	uint8_t dn_bonus[DN_MAX_BONUSLEN - sizeof (blkptr_t)];
779 	blkptr_t dn_spill;
780 } dnode_phys_t;
781 
782 typedef enum dmu_object_type {
783 	DMU_OT_NONE,
784 	/* general: */
785 	DMU_OT_OBJECT_DIRECTORY,	/* ZAP */
786 	DMU_OT_OBJECT_ARRAY,		/* UINT64 */
787 	DMU_OT_PACKED_NVLIST,		/* UINT8 (XDR by nvlist_pack/unpack) */
788 	DMU_OT_PACKED_NVLIST_SIZE,	/* UINT64 */
789 	DMU_OT_BPLIST,			/* UINT64 */
790 	DMU_OT_BPLIST_HDR,		/* UINT64 */
791 	/* spa: */
792 	DMU_OT_SPACE_MAP_HEADER,	/* UINT64 */
793 	DMU_OT_SPACE_MAP,		/* UINT64 */
794 	/* zil: */
795 	DMU_OT_INTENT_LOG,		/* UINT64 */
796 	/* dmu: */
797 	DMU_OT_DNODE,			/* DNODE */
798 	DMU_OT_OBJSET,			/* OBJSET */
799 	/* dsl: */
800 	DMU_OT_DSL_DIR,			/* UINT64 */
801 	DMU_OT_DSL_DIR_CHILD_MAP,	/* ZAP */
802 	DMU_OT_DSL_DS_SNAP_MAP,		/* ZAP */
803 	DMU_OT_DSL_PROPS,		/* ZAP */
804 	DMU_OT_DSL_DATASET,		/* UINT64 */
805 	/* zpl: */
806 	DMU_OT_ZNODE,			/* ZNODE */
807 	DMU_OT_OLDACL,			/* Old ACL */
808 	DMU_OT_PLAIN_FILE_CONTENTS,	/* UINT8 */
809 	DMU_OT_DIRECTORY_CONTENTS,	/* ZAP */
810 	DMU_OT_MASTER_NODE,		/* ZAP */
811 	DMU_OT_UNLINKED_SET,		/* ZAP */
812 	/* zvol: */
813 	DMU_OT_ZVOL,			/* UINT8 */
814 	DMU_OT_ZVOL_PROP,		/* ZAP */
815 	/* other; for testing only! */
816 	DMU_OT_PLAIN_OTHER,		/* UINT8 */
817 	DMU_OT_UINT64_OTHER,		/* UINT64 */
818 	DMU_OT_ZAP_OTHER,		/* ZAP */
819 	/* new object types: */
820 	DMU_OT_ERROR_LOG,		/* ZAP */
821 	DMU_OT_SPA_HISTORY,		/* UINT8 */
822 	DMU_OT_SPA_HISTORY_OFFSETS,	/* spa_his_phys_t */
823 	DMU_OT_POOL_PROPS,		/* ZAP */
824 	DMU_OT_DSL_PERMS,		/* ZAP */
825 	DMU_OT_ACL,			/* ACL */
826 	DMU_OT_SYSACL,			/* SYSACL */
827 	DMU_OT_FUID,			/* FUID table (Packed NVLIST UINT8) */
828 	DMU_OT_FUID_SIZE,		/* FUID table size UINT64 */
829 	DMU_OT_NEXT_CLONES,		/* ZAP */
830 	DMU_OT_SCAN_QUEUE,		/* ZAP */
831 	DMU_OT_USERGROUP_USED,		/* ZAP */
832 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
833 	DMU_OT_USERREFS,		/* ZAP */
834 	DMU_OT_DDT_ZAP,			/* ZAP */
835 	DMU_OT_DDT_STATS,		/* ZAP */
836 	DMU_OT_SA,			/* System attr */
837 	DMU_OT_SA_MASTER_NODE,		/* ZAP */
838 	DMU_OT_SA_ATTR_REGISTRATION,	/* ZAP */
839 	DMU_OT_SA_ATTR_LAYOUTS,		/* ZAP */
840 	DMU_OT_SCAN_XLATE,		/* ZAP */
841 	DMU_OT_DEDUP,			/* fake dedup BP from ddt_bp_create() */
842 	DMU_OT_NUMTYPES
843 } dmu_object_type_t;
844 
845 typedef enum dmu_objset_type {
846 	DMU_OST_NONE,
847 	DMU_OST_META,
848 	DMU_OST_ZFS,
849 	DMU_OST_ZVOL,
850 	DMU_OST_OTHER,			/* For testing only! */
851 	DMU_OST_ANY,			/* Be careful! */
852 	DMU_OST_NUMTYPES
853 } dmu_objset_type_t;
854 
855 /*
856  * header for all bonus and spill buffers.
