X-Git-Url: http://git.ipfire.org/?p=thirdparty%2Fmdadm.git;a=blobdiff_plain;f=super-ddf.c;h=ad21e74a3c82fee0ded5074d1b737ac921b330db;hp=b494647e04dab00f9c28dccf094010a6f2076b40;hb=ff54de6e47163944185f231700e72d3122b58f4c;hpb=598f0d58ac3544e25f90f2d34337e15764f4d877 diff --git a/super-ddf.c b/super-ddf.c index b494647e..ad21e74a 100644 --- a/super-ddf.c +++ b/super-ddf.c @@ -27,14 +27,10 @@ #define HAVE_STDINT_H 1 #include "mdadm.h" +#include "mdmon.h" #include "sha1.h" #include -static inline int ROUND_UP(int a, int base) -{ - return ((a+base-1)/base)*base; -} - /* a non-official T10 name for creation GUIDs */ static char T10[] = "Linux-MD"; @@ -70,7 +66,7 @@ unsigned long crc32( #define DDF_CONCAT 0x1f #define DDF_RAID5E 0x15 #define DDF_RAID5EE 0x25 -#define DDF_RAID6 0x16 /* Vendor unique layout */ +#define DDF_RAID6 0x06 /* Raid Level Qualifier (RLQ) */ #define DDF_RAID0_SIMPLE 0x00 @@ -82,6 +78,7 @@ unsigned long crc32( #define DDF_RAID4_N 0x01 /* parity in last extent */ /* these apply to raid5e and raid5ee as well */ #define DDF_RAID5_0_RESTART 0x00 /* same as 'right asymmetric' - layout 1 */ +#define DDF_RAID6_0_RESTART 0x01 /* raid6 different from raid5 here!!! */ #define DDF_RAID5_N_RESTART 0x02 /* same as 'left asymmetric' - layout 0 */ #define DDF_RAID5_N_CONTINUE 0x03 /* same as 'left symmetric' - layout 2 */ @@ -107,13 +104,14 @@ unsigned long crc32( #define DDF_BBM_LOG_MAGIC __cpu_to_be32(0xABADB10C) #define DDF_GUID_LEN 24 -#define DDF_REVISION "01.00.00" +#define DDF_REVISION_0 "01.00.00" +#define DDF_REVISION_2 "01.02.00" struct ddf_header { - __u32 magic; + __u32 magic; /* DDF_HEADER_MAGIC */ __u32 crc; char guid[DDF_GUID_LEN]; - char revision[8]; /* 01.00.00 */ + char revision[8]; /* 01.02.00 */ __u32 seq; /* starts at '1' */ __u32 timestamp; __u8 openflag; @@ -166,7 +164,7 @@ struct ddf_header { /* The content of the 'controller section' - global scope */ struct ddf_controller_data { - __u32 magic; + __u32 magic; /* DDF_CONTROLLER_MAGIC */ __u32 crc; char guid[DDF_GUID_LEN]; struct controller_type { @@ -182,7 +180,7 @@ struct ddf_controller_data { /* The content of phys_section - global scope */ struct phys_disk { - __u32 magic; + __u32 magic; /* DDF_PHYS_RECORDS_MAGIC */ __u32 crc; __u16 used_pdes; __u16 max_pdes; @@ -201,7 +199,7 @@ struct phys_disk { /* phys_disk_entry.type is a bitmap - bigendian remember */ #define DDF_Forced_PD_GUID 1 #define DDF_Active_in_VD 2 -#define DDF_Global_Spare 4 +#define DDF_Global_Spare 4 /* VD_CONF records are ignored */ #define DDF_Spare 8 /* overrides Global_spare */ #define DDF_Foreign 16 #define DDF_Legacy 32 /* no DDF on this device */ @@ -223,7 +221,7 @@ struct phys_disk { /* The content of the virt_section global scope */ struct virtual_disk { - __u32 magic; + __u32 magic; /* DDF_VIRT_RECORDS_MAGIC */ __u32 crc; __u16 populated_vdes; __u16 max_vdes; @@ -254,6 +252,7 @@ struct virtual_disk { #define DDF_state_deleted 0x2 #define DDF_state_missing 0x3 #define DDF_state_failed 0x4 +#define DDF_state_part_optimal 0x5 #define DDF_state_morphing 0x8 #define DDF_state_inconsistent 0x10 @@ -261,7 +260,8 @@ struct virtual_disk { /* virtual_entry.init_state is a bigendian bitmap */ #define DDF_initstate_mask 0x03 #define DDF_init_not 0x00 -#define DDF_init_quick 0x01 +#define DDF_init_quick 0x01 /* initialisation is progress. + * i.e. 'state_inconsistent' */ #define DDF_init_full 0x02 #define DDF_access_mask 0xc0 @@ -275,7 +275,7 @@ struct virtual_disk { */ struct vd_config { - __u32 magic; + __u32 magic; /* DDF_VD_CONF_MAGIC */ __u32 crc; char guid[DDF_GUID_LEN]; __u32 timestamp; @@ -319,7 +319,7 @@ struct vd_config { #define DDF_cache_rallowed 64 /* enable read caching */ struct spare_assign { - __u32 magic; + __u32 magic; /* DDF_SPARE_ASSIGN_MAGIC */ __u32 crc; __u32 timestamp; __u8 reserved[7]; @@ -341,7 +341,7 @@ struct spare_assign { /* The data_section contents - local scope */ struct disk_data { - __u32 magic; + __u32 magic; /* DDF_PHYS_DATA_MAGIC */ __u32 crc; char guid[DDF_GUID_LEN]; __u32 refnum; /* crc of some magic drive data ... */ @@ -390,33 +390,51 @@ struct bad_block_log { * built in Create or Assemble to describe the whole array. */ struct ddf_super { - struct ddf_header anchor, primary, secondary, *active; + struct ddf_header anchor, primary, secondary; struct ddf_controller_data controller; + struct ddf_header *active; struct phys_disk *phys; struct virtual_disk *virt; int pdsize, vdsize; - int max_part; + int max_part, mppe, conf_rec_len; + int currentdev; + int updates_pending; struct vcl { - struct vcl *next; - __u64 *lba_offset; /* location in 'conf' of - * the lba table */ + union { + char space[512]; + struct { + struct vcl *next; + __u64 *lba_offset; /* location in 'conf' of + * the lba table */ + int vcnum; /* index into ->virt */ + __u64 *block_sizes; /* NULL if all the same */ + }; + }; struct vd_config conf; - } *conflist, *newconf; + } *conflist, *currentconf; struct dl { - struct dl *next; + union { + char space[512]; + struct { + struct dl *next; + int major, minor; + char *devname; + int fd; + unsigned long long size; /* sectors */ + int pdnum; /* index in ->phys */ + struct spare_assign *spare; + }; + }; struct disk_data disk; - int major, minor; - char *devname; - int fd; - struct vcl *vlist[0]; /* max_part+1 in size */ - } *dlist; + void *mdupdate; /* hold metadata update */ + struct vcl *vlist[0]; /* max_part in size */ + } *dlist, *add_list; }; #ifndef offsetof #define offsetof(t,f) ((size_t)&(((t*)0)->f)) #endif -struct superswitch super_ddf_container, super_ddf_bvd; static int calc_crc(void *buf, int len) { @@ -486,8 +504,10 @@ static void *load_section(int fd, struct ddf_super *super, void *buf, /* All pre-allocated sections are a single block */ if (len != 1) return NULL; - } else - buf = malloc(len<<9); + } else { + posix_memalign(&buf, 512, len<<9); + } + if (!buf) return NULL; @@ -541,11 +561,12 @@ static int load_ddf_headers(int fd, struct ddf_super *super, char *devname) devname); return 2; } - if (memcmp(super->anchor.revision, DDF_REVISION, 8) != 0) { + if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 && + memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) { if (devname) fprintf(stderr, Name ": can only support super revision" - " %.8s, not %.8s on %s\n", - DDF_REVISION, super->anchor.revision, devname); + " %.8s and earlier, not %.8s on %s\n", + DDF_REVISION_2, super->anchor.revision,devname); return 2; } if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba), @@ -596,10 +617,16 @@ static int load_ddf_global(int fd, struct ddf_super *super, char *devname) !super->virt) { free(super->phys); free(super->virt); + super->phys = NULL; + super->virt = NULL; return 2; } super->conflist = NULL; super->dlist = NULL; + + super->max_part = __be16_to_cpu(super->active->max_partitions); + super->mppe = __be16_to_cpu(super->active->max_primary_element_entries); + super->conf_rec_len = __be16_to_cpu(super->active->config_record_len); return 0; } @@ -610,13 +637,14 @@ static int load_ddf_local(int fd, struct ddf_super *super, struct stat stb; char *conf; int i; - int conflen; - int mppe; + int vnum; + int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries); + unsigned long long dsize; /* First the local disk info */ - super->max_part = __be16_to_cpu(super->active->max_partitions); - dl = malloc(sizeof(*dl) + - (super->max_part+1) * sizeof(dl->vlist[0])); + posix_memalign((void**)&dl, 512, + sizeof(*dl) + + (super->max_part) * sizeof(dl->vlist[0])); load_section(fd, super, &dl->disk, super->active->data_section_offset, @@ -629,29 +657,47 @@ static int load_ddf_local(int fd, struct ddf_super *super, dl->minor = minor(stb.st_rdev); dl->next = super->dlist; dl->fd = keep ? fd : -1; - for (i=0 ; i < super->max_part + 1 ; i++) + + dl->size = 0; + if (get_dev_size(fd, devname, &dsize)) + dl->size = dsize >> 9; + dl->spare = NULL; + for (i=0 ; i < super->max_part ; i++) dl->vlist[i] = NULL; super->dlist = dl; + dl->pdnum = -1; + for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++) + if (memcmp(super->phys->entries[i].guid, + dl->disk.guid, DDF_GUID_LEN) == 0) + dl->pdnum = i; /* Now the config list. */ /* 'conf' is an array of config entries, some of which are * probably invalid. Those which are good need to be copied into * the conflist */ - conflen = __be16_to_cpu(super->active->config_record_len); conf = load_section(fd, super, NULL, super->active->config_section_offset, super->active->config_section_length, 0); + vnum = 0; for (i = 0; i < __be32_to_cpu(super->active->config_section_length); - i += conflen) { + i += super->conf_rec_len) { struct vd_config *vd = (struct vd_config *)((char*)conf + i*512); struct vcl *vcl; + if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) { + if (dl->spare) + continue; + posix_memalign((void**)&dl->spare, 512, + super->conf_rec_len*512); + memcpy(dl->spare, vd, super->conf_rec_len*512); + continue; + } if (vd->magic != DDF_VD_CONF_MAGIC) continue; for (vcl = super->conflist; vcl; vcl = vcl->next) { @@ -661,20 +707,29 @@ static int load_ddf_local(int fd, struct ddf_super *super, } if (vcl) { - dl->vlist[i/conflen] = vcl; + dl->vlist[vnum++] = vcl; if (__be32_to_cpu(vd->seqnum) <= __be32_to_cpu(vcl->conf.seqnum)) continue; - } else { - vcl = malloc(conflen*512 + offsetof(struct vcl, conf)); + } else { + posix_memalign((void**)&vcl, 512, + (super->conf_rec_len*512 + + offsetof(struct vcl, conf))); vcl->next = super->conflist; + vcl->block_sizes = NULL; /* FIXME not for CONCAT */ super->conflist = vcl; + dl->vlist[vnum++] = vcl; } - memcpy(&vcl->conf, vd, conflen*512); - mppe = __be16_to_cpu(super->anchor.max_primary_element_entries); + memcpy(&vcl->conf, vd, super->conf_rec_len*512); vcl->lba_offset = (__u64*) - &vcl->conf.phys_refnum[mppe]; - dl->vlist[i/conflen] = vcl; + &vcl->conf.phys_refnum[super->mppe]; + + for (i=0; i < max_virt_disks ; i++) + if (memcmp(super->virt->entries[i].guid, + vcl->conf.guid, DDF_GUID_LEN)==0) + break; + if (i < max_virt_disks) + vcl->vcnum = i; } free(conf); @@ -693,9 +748,12 @@ static int load_super_ddf(struct supertype *st, int fd, int rv; #ifndef MDASSEMBLE - if (load_super_ddf_all(st, fd, &st->sb, devname, 0) == 0) + /* if 'fd' is a container, load metadata from all the devices */ + if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0) return 0; #endif + if (st->subarray[0]) + return 1; /* FIXME Is this correct */ if (get_dev_size(fd, devname, &dsize) == 0) return 1; @@ -720,12 +778,12 @@ static int load_super_ddf(struct supertype *st, int fd, } } - super = malloc(sizeof(*super)); - if (!super) { + if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) { fprintf(stderr, Name ": malloc of %zu failed.