4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2014, 2016, Intel Corporation.
26 * lustre/include/lustre_lmv.h
28 * Lustre LMV structures and functions.
30 * Author: Di Wang <di.wang@intel.com>
35 #include <uapi/linux/lustre/lustre_idl.h>
38 struct lu_fid lmo_fid;
40 struct inode *lmo_root;
43 struct lmv_stripe_md {
45 __u32 lsm_md_stripe_count;
46 __u32 lsm_md_master_mdt_index;
47 __u32 lsm_md_hash_type;
48 __u8 lsm_md_max_inherit;
49 __u8 lsm_md_max_inherit_rr;
50 __u32 lsm_md_layout_version;
51 __u32 lsm_md_migrate_offset;
52 __u32 lsm_md_migrate_hash;
53 __u32 lsm_md_default_count;
54 __u32 lsm_md_default_index;
55 char lsm_md_pool_name[LOV_MAXPOOLNAME + 1];
56 struct lmv_oinfo lsm_md_oinfo[0];
59 static inline bool lmv_dir_striped(const struct lmv_stripe_md *lsm)
61 return lsm && lsm->lsm_md_magic == LMV_MAGIC;
64 static inline bool lmv_dir_foreign(const struct lmv_stripe_md *lsm)
66 return lsm && lsm->lsm_md_magic == LMV_MAGIC_FOREIGN;
69 static inline bool lmv_dir_layout_changing(const struct lmv_stripe_md *lsm)
71 return lmv_dir_striped(lsm) &&
72 lmv_hash_is_layout_changing(lsm->lsm_md_hash_type);
75 static inline bool lmv_dir_bad_hash(const struct lmv_stripe_md *lsm)
77 if (!lmv_dir_striped(lsm))
80 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_BAD_TYPE)
83 return !lmv_is_known_hash_type(lsm->lsm_md_hash_type);
87 lsm_md_eq(const struct lmv_stripe_md *lsm1, const struct lmv_stripe_md *lsm2)
91 if (lsm1->lsm_md_magic != lsm2->lsm_md_magic ||
92 lsm1->lsm_md_stripe_count != lsm2->lsm_md_stripe_count ||
93 lsm1->lsm_md_master_mdt_index !=
94 lsm2->lsm_md_master_mdt_index ||
95 lsm1->lsm_md_hash_type != lsm2->lsm_md_hash_type ||
96 lsm1->lsm_md_layout_version !=
97 lsm2->lsm_md_layout_version ||
98 lsm1->lsm_md_migrate_offset !=
99 lsm2->lsm_md_migrate_offset ||
100 lsm1->lsm_md_migrate_hash !=
101 lsm2->lsm_md_migrate_hash ||
102 strncmp(lsm1->lsm_md_pool_name, lsm2->lsm_md_pool_name,
103 sizeof(lsm1->lsm_md_pool_name)) != 0)
106 if (lmv_dir_striped(lsm1)) {
107 for (idx = 0; idx < lsm1->lsm_md_stripe_count; idx++) {
108 if (!lu_fid_eq(&lsm1->lsm_md_oinfo[idx].lmo_fid,
109 &lsm2->lsm_md_oinfo[idx].lmo_fid))
117 static inline void lsm_md_dump(int mask, const struct lmv_stripe_md *lsm)
119 bool valid_hash = lmv_dir_bad_hash(lsm);
122 /* If lsm_md_magic == LMV_MAGIC_FOREIGN pool_name may not be a null
123 * terminated string so only print LOV_MAXPOOLNAME bytes.