857  * The header has a fixed portion with a variable number
858  * of "lengths" depending on the number of variable sized
859  * attribues which are determined by the "layout number"
860  */
861 
862 #define	SA_MAGIC	0x2F505A  /* ZFS SA */
863 typedef struct sa_hdr_phys {
864 	uint32_t sa_magic;
865 	uint16_t sa_layout_info;  /* Encoded with hdrsize and layout number */
866 	uint16_t sa_lengths[1];	/* optional sizes for variable length attrs */
867 	/* ... Data follows the lengths.  */
868 } sa_hdr_phys_t;
869 
870 /*
871  * sa_hdr_phys -> sa_layout_info
872  *
873  * 16      10       0
874  * +--------+-------+
875  * | hdrsz  |layout |
876  * +--------+-------+
877  *
878  * Bits 0-10 are the layout number
879  * Bits 11-16 are the size of the header.
880  * The hdrsize is the number * 8
881  *
882  * For example.
883  * hdrsz of 1 ==> 8 byte header
884  *          2 ==> 16 byte header
885  *
886  */
887 
888 #define	SA_HDR_LAYOUT_NUM(hdr) BF32_GET(hdr->sa_layout_info, 0, 10)
889 #define	SA_HDR_SIZE(hdr) BF32_GET_SB(hdr->sa_layout_info, 10, 16, 3, 0)
890 #define	SA_HDR_LAYOUT_INFO_ENCODE(x, num, size) \
891 { \
892 	BF32_SET_SB(x, 10, 6, 3, 0, size); \
893 	BF32_SET(x, 0, 10, num); \
894 }
895 
896 #define	SA_MODE_OFFSET		0
897 #define	SA_SIZE_OFFSET		8
898 #define	SA_GEN_OFFSET		16
899 #define	SA_UID_OFFSET		24
900 #define	SA_GID_OFFSET		32
901 #define	SA_PARENT_OFFSET	40
902 
903 /*
904  * Intent log header - this on disk structure holds fields to manage
905  * the log.  All fields are 64 bit to easily handle cross architectures.
906  */
907 typedef struct zil_header {
908 	uint64_t zh_claim_txg;	/* txg in which log blocks were claimed */
909 	uint64_t zh_replay_seq;	/* highest replayed sequence number */
910 	blkptr_t zh_log;	/* log chain */
911 	uint64_t zh_claim_seq;	/* highest claimed sequence number */
912 	uint64_t zh_pad[5];
913 } zil_header_t;
914 
915 #define	OBJSET_PHYS_SIZE 2048
916 
917 typedef struct objset_phys {
918 	dnode_phys_t os_meta_dnode;
919 	zil_header_t os_zil_header;
920 	uint64_t os_type;
921 	uint64_t os_flags;
922 	char os_pad[OBJSET_PHYS_SIZE - sizeof (dnode_phys_t)*3 -
923 	    sizeof (zil_header_t) - sizeof (uint64_t)*2];
924 	dnode_phys_t os_userused_dnode;
925 	dnode_phys_t os_groupused_dnode;
926 } objset_phys_t;
927 
928 typedef struct dsl_dir_phys {
929 	uint64_t dd_creation_time; /* not actually used */
930 	uint64_t dd_head_dataset_obj;
931 	uint64_t dd_parent_obj;
932 	uint64_t dd_clone_parent_obj;
933 	uint64_t dd_child_dir_zapobj;
934 	/*
935 	 * how much space our children are accounting for; for leaf
936 	 * datasets, == physical space used by fs + snaps
937 	 */
938 	uint64_t dd_used_bytes;
939 	uint64_t dd_compressed_bytes;
940 	uint64_t dd_uncompressed_bytes;
941 	/* Administrative quota setting */
