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 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
25  * Copyright 2017 Nexenta Systems, Inc.
26  */
27 
28 #ifndef	_SYS_ZAP_H
29 #define	_SYS_ZAP_H
30 
31 /*
32  * ZAP - ZFS Attribute Processor
33  *
34  * The ZAP is a module which sits on top of the DMU (Data Management
35  * Unit) and implements a higher-level storage primitive using DMU
36  * objects.  Its primary consumer is the ZPL (ZFS Posix Layer).
37  *
38  * A "zapobj" is a DMU object which the ZAP uses to stores attributes.
39  * Users should use only zap routines to access a zapobj - they should
40  * not access the DMU object directly using DMU routines.
41  *
42  * The attributes stored in a zapobj are name-value pairs.  The name is
43  * a zero-terminated string of up to ZAP_MAXNAMELEN bytes (including
44  * terminating NULL).  The value is an array of integers, which may be
45  * 1, 2, 4, or 8 bytes long.  The total space used by the array (number
46  * of integers * integer length) can be up to ZAP_MAXVALUELEN bytes.
47  * Note that an 8-byte integer value can be used to store the location
48  * (object number) of another dmu object (which may be itself a zapobj).
49  * Note that you can use a zero-length attribute to store a single bit
50  * of information - the attribute is present or not.
51  *
52  * The ZAP routines are thread-safe.  However, you must observe the
53  * DMU's restriction that a transaction may not be operated on
54  * concurrently.
55  *
56  * Any of the routines that return an int may return an I/O error (EIO
57  * or ECHECKSUM).
58  *
59  *
60  * Implementation / Performance Notes:
61  *
62  * The ZAP is intended to operate most efficiently on attributes with
63  * short (49 bytes or less) names and single 8-byte values, for which
64  * the microzap will be used.  The ZAP should be efficient enough so
65  * that the user does not need to cache these attributes.
66  *
67  * The ZAP's locking scheme makes its routines thread-safe.  Operations
68  * on different zapobjs will be processed concurrently.  Operations on
69  * the same zapobj which only read data will be processed concurrently.
70  * Operations on the same zapobj which modify data will be processed
71  * concurrently when there are many attributes in the zapobj (because
72  * the ZAP uses per-block locking - more than 128 * (number of cpus)
73  * small attributes will suffice).
74  */
75 
76 /*
77  * We're using zero-terminated byte strings (ie. ASCII or UTF-8 C
78  * strings) for the names of attributes, rather than a byte string
79  * bounded by an explicit length.  If some day we want to support names
80  * in character sets which have embedded zeros (eg. UTF-16, UTF-32),
81  * we'll have to add routines for using length-bounded strings.
82  */
83 
84 #include <sys/dmu.h>
85 #include <sys/refcount.h>
86 
87 #ifdef	__cplusplus
88 extern "C" {
89 #endif
90 
91 /*
92  * Specifies matching criteria for ZAP lookups.
93  * MT_NORMALIZE		Use ZAP normalization flags, which can include both
94  *			unicode normalization and case-insensitivity.
95  * MT_MATCH_CASE	Do case-sensitive lookups even if MT_NORMALIZE is
96  *			specified and ZAP normalization flags include
97  *			U8_TEXTPREP_TOUPPER.
98  */
99 typedef enum matchtype {
100 	MT_NORMALIZE = 1 << 0,
101 	MT_MATCH_CASE = 1 << 1,
102 } matchtype_t;
103 
104 typedef enum zap_flags {
105 	/* Use 64-bit hash value (serialized cursors will always use 64-bits) */
106 	ZAP_FLAG_HASH64 = 1 << 0,
107 	/* Key is binary, not string (zap_add_uint64() can be used) */
108 	ZAP_FLAG_UINT64_KEY = 1 << 1,
109 	/*
110 	 * First word of key (which must be an array of uint64) is
111 	 * already randomly distributed.
112 	 */
113 	ZAP_FLAG_PRE_HASHED_KEY = 1 << 2,
114 } zap_flags_t;
115 
116 /*
117  * Create a new zapobj with no attributes and return its object number.
118  *
119  * dnodesize specifies the on-disk size of the dnode for the new zapobj.
