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);
53 /* Find minimum stripe maxbytes value. For inactive or
54 * reconnecting targets use LUSTRE_EXT3_STRIPE_MAXBYTES. */
55 static loff_t lov_tgt_maxbytes(struct lov_tgt_desc *tgt)
57 struct obd_import *imp;
58 loff_t maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES;
63 imp = tgt->ltd_obd->u.cli.cl_import;
67 spin_lock(&imp->imp_lock);
68 if ((imp->imp_state == LUSTRE_IMP_FULL ||
69 imp->imp_state == LUSTRE_IMP_IDLE) &&
70 (imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_MAXBYTES) &&
71 imp->imp_connect_data.ocd_maxbytes > 0)
72 maxbytes = imp->imp_connect_data.ocd_maxbytes;
74 spin_unlock(&imp->imp_lock);
79 static int lsm_lmm_verify_v1v3(struct lov_mds_md *lmm, size_t lmm_size,
82 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
83 CERROR("bad stripe count %d\n", stripe_count);
84 lov_dump_lmm_common(D_WARNING, lmm);
88 if (lmm_oi_id(&lmm->lmm_oi) == 0) {
89 CERROR("zero object id\n");
90 lov_dump_lmm_common(D_WARNING, lmm);
94 if (lov_pattern(le32_to_cpu(lmm->lmm_pattern)) != LOV_PATTERN_MDT &&
95 lov_pattern(le32_to_cpu(lmm->lmm_pattern)) != LOV_PATTERN_RAID0) {
96 CERROR("bad striping pattern\n");
97 lov_dump_lmm_common(D_WARNING, lmm);
101 if (lmm->lmm_stripe_size == 0 ||
102 (le32_to_cpu(lmm->lmm_stripe_size)&(LOV_MIN_STRIPE_SIZE-1)) != 0) {
103 CERROR("bad stripe size %u\n",
104 le32_to_cpu(lmm->lmm_stripe_size));
105 lov_dump_lmm_common(D_WARNING, lmm);
111 static void lsme_free(struct lov_stripe_md_entry *lsme)
113 unsigned int stripe_count = lsme->lsme_stripe_count;
117 if (!lsme_inited(lsme) ||
118 lsme->lsme_pattern & LOV_PATTERN_F_RELEASED)
120 for (i = 0; i < stripe_count; i++)
121 OBD_SLAB_FREE_PTR(lsme->lsme_oinfo[i], lov_oinfo_slab);
123 lsme_size = offsetof(typeof(*lsme), lsme_oinfo[stripe_count]);
124 OBD_FREE_LARGE(lsme, lsme_size);
127 void lsm_free(struct lov_stripe_md *lsm)
129 unsigned int entry_count = lsm->lsm_entry_count;
133 for (i = 0; i < entry_count; i++)
134 lsme_free(lsm->lsm_entries[i]);
136 lsm_size = offsetof(typeof(*lsm), lsm_entries[entry_count]);
137 OBD_FREE(lsm, lsm_size);
141 * Unpack a struct lov_mds_md into a struct lov_stripe_md_entry.
143 * The caller should set id and extent.
145 static struct lov_stripe_md_entry *
146 lsme_unpack(struct lov_obd *lov, struct lov_mds_md *lmm, size_t buf_size,
147 const char *pool_name, bool inited, struct lov_ost_data_v1 *objects,
150 struct lov_stripe_md_entry *lsme;
152 loff_t min_stripe_maxbytes = 0;
156 unsigned int stripe_count;
160 magic = le32_to_cpu(lmm->lmm_magic);
161 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
162 RETURN(ERR_PTR(-EINVAL));
164 pattern = le32_to_cpu(lmm->lmm_pattern);
165 if (pattern & LOV_PATTERN_F_RELEASED || !inited)
168 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
170 if (buf_size < (magic == LOV_MAGIC_V1 ? sizeof(struct lov_mds_md_v1) :
171 sizeof(struct lov_mds_md_v3))) {
172 CERROR("LOV EA %s too small: %zu, need %u\n",
173 magic == LOV_MAGIC_V1 ? "V1" : "V3", buf_size,
174 lov_mds_md_size(stripe_count, magic == LOV_MAGIC_V1 ?
