4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * All rights reserved. This program and the accompanying materials
7 * are made available under the terms of the GNU Lesser General Public License
8 * (LGPL) version 2.1 or (at your discretion) any later version.
9 * (LGPL) version 2.1 accompanies this distribution, and is available at
10 * http://www.gnu.org/licenses/lgpl-2.1.html
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
20 * lustre/utils/liblustreapi_layout.c
22 * lustreapi library for layout calls for interacting with the layout of
23 * Lustre files while hiding details of the internal data structures
26 * Copyright (c) 2016, 2017, Intel Corporation.
28 * Author: Ned Bass <bass6@llnl.gov>
38 #include <sys/xattr.h>
39 #include <sys/param.h>
41 #include <libcfs/util/list.h>
42 #include <lustre/lustreapi.h>
43 #include "lustreapi_internal.h"
46 * Layout component, which contains all attributes of a plain
49 struct llapi_layout_comp {
51 uint64_t llc_stripe_size;
52 uint64_t llc_stripe_count;
53 uint64_t llc_stripe_offset;
54 /* Add 1 so user always gets back a null terminated string. */
55 char llc_pool_name[LOV_MAXPOOLNAME + 1];
56 /** Number of objects in llc_objects array if was initialized. */
57 uint32_t llc_objects_count;
58 struct lov_user_ost_data_v1 *llc_objects;
59 /* fields used only for composite layouts */
60 struct lu_extent llc_extent; /* [start, end) of component */
61 uint32_t llc_id; /* unique ID of component */
62 uint32_t llc_flags; /* LCME_FL_* flags */
63 uint64_t llc_timestamp; /* snapshot timestamp */
64 struct list_head llc_list; /* linked to the llapi_layout
69 * An Opaque data type abstracting the layout of a Lustre file.
72 uint32_t llot_magic; /* LLAPI_LAYOUT_MAGIC */
75 bool llot_is_composite;
76 uint16_t llot_mirror_count;
77 /* Cursor pointing to one of the components in llot_comp_list */
78 struct llapi_layout_comp *llot_cur_comp;
79 struct list_head llot_comp_list;
83 * Compute the number of elements in the lmm_objects array of \a lum
84 * with size \a lum_size.
86 * \param[in] lum the struct lov_user_md to check
87 * \param[in] lum_size the number of bytes in \a lum
89 * \retval number of elements in array lum->lmm_objects
91 static int llapi_layout_objects_in_lum(struct lov_user_md *lum, size_t lum_size)
96 if (lum_size < lov_user_md_size(0, LOV_MAGIC_V1))
99 if (lum->lmm_magic == __swab32(LOV_MAGIC_V1) ||
100 lum->lmm_magic == __swab32(LOV_MAGIC_V3))
101 magic = __swab32(lum->lmm_magic);
103 magic = lum->lmm_magic;
105 base_size = lov_user_md_size(0, magic);
107 if (lum_size <= base_size)
110 return (lum_size - base_size) / sizeof(lum->lmm_objects[0]);
114 * Byte-swap the fields of struct lov_user_md.
116 * XXX Rather than duplicating swabbing code here, we should eventually
117 * refactor the needed functions in lustre/ptlrpc/pack_generic.c
118 * into a library that can be shared between kernel and user code.
121 llapi_layout_swab_lov_user_md(struct lov_user_md *lum, int lum_size)
123 int i, j, ent_count, obj_count;
124 struct lov_comp_md_v1 *comp_v1 = NULL;
125 struct lov_comp_md_entry_v1 *ent;
126 struct lov_user_ost_data *lod;
128 if (lum->lmm_magic != __swab32(LOV_MAGIC_V1) &&
129 lum->lmm_magic != __swab32(LOV_MAGIC_V3) &&
130 lum->lmm_magic != __swab32(LOV_MAGIC_COMP_V1))
133 if (lum->lmm_magic == __swab32(LOV_MAGIC_COMP_V1))
134 comp_v1 = (struct lov_comp_md_v1 *)lum;
136 if (comp_v1 != NULL) {
137 __swab32s(&comp_v1->lcm_magic);
138 __swab32s(&comp_v1->lcm_size);
139 __swab32s(&comp_v1->lcm_layout_gen);
140 __swab16s(&comp_v1->lcm_flags);
141 __swab16s(&comp_v1->lcm_entry_count);
142 ent_count = comp_v1->lcm_entry_count;
147 for (i = 0; i < ent_count; i++) {
148 if (comp_v1 != NULL) {
149 ent = &comp_v1->lcm_entries[i];
150 __swab32s(&ent->lcme_id);
151 __swab32s(&ent->lcme_flags);
152 __swab64s(&ent->lcme_timestamp);
153 __swab64s(&ent->lcme_extent.e_start);
154 __swab64s(&ent->lcme_extent.e_end);
155 __swab32s(&ent->lcme_offset);
156 __swab32s(&ent->lcme_size);
158 lum = (struct lov_user_md *)((char *)comp_v1 +
160 lum_size = ent->lcme_size;
162 obj_count = llapi_layout_objects_in_lum(lum, lum_size);
164 __swab32s(&lum->lmm_magic);
165 __swab32s(&lum->lmm_pattern);
166 __swab32s(&lum->lmm_stripe_size);
167 __swab16s(&lum->lmm_stripe_count);
168 __swab16s(&lum->lmm_stripe_offset);
170 if (lum->lmm_magic != LOV_MAGIC_V1) {
171 struct lov_user_md_v3 *v3;
172 v3 = (struct lov_user_md_v3 *)lum;
173 lod = v3->lmm_objects;
175 lod = lum->lmm_objects;
178 for (j = 0; j < obj_count; j++)
179 __swab32s(&lod[j].l_ost_idx);
184 * (Re-)allocate llc_objects[] to \a num_stripes stripes.
186 * Copy over existing llc_objects[], if any, to the new llc_objects[].
188 * \param[in] layout existing layout to be modified
189 * \param[in] num_stripes number of stripes in new layout
191 * \retval 0 if the objects are re-allocated successfully
192 * \retval -1 on error with errno set
194 static int __llapi_comp_objects_realloc(struct llapi_layout_comp *comp,
195 unsigned int new_stripes)
197 struct lov_user_ost_data_v1 *new_objects;
200 if (new_stripes > LOV_MAX_STRIPE_COUNT) {
205 if (new_stripes == comp->llc_objects_count)
208 if (new_stripes != 0 && new_stripes <= comp->llc_objects_count)
211 new_objects = realloc(comp->llc_objects,
212 sizeof(*new_objects) * new_stripes);
213 if (new_objects == NULL && new_stripes != 0) {
218 for (i = comp->llc_objects_count; i < new_stripes; i++)
219 new_objects[i].l_ost_idx = LLAPI_LAYOUT_IDX_MAX;
221 comp->llc_objects = new_objects;
222 comp->llc_objects_count = new_stripes;
228 * Allocate storage for a llapi_layout_comp with \a num_stripes stripes.
230 * \param[in] num_stripes number of stripes in new layout
232 * \retval valid pointer if allocation succeeds
233 * \retval NULL if allocation fails
235 static struct llapi_layout_comp *__llapi_comp_alloc(unsigned int num_stripes)
237 struct llapi_layout_comp *comp;
239 if (num_stripes > LOV_MAX_STRIPE_COUNT) {
244 comp = calloc(1, sizeof(*comp));
250 comp->llc_objects = NULL;
251 comp->llc_objects_count = 0;
253 if (__llapi_comp_objects_realloc(comp, num_stripes) < 0) {
259 comp->llc_pattern = LLAPI_LAYOUT_DEFAULT;
260 comp->llc_stripe_size = LLAPI_LAYOUT_DEFAULT;
261 comp->llc_stripe_count = LLAPI_LAYOUT_DEFAULT;
262 comp->llc_stripe_offset = LLAPI_LAYOUT_DEFAULT;
263 comp->llc_pool_name[0] = '\0';
264 comp->llc_extent.e_start = 0;
265 comp->llc_extent.e_end = LUSTRE_EOF;
268 INIT_LIST_HEAD(&comp->llc_list);
274 * Free memory allocated for \a comp
276 * \param[in] comp previously allocated by __llapi_comp_alloc()
278 static void __llapi_comp_free(struct llapi_layout_comp *comp)
280 if (comp->llc_objects != NULL)
281 free(comp->llc_objects);
286 * Free memory allocated for \a layout.
288 * \param[in] layout previously allocated by llapi_layout_alloc()
290 void llapi_layout_free(struct llapi_layout *layout)
292 struct llapi_layout_comp *comp, *n;
297 list_for_each_entry_safe(comp, n, &layout->llot_comp_list, llc_list) {
298 list_del_init(&comp->llc_list);
299 __llapi_comp_free(comp);
305 * Allocate and initialize a llapi_layout structure.
