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
70 * An Opaque data type abstracting the layout of a Lustre file.
73 uint32_t llot_magic; /* LLAPI_LAYOUT_MAGIC */
76 bool llot_is_composite;
77 uint16_t llot_mirror_count;
78 /* Cursor pointing to one of the components in llot_comp_list */
79 struct llapi_layout_comp *llot_cur_comp;
80 struct list_head llot_comp_list;
84 * Compute the number of elements in the lmm_objects array of \a lum
85 * with size \a lum_size.
87 * \param[in] lum the struct lov_user_md to check
88 * \param[in] lum_size the number of bytes in \a lum
90 * \retval number of elements in array lum->lmm_objects
92 static int llapi_layout_objects_in_lum(struct lov_user_md *lum, size_t lum_size)
97 if (lum_size < lov_user_md_size(0, LOV_MAGIC_V1))
100 if (lum->lmm_magic == __swab32(LOV_MAGIC_V1) ||
101 lum->lmm_magic == __swab32(LOV_MAGIC_V3))
102 magic = __swab32(lum->lmm_magic);
104 magic = lum->lmm_magic;
106 base_size = lov_user_md_size(0, magic);
108 if (lum_size <= base_size)
111 return (lum_size - base_size) / sizeof(lum->lmm_objects[0]);
115 * Byte-swap the fields of struct lov_user_md.
117 * XXX Rather than duplicating swabbing code here, we should eventually
118 * refactor the needed functions in lustre/ptlrpc/pack_generic.c
119 * into a library that can be shared between kernel and user code.
122 llapi_layout_swab_lov_user_md(struct lov_user_md *lum, int lum_size)
124 int i, j, ent_count, obj_count;
125 struct lov_comp_md_v1 *comp_v1 = NULL;
126 struct lov_comp_md_entry_v1 *ent;
127 struct lov_user_ost_data *lod;
129 if (lum->lmm_magic != __swab32(LOV_MAGIC_V1) &&
130 lum->lmm_magic != __swab32(LOV_MAGIC_V3) &&
131 lum->lmm_magic != __swab32(LOV_MAGIC_COMP_V1))
134 if (lum->lmm_magic == __swab32(LOV_MAGIC_COMP_V1))
135 comp_v1 = (struct lov_comp_md_v1 *)lum;
137 if (comp_v1 != NULL) {
138 __swab32s(&comp_v1->lcm_magic);
139 __swab32s(&comp_v1->lcm_size);
140 __swab32s(&comp_v1->lcm_layout_gen);
141 __swab16s(&comp_v1->lcm_flags);
142 __swab16s(&comp_v1->lcm_entry_count);
143 ent_count = comp_v1->lcm_entry_count;
148 for (i = 0; i < ent_count; i++) {
149 if (comp_v1 != NULL) {
150 ent = &comp_v1->lcm_entries[i];
151 __swab32s(&ent->lcme_id);
152 __swab32s(&ent->lcme_flags);
153 __swab64s(&ent->lcme_timestamp);
154 __swab64s(&ent->lcme_extent.e_start);
155 __swab64s(&ent->lcme_extent.e_end);
156 __swab32s(&ent->lcme_offset);
157 __swab32s(&ent->lcme_size);
159 lum = (struct lov_user_md *)((char *)comp_v1 +
161 lum_size = ent->lcme_size;
163 obj_count = llapi_layout_objects_in_lum(lum, lum_size);
165 __swab32s(&lum->lmm_magic);
166 __swab32s(&lum->lmm_pattern);
167 __swab32s(&lum->lmm_stripe_size);
168 __swab16s(&lum->lmm_stripe_count);
169 __swab16s(&lum->lmm_stripe_offset);
171 if (lum->lmm_magic != LOV_MAGIC_V1) {
172 struct lov_user_md_v3 *v3;
173 v3 = (struct lov_user_md_v3 *)lum;
174 lod = v3->lmm_objects;
176 lod = lum->lmm_objects;
179 for (j = 0; j < obj_count; j++)
180 __swab32s(&lod[j].l_ost_idx);
185 * (Re-)allocate llc_objects[] to \a num_stripes stripes.
187 * Copy over existing llc_objects[], if any, to the new llc_objects[].
189 * \param[in] layout existing layout to be modified
190 * \param[in] num_stripes number of stripes in new layout
192 * \retval 0 if the objects are re-allocated successfully
193 * \retval -1 on error with errno set
195 static int __llapi_comp_objects_realloc(struct llapi_layout_comp *comp,
196 unsigned int new_stripes)
198 struct lov_user_ost_data_v1 *new_objects;
201 if (new_stripes > LOV_MAX_STRIPE_COUNT) {
206 if (new_stripes == comp->llc_objects_count)
209 if (new_stripes != 0 && new_stripes <= comp->llc_objects_count)
212 new_objects = realloc(comp->llc_objects,
213 sizeof(*new_objects) * new_stripes);
214 if (new_objects == NULL && new_stripes != 0) {
219 for (i = comp->llc_objects_count; i < new_stripes; i++)
220 new_objects[i].l_ost_idx = LLAPI_LAYOUT_IDX_MAX;
222 comp->llc_objects = new_objects;
223 comp->llc_objects_count = new_stripes;
229 * Allocate storage for a llapi_layout_comp with \a num_stripes stripes.
231 * \param[in] num_stripes number of stripes in new layout
233 * \retval valid pointer if allocation succeeds
234 * \retval NULL if allocation fails
236 static struct llapi_layout_comp *__llapi_comp_alloc(unsigned int num_stripes)
238 struct llapi_layout_comp *comp;
240 if (num_stripes > LOV_MAX_STRIPE_COUNT) {
245 comp = calloc(1, sizeof(*comp));
251 comp->llc_objects = NULL;
252 comp->llc_objects_count = 0;
254 if (__llapi_comp_objects_realloc(comp, num_stripes) < 0) {
260 comp->llc_pattern = LLAPI_LAYOUT_DEFAULT;
261 comp->llc_stripe_size = LLAPI_LAYOUT_DEFAULT;
262 comp->llc_stripe_count = LLAPI_LAYOUT_DEFAULT;
263 comp->llc_stripe_offset = LLAPI_LAYOUT_DEFAULT;
264 comp->llc_pool_name[0] = '\0';
265 comp->llc_extent.e_start = 0;
266 comp->llc_extent.e_end = LUSTRE_EOF;
269 INIT_LIST_HEAD(&comp->llc_list);
275 * Free memory allocated for \a comp
277 * \param[in] comp previously allocated by __llapi_comp_alloc()
279 static void __llapi_comp_free(struct llapi_layout_comp *comp)
281 if (comp->llc_objects != NULL)
282 free(comp->llc_objects);
287 * Free memory allocated for \a layout.
289 * \param[in] layout previously allocated by llapi_layout_alloc()
291 void llapi_layout_free(struct llapi_layout *layout)
293 struct llapi_layout_comp *comp, *n;
298 list_for_each_entry_safe(comp, n, &layout->llot_comp_list, llc_list) {
299 list_del_init(&comp->llc_list);
300 __llapi_comp_free(comp);
306 * Allocate and initialize a llapi_layout structure.
308 * \retval valid llapi_layout pointer on success
309 * \retval NULL if memory allocation fails
311 static struct llapi_layout *__llapi_layout_alloc(void)
313 struct llapi_layout *layout;
315 layout = calloc(1, sizeof(*layout));
316 if (layout == NULL) {
322 layout->llot_magic = LLAPI_LAYOUT_MAGIC;
323 layout->llot_gen = 0;
324 layout->llot_flags = 0;
325 layout->llot_is_composite = false;
326 layout->llot_mirror_count = 1;
327 layout->llot_cur_comp = NULL;
328 INIT_LIST_HEAD(&layout->llot_comp_list);
334 * Allocate and initialize a new plain layout.
336 * \retval valid llapi_layout pointer on success
337 * \retval NULL if memory allocation fails
339 struct llapi_layout *llapi_layout_alloc(void)
341 struct llapi_layout_comp *comp;
342 struct llapi_layout *layout;
344 layout = __llapi_layout_alloc();
348 comp = __llapi_comp_alloc(0);
354 list_add_tail(&comp->llc_list, &layout->llot_comp_list);
355 layout->llot_cur_comp = comp;
361 * Check if the given \a lum_size is large enough to hold the required
364 * \param[in] lum the struct lov_user_md to check
365 * \param[in] lum_size the number of bytes in \a lum
367 * \retval true the \a lum_size is too small
368 * \retval false the \a lum_size is large enough
370 static bool llapi_layout_lum_truncated(struct lov_user_md *lum, size_t lum_size)
374 if (lum_size < sizeof(lum->lmm_magic))
377 if (lum->lmm_magic == LOV_MAGIC_V1 ||
378 lum->lmm_magic == __swab32(LOV_MAGIC_V1))
379 magic = LOV_MAGIC_V1;
380 else if (lum->lmm_magic == LOV_MAGIC_V3 ||
381 lum->lmm_magic == __swab32(LOV_MAGIC_V3))
382 magic = LOV_MAGIC_V3;
383 else if (lum->lmm_magic == LOV_MAGIC_COMP_V1 ||
384 lum->lmm_magic == __swab32(LOV_MAGIC_COMP_V1))
385 magic = LOV_MAGIC_COMP_V1;
389 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3)
390 return lum_size < lov_user_md_size(0, magic);
392 return lum_size < sizeof(struct lov_comp_md_v1);
395 /* Verify if the objects count in lum is consistent with the
396 * stripe count in lum. It applies to regular file only. */
397 static bool llapi_layout_lum_valid(struct lov_user_md *lum, int lum_size)
399 struct lov_comp_md_v1 *comp_v1 = NULL;
400 int i, ent_count, obj_count;
402 if (lum->lmm_magic == LOV_MAGIC_COMP_V1) {
403 comp_v1 = (struct lov_comp_md_v1 *)lum;
404 ent_count = comp_v1->lcm_entry_count;
405 } else if (lum->lmm_magic == LOV_MAGIC_V1 ||
406 lum->lmm_magic == LOV_MAGIC_V3) {
412 for (i = 0; i < ent_count; i++) {
414 lum = (struct lov_user_md *)((char *)comp_v1 +
415 comp_v1->lcm_entries[i].lcme_offset);
416 lum_size = comp_v1->lcm_entries[i].lcme_size;
418 obj_count = llapi_layout_objects_in_lum(lum, lum_size);
421 if (!(comp_v1->lcm_entries[i].lcme_flags &
422 LCME_FL_INIT) && obj_count != 0)
424 } else if (obj_count != lum->lmm_stripe_count) {
432 * Convert the data from a lov_user_md to a newly allocated llapi_layout.