126 "magic %#x stripe count %d master mdt %d hash type %s:%#x max inherit %hhu version %d migrate offset %d migrate hash %#x pool %.*s\n",
127 lsm->lsm_md_magic, lsm->lsm_md_stripe_count,
128 lsm->lsm_md_master_mdt_index,
129 valid_hash ? "invalid hash" :
130 mdt_hash_name[lsm->lsm_md_hash_type & (LMV_HASH_TYPE_MAX - 1)],
131 lsm->lsm_md_hash_type, lsm->lsm_md_max_inherit,
132 lsm->lsm_md_layout_version,
133 lsm->lsm_md_migrate_offset, lsm->lsm_md_migrate_hash,
134 LOV_MAXPOOLNAME, lsm->lsm_md_pool_name);
136 if (!lmv_dir_striped(lsm))
139 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
140 CDEBUG(mask, "stripe[%d] "DFID"\n",
141 i, PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
146 void lmv_free_memmd(struct lmv_stripe_md *lsm);
148 static inline void lmv1_le_to_cpu(struct lmv_mds_md_v1 *lmv_dst,
149 const struct lmv_mds_md_v1 *lmv_src)
153 lmv_dst->lmv_magic = le32_to_cpu(lmv_src->lmv_magic);
154 lmv_dst->lmv_stripe_count = le32_to_cpu(lmv_src->lmv_stripe_count);
155 lmv_dst->lmv_master_mdt_index =
156 le32_to_cpu(lmv_src->lmv_master_mdt_index);
157 lmv_dst->lmv_hash_type = le32_to_cpu(lmv_src->lmv_hash_type);
158 lmv_dst->lmv_layout_version = le32_to_cpu(lmv_src->lmv_layout_version);
159 if (lmv_src->lmv_stripe_count > LMV_MAX_STRIPE_COUNT)
161 for (i = 0; i < lmv_src->lmv_stripe_count; i++)
162 fid_le_to_cpu(&lmv_dst->lmv_stripe_fids[i],
163 &lmv_src->lmv_stripe_fids[i]);
166 static inline void lmv_le_to_cpu(union lmv_mds_md *lmv_dst,
167 const union lmv_mds_md *lmv_src)
169 switch (le32_to_cpu(lmv_src->lmv_magic)) {
171 lmv1_le_to_cpu(&lmv_dst->lmv_md_v1, &lmv_src->lmv_md_v1);
178 /* This hash is only for testing purpose */
179 static inline unsigned int
180 lmv_hash_all_chars(unsigned int count, const char *name, int namelen)
183 const unsigned char *p = (const unsigned char *)name;
185 while (--namelen >= 0)
193 static inline unsigned int
194 lmv_hash_fnv1a(unsigned int count, const char *name, int namelen)
198 hash = lustre_hash_fnv_1a_64(name, namelen);
200 return do_div(hash, count);
204 * Robert Jenkins' function for mixing 32-bit values
205 * http://burtleburtle.net/bob/hash/evahash.html
206 * a, b = random bits, c = input and output
208 * Mixing inputs to generate an evenly distributed hash.
210 #define crush_hashmix(a, b, c) \
212 a = a - b; a = a - c; a = a ^ (c >> 13); \
213 b = b - c; b = b - a; b = b ^ (a << 8); \
214 c = c - a; c = c - b; c = c ^ (b >> 13); \
215 a = a - b; a = a - c; a = a ^ (c >> 12); \
216 b = b - c; b = b - a; b = b ^ (a << 16); \
217 c = c - a; c = c - b; c = c ^ (b >> 5); \
218 a = a - b; a = a - c; a = a ^ (c >> 3); \
219 b = b - c; b = b - a; b = b ^ (a << 10); \
220 c = c - a; c = c - b; c = c ^ (b >> 15); \
223 #define crush_hash_seed 1315423911
225 static inline __u32 crush_hash(__u32 a, __u32 b)
227 __u32 hash = crush_hash_seed ^ a ^ b;
231 crush_hashmix(a, b, hash);
232 crush_hashmix(x, a, hash);
233 crush_hashmix(b, y, hash);
238 /* refer to https://github.com/ceph/ceph/blob/master/src/crush/hash.c and
239 * https://www.ssrc.ucsc.edu/Papers/weil-sc06.pdf for details of CRUSH
242 static inline unsigned int
243 lmv_hash_crush(unsigned int count, const char *name, int namelen)
245 unsigned long long straw;
246 unsigned long long highest_straw = 0;
248 unsigned int idx = 0;
251 /* put temp and backup file on the same MDT where target is located.
252 * temporary file naming rule:
253 * 1. rsync: .<target>.XXXXXX
254 * 2. dstripe: <target>.XXXXXXXX
256 if (lu_name_is_temp_file(name, namelen, true, 6)) {
259 } else if (lu_name_is_temp_file(name, namelen, false, 8)) {
261 } else if (lu_name_is_backup_file(name, namelen, &i)) {
262 LASSERT(i < namelen);
266 pg_id = lmv_hash_fnv1a(LMV_CRUSH_PG_COUNT, name, namelen);
268 /* distribute PG among all stripes pseudo-randomly, so they are almost
269 * evenly distributed, and when stripe count changes, only (delta /
270 * total) sub files need to be moved, herein 'delta' is added or removed
271 * stripe count, 'total' is total stripe count before change for
272 * removal, or count after change for addition.