942 	uint64_t dd_quota;
943 	/* Administrative reservation setting */
944 	uint64_t dd_reserved;
945 	uint64_t dd_props_zapobj;
946 	uint64_t dd_pad[21]; /* pad out to 256 bytes for good measure */
947 } dsl_dir_phys_t;
948 
949 typedef struct dsl_dataset_phys {
950 	uint64_t ds_dir_obj;
951 	uint64_t ds_prev_snap_obj;
952 	uint64_t ds_prev_snap_txg;
953 	uint64_t ds_next_snap_obj;
954 	uint64_t ds_snapnames_zapobj;	/* zap obj of snaps; ==0 for snaps */
955 	uint64_t ds_num_children;	/* clone/snap children; ==0 for head */
956 	uint64_t ds_creation_time;	/* seconds since 1970 */
957 	uint64_t ds_creation_txg;
958 	uint64_t ds_deadlist_obj;
959 	uint64_t ds_used_bytes;
960 	uint64_t ds_compressed_bytes;
961 	uint64_t ds_uncompressed_bytes;
962 	uint64_t ds_unique_bytes;	/* only relevant to snapshots */
963 	/*
964 	 * The ds_fsid_guid is a 56-bit ID that can change to avoid
965 	 * collisions.  The ds_guid is a 64-bit ID that will never
966 	 * change, so there is a small probability that it will collide.
967 	 */
968 	uint64_t ds_fsid_guid;
969 	uint64_t ds_guid;
970 	uint64_t ds_flags;
971 	blkptr_t ds_bp;
972 	uint64_t ds_pad[8]; /* pad out to 320 bytes for good measure */
973 } dsl_dataset_phys_t;
974 
975 /*
976  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
977  */
978 #define	DMU_POOL_DIRECTORY_OBJECT	1
979 #define	DMU_POOL_CONFIG			"config"
980 #define	DMU_POOL_ROOT_DATASET		"root_dataset"
981 #define	DMU_POOL_SYNC_BPLIST		"sync_bplist"
982 #define	DMU_POOL_ERRLOG_SCRUB		"errlog_scrub"
983 #define	DMU_POOL_ERRLOG_LAST		"errlog_last"
984 #define	DMU_POOL_SPARES			"spares"
985 #define	DMU_POOL_DEFLATE		"deflate"
986 #define	DMU_POOL_HISTORY		"history"
987 #define	DMU_POOL_PROPS			"pool_props"
988 
989 #define	ZAP_MAGIC 0x2F52AB2ABULL
990 
991 #define	FZAP_BLOCK_SHIFT(zap)	((zap)->zap_block_shift)
992 
993 #define	ZAP_MAXCD		(uint32_t)(-1)
994 #define	ZAP_HASHBITS		28
995 #define	MZAP_ENT_LEN		64
996 #define	MZAP_NAME_LEN		(MZAP_ENT_LEN - 8 - 4 - 2)
997 #define	MZAP_MAX_BLKSHIFT	SPA_MAXBLOCKSHIFT
998 #define	MZAP_MAX_BLKSZ		(1 << MZAP_MAX_BLKSHIFT)
999 
1000 typedef struct mzap_ent_phys {
1001 	uint64_t mze_value;
1002 	uint32_t mze_cd;
1003 	uint16_t mze_pad;	/* in case we want to chain them someday */
1004 	char mze_name[MZAP_NAME_LEN];
1005 } mzap_ent_phys_t;
1006 
1007 typedef struct mzap_phys {
1008 	uint64_t mz_block_type;	/* ZBT_MICRO */
1009 	uint64_t mz_salt;
1010 	uint64_t mz_pad[6];
1011 	mzap_ent_phys_t mz_chunk[1];
1012 	/* actually variable size depending on block size */
1013 } mzap_phys_t;
1014 
1015 /*
1016  * The (fat) zap is stored in one object. It is an array of
1017  * 1<<FZAP_BLOCK_SHIFT byte blocks. The layout looks like one of:
1018  *
1019  * ptrtbl fits in first block:
1020  * 	[zap_phys_t zap_ptrtbl_shift < 6] [zap_leaf_t] ...