433 * The caller is responsible for freeing the returned pointer.
435 * \param[in] lov_xattr LOV user metadata xattr to copy data from
436 * \param[in] lov_xattr_size size the lov_xattr_size passed in
437 * \param[in] flags bitwise-or'd flags to control the behavior
439 * \retval valid llapi_layout pointer on success
440 * \retval NULL if memory allocation fails
442 struct llapi_layout *llapi_layout_get_by_xattr(void *lov_xattr,
443 ssize_t lov_xattr_size,
446 struct lov_user_md *lum = lov_xattr;
447 struct lov_comp_md_v1 *comp_v1 = NULL;
448 struct lov_comp_md_entry_v1 *ent;
449 struct lov_user_md *v1;
450 struct llapi_layout *layout = NULL;
451 struct llapi_layout_comp *comp;
452 int i, ent_count = 0, obj_count;
454 if (lov_xattr == NULL || lov_xattr_size <= 0) {
459 /* Return an error if we got back a partial layout. */
460 if (llapi_layout_lum_truncated(lov_xattr, lov_xattr_size)) {
465 #if __BYTE_ORDER == __BIG_ENDIAN
466 if (flags & LLAPI_LXF_COPY) {
467 lum = malloc(lov_xattr_size);
472 memcpy(lum, lov_xattr, lov_xattr_size);
476 llapi_layout_swab_lov_user_md(lum, lov_xattr_size);
478 if ((flags & LLAPI_LXF_CHECK) &&
479 !llapi_layout_lum_valid(lum, lov_xattr_size)) {
484 layout = __llapi_layout_alloc();
485 if (layout == NULL) {
490 if (lum->lmm_magic == LOV_MAGIC_COMP_V1) {
491 comp_v1 = (struct lov_comp_md_v1 *)lum;
492 ent_count = comp_v1->lcm_entry_count;
493 layout->llot_gen = comp_v1->lcm_layout_gen;
494 layout->llot_is_composite = true;
495 layout->llot_mirror_count = comp_v1->lcm_mirror_count + 1;
496 layout->llot_gen = comp_v1->lcm_layout_gen;
497 layout->llot_flags = comp_v1->lcm_flags;
498 } else if (lum->lmm_magic == LOV_MAGIC_V1 ||
499 lum->lmm_magic == LOV_MAGIC_V3) {
501 layout->llot_is_composite = false;
503 if (lov_xattr_size <= 0) {
512 if (ent_count == 0) {
517 v1 = (struct lov_user_md *)lum;
518 for (i = 0; i < ent_count; i++) {
519 if (comp_v1 != NULL) {
520 ent = &comp_v1->lcm_entries[i];
521 v1 = (struct lov_user_md *)((char *)comp_v1 +
523 lov_xattr_size = ent->lcme_size;
528 obj_count = llapi_layout_objects_in_lum(v1, lov_xattr_size);
529 comp = __llapi_comp_alloc(obj_count);
534 comp->llc_extent.e_start = ent->lcme_extent.e_start;
535 comp->llc_extent.e_end = ent->lcme_extent.e_end;
536 comp->llc_id = ent->lcme_id;
537 comp->llc_flags = ent->lcme_flags;
538 if (comp->llc_flags & LCME_FL_NOSYNC)
539 comp->llc_timestamp = ent->lcme_timestamp;
541 comp->llc_extent.e_start = 0;
542 comp->llc_extent.e_end = LUSTRE_EOF;
547 if (v1->lmm_pattern == LOV_PATTERN_RAID0)
548 comp->llc_pattern = LLAPI_LAYOUT_RAID0;
549 else if (v1->lmm_pattern == (LOV_PATTERN_RAID0 |
550 LOV_PATTERN_OVERSTRIPING))
551 comp->llc_pattern = LLAPI_LAYOUT_OVERSTRIPING;
553 /* Lustre only supports RAID0 for now. */
554 comp->llc_pattern = v1->lmm_pattern;
556 if (v1->lmm_stripe_size == 0)
557 comp->llc_stripe_size = LLAPI_LAYOUT_DEFAULT;
559 comp->llc_stripe_size = v1->lmm_stripe_size;
561 if (v1->lmm_stripe_count == (typeof(v1->lmm_stripe_count))-1)
562 comp->llc_stripe_count = LLAPI_LAYOUT_WIDE;
563 else if (v1->lmm_stripe_count == 0)
564 comp->llc_stripe_count = LLAPI_LAYOUT_DEFAULT;
566 comp->llc_stripe_count = v1->lmm_stripe_count;
568 if (v1->lmm_stripe_offset ==
569 (typeof(v1->lmm_stripe_offset))-1)
570 comp->llc_stripe_offset = LLAPI_LAYOUT_DEFAULT;
572 comp->llc_stripe_offset = v1->lmm_stripe_offset;
574 if (v1->lmm_magic != LOV_USER_MAGIC_V1) {
575 const struct lov_user_md_v3 *lumv3;
576 lumv3 = (struct lov_user_md_v3 *)v1;
577 snprintf(comp->llc_pool_name,
578 sizeof(comp->llc_pool_name),
579 "%s", lumv3->lmm_pool_name);
580 memcpy(comp->llc_objects, lumv3->lmm_objects,
581 obj_count * sizeof(lumv3->lmm_objects[0]));
583 const struct lov_user_md_v1 *lumv1;
584 lumv1 = (struct lov_user_md_v1 *)v1;
585 memcpy(comp->llc_objects, lumv1->lmm_objects,
586 obj_count * sizeof(lumv1->lmm_objects[0]));
590 comp->llc_stripe_offset =
591 comp->llc_objects[0].l_ost_idx;
593 comp->llc_ondisk = true;
594 list_add_tail(&comp->llc_list, &layout->llot_comp_list);
595 layout->llot_cur_comp = comp;
599 if (lum != lov_xattr)
603 llapi_layout_free(layout);
608 __u32 llapi_pattern_to_lov(uint64_t pattern)
613 case LLAPI_LAYOUT_DEFAULT:
614 lov_pattern = LOV_PATTERN_RAID0;
616 case LLAPI_LAYOUT_RAID0:
617 lov_pattern = LOV_PATTERN_RAID0;
619 case LLAPI_LAYOUT_MDT:
620 lov_pattern = LOV_PATTERN_MDT;
622 case LLAPI_LAYOUT_OVERSTRIPING:
623 lov_pattern = LOV_PATTERN_OVERSTRIPING | LOV_PATTERN_RAID0;
626 lov_pattern = EINVAL;
633 * Convert the data from a llapi_layout to a newly allocated lov_user_md.
634 * The caller is responsible for freeing the returned pointer.
636 * \param[in] layout the layout to copy from
638 * \retval valid lov_user_md pointer on success
639 * \retval NULL if memory allocation fails or the layout is invalid
641 static struct lov_user_md *
642 llapi_layout_to_lum(const struct llapi_layout *layout)
644 struct llapi_layout_comp *comp;
645 struct lov_comp_md_v1 *comp_v1 = NULL;
646 struct lov_comp_md_entry_v1 *ent;
647 struct lov_user_md *lum = NULL;
652 if (layout == NULL ||
653 list_empty((struct list_head *)&layout->llot_comp_list)) {
658 /* Allocate header of lov_comp_md_v1 if necessary */
659 if (layout->llot_is_composite) {
662 list_for_each_entry(comp, &layout->llot_comp_list, llc_list)
665 lum_size = sizeof(*comp_v1) + comp_cnt * sizeof(*ent);
666 lum = calloc(lum_size, 1);
671 comp_v1 = (struct lov_comp_md_v1 *)lum;
672 comp_v1->lcm_magic = LOV_USER_MAGIC_COMP_V1;
673 comp_v1->lcm_size = lum_size;
674 comp_v1->lcm_layout_gen = 0;
675 comp_v1->lcm_flags = layout->llot_flags;
676 comp_v1->lcm_entry_count = comp_cnt;
677 comp_v1->lcm_mirror_count = layout->llot_mirror_count - 1;
681 list_for_each_entry(comp, &layout->llot_comp_list, llc_list) {
682 struct lov_user_md *blob;
685 int i, obj_count = 0;
686 struct lov_user_ost_data *lmm_objects;
687 uint64_t pattern = comp->llc_pattern;
689 if ((pattern & LLAPI_LAYOUT_SPECIFIC) != 0) {
690 if (comp->llc_objects_count <
691 comp->llc_stripe_count) {
695 magic = LOV_USER_MAGIC_SPECIFIC;
696 obj_count = comp->llc_stripe_count;
697 pattern &= ~LLAPI_LAYOUT_SPECIFIC;
698 } else if (strlen(comp->llc_pool_name) != 0) {
699 magic = LOV_USER_MAGIC_V3;
701 magic = LOV_USER_MAGIC_V1;
703 /* All stripes must be specified when the pattern contains
704 * LLAPI_LAYOUT_SPECIFIC */
705 for (i = 0; i < obj_count; i++) {
706 if (comp->llc_objects[i].l_ost_idx ==
707 LLAPI_LAYOUT_IDX_MAX) {
713 blob_size = lov_user_md_size(obj_count, magic);
714 blob = realloc(lum, lum_size + blob_size);
720 comp_v1 = (struct lov_comp_md_v1 *)lum;
721 blob = (struct lov_user_md *)((char *)lum + lum_size);
722 lum_size += blob_size;
725 blob->lmm_magic = magic;
726 blob->lmm_pattern = llapi_pattern_to_lov(pattern);
727 if (blob->lmm_pattern == EINVAL) {
732 if (comp->llc_stripe_size == LLAPI_LAYOUT_DEFAULT)
733 blob->lmm_stripe_size = 0;
735 blob->lmm_stripe_size = comp->llc_stripe_size;
737 if (comp->llc_stripe_count == LLAPI_LAYOUT_DEFAULT)
738 blob->lmm_stripe_count = 0;
739 else if (comp->llc_stripe_count == LLAPI_LAYOUT_WIDE)
740 blob->lmm_stripe_count = LOV_ALL_STRIPES;
742 blob->lmm_stripe_count = comp->llc_stripe_count;
744 if (comp->llc_stripe_offset == LLAPI_LAYOUT_DEFAULT)
745 blob->lmm_stripe_offset = -1;
747 blob->lmm_stripe_offset = comp->llc_stripe_offset;
749 if (magic == LOV_USER_MAGIC_V3 ||
750 magic == LOV_USER_MAGIC_SPECIFIC) {
751 struct lov_user_md_v3 *lumv3 =
752 (struct lov_user_md_v3 *)blob;
754 if (comp->llc_pool_name[0] != '\0') {
755 strncpy(lumv3->lmm_pool_name,
757 sizeof(lumv3->lmm_pool_name));
759 memset(lumv3->lmm_pool_name, 0,
760 sizeof(lumv3->lmm_pool_name));
762 lmm_objects = lumv3->lmm_objects;
764 lmm_objects = blob->lmm_objects;
767 for (i = 0; i < obj_count; i++)
768 lmm_objects[i].l_ost_idx =
769 comp->llc_objects[i].l_ost_idx;
771 if (layout->llot_is_composite) {
772 ent = &comp_v1->lcm_entries[ent_idx];
773 ent->lcme_id = comp->llc_id;
774 ent->lcme_flags = comp->llc_flags;
775 if (ent->lcme_flags & LCME_FL_NOSYNC)
776 ent->lcme_timestamp = comp->llc_timestamp;
777 ent->lcme_extent.e_start = comp->llc_extent.e_start;
778 ent->lcme_extent.e_end = comp->llc_extent.e_end;
779 ent->lcme_size = blob_size;
780 ent->lcme_offset = offset;
782 comp_v1->lcm_size += blob_size;
796 * Get the parent directory of a path.