175 LOV_MAGIC_V1 : LOV_MAGIC_V3));
176 lov_dump_lmm_common(D_WARNING, lmm);
177 return ERR_PTR(-EINVAL);
180 rc = lsm_lmm_verify_v1v3(lmm, buf_size, stripe_count);
184 lsme_size = offsetof(typeof(*lsme), lsme_oinfo[stripe_count]);
185 OBD_ALLOC_LARGE(lsme, lsme_size);
187 RETURN(ERR_PTR(-ENOMEM));
189 lsme->lsme_magic = magic;
190 lsme->lsme_pattern = pattern;
191 lsme->lsme_flags = 0;
192 lsme->lsme_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
193 /* preserve the possible -1 stripe count for uninstantiated component */
194 lsme->lsme_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
195 lsme->lsme_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
197 if (pool_name != NULL) {
198 size_t pool_name_len;
200 pool_name_len = strlcpy(lsme->lsme_pool_name, pool_name,
201 sizeof(lsme->lsme_pool_name));
202 if (pool_name_len >= sizeof(lsme->lsme_pool_name))
203 GOTO(out_lsme, rc = -E2BIG);
206 /* with Data-on-MDT set maxbytes to stripe size */
207 if (lsme_is_dom(lsme)) {
208 lov_bytes = lsme->lsme_stripe_size;
212 for (i = 0; i < stripe_count; i++) {
213 struct lov_oinfo *loi;
214 struct lov_tgt_desc *ltd;
216 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
218 GOTO(out_lsme, rc = -ENOMEM);
220 lsme->lsme_oinfo[i] = loi;
222 ostid_le_to_cpu(&objects[i].l_ost_oi, &loi->loi_oi);
223 loi->loi_ost_idx = le32_to_cpu(objects[i].l_ost_idx);
224 loi->loi_ost_gen = le32_to_cpu(objects[i].l_ost_gen);
225 if (lov_oinfo_is_dummy(loi))
228 if (loi->loi_ost_idx >= lov->desc.ld_tgt_count &&
229 !lov2obd(lov)->obd_process_conf) {
230 CERROR("%s: OST index %d more than OST count %d\n",
231 (char*)lov->desc.ld_uuid.uuid,
232 loi->loi_ost_idx, lov->desc.ld_tgt_count);
233 lov_dump_lmm_v1(D_WARNING, lmm);
234 GOTO(out_lsme, rc = -EINVAL);
237 ltd = lov->lov_tgts[loi->loi_ost_idx];
239 CERROR("%s: OST index %d missing\n",
240 (char*)lov->desc.ld_uuid.uuid, loi->loi_ost_idx);
241 lov_dump_lmm_v1(D_WARNING, lmm);
245 lov_bytes = lov_tgt_maxbytes(ltd);
246 if (min_stripe_maxbytes == 0 || lov_bytes < min_stripe_maxbytes)
247 min_stripe_maxbytes = lov_bytes;
250 if (min_stripe_maxbytes == 0)
251 min_stripe_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES;
253 lov_bytes = min_stripe_maxbytes * stripe_count;
256 if (maxbytes != NULL) {
257 if (lov_bytes < min_stripe_maxbytes) /* handle overflow */
258 *maxbytes = MAX_LFS_FILESIZE;
260 *maxbytes = lov_bytes;
266 for (i = 0; i < stripe_count; i++) {
267 struct lov_oinfo *loi = lsme->lsme_oinfo[i];
270 OBD_SLAB_FREE_PTR(lsme->lsme_oinfo[i], lov_oinfo_slab);
272 OBD_FREE_LARGE(lsme, lsme_size);
278 lov_stripe_md *lsm_unpackmd_v1v3(struct lov_obd *lov, struct lov_mds_md *lmm,
279 size_t buf_size, const char *pool_name,
280 struct lov_ost_data_v1 *objects)
282 struct lov_stripe_md *lsm;
283 struct lov_stripe_md_entry *lsme;
289 pattern = le32_to_cpu(lmm->lmm_pattern);
291 lsme = lsme_unpack(lov, lmm, buf_size, pool_name, true, objects,
294 RETURN(ERR_CAST(lsme));
296 lsme->lsme_flags = LCME_FL_INIT;
297 lsme->lsme_extent.e_start = 0;
298 lsme->lsme_extent.e_end = LUSTRE_EOF;
300 lsm_size = offsetof(typeof(*lsm), lsm_entries[1]);
301 OBD_ALLOC(lsm, lsm_size);
303 GOTO(out_lsme, rc = -ENOMEM);