1021  *
1022  * ptrtbl too big for first block:
1023  * 	[zap_phys_t zap_ptrtbl_shift >= 6] [zap_leaf_t] [ptrtbl] ...
1024  *
1025  */
1026 
1027 #define	ZBT_LEAF		((1ULL << 63) + 0)
1028 #define	ZBT_HEADER		((1ULL << 63) + 1)
1029 #define	ZBT_MICRO		((1ULL << 63) + 3)
1030 /* any other values are ptrtbl blocks */
1031 
1032 /*
1033  * the embedded pointer table takes up half a block:
1034  * block size / entry size (2^3) / 2
1035  */
1036 #define	ZAP_EMBEDDED_PTRTBL_SHIFT(zap) (FZAP_BLOCK_SHIFT(zap) - 3 - 1)
1037 
1038 /*
1039  * The embedded pointer table starts half-way through the block.  Since
1040  * the pointer table itself is half the block, it starts at (64-bit)
1041  * word number (1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)).
1042  */
1043 #define	ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) \
1044 	((uint64_t *)(zap)->zap_phys) \
1045 	[(idx) + (1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap))]
1046 
1047 /*
1048  * TAKE NOTE:
1049  * If zap_phys_t is modified, zap_byteswap() must be modified.
1050  */
1051 typedef struct zap_phys {
1052 	uint64_t zap_block_type;	/* ZBT_HEADER */
1053 	uint64_t zap_magic;		/* ZAP_MAGIC */
1054 
1055 	struct zap_table_phys {
1056 		uint64_t zt_blk;	/* starting block number */
1057 		uint64_t zt_numblks;	/* number of blocks */
1058 		uint64_t zt_shift;	/* bits to index it */
1059 		uint64_t zt_nextblk;	/* next (larger) copy start block */
1060 		uint64_t zt_blks_copied; /* number source blocks copied */
1061 	} zap_ptrtbl;
1062 
1063 	uint64_t zap_freeblk;		/* the next free block */
1064 	uint64_t zap_num_leafs;		/* number of leafs */
1065 	uint64_t zap_num_entries;	/* number of entries */
1066 	uint64_t zap_salt;		/* salt to stir into hash function */
1067 	/*
1068 	 * This structure is followed by padding, and then the embedded
1069 	 * pointer table.  The embedded pointer table takes up second
1070 	 * half of the block.  It is accessed using the
1071 	 * ZAP_EMBEDDED_PTRTBL_ENT() macro.
1072 	 */
1073 } zap_phys_t;
1074 
1075 typedef struct zap_table_phys zap_table_phys_t;
1076 
1077 typedef struct fat_zap {
1078 	int zap_block_shift;			/* block size shift */
1079 	zap_phys_t *zap_phys;
1080 } fat_zap_t;
1081 
1082 #define	ZAP_LEAF_MAGIC 0x2AB1EAF
1083 
1084 /* chunk size = 24 bytes */
1085 #define	ZAP_LEAF_CHUNKSIZE 24
1086 
1087 /*
1088  * The amount of space available for chunks is:
1089  * block size (1<<l->l_bs) - hash entry size (2) * number of hash
1090  * entries - header space (2*chunksize)
1091  */
1092 #define	ZAP_LEAF_NUMCHUNKS(l) \
1093 	(((1<<(l)->l_bs) - 2*ZAP_LEAF_HASH_NUMENTRIES(l)) / \
1094 	ZAP_LEAF_CHUNKSIZE - 2)
1095 
1096 /*
1097  * The amount of space within the chunk available for the array is:
1098  * chunk size - space for type (1) - space for next pointer (2)
1099  */
1100 #define	ZAP_LEAF_ARRAY_BYTES (ZAP_LEAF_CHUNKSIZE - 3)
1101 
1102 #define	ZAP_LEAF_ARRAY_NCHUNKS(bytes) \
1103 	(((bytes)+ZAP_LEAF_ARRAY_BYTES-1)/ZAP_LEAF_ARRAY_BYTES)
1104 
1105 /*
1106  * Low water mark:  when there are only this many chunks free, start
1107  * growing the ptrtbl.  Ideally, this should be larger than a
1108  * "reasonably-sized" entry.  20 chunks is more than enough for the
1109  * largest directory entry (MAXNAMELEN (256) byte name, 8-byte value),
1110  * while still being only around 3% for 16k blocks.