307 * \retval valid llapi_layout pointer on success
308 * \retval NULL if memory allocation fails
310 static struct llapi_layout *__llapi_layout_alloc(void)
312 struct llapi_layout *layout;
314 layout = calloc(1, sizeof(*layout));
315 if (layout == NULL) {
321 layout->llot_magic = LLAPI_LAYOUT_MAGIC;
322 layout->llot_gen = 0;
323 layout->llot_flags = 0;
324 layout->llot_is_composite = false;
325 layout->llot_mirror_count = 1;
326 layout->llot_cur_comp = NULL;
327 INIT_LIST_HEAD(&layout->llot_comp_list);
333 * Allocate and initialize a new plain layout.
335 * \retval valid llapi_layout pointer on success
336 * \retval NULL if memory allocation fails
338 struct llapi_layout *llapi_layout_alloc(void)
340 struct llapi_layout_comp *comp;
341 struct llapi_layout *layout;
343 layout = __llapi_layout_alloc();
347 comp = __llapi_comp_alloc(0);
353 list_add_tail(&comp->llc_list, &layout->llot_comp_list);
354 layout->llot_cur_comp = comp;
360 * Check if the given \a lum_size is large enough to hold the required
363 * \param[in] lum the struct lov_user_md to check
364 * \param[in] lum_size the number of bytes in \a lum
366 * \retval true the \a lum_size is too small
367 * \retval false the \a lum_size is large enough
369 static bool llapi_layout_lum_truncated(struct lov_user_md *lum, size_t lum_size)
373 if (lum_size < sizeof(lum->lmm_magic))
376 if (lum->lmm_magic == LOV_MAGIC_V1 ||
377 lum->lmm_magic == __swab32(LOV_MAGIC_V1))
378 magic = LOV_MAGIC_V1;
379 else if (lum->lmm_magic == LOV_MAGIC_V3 ||
380 lum->lmm_magic == __swab32(LOV_MAGIC_V3))
381 magic = LOV_MAGIC_V3;
382 else if (lum->lmm_magic == LOV_MAGIC_COMP_V1 ||
383 lum->lmm_magic == __swab32(LOV_MAGIC_COMP_V1))
384 magic = LOV_MAGIC_COMP_V1;
388 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3)
389 return lum_size < lov_user_md_size(0, magic);
391 return lum_size < sizeof(struct lov_comp_md_v1);
394 /* Verify if the objects count in lum is consistent with the
395 * stripe count in lum. It applies to regular file only. */
396 static bool llapi_layout_lum_valid(struct lov_user_md *lum, int lum_size)
398 struct lov_comp_md_v1 *comp_v1 = NULL;
399 int i, ent_count, obj_count;
401 if (lum->lmm_magic == LOV_MAGIC_COMP_V1) {
402 comp_v1 = (struct lov_comp_md_v1 *)lum;
403 ent_count = comp_v1->lcm_entry_count;
404 } else if (lum->lmm_magic == LOV_MAGIC_V1 ||
405 lum->lmm_magic == LOV_MAGIC_V3) {
411 for (i = 0; i < ent_count; i++) {
413 lum = (struct lov_user_md *)((char *)comp_v1 +
414 comp_v1->lcm_entries[i].lcme_offset);
415 lum_size = comp_v1->lcm_entries[i].lcme_size;
417 obj_count = llapi_layout_objects_in_lum(lum, lum_size);
420 if (!(comp_v1->lcm_entries[i].lcme_flags &
421 LCME_FL_INIT) && obj_count != 0)
423 } else if (obj_count != lum->lmm_stripe_count) {
431 * Convert the data from a lov_user_md to a newly allocated llapi_layout.
432 * The caller is responsible for freeing the returned pointer.
434 * \param[in] lov_xattr LOV user metadata xattr to copy data from
435 * \param[in] lov_xattr_size size the lov_xattr_size passed in
436 * \param[in] flags bitwise-or'd flags to control the behavior
438 * \retval valid llapi_layout pointer on success
439 * \retval NULL if memory allocation fails
441 struct llapi_layout *llapi_layout_get_by_xattr(void *lov_xattr,
442 ssize_t lov_xattr_size,
445 struct lov_user_md *lum = lov_xattr;
446 struct lov_comp_md_v1 *comp_v1 = NULL;
447 struct lov_comp_md_entry_v1 *ent;
448 struct lov_user_md *v1;
449 struct llapi_layout *layout = NULL;
450 struct llapi_layout_comp *comp;
451 int i, ent_count = 0, obj_count;
453 if (lov_xattr == NULL || lov_xattr_size <= 0) {
458 /* Return an error if we got back a partial layout. */
459 if (llapi_layout_lum_truncated(lov_xattr, lov_xattr_size)) {
464 #if __BYTE_ORDER == __BIG_ENDIAN
465 if (flags & LLAPI_LXF_COPY) {
466 lum = malloc(lov_xattr_size);
471 memcpy(lum, lov_xattr, lov_xattr_size);
475 llapi_layout_swab_lov_user_md(lum, lov_xattr_size);
477 if ((flags & LLAPI_LXF_CHECK) &&
478 !llapi_layout_lum_valid(lum, lov_xattr_size)) {
483 layout = __llapi_layout_alloc();
484 if (layout == NULL) {
489 if (lum->lmm_magic == LOV_MAGIC_COMP_V1) {
490 comp_v1 = (struct lov_comp_md_v1 *)lum;
491 ent_count = comp_v1->lcm_entry_count;
492 layout->llot_gen = comp_v1->lcm_layout_gen;
493 layout->llot_is_composite = true;
494 layout->llot_mirror_count = comp_v1->lcm_mirror_count + 1;
495 layout->llot_gen = comp_v1->lcm_layout_gen;
496 layout->llot_flags = comp_v1->lcm_flags;
497 } else if (lum->lmm_magic == LOV_MAGIC_V1 ||
498 lum->lmm_magic == LOV_MAGIC_V3) {
500 layout->llot_is_composite = false;
502 if (lov_xattr_size <= 0) {
511 if (ent_count == 0) {
516 v1 = (struct lov_user_md *)lum;
517 for (i = 0; i < ent_count; i++) {
518 if (comp_v1 != NULL) {
519 ent = &comp_v1->lcm_entries[i];
520 v1 = (struct lov_user_md *)((char *)comp_v1 +
522 lov_xattr_size = ent->lcme_size;
527 obj_count = llapi_layout_objects_in_lum(v1, lov_xattr_size);
528 comp = __llapi_comp_alloc(obj_count);
533 comp->llc_extent.e_start = ent->lcme_extent.e_start;
534 comp->llc_extent.e_end = ent->lcme_extent.e_end;
535 comp->llc_id = ent->lcme_id;
536 comp->llc_flags = ent->lcme_flags;
537 if (comp->llc_flags & LCME_FL_NOSYNC)
538 comp->llc_timestamp = ent->lcme_timestamp;
540 comp->llc_extent.e_start = 0;
541 comp->llc_extent.e_end = LUSTRE_EOF;
546 if (v1->lmm_pattern == LOV_PATTERN_RAID0)
547 comp->llc_pattern = LLAPI_LAYOUT_RAID0;
549 /* Lustre only supports RAID0 for now. */
550 comp->llc_pattern = v1->lmm_pattern;
552 if (v1->lmm_stripe_size == 0)
553 comp->llc_stripe_size = LLAPI_LAYOUT_DEFAULT;
555 comp->llc_stripe_size = v1->lmm_stripe_size;
557 if (v1->lmm_stripe_count == (typeof(v1->lmm_stripe_count))-1)
558 comp->llc_stripe_count = LLAPI_LAYOUT_WIDE;
559 else if (v1->lmm_stripe_count == 0)
560 comp->llc_stripe_count = LLAPI_LAYOUT_DEFAULT;
562 comp->llc_stripe_count = v1->lmm_stripe_count;
564 if (v1->lmm_stripe_offset ==
565 (typeof(v1->lmm_stripe_offset))-1)
566 comp->llc_stripe_offset = LLAPI_LAYOUT_DEFAULT;
568 comp->llc_stripe_offset = v1->lmm_stripe_offset;
570 if (v1->lmm_magic != LOV_USER_MAGIC_V1) {
571 const struct lov_user_md_v3 *lumv3;
572 lumv3 = (struct lov_user_md_v3 *)v1;
573 snprintf(comp->llc_pool_name,
574 sizeof(comp->llc_pool_name),
575 "%s", lumv3->lmm_pool_name);
576 memcpy(comp->llc_objects, lumv3->lmm_objects,
577 obj_count * sizeof(lumv3->lmm_objects[0]));
579 const struct lov_user_md_v1 *lumv1;
580 lumv1 = (struct lov_user_md_v1 *)v1;
581 memcpy(comp->llc_objects, lumv1->lmm_objects,
582 obj_count * sizeof(lumv1->lmm_objects[0]));
586 comp->llc_stripe_offset =
587 comp->llc_objects[0].l_ost_idx;
589 list_add_tail(&comp->llc_list, &layout->llot_comp_list);
590 layout->llot_cur_comp = comp;
594 if (lum != lov_xattr)
598 llapi_layout_free(layout);
604 * Convert the data from a llapi_layout to a newly allocated lov_user_md.