274 for (i = 0; i < count; i++) {
275 straw = crush_hash(pg_id, i);
276 if (straw > highest_straw) {
277 highest_straw = straw;
281 LASSERT(idx < count);
286 /* directory layout may change in three ways:
287 * 1. directory migration, in its LMV source stripes are appended after
288 * target stripes, \a migrate_hash is source hash type, \a migrate_offset is
289 * target stripe count,
290 * 2. directory split, \a migrate_hash is hash type before split,
291 * \a migrate_offset is stripe count before split.
292 * 3. directory merge, \a migrate_hash is hash type after merge,
293 * \a migrate_offset is stripe count after merge.
296 __lmv_name_to_stripe_index(__u32 hash_type, __u32 stripe_count,
297 __u32 migrate_hash, __u32 migrate_offset,
298 const char *name, int namelen, bool new_layout)
300 __u32 saved_hash = hash_type;
301 __u32 saved_count = stripe_count;
302 int stripe_index = 0;
304 LASSERT(namelen > 0);
305 LASSERT(stripe_count > 0);
307 if (lmv_hash_is_splitting(hash_type)) {
309 hash_type = migrate_hash;
310 stripe_count = migrate_offset;
312 } else if (lmv_hash_is_merging(hash_type)) {
314 hash_type = migrate_hash;
315 stripe_count = migrate_offset;
317 } else if (lmv_hash_is_migrating(hash_type)) {
319 stripe_count = migrate_offset;
321 hash_type = migrate_hash;
322 stripe_count -= migrate_offset;
326 if (stripe_count > 1) {
327 switch (hash_type & LMV_HASH_TYPE_MASK) {
328 case LMV_HASH_TYPE_ALL_CHARS:
329 stripe_index = lmv_hash_all_chars(stripe_count, name,
332 case LMV_HASH_TYPE_FNV_1A_64:
333 stripe_index = lmv_hash_fnv1a(stripe_count, name,
336 case LMV_HASH_TYPE_CRUSH:
337 stripe_index = lmv_hash_crush(stripe_count, name,
345 LASSERT(stripe_index < stripe_count);
347 if (!new_layout && lmv_hash_is_migrating(saved_hash))
348 stripe_index += migrate_offset;
350 LASSERT(stripe_index < saved_count);
352 CDEBUG(D_INFO, "name %.*s hash=%#x/%#x idx=%d/%u/%u under %s layout\n",
353 namelen, name, saved_hash, migrate_hash, stripe_index,
354 saved_count, migrate_offset, new_layout ? "new" : "old");
359 static inline int lmv_name_to_stripe_index(struct lmv_mds_md_v1 *lmv,
360 const char *name, int namelen)
362 if (lmv->lmv_magic == LMV_MAGIC_V1)
363 return __lmv_name_to_stripe_index(lmv->lmv_hash_type,
364 lmv->lmv_stripe_count,
365 lmv->lmv_migrate_hash,
366 lmv->lmv_migrate_offset,
367 name, namelen, true);
369 if (lmv->lmv_magic == cpu_to_le32(LMV_MAGIC_V1))
370 return __lmv_name_to_stripe_index(
371 le32_to_cpu(lmv->lmv_hash_type),
372 le32_to_cpu(lmv->lmv_stripe_count),
373 le32_to_cpu(lmv->lmv_migrate_hash),
374 le32_to_cpu(lmv->lmv_migrate_offset),
375 name, namelen, true);
380 static inline int lmv_name_to_stripe_index_old(struct lmv_mds_md_v1 *lmv,
381 const char *name, int namelen)
383 if (lmv->lmv_magic == LMV_MAGIC_V1 ||
384 lmv->lmv_magic == LMV_MAGIC_STRIPE)
385 return __lmv_name_to_stripe_index(lmv->lmv_hash_type,
386 lmv->lmv_stripe_count,
387 lmv->lmv_migrate_hash,