798 * \param[in] path path to get parent of
799 * \param[out] buf buffer in which to store parent path
800 * \param[in] size size in bytes of buffer \a buf
802 static void get_parent_dir(const char *path, char *buf, size_t size)
806 strncpy(buf, path, size - 1);
807 p = strrchr(buf, '/');
811 } else if (size >= 2) {
812 strncpy(buf, ".", 2);
813 buf[size - 1] = '\0';
818 * Substitute unspecified attribute values in \a layout with values
819 * from fs global settings. (lov.stripesize, lov.stripecount,
822 * \param[in] layout layout to inherit values from
823 * \param[in] path file path of the filesystem
825 static void inherit_sys_attributes(struct llapi_layout *layout,
828 struct llapi_layout_comp *comp;
829 unsigned int ssize, scount, soffset;
832 rc = sattr_cache_get_defaults(NULL, path, &scount, &ssize, &soffset);
836 list_for_each_entry(comp, &layout->llot_comp_list, llc_list) {
837 if (comp->llc_pattern == LLAPI_LAYOUT_DEFAULT)
838 comp->llc_pattern = LLAPI_LAYOUT_RAID0;
839 if (comp->llc_stripe_size == LLAPI_LAYOUT_DEFAULT)
840 comp->llc_stripe_size = ssize;
841 if (comp->llc_stripe_count == LLAPI_LAYOUT_DEFAULT)
842 comp->llc_stripe_count = scount;
843 if (comp->llc_stripe_offset == LLAPI_LAYOUT_DEFAULT)
844 comp->llc_stripe_offset = soffset;
849 * Get the current component of \a layout.
851 * \param[in] layout layout to get current component
853 * \retval valid llapi_layout_comp pointer on success
854 * \retval NULL on error
856 static struct llapi_layout_comp *
857 __llapi_layout_cur_comp(const struct llapi_layout *layout)
859 struct llapi_layout_comp *comp;
861 if (layout == NULL || layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
865 if (layout->llot_cur_comp == NULL) {
869 /* Verify data consistency */
870 list_for_each_entry(comp, &layout->llot_comp_list, llc_list)
871 if (comp == layout->llot_cur_comp)
878 * Test if any attributes of \a layout are specified.
880 * \param[in] layout the layout to check
882 * \retval true any attributes are specified
883 * \retval false all attributes are unspecified
885 static bool is_any_specified(const struct llapi_layout *layout)
887 struct llapi_layout_comp *comp;
889 comp = __llapi_layout_cur_comp(layout);
893 if (layout->llot_is_composite || layout->llot_mirror_count != 1)
896 return comp->llc_pattern != LLAPI_LAYOUT_DEFAULT ||
897 comp->llc_stripe_size != LLAPI_LAYOUT_DEFAULT ||
898 comp->llc_stripe_count != LLAPI_LAYOUT_DEFAULT ||
899 comp->llc_stripe_offset != LLAPI_LAYOUT_DEFAULT ||
900 strlen(comp->llc_pool_name);
904 * Get the striping layout for the file referenced by file descriptor \a fd.
906 * If the filesystem does not support the "lustre." xattr namespace, the
907 * file must be on a non-Lustre filesystem, so set errno to ENOTTY per
908 * convention. If the file has no "lustre.lov" data, the file will
909 * inherit default values, so return a default layout.
911 * If the kernel gives us back less than the expected amount of data,
912 * we fail with errno set to EINTR.
914 * \param[in] fd open file descriptor
915 * \param[in] flags open file descriptor
917 * \retval valid llapi_layout pointer on success
918 * \retval NULL if an error occurs
920 struct llapi_layout *llapi_layout_get_by_fd(int fd, uint32_t flags)
923 struct lov_user_md *lum;
924 struct llapi_layout *layout = NULL;
928 lum_len = XATTR_SIZE_MAX;
929 lum = malloc(lum_len);
933 bytes_read = fgetxattr(fd, XATTR_LUSTRE_LOV, lum, lum_len);
934 if (bytes_read < 0) {
935 if (errno == EOPNOTSUPP)
937 else if (errno == ENODATA)
938 layout = llapi_layout_alloc();
942 /* Directories may have a positive non-zero lum->lmm_stripe_count
943 * yet have an empty lum->lmm_objects array. For non-directories the
944 * amount of data returned from the kernel must be consistent
945 * with the stripe count. */
946 if (fstat(fd, &st) < 0)
949 layout = llapi_layout_get_by_xattr(lum, bytes_read,
950 S_ISDIR(st.st_mode) ? 0 : LLAPI_LXF_CHECK);
957 * Get the expected striping layout for a file at \a path.
959 * Substitute expected inherited attribute values for unspecified
960 * attributes. Unspecified attributes may belong to directories and
961 * never-written-to files, and indicate that default values will be
962 * assigned when files are created or first written to. A default value
963 * is inherited from the parent directory if the attribute is specified
964 * there, otherwise it is inherited from the filesystem root.
965 * Unspecified attributes normally have the value LLAPI_LAYOUT_DEFAULT.
967 * The complete \a path need not refer to an existing file or directory,
968 * but some leading portion of it must reside within a lustre filesystem.
969 * A use case for this interface would be to obtain the literal striping
970 * values that would be assigned to a new file in a given directory.
972 * \param[in] path path for which to get the expected layout
974 * \retval valid llapi_layout pointer on success
975 * \retval NULL if an error occurs
977 static struct llapi_layout *llapi_layout_expected(const char *path)
979 struct llapi_layout *path_layout = NULL;
980 char donor_path[PATH_MAX];
985 fd = open(path, O_RDONLY);
986 if (fd < 0 && errno != ENOENT)
992 path_layout = llapi_layout_get_by_fd(fd, 0);
998 if (path_layout == NULL) {
999 if (errno != ENODATA && errno != ENOENT)
1002 path_layout = llapi_layout_alloc();
1003 if (path_layout == NULL)
1007 if (is_any_specified(path_layout)) {
1008 inherit_sys_attributes(path_layout, path);
1012 llapi_layout_free(path_layout);
1014 rc = stat(path, &st);
1015 if (rc < 0 && errno != ENOENT)
1018 /* If path is a not a directory or doesn't exist, inherit layout
1019 * from parent directory. */
1020 if ((rc == 0 && !S_ISDIR(st.st_mode)) ||
1021 (rc < 0 && errno == ENOENT)) {
1022 get_parent_dir(path, donor_path, sizeof(donor_path));
1023 path_layout = llapi_layout_get_by_path(donor_path, 0);
1024 if (path_layout != NULL) {
1025 if (is_any_specified(path_layout)) {
1026 inherit_sys_attributes(path_layout, donor_path);
1029 llapi_layout_free(path_layout);
1033 /* Inherit layout from the filesystem root. */
1034 rc = llapi_search_mounts(path, 0, donor_path, NULL);
1037 path_layout = llapi_layout_get_by_path(donor_path, 0);
1038 if (path_layout == NULL)
1041 inherit_sys_attributes(path_layout, donor_path);
1046 * Get the striping layout for the file at \a path.
1048 * If \a flags contains LAYOUT_GET_EXPECTED, substitute
1049 * expected inherited attribute values for unspecified attributes. See
1050 * llapi_layout_expected().
1052 * \param[in] path path for which to get the layout
1053 * \param[in] flags flags to control how layout is retrieved
1055 * \retval valid llapi_layout pointer on success
1056 * \retval NULL if an error occurs
1058 struct llapi_layout *llapi_layout_get_by_path(const char *path, uint32_t flags)
1060 struct llapi_layout *layout = NULL;
1064 if (flags & LAYOUT_GET_EXPECTED)
1065 return llapi_layout_expected(path);
1067 fd = open(path, O_RDONLY);
1071 layout = llapi_layout_get_by_fd(fd, flags);
1080 * Get the layout for the file with FID \a fidstr in filesystem \a lustre_dir.
1082 * \param[in] lustre_dir path within Lustre filesystem containing \a fid
1083 * \param[in] fid Lustre identifier of file to get layout for
1085 * \retval valid llapi_layout pointer on success
1086 * \retval NULL if an error occurs
1088 struct llapi_layout *llapi_layout_get_by_fid(const char *lustre_dir,
1089 const struct lu_fid *fid,
1094 int saved_msg_level = llapi_msg_get_level();
1095 struct llapi_layout *layout = NULL;
1097 /* Prevent llapi internal routines from writing to console
1098 * while executing this function, then restore previous message
1100 llapi_msg_set_level(LLAPI_MSG_OFF);
1101 fd = llapi_open_by_fid(lustre_dir, fid, O_RDONLY);
1102 llapi_msg_set_level(saved_msg_level);
1107 layout = llapi_layout_get_by_fd(fd, flags);
1116 * Get the stripe count of \a layout.