\n", sizeof(*super)); return 1; } + memset(super, 0, sizeof(*super)); rv = load_ddf_headers(fd, super, devname); if (rv) { @@ -756,6 +814,7 @@ static int load_super_ddf(struct supertype *st, int fd, st->minor_version = 0; st->max_devs = 512; } + st->loaded_container = 0; return 0; } @@ -770,6 +829,8 @@ static void free_super_ddf(struct supertype *st) while (ddf->conflist) { struct vcl *v = ddf->conflist; ddf->conflist = v->next; + if (v->block_sizes) + free(v->block_sizes); free(v); } while (ddf->dlist) { @@ -777,6 +838,8 @@ static void free_super_ddf(struct supertype *st) ddf->dlist = d->next; if (d->fd >= 0) close(d->fd); + if (d->spare) + free(d->spare); free(d); } free(ddf); @@ -793,6 +856,7 @@ static struct supertype *match_metadata_desc_ddf(char *arg) return NULL; st = malloc(sizeof(*st)); + memset(st, 0, sizeof(*st)); st->ss = &super_ddf; st->max_devs = 512; st->minor_version = 0; @@ -800,38 +864,6 @@ static struct supertype *match_metadata_desc_ddf(char *arg) return st; } -static struct supertype *match_metadata_desc_ddf_bvd(char *arg) -{ - struct supertype *st; - if (strcmp(arg, "ddf/bvd") != 0 && - strcmp(arg, "bvd") != 0 && - strcmp(arg, "default") != 0 - ) - return NULL; - - st = malloc(sizeof(*st)); - st->ss = &super_ddf_bvd; - st->max_devs = 512; - st->minor_version = 0; - st->sb = NULL; - return st; -} -static struct supertype *match_metadata_desc_ddf_svd(char *arg) -{ - struct supertype *st; - if (strcmp(arg, "ddf/svd") != 0 && - strcmp(arg, "svd") != 0 && - strcmp(arg, "default") != 0 - ) - return NULL; - - st = malloc(sizeof(*st)); - st->ss = &super_ddf_svd; - st->max_devs = 512; - st->minor_version = 0; - st->sb = NULL; - return st; -} #ifndef MDASSEMBLE @@ -892,7 +924,8 @@ static struct num_mapping ddf_level_num[] = { { DDF_RAID0, 0 }, { DDF_RAID1, 1 }, { DDF_RAID3, LEVEL_UNSUPPORTED }, - { DDF_RAID5, 4 }, + { DDF_RAID4, 4 }, + { DDF_RAID5, 5 }, { DDF_RAID1E, LEVEL_UNSUPPORTED }, { DDF_JBOD, LEVEL_UNSUPPORTED }, { DDF_CONCAT, LEVEL_LINEAR }, @@ -916,34 +949,41 @@ static void print_guid(char *guid, int tstamp) { /* A GUIDs are part (or all) ASCII and part binary. * They tend to be space padded. - * We ignore trailing spaces and print numbers - * <0x20 and >=0x7f as \xXX - * Some GUIDs have a time stamp in bytes 16-19. - * We print that if appropriate + * We print the GUID in HEX, then in parentheses add + * any initial ASCII sequence, and a possible + * time stamp from bytes 16-19 */ int l = DDF_GUID_LEN; int i; + + for (i=0 ; i= 0x20 && guid[i] < 0x7f) fputc(guid[i], stdout); else - fprintf(stdout, "\\x%02x", guid[i]&255); + break; } if (tstamp) { time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE; char tbuf[100]; struct tm *tm; tm = localtime(&then); - strftime(tbuf, 100, " (%D %T)",tm); + strftime(tbuf, 100, " %D %T",tm); fputs(tbuf, stdout); } + printf(")"); } static void examine_vd(int n, struct ddf_super *sb, char *guid) { - int crl = __be16_to_cpu(sb->anchor.config_record_len); + int crl = sb->conf_rec_len; struct vcl *vcl; for (vcl = sb->conflist ; vcl ; vcl = vcl->next) { @@ -1060,20 +1100,18 @@ static void examine_super_ddf(struct supertype *st, char *homehost) examine_pds(sb); } +static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info); + + static void brief_examine_super_ddf(struct supertype *st) { /* We just write a generic DDF ARRAY entry - * The uuid is all hex, 6 groups of 4 bytes */ - struct ddf_super *ddf = st->sb; - int i; - printf("ARRAY /dev/ddf UUID="); - for (i = 0; i < DDF_GUID_LEN; i++) { - printf("%02x", ddf->anchor.guid[i]); - if ((i&3) == 0 && i != 0) - printf(":"); - } - printf("\n"); + struct mdinfo info; + char nbuf[64]; + getinfo_super_ddf(st, &info); + fname_from_uuid(st, &info, nbuf, ':'); + printf("ARRAY /dev/ddf metadata=ddf UUID=%s\n", nbuf + 5); } static void detail_super_ddf(struct supertype *st, char *homehost) @@ -1092,9 +1130,12 @@ static void brief_detail_super_ddf(struct supertype *st) * Can that be stored in ddf_super?? */ // struct ddf_super *ddf = st->sb; + struct mdinfo info; + char nbuf[64]; + getinfo_super_ddf(st, &info); + fname_from_uuid(st, &info, nbuf,':'); + printf(" UUID=%s", nbuf + 5); } - - #endif static int match_home_ddf(struct supertype *st, char *homehost) @@ -1112,10 +1153,28 @@ static int match_home_ddf(struct supertype *st, char *homehost) ddf->controller.vendor_data[len] == 0); } -static struct vd_config *find_vdcr(struct ddf_super *ddf) +#ifndef MDASSEMBLE +static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst) +{ + struct vcl *v; + + for (v = ddf->conflist; v; v = v->next) + if (inst == v->vcnum) + return &v->conf; + return NULL; +} +#endif + +static int find_phys(struct ddf_super *ddf, __u32 phys_refnum) { - /* FIXME this just picks off the first one */ - return &ddf->conflist->conf; + /* Find the entry in phys_disk which has the given refnum + * and return it's index + */ + int i; + for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++) + if (ddf->phys->entries[i].refnum == phys_refnum) + return i; + return -1; } static void uuid_from_super_ddf(struct supertype *st, int uuid[4]) @@ -1124,11 +1183,14 @@ static void uuid_from_super_ddf(struct supertype *st, int uuid[4]) * uuid to put into bitmap file (Create, Grow) * uuid for backup header when saving critical section (Grow) * comparing uuids when re-adding a device into an array + * In these cases the uuid required is that of the data-array, + * not the device-set. + * uuid to recognise same set when adding a missing device back + * to an array. This is a uuid for the device-set. + * * For each of these we can make do with a truncated * or hashed uuid rather than the original, as long as * everyone agrees. - * In each case the uuid required is that of the data-array, - * not the device-set. * In the case of SVD we assume the BVD is of interest, * though that might be the case if a bitmap were made for * a mirrored SVD - worry about that later. @@ -1137,30 +1199,35 @@ static void uuid_from_super_ddf(struct supertype *st, int uuid[4]) * The first 16 bytes of the sha1 of these is used. */ struct ddf_super *ddf = st->sb; - struct vd_config *vd = find_vdcr(ddf); - - if (!vd) - memset(uuid, 