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, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Wang Di <wangdi@clusterfs.com>
37 #define DEBUG_SUBSYSTEM S_LOV
39 #include <linux/math64.h>
40 #include <linux/sort.h>
41 #include <libcfs/libcfs.h>
43 #include <obd_class.h>
44 #include "lov_internal.h"
47 lu_extent_le_to_cpu(struct lu_extent *dst, const struct lu_extent *src)
49 dst->e_start = le64_to_cpu(src->e_start);
50 dst->e_end = le64_to_cpu(src->e_end);
54 * Find minimum stripe maxbytes value. For inactive or
55 * reconnecting targets use LUSTRE_EXT3_STRIPE_MAXBYTES.
57 static loff_t lov_tgt_maxbytes(struct lov_tgt_desc *tgt)
59 struct obd_import *imp;
60 loff_t maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES;
65 imp = tgt->ltd_obd->u.cli.cl_import;
69 spin_lock(&imp->imp_lock);
70 if ((imp->imp_state == LUSTRE_IMP_FULL ||
71 imp->imp_state == LUSTRE_IMP_IDLE) &&
72 (imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_MAXBYTES) &&
73 imp->imp_connect_data.ocd_maxbytes > 0)
74 maxbytes = imp->imp_connect_data.ocd_maxbytes;
76 spin_unlock(&imp->imp_lock);
81 static int lsm_lmm_verify_v1v3(struct lov_mds_md *lmm, size_t lmm_size,
86 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
88 CERROR("lov: bad stripe count %d: rc = %d\n",
90 lov_dump_lmm_common(D_WARNING, lmm);
94 if (lmm_oi_id(&lmm->lmm_oi) == 0) {
96 CERROR("lov: zero object id: rc = %d\n", rc);
97 lov_dump_lmm_common(D_WARNING, lmm);
101 if (lov_pattern(le32_to_cpu(lmm->lmm_pattern)) != LOV_PATTERN_MDT &&
102 lov_pattern(le32_to_cpu(lmm->lmm_pattern)) != LOV_PATTERN_RAID0 &&
103 lov_pattern(le32_to_cpu(lmm->lmm_pattern)) !=
104 (LOV_PATTERN_RAID0 | LOV_PATTERN_OVERSTRIPING)) {
106 CERROR("lov: unrecognized striping pattern: rc = %d\n", rc);
107 lov_dump_lmm_common(D_WARNING, lmm);
111 if (lmm->lmm_stripe_size == 0 ||
112 (le32_to_cpu(lmm->lmm_stripe_size)&(LOV_MIN_STRIPE_SIZE-1)) != 0) {
114 CERROR("lov: bad stripe size %u: rc = %d\n",
115 le32_to_cpu(lmm->lmm_stripe_size), rc);
116 lov_dump_lmm_common(D_WARNING, lmm);
124 static void lsme_free(struct lov_stripe_md_entry *lsme)
126 unsigned int stripe_count = lsme->lsme_stripe_count;
130 if (!lsme_inited(lsme) ||
131 lsme->lsme_pattern & LOV_PATTERN_F_RELEASED)
133 for (i = 0; i < stripe_count; i++)
134 OBD_SLAB_FREE_PTR(lsme->lsme_oinfo[i], lov_oinfo_slab);
136 lsme_size = offsetof(typeof(*lsme), lsme_oinfo[stripe_count]);
137 OBD_FREE_LARGE(lsme, lsme_size);
140 void lsm_free(struct lov_stripe_md *lsm)
142 unsigned int entry_count = lsm->lsm_entry_count;
146 if (lsm->lsm_magic == LOV_MAGIC_FOREIGN) {
147 OBD_FREE_LARGE(lsm_foreign(lsm), lsm->lsm_foreign_size);
149 for (i = 0; i < entry_count; i++)
150 lsme_free(lsm->lsm_entries[i]);
153 lsm_size = lsm->lsm_magic == LOV_MAGIC_FOREIGN ?
154 offsetof(typeof(*lsm), lsm_entries[1]) :
155 offsetof(typeof(*lsm), lsm_entries[entry_count]);
156 OBD_FREE(lsm, lsm_size);
160 * Unpack a struct lov_mds_md into a struct lov_stripe_md_entry.