1118 * \param[in] layout layout to get stripe count from
1119 * \param[out] count integer to store stripe count in
1121 * \retval 0 on success
1122 * \retval -1 if arguments are invalid
1124 int llapi_layout_stripe_count_get(const struct llapi_layout *layout,
1127 struct llapi_layout_comp *comp;
1129 comp = __llapi_layout_cur_comp(layout);
1133 if (count == NULL) {
1138 *count = comp->llc_stripe_count;
1144 * The llapi_layout API functions have these extra validity checks since
1145 * they use intuitively named macros to denote special behavior, whereas
1146 * the old API uses 0 and -1.
1149 static bool llapi_layout_stripe_count_is_valid(int64_t stripe_count)
1151 return stripe_count == LLAPI_LAYOUT_DEFAULT ||
1152 stripe_count == LLAPI_LAYOUT_WIDE ||
1153 (stripe_count != 0 && stripe_count != -1 &&
1154 llapi_stripe_count_is_valid(stripe_count));
1157 static bool llapi_layout_extension_size_is_valid(uint64_t ext_size)
1159 return (ext_size != 0 &&
1160 llapi_stripe_size_is_aligned(ext_size) &&
1161 !llapi_stripe_size_is_too_big(ext_size));
1164 static bool llapi_layout_stripe_size_is_valid(uint64_t stripe_size)
1166 return stripe_size == LLAPI_LAYOUT_DEFAULT ||
1167 (stripe_size != 0 &&
1168 llapi_stripe_size_is_aligned(stripe_size) &&
1169 !llapi_stripe_size_is_too_big(stripe_size));
1172 static bool llapi_layout_stripe_index_is_valid(int64_t stripe_index)
1174 return stripe_index == LLAPI_LAYOUT_DEFAULT ||
1175 (stripe_index >= 0 &&
1176 llapi_stripe_index_is_valid(stripe_index));
1180 * Set the stripe count of \a layout.
1182 * \param[in] layout layout to set stripe count in
1183 * \param[in] count value to be set
1185 * \retval 0 on success
1186 * \retval -1 if arguments are invalid
1188 int llapi_layout_stripe_count_set(struct llapi_layout *layout,
1191 struct llapi_layout_comp *comp;
1193 comp = __llapi_layout_cur_comp(layout);
1197 if (!llapi_layout_stripe_count_is_valid(count)) {
1202 comp->llc_stripe_count = count;
1208 * Get the stripe/extension size of \a layout.
1210 * \param[in] layout layout to get stripe size from
1211 * \param[out] size integer to store stripe size in
1212 * \param[in] extension flag if extenion size is requested
1214 * \retval 0 on success
1215 * \retval -1 if arguments are invalid
1217 static int layout_stripe_size_get(const struct llapi_layout *layout,
1218 uint64_t *size, bool extension)
1220 struct llapi_layout_comp *comp;
1223 comp = __llapi_layout_cur_comp(layout);
1232 comp_ext = comp->llc_flags & LCME_FL_EXTENSION;
1233 if ((comp_ext && !extension) || (!comp_ext && extension)) {
1238 *size = comp->llc_stripe_size;
1239 if (comp->llc_flags & LCME_FL_EXTENSION)
1240 *size *= SEL_UNIT_SIZE;
1245 int llapi_layout_stripe_size_get(const struct llapi_layout *layout,
1248 return layout_stripe_size_get(layout, size, false);
1251 int llapi_layout_extension_size_get(const struct llapi_layout *layout,
1254 return layout_stripe_size_get(layout, size, true);
1258 * Set the stripe/extension size of \a layout.
1260 * \param[in] layout layout to set stripe size in
1261 * \param[in] size value to be set
1262 * \param[in] extension flag if extenion size is passed
1264 * \retval 0 on success
1265 * \retval -1 if arguments are invalid
1267 static int layout_stripe_size_set(struct llapi_layout *layout,
1268 uint64_t size, bool extension)
1270 struct llapi_layout_comp *comp;
1273 comp = __llapi_layout_cur_comp(layout);
1277 comp_ext = comp->llc_flags & LCME_FL_EXTENSION;
1278 if ((comp_ext && !extension) || (!comp_ext && extension)) {
1284 size /= SEL_UNIT_SIZE;
1286 if ((comp_ext && !llapi_layout_extension_size_is_valid(size)) ||
1287 (!comp_ext && !llapi_layout_stripe_size_is_valid(size))) {
1292 comp->llc_stripe_size = size;
1296 int llapi_layout_stripe_size_set(struct llapi_layout *layout,
1299 return layout_stripe_size_set(layout, size, false);
1302 int llapi_layout_extension_size_set(struct llapi_layout *layout,
1305 return layout_stripe_size_set(layout, size, true);
1309 * Get the RAID pattern of \a layout.
1311 * \param[in] layout layout to get pattern from
1312 * \param[out] pattern integer to store pattern in
1314 * \retval 0 on success
1315 * \retval -1 if arguments are invalid
1317 int llapi_layout_pattern_get(const struct llapi_layout *layout,
1320 struct llapi_layout_comp *comp;
1322 comp = __llapi_layout_cur_comp(layout);
1326 if (pattern == NULL) {
1331 *pattern = comp->llc_pattern;
1337 * Set the pattern of \a layout.
1339 * \param[in] layout layout to set pattern in
1340 * \param[in] pattern value to be set
1342 * \retval 0 on success
1343 * \retval -1 if arguments are invalid or RAID pattern
1346 int llapi_layout_pattern_set(struct llapi_layout *layout, uint64_t pattern)
1348 struct llapi_layout_comp *comp;
1350 comp = __llapi_layout_cur_comp(layout);
1354 if (pattern != LLAPI_LAYOUT_DEFAULT &&
1355 pattern != LLAPI_LAYOUT_RAID0 && pattern != LLAPI_LAYOUT_MDT
1356 && pattern != LLAPI_LAYOUT_OVERSTRIPING) {
1361 comp->llc_pattern = pattern |
1362 (comp->llc_pattern & LLAPI_LAYOUT_SPECIFIC);
1367 static inline int stripe_number_roundup(int stripe_number)
1369 unsigned int round_up = (stripe_number + 8) & ~7;
1370 return round_up > LOV_MAX_STRIPE_COUNT ?
1371 LOV_MAX_STRIPE_COUNT : round_up;
1375 * Set the OST index of stripe number \a stripe_number to \a ost_index.
1377 * If only the starting stripe's OST index is specified, then this can use
1378 * the normal LOV_MAGIC_{V1,V3} layout type. If multiple OST indices are
1379 * given, then allocate an array to hold the list of indices and ensure that
1380 * the LOV_USER_MAGIC_SPECIFIC layout is used when creating the file.
1382 * \param[in] layout layout to set OST index in
1383 * \param[in] stripe_number stripe number to set index for
1384 * \param[in] ost_index the index to set
1386 * \retval 0 on success
1387 * \retval -1 if arguments are invalid or an unsupported stripe number
1388 * was specified, error returned in errno
1390 int llapi_layout_ost_index_set(struct llapi_layout *layout, int stripe_number,
1393 struct llapi_layout_comp *comp;
1395 comp = __llapi_layout_cur_comp(layout);
1399 if (!llapi_layout_stripe_index_is_valid(ost_index)) {
1404 if (stripe_number == 0 && ost_index == LLAPI_LAYOUT_DEFAULT) {
1405 comp->llc_stripe_offset = ost_index;
1406 comp->llc_pattern &= ~LLAPI_LAYOUT_SPECIFIC;
1407 __llapi_comp_objects_realloc(comp, 0);
1408 } else if (stripe_number >= 0 &&
1409 stripe_number < LOV_MAX_STRIPE_COUNT) {
1410 if (ost_index >= LLAPI_LAYOUT_IDX_MAX) {
1415 /* Preallocate a few more stripes to avoid realloc() overhead.*/
1416 if (__llapi_comp_objects_realloc(comp,
1417 stripe_number_roundup(stripe_number)) < 0)
1420 comp->llc_objects[stripe_number].l_ost_idx = ost_index;
1422 if (stripe_number == 0)
1423 comp->llc_stripe_offset = ost_index;
1425 comp->llc_pattern |= LLAPI_LAYOUT_SPECIFIC;
1427 if (comp->llc_stripe_count == LLAPI_LAYOUT_DEFAULT ||
1428 comp->llc_stripe_count <= stripe_number)
1429 comp->llc_stripe_count = stripe_number + 1;
1439 * Get the OST index associated with stripe \a stripe_number.
1441 * Stripes are indexed starting from zero.
1443 * \param[in] layout layout to get index from
1444 * \param[in] stripe_number stripe number to get index for
1445 * \param[out] index integer to store index in
1447 * \retval 0 on success
1448 * \retval -1 if arguments are invalid
1450 int llapi_layout_ost_index_get(const struct llapi_layout *layout,
1451 uint64_t stripe_number, uint64_t *index)
1453 struct llapi_layout_comp *comp;
1455 comp = __llapi_layout_cur_comp(layout);
1459 if (index == NULL) {
1464 if (stripe_number >= comp->llc_stripe_count ||
1465 stripe_number >= comp->llc_objects_count) {
1470 if (comp->llc_stripe_offset == LLAPI_LAYOUT_DEFAULT)
1471 *index = LLAPI_LAYOUT_DEFAULT;
1473 *index = comp->llc_objects[stripe_number].l_ost_idx;
1480 * Get the pool name of layout \a layout.
1482 * \param[in] layout layout to get pool name from
1483 * \param[out] dest buffer to store pool name in
1484 * \param[in] n size in bytes of buffer \a dest
1486 * \retval 0 on success
1487 * \retval -1 if arguments are invalid
1489 int llapi_layout_pool_name_get(const struct llapi_layout *layout, char *dest,
1492 struct llapi_layout_comp *comp;
1494 comp = __llapi_layout_cur_comp(layout);
1503 strncpy(dest, comp->llc_pool_name, n);
1509 * Set the name of the pool of layout \a layout.
1511 * \param[in] layout layout to set pool name in
1512 * \param[in] pool_name pool name to set
1514 * \retval 0 on success
1515 * \retval -1 if arguments are invalid or pool name is too long
1517 int llapi_layout_pool_name_set(struct llapi_layout *layout,
1518 const char *pool_name)
1520 struct llapi_layout_comp *comp;
1523 comp = __llapi_layout_cur_comp(layout);
1527 if (pool_name == NULL) {
1532 /* Strip off any 'fsname.' portion. */
1533 ptr = strchr(pool_name, '.');
1535 pool_name = ptr + 1;
1537 if (strlen(pool_name) > LOV_MAXPOOLNAME) {
1542 strncpy(comp->llc_pool_name, pool_name, sizeof(comp->llc_pool_name));
1548 * Open and possibly create a file with a given \a layout.