388 lmv->lmv_migrate_offset,
389 name, namelen, false);
391 if (lmv->lmv_magic == cpu_to_le32(LMV_MAGIC_V1) ||
392 lmv->lmv_magic == cpu_to_le32(LMV_MAGIC_STRIPE))
393 return __lmv_name_to_stripe_index(
394 le32_to_cpu(lmv->lmv_hash_type),
395 le32_to_cpu(lmv->lmv_stripe_count),
396 le32_to_cpu(lmv->lmv_migrate_hash),
397 le32_to_cpu(lmv->lmv_migrate_offset),
398 name, namelen, false);
403 static inline bool lmv_user_magic_supported(__u32 lum_magic)
405 return lum_magic == LMV_USER_MAGIC ||
406 lum_magic == LMV_USER_MAGIC_SPECIFIC ||
407 lum_magic == LMV_MAGIC_FOREIGN;
410 #define LMV_DEBUG(mask, lmv, msg) \
412 "%s LMV: magic=%#x count=%u index=%u hash=%s:%#x version=%u migrate offset=%u migrate hash=%s:%u.\n",\
413 msg, (lmv)->lmv_magic, (lmv)->lmv_stripe_count, \
414 (lmv)->lmv_master_mdt_index, \
415 mdt_hash_name[(lmv)->lmv_hash_type & (LMV_HASH_TYPE_MAX - 1)],\
416 (lmv)->lmv_hash_type, (lmv)->lmv_layout_version, \
417 (lmv)->lmv_migrate_offset, \
418 mdt_hash_name[(lmv)->lmv_migrate_hash & (LMV_HASH_TYPE_MAX - 1)],\
419 (lmv)->lmv_migrate_hash)
421 /* master LMV is sane */
422 static inline bool lmv_is_sane(const struct lmv_mds_md_v1 *lmv)
427 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1)
430 if (le32_to_cpu(lmv->lmv_stripe_count) == 0)
433 if (!lmv_is_known_hash_type(le32_to_cpu(lmv->lmv_hash_type)))
438 LMV_DEBUG(D_ERROR, lmv, "insane");
442 /* LMV can be either master or stripe LMV */
443 static inline bool lmv_is_sane2(const struct lmv_mds_md_v1 *lmv)
448 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1 &&
449 le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_STRIPE)
452 if (le32_to_cpu(lmv->lmv_stripe_count) == 0)
455 if (!lmv_is_known_hash_type(le32_to_cpu(lmv->lmv_hash_type)))
460 LMV_DEBUG(D_ERROR, lmv, "insane");
464 static inline bool lmv_is_splitting(const struct lmv_mds_md_v1 *lmv)
466 if (!lmv_is_sane2(lmv))
469 return lmv_hash_is_splitting(cpu_to_le32(lmv->lmv_hash_type));
472 static inline bool lmv_is_merging(const struct lmv_mds_md_v1 *lmv)
474 if (!lmv_is_sane2(lmv))
477 return lmv_hash_is_merging(cpu_to_le32(lmv->lmv_hash_type));
480 static inline bool lmv_is_migrating(const struct lmv_mds_md_v1 *lmv)
482 if (!lmv_is_sane(lmv))
485 return lmv_hash_is_migrating(cpu_to_le32(lmv->lmv_hash_type));
488 static inline bool lmv_is_restriping(const struct lmv_mds_md_v1 *lmv)
490 if (!lmv_is_sane2(lmv))
493 return lmv_hash_is_splitting(cpu_to_le32(lmv->lmv_hash_type)) ||
494 lmv_hash_is_merging(cpu_to_le32(lmv->lmv_hash_type));
497 static inline bool lmv_is_layout_changing(const struct lmv_mds_md_v1 *lmv)
499 if (!lmv_is_sane2(lmv))
502 return lmv_hash_is_splitting(cpu_to_le32(lmv->lmv_hash_type)) ||
503 lmv_hash_is_merging(cpu_to_le32(lmv->lmv_hash_type)) ||
504 lmv_hash_is_migrating(cpu_to_le32(lmv->lmv_hash_type));
507 static inline bool lmv_is_fixed(const struct lmv_mds_md_v1 *lmv)
509 return cpu_to_le32(lmv->lmv_hash_type) & LMV_HASH_FLAG_FIXED;