605 * The caller is responsible for freeing the returned pointer.
607 * \param[in] layout the layout to copy from
609 * \retval valid lov_user_md pointer on success
610 * \retval NULL if memory allocation fails or the layout is invalid
612 static struct lov_user_md *
613 llapi_layout_to_lum(const struct llapi_layout *layout)
615 struct llapi_layout_comp *comp;
616 struct lov_comp_md_v1 *comp_v1 = NULL;
617 struct lov_comp_md_entry_v1 *ent;
618 struct lov_user_md *lum = NULL;
623 if (layout == NULL ||
624 list_empty((struct list_head *)&layout->llot_comp_list)) {
629 /* Allocate header of lov_comp_md_v1 if necessary */
630 if (layout->llot_is_composite) {
633 list_for_each_entry(comp, &layout->llot_comp_list, llc_list)
636 lum_size = sizeof(*comp_v1) + comp_cnt * sizeof(*ent);
637 lum = calloc(lum_size, 1);
642 comp_v1 = (struct lov_comp_md_v1 *)lum;
643 comp_v1->lcm_magic = LOV_USER_MAGIC_COMP_V1;
644 comp_v1->lcm_size = lum_size;
645 comp_v1->lcm_layout_gen = 0;
646 comp_v1->lcm_flags = layout->llot_flags;
647 comp_v1->lcm_entry_count = comp_cnt;
648 comp_v1->lcm_mirror_count = layout->llot_mirror_count - 1;
652 list_for_each_entry(comp, &layout->llot_comp_list, llc_list) {
653 struct lov_user_md *blob;
656 int i, obj_count = 0;
657 struct lov_user_ost_data *lmm_objects;
658 uint64_t pattern = comp->llc_pattern;
660 if ((pattern & LLAPI_LAYOUT_SPECIFIC) != 0) {
661 if (comp->llc_objects_count <
662 comp->llc_stripe_count) {
666 magic = LOV_USER_MAGIC_SPECIFIC;
667 obj_count = comp->llc_stripe_count;
668 pattern &= ~LLAPI_LAYOUT_SPECIFIC;
669 } else if (strlen(comp->llc_pool_name) != 0) {
670 magic = LOV_USER_MAGIC_V3;
672 magic = LOV_USER_MAGIC_V1;
674 /* All stripes must be specified when the pattern contains
675 * LLAPI_LAYOUT_SPECIFIC */
676 for (i = 0; i < obj_count; i++) {
677 if (comp->llc_objects[i].l_ost_idx ==
678 LLAPI_LAYOUT_IDX_MAX) {
684 blob_size = lov_user_md_size(obj_count, magic);
685 blob = realloc(lum, lum_size + blob_size);
691 comp_v1 = (struct lov_comp_md_v1 *)lum;
692 blob = (struct lov_user_md *)((char *)lum + lum_size);
693 lum_size += blob_size;
696 blob->lmm_magic = magic;
697 if (pattern == LLAPI_LAYOUT_DEFAULT)
698 blob->lmm_pattern = LOV_PATTERN_RAID0;
699 else if (pattern == LLAPI_LAYOUT_MDT)
700 blob->lmm_pattern = LOV_PATTERN_MDT;
702 blob->lmm_pattern = pattern;
704 if (comp->llc_stripe_size == LLAPI_LAYOUT_DEFAULT)
705 blob->lmm_stripe_size = 0;
707 blob->lmm_stripe_size = comp->llc_stripe_size;
709 if (comp->llc_stripe_count == LLAPI_LAYOUT_DEFAULT)
710 blob->lmm_stripe_count = 0;
711 else if (comp->llc_stripe_count == LLAPI_LAYOUT_WIDE)
712 blob->lmm_stripe_count = LOV_ALL_STRIPES;
714 blob->lmm_stripe_count = comp->llc_stripe_count;
716 if (comp->llc_stripe_offset == LLAPI_LAYOUT_DEFAULT)
717 blob->lmm_stripe_offset = -1;
719 blob->lmm_stripe_offset = comp->llc_stripe_offset;
721 if (magic == LOV_USER_MAGIC_V3 ||
722 magic == LOV_USER_MAGIC_SPECIFIC) {
723 struct lov_user_md_v3 *lumv3 =
724 (struct lov_user_md_v3 *)blob;
726 if (comp->llc_pool_name[0] != '\0') {
727 strncpy(lumv3->lmm_pool_name,
729 sizeof(lumv3->lmm_pool_name));
731 memset(lumv3->lmm_pool_name, 0,
732 sizeof(lumv3->lmm_pool_name));
734 lmm_objects = lumv3->lmm_objects;
736 lmm_objects = blob->lmm_objects;
739 for (i = 0; i < obj_count; i++)
740 lmm_objects[i].l_ost_idx =
741 comp->llc_objects[i].l_ost_idx;
743 if (layout->llot_is_composite) {
744 ent = &comp_v1->lcm_entries[ent_idx];
745 ent->lcme_id = comp->llc_id;
746 ent->lcme_flags = comp->llc_flags;
747 if (ent->lcme_flags & LCME_FL_NOSYNC)
748 ent->lcme_timestamp = comp->llc_timestamp;
749 ent->lcme_extent.e_start = comp->llc_extent.e_start;
750 ent->lcme_extent.e_end = comp->llc_extent.e_end;
751 ent->lcme_size = blob_size;
752 ent->lcme_offset = offset;
754 comp_v1->lcm_size += blob_size;
768 * Get the parent directory of a path.
770 * \param[in] path path to get parent of
771 * \param[out] buf buffer in which to store parent path
772 * \param[in] size size in bytes of buffer \a buf
774 static void get_parent_dir(const char *path, char *buf, size_t size)
778 strncpy(buf, path, size);
779 p = strrchr(buf, '/');
783 } else if (size >= 2) {
784 strncpy(buf, ".", 2);
785 buf[size - 1] = '\0';
790 * Substitute unspecified attribute values in \a layout with values
791 * from fs global settings. (lov.stripesize, lov.stripecount,
794 * \param[in] layout layout to inherit values from
795 * \param[in] path file path of the filesystem
797 static void inherit_sys_attributes(struct llapi_layout *layout,
800 struct llapi_layout_comp *comp;
801 unsigned int ssize, scount, soffset;
804 rc = sattr_cache_get_defaults(NULL, path, &scount, &ssize, &soffset);
808 list_for_each_entry(comp, &layout->llot_comp_list, llc_list) {
809 if (comp->llc_pattern == LLAPI_LAYOUT_DEFAULT)
810 comp->llc_pattern = LLAPI_LAYOUT_RAID0;
811 if (comp->llc_stripe_size == LLAPI_LAYOUT_DEFAULT)
812 comp->llc_stripe_size = ssize;
813 if (comp->llc_stripe_count == LLAPI_LAYOUT_DEFAULT)
814 comp->llc_stripe_count = scount;
815 if (comp->llc_stripe_offset == LLAPI_LAYOUT_DEFAULT)
816 comp->llc_stripe_offset = soffset;
821 * Get the current component of \a layout.
823 * \param[in] layout layout to get current component
825 * \retval valid llapi_layout_comp pointer on success
826 * \retval NULL on error
828 static struct llapi_layout_comp *
829 __llapi_layout_cur_comp(const struct llapi_layout *layout)
831 struct llapi_layout_comp *comp;
833 if (layout == NULL || layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
837 if (layout->llot_cur_comp == NULL) {
841 /* Verify data consistency */
842 list_for_each_entry(comp, &layout->llot_comp_list, llc_list)
843 if (comp == layout->llot_cur_comp)
850 * Test if any attributes of \a layout are specified.