1550 * If \a layout is NULL this function acts as a simple wrapper for
1551 * open(). By convention, ENOTTY is returned in errno if \a path
1552 * refers to a non-Lustre file.
1554 * \param[in] path name of the file to open
1555 * \param[in] open_flags open() flags
1556 * \param[in] mode permissions to create file, filtered by umask
1557 * \param[in] layout layout to create new file with
1559 * \retval non-negative file descriptor on successful open
1560 * \retval -1 if an error occurred
1562 int llapi_layout_file_open(const char *path, int open_flags, mode_t mode,
1563 const struct llapi_layout *layout)
1568 struct lov_user_md *lum;
1572 (layout != NULL && layout->llot_magic != LLAPI_LAYOUT_MAGIC)) {
1578 rc = llapi_layout_sanity((struct llapi_layout *)layout, false,
1579 !!(layout->llot_mirror_count > 1));
1581 llapi_layout_sanity_perror(rc);
1586 /* Object creation must be postponed until after layout attributes
1587 * have been applied. */
1588 if (layout != NULL && (open_flags & O_CREAT))
1589 open_flags |= O_LOV_DELAY_CREATE;
1591 fd = open(path, open_flags, mode);
1593 if (layout == NULL || fd < 0)
1596 lum = llapi_layout_to_lum(layout);
1605 if (lum->lmm_magic == LOV_USER_MAGIC_COMP_V1)
1606 lum_size = ((struct lov_comp_md_v1 *)lum)->lcm_size;
1607 else if (lum->lmm_magic == LOV_USER_MAGIC_SPECIFIC)
1608 lum_size = lov_user_md_size(lum->lmm_stripe_count,
1611 lum_size = lov_user_md_size(0, lum->lmm_magic);
1613 rc = fsetxattr(fd, XATTR_LUSTRE_LOV, lum, lum_size, 0);
1622 errno = errno == EOPNOTSUPP ? ENOTTY : errno;
1628 * Create a file with a given \a layout.
1630 * Force O_CREAT and O_EXCL flags on so caller is assured that file was
1631 * created with the given \a layout on successful function return.
1633 * \param[in] path name of the file to open
1634 * \param[in] open_flags open() flags
1635 * \param[in] mode permissions to create new file with
1636 * \param[in] layout layout to create new file with
1638 * \retval non-negative file descriptor on successful open
1639 * \retval -1 if an error occurred
1641 int llapi_layout_file_create(const char *path, int open_flags, int mode,
1642 const struct llapi_layout *layout)
1644 return llapi_layout_file_open(path, open_flags|O_CREAT|O_EXCL, mode,
1648 int llapi_layout_flags_get(struct llapi_layout *layout, uint32_t *flags)
1650 if (layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
1655 *flags = layout->llot_flags;
1660 * Set flags to the header of a component layout.
1662 int llapi_layout_flags_set(struct llapi_layout *layout, uint32_t flags)
1664 if (layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
1669 layout->llot_flags = flags;
1673 const char *llapi_layout_flags_string(uint32_t flags)
1675 switch (flags & LCM_FL_FLR_MASK) {
1678 case LCM_FL_WRITE_PENDING:
1680 case LCM_FL_SYNC_PENDING:
1687 const __u16 llapi_layout_string_flags(char *string)
1689 if (strncmp(string, "ro", strlen(string)) == 0)
1690 return LCM_FL_RDONLY;
1691 if (strncmp(string, "wp", strlen(string)) == 0)
1692 return LCM_FL_WRITE_PENDING;
1693 if (strncmp(string, "sp", strlen(string)) == 0)
1694 return LCM_FL_SYNC_PENDING;
1700 * llapi_layout_mirror_count_is_valid() - Check the validity of mirror count.
1701 * @count: Mirror count value to be checked.
1703 * This function checks the validity of mirror count.
1705 * Return: true on success or false on failure.
1707 static bool llapi_layout_mirror_count_is_valid(uint16_t count)
1709 return count >= 0 && count <= LUSTRE_MIRROR_COUNT_MAX;
1713 * llapi_layout_mirror_count_get() - Get mirror count from the header of
1715 * @layout: Layout to get mirror count from.
1716 * @count: Returned mirror count value.
1718 * This function gets mirror count from the header of a layout.
1720 * Return: 0 on success or -1 on failure.
1722 int llapi_layout_mirror_count_get(struct llapi_layout *layout,
1725 if (layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
1730 *count = layout->llot_mirror_count;
1735 * llapi_layout_mirror_count_set() - Set mirror count to the header of a layout.
1736 * @layout: Layout to set mirror count in.
1737 * @count: Mirror count value to be set.
1739 * This function sets mirror count to the header of a layout.
1741 * Return: 0 on success or -1 on failure.
1743 int llapi_layout_mirror_count_set(struct llapi_layout *layout,
1746 if (layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
1751 if (!llapi_layout_mirror_count_is_valid(count)) {
1756 layout->llot_mirror_count = count;
1761 * Fetch the start and end offset of the current layout component.
1763 * \param[in] layout the layout component
1764 * \param[out] start extent start, inclusive
1765 * \param[out] end extent end, exclusive
1767 * \retval 0 on success
1768 * \retval <0 if error occurs
1770 int llapi_layout_comp_extent_get(const struct llapi_layout *layout,
1771 uint64_t *start, uint64_t *end)
1773 struct llapi_layout_comp *comp;
1775 comp = __llapi_layout_cur_comp(layout);
1779 if (start == NULL || end == NULL) {
1784 *start = comp->llc_extent.e_start;
1785 *end = comp->llc_extent.e_end;
1791 * Set the layout extent of a layout.
1793 * \param[in] layout the layout to be set
1794 * \param[in] start extent start, inclusive
1795 * \param[in] end extent end, exclusive
1797 * \retval 0 on success
1798 * \retval <0 if error occurs
1800 int llapi_layout_comp_extent_set(struct llapi_layout *layout,
1801 uint64_t start, uint64_t end)
1803 struct llapi_layout_comp *comp;
1805 comp = __llapi_layout_cur_comp(layout);
1814 comp->llc_extent.e_start = start;
1815 comp->llc_extent.e_end = end;
1816 layout->llot_is_composite = true;
1822 * Gets the attribute flags of the current component.
1824 * \param[in] layout the layout component
1825 * \param[out] flags stored the returned component flags
1827 * \retval 0 on success
1828 * \retval <0 if error occurs
1830 int llapi_layout_comp_flags_get(const struct llapi_layout *layout,
1833 struct llapi_layout_comp *comp;
1835 comp = __llapi_layout_cur_comp(layout);
1839 if (flags == NULL) {
1844 *flags = comp->llc_flags;
1850 * Sets the specified flags of the current component leaving other flags as-is.
1852 * \param[in] layout the layout component
1853 * \param[in] flags component flags to be set
1855 * \retval 0 on success
1856 * \retval <0 if error occurs
1858 int llapi_layout_comp_flags_set(struct llapi_layout *layout, uint32_t flags)
1860 struct llapi_layout_comp *comp;
1862 comp = __llapi_layout_cur_comp(layout);
1866 comp->llc_flags |= flags;
1872 * Clears the flags specified in the flags leaving other flags as-is.
1874 * \param[in] layout the layout component
1875 * \param[in] flags component flags to be cleared
1877 * \retval 0 on success
1878 * \retval <0 if error occurs
1880 int llapi_layout_comp_flags_clear(struct llapi_layout *layout,
1883 struct llapi_layout_comp *comp;
1885 comp = __llapi_layout_cur_comp(layout);
1889 comp->llc_flags &= ~flags;
1895 * Fetches the file-unique component ID of the current layout component.
1897 * \param[in] layout the layout component
1898 * \param[out] id stored the returned component ID
1900 * \retval 0 on success
1901 * \retval <0 if error occurs
1903 int llapi_layout_comp_id_get(const struct llapi_layout *layout, uint32_t *id)
1905 struct llapi_layout_comp *comp;
1907 comp = __llapi_layout_cur_comp(layout);
1921 * Return the mirror id of the current layout component.
1923 * \param[in] layout the layout component
1924 * \param[out] id stored the returned mirror ID
1926 * \retval 0 on success
1927 * \retval <0 if error occurs
1929 int llapi_layout_mirror_id_get(const struct llapi_layout *layout, uint32_t *id)
1931 struct llapi_layout_comp *comp;
1933 comp = __llapi_layout_cur_comp(layout);
1942 *id = mirror_id_of(comp->llc_id);
1948 * Adds a component to \a layout, the new component will be added to
1949 * the tail of components list and it'll inherit attributes of existing
1950 * ones. The \a layout will change it's current component pointer to
1951 * the newly added component, and it'll be turned into a composite
1952 * layout if it was not before the adding.
1954 * \param[in] layout existing composite or plain layout
1956 * \retval 0 on success
1957 * \retval <0 if error occurs
1959 int llapi_layout_comp_add(struct llapi_layout *layout)
1961 struct llapi_layout_comp *last, *comp, *new;
1962 bool composite = layout->llot_is_composite;
1964 comp = __llapi_layout_cur_comp(layout);
1968 new = __llapi_comp_alloc(0);
1972 last = list_entry(layout->llot_comp_list.prev, typeof(*last),
1975 list_add_tail(&new->llc_list, &layout->llot_comp_list);
1977 /* We must mark the layout composite for the sanity check, but it may
1978 * not stay that way if the check fails */
1979 layout->llot_is_composite = true;
1980 layout->llot_cur_comp = new;
1982 /* We need to set a temporary non-zero value for "end" when we call
1983 * comp_extent_set, so we use LUSTRE_EOF-1, which is > all allowed
1984 * for the end of the previous component. (If we're adding this
1985 * component, the end of the previous component cannot be EOF.) */
1986 if (llapi_layout_comp_extent_set(layout, last->llc_extent.e_end,
1988 llapi_layout_comp_del(layout);
1989 layout->llot_is_composite = composite;
1996 * Adds a first component of a mirror to \a layout.
1997 * The \a layout will change it's current component pointer to
1998 * the newly added component, and it'll be turned into a composite
1999 * layout if it was not before the adding.