305 atomic_set(&lsm->lsm_refc, 1);
306 spin_lock_init(&lsm->lsm_lock);
307 lsm->lsm_maxbytes = maxbytes;
308 lmm_oi_le_to_cpu(&lsm->lsm_oi, &lmm->lmm_oi);
309 lsm->lsm_magic = le32_to_cpu(lmm->lmm_magic);
310 lsm->lsm_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
311 lsm->lsm_entry_count = 1;
312 lsm->lsm_is_released = pattern & LOV_PATTERN_F_RELEASED;
313 lsm->lsm_entries[0] = lsme;
323 static inline struct lov_stripe_md *
324 lsm_unpackmd_v1(struct lov_obd *lov, void *buf, size_t buf_size)
326 struct lov_mds_md_v1 *lmm = buf;
328 return lsm_unpackmd_v1v3(lov, buf, buf_size, NULL, lmm->lmm_objects);
331 const struct lsm_operations lsm_v1_ops = {
332 .lsm_unpackmd = lsm_unpackmd_v1,
336 struct lov_stripe_md *lsm_unpackmd_v3(struct lov_obd *lov, void *buf,
339 struct lov_mds_md_v3 *lmm = buf;
341 return lsm_unpackmd_v1v3(lov, buf, buf_size, lmm->lmm_pool_name,
345 const struct lsm_operations lsm_v3_ops = {
346 .lsm_unpackmd = lsm_unpackmd_v3,
349 static int lsm_verify_comp_md_v1(struct lov_comp_md_v1 *lcm,
352 unsigned int entry_count;
356 lcm_size = le32_to_cpu(lcm->lcm_size);
357 if (lcm_buf_size < lcm_size) {
358 CERROR("bad LCM buffer size %zu, expected %zu\n",
359 lcm_buf_size, lcm_size);
363 entry_count = le16_to_cpu(lcm->lcm_entry_count);
364 for (i = 0; i < entry_count; i++) {
365 struct lov_comp_md_entry_v1 *lcme = &lcm->lcm_entries[i];
369 blob_offset = le32_to_cpu(lcme->lcme_offset);
370 blob_size = le32_to_cpu(lcme->lcme_size);
372 if (lcm_size < blob_offset || lcm_size < blob_size ||
373 lcm_size < blob_offset + blob_size) {
374 CERROR("LCM entry %u has invalid blob: "
375 "LCM size = %zu, offset = %zu, size = %zu\n",
376 le32_to_cpu(lcme->lcme_id),
377 lcm_size, blob_offset, blob_size);
385 static struct lov_stripe_md_entry *
386 lsme_unpack_comp(struct lov_obd *lov, struct lov_mds_md *lmm,
387 size_t lmm_buf_size, bool inited, loff_t *maxbytes)
390 unsigned int stripe_count;
392 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
393 if (stripe_count == 0 &&
394 lov_pattern(le32_to_cpu(lmm->lmm_pattern)) != LOV_PATTERN_MDT)
395 RETURN(ERR_PTR(-EINVAL));
396 /* un-instantiated lmm contains no ost id info, i.e. lov_ost_data_v1 */
400 magic = le32_to_cpu(lmm->lmm_magic);
401 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
402 RETURN(ERR_PTR(-EINVAL));
404 if (lmm_buf_size < lov_mds_md_size(stripe_count, magic))
405 RETURN(ERR_PTR(-EINVAL));
407 if (magic == LOV_MAGIC_V1) {
408 return lsme_unpack(lov, lmm, lmm_buf_size, NULL,
409 inited, lmm->lmm_objects, maxbytes);
411 struct lov_mds_md_v3 *lmm3 = (struct lov_mds_md_v3 *)lmm;
413 return lsme_unpack(lov, lmm, lmm_buf_size, lmm3->lmm_pool_name,
414 inited, lmm3->lmm_objects, maxbytes);
418 static struct lov_stripe_md *
419 lsm_unpackmd_comp_md_v1(struct lov_obd *lov, void *buf, size_t buf_size)
421 struct lov_comp_md_v1 *lcm = buf;
422 struct lov_stripe_md *lsm;
424 unsigned int entry_count = 0;
429 rc = lsm_verify_comp_md_v1(buf, buf_size);
433 entry_count = le16_to_cpu(lcm->lcm_entry_count);
435 lsm_size = offsetof(typeof(*lsm), lsm_entries[entry_count]);
436 OBD_ALLOC(lsm, lsm_size);
438 return ERR_PTR(-ENOMEM);
440 atomic_set(&lsm->lsm_refc, 1);
441 spin_lock_init(&lsm->lsm_lock);
