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_connect_data.ocd_connect_flags & OBD_CONNECT_MAXBYTES) &&
70 imp->imp_connect_data.ocd_maxbytes > 0)
71 maxbytes = imp->imp_connect_data.ocd_maxbytes;
73 spin_unlock(&imp->imp_lock);
78 static int lsm_lmm_verify_v1v3(struct lov_mds_md *lmm, size_t lmm_size,
81 if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
82 CERROR("bad stripe count %d\n", stripe_count);
83 lov_dump_lmm_common(D_WARNING, lmm);
87 if (lmm_oi_id(&lmm->lmm_oi) == 0) {
88 CERROR("zero object id\n");
89 lov_dump_lmm_common(D_WARNING, lmm);
93 if (lov_pattern(le32_to_cpu(lmm->lmm_pattern)) != LOV_PATTERN_MDT &&
94 lov_pattern(le32_to_cpu(lmm->lmm_pattern)) != LOV_PATTERN_RAID0) {
95 CERROR("bad striping pattern\n");
96 lov_dump_lmm_common(D_WARNING, lmm);
100 if (lmm->lmm_stripe_size == 0 ||
101 (le32_to_cpu(lmm->lmm_stripe_size)&(LOV_MIN_STRIPE_SIZE-1)) != 0) {
102 CERROR("bad stripe size %u\n",
103 le32_to_cpu(lmm->lmm_stripe_size));
104 lov_dump_lmm_common(D_WARNING, lmm);
110 static void lsme_free(struct lov_stripe_md_entry *lsme)
112 unsigned int stripe_count = lsme->lsme_stripe_count;
116 if (!lsme_inited(lsme) ||
117 lsme->lsme_pattern & LOV_PATTERN_F_RELEASED)
119 for (i = 0; i < stripe_count; i++)
120 OBD_SLAB_FREE_PTR(lsme->lsme_oinfo[i], lov_oinfo_slab);
122 lsme_size = offsetof(typeof(*lsme), lsme_oinfo[stripe_count]);
123 OBD_FREE_LARGE(lsme, lsme_size);
126 void lsm_free(struct lov_stripe_md *lsm)
128 unsigned int entry_count = lsm->lsm_entry_count;
132 for (i = 0; i < entry_count; i++)
133 lsme_free(lsm->lsm_entries[i]);
135 lsm_size = offsetof(typeof(*lsm), lsm_entries[entry_count]);
136 OBD_FREE(lsm, lsm_size);
140 * Unpack a struct lov_mds_md into a struct lov_stripe_md_entry.
142 * The caller should set id and extent.
144 static struct lov_stripe_md_entry *
145 lsme_unpack(struct lov_obd *lov, struct lov_mds_md *lmm, size_t buf_size,
146 const char *pool_name, bool inited, struct lov_ost_data_v1 *objects,
149 struct lov_stripe_md_entry *lsme;
151 loff_t min_stripe_maxbytes = 0;
155 unsigned int stripe_count;
159 magic = le32_to_cpu(lmm->lmm_magic);
160 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
161 RETURN(ERR_PTR(-EINVAL));
163 pattern = le32_to_cpu(lmm->lmm_pattern);
164 if (pattern & LOV_PATTERN_F_RELEASED || !inited)
167 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
169 if (buf_size < (magic == LOV_MAGIC_V1 ? sizeof(struct lov_mds_md_v1) :
170 sizeof(struct lov_mds_md_v3))) {
171 CERROR("LOV EA %s too small: %zu, need %u\n",
172 magic == LOV_MAGIC_V1 ? "V1" : "V3", buf_size,
173 lov_mds_md_size(stripe_count, magic == LOV_MAGIC_V1 ?