1111  */
1112 #define	ZAP_LEAF_LOW_WATER (20)
1113 
1114 /*
1115  * The leaf hash table has block size / 2^5 (32) number of entries,
1116  * which should be more than enough for the maximum number of entries,
1117  * which is less than block size / CHUNKSIZE (24) / minimum number of
1118  * chunks per entry (3).
1119  */
1120 #define	ZAP_LEAF_HASH_SHIFT(l) ((l)->l_bs - 5)
1121 #define	ZAP_LEAF_HASH_NUMENTRIES(l) (1 << ZAP_LEAF_HASH_SHIFT(l))
1122 
1123 /*
1124  * The chunks start immediately after the hash table.  The end of the
1125  * hash table is at l_hash + HASH_NUMENTRIES, which we simply cast to a
1126  * chunk_t.
1127  */
1128 #define	ZAP_LEAF_CHUNK(l, idx) \
1129 	((zap_leaf_chunk_t *) \
1130 	((l)->l_phys->l_hash + ZAP_LEAF_HASH_NUMENTRIES(l)))[idx]
1131 #define	ZAP_LEAF_ENTRY(l, idx) (&ZAP_LEAF_CHUNK(l, idx).l_entry)
1132 
1133 typedef enum zap_chunk_type {
1134 	ZAP_CHUNK_FREE = 253,
1135 	ZAP_CHUNK_ENTRY = 252,
1136 	ZAP_CHUNK_ARRAY = 251,
1137 	ZAP_CHUNK_TYPE_MAX = 250
1138 } zap_chunk_type_t;
1139 
1140 /*
1141  * TAKE NOTE:
1142  * If zap_leaf_phys_t is modified, zap_leaf_byteswap() must be modified.
1143  */
1144 typedef struct zap_leaf_phys {
1145 	struct zap_leaf_header {
1146 		uint64_t lh_block_type;		/* ZBT_LEAF */
1147 		uint64_t lh_pad1;
1148 		uint64_t lh_prefix;		/* hash prefix of this leaf */
1149 		uint32_t lh_magic;		/* ZAP_LEAF_MAGIC */
1150 		uint16_t lh_nfree;		/* number free chunks */
1151 		uint16_t lh_nentries;		/* number of entries */
1152 		uint16_t lh_prefix_len;		/* num bits used to id this */
1153 
1154 /* above is accessable to zap, below is zap_leaf private */
1155 
1156 		uint16_t lh_freelist;		/* chunk head of free list */
1157 		uint8_t lh_pad2[12];
1158 	} l_hdr; /* 2 24-byte chunks */
1159 
1160 	/*
1161 	 * The header is followed by a hash table with
1162 	 * ZAP_LEAF_HASH_NUMENTRIES(zap) entries.  The hash table is
1163 	 * followed by an array of ZAP_LEAF_NUMCHUNKS(zap)
1164 	 * zap_leaf_chunk structures.  These structures are accessed
1165 	 * with the ZAP_LEAF_CHUNK() macro.