162 * The caller should set id and extent.
164 static struct lov_stripe_md_entry *
165 lsme_unpack(struct lov_obd *lov, struct lov_mds_md *lmm, size_t buf_size,
166 const char *pool_name, bool inited, struct lov_ost_data_v1 *objects,
169 struct lov_stripe_md_entry *lsme;
171 loff_t min_stripe_maxbytes = 0;
175 unsigned int stripe_count;
179 magic = le32_to_cpu(lmm->lmm_magic);
180 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
181 RETURN(ERR_PTR(-EINVAL));
183 pattern = le32_to_cpu(lmm->lmm_pattern);
184 if (pattern & LOV_PATTERN_F_RELEASED || !inited)
187 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
189 if (buf_size < (magic == LOV_MAGIC_V1 ? sizeof(struct lov_mds_md_v1) :
190 sizeof(struct lov_mds_md_v3))) {
191 CERROR("LOV EA %s too small: %zu, need %u\n",
192 magic == LOV_MAGIC_V1 ? "V1" : "V3", buf_size,
193 lov_mds_md_size(stripe_count, magic == LOV_MAGIC_V1 ?
194 LOV_MAGIC_V1 : LOV_MAGIC_V3));
195 lov_dump_lmm_common(D_WARNING, lmm);
196 return ERR_PTR(-EINVAL);
199 rc = lsm_lmm_verify_v1v3(lmm, buf_size, stripe_count);
203 lsme_size = offsetof(typeof(*lsme), lsme_oinfo[stripe_count]);
204 OBD_ALLOC_LARGE(lsme, lsme_size);
206 RETURN(ERR_PTR(-ENOMEM));
208 lsme->lsme_magic = magic;
209 lsme->lsme_pattern = pattern;
210 lsme->lsme_flags = 0;
211 lsme->lsme_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
212 /* preserve the possible -1 stripe count for uninstantiated component */
213 lsme->lsme_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
214 lsme->lsme_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
217 size_t pool_name_len;
219 pool_name_len = strlcpy(lsme->lsme_pool_name, pool_name,
220 sizeof(lsme->lsme_pool_name));
221 if (pool_name_len >= sizeof(lsme->lsme_pool_name))
222 GOTO(out_lsme, rc = -E2BIG);
225 /* with Data-on-MDT set maxbytes to stripe size */
226 if (lsme_is_dom(lsme)) {
228 lov_bytes = lsme->lsme_stripe_size;
235 for (i = 0; i < stripe_count; i++) {
236 struct lov_oinfo *loi;
237 struct lov_tgt_desc *ltd;
239 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
241 GOTO(out_lsme, rc = -ENOMEM);
243 lsme->lsme_oinfo[i] = loi;
245 ostid_le_to_cpu(&objects[i].l_ost_oi, &loi->loi_oi);
246 loi->loi_ost_idx = le32_to_cpu(objects[i].l_ost_idx);
247 loi->loi_ost_gen = le32_to_cpu(objects[i].l_ost_gen);
248 if (lov_oinfo_is_dummy(loi))
251 if (loi->loi_ost_idx >= lov->desc.ld_tgt_count &&
252 !lov2obd(lov)->obd_process_conf) {
253 CERROR("%s: OST index %d more than OST count %d\n",
254 (char*)lov->desc.ld_uuid.uuid,
255 loi->loi_ost_idx, lov->desc.ld_tgt_count);
256 lov_dump_lmm_v1(D_WARNING, lmm);
257 GOTO(out_lsme, rc = -EINVAL);
260 ltd = lov->lov_tgts[loi->loi_ost_idx];
262 CERROR("%s: OST index %d missing\n",
263 (char*)lov->desc.ld_uuid.uuid, loi->loi_ost_idx);
264 lov_dump_lmm_v1(D_WARNING, lmm);
268 lov_bytes = lov_tgt_maxbytes(ltd);
269 if (min_stripe_maxbytes == 0 || lov_bytes < min_stripe_maxbytes)