0, sizeof (uuid)); - else { - char buf[20]; - struct sha1_ctx ctx; - sha1_init_ctx(&ctx); - sha1_process_bytes(&vd->guid, DDF_GUID_LEN, &ctx); - if (vd->sec_elmnt_count > 1) - sha1_process_bytes(&vd->sec_elmnt_seq, 1, &ctx); - sha1_finish_ctx(&ctx, buf); - memcpy(uuid, buf, sizeof(uuid)); - } + struct vcl *vcl = ddf->currentconf; + char *guid; + char buf[20]; + struct sha1_ctx ctx; + + if (vcl) + guid = vcl->conf.guid; + else + guid = ddf->anchor.guid; + + sha1_init_ctx(&ctx); + sha1_process_bytes(guid, DDF_GUID_LEN, &ctx); + if (vcl && vcl->conf.sec_elmnt_count > 1) + sha1_process_bytes(&vcl->conf.sec_elmnt_seq, 1, &ctx); + sha1_finish_ctx(&ctx, buf); + memcpy(uuid, buf, 4*4); } +static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info); + static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info) { struct ddf_super *ddf = st->sb; - int i; - info->array.major_version = 1000; - info->array.minor_version = 0; /* FIXME use ddf->revision somehow */ - info->array.patch_version = 0; + if (ddf->currentconf) { + getinfo_super_ddf_bvd(st, info); + return; + } + info->array.raid_disks = __be16_to_cpu(ddf->phys->used_pdes); info->array.level = LEVEL_CONTAINER; info->array.layout = 0; @@ -1170,36 +1237,34 @@ static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info) info->array.utime = 0; info->array.chunk_size = 0; -// info->data_offset = ???; -// info->component_size = ???; info->disk.major = 0; info->disk.minor = 0; - info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum); -// info->disk.raid_disk = find refnum in the table and use index; - info->disk.raid_disk = -1; - for (i = 0; i < __be16_to_cpu(ddf->phys->max_pdes) ; i++) - if (ddf->phys->entries[i].refnum == ddf->dlist->disk.refnum) { - info->disk.raid_disk = i; - break; - } + if (ddf->dlist) { + info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum); + info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum); + + info->data_offset = __be64_to_cpu(ddf->phys-> + entries[info->disk.raid_disk]. + config_size); + info->component_size = ddf->dlist->size - info->data_offset; + } else { + info->disk.number = -1; +// info->disk.raid_disk = find refnum in the table and use index; + } info->disk.state = (1 << MD_DISK_SYNC); - info->reshape_active = 0; -// uuid_from_super_ddf(info->uuid, sbv); + info->reshape_active = 0; + info->name[0] = 0; -// info->name[] ?? ; -} + info->array.major_version = -1; + info->array.minor_version = -2; + strcpy(info->text_version, "ddf"); + info->safe_mode_delay = 0; -static void getinfo_super_n_container(struct supertype *st, struct mdinfo *info) -{ - /* just need offset and size */ - struct ddf_super *ddf = st->sb; - int n = info->disk.number; + uuid_from_super_ddf(st, info->uuid); - info->data_offset = __be64_to_cpu(ddf->phys->entries[n].config_size); - info->component_size = 32*1024*1024 / 512; } static int rlq_to_layout(int rlq, int prl, int raiddisks); @@ -1207,24 +1272,28 @@ static int rlq_to_layout(int rlq, int prl, int raiddisks); static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info) { struct ddf_super *ddf = st->sb; - struct vd_config *vd = find_vdcr(ddf); + struct vcl *vc = ddf->currentconf; + int cd = ddf->currentdev; /* FIXME this returns BVD info - what if we want SVD ?? */ - info->array.major_version = 1000; - info->array.minor_version = 0; /* FIXME use ddf->revision somehow */ - info->array.patch_version = 0; - info->array.raid_disks = __be16_to_cpu(vd->prim_elmnt_count); - info->array.level = map_num1(ddf_level_num, vd->prl); - info->array.layout = rlq_to_layout(vd->rlq, vd->prl, + info->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count); + info->array.level = map_num1(ddf_level_num, vc->conf.prl); + info->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl, info->array.raid_disks); info->array.md_minor = -1; - info->array.ctime = DECADE + __be32_to_cpu(*(__u32*)(vd->guid+16)); - info->array.utime = DECADE + __be32_to_cpu(vd->timestamp); - info->array.chunk_size = 512 << vd->chunk_shift; - -// info->data_offset = ???; -// info->component_size = ???; + info->array.ctime = DECADE + + __be32_to_cpu(*(__u32*)(vc->conf.guid+16)); + info->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp); + info->array.chunk_size = 512 << vc->conf.chunk_shift; + + if (cd >= 0 && cd < ddf->mppe) { + info->data_offset = __be64_to_cpu(vc->lba_offset[cd]); + if (vc->block_sizes) + info->component_size = vc->block_sizes[cd]; + else + info->component_size = __be64_to_cpu(vc->conf.blocks); + } info->disk.major = 0; info->disk.minor = 0; @@ -1232,23 +1301,30 @@ static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info) // info->disk.raid_disk = find refnum in the table and use index; // info->disk.state = ???; + info->container_member = ddf->currentconf->vcnum; + + info->resync_start = 0; + if (!