1166 	 */
1167 
1168 	uint16_t l_hash[1];
1169 } zap_leaf_phys_t;
1170 
1171 typedef union zap_leaf_chunk {
1172 	struct zap_leaf_entry {
1173 		uint8_t le_type; 		/* always ZAP_CHUNK_ENTRY */
1174 		uint8_t le_int_size;		/* size of ints */
1175 		uint16_t le_next;		/* next entry in hash chain */
1176 		uint16_t le_name_chunk;		/* first chunk of the name */
1177 		uint16_t le_name_length;	/* bytes in name, incl null */
1178 		uint16_t le_value_chunk;	/* first chunk of the value */
1179 		uint16_t le_value_length;	/* value length in ints */
1180 		uint32_t le_cd;			/* collision differentiator */
1181 		uint64_t le_hash;		/* hash value of the name */
1182 	} l_entry;
1183 	struct zap_leaf_array {
1184 		uint8_t la_type;		/* always ZAP_CHUNK_ARRAY */
1185 		uint8_t la_array[ZAP_LEAF_ARRAY_BYTES];
1186 		uint16_t la_next;		/* next blk or CHAIN_END */
1187 	} l_array;
1188 	struct zap_leaf_free {
1189 		uint8_t lf_type;		/* always ZAP_CHUNK_FREE */
1190 		uint8_t lf_pad[ZAP_LEAF_ARRAY_BYTES];
1191 		uint16_t lf_next;	/* next in free list, or CHAIN_END */
1192 	} l_free;
1193 } zap_leaf_chunk_t;
1194 
1195 typedef struct zap_leaf {
1196 	int l_bs;			/* block size shift */
1197 	zap_leaf_phys_t *l_phys;
1198 } zap_leaf_t;
1199 
1200 /*
1201  * Define special zfs pflags
1202  */
1203 #define	ZFS_XATTR	0x1		/* is an extended attribute */
1204 #define	ZFS_INHERIT_ACE	0x2		/* ace has inheritable ACEs */
1205 #define	ZFS_ACL_TRIVIAL 0x4		/* files ACL is trivial */
1206 
1207 #define	MASTER_NODE_OBJ	1
1208 
1209 /*
1210  * special attributes for master node.
1211  */
1212 
1213 #define	ZFS_FSID		"FSID"
1214 #define	ZFS_UNLINKED_SET	"DELETE_QUEUE"
1215 #define	ZFS_ROOT_OBJ		"ROOT"
1216 #define	ZPL_VERSION_OBJ		"VERSION"
1217 #define	ZFS_PROP_BLOCKPERPAGE	"BLOCKPERPAGE"
1218 #define	ZFS_PROP_NOGROWBLOCKS	"NOGROWBLOCKS"
1219 
1220 #define	ZFS_FLAG_BLOCKPERPAGE	0x1
1221 #define	ZFS_FLAG_NOGROWBLOCKS	0x2
1222 
1223 /*
1224  * ZPL version - rev'd whenever an incompatible on-disk format change
1225  * occurs.  Independent of SPA/DMU/ZAP versioning.
1226  */
1227 
1228 #define	ZPL_VERSION		1ULL
1229 
1230 /*
1231  * The directory entry has the type (currently unused on Solaris) in the
1232  * top 4 bits, and the object number in the low 48 bits.  The "middle"
1233  * 12 bits are unused.
1234  */
1235 #define	ZFS_DIRENT_TYPE(de) BF64_GET(de, 60, 4)
1236 #define	ZFS_DIRENT_OBJ(de) BF64_GET(de, 0, 48)
1237 #define	ZFS_DIRENT_MAKE(type, obj) (((uint64_t)type << 60) | obj)
1238 
1239 typedef struct ace {
1240 	uid_t		a_who;		/* uid or gid */
1241 	uint32_t	a_access_mask;	/* read,write,... */
1242 	uint16_t	a_flags;	/* see below */
1243 	uint16_t	a_type;		/* allow or deny */
1244 } ace_t;
1245 
1246 #define ACE_SLOT_CNT	6
1247 
1248 typedef struct zfs_znode_acl {
1249 	uint64_t	z_acl_extern_obj;	  /* ext acl pieces */
1250 	uint32_t	z_acl_count;		  /* Number of ACEs */
1251 	uint16_t	z_acl_version;		  /* acl version */
1252 	uint16_t	z_acl_pad;		  /* pad */
1253 	ace_t		z_ace_data[ACE_SLOT_CNT]; /* 6 standard ACEs */
1254 } zfs_znode_acl_t;
1255 
1256 /*
1257  * This is the persistent portion of the znode.  It is stored
1258  * in the "bonus buffer" of the file.  Short symbolic links
1259  * are also stored in the bonus buffer.