270 min_stripe_maxbytes = lov_bytes;
274 if (min_stripe_maxbytes == 0)
275 min_stripe_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES;
277 if (stripe_count == 0)
278 stripe_count = lov->desc.ld_tgt_count;
280 if (min_stripe_maxbytes <= LLONG_MAX / stripe_count)
281 lov_bytes = min_stripe_maxbytes * stripe_count;
283 lov_bytes = MAX_LFS_FILESIZE;
285 *maxbytes = min_t(loff_t, lov_bytes, MAX_LFS_FILESIZE);
292 for (i = 0; i < stripe_count; i++) {
293 struct lov_oinfo *loi = lsme->lsme_oinfo[i];
296 OBD_SLAB_FREE_PTR(lsme->lsme_oinfo[i], lov_oinfo_slab);
298 OBD_FREE_LARGE(lsme, lsme_size);
304 lov_stripe_md *lsm_unpackmd_v1v3(struct lov_obd *lov, struct lov_mds_md *lmm,
305 size_t buf_size, const char *pool_name,
306 struct lov_ost_data_v1 *objects)
308 struct lov_stripe_md *lsm;
309 struct lov_stripe_md_entry *lsme;
315 pattern = le32_to_cpu(lmm->lmm_pattern);
317 lsme = lsme_unpack(lov, lmm, buf_size, pool_name, true, objects,
320 RETURN(ERR_CAST(lsme));
322 lsme->lsme_flags = LCME_FL_INIT;
323 lsme->lsme_extent.e_start = 0;
324 lsme->lsme_extent.e_end = LUSTRE_EOF;
326 lsm_size = offsetof(typeof(*lsm), lsm_entries[1]);
327 OBD_ALLOC(lsm, lsm_size);
329 GOTO(out_lsme, rc = -ENOMEM);
331 atomic_set(&lsm->lsm_refc, 1);
332 spin_lock_init(&lsm->lsm_lock);
333 lsm->lsm_maxbytes = maxbytes;
334 lmm_oi_le_to_cpu(&lsm->lsm_oi, &lmm->lmm_oi);
335 lsm->lsm_magic = le32_to_cpu(lmm->lmm_magic);
336 lsm->lsm_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
337 lsm->lsm_entry_count = 1;
338 lsm->lsm_is_released = pattern & LOV_PATTERN_F_RELEASED;
339 lsm->lsm_entries[0] = lsme;
349 static inline struct lov_stripe_md *
350 lsm_unpackmd_v1(struct lov_obd *lov, void *buf, size_t buf_size)
352 struct lov_mds_md_v1 *lmm = buf;
354 return lsm_unpackmd_v1v3(lov, buf, buf_size, NULL, lmm->lmm_objects);
357 const struct lsm_operations lsm_v1_ops = {
358 .lsm_unpackmd = lsm_unpackmd_v1,
362 struct lov_stripe_md *lsm_unpackmd_v3(struct lov_obd *lov, void *buf,
365 struct lov_mds_md_v3 *lmm = buf;
367 return lsm_unpackmd_v1v3(lov, buf, buf_size, lmm->lmm_pool_name,
371 const struct lsm_operations lsm_v3_ops = {
372 .lsm_unpackmd = lsm_unpackmd_v3,
375 static int lsm_verify_comp_md_v1(struct lov_comp_md_v1 *lcm,
378 unsigned int entry_count;
382 lcm_size = le32_to_cpu(lcm->lcm_size);
383 if (lcm_buf_size < lcm_size) {
384 CERROR("bad LCM buffer size %zu, expected %zu\n",
385 lcm_buf_size, lcm_size);
389 entry_count = le16_to_cpu(lcm->lcm_entry_count);
390 for (i = 0; i < entry_count; i++) {
391 struct lov_comp_md_entry_v1 *lcme = &lcm->lcm_entries[i];
395 blob_offset = le32_to_cpu(lcme->lcme_offset);
396 blob_size = le32_to_cpu(lcme->lcme_size);
398 if (lcm_size < blob_offset || lcm_size < blob_size ||
399 lcm_size < blob_offset + blob_size) {
400 CERROR("LCM entry %u has invalid blob: "
401 "LCM size = %zu, offset = %zu, size = %zu\n",