120  * Valid values are multiples of 512 up to DNODE_MAX_SIZE.
121  */
122 uint64_t zap_create(objset_t *ds, dmu_object_type_t ot,
123     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
124 uint64_t zap_create_dnsize(objset_t *ds, dmu_object_type_t ot,
125     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx);
126 uint64_t zap_create_norm(objset_t *ds, int normflags, dmu_object_type_t ot,
127     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
128 uint64_t zap_create_norm_dnsize(objset_t *ds, int normflags,
129     dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
130     int dnodesize, dmu_tx_t *tx);
131 uint64_t zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
132     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
133     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
134 uint64_t zap_create_flags_dnsize(objset_t *os, int normflags,
135     zap_flags_t flags, dmu_object_type_t ot, int leaf_blockshift,
136     int indirect_blockshift, dmu_object_type_t bonustype, int bonuslen,
137     int dnodesize, dmu_tx_t *tx);
138 uint64_t zap_create_link(objset_t *os, dmu_object_type_t ot,
139      uint64_t parent_obj, const char *name, dmu_tx_t *tx);
140 uint64_t zap_create_link_dnsize(objset_t *os, dmu_object_type_t ot,
141     uint64_t parent_obj, const char *name, int dnodesize, dmu_tx_t *tx);
142 uint64_t zap_create_link_dnsize(objset_t *os, dmu_object_type_t ot,
143     uint64_t parent_obj, const char *name, int dnodesize, dmu_tx_t *tx);
144 
145 /*
146  * Initialize an already-allocated object.
147  */
148 void mzap_create_impl(objset_t *os, uint64_t obj, int normflags,
149     zap_flags_t flags, dmu_tx_t *tx);
150 
151 /*
152  * Create a new zapobj with no attributes from the given (unallocated)
153  * object number.
154  */
155 int zap_create_claim(objset_t *ds, uint64_t obj, dmu_object_type_t ot,
156     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
157 int zap_create_claim_dnsize(objset_t *ds, uint64_t obj, dmu_object_type_t ot,
158     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx);
159 int zap_create_claim_norm(objset_t *ds, uint64_t obj,
160     int normflags, dmu_object_type_t ot,
161     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
162 int zap_create_claim_norm_dnsize(objset_t *ds, uint64_t obj,
163     int normflags, dmu_object_type_t ot,
164     dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx);
165 
166 /*
167  * The zapobj passed in must be a valid ZAP object for all of the
168  * following routines.
169  */
170 
171 /*
172  * Destroy this zapobj and all its attributes.
173  *
174  * Frees the object number using dmu_object_free.
175  */
176 int zap_destroy(objset_t *ds, uint64_t zapobj, dmu_tx_t *tx);
177 
178 /*
179  * Manipulate attributes.
180  *
181  * 'integer_size' is in bytes, and must be 1, 2, 4, or 8.
182  */
183 
184 /*
185  * Retrieve the contents of the attribute with the given name.
186  *
187  * If the requested attribute does not exist, the call will fail and
188  * return ENOENT.
189  *
190  * If 'integer_size' is smaller than the attribute's integer size, the
191  * call will fail and return EINVAL.
192  *
193  * If 'integer_size' is equal to or larger than the attribute's integer
194  * size, the call will succeed and return 0.
195  *
196  * When converting to a larger integer size, the integers will be treated as
197  * unsigned (ie. no sign-extension will be performed).
198  *
199  * 'num_integers' is the length (in integers) of 'buf'.
200  *
201  * If the attribute is longer than the buffer, as many integers as will
202  * fit will be transferred to 'buf'.  If the entire attribute was not
203  * transferred, the call will return EOVERFLOW.
204  */
205 int zap_lookup(objset_t *ds, uint64_t zapobj, const char *name,
206     uint64_t integer_size, uint64_t num_integers, void *buf);
207 
208 /*
209  * If rn_len is nonzero, realname will be set to the name of the found
210  * entry (which may be different from the requested name if matchtype is
211  * not MT_EXACT).
212  *
213  * If normalization_conflictp is not NULL, it will be set if there is
214  * another name with the same case/unicode normalized form.