442 lsm->lsm_magic = le32_to_cpu(lcm->lcm_magic);
443 lsm->lsm_layout_gen = le32_to_cpu(lcm->lcm_layout_gen);
444 lsm->lsm_entry_count = entry_count;
445 lsm->lsm_mirror_count = le16_to_cpu(lcm->lcm_mirror_count);
446 lsm->lsm_flags = le16_to_cpu(lcm->lcm_flags);
447 lsm->lsm_is_released = true;
448 lsm->lsm_maxbytes = LLONG_MIN;
450 for (i = 0; i < entry_count; i++) {
451 struct lov_comp_md_entry_v1 *lcme = &lcm->lcm_entries[i];
452 struct lov_stripe_md_entry *lsme;
457 blob_offset = le32_to_cpu(lcme->lcme_offset);
458 blob_size = le32_to_cpu(lcme->lcme_size);
459 blob = (char *)lcm + blob_offset;
461 lsme = lsme_unpack_comp(lov, blob, blob_size,
462 le32_to_cpu(lcme->lcme_flags) &
464 (i == entry_count - 1) ? &maxbytes :
467 GOTO(out_lsm, rc = PTR_ERR(lsme));
469 if (!(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
470 lsm->lsm_is_released = false;
472 lsm->lsm_entries[i] = lsme;
473 lsme->lsme_id = le32_to_cpu(lcme->lcme_id);
474 lsme->lsme_flags = le32_to_cpu(lcme->lcme_flags);
475 if (lsme->lsme_flags & LCME_FL_NOSYNC)
476 lsme->lsme_timestamp =
477 le64_to_cpu(lcme->lcme_timestamp);
478 lu_extent_le_to_cpu(&lsme->lsme_extent, &lcme->lcme_extent);
480 if (i == entry_count - 1) {
481 lsm->lsm_maxbytes = (loff_t)lsme->lsme_extent.e_start +
483 /* the last component hasn't been defined, or
484 * lsm_maxbytes overflowed. */
485 if (!lsme_is_dom(lsme) &&
486 (lsme->lsme_extent.e_end != LUSTRE_EOF ||
488 (loff_t)lsme->lsme_extent.e_start))
489 lsm->lsm_maxbytes = MAX_LFS_FILESIZE;
496 for (i = 0; i < entry_count; i++)
497 if (lsm->lsm_entries[i] != NULL)
498 lsme_free(lsm->lsm_entries[i]);
500 OBD_FREE(lsm, lsm_size);
505 const struct lsm_operations lsm_comp_md_v1_ops = {
506 .lsm_unpackmd = lsm_unpackmd_comp_md_v1,
509 void dump_lsm(unsigned int level, const struct lov_stripe_md *lsm)
513 CDEBUG(level, "lsm %p, objid "DOSTID", maxbytes %#llx, magic 0x%08X, "
514 "refc: %d, entry: %u, layout_gen %u\n",
515 lsm, POSTID(&lsm->lsm_oi), lsm->lsm_maxbytes, lsm->lsm_magic,
516 atomic_read(&lsm->lsm_refc), lsm->lsm_entry_count,
517 lsm->lsm_layout_gen);
519 for (i = 0; i < lsm->lsm_entry_count; i++) {
520 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
522 CDEBUG(level, DEXT ": id: %u, flags: %x, "
523 "magic 0x%08X, layout_gen %u, "
524 "stripe count %u, sstripe size %u, "
525 "pool: ["LOV_POOLNAMEF"]\n",
526 PEXT(&lse->lsme_extent), lse->lsme_id, lse->lsme_flags,
527 lse->lsme_magic, lse->lsme_layout_gen,
528 lse->lsme_stripe_count, lse->lsme_stripe_size,
529 lse->lsme_pool_name);
530 if (!lsme_inited(lse) ||
531 lse->lsme_pattern & LOV_PATTERN_F_RELEASED)
533 for (j = 0; j < lse->lsme_stripe_count; j++) {
534 CDEBUG(level, " oinfo:%p: ostid: "DOSTID
535 " ost idx: %d gen: %d\n",
537 POSTID(&lse->lsme_oinfo[j]->loi_oi),
538 lse->lsme_oinfo[j]->loi_ost_idx,
539 lse->lsme_oinfo[j]->loi_ost_gen);
544 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset)
548 for (i = 0; i < lsm->lsm_entry_count; i++) {
549 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
551 if ((offset >= lse->lsme_extent.e_start &&
552 offset < lse->lsme_extent.e_end) ||
553 (offset == OBD_OBJECT_EOF &&
554 lse->lsme_extent.e_end == OBD_OBJECT_EOF))