852 * \param[in] layout the layout to check
854 * \retval true any attributes are specified
855 * \retval false all attributes are unspecified
857 static bool is_any_specified(const struct llapi_layout *layout)
859 struct llapi_layout_comp *comp;
861 comp = __llapi_layout_cur_comp(layout);
865 if (layout->llot_is_composite || layout->llot_mirror_count != 1)
868 return comp->llc_pattern != LLAPI_LAYOUT_DEFAULT ||
869 comp->llc_stripe_size != LLAPI_LAYOUT_DEFAULT ||
870 comp->llc_stripe_count != LLAPI_LAYOUT_DEFAULT ||
871 comp->llc_stripe_offset != LLAPI_LAYOUT_DEFAULT ||
872 strlen(comp->llc_pool_name);
876 * Get the striping layout for the file referenced by file descriptor \a fd.
878 * If the filesystem does not support the "lustre." xattr namespace, the
879 * file must be on a non-Lustre filesystem, so set errno to ENOTTY per
880 * convention. If the file has no "lustre.lov" data, the file will
881 * inherit default values, so return a default layout.
883 * If the kernel gives us back less than the expected amount of data,
884 * we fail with errno set to EINTR.
886 * \param[in] fd open file descriptor
887 * \param[in] flags open file descriptor
889 * \retval valid llapi_layout pointer on success
890 * \retval NULL if an error occurs
892 struct llapi_layout *llapi_layout_get_by_fd(int fd, uint32_t flags)
895 struct lov_user_md *lum;
896 struct llapi_layout *layout = NULL;
900 lum_len = XATTR_SIZE_MAX;
901 lum = malloc(lum_len);
905 bytes_read = fgetxattr(fd, XATTR_LUSTRE_LOV, lum, lum_len);
906 if (bytes_read < 0) {
907 if (errno == EOPNOTSUPP)
909 else if (errno == ENODATA)
910 layout = llapi_layout_alloc();
914 /* Directories may have a positive non-zero lum->lmm_stripe_count
915 * yet have an empty lum->lmm_objects array. For non-directories the
916 * amount of data returned from the kernel must be consistent
917 * with the stripe count. */
918 if (fstat(fd, &st) < 0)
921 layout = llapi_layout_get_by_xattr(lum, bytes_read,
922 S_ISDIR(st.st_mode) ? 0 : LLAPI_LXF_CHECK);
929 * Get the expected striping layout for a file at \a path.
931 * Substitute expected inherited attribute values for unspecified
932 * attributes. Unspecified attributes may belong to directories and
933 * never-written-to files, and indicate that default values will be
934 * assigned when files are created or first written to. A default value
935 * is inherited from the parent directory if the attribute is specified
936 * there, otherwise it is inherited from the filesystem root.
937 * Unspecified attributes normally have the value LLAPI_LAYOUT_DEFAULT.
939 * The complete \a path need not refer to an existing file or directory,
940 * but some leading portion of it must reside within a lustre filesystem.
941 * A use case for this interface would be to obtain the literal striping
942 * values that would be assigned to a new file in a given directory.
944 * \param[in] path path for which to get the expected layout
946 * \retval valid llapi_layout pointer on success
947 * \retval NULL if an error occurs
949 static struct llapi_layout *llapi_layout_expected(const char *path)
951 struct llapi_layout *path_layout = NULL;
952 char donor_path[PATH_MAX];
957 fd = open(path, O_RDONLY);
958 if (fd < 0 && errno != ENOENT)
964 path_layout = llapi_layout_get_by_fd(fd, 0);
970 if (path_layout == NULL) {
971 if (errno != ENODATA && errno != ENOENT)
974 path_layout = llapi_layout_alloc();
975 if (path_layout == NULL)
979 if (is_any_specified(path_layout)) {
980 inherit_sys_attributes(path_layout, path);
984 llapi_layout_free(path_layout);
986 rc = stat(path, &st);
987 if (rc < 0 && errno != ENOENT)
990 /* If path is a not a directory or doesn't exist, inherit layout
991 * from parent directory. */
992 if ((rc == 0 && !S_ISDIR(st.st_mode)) ||
993 (rc < 0 && errno == ENOENT)) {
994 get_parent_dir(path, donor_path, sizeof(donor_path));
995 path_layout = llapi_layout_get_by_path(donor_path, 0);
996 if (path_layout != NULL) {
997 if (is_any_specified(path_layout)) {
998 inherit_sys_attributes(path_layout, donor_path);
1001 llapi_layout_free(path_layout);
1005 /* Inherit layout from the filesystem root. */
1006 rc = llapi_search_mounts(path, 0, donor_path, NULL);
1009 path_layout = llapi_layout_get_by_path(donor_path, 0);
1010 if (path_layout == NULL)
1013 inherit_sys_attributes(path_layout, donor_path);
1018 * Get the striping layout for the file at \a path.
1020 * If \a flags contains LAYOUT_GET_EXPECTED, substitute
1021 * expected inherited attribute values for unspecified attributes. See
1022 * llapi_layout_expected().
1024 * \param[in] path path for which to get the layout
1025 * \param[in] flags flags to control how layout is retrieved
1027 * \retval valid llapi_layout pointer on success
1028 * \retval NULL if an error occurs
1030 struct llapi_layout *llapi_layout_get_by_path(const char *path, uint32_t flags)
1032 struct llapi_layout *layout = NULL;
1036 if (flags & LAYOUT_GET_EXPECTED)
1037 return llapi_layout_expected(path);
1039 fd = open(path, O_RDONLY);
1043 layout = llapi_layout_get_by_fd(fd, flags);
1052 * Get the layout for the file with FID \a fidstr in filesystem \a lustre_dir.
1054 * \param[in] lustre_dir path within Lustre filesystem containing \a fid
1055 * \param[in] fid Lustre identifier of file to get layout for
1057 * \retval valid llapi_layout pointer on success
1058 * \retval NULL if an error occurs
1060 struct llapi_layout *llapi_layout_get_by_fid(const char *lustre_dir,
1061 const struct lu_fid *fid,
1066 int saved_msg_level = llapi_msg_get_level();
1067 struct llapi_layout *layout = NULL;
1069 /* Prevent llapi internal routines from writing to console
1070 * while executing this function, then restore previous message
1072 llapi_msg_set_level(LLAPI_MSG_OFF);
1073 fd = llapi_open_by_fid(lustre_dir, fid, O_RDONLY);
1074 llapi_msg_set_level(saved_msg_level);
1079 layout = llapi_layout_get_by_fd(fd, flags);
1088 * Get the stripe count of \a layout.
1090 * \param[in] layout layout to get stripe count from
1091 * \param[out] count integer to store stripe count in
1093 * \retval 0 on success
1094 * \retval -1 if arguments are invalid
1096 int llapi_layout_stripe_count_get(const struct llapi_layout *layout,
1099 struct llapi_layout_comp *comp;
1101 comp = __llapi_layout_cur_comp(layout);
1105 if (count == NULL) {
1110 *count = comp->llc_stripe_count;
1116 * The llapi_layout API functions have these extra validity checks since
1117 * they use intuitively named macros to denote special behavior, whereas
1118 * the old API uses 0 and -1.
1121 static bool llapi_layout_stripe_count_is_valid(int64_t stripe_count)
1123 return stripe_count == LLAPI_LAYOUT_DEFAULT ||
1124 stripe_count == LLAPI_LAYOUT_WIDE ||
1125 (stripe_count != 0 && stripe_count != -1 &&
1126 llapi_stripe_count_is_valid(stripe_count));
1129 static bool llapi_layout_stripe_size_is_valid(uint64_t stripe_size)
1131 return stripe_size == LLAPI_LAYOUT_DEFAULT ||
1132 (stripe_size != 0 &&
1133 llapi_stripe_size_is_aligned(stripe_size) &&
1134 !llapi_stripe_size_is_too_big(stripe_size));
1137 static bool llapi_layout_stripe_index_is_valid(int64_t stripe_index)
1139 return stripe_index == LLAPI_LAYOUT_DEFAULT ||
1140 (stripe_index >= 0 &&
1141 llapi_stripe_index_is_valid(stripe_index));
1145 * Set the stripe count of \a layout.
1147 * \param[in] layout layout to set stripe count in
1148 * \param[in] count value to be set
1150 * \retval 0 on success
1151 * \retval -1 if arguments are invalid
1153 int llapi_layout_stripe_count_set(struct llapi_layout *layout,
1156 struct llapi_layout_comp *comp;
1158 comp = __llapi_layout_cur_comp(layout);
1162 if (!llapi_layout_stripe_count_is_valid(count)) {
1167 comp->llc_stripe_count = count;
1173 * Get the stripe size of \a layout.