2001 * \param[in] layout existing composite or plain layout
2003 * \retval 0 on success
2004 * \retval <0 if error occurs
2006 int llapi_layout_add_first_comp(struct llapi_layout *layout)
2008 struct llapi_layout_comp *comp, *new;
2010 comp = __llapi_layout_cur_comp(layout);
2014 new = __llapi_comp_alloc(0);
2018 new->llc_extent.e_start = 0;
2020 list_add_tail(&new->llc_list, &layout->llot_comp_list);
2021 layout->llot_cur_comp = new;
2022 layout->llot_is_composite = true;
2028 * Deletes current component from the composite layout. The component
2029 * to be deleted must be the tail of components list, and it can't be
2030 * the only component in the layout.
2032 * \param[in] layout composite layout
2034 * \retval 0 on success
2035 * \retval <0 if error occurs
2037 int llapi_layout_comp_del(struct llapi_layout *layout)
2039 struct llapi_layout_comp *comp;
2041 comp = __llapi_layout_cur_comp(layout);
2045 if (!layout->llot_is_composite) {
2050 /* It must be the tail of the list (for PFL, can be relaxed
2051 * once we get mirrored components) */
2052 if (comp->llc_list.next != &layout->llot_comp_list) {
2056 layout->llot_cur_comp =
2057 list_entry(comp->llc_list.prev, typeof(*comp), llc_list);
2058 if (comp->llc_list.prev == &layout->llot_comp_list)
2059 layout->llot_cur_comp = NULL;
2061 list_del_init(&comp->llc_list);
2062 __llapi_comp_free(comp);
2068 * Move the current component pointer to the component with
2069 * specified component ID.
2071 * \param[in] layout composite layout
2072 * \param[in] id component ID
2074 * \retval =0 : moved successfully
2075 * \retval <0 if error occurs
2077 int llapi_layout_comp_use_id(struct llapi_layout *layout, uint32_t comp_id)
2079 struct llapi_layout_comp *comp;
2081 comp = __llapi_layout_cur_comp(layout);
2083 return -1; /* use previously set errno */
2085 if (!layout->llot_is_composite) {
2090 if (comp_id == LCME_ID_INVAL) {
2095 list_for_each_entry(comp, &layout->llot_comp_list, llc_list) {
2096 if (comp->llc_id == comp_id) {
2097 layout->llot_cur_comp = comp;
2106 * Move the current component pointer to a specified position.
2108 * \param[in] layout composite layout
2109 * \param[in] pos the position to be moved, it can be:
2110 * LLAPI_LAYOUT_COMP_USE_FIRST: use first component
2111 * LLAPI_LAYOUT_COMP_USE_LAST: use last component
2112 * LLAPI_LAYOUT_COMP_USE_NEXT: use component after current
2113 * LLAPI_LAYOUT_COMP_USE_PREV: use component before current
2115 * \retval =0 : moved successfully
2116 * \retval =1 : at last component with NEXT, at first component with PREV
2117 * \retval <0 if error occurs
2119 int llapi_layout_comp_use(struct llapi_layout *layout,
2120 enum llapi_layout_comp_use pos)
2122 struct llapi_layout_comp *comp, *head, *tail;
2124 comp = __llapi_layout_cur_comp(layout);
2128 if (!layout->llot_is_composite) {
2129 if (pos == LLAPI_LAYOUT_COMP_USE_FIRST ||
2130 pos == LLAPI_LAYOUT_COMP_USE_LAST)
2136 head = list_entry(layout->llot_comp_list.next, typeof(*head), llc_list);
2137 tail = list_entry(layout->llot_comp_list.prev, typeof(*tail), llc_list);
2139 case LLAPI_LAYOUT_COMP_USE_FIRST:
2140 layout->llot_cur_comp = head;
2142 case LLAPI_LAYOUT_COMP_USE_NEXT:
2147 layout->llot_cur_comp = list_entry(comp->llc_list.next,
2148 typeof(*comp), llc_list);
2150 case LLAPI_LAYOUT_COMP_USE_LAST:
2151 layout->llot_cur_comp = tail;
2153 case LLAPI_LAYOUT_COMP_USE_PREV:
2158 layout->llot_cur_comp = list_entry(comp->llc_list.prev,
2159 typeof(*comp), llc_list);
2170 * Add layout component(s) to an existing file.
2172 * \param[in] path The path name of the file
2173 * \param[in] layout The layout component(s) to be added
2175 int llapi_layout_file_comp_add(const char *path,
2176 const struct llapi_layout *layout)
2178 int rc, fd = -1, lum_size, tmp_errno = 0;
2179 struct llapi_layout *existing_layout = NULL;
2180 struct lov_user_md *lum = NULL;
2182 if (path == NULL || layout == NULL ||
2183 layout->llot_magic != LLAPI_LAYOUT_MAGIC) {
2188 fd = open(path, O_RDWR);
2195 existing_layout = llapi_layout_get_by_fd(fd, 0);
2196 if (existing_layout == NULL) {
2202 rc = llapi_layout_merge(&existing_layout, layout);
2209 rc = llapi_layout_sanity(existing_layout, false, false);
2212 llapi_layout_sanity_perror(rc);
2217 lum = llapi_layout_to_lum(layout);
2224 if (lum->lmm_magic != LOV_USER_MAGIC_COMP_V1) {
2229 lum_size = ((struct lov_comp_md_v1 *)lum)->lcm_size;
2231 rc = fsetxattr(fd, XATTR_LUSTRE_LOV".add", lum, lum_size, 0);
2241 llapi_layout_free(existing_layout);
2247 * Delete component(s) by the specified component id or component flags
2248 * from an existing file.
2250 * \param[in] path path name of the file
2251 * \param[in] id unique component ID
2252 * \param[in] flags flags: LCME_FL_* or;
2253 * negative flags: (LCME_FL_NEG|LCME_FL_*)
2255 int llapi_layout_file_comp_del(const char *path, uint32_t id, uint32_t flags)
2257 int rc = 0, fd = -1, lum_size, tmp_errno = 0;
2258 struct llapi_layout *layout;
2259 struct llapi_layout_comp *comp, *next;
2260 struct llapi_layout *existing_layout = NULL;
2261 struct lov_user_md *lum = NULL;
2263 if (path == NULL || id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2268 /* Can only specify ID or flags, not both, not none. */
2269 if ((id != LCME_ID_INVAL && flags != 0) ||
2270 (id == LCME_ID_INVAL && flags == 0)) {
2275 layout = llapi_layout_alloc();
2279 llapi_layout_comp_extent_set(layout, 0, LUSTRE_EOF);
2280 comp = __llapi_layout_cur_comp(layout);
2288 comp->llc_flags = flags;
2290 lum = llapi_layout_to_lum(layout);
2296 lum_size = ((struct lov_comp_md_v1 *)lum)->lcm_size;
2298 fd = open(path, O_RDWR);
2305 existing_layout = llapi_layout_get_by_fd(fd, 0);
2306 if (existing_layout == NULL) {
2314 while (rc == 0 && existing_layout->llot_cur_comp != NULL) {
2315 rc = llapi_layout_comp_use(existing_layout, comp ?
2316 LLAPI_LAYOUT_COMP_USE_PREV :
2317 LLAPI_LAYOUT_COMP_USE_LAST);
2322 comp = __llapi_layout_cur_comp(existing_layout);
2328 if (id != LCME_ID_INVAL && id != comp->llc_id)
2330 else if ((flags & LCME_FL_NEG) && (flags & comp->llc_flags))
2332 else if (flags && !(flags & comp->llc_flags))
2335 rc = llapi_layout_comp_del(existing_layout);
2336 /* the layout position is moved to previous one, adjust */
2344 rc = llapi_layout_sanity(existing_layout, false, false);
2347 llapi_layout_sanity_perror(rc);
2352 rc = fsetxattr(fd, XATTR_LUSTRE_LOV".del", lum, lum_size, 0);
2363 llapi_layout_free(layout);
2364 llapi_layout_free(existing_layout);
2370 /* Internal utility function to apply flags for sanity checking */
2371 static void llapi_layout_comp_apply_flags(struct llapi_layout_comp *comp,
2374 if (flags & LCME_FL_NEG)
2375 comp->llc_flags &= ~flags;
2377 comp->llc_flags |= flags;
2380 struct llapi_layout_apply_flags_args {
2382 uint32_t *lfa_flags;
2388 static int llapi_layout_apply_flags_cb(struct llapi_layout *layout,
2391 struct llapi_layout_apply_flags_args *args = arg;
2392 struct llapi_layout_comp *comp;
2395 comp = __llapi_layout_cur_comp(layout);
2398 return LLAPI_LAYOUT_ITER_STOP;
2401 for (i = 0; i < args->lfa_count; i++) {
2402 if (comp->llc_id == args->lfa_ids[i])
2403 llapi_layout_comp_apply_flags(comp, args->lfa_flags[i]);
2406 return LLAPI_LAYOUT_ITER_CONT;
2409 /* Apply flags to the layout for sanity checking */
2410 static int llapi_layout_apply_flags(struct llapi_layout *layout, uint32_t *ids,
2411 uint32_t *flags, int count)
2413 struct llapi_layout_apply_flags_args args;
2416 if (!ids || !flags || count == 0) {
2422 args.lfa_flags = flags;
2423 args.lfa_count = count;
2426 rc = llapi_layout_comp_iterate(layout,
2427 llapi_layout_apply_flags_cb,
2429 if (errno == ENOENT)
2432 if (rc != LLAPI_LAYOUT_ITER_CONT)
2438 * Change flags by component ID of components of an existing file.
2439 * The component to be modified is specified by the comp->lcme_id value,
2440 * which must be a unique component ID.