215  */
216 int zap_lookup_norm(objset_t *ds, uint64_t zapobj, const char *name,
217     uint64_t integer_size, uint64_t num_integers, void *buf,
218     matchtype_t mt, char *realname, int rn_len,
219     boolean_t *normalization_conflictp);
220 int zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
221     int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf);
222 int zap_contains(objset_t *ds, uint64_t zapobj, const char *name);
223 int zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
224     int key_numints);
225 int zap_lookup_by_dnode(dnode_t *dn, const char *name,
226     uint64_t integer_size, uint64_t num_integers, void *buf);
227 int zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
228     uint64_t integer_size, uint64_t num_integers, void *buf,
229     matchtype_t mt, char *realname, int rn_len,
230     boolean_t *ncp);
231 
232 int zap_count_write_by_dnode(dnode_t *dn, const char *name,
233     int add, zfs_refcount_t *towrite, zfs_refcount_t *tooverwrite);
234 
235 /*
236  * Create an attribute with the given name and value.
237  *
238  * If an attribute with the given name already exists, the call will
239  * fail and return EEXIST.
240  */
241 int zap_add(objset_t *ds, uint64_t zapobj, const char *key,
242     int integer_size, uint64_t num_integers,
243     const void *val, dmu_tx_t *tx);
244 int zap_add_by_dnode(dnode_t *dn, const char *key,
245     int integer_size, uint64_t num_integers,
246     const void *val, dmu_tx_t *tx);
247 int zap_add_uint64(objset_t *ds, uint64_t zapobj, const uint64_t *key,
248     int key_numints, int integer_size, uint64_t num_integers,
249     const void *val, dmu_tx_t *tx);
250 
251 /*
252  * Set the attribute with the given name to the given value.  If an
253  * attribute with the given name does not exist, it will be created.  If
254  * an attribute with the given name already exists, the previous value
255  * will be overwritten.  The integer_size may be different from the
256  * existing attribute's integer size, in which case the attribute's
257  * integer size will be updated to the new value.
258  */
259 int zap_update(objset_t *ds, uint64_t zapobj, const char *name,
260     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx);
261 int zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
262     int key_numints,
263     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx);
264 
265 /*
266  * Get the length (in integers) and the integer size of the specified
267  * attribute.
268  *
269  * If the requested attribute does not exist, the call will fail and
270  * return ENOENT.
271  */
272 int zap_length(objset_t *ds, uint64_t zapobj, const char *name,
273     uint64_t *integer_size, uint64_t *num_integers);
274 int zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
275     int key_numints, uint64_t *integer_size, uint64_t *num_integers);
276 
277 /*
278  * Remove the specified attribute.
279  *
280  * If the specified attribute does not exist, the call will fail and
281  * return ENOENT.
282  */
283 int zap_remove(objset_t *ds, uint64_t zapobj, const char *name, dmu_tx_t *tx);
284 int zap_remove_norm(objset_t *ds, uint64_t zapobj, const char *name,
285     matchtype_t mt, dmu_tx_t *tx);
286 int zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx);
287 int zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
288     int key_numints, dmu_tx_t *tx);
289 
290 /*
291  * Returns (in *count) the number of attributes in the specified zap
292  * object.
293  */
294 int zap_count(objset_t *ds, uint64_t zapobj, uint64_t *count);
295 
296 /*
297  * Returns (in name) the name of the entry whose (value & mask)
298  * (za_first_integer) is value, or ENOENT if not found.  The string
299  * pointed to by name must be at least 256 bytes long.  If mask==0, the
300  * match must be exact (ie, same as mask=-1ULL).
301  */
302 int zap_value_search(objset_t *os, uint64_t zapobj,
303     uint64_t value, uint64_t mask, char *name);
304 
305 /*
306  * Transfer all the entries from fromobj into intoobj.  Only works on
307  * int_size=8 num_integers=1 values.  Fails if there are any duplicated
308  * entries.
309  */
310 int zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx);
311 
312 /* Same as zap_join, but set the values to 'value'. */
313 int zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
314     uint64_t value, dmu_tx_t *tx);
315 
316 /* Same as zap_join, but add together any duplicated entries. */
317 int zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
318     dmu_tx_t *tx);
319 
320 /*
321  * Manipulate entries where the name + value are the "same" (the name is
322  * a stringified version of the value).