174 LOV_MAGIC_V1 : LOV_MAGIC_V3));
175 lov_dump_lmm_common(D_WARNING, lmm);
176 return ERR_PTR(-EINVAL);
179 rc = lsm_lmm_verify_v1v3(lmm, buf_size, stripe_count);
183 lsme_size = offsetof(typeof(*lsme), lsme_oinfo[stripe_count]);
184 OBD_ALLOC_LARGE(lsme, lsme_size);
186 RETURN(ERR_PTR(-ENOMEM));
188 lsme->lsme_magic = magic;
189 lsme->lsme_pattern = pattern;
190 lsme->lsme_flags = 0;
191 lsme->lsme_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
192 /* preserve the possible -1 stripe count for uninstantiated component */
193 lsme->lsme_stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
194 lsme->lsme_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
196 if (pool_name != NULL) {
197 size_t pool_name_len;
199 pool_name_len = strlcpy(lsme->lsme_pool_name, pool_name,
200 sizeof(lsme->lsme_pool_name));
201 if (pool_name_len >= sizeof(lsme->lsme_pool_name))
202 GOTO(out_lsme, rc = -E2BIG);
205 /* with Data-on-MDT set maxbytes to stripe size */
206 if (lsme_is_dom(lsme)) {
207 lov_bytes = lsme->lsme_stripe_size;
211 for (i = 0; i < stripe_count; i++) {
212 struct lov_oinfo *loi;
213 struct lov_tgt_desc *ltd;
215 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
217 GOTO(out_lsme, rc = -ENOMEM);
219 lsme->lsme_oinfo[i] = loi;
221 ostid_le_to_cpu(&objects[i].l_ost_oi, &loi->loi_oi);
222 loi->loi_ost_idx = le32_to_cpu(objects[i].l_ost_idx);
223 loi->loi_ost_gen = le32_to_cpu(objects[i].l_ost_gen);
224 if (lov_oinfo_is_dummy(loi))
227 if (loi->loi_ost_idx >= lov->desc.ld_tgt_count &&
228 !lov2obd(lov)->obd_process_conf) {
229 CERROR("%s: OST index %d more than OST count %d\n",
230 (char*)lov->desc.ld_uuid.uuid,
231 loi->loi_ost_idx, lov->desc.ld_tgt_count);
232 lov_dump_lmm_v1(D_WARNING, lmm);
233 GOTO(out_lsme, rc = -EINVAL);
236 ltd = lov->lov_tgts[loi->loi_ost_idx];
238 CERROR("%s: OST index %d missing\n",
239 (char*)lov->desc.ld_uuid.uuid, loi->loi_ost_idx);
240 lov_dump_lmm_v1(D_WARNING, lmm);
244 lov_bytes = lov_tgt_maxbytes(ltd);
245 if (min_stripe_maxbytes == 0 || lov_bytes < min_stripe_maxbytes)
246 min_stripe_maxbytes = lov_bytes;
249 if (min_stripe_maxbytes == 0)
250 min_stripe_maxbytes = LUSTRE_EXT3_STRIPE_MAXBYTES;
252 lov_bytes = min_stripe_maxbytes * stripe_count;
255 if (maxbytes != NULL) {
256 if (lov_bytes < min_stripe_maxbytes) /* handle overflow */
257 *maxbytes = MAX_LFS_FILESIZE;
259 *maxbytes = lov_bytes;
265 for (i = 0; i < stripe_count; i++) {
266 struct lov_oinfo *loi = lsme->lsme_oinfo[i];
269 OBD_SLAB_FREE_PTR(lsme->lsme_oinfo[i], lov_oinfo_slab);
271 OBD_FREE_LARGE(lsme, lsme_size);
277 lov_stripe_md *lsm_unpackmd_v1v3(struct lov_obd *lov, struct lov_mds_md *lmm,
278 size_t buf_size, const char *pool_name,
279 struct lov_ost_data_v1 *objects)
281 struct lov_stripe_md *lsm;
282 struct lov_stripe_md_entry *lsme;
288 pattern = le32_to_cpu(lmm->lmm_pattern);
290 lsme = lsme_unpack(lov, lmm, buf_size, pool_name, true, objects,
293 RETURN(ERR_CAST(lsme));
295 lsme->lsme_flags = LCME_FL_INIT;
296 lsme->lsme_extent.e_start = 0;
297 lsme->lsme_extent.e_end = LUSTRE_EOF;
299 lsm_size = offsetof(typeof(*lsm), lsm_entries[1]);
300 OBD_ALLOC(lsm, lsm_size);