2442 * \param[in] path path name of the file
2443 * \param[in] ids An array of component IDs
2444 * \param[in] flags flags: LCME_FL_* or;
2445 * negative flags: (LCME_FL_NEG|LCME_FL_*)
2446 * \param[in] count Number of elements in ids and flags array
2448 int llapi_layout_file_comp_set(const char *path, uint32_t *ids, uint32_t *flags,
2451 int rc = -1, fd = -1, i, tmp_errno = 0;
2453 struct llapi_layout *existing_layout = NULL;
2454 struct llapi_layout *layout = NULL;
2455 struct llapi_layout_comp *comp;
2456 struct lov_user_md *lum = NULL;
2466 for (i = 0; i < count; i++) {
2467 if (!ids[i] || !flags[i]) {
2472 if (ids[i] > LCME_ID_MAX || (flags[i] & ~LCME_KNOWN_FLAGS)) {
2477 /* do not allow to set or clear INIT flag */
2478 if (flags[i] & LCME_FL_INIT) {
2484 fd = open(path, O_RDWR);
2491 existing_layout = llapi_layout_get_by_fd(fd, 0);
2492 if (existing_layout == NULL) {
2498 if (llapi_layout_apply_flags(existing_layout, ids, flags, count)) {
2504 rc = llapi_layout_sanity(existing_layout, false, false);
2507 llapi_layout_sanity_perror(rc);
2512 layout = __llapi_layout_alloc();
2513 if (layout == NULL) {
2519 layout->llot_is_composite = true;
2520 for (i = 0; i < count; i++) {
2521 comp = __llapi_comp_alloc(0);
2528 comp->llc_id = ids[i];
2529 comp->llc_flags = flags[i];
2531 list_add_tail(&comp->llc_list, &layout->llot_comp_list);
2532 layout->llot_cur_comp = comp;
2535 lum = llapi_layout_to_lum(layout);
2542 lum_size = ((struct lov_comp_md_v1 *)lum)->lcm_size;
2544 /* flush cached pages from clients */
2545 rc = llapi_file_flush(fd);
2552 rc = fsetxattr(fd, XATTR_LUSTRE_LOV".set.flags", lum, lum_size, 0);
2565 llapi_layout_free(existing_layout);
2566 llapi_layout_free(layout);
2572 * Check if the file layout is composite.
2574 * \param[in] layout the file layout to check
2576 * \retval true composite
2577 * \retval false not composite
2579 bool llapi_layout_is_composite(struct llapi_layout *layout)
2581 return layout->llot_is_composite;
2585 * Iterate every components in the @layout and call callback function @cb.
2587 * \param[in] layout component layout list.
2588 * \param[in] cb callback for each component
2589 * \param[in] cbdata callback data
2591 * \retval < 0 error happens during the iteration
2592 * \retval LLAPI_LAYOUT_ITER_CONT finished the iteration w/o error
2593 * \retval LLAPI_LAYOUT_ITER_STOP got something, stop the iteration
2595 int llapi_layout_comp_iterate(struct llapi_layout *layout,
2596 llapi_layout_iter_cb cb, void *cbdata)
2600 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_FIRST);
2605 * make sure on success llapi_layout_comp_use() API returns 0 with
2611 rc = cb(layout, cbdata);
2612 if (rc != LLAPI_LAYOUT_ITER_CONT)
2615 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_NEXT);
2618 else if (rc == 1) /* reached the last comp */
2619 return LLAPI_LAYOUT_ITER_CONT;
2626 * llapi_layout_merge() - Merge a composite layout into another one.
2627 * @dst_layout: Destination composite layout.
2628 * @src_layout: Source composite layout.
2630 * This function copies all of the components from @src_layout and
2631 * appends them to @dst_layout.
2633 * Return: 0 on success or -1 on failure.
2635 int llapi_layout_merge(struct llapi_layout **dst_layout,
2636 const struct llapi_layout *src_layout)
2638 struct llapi_layout *new_layout = *dst_layout;
2639 struct llapi_layout_comp *new = NULL;
2640 struct llapi_layout_comp *comp = NULL;
2643 if (src_layout == NULL ||
2644 list_empty((struct list_head *)&src_layout->llot_comp_list))
2647 if (new_layout == NULL) {
2648 new_layout = __llapi_layout_alloc();
2649 if (new_layout == NULL) {
2655 list_for_each_entry(comp, &src_layout->llot_comp_list, llc_list) {
2656 new = __llapi_comp_alloc(0);
2662 new->llc_pattern = comp->llc_pattern;
2663 new->llc_stripe_size = comp->llc_stripe_size;
2664 new->llc_stripe_count = comp->llc_stripe_count;
2665 new->llc_stripe_offset = comp->llc_stripe_offset;
2667 if (comp->llc_pool_name[0] != '\0')
2668 strncpy(new->llc_pool_name, comp->llc_pool_name,
2669 sizeof(new->llc_pool_name));
2671 for (i = 0; i < comp->llc_objects_count; i++) {
2672 if (__llapi_comp_objects_realloc(new,
2673 stripe_number_roundup(i)) < 0) {
2675 __llapi_comp_free(new);
2678 new->llc_objects[i].l_ost_idx = \
2679 comp->llc_objects[i].l_ost_idx;
2682 new->llc_objects_count = comp->llc_objects_count;
2683 new->llc_extent.e_start = comp->llc_extent.e_start;
2684 new->llc_extent.e_end = comp->llc_extent.e_end;
2685 new->llc_id = comp->llc_id;
2686 new->llc_flags = comp->llc_flags;
2688 list_add_tail(&new->llc_list, &new_layout->llot_comp_list);
2689 new_layout->llot_cur_comp = new;
2691 new_layout->llot_is_composite = true;
2693 *dst_layout = new_layout;
2696 llapi_layout_free(new_layout);
2701 * Find all stale components.
2703 * \param[in] layout component layout list.
2704 * \param[out] comp array of stale component info.
2705 * \param[in] comp_size array size of @comp.
2706 * \param[in] mirror_ids array of mirror id that only components
2707 * belonging to these mirror will be collected.
2708 * \param[in] ids_nr number of mirror ids array.
2710 * \retval number of component info collected on sucess or
2711 * an error code on failure.
2713 int llapi_mirror_find_stale(struct llapi_layout *layout,
2714 struct llapi_resync_comp *comp, size_t comp_size,
2715 __u16 *mirror_ids, int ids_nr)
2720 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_FIRST);
2728 uint64_t start, end;
2730 rc = llapi_layout_comp_flags_get(layout, &flags);
2734 if (!(flags & LCME_FL_STALE))
2737 rc = llapi_layout_mirror_id_get(layout, &mirror_id);
2741 /* the caller only wants stale components from specific
2746 for (j = 0; j < ids_nr; j++) {
2747 if (mirror_ids[j] == mirror_id)
2751 /* not in the specified mirror */
2754 } else if (flags & LCME_FL_NOSYNC) {
2755 /* if not specified mirrors, do not resync "nosync"
2760 rc = llapi_layout_comp_id_get(layout, &id);
2764 rc = llapi_layout_comp_extent_get(layout, &start, &end);
2768 /* pack this component into @comp array */
2769 comp[idx].lrc_id = id;
2770 comp[idx].lrc_mirror_id = mirror_id;
2771 comp[idx].lrc_start = start;
2772 comp[idx].lrc_end = end;
2775 if (idx >= comp_size) {
2781 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_NEXT);
2788 return rc < 0 ? rc : idx;
2791 /* locate @layout to a valid component covering file [file_start, file_end) */
2792 uint32_t llapi_mirror_find(struct llapi_layout *layout,
2793 uint64_t file_start, uint64_t file_end,
2796 uint32_t mirror_id = 0;
2799 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_FIRST);
2805 uint64_t start, end;
2806 uint32_t flags, id, rid;
2808 rc = llapi_layout_comp_flags_get(layout, &flags);
2812 if (flags & LCME_FL_STALE)
2815 rc = llapi_layout_mirror_id_get(layout, &rid);
2819 rc = llapi_layout_comp_id_get(layout, &id);
2823 rc = llapi_layout_comp_extent_get(layout, &start, &end);
2827 if (file_start >= start && file_start < end) {
2830 else if (mirror_id != rid || *endp != start)
2833 file_start = *endp = end;
2834 if (end >= file_end)
2839 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_NEXT);
2847 int llapi_mirror_resync_many(int fd, struct llapi_layout *layout,
2848 struct llapi_resync_comp *comp_array,
2849 int comp_size, uint64_t start, uint64_t end)
2852 size_t page_size = sysconf(_SC_PAGESIZE);
2853 const size_t buflen = 4 << 20; /* 4M */
2855 uint64_t pos = start;
2860 rc = posix_memalign(&buf, page_size, buflen);
2864 if (end == OBD_OBJECT_EOF)
2865 count = OBD_OBJECT_EOF;
2867 count = end - start;
2871 uint64_t mirror_end = 0;
2872 uint64_t bytes_left;
2877 src = llapi_mirror_find(layout, pos, end, &mirror_end);
2881 if (mirror_end == OBD_OBJECT_EOF) {
2884 bytes_left = MIN(count, mirror_end - pos);
2885 bytes_left = ((bytes_left - 1) | (page_size - 1)) + 1;
2887 to_read = MIN(buflen, bytes_left);
2889 bytes_read = llapi_mirror_read(fd, src, buf, to_read, pos);
2890 if (bytes_read == 0) {
2894 if (bytes_read < 0) {
2899 /* round up to page align to make direct IO happy. */
2900 to_write = ((bytes_read - 1) | (page_size - 1)) + 1;
2902 for (i = 0; i < comp_size; i++) {
2905 size_t to_write2 = to_write;
2907 /* skip non-overlapped component */
2908 if (pos >= comp_array[i].lrc_end ||
2909 pos + to_write <= comp_array[i].lrc_start)
2912 if (pos < comp_array[i].lrc_start)
2913 pos2 = comp_array[i].lrc_start;
2915 to_write2 -= pos2 - pos;
2917 if ((pos + to_write) > comp_array[i].lrc_end)
2918 to_write2 -= pos + to_write -
2919 comp_array[i].lrc_end;
2921 written = llapi_mirror_write(fd,
2922 comp_array[i].lrc_mirror_id,
2927 * this component is not written successfully,
2928 * mark it using its lrc_synced, it is supposed
2929 * to be false before getting here.
2931 * And before this function returns, all
2932 * elements of comp_array will reverse their
2933 * lrc_synced flag to reflect their true
2936 comp_array[i].lrc_synced = true;
2937 llapi_error(LLAPI_MSG_ERROR, written,
2938 "component %u not synced\n",
2939 comp_array[i].lrc_id);
2944 assert(written == to_write2);
2948 count -= bytes_read;
2954 /* fatal error happens */
2955 for (i = 0; i < comp_size; i++)
2956 comp_array[i].lrc_synced = false;
2961 * no fatal error happens, each lrc_synced tells whether the component
2962 * has been resync successfully (note: we'd reverse the value to
2963 * reflect its true meaning.