323  */
324 int zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx);
325 int zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx);
326 int zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value);
327 int zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
328     dmu_tx_t *tx);
329 
330 /* Here the key is an int and the value is a different int. */
331 int zap_add_int_key(objset_t *os, uint64_t obj,
332     uint64_t key, uint64_t value, dmu_tx_t *tx);
333 int zap_update_int_key(objset_t *os, uint64_t obj,
334     uint64_t key, uint64_t value, dmu_tx_t *tx);
335 int zap_lookup_int_key(objset_t *os, uint64_t obj,
336     uint64_t key, uint64_t *valuep);
337 
338 int zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
339     dmu_tx_t *tx);
340 
341 struct zap;
342 struct zap_leaf;
343 typedef struct zap_cursor {
344 	/* This structure is opaque! */
345 	objset_t *zc_objset;
346 	struct zap *zc_zap;
347 	struct zap_leaf *zc_leaf;
348 	uint64_t zc_zapobj;
349 	uint64_t zc_serialized;
350 	uint64_t zc_hash;
351 	uint32_t zc_cd;
352 	boolean_t zc_prefetch;
353 } zap_cursor_t;
354 
355 typedef struct {
356 	int za_integer_length;
357 	/*
358 	 * za_normalization_conflict will be set if there are additional
359 	 * entries with this normalized form (eg, "foo" and "Foo").
360 	 */
361 	boolean_t za_normalization_conflict;
362 	uint64_t za_num_integers;
363 	uint64_t za_first_integer;	/* no sign extension for <8byte ints */
364 	char za_name[ZAP_MAXNAMELEN];
365 } zap_attribute_t;
366 
367 /*
368  * The interface for listing all the attributes of a zapobj can be
369  * thought of as cursor moving down a list of the attributes one by
370  * one.  The cookie returned by the zap_cursor_serialize routine is
371  * persistent across system calls (and across reboot, even).
372  */
373 
374 /*
375  * Initialize a zap cursor, pointing to the "first" attribute of the
376  * zapobj.  You must _fini the cursor when you are done with it.
377  */
378 void zap_cursor_init(zap_cursor_t *zc, objset_t *ds, uint64_t zapobj);
379 void zap_cursor_init_noprefetch(zap_cursor_t *zc, objset_t *os,
380     uint64_t zapobj);
381 void zap_cursor_fini(zap_cursor_t *zc);
382 
383 /*
384  * Get the attribute currently pointed to by the cursor.  Returns
385  * ENOENT if at the end of the attributes.
386  */
387 int zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za);
388 
389 /*
390  * Advance the cursor to the next attribute.
391  */
392 void zap_cursor_advance(zap_cursor_t *zc);
393 
394 /*
395  * Get a persistent cookie pointing to the current position of the zap
396  * cursor.  The low 4 bits in the cookie are always zero, and thus can
397  * be used as to differentiate a serialized cookie from a different type
398  * of value.  The cookie will be less than 2^32 as long as there are
399  * fewer than 2^22 (4.2 million) entries in the zap object.
400  */
401 uint64_t zap_cursor_serialize(zap_cursor_t *zc);
402 
403 /*
404  * Advance the cursor to the attribute having the given key.
405  */
406 int zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt);
407 
408 /*
409  * Initialize a zap cursor pointing to the position recorded by
410  * zap_cursor_serialize (in the "serialized" argument).  You can also
411  * use a "serialized" argument of 0 to start at the beginning of the
412  * zapobj (ie.  zap_cursor_init_serialized(..., 0) is equivalent to
413  * zap_cursor_init(...).)
414  */
415 void zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *ds,
416     uint64_t zapobj, uint64_t serialized);
417 
418 
419 #define	ZAP_HISTOGRAM_SIZE 10
420 
421 typedef struct zap_stats {
422 	/*
423 	 * Size of the pointer table (in number of entries).
424 	 * This is always a power of 2, or zero if it's a microzap.
425 	 * In general, it should be considerably greater than zs_num_leafs.