302 GOTO(out_lsme, rc = -ENOMEM);
304 atomic_set(&lsm->lsm_refc, 1);
305 spin_lock_init(&lsm->lsm_lock);
306 lsm->lsm_maxbytes = maxbytes;
307 lmm_oi_le_to_cpu(&lsm->lsm_oi, &lmm->lmm_oi);
308 lsm->lsm_magic = le32_to_cpu(lmm->lmm_magic);
309 lsm->lsm_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
310 lsm->lsm_entry_count = 1;
311 lsm->lsm_is_released = pattern & LOV_PATTERN_F_RELEASED;
312 lsm->lsm_entries[0] = lsme;
322 static inline struct lov_stripe_md *
323 lsm_unpackmd_v1(struct lov_obd *lov, void *buf, size_t buf_size)
325 struct lov_mds_md_v1 *lmm = buf;
327 return lsm_unpackmd_v1v3(lov, buf, buf_size, NULL, lmm->lmm_objects);
330 const struct lsm_operations lsm_v1_ops = {
331 .lsm_unpackmd = lsm_unpackmd_v1,
335 struct lov_stripe_md *lsm_unpackmd_v3(struct lov_obd *lov, void *buf,
338 struct lov_mds_md_v3 *lmm = buf;
340 return lsm_unpackmd_v1v3(lov, buf, buf_size, lmm->lmm_pool_name,
344 const struct lsm_operations lsm_v3_ops = {
345 .lsm_unpackmd = lsm_unpackmd_v3,
348 static int lsm_verify_comp_md_v1(struct lov_comp_md_v1 *lcm,
351 unsigned int entry_count;
355 lcm_size = le32_to_cpu(lcm->lcm_size);
356 if (lcm_buf_size < lcm_size) {
357 CERROR("bad LCM buffer size %zu, expected %zu\n",
358 lcm_buf_size, lcm_size);
362 entry_count = le16_to_cpu(lcm->lcm_entry_count);
363 for (i = 0; i < entry_count; i++) {
364 struct lov_comp_md_entry_v1 *lcme = &lcm->lcm_entries[i];
368 blob_offset = le32_to_cpu(lcme->lcme_offset);
369 blob_size = le32_to_cpu(lcme->lcme_size);
371 if (lcm_size < blob_offset || lcm_size < blob_size ||
372 lcm_size < blob_offset + blob_size) {
373 CERROR("LCM entry %u has invalid blob: "
374 "LCM size = %zu, offset = %zu, size = %zu\n",
375 le32_to_cpu(lcme->lcme_id),
376 lcm_size, blob_offset, blob_size);
384 static struct lov_stripe_md_entry *
385 lsme_unpack_comp(struct lov_obd *lov, struct lov_mds_md *lmm,
386 size_t lmm_buf_size, bool inited, loff_t *maxbytes)
389 unsigned int stripe_count;
391 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
392 if (stripe_count == 0 &&
393 lov_pattern(le32_to_cpu(lmm->lmm_pattern)) != LOV_PATTERN_MDT)
394 RETURN(ERR_PTR(-EINVAL));
395 /* un-instantiated lmm contains no ost id info, i.e. lov_ost_data_v1 */
399 magic = le32_to_cpu(lmm->lmm_magic);
400 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
401 RETURN(ERR_PTR(-EINVAL));
403 if (lmm_buf_size < lov_mds_md_size(stripe_count, magic))
404 RETURN(ERR_PTR(-EINVAL));
406 if (magic == LOV_MAGIC_V1) {
407 return lsme_unpack(lov, lmm, lmm_buf_size, NULL,
408 inited, lmm->lmm_objects, maxbytes);
410 struct lov_mds_md_v3 *lmm3 = (struct lov_mds_md_v3 *)lmm;
412 return lsme_unpack(lov, lmm, lmm_buf_size, lmm3->lmm_pool_name,
413 inited, lmm3->lmm_objects, maxbytes);
417 static struct lov_stripe_md *
418 lsm_unpackmd_comp_md_v1(struct lov_obd *lov, void *buf, size_t buf_size)
420 struct lov_comp_md_v1 *lcm = buf;
421 struct lov_stripe_md *lsm;
423 unsigned int entry_count = 0;
428 rc = lsm_verify_comp_md_v1(buf, buf_size);
432 entry_count = le16_to_cpu(lcm->lcm_entry_count);
434 lsm_size = offsetof(typeof(*lsm), lsm_entries[entry_count]);
435 OBD_ALLOC(lsm, lsm_size);
437 return ERR_PTR(-ENOMEM);