402 le32_to_cpu(lcme->lcme_id),
403 lcm_size, blob_offset, blob_size);
411 static struct lov_stripe_md_entry *
412 lsme_unpack_comp(struct lov_obd *lov, struct lov_mds_md *lmm,
413 size_t lmm_buf_size, bool inited, loff_t *maxbytes)
416 unsigned int stripe_count;
418 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
419 if (stripe_count == 0 &&
420 lov_pattern(le32_to_cpu(lmm->lmm_pattern)) != LOV_PATTERN_MDT)
421 RETURN(ERR_PTR(-EINVAL));
422 /* un-instantiated lmm contains no ost id info, i.e. lov_ost_data_v1 */
426 magic = le32_to_cpu(lmm->lmm_magic);
427 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
428 RETURN(ERR_PTR(-EINVAL));
430 if (lmm_buf_size < lov_mds_md_size(stripe_count, magic))
431 RETURN(ERR_PTR(-EINVAL));
433 if (magic == LOV_MAGIC_V1) {
434 return lsme_unpack(lov, lmm, lmm_buf_size, NULL,
435 inited, lmm->lmm_objects, maxbytes);
437 struct lov_mds_md_v3 *lmm3 = (struct lov_mds_md_v3 *)lmm;
439 return lsme_unpack(lov, lmm, lmm_buf_size, lmm3->lmm_pool_name,
440 inited, lmm3->lmm_objects, maxbytes);
444 static struct lov_stripe_md *
445 lsm_unpackmd_comp_md_v1(struct lov_obd *lov, void *buf, size_t buf_size)
447 struct lov_comp_md_v1 *lcm = buf;
448 struct lov_stripe_md *lsm;
450 unsigned int entry_count = 0;
455 rc = lsm_verify_comp_md_v1(buf, buf_size);
459 entry_count = le16_to_cpu(lcm->lcm_entry_count);
461 lsm_size = offsetof(typeof(*lsm), lsm_entries[entry_count]);
462 OBD_ALLOC(lsm, lsm_size);
464 return ERR_PTR(-ENOMEM);
466 atomic_set(&lsm->lsm_refc, 1);
467 spin_lock_init(&lsm->lsm_lock);
468 lsm->lsm_magic = le32_to_cpu(lcm->lcm_magic);
469 lsm->lsm_layout_gen = le32_to_cpu(lcm->lcm_layout_gen);
470 lsm->lsm_entry_count = entry_count;
471 lsm->lsm_mirror_count = le16_to_cpu(lcm->lcm_mirror_count);
472 lsm->lsm_flags = le16_to_cpu(lcm->lcm_flags);
473 lsm->lsm_is_released = true;
474 lsm->lsm_maxbytes = LLONG_MIN;
476 for (i = 0; i < entry_count; i++) {
477 struct lov_comp_md_entry_v1 *lcme = &lcm->lcm_entries[i];
478 struct lov_stripe_md_entry *lsme;
483 blob_offset = le32_to_cpu(lcme->lcme_offset);
484 blob_size = le32_to_cpu(lcme->lcme_size);
485 blob = (char *)lcm + blob_offset;
487 lsme = lsme_unpack_comp(lov, blob, blob_size,
488 le32_to_cpu(lcme->lcme_flags) &
490 (i == entry_count - 1) ? &maxbytes :
493 GOTO(out_lsm, rc = PTR_ERR(lsme));
495 if (!(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
496 lsm->lsm_is_released = false;
498 lsm->lsm_entries[i] = lsme;
499 lsme->lsme_id = le32_to_cpu(lcme->lcme_id);
500 lsme->lsme_flags = le32_to_cpu(lcme->lcme_flags);
501 if (lsme->lsme_flags & LCME_FL_NOSYNC)
502 lsme->lsme_timestamp =
503 le64_to_cpu(lcme->lcme_timestamp);
504 lu_extent_le_to_cpu(&lsme->lsme_extent, &lcme->lcme_extent);
506 if (i == entry_count - 1) {
507 lsm->lsm_maxbytes = (loff_t)lsme->lsme_extent.e_start +
510 * the last component hasn't been defined, or
511 * lsm_maxbytes overflowed.