1175 * \param[in] layout layout to get stripe size from
1176 * \param[out] size integer to store stripe size in
1178 * \retval 0 on success
1179 * \retval -1 if arguments are invalid
1181 int llapi_layout_stripe_size_get(const struct llapi_layout *layout,
1184 struct llapi_layout_comp *comp;
1186 comp = __llapi_layout_cur_comp(layout);
1195 *size = comp->llc_stripe_size;
1201 * Set the stripe size of \a layout.
1203 * \param[in] layout layout to set stripe size in
1204 * \param[in] size value to be set
1206 * \retval 0 on success
1207 * \retval -1 if arguments are invalid
1209 int llapi_layout_stripe_size_set(struct llapi_layout *layout,
1212 struct llapi_layout_comp *comp;
1214 comp = __llapi_layout_cur_comp(layout);
1218 if (!llapi_layout_stripe_size_is_valid(size)) {
1223 comp->llc_stripe_size = size;
1229 * Get the RAID pattern of \a layout.
1231 * \param[in] layout layout to get pattern from
1232 * \param[out] pattern integer to store pattern in
1234 * \retval 0 on success
1235 * \retval -1 if arguments are invalid
1237 int llapi_layout_pattern_get(const struct llapi_layout *layout,
1240 struct llapi_layout_comp *comp;
1242 comp = __llapi_layout_cur_comp(layout);
1246 if (pattern == NULL) {
1251 *pattern = comp->llc_pattern;
1257 * Set the pattern of \a layout.
1259 * \param[in] layout layout to set pattern in
1260 * \param[in] pattern value to be set
1262 * \retval 0 on success
1263 * \retval -1 if arguments are invalid or RAID pattern
1266 int llapi_layout_pattern_set(struct llapi_layout *layout, uint64_t pattern)
1268 struct llapi_layout_comp *comp;
1270 comp = __llapi_layout_cur_comp(layout);
1274 if (pattern != LLAPI_LAYOUT_DEFAULT &&
1275 pattern != LLAPI_LAYOUT_RAID0 && pattern != LLAPI_LAYOUT_MDT) {
1280 comp->llc_pattern = pattern |
1281 (comp->llc_pattern & LLAPI_LAYOUT_SPECIFIC);
1286 static inline int stripe_number_roundup(int stripe_number)
1288 unsigned int round_up = (stripe_number + 8) & ~7;
1289 return round_up > LOV_MAX_STRIPE_COUNT ?
1290 LOV_MAX_STRIPE_COUNT : round_up;
1294 * Set the OST index of stripe number \a stripe_number to \a ost_index.
1296 * If only the starting stripe's OST index is specified, then this can use
1297 * the normal LOV_MAGIC_{V1,V3} layout type. If multiple OST indices are
1298 * given, then allocate an array to hold the list of indices and ensure that
1299 * the LOV_USER_MAGIC_SPECIFIC layout is used when creating the file.
1301 * \param[in] layout layout to set OST index in
1302 * \param[in] stripe_number stripe number to set index for
1303 * \param[in] ost_index the index to set
1305 * \retval 0 on success
1306 * \retval -1 if arguments are invalid or an unsupported stripe number
1307 * was specified, error returned in errno
1309 int llapi_layout_ost_index_set(struct llapi_layout *layout, int stripe_number,
1312 struct llapi_layout_comp *comp;
1314 comp = __llapi_layout_cur_comp(layout);
1318 if (!llapi_layout_stripe_index_is_valid(ost_index)) {
1323 if (stripe_number == 0 && ost_index == LLAPI_LAYOUT_DEFAULT) {
1324 comp->llc_stripe_offset = ost_index;
1325 comp->llc_pattern &= ~LLAPI_LAYOUT_SPECIFIC;
1326 __llapi_comp_objects_realloc(comp, 0);
1327 } else if (stripe_number >= 0 &&
1328 stripe_number < LOV_MAX_STRIPE_COUNT) {
1329 if (ost_index >= LLAPI_LAYOUT_IDX_MAX) {
1334 /* Preallocate a few more stripes to avoid realloc() overhead.*/
1335 if (__llapi_comp_objects_realloc(comp,
1336 stripe_number_roundup(stripe_number)) < 0)
1339 comp->llc_objects[stripe_number].l_ost_idx = ost_index;
1341 if (stripe_number == 0)
1342 comp->llc_stripe_offset = ost_index;
1344 comp->llc_pattern |= LLAPI_LAYOUT_SPECIFIC;
1346 if (comp->llc_stripe_count == LLAPI_LAYOUT_DEFAULT ||
1347 comp->llc_stripe_count <= stripe_number)
1348 comp->llc_stripe_count = stripe_number + 1;
1358 * Get the OST index associated with stripe \a stripe_number.
1360 * Stripes are indexed starting from zero.
1362 * \param[in] layout layout to get index from
1363 * \param[in] stripe_number stripe number to get index for
1364 * \param[out] index integer to store index in
1366 * \retval 0 on success
1367 * \retval -1 if arguments are invalid
1369 int llapi_layout_ost_index_get(const struct llapi_layout *layout,
1370 uint64_t stripe_number, uint64_t *index)
1372 struct llapi_layout_comp *comp;
1374 comp = __llapi_layout_cur_comp(layout);
1378 if (index == NULL) {
1383 if (stripe_number >= comp->llc_stripe_count ||
1384 stripe_number >= comp->llc_objects_count) {
1389 if (comp->llc_stripe_offset == LLAPI_LAYOUT_DEFAULT)
1390 *index = LLAPI_LAYOUT_DEFAULT;
1392 *index = comp->llc_objects[stripe_number].l_ost_idx;
1399 * Get the pool name of layout \a layout.
1401 * \param[in] layout layout to get pool name from
1402 * \param[out] dest buffer to store pool name in
1403 * \param[in] n size in bytes of buffer \a dest
1405 * \retval 0 on success
1406 * \retval -1 if arguments are invalid
1408 int llapi_layout_pool_name_get(const struct llapi_layout *layout, char *dest,
1411 struct llapi_layout_comp *comp;
1413 comp = __llapi_layout_cur_comp(layout);
1422 strncpy(dest, comp->llc_pool_name, n);
1428 * Set the name of the pool of layout \a layout.
1430 * \param[in] layout layout to set pool name in
1431 * \param[in] pool_name pool name to set
1433 * \retval 0 on success
1434 * \retval -1 if arguments are invalid or pool name is too long
1436 int llapi_layout_pool_name_set(struct llapi_layout *layout,
1437 const char *pool_name)
1439 struct llapi_layout_comp *comp;
1442 comp = __llapi_layout_cur_comp(layout);
1446 if (pool_name == NULL) {
1451 /* Strip off any 'fsname.' portion. */
1452 ptr = strchr(pool_name, '.');
1454 pool_name = ptr + 1;
1456 if (strlen(pool_name) > LOV_MAXPOOLNAME) {
1461 strncpy(comp->llc_pool_name, pool_name, sizeof(comp->llc_pool_name));
1467 * Open and possibly create a file with a given \a layout.
1469 * If \a layout is NULL this function acts as a simple wrapper for
1470 * open(). By convention, ENOTTY is returned in errno if \a path
1471 * refers to a non-Lustre file.
1473 * \param[in] path name of the file to open
1474 * \param[in] open_flags open() flags
1475 * \param[in] mode permissions to create file, filtered by umask
1476 * \param[in] layout layout to create new file with
1478 * \retval non-negative file descriptor on successful open
1479 * \retval -1 if an error occurred
1481 int llapi_layout_file_open(const char *path, int open_flags, mode_t mode,
1482 const struct llapi_layout *layout)
1487 struct lov_user_md *lum;
1491 (layout != NULL && layout->llot_magic != LLAPI_LAYOUT_MAGIC)) {
1496 /* Object creation must be postponed until after layout attributes
1497 * have been applied. */
1498 if (layout != NULL && (open_flags & O_CREAT))
1499 open_flags |= O_LOV_DELAY_CREATE;
1501 fd = open(path, open_flags, mode);
1503 if (layout == NULL || fd < 0)
1506 lum = llapi_layout_to_lum(layout);
1515 if (lum->lmm_magic == LOV_USER_MAGIC_COMP_V1)
1516 lum_size = ((struct lov_comp_md_v1 *)lum)->lcm_size;
1517 else if (lum->lmm_magic == LOV_USER_MAGIC_SPECIFIC)
1518 lum_size = lov_user_md_size(lum->lmm_stripe_count,
1521 lum_size = lov_user_md_size(0, lum->lmm_magic);
1523 rc = fsetxattr(fd, XATTR_LUSTRE_LOV, lum, lum_size, 0);
1532 errno = errno == EOPNOTSUPP ? ENOTTY : errno;
1538 * Create a file with a given \a layout.