2965 for (i = 0; i < comp_size; i++) {
2966 comp_array[i].lrc_synced = !comp_array[i].lrc_synced;
2967 if (comp_array[i].lrc_synced && pos & (page_size - 1)) {
2968 rc = llapi_mirror_truncate(fd,
2969 comp_array[i].lrc_mirror_id, pos);
2971 comp_array[i].lrc_synced = false;
2976 * returns the first error code for partially successful resync if
2982 enum llapi_layout_comp_sanity_error {
2984 LSE_INCOMPLETE_MIRROR,
2985 LSE_ADJACENT_EXTENSION,
2989 LSE_DOM_EXTENSION_FOLLOWING,
2992 LSE_NOT_ZERO_LENGTH_EXTENDABLE,
2993 LSE_END_NOT_GREATER,
2994 LSE_ZERO_LENGTH_NORMAL,
2995 LSE_NOT_ADJACENT_PREV,
3002 const char *llapi_layout_strerror[] =
3005 [LSE_INCOMPLETE_MIRROR] =
3006 "Incomplete mirror - must go to EOF",
3007 [LSE_ADJACENT_EXTENSION] =
3008 "No adjacent extension space components",
3009 [LSE_INIT_EXTENSION] =
3010 "Cannot apply extension flag to init components",
3013 [LSE_DOM_EXTENSION] =
3014 "DoM components can't be extension space",
3015 [LSE_DOM_EXTENSION_FOLLOWING] =
3016 "DoM components cannot be followed by extension space",
3018 "DoM component should be the first one in a file/mirror",
3019 [LSE_SET_COMP_START] =
3020 "Must set previous component extent before adding next",
3021 [LSE_NOT_ZERO_LENGTH_EXTENDABLE] =
3022 "Extendable component must start out zero-length",
3023 [LSE_END_NOT_GREATER] =
3024 "Component end is before end of previous component",
3025 [LSE_ZERO_LENGTH_NORMAL] =
3026 "Zero length components must be followed by extension",
3027 [LSE_NOT_ADJACENT_PREV] =
3028 "Components not adjacent (end != next->start",
3029 [LSE_START_GT_END] =
3030 "Component start is > end",
3032 "The component end must be aligned by the stripe size",
3034 "The extension size must be aligned by the stripe size",
3037 struct llapi_layout_sanity_args {
3038 bool lsa_incomplete;
3044 static int llapi_layout_sanity_cb(struct llapi_layout *layout,
3047 struct llapi_layout_comp *comp, *next, *prev;
3048 struct llapi_layout_sanity_args *args = arg;
3049 bool first_comp = false;
3051 comp = __llapi_layout_cur_comp(layout);
3057 if (comp->llc_list.prev != &layout->llot_comp_list)
3058 prev = list_entry(comp->llc_list.prev, typeof(*prev),
3063 if (comp->llc_list.next != &layout->llot_comp_list)
3064 next = list_entry(comp->llc_list.next, typeof(*next),
3069 /* Start of zero implies a new mirror */
3070 if (comp->llc_extent.e_start == 0) {
3072 /* Most checks apply only within one mirror, this is an
3074 if (prev && prev->llc_extent.e_end != LUSTRE_EOF) {
3075 args->lsa_rc = LSE_INCOMPLETE_MIRROR;
3082 if (next && next->llc_extent.e_start == 0)
3085 /* Flag sanity checks */
3086 /* No adjacent extension components */
3087 if ((comp->llc_flags & LCME_FL_EXTENSION) && next &&
3088 (next->llc_flags & LCME_FL_EXTENSION)) {
3089 args->lsa_rc = LSE_ADJACENT_EXTENSION;
3093 /* Extension flag cannot be applied to init components and the first
3094 * component of each mirror is automatically init */
3095 if ((comp->llc_flags & LCME_FL_EXTENSION) &&
3096 (comp->llc_flags & LCME_FL_INIT || first_comp)) {
3097 args->lsa_rc = LSE_INIT_EXTENSION;
3101 if (comp->llc_ondisk) {
3102 if (comp->llc_flags & LCME_FL_NEG)
3103 args->lsa_rc = LSE_FLAGS;
3104 } else if (!args->lsa_incomplete) {
3105 if (args->lsa_flr) {
3106 if (comp->llc_flags & ~LCME_USER_COMP_FLAGS)
3107 args->lsa_rc = LSE_FLAGS;
3109 if (comp->llc_flags & ~LCME_FL_EXTENSION)
3110 args->lsa_rc = LSE_FLAGS;
3116 /* DoM sanity checks */
3117 if (comp->llc_pattern == LLAPI_LAYOUT_MDT ||
3118 comp->llc_pattern == LOV_PATTERN_MDT) {
3119 /* DoM components can't be extension components */
3120 if (comp->llc_flags & LCME_FL_EXTENSION) {
3121 args->lsa_rc = LSE_DOM_EXTENSION;
3124 /* DoM components cannot be followed by an extension comp */
3125 if (next && (next->llc_flags & LCME_FL_EXTENSION)) {
3126 args->lsa_rc = LSE_DOM_EXTENSION_FOLLOWING;
3130 /* DoM should be the first component in a mirror */
3132 args->lsa_rc = LSE_DOM_FIRST;
3138 /* Extent sanity checks */
3139 /* Must set previous component extent before adding another */
3140 if (prev && prev->llc_extent.e_start == 0 &&
3141 prev->llc_extent.e_end == 0) {
3142 args->lsa_rc = LSE_SET_COMP_START;
3146 if (!args->lsa_incomplete) {
3147 /* Components followed by extension space (extendable
3148 * components) must be zero length before initialization.
3149 * (Except for first comp, which will be initialized on
3151 if (next && (next->llc_flags & LCME_FL_EXTENSION) &&
3152 !first_comp && !(comp->llc_flags & LCME_FL_INIT) &&
3153 comp->llc_extent.e_start != comp->llc_extent.e_end) {
3154 args->lsa_rc = LSE_NOT_ZERO_LENGTH_EXTENDABLE;
3158 /* End must come after end of previous comp */
3159 if (prev && comp->llc_extent.e_end < prev->llc_extent.e_end) {
3160 args->lsa_rc = LSE_END_NOT_GREATER;
3164 /* Components not followed by ext space must have length > 0. */
3165 if (comp->llc_extent.e_start == comp->llc_extent.e_end &&
3166 (next == NULL || !(next->llc_flags & LCME_FL_EXTENSION))) {
3167 args->lsa_rc = LSE_ZERO_LENGTH_NORMAL;
3171 /* The component end must be aligned by the stripe size */
3172 if ((comp->llc_flags & LCME_FL_EXTENSION) &&
3173 (prev->llc_stripe_size != LLAPI_LAYOUT_DEFAULT)) {
3174 if (comp->llc_extent.e_end != LUSTRE_EOF &&
3175 comp->llc_extent.e_end % prev->llc_stripe_size) {
3176 args->lsa_rc = LSE_ALIGN_END;
3179 if ((comp->llc_stripe_size * SEL_UNIT_SIZE) %
3180 prev->llc_stripe_size) {
3181 args->lsa_rc = LSE_ALIGN_EXT;
3184 } else if (!(comp->llc_flags & LCME_FL_EXTENSION) &&
3185 (comp->llc_stripe_size != LLAPI_LAYOUT_DEFAULT)) {
3186 if (comp->llc_extent.e_end != LUSTRE_EOF &&
3187 comp->llc_extent.e_end % comp->llc_stripe_size) {
3188 args->lsa_rc = LSE_ALIGN_END;
3194 /* Components must have start == prev->end */
3195 if (prev && comp->llc_extent.e_start != 0 &&
3196 comp->llc_extent.e_start != prev->llc_extent.e_end) {
3197 args->lsa_rc = LSE_NOT_ADJACENT_PREV;
3201 /* Components must have start <= end */
3202 if (comp->llc_extent.e_start > comp->llc_extent.e_end) {
3203 args->lsa_rc = LSE_START_GT_END;
3207 return LLAPI_LAYOUT_ITER_CONT;
3210 errno = errno ? errno : EINVAL;
3211 return LLAPI_LAYOUT_ITER_STOP;
3214 /* Print explanation of layout error */
3215 void llapi_layout_sanity_perror(int error)
3217 if (error >= LSE_LAST || error < 0) {
3218 fprintf(stdout, "Invalid layout, unrecognized error: %d\n",
3221 fprintf(stdout, "Invalid layout: %s\n",
3222 llapi_layout_strerror[error]);
3226 /* Walk a layout and enforce sanity checks that apply to > 1 component
3228 * The core idea here is that of sanity checking individual tokens vs semantic
3230 * We cannot check everything at the individual component level ('token'),
3231 * instead we must check whether or not the full layout has a valid meaning.
3233 * An example of a component level check is "is stripe size valid?". That is
3234 * handled when setting stripe size.
3236 * An example of a layout level check is "are the extents of these components
3237 * valid when adjacent to one another", or "can we set these flags on adjacent
3240 * \param[in] layout component layout list.
3241 * \param[in] incomplete if layout is complete or not - some checks can
3242 * only be done on complete layouts.
3243 * \param[in] flr set when this is called from FLR mirror create
3245 * \retval 0, success, positive: various errors, see
3246 * llapi_layout_sanity_perror, -1, failure
3248 int llapi_layout_sanity(struct llapi_layout *layout, bool incomplete, bool flr)
3250 struct llapi_layout_sanity_args args;
3251 struct llapi_layout_comp *curr;
3257 curr = layout->llot_cur_comp;
3264 args.lsa_incomplete = incomplete;
3266 /* When we modify an existing layout, this tells us if it's FLR */
3267 if (mirror_id_of(curr->llc_id) > 0)
3268 args.lsa_flr = true;
3271 rc = llapi_layout_comp_iterate(layout,
3272 llapi_layout_sanity_cb,
3274 if (errno == ENOENT)
3277 if (rc != LLAPI_LAYOUT_ITER_CONT)
3280 layout->llot_cur_comp = curr;
3285 int llapi_layout_dom_size(struct llapi_layout *layout, uint64_t *size)
3287 uint64_t pattern, start;
3290 if (!layout || !llapi_layout_is_composite(layout)) {
3295 rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_FIRST);
3299 rc = llapi_layout_pattern_get(layout, &pattern);
3303 if (pattern != LOV_PATTERN_MDT && pattern != LLAPI_LAYOUT_MDT) {
3308 rc = llapi_layout_comp_extent_get(layout, &start, size);