439 atomic_set(&lsm->lsm_refc, 1);
440 spin_lock_init(&lsm->lsm_lock);
441 lsm->lsm_magic = le32_to_cpu(lcm->lcm_magic);
442 lsm->lsm_layout_gen = le32_to_cpu(lcm->lcm_layout_gen);
443 lsm->lsm_entry_count = entry_count;
444 lsm->lsm_mirror_count = le16_to_cpu(lcm->lcm_mirror_count);
445 lsm->lsm_flags = le16_to_cpu(lcm->lcm_flags);
446 lsm->lsm_is_released = true;
447 lsm->lsm_maxbytes = LLONG_MIN;
449 for (i = 0; i < entry_count; i++) {
450 struct lov_comp_md_entry_v1 *lcme = &lcm->lcm_entries[i];
451 struct lov_stripe_md_entry *lsme;
456 blob_offset = le32_to_cpu(lcme->lcme_offset);
457 blob_size = le32_to_cpu(lcme->lcme_size);
458 blob = (char *)lcm + blob_offset;
460 lsme = lsme_unpack_comp(lov, blob, blob_size,
461 le32_to_cpu(lcme->lcme_flags) &
463 (i == entry_count - 1) ? &maxbytes :
466 GOTO(out_lsm, rc = PTR_ERR(lsme));
468 if (!(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
469 lsm->lsm_is_released = false;
471 lsm->lsm_entries[i] = lsme;
472 lsme->lsme_id = le32_to_cpu(lcme->lcme_id);
473 lsme->lsme_flags = le32_to_cpu(lcme->lcme_flags);
474 lu_extent_le_to_cpu(&lsme->lsme_extent, &lcme->lcme_extent);
476 if (i == entry_count - 1) {
477 lsm->lsm_maxbytes = (loff_t)lsme->lsme_extent.e_start +
479 /* the last component hasn't been defined, or
480 * lsm_maxbytes overflowed. */
481 if (!lsme_is_dom(lsme) &&
482 (lsme->lsme_extent.e_end != LUSTRE_EOF ||
484 (loff_t)lsme->lsme_extent.e_start))
485 lsm->lsm_maxbytes = MAX_LFS_FILESIZE;
492 for (i = 0; i < entry_count; i++)
493 if (lsm->lsm_entries[i] != NULL)
494 lsme_free(lsm->lsm_entries[i]);
496 OBD_FREE(lsm, lsm_size);
501 const struct lsm_operations lsm_comp_md_v1_ops = {
502 .lsm_unpackmd = lsm_unpackmd_comp_md_v1,
505 void dump_lsm(unsigned int level, const struct lov_stripe_md *lsm)
509 CDEBUG(level, "lsm %p, objid "DOSTID", maxbytes %#llx, magic 0x%08X, "
510 "refc: %d, entry: %u, layout_gen %u\n",
511 lsm, POSTID(&lsm->lsm_oi), lsm->lsm_maxbytes, lsm->lsm_magic,
512 atomic_read(&lsm->lsm_refc), lsm->lsm_entry_count,
513 lsm->lsm_layout_gen);
515 for (i = 0; i < lsm->lsm_entry_count; i++) {
516 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
518 CDEBUG(level, DEXT ": id: %u, flags: %x, "
519 "magic 0x%08X, layout_gen %u, "
520 "stripe count %u, sstripe size %u, "
521 "pool: ["LOV_POOLNAMEF"]\n",
522 PEXT(&lse->lsme_extent), lse->lsme_id, lse->lsme_flags,
523 lse->lsme_magic, lse->lsme_layout_gen,
524 lse->lsme_stripe_count, lse->lsme_stripe_size,
525 lse->lsme_pool_name);
526 if (!lsme_inited(lse) ||
527 lse->lsme_pattern & LOV_PATTERN_F_RELEASED)
529 for (j = 0; j < lse->lsme_stripe_count; j++) {
530 CDEBUG(level, " oinfo:%p: ostid: "DOSTID
531 " ost idx: %d gen: %d\n",
533 POSTID(&lse->lsme_oinfo[j]->loi_oi),
534 lse->lsme_oinfo[j]->loi_ost_idx,
535 lse->lsme_oinfo[j]->loi_ost_gen);
540 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset)
544 for (i = 0; i < lsm->lsm_entry_count; i++) {
545 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
547 if ((offset >= lse->lsme_extent.e_start &&
548 offset < lse->lsme_extent.e_end) ||
549 (offset == OBD_OBJECT_EOF &&
550 lse->lsme_extent.e_end == OBD_OBJECT_EOF))