1260  */
1261 typedef struct znode_phys {
1262 	uint64_t zp_atime[2];		/*  0 - last file access time */
1263 	uint64_t zp_mtime[2];		/* 16 - last file modification time */
1264 	uint64_t zp_ctime[2];		/* 32 - last file change time */
1265 	uint64_t zp_crtime[2];		/* 48 - creation time */
1266 	uint64_t zp_gen;		/* 64 - generation (txg of creation) */
1267 	uint64_t zp_mode;		/* 72 - file mode bits */
1268 	uint64_t zp_size;		/* 80 - size of file */
1269 	uint64_t zp_parent;		/* 88 - directory parent (`..') */
1270 	uint64_t zp_links;		/* 96 - number of links to file */
1271 	uint64_t zp_xattr;		/* 104 - DMU object for xattrs */
1272 	uint64_t zp_rdev;		/* 112 - dev_t for VBLK & VCHR files */
1273 	uint64_t zp_flags;		/* 120 - persistent flags */
1274 	uint64_t zp_uid;		/* 128 - file owner */
1275 	uint64_t zp_gid;		/* 136 - owning group */
1276 	uint64_t zp_pad[4];		/* 144 - future */
1277 	zfs_znode_acl_t zp_acl;		/* 176 - 263 ACL */
1278 	/*
1279 	 * Data may pad out any remaining bytes in the znode buffer, eg:
1280 	 *
1281 	 * |<---------------------- dnode_phys (512) ------------------------>|
1282 	 * |<-- dnode (192) --->|<----------- "bonus" buffer (320) ---------->|
1283 	 *			|<---- znode (264) ---->|<---- data (56) ---->|
1284 	 *
1285 	 * At present, we only use this space to store symbolic links.
1286 	 */
1287 } znode_phys_t;
1288 
1289 /*
1290  * In-core vdev representation.
1291  */
1292 struct vdev;
1293 typedef int vdev_phys_read_t(struct vdev *vdev, void *priv,
1294     off_t offset, void *buf, size_t bytes);
1295 typedef int vdev_read_t(struct vdev *vdev, const blkptr_t *bp,
1296     void *buf, off_t offset, size_t bytes);
1297 
1298 typedef STAILQ_HEAD(vdev_list, vdev) vdev_list_t;
1299 
1300 typedef struct vdev {
1301 	STAILQ_ENTRY(vdev) v_childlink;	/* link in parent's child list */
1302 	STAILQ_ENTRY(vdev) v_alllink;	/* link in global vdev list */
1303 	vdev_list_t	v_children;	/* children of this vdev */
1304 	const char	*v_name;	/* vdev name */
1305 	uint64_t	v_guid;		/* vdev guid */
1306 	int		v_id;		/* index in parent */
1307 	int		v_ashift;	/* offset to block shift */
1308 	int		v_nparity;	/* # parity for raidz */
1309 	struct vdev	*v_top;		/* parent vdev */
1310 	int		v_nchildren;	/* # children */
1311 	vdev_state_t	v_state;	/* current state */
1312 	vdev_phys_read_t *v_phys_read;	/* read from raw leaf vdev */
1313 	vdev_read_t	*v_read;	/* read from vdev */
1314 	void		*v_read_priv;	/* private data for read function */
1315 } vdev_t;
1316 
1317 /*
1318  * In-core pool representation.
1319  */
1320 typedef STAILQ_HEAD(spa_list, spa) spa_list_t;
1321 
1322 typedef struct spa {
1323 	STAILQ_ENTRY(spa) spa_link;	/* link in global pool list */
1324 	char		*spa_name;	/* pool name */
1325 	uint64_t	spa_guid;	/* pool guid */
1326 	uint64_t	spa_txg;	/* most recent transaction */
1327 	struct uberblock spa_uberblock;	/* best uberblock so far */
1328 	vdev_list_t	spa_vdevs;	/* list of all toplevel vdevs */
1329 	objset_phys_t	spa_mos;	/* MOS for this pool */
1330 	objset_phys_t	spa_root_objset; /* current mounted ZPL objset */
1331 } spa_t;
1332