1540 * Force O_CREAT and O_EXCL flags on so caller is assured that file was
1541 * created with the given \a layout on successful function return.
1543 * \param[in] path name of the file to open
1544 * \param[in] open_flags open() flags
1545 * \param[in] mode permissions to create new file with
1546 * \param[in] layout layout to create new file with
1548 * \retval non-negative file descriptor on successful open
1549 * \retval -1 if an error occurred
1551 int llapi_layout_file_create(const char *path, int open_flags, int mode,
1552 const struct llapi_layout *layout)
1554 return llapi_layout_file_open(path, open_flags|O_CREAT|O_EXCL, mode,
1558 int llapi_layout_flags_get(struct llapi_layout *layout, uint32_t *flags)
1560 if (layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
1565 *flags = layout->llot_flags;
1570 * Set flags to the header of a component layout.
1572 int llapi_layout_flags_set(struct llapi_layout *layout, uint32_t flags)
1574 if (layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
1579 layout->llot_flags = flags;
1583 const char *llapi_layout_flags_string(uint32_t flags)
1585 switch (flags & LCM_FL_FLR_MASK) {
1588 case LCM_FL_WRITE_PENDING:
1590 case LCM_FL_SYNC_PENDING:
1597 const __u16 llapi_layout_string_flags(char *string)
1599 if (strncmp(string, "ro", strlen(string)) == 0)
1600 return LCM_FL_RDONLY;
1601 if (strncmp(string, "wp", strlen(string)) == 0)
1602 return LCM_FL_WRITE_PENDING;
1603 if (strncmp(string, "sp", strlen(string)) == 0)
1604 return LCM_FL_SYNC_PENDING;
1610 * llapi_layout_mirror_count_is_valid() - Check the validity of mirror count.
1611 * @count: Mirror count value to be checked.
1613 * This function checks the validity of mirror count.
1615 * Return: true on success or false on failure.
1617 static bool llapi_layout_mirror_count_is_valid(uint16_t count)
1619 return count >= 0 && count <= LUSTRE_MIRROR_COUNT_MAX;
1623 * llapi_layout_mirror_count_get() - Get mirror count from the header of
1625 * @layout: Layout to get mirror count from.
1626 * @count: Returned mirror count value.
1628 * This function gets mirror count from the header of a layout.
1630 * Return: 0 on success or -1 on failure.
1632 int llapi_layout_mirror_count_get(struct llapi_layout *layout,
1635 if (layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
1640 *count = layout->llot_mirror_count;
1645 * llapi_layout_mirror_count_set() - Set mirror count to the header of a layout.
1646 * @layout: Layout to set mirror count in.
1647 * @count: Mirror count value to be set.
1649 * This function sets mirror count to the header of a layout.
1651 * Return: 0 on success or -1 on failure.
1653 int llapi_layout_mirror_count_set(struct llapi_layout *layout,
1656 if (layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
1661 if (!llapi_layout_mirror_count_is_valid(count)) {
1666 layout->llot_mirror_count = count;
1671 * Fetch the start and end offset of the current layout component.
1673 * \param[in] layout the layout component
1674 * \param[out] start extent start, inclusive
1675 * \param[out] end extent end, exclusive
1677 * \retval 0 on success
1678 * \retval <0 if error occurs
1680 int llapi_layout_comp_extent_get(const struct llapi_layout *layout,
1681 uint64_t *start, uint64_t *end)
1683 struct llapi_layout_comp *comp;
1685 comp = __llapi_layout_cur_comp(layout);
1689 if (start == NULL || end == NULL) {
1694 *start = comp->llc_extent.e_start;
1695 *end = comp->llc_extent.e_end;
1701 * Set the layout extent of a layout.
1703 * \param[in] layout the layout to be set
1704 * \param[in] start extent start, inclusive
1705 * \param[in] end extent end, exclusive
1707 * \retval 0 on success
1708 * \retval <0 if error occurs
1710 int llapi_layout_comp_extent_set(struct llapi_layout *layout,
1711 uint64_t start, uint64_t end)
1713 struct llapi_layout_comp *prev, *next, *comp;
1715 comp = __llapi_layout_cur_comp(layout);
1725 * We need to make sure the extent to be set is valid: the new
1726 * extent must be adjacent with the prev & next component.
1728 if (comp->llc_list.prev != &layout->llot_comp_list) {
1729 prev = list_entry(comp->llc_list.prev, typeof(*prev),
1731 if (start != 0 && start != prev->llc_extent.e_end) {
1737 if (comp->llc_list.next != &layout->llot_comp_list) {
1738 next = list_entry(comp->llc_list.next, typeof(*next),
1740 if (next->llc_extent.e_start != 0 &&
1741 end != next->llc_extent.e_start) {
1747 comp->llc_extent.e_start = start;
1748 comp->llc_extent.e_end = end;
1749 layout->llot_is_composite = true;
1755 * Gets the attribute flags of the current component.
1757 * \param[in] layout the layout component
1758 * \param[out] flags stored the returned component flags
1760 * \retval 0 on success
1761 * \retval <0 if error occurs
1763 int llapi_layout_comp_flags_get(const struct llapi_layout *layout,
1766 struct llapi_layout_comp *comp;
1768 comp = __llapi_layout_cur_comp(layout);
1772 if (flags == NULL) {
1777 *flags = comp->llc_flags;
1783 * Sets the specified flags of the current component leaving other flags as-is.
1785 * \param[in] layout the layout component
1786 * \param[in] flags component flags to be set
1788 * \retval 0 on success
1789 * \retval <0 if error occurs
1791 int llapi_layout_comp_flags_set(struct llapi_layout *layout, uint32_t flags)
1793 struct llapi_layout_comp *comp;
1795 comp = __llapi_layout_cur_comp(layout);
1799 comp->llc_flags |= flags;
1805 * Clears the flags specified in the flags leaving other flags as-is.
1807 * \param[in] layout the layout component
1808 * \param[in] flags component flags to be cleared
1810 * \retval 0 on success
1811 * \retval <0 if error occurs
1813 int llapi_layout_comp_flags_clear(struct llapi_layout *layout,
1816 struct llapi_layout_comp *comp;
1818 comp = __llapi_layout_cur_comp(layout);
1822 comp->llc_flags &= ~flags;
1828 * Fetches the file-unique component ID of the current layout component.
1830 * \param[in] layout the layout component
1831 * \param[out] id stored the returned component ID
1833 * \retval 0 on success
1834 * \retval <0 if error occurs
1836 int llapi_layout_comp_id_get(const struct llapi_layout *layout, uint32_t *id)
1838 struct llapi_layout_comp *comp;
1840 comp = __llapi_layout_cur_comp(layout);
1854 * Return the mirror id of the current layout component.
1856 * \param[in] layout the layout component
1857 * \param[out] id stored the returned mirror ID
1859 * \retval 0 on success
1860 * \retval <0 if error occurs
1862 int llapi_layout_mirror_id_get(const struct llapi_layout *layout, uint32_t *id)
1864 struct llapi_layout_comp *comp;
1866 comp = __llapi_layout_cur_comp(layout);
1875 *id = mirror_id_of(comp->llc_id);
1881 * Adds a component to \a layout, the new component will be added to
1882 * the tail of components list and it'll inherit attributes of existing
1883 * ones. The \a layout will change it's current component pointer to
1884 * the newly added component, and it'll be turned into a composite
1885 * layout if it was not before the adding.
1887 * \param[in] layout existing composite or plain layout
1889 * \retval 0 on success
1890 * \retval <0 if error occurs
1892 int llapi_layout_comp_add(struct llapi_layout *layout)
1894 struct llapi_layout_comp *last, *comp, *new;
1896 comp = __llapi_layout_cur_comp(layout);
1900 new = __llapi_comp_alloc(0);
1904 last = list_entry(layout->llot_comp_list.prev, typeof(*last),
1907 if (new->llc_extent.e_end <= last->llc_extent.e_end) {
1908 __llapi_comp_free(new);
1912 new->llc_extent.e_start = last->llc_extent.e_end;
1914 list_add_tail(&new->llc_list, &layout->llot_comp_list);
1915 layout->llot_cur_comp = new;
1916 layout->llot_is_composite = true;
1921 * Adds a first component of a mirror to \a layout.
1922 * The \a layout will change it's current component pointer to
1923 * the newly added component, and it'll be turned into a composite
1924 * layout if it was not before the adding.