513 if (!lsme_is_dom(lsme) &&
514 (lsme->lsme_extent.e_end != LUSTRE_EOF ||
516 (loff_t)lsme->lsme_extent.e_start))
517 lsm->lsm_maxbytes = MAX_LFS_FILESIZE;
524 for (i = 0; i < entry_count; i++)
525 if (lsm->lsm_entries[i])
526 lsme_free(lsm->lsm_entries[i]);
528 OBD_FREE(lsm, lsm_size);
533 const struct lsm_operations lsm_comp_md_v1_ops = {
534 .lsm_unpackmd = lsm_unpackmd_comp_md_v1,
538 lov_stripe_md *lsm_unpackmd_foreign(struct lov_obd *lov, void *buf,
541 struct lov_foreign_md *lfm = buf;
542 struct lov_stripe_md *lsm;
544 struct lov_stripe_md_entry *lsme;
546 lsm_size = offsetof(typeof(*lsm), lsm_entries[1]);
547 OBD_ALLOC(lsm, lsm_size);
549 RETURN(ERR_PTR(-ENOMEM));
551 atomic_set(&lsm->lsm_refc, 1);
552 spin_lock_init(&lsm->lsm_lock);
553 lsm->lsm_magic = le32_to_cpu(lfm->lfm_magic);
554 lsm->lsm_foreign_size = foreign_size_le(lfm);
556 /* alloc for full foreign EA including format fields */
557 OBD_ALLOC_LARGE(lsme, lsm->lsm_foreign_size);
559 OBD_FREE(lsm, lsm_size);
560 RETURN(ERR_PTR(-ENOMEM));
563 /* copy full foreign EA including format fields */
564 memcpy(lsme, buf, lsm->lsm_foreign_size);
566 lsm_foreign(lsm) = lsme;
571 const struct lsm_operations lsm_foreign_ops = {
572 .lsm_unpackmd = lsm_unpackmd_foreign,
575 void dump_lsm(unsigned int level, const struct lov_stripe_md *lsm)
580 "lsm %p, objid "DOSTID", maxbytes %#llx, magic 0x%08X, refc: %d, entry: %u, layout_gen %u\n",
581 lsm, POSTID(&lsm->lsm_oi), lsm->lsm_maxbytes, lsm->lsm_magic,
582 atomic_read(&lsm->lsm_refc), lsm->lsm_entry_count,
583 lsm->lsm_layout_gen);
585 if (lsm->lsm_magic == LOV_MAGIC_FOREIGN) {
586 struct lov_foreign_md *lfm = (void *)lsm_foreign(lsm);
589 "foreign LOV EA, magic %x, length %u, type %x, flags %x, value '%.*s'\n",
590 lfm->lfm_magic, lfm->lfm_length, lfm->lfm_type,
591 lfm->lfm_flags, lfm->lfm_length, lfm->lfm_value);
595 for (i = 0; i < lsm->lsm_entry_count; i++) {
596 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
598 CDEBUG(level, DEXT ": id: %u, flags: %x, "
599 "magic 0x%08X, layout_gen %u, "
600 "stripe count %u, sstripe size %u, "
601 "pool: ["LOV_POOLNAMEF"]\n",
602 PEXT(&lse->lsme_extent), lse->lsme_id, lse->lsme_flags,
603 lse->lsme_magic, lse->lsme_layout_gen,
604 lse->lsme_stripe_count, lse->lsme_stripe_size,
605 lse->lsme_pool_name);
606 if (!lsme_inited(lse) ||
607 lse->lsme_pattern & LOV_PATTERN_F_RELEASED)
609 for (j = 0; j < lse->lsme_stripe_count; j++) {
610 CDEBUG(level, " oinfo:%p: ostid: "DOSTID
611 " ost idx: %d gen: %d\n",
613 POSTID(&lse->lsme_oinfo[j]->loi_oi),
614 lse->lsme_oinfo[j]->loi_ost_idx,
615 lse->lsme_oinfo[j]->loi_ost_gen);
620 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset)
624 for (i = 0; i < lsm->lsm_entry_count; i++) {
625 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
627 if ((offset >= lse->lsme_extent.e_start &&
628 offset < lse->lsme_extent.e_end) ||
629 (offset == OBD_OBJECT_EOF &&
630 lse->lsme_extent.e_end == OBD_OBJECT_EOF))