426 	 */
427 	uint64_t zs_ptrtbl_len;
428 
429 	uint64_t zs_blocksize;		/* size of zap blocks */
430 
431 	/*
432 	 * The number of blocks used.  Note that some blocks may be
433 	 * wasted because old ptrtbl's and large name/value blocks are
434 	 * not reused.  (Although their space is reclaimed, we don't
435 	 * reuse those offsets in the object.)
436 	 */
437 	uint64_t zs_num_blocks;
438 
439 	/*
440 	 * Pointer table values from zap_ptrtbl in the zap_phys_t
441 	 */
442 	uint64_t zs_ptrtbl_nextblk;	  /* next (larger) copy start block */
443 	uint64_t zs_ptrtbl_blks_copied;   /* number source blocks copied */
444 	uint64_t zs_ptrtbl_zt_blk;	  /* starting block number */
445 	uint64_t zs_ptrtbl_zt_numblks;    /* number of blocks */
446 	uint64_t zs_ptrtbl_zt_shift;	  /* bits to index it */
447 
448 	/*
449 	 * Values of the other members of the zap_phys_t
450 	 */
451 	uint64_t zs_block_type;		/* ZBT_HEADER */
452 	uint64_t zs_magic;		/* ZAP_MAGIC */
453 	uint64_t zs_num_leafs;		/* The number of leaf blocks */
454 	uint64_t zs_num_entries;	/* The number of zap entries */
455 	uint64_t zs_salt;		/* salt to stir into hash function */
456 
457 	/*
458 	 * Histograms.  For all histograms, the last index
459 	 * (ZAP_HISTOGRAM_SIZE-1) includes any values which are greater
460 	 * than what can be represented.  For example
461 	 * zs_leafs_with_n5_entries[ZAP_HISTOGRAM_SIZE-1] is the number
462 	 * of leafs with more than 45 entries.
463 	 */
464 
465 	/*
466 	 * zs_leafs_with_n_pointers[n] is the number of leafs with
467 	 * 2^n pointers to it.
468 	 */
469 	uint64_t zs_leafs_with_2n_pointers[ZAP_HISTOGRAM_SIZE];
470 
471 	/*
472 	 * zs_leafs_with_n_entries[n] is the number of leafs with
473 	 * [n*5, (n+1)*5) entries.  In the current implementation, there
474 	 * can be at most 55 entries in any block, but there may be
475 	 * fewer if the name or value is large, or the block is not
476 	 * completely full.
477 	 */
478 	uint64_t zs_blocks_with_n5_entries[ZAP_HISTOGRAM_SIZE];
479 
480 	/*
481 	 * zs_leafs_n_tenths_full[n] is the number of leafs whose
482 	 * fullness is in the range [n/10, (n+1)/10).
483 	 */
484 	uint64_t zs_blocks_n_tenths_full[ZAP_HISTOGRAM_SIZE];
485 
486 	/*
487 	 * zs_entries_using_n_chunks[n] is the number of entries which
488 	 * consume n 24-byte chunks.  (Note, large names/values only use
489 	 * one chunk, but contribute to zs_num_blocks_large.)
490 	 */
491 	uint64_t zs_entries_using_n_chunks[ZAP_HISTOGRAM_SIZE];
492 
493 	/*
494 	 * zs_buckets_with_n_entries[n] is the number of buckets (each
495 	 * leaf has 64 buckets) with n entries.
496 	 * zs_buckets_with_n_entries[1] should be very close to
497 	 * zs_num_entries.
498 	 */
499 	uint64_t zs_buckets_with_n_entries[ZAP_HISTOGRAM_SIZE];
500 } zap_stats_t;
501 
502 /*
503  * Get statistics about a ZAP object.  Note: you need to be aware of the
504  * internal implementation of the ZAP to correctly interpret some of the
505  * statistics.  This interface shouldn't be relied on unless you really
506  * know what you're doing.
507  */
508 int zap_get_stats(objset_t *ds, uint64_t zapobj, zap_stats_t *zs);
509 
510 #ifdef	__cplusplus
511 }
512 #endif
513 
514 #endif	/* _SYS_ZAP_H */
515