1926 * \param[in] layout existing composite or plain layout
1928 * \retval 0 on success
1929 * \retval <0 if error occurs
1931 int llapi_layout_add_first_comp(struct llapi_layout *layout)
1933 struct llapi_layout_comp *comp, *new;
1935 comp = __llapi_layout_cur_comp(layout);
1939 new = __llapi_comp_alloc(0);
1943 new->llc_extent.e_start = 0;
1945 list_add_tail(&new->llc_list, &layout->llot_comp_list);
1946 layout->llot_cur_comp = new;
1947 layout->llot_is_composite = true;
1953 * Deletes current component from the composite layout. The component
1954 * to be deleted must be the tail of components list, and it can't be
1955 * the only component in the layout.
1957 * \param[in] layout composite layout
1959 * \retval 0 on success
1960 * \retval <0 if error occurs
1962 int llapi_layout_comp_del(struct llapi_layout *layout)
1964 struct llapi_layout_comp *comp;
1966 comp = __llapi_layout_cur_comp(layout);
1970 if (!layout->llot_is_composite) {
1975 /* It must be the tail of the list (for PFL, can be relaxed
1976 * once we get mirrored components) */
1977 if (comp->llc_list.next != &layout->llot_comp_list) {
1981 /* It can't be the only one on the list */
1982 if (comp->llc_list.prev == &layout->llot_comp_list) {
1987 layout->llot_cur_comp =
1988 list_entry(comp->llc_list.prev, typeof(*comp), llc_list);
1989 list_del_init(&comp->llc_list);
1990 __llapi_comp_free(comp);
1996 * Move the current component pointer to the component with
1997 * specified component ID.
1999 * \param[in] layout composite layout
2000 * \param[in] id component ID
2002 * \retval =0 : moved successfully
2003 * \retval <0 if error occurs
2005 int llapi_layout_comp_use_id(struct llapi_layout *layout, uint32_t comp_id)
2007 struct llapi_layout_comp *comp;
2009 comp = __llapi_layout_cur_comp(layout);
2011 return -1; /* use previously set errno */
2013 if (!layout->llot_is_composite) {
2018 if (comp_id == LCME_ID_INVAL) {
2023 list_for_each_entry(comp, &layout->llot_comp_list, llc_list) {
2024 if (comp->llc_id == comp_id) {
2025 layout->llot_cur_comp = comp;
2034 * Move the current component pointer to a specified position.
2036 * \param[in] layout composite layout
2037 * \param[in] pos the position to be moved, it can be:
2038 * LLAPI_LAYOUT_COMP_USE_FIRST: use first component
2039 * LLAPI_LAYOUT_COMP_USE_LAST: use last component
2040 * LLAPI_LAYOUT_COMP_USE_NEXT: use component after current
2041 * LLAPI_LAYOUT_COMP_USE_PREV: use component before current
2043 * \retval =0 : moved successfully
2044 * \retval =1 : at last component with NEXT, at first component with PREV
2045 * \retval <0 if error occurs
2047 int llapi_layout_comp_use(struct llapi_layout *layout,
2048 enum llapi_layout_comp_use pos)
2050 struct llapi_layout_comp *comp, *head, *tail;
2052 comp = __llapi_layout_cur_comp(layout);
2056 if (!layout->llot_is_composite) {
2057 if (pos == LLAPI_LAYOUT_COMP_USE_FIRST ||
2058 pos == LLAPI_LAYOUT_COMP_USE_LAST)
2064 head = list_entry(layout->llot_comp_list.next, typeof(*head), llc_list);
2065 tail = list_entry(layout->llot_comp_list.prev, typeof(*tail), llc_list);
2067 case LLAPI_LAYOUT_COMP_USE_FIRST:
2068 layout->llot_cur_comp = head;
2070 case LLAPI_LAYOUT_COMP_USE_NEXT:
2075 layout->llot_cur_comp = list_entry(comp->llc_list.next,
2076 typeof(*comp), llc_list);
2078 case LLAPI_LAYOUT_COMP_USE_LAST:
2079 layout->llot_cur_comp = tail;
2081 case LLAPI_LAYOUT_COMP_USE_PREV:
2086 layout->llot_cur_comp = list_entry(comp->llc_list.prev,
2087 typeof(*comp), llc_list);
2098 * Add layout component(s) to an existing file.
2100 * \param[in] path The path name of the file
2101 * \param[in] layout The layout component(s) to be added
2103 int llapi_layout_file_comp_add(const char *path,
2104 const struct llapi_layout *layout)
2106 int rc, fd, lum_size, tmp_errno = 0;
2107 struct lov_user_md *lum;
2109 if (path == NULL || layout == NULL ||
2110 layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
2115 lum = llapi_layout_to_lum(layout);
2119 if (lum->lmm_magic != LOV_USER_MAGIC_COMP_V1) {
2124 lum_size = ((struct lov_comp_md_v1 *)lum)->lcm_size;
2126 fd = open(path, O_RDWR);
2133 rc = fsetxattr(fd, XATTR_LUSTRE_LOV".add", lum, lum_size, 0);
2148 * Delete component(s) by the specified component id or component flags
2149 * from an existing file.
2151 * \param[in] path path name of the file
2152 * \param[in] id unique component ID
2153 * \param[in] flags flags: LCME_FL_* or;
2154 * negative flags: (LCME_FL_NEG|LCME_FL_*)
2156 int llapi_layout_file_comp_del(const char *path, uint32_t id, uint32_t flags)
2158 int rc, fd, lum_size;
2159 struct llapi_layout *layout;
2160 struct llapi_layout_comp *comp;
2161 struct lov_user_md *lum;
2163 if (path == NULL || id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2168 /* Can only specify ID or flags, not both. */
2169 if (id != 0 && flags != 0) {
2174 layout = llapi_layout_alloc();
2178 llapi_layout_comp_extent_set(layout, 0, LUSTRE_EOF);
2179 comp = __llapi_layout_cur_comp(layout);
2181 llapi_layout_free(layout);
2186 comp->llc_flags = flags;
2188 lum = llapi_layout_to_lum(layout);
2190 llapi_layout_free(layout);
2193 lum_size = ((struct lov_comp_md_v1 *)lum)->lcm_size;
2195 fd = open(path, O_RDWR);
2201 rc = fsetxattr(fd, XATTR_LUSTRE_LOV".del", lum, lum_size, 0);
2203 int tmp_errno = errno;
2212 llapi_layout_free(layout);
2217 * Change flags or other parameters of the component(s) by component ID of an
2218 * existing file. The component to be modified is specified by the
2219 * comp->lcme_id value, which must be an unique component ID. The new
2220 * attributes are passed in by @comp and @valid is used to specify which
2221 * attributes in the component are going to be changed.
2223 * \param[in] path path name of the file
2224 * \param[in] ids An array of component IDs
2225 * \param[in] flags flags: LCME_FL_* or;
2226 * negative flags: (LCME_FL_NEG|LCME_FL_*)
2227 * \param[in] count Number of elements in ids and flags array
2229 int llapi_layout_file_comp_set(const char *path, uint32_t *ids, uint32_t *flags,
2232 int rc = -1, fd = -1, i;
2234 struct llapi_layout *layout;
2235 struct llapi_layout_comp *comp;
2236 struct lov_user_md *lum = NULL;
2246 for (i = 0; i < count; i++) {
2247 if (!ids[i] || !flags[i]) {
2252 if (ids[i] > LCME_ID_MAX || (flags[i] & ~LCME_KNOWN_FLAGS)) {
2257 /* do not allow to set or clear INIT flag */
2258 if (flags[i] & LCME_FL_INIT) {
2264 layout = __llapi_layout_alloc();
2268 layout->llot_is_composite = true;
2269 for (i = 0; i < count; i++) {
2270 comp = __llapi_comp_alloc(0);
2274 comp->llc_id = ids[i];
2275 comp->llc_flags = flags[i];
2277 list_add_tail(&comp->llc_list, &layout->llot_comp_list);
2278 layout->llot_cur_comp = comp;
2281 lum = llapi_layout_to_lum(layout);
2285 lum_size = ((struct lov_comp_md_v1 *)lum)->lcm_size;
2287 fd = open(path, O_RDWR);
2291 /* flush cached pages from clients */
2292 rc = llapi_file_flush(fd);
2299 rc = fsetxattr(fd, XATTR_LUSTRE_LOV".set.flags", lum, lum_size, 0);
2307 int tmp_errno = errno;
2314 llapi_layout_free(layout);
2319 * Check if the file layout is composite.
2321 * \param[in] layout the file layout to check
2323 * \retval true composite
2324 * \retval false not composite
2326 bool llapi_layout_is_composite(struct llapi_layout *layout)
2328 return layout->llot_is_composite;
2332 * Iterate every components in the @layout and call callback function @cb.
2334 * \param[in] layout component layout list.
2335 * \param[in] cb callback for each component
2336 * \param[in] cbdata callback data
2338 * \retval < 0 error happens during the iteration
2339 * \retval LLAPI_LAYOUT_ITER_CONT finished the iteration w/o error
2340 * \retval LLAPI_LAYOUT_ITER_STOP got something, stop the iteration
2342 int llapi_layout_comp_iterate(struct llapi_layout *layout,
2343 llapi_layout_iter_cb cb, void *cbdata)
2347 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_FIRST);
2352 * make sure on success llapi_layout_comp_use() API returns 0 with
2358 rc = cb(layout, cbdata);
2359 if (rc != LLAPI_LAYOUT_ITER_CONT)
2362 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_NEXT);
2365 else if (rc == 1) /* reached the last comp */
2366 return LLAPI_LAYOUT_ITER_CONT;
2373 * llapi_layout_merge() - Merge a composite layout into another one.
2374 * @dst_layout: Destination composite layout.
2375 * @src_layout: Source composite layout.
2377 * This function copies all of the components from @src_layout and
2378 * appends them to @dst_layout.
2380 * Return: 0 on success or -1 on failure.
2382 int llapi_layout_merge(struct llapi_layout **dst_layout,
2383 const struct llapi_layout *src_layout)
2385 struct llapi_layout *new_layout = *dst_layout;
2386 struct llapi_layout_comp *new = NULL;
2387 struct llapi_layout_comp *comp = NULL;
2390 if (src_layout == NULL ||
2391 list_empty((struct list_head *)&src_layout->llot_comp_list))
2394 if (new_layout == NULL) {
2395 new_layout = __llapi_layout_alloc();
2396 if (new_layout == NULL) {
2402 list_for_each_entry(comp, &src_layout->llot_comp_list, llc_list) {
2403 new = __llapi_comp_alloc(0);
2409 new->llc_pattern = comp->llc_pattern;
2410 new->llc_stripe_size = comp->llc_stripe_size;
2411 new->llc_stripe_count = comp->llc_stripe_count;
2412 new->llc_stripe_offset = comp->llc_stripe_offset;
2414 if (comp->llc_pool_name[0] != '\0')
2415 strncpy(new->llc_pool_name, comp->llc_pool_name,
2416 sizeof(new->llc_pool_name));
2418 for (i = 0; i < comp->llc_objects_count; i++) {
2419 if (__llapi_comp_objects_realloc(new,
2420 stripe_number_roundup(i)) < 0) {
2422 __llapi_comp_free(new);
2425 new->llc_objects[i].l_ost_idx = \
2426 comp->llc_objects[i].l_ost_idx;
2429 new->llc_objects_count = comp->llc_objects_count;
2430 new->llc_extent.e_start = comp->llc_extent.e_start;
2431 new->llc_extent.e_end = comp->llc_extent.e_end;
2432 new->llc_id = comp->llc_id;
2433 new->llc_flags = comp->llc_flags;
2435 list_add_tail(&new->llc_list, &new_layout->llot_comp_list);
2436 new_layout->llot_cur_comp = new;
2438 new_layout->llot_is_composite = true;
2440 *dst_layout = new_layout;
2443 llapi_layout_free(new_layout);
2448 * Find all stale components.
2450 * \param[in] layout component layout list.
2451 * \param[out] comp array of stale component info.
2452 * \param[in] comp_size array size of @comp.
2453 * \param[in] mirror_ids array of mirror id that only components
2454 * belonging to these mirror will be collected.
2455 * \param[in] ids_nr number of mirror ids array.
2457 * \retval number of component info collected on sucess or
2458 * an error code on failure.
2460 int llapi_mirror_find_stale(struct llapi_layout *layout,
2461 struct llapi_resync_comp *comp, size_t comp_size,
2462 __u16 *mirror_ids, int ids_nr)
2467 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_FIRST);
2475 uint64_t start, end;
2477 rc = llapi_layout_comp_flags_get(layout, &flags);
2481 if (!(flags & LCME_FL_STALE))
2484 rc = llapi_layout_mirror_id_get(layout, &mirror_id);
2488 /* the caller only wants stale components from specific
2493 for (j = 0; j < ids_nr; j++) {
2494 if (mirror_ids[j] == mirror_id)
2498 /* not in the specified mirror */
2501 } else if (flags & LCME_FL_NOSYNC) {
2502 /* if not specified mirrors, do not resync "nosync"
2507 rc = llapi_layout_comp_id_get(layout, &id);
2511 rc = llapi_layout_comp_extent_get(layout, &start, &end);
2515 /* pack this component into @comp array */
2516 comp[idx].lrc_id = id;
2517 comp[idx].lrc_mirror_id = mirror_id;
2518 comp[idx].lrc_start = start;
2519 comp[idx].lrc_end = end;
2522 if (idx >= comp_size) {
2528 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_NEXT);
2535 return rc < 0 ? rc : idx;
2538 /* locate @layout to a valid component covering file [file_start, file_end) */
2539 uint32_t llapi_mirror_find(struct llapi_layout *layout,
2540 uint64_t file_start, uint64_t file_end,
2543 uint32_t mirror_id = 0;
2546 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_FIRST);
2552 uint64_t start, end;
2553 uint32_t flags, id, rid;
2555 rc = llapi_layout_comp_flags_get(layout, &flags);
2559 if (flags & LCME_FL_STALE)
2562 rc = llapi_layout_mirror_id_get(layout, &rid);
2566 rc = llapi_layout_comp_id_get(layout, &id);
2570 rc = llapi_layout_comp_extent_get(layout, &start, &end);
2574 if (file_start >= start && file_start < end) {
2577 else if (mirror_id != rid || *endp != start)
2580 file_start = *endp = end;
2581 if (end >= file_end)
2586 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_NEXT);
2594 int llapi_mirror_resync_many(int fd, struct llapi_layout *layout,
2595 struct llapi_resync_comp *comp_array,
2596 int comp_size, uint64_t start, uint64_t end)
2599 size_t page_size = sysconf(_SC_PAGESIZE);
2600 const size_t buflen = 4 << 20; /* 4M */
2602 uint64_t pos = start;
2606 rc = posix_memalign(&buf, page_size, buflen);
2610 if (end == OBD_OBJECT_EOF)
2611 count = OBD_OBJECT_EOF;
2613 count = end - start;
2617 uint64_t mirror_end = 0;
2622 src = llapi_mirror_find(layout, pos, end, &mirror_end);
2626 if (mirror_end == OBD_OBJECT_EOF) {
2629 to_read = MIN(count, mirror_end - pos);
2630 to_read = (to_read + page_size - 1) & ~(page_size - 1);
2632 to_read = MIN(buflen, to_read);
2634 bytes_read = llapi_mirror_read(fd, src, buf, to_read, pos);
2635 if (bytes_read == 0) {
2639 if (bytes_read < 0) {
2644 /* round up to page align to make direct IO happy. */
2645 to_write = (bytes_read + page_size - 1) & ~(page_size - 1);
2647 for (i = 0; i < comp_size; i++) {
2650 size_t to_write2 = to_write;
2652 /* skip non-overlapped component */
2653 if (pos >= comp_array[i].lrc_end ||
2654 pos + to_write <= comp_array[i].lrc_start)
2657 if (pos < comp_array[i].lrc_start)
2658 pos2 = comp_array[i].lrc_start;
2660 to_write2 -= pos2 - pos;
2662 if ((pos + to_write) > comp_array[i].lrc_end)
2663 to_write2 -= pos + to_write -
2664 comp_array[i].lrc_end;
2666 written = llapi_mirror_write(fd,
2667 comp_array[i].lrc_mirror_id,
2672 * this component is not written successfully,
2673 * mark it using its lrc_synced, it is supposed
2674 * to be false before getting here.
2676 * And before this function returns, all
2677 * elements of comp_array will reverse their
2678 * lrc_synced flag to reflect their true
2681 comp_array[i].lrc_synced = true;
2684 assert(written == to_write2);
2688 count -= bytes_read;
2694 for (i = 0; i < comp_size; i++)
2695 comp_array[i].lrc_synced = false;
2699 for (i = 0; i < comp_size; i++) {
2700 comp_array[i].lrc_synced = !comp_array[i].lrc_synced;
2701 if (comp_array[i].lrc_synced && pos & (page_size - 1)) {
2702 rc = llapi_mirror_truncate(fd,
2703 comp_array[i].lrc_mirror_id, pos);
2705 comp_array[i].lrc_synced = false;
2709 /* partially successful is successful */