(ddf->virt->entries[info->container_member].state + & DDF_state_inconsistent) && + (ddf->virt->entries[info->container_member].init_state + & DDF_initstate_mask) + == DDF_init_full) + info->resync_start = ~0ULL; + uuid_from_super_ddf(st, info->uuid); -// info->name[] ?? ; -} + info->container_member = atoi(st->subarray); + info->array.major_version = -1; + info->array.minor_version = -2; + sprintf(info->text_version, "/%s/%s", + devnum2devname(st->container_dev), + st->subarray); + info->safe_mode_delay = 200; -static void getinfo_super_n_bvd(struct supertype *st, struct mdinfo *info) -{ - /* Find the particular details for info->disk.raid_disk. - * This includes data_offset, component_size, - */ - struct ddf_super *ddf = st->sb; - __u64 *lba_offset = ddf->newconf->lba_offset; - struct vd_config *conf = &ddf->newconf->conf; - info->data_offset = __be64_to_cpu(lba_offset[info->disk.raid_disk]); - info->component_size = __be64_to_cpu(conf->blocks); + info->name[0] = 0; } + static int update_super_ddf(struct supertype *st, struct mdinfo *info, char *update, char *devname, int verbose, @@ -1268,7 +1344,7 @@ static int update_super_ddf(struct supertype *st, struct mdinfo *info, * grow: Array has gained a new device - this is currently for * linear only * resync: mark as dirty so a resync will happen. - * uuid: Change the uuid of the array to match watch is given + * uuid: Change the uuid of the array to match what is given * homehost: update the recorded homehost * name: update the name - preserving the homehost * _reshape_progress: record new reshape_progress position. @@ -1280,10 +1356,9 @@ static int update_super_ddf(struct supertype *st, struct mdinfo *info, */ int rv = 0; // struct ddf_super *ddf = st->sb; -// struct vd_config *vd = find_vdcr(ddf); +// struct vd_config *vd = find_vdcr(ddf, info->container_member); // struct virtual_entry *ve = find_ve(ddf); - /* we don't need to handle "force-*" or "assemble" as * there is no need to 'trick' the kernel. We the metadata is * first updated to activate the array, all the implied modifications @@ -1344,6 +1419,13 @@ static void make_header_guid(char *guid) memcpy(guid+20, &stamp, 4); if (rfd >= 0) close(rfd); } + +static int init_super_ddf_bvd(struct supertype *st, + mdu_array_info_t *info, + unsigned long long size, + char *name, char *homehost, + int *uuid); + static int init_super_ddf(struct supertype *st, mdu_array_info_t *info, unsigned long long size, char *name, char *homehost, @@ -1383,7 +1465,16 @@ static int init_super_ddf(struct supertype *st, struct phys_disk *pd; struct virtual_disk *vd; - ddf = malloc(sizeof(*ddf)); + if (!info) { + st->sb = NULL; + return 0; + } + if (st->sb) + return init_super_ddf_bvd(st, info, size, name, homehost, + uuid); + + posix_memalign((void**)&ddf, 512, sizeof(*ddf)); + memset(ddf, 0, sizeof(*ddf)); ddf->dlist = NULL; /* no physical disks yet */ ddf->conflist = NULL; /* No virtual disks yet */ @@ -1398,7 +1489,7 @@ static int init_super_ddf(struct supertype *st, ddf->anchor.magic = DDF_HEADER_MAGIC; make_header_guid(ddf->anchor.guid); - memcpy(ddf->anchor.revision, DDF_REVISION, 8); + memcpy(ddf->anchor.revision, DDF_REVISION_2, 8); ddf->anchor.seq = __cpu_to_be32(1); ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE); ddf->anchor.openflag = 0xFF; @@ -1420,10 +1511,11 @@ static int init_super_ddf(struct supertype *st, ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */ ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */ ddf->max_part = 64; - ddf->anchor.config_record_len = __cpu_to_be16(1 + 256*12/512); - ddf->anchor.max_primary_element_entries = __cpu_to_be16(256); + ddf->mppe = 256; + ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512; + ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len); + ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe); memset(ddf->anchor.pad3, 0xff, 54); - /* controller sections is one sector long immediately * after the ddf header */ sector = 1; @@ -1457,7 +1549,7 @@ static int init_super_ddf(struct supertype *st, __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */ sector += vdsize/512; - clen = (1 + 256*12/512) * (64+1); + clen = ddf->conf_rec_len * (ddf->max_part+1); ddf->anchor.config_section_offset = __cpu_to_be32(sector); ddf->anchor.config_section_length = __cpu_to_be32(clen); sector += clen; @@ -1493,8 +1585,8 @@ static int init_super_ddf(struct supertype *st, * Remaining 16 are serial number.... maybe a hostname would do? */ memcpy(ddf->controller.guid, T10, sizeof(T10)); - gethostname(hostname, 17); - hostname[17] = 0; + gethostname(hostname, sizeof(hostname)); + hostname[sizeof(hostname) - 1] = 0; hostlen = strlen(hostname); memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen); for (i = strlen(T10) ; i+hostlen < 24; i++) @@ -1508,7 +1600,8 @@ static int init_super_ddf(struct supertype *st, memset(ddf->controller.pad, 0xff, 8); memset(ddf->controller.vendor_data, 0xff, 448); - pd = ddf->phys = malloc(pdsize); + posix_memalign((void**)&pd, 512, pdsize); + ddf->phys = pd; ddf->pdsize = pdsize; memset(pd, 0xff, pdsize); @@ -1518,7 +1611,8 @@ static int init_super_ddf(struct supertype *st, pd->max_pdes = __cpu_to_be16(max_phys_disks); memset(pd->pad, 0xff, 52); - vd = ddf->virt = malloc(vdsize); + posix_memalign((void**)&vd, 512, vdsize); + ddf->virt = vd; ddf->vdsize = vdsize; memset(vd, 0, vdsize); vd->magic = DDF_VIRT_RECORDS_MAGIC; @@ -1530,6 +1624,7 @@ static int init_super_ddf(struct supertype *st, memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry)); st->sb = ddf; + ddf->updates_pending = 1; return 1; } @@ -1580,7 +1675,10 @@ static int layout_to_rlq(int level, int layout, int raiddisks) case ALGORITHM_LEFT_ASYMMETRIC: return DDF_RAID5_N_RESTART; case ALGORITHM_RIGHT_ASYMMETRIC: - return DDF_RAID5_0_RESTART; + if (level == 5) + return DDF_RAID5_0_RESTART; + else + return DDF_RAID6_0_RESTART; case ALGORITHM_LEFT_SYMMETRIC: return DDF_RAID5_N_CONTINUE; case ALGORITHM_RIGHT_SYMMETRIC: @@ -1607,7 +1705,6 @@ static int rlq_to_layout(int rlq, int prl, int raiddisks) return -1; /* FIXME this isn't checked */ } case DDF_RAID5: - case DDF_RAID6: switch(rlq) { case DDF_RAID5_N_RESTART: return ALGORITHM_LEFT_ASYMMETRIC; @@ -1618,10 +1715,74 @@ static int rlq_to_layout(int rlq, int prl, int raiddisks) default: return -1; } + case DDF_RAID6: + switch(rlq) { + case DDF_RAID5_N_RESTART: + return ALGORITHM_LEFT_ASYMMETRIC; + case DDF_RAID6_0_RESTART: + return ALGORITHM_RIGHT_ASYMMETRIC; + case DDF_RAID5_N_CONTINUE: + return ALGORITHM_LEFT_SYMMETRIC; + default: + return -1; + } } return -1; } +#ifndef MDASSEMBLE +struct extent { + unsigned long long start, size; +}; +static int cmp_extent(const void *av, const void *bv) +{ + const struct extent *a = av; + const struct extent *b = bv; + if (a->start < b->start) + return -1; + if (a->start > b->start) + return 1; + return 0; +} + +static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl) +{ + /* find a list of used extents on the give physical device + * (dnum) of the given ddf. + * Return a malloced array of 'struct extent' + +FIXME ignore DDF_Legacy devices? + + */ + struct extent *rv; + int n = 0; + int i, j; + + rv = malloc(sizeof(struct extent) * (ddf->max_part + 2)); + if (!rv) + return NULL; + + for (i = 0; i < ddf->max_part; i++) { + struct vcl *v = dl->vlist[i]; + if (v == NULL) + continue; + for (j=0; j < v->conf.prim_elmnt_count; j++) + if (v->conf.phys_refnum[j] == dl->disk.refnum) { + /* This device plays role 'j' in 'v'. */ + rv[n].start = __be64_to_cpu(v->lba_offset[j]); + rv[n].size = __be64_to_cpu(v->conf.blocks); + n++; + break; + } + } + qsort(rv, n, sizeof(*rv), cmp_extent); + + rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size); + rv[n].size = 0; + return rv; +} +#endif + static int init_super_ddf_bvd(struct supertype *st, mdu_array_info_t *info, unsigned long long size, @@ -1637,8 +1798,6 @@ static int init_super_ddf_bvd(struct supertype *st, struct virtual_entry *ve; struct vcl *vcl; struct vd_config *vc; - int mppe; - int conflen; if (__be16_to_cpu(ddf->virt->populated_vdes) >= __be16_to_cpu(ddf->virt->max_vdes)) { @@ -1666,10 +1825,12 @@ static int init_super_ddf_bvd(struct supertype *st, ve->pad0 = 0xFFFF; ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN); ve->type = 0; - ve->state = 0; - ve->init_state = 0; - if (!(info->state & 1)) - ve->init_state = DDF_state_inconsistent; + ve->state = DDF_state_degraded; /* Will be modified as devices are added */ + if (info->state & 1) /* clean */ + ve->init_state = DDF_init_full; + else + ve->init_state = DDF_init_not; + memset(ve->pad1, 0xff, 14); memset(ve->name, ' ', 16); if (name) @@ -1678,10 +1839,12 @@ static int init_super_ddf_bvd(struct supertype *st, __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1); /* Now create a new vd_config */ - conflen = __be16_to_cpu(ddf->active->config_record_len); - vcl = malloc(offsetof(struct vcl, conf) + conflen * 512); - mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries); - vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[mppe]; + posix_memalign((void**)&vcl, 512, + (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)); + vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe]; + vcl->vcnum = venum; + sprintf(st->subarray, "%d", venum); + vcl->block_sizes = NULL; /* FIXME not for CONCAT */ vc = &vcl->conf; @@ -1721,15 +1884,17 @@ static int init_super_ddf_bvd(struct supertype *st, memset(vc->v3, 0xff, 16); memset(vc->vendor, 0xff, 32); - memset(vc->phys_refnum, 0xff, 4*mppe); - memset(vc->phys_refnum+mppe, 0x00, 8*mppe); + memset(vc->phys_refnum, 0xff, 4*ddf->mppe); + memset(vc->phys_refnum+(ddf->mppe * 4), 0x00, 8*ddf->mppe); vcl->next = ddf->conflist; ddf->conflist = vcl; - ddf->newconf = vcl; + ddf->currentconf = vcl; + ddf->updates_pending = 1; return 1; } +#ifndef MDASSEMBLE static void add_to_super_ddf_bvd(struct supertype *st, mdu_disk_info_t *dk, int fd, char *devname) { @@ -1738,13 +1903,16 @@ static void add_to_super_ddf_bvd(struct supertype *st, * We need to find suitable free space in that device and update * the phys_refnum and lba_offset for the newly created vd_config. * We might also want to update the type in the phys_disk - * section. FIXME + * section. */ struct dl *dl; struct ddf_super *ddf = st->sb; struct vd_config *vc; __u64 *lba_offset; - int mppe; + int working; + int i; + unsigned long long blocks, pos, esize; + struct extent *ex; for (dl = ddf->dlist; dl ; dl = dl->next) if (dl->major == dk->major && @@ -1753,16 +1921,69 @@ static void add_to_super_ddf_bvd(struct supertype *st, if (!dl || ! (dk->state & (1<newconf->conf; + vc = &ddf->currentconf->conf; + lba_offset = ddf->currentconf->lba_offset; + + ex = get_extents(ddf, dl); + if (!ex) + return; + + i = 0; pos = 0; + blocks = __be64_to_cpu(vc->blocks); + if (ddf->currentconf->block_sizes) + blocks = ddf->currentconf->block_sizes[dk->raid_disk]; + + do { + esize = ex[i].start - pos; + if (esize >= blocks) + break; + pos = ex[i].start + ex[i].size; + i++; + } while (ex[i-1].size); + + free(ex); + if (esize < blocks) + return; + + ddf->currentdev = dk->raid_disk; vc->phys_refnum[dk->raid_disk] = dl->disk.refnum; - mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries); - lba_offset = (__u64*)(vc->phys_refnum + mppe); - lba_offset[dk->raid_disk] = 0; /* FIXME */ + lba_offset[dk->raid_disk] = __cpu_to_be64(pos); - dl->vlist[0] =ddf->newconf; /* FIXME */ + for (i=0; i < ddf->max_part ; i++) + if (dl->vlist[i] == NULL) + break; + if (i == ddf->max_part) + return; + dl->vlist[i] = ddf->currentconf; dl->fd = fd; dl->devname = devname; + + /* Check how many working raid_disks, and if we can mark + * array as optimal yet + */ + working = 0; + + for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++) + if (vc->phys_refnum[i] != 0xffffffff) + working++; + + /* Find which virtual_entry */ + i = ddf->currentconf->vcnum; + if (working == __be16_to_cpu(vc->prim_elmnt_count)) + ddf->virt->entries[i].state = + (ddf->virt->entries[i].state & ~DDF_state_mask) + | DDF_state_optimal; + + if (vc->prl == DDF_RAID6 && + working+1 == __be16_to_cpu(vc->prim_elmnt_count)) + ddf->virt->entries[i].state = + (ddf->virt->entries[i].state & ~DDF_state_mask) + | DDF_state_part_optimal; + + ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare); + ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD); + ddf->updates_pending = 1; } /* add a device to a container, either while creating it or while @@ -1780,16 +2001,22 @@ static void add_to_super_ddf(struct supertype *st, int n, i; struct stat stb; + if (ddf->currentconf) { + add_to_super_ddf_bvd(st, dk, fd, devname); + return; + } + /* This is device numbered dk->number. We need to create * a phys_disk entry and a more detailed disk_data entry. */ fstat(fd, &stb); - dd = malloc(sizeof(*dd) + sizeof(dd->vlist[0]) * (ddf->max_part+1)); + posix_memalign((void**)&dd, 512, + sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part); dd->major = major(stb.st_rdev); dd->minor = minor(stb.st_rdev); dd->devname = devname; - dd->next = ddf->dlist; dd->fd = fd; + dd->spare = NULL; dd->disk.magic = DDF_PHYS_DATA_MAGIC; now = time(0); @@ -1799,23 +2026,45 @@ static void add_to_super_ddf(struct supertype *st, *(__u32*)(dd->disk.guid + 16) = random(); *(__u32*)(dd->disk.guid + 20) = random(); - dd->disk.refnum = random(); /* and hope for the best FIXME check this is unique!!*/ + do { + /* Cannot be bothered finding a CRC of some irrelevant details*/ + dd->disk.refnum = random(); + for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1; + i >= 0; i--) + if (ddf->phys->entries[i].refnum == dd->disk.refnum) + break; + } while (i >= 0); + dd->disk.forced_ref = 1; dd->disk.forced_guid = 1; memset(dd->disk.vendor, ' ', 32); memcpy(dd->disk.vendor, "Linux", 5); memset(dd->disk.pad, 0xff, 442); - for (i = 0; i < ddf->max_part+1 ; i++) + for (i = 0; i < ddf->max_part ; i++) dd->vlist[i] = NULL; n = __be16_to_cpu(ddf->phys->used_pdes); pde = &ddf->phys->entries[n]; - n++; - ddf->phys->used_pdes = __cpu_to_be16(n); + dd->pdnum = n; + + if (st->update_tail) { + int len = (sizeof(struct phys_disk) + + sizeof(struct phys_disk_entry)); + struct phys_disk *pd; + + pd = malloc(len); + pd->magic = DDF_PHYS_RECORDS_MAGIC; + pd->used_pdes = __cpu_to_be16(n); + pde = &pd->entries[0]; + dd->mdupdate = pd; + } else { + n++; + ddf->phys->used_pdes = __cpu_to_be16(n); + } memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN); pde->refnum = dd->disk.refnum; - pde->type = __cpu_to_be16(DDF_Forced_PD_GUID |DDF_Global_Spare); + pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare); pde->state = __cpu_to_be16(DDF_Online); get_dev_size(fd, NULL, &size); /* We are required to reserve 32Meg, and record the size in sectors */ @@ -1823,7 +2072,15 @@ static void add_to_super_ddf(struct supertype *st, sprintf(pde->path, "%17.17s","Information: nil") ; memset(pde->pad, 0xff, 6); - ddf->dlist = dd; + dd->size = size >> 9; + if (st->update_tail) { + dd->next = ddf->add_list; + ddf->add_list = dd; + } else { + dd->next = ddf->dlist; + ddf->dlist = dd; + ddf->updates_pending = 1; + } } /* @@ -1832,8 +2089,9 @@ static void add_to_super_ddf(struct supertype *st, * container. */ -#ifndef MDASSEMBLE -static int write_init_super_ddf(struct supertype *st) +static unsigned char null_conf[4096+512]; + +static int __write_init_super_ddf(struct supertype *st, int do_close) { struct ddf_super *ddf = st->sb; @@ -1892,18 +2150,26 @@ static int write_init_super_ddf(struct supertype *st) write(fd, ddf->virt, ddf->vdsize); /* Now write lots of config records. */ - n_config = __be16_to_cpu(ddf->active->max_partitions); - conf_size = __be16_to_cpu(ddf->active->config_record_len) * 512; + n_config = ddf->max_part; + conf_size = ddf->conf_rec_len * 512; for (i = 0 ; i <= n_config ; i++) { struct vcl *c = d->vlist[i]; + if (i == n_config) + c = (struct vcl*)d->spare; if (c) { c->conf.crc = calc_crc(&c->conf, conf_size); write(fd, &c->conf, conf_size); } else { - __u32 sig = 0xffffffff; - write(fd, &sig, 4); - lseek64(fd, conf_size-4, SEEK_CUR); + char *null_aligned = (char*)((((unsigned long)null_conf)+511)&~511UL); + if (null_conf[0] != 0xff) + memset(null_conf, 0xff, sizeof(null_conf)); + int togo = conf_size; + while (togo > sizeof(null_conf)-512) { + write(fd, null_aligned, sizeof(null_conf)-512); + togo -= sizeof(null_conf)-512; + } + write(fd, null_aligned, togo); } } d->disk.crc = calc_crc(&d->disk, 512); @@ -1913,10 +2179,59 @@ static int write_init_super_ddf(struct supertype *st) lseek64(fd, (size-1)*512, SEEK_SET); write(fd, &ddf->anchor, 512); - close(fd); + if (do_close) { + close(fd); + d->fd = -1; + } } return 1; } + +static int write_init_super_ddf(struct supertype *st) +{ + + if (st->update_tail) { + /* queue the virtual_disk and vd_config as metadata updates */ + struct virtual_disk *vd; + struct vd_config *vc; + struct ddf_super *ddf = st->sb; + int len; + + if (!ddf->currentconf) { + int len = (sizeof(struct phys_disk) + + sizeof(struct phys_disk_entry)); + + /* adding a disk to the container. */ + if (!ddf->add_list) + return 0; + + append_metadata_update(st, ddf->add_list->mdupdate, len); + ddf->add_list->mdupdate = NULL; + return 0; + } + + /* Newly created VD */ + + /* First the virtual disk. We have a slightly fake header */ + len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry); + vd = malloc(len); + *vd = *ddf->virt; + vd->entries[0] = ddf->virt->entries[ddf->currentconf->vcnum]; + vd->populated_vdes = __cpu_to_be16(ddf->currentconf->vcnum); + append_metadata_update(st, vd, len); + + /* Then the vd_config */ + len = ddf->conf_rec_len * 512; + vc = malloc(len); + memcpy(vc, &ddf->currentconf->conf, len); + append_metadata_update(st, vc, len); + + /* FIXME I need to close the fds! */ + return 0; + } else + return __write_init_super_ddf(st, 1); +} + #endif static __u64 avail_size_ddf(struct supertype *st, __u64 devsize) @@ -1928,10 +2243,24 @@ static __u64 avail_size_ddf(struct supertype *st, __u64 devsize) } #ifndef MDASSEMBLE -int validate_geometry_ddf(struct supertype *st, - int level, int layout, int raiddisks, - int chunk, unsigned long long size, - char *dev, unsigned long long *freesize) +static int +validate_geometry_ddf_container(struct supertype *st, + int level, int layout, int raiddisks, + int chunk, unsigned long long size, + char *dev, unsigned long long *freesize, + int verbose); + +static int validate_geometry_ddf_bvd(struct supertype *st, + int level, int layout, int raiddisks, + int chunk, unsigned long long size, + char *dev, unsigned long long *freesize, + int verbose); + +static int validate_geometry_ddf(struct supertype *st, + int level, int layout, int raiddisks, + int chunk, unsigned long long size, + char *dev, unsigned long long *freesize, + int verbose) { int fd; struct mdinfo *sra; @@ -1945,71 +2274,74 @@ int validate_geometry_ddf(struct supertype *st, */ if (level == LEVEL_CONTAINER) { - st->ss = &super_ddf_container; - if (dev) { - int rv =st->ss->validate_geometry(st, level, layout, - raiddisks, chunk, - size, - NULL, freesize); - if (rv) - return rv; - } - return st->ss->validate_geometry(st, level, layout, raiddisks, - chunk, size, dev, freesize); + /* Must be a fresh device to add to a container */ + return validate_geometry_ddf_container(st, level, layout, + raiddisks, chunk, + size, dev, freesize, + verbose); } if (st->sb) { - /* creating in a given container */ - st->ss = &super_ddf_bvd; - if (dev) { - int rv =st->ss->validate_geometry(st, level, layout, - raiddisks, chunk, - size, - NULL, freesize); - if (rv) - return rv; - } - return st->ss->validate_geometry(st, level, layout, raiddisks, - chunk, size, dev, freesize); + /* A container has already been opened, so we are + * creating in there. Maybe a BVD, maybe an SVD. + * Should make a distinction one day. + */ + return validate_geometry_ddf_bvd(st, level, layout, raiddisks, + chunk, size, dev, freesize, + verbose); } - /* FIXME should exclude MULTIPATH, or more appropriately, allow - * only known levels. - */ - if (!dev) + if (!dev) { + /* Initial sanity check. Exclude illegal levels. */ + int i; + for (i=0; ddf_level_num[i].num1 != MAXINT; i++) + if (ddf_level_num[i].num2 == level) + break; + if (ddf_level_num[i].num1 == MAXINT) + return 0; + /* Should check layout? etc */ return 1; + } - /* This device needs to be either a device in a 'ddf' container, - * or it needs to be a 'ddf-bvd' array. + /* This is the first device for the array. + * If it is a container, we read it in and do automagic allocations, + * no other devices should be given. + * Otherwise it must be a member device of a container, and we + * do manual allocation. + * Later we should check for a BVD and make an SVD. */ - fd = open(dev, O_RDONLY|O_EXCL, 0); if (fd >= 0) { sra = sysfs_read(fd, 0, GET_VERSION); close(fd); if (sra && sra->array.major_version == -1 && - strcmp(sra->text_version, "ddf-bvd") == 0) { - st->ss = &super_ddf_svd; - return st->ss->validate_geometry(st, level, layout, - raiddisks, chunk, size, - dev, freesize); + strcmp(sra->text_version, "ddf") == 0) { + + /* load super */ + /* find space for 'n' devices. */ + /* remember the devices */ + /* Somehow return the fact that we have enough */ } - fprintf(stderr, - Name ": Cannot create this array on device %s\n", - dev); + if (verbose) + fprintf(stderr, + Name ": ddf: Cannot create this array " + "on device %s\n", + dev); return 0; } if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) { - fprintf(stderr, Name ": Cannot open %s: %s\n", - dev, strerror(errno)); + if (verbose) + fprintf(stderr, Name ": ddf: Cannot open %s: %s\n", + dev, strerror(errno)); return 0; } /* Well, it is in use by someone, maybe a 'ddf' container. */ cfd = open_container(fd); if (cfd < 0) { close(fd); - fprintf(stderr, Name ": Cannot use %s: It is busy\n", - dev); + if (verbose) + fprintf(stderr, Name ": ddf: Cannot use %s: %s\n", + dev, strerror(EBUSY)); return 0; } sra = sysfs_read(cfd, 0, GET_VERSION); @@ -2020,25 +2352,28 @@ int validate_geometry_ddf(struct supertype *st, * and try to create a bvd */ struct ddf_super *ddf; - st->ss = &super_ddf_bvd; if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) { st->sb = ddf; + st->container_dev = fd2devnum(cfd); close(cfd); - return st->ss->validate_geometry(st, level, layout, + return validate_geometry_ddf_bvd(st, level, layout, raiddisks, chunk, size, - dev, freesize); + dev, freesize, + verbose); } close(cfd); - } - fprintf(stderr, Name ": Cannot use %s: Already in use\n", - dev); + } else /* device may belong to a different container */ + return 0; + return 1; } -int validate_geometry_ddf_container(struct supertype *st, - int level, int layout, int raiddisks, - int chunk, unsigned long long size, - char *dev, unsigned long long *freesize) +static int +validate_geometry_ddf_container(struct supertype *st, + int level, int layout, int raiddisks, + int chunk, unsigned long long size, + char *dev, unsigned long long *freesize, + int verbose) { int fd; unsigned long long ldsize; @@ -2050,8 +2385,9 @@ int validate_geometry_ddf_container(struct supertype *st, fd = open(dev, O_RDONLY|O_EXCL, 0); if (fd < 0) { - fprintf(stderr, Name ": Cannot open %s: %s\n", - dev, strerror(errno)); + if (verbose) + fprintf(stderr, Name ": ddf: Cannot open %s: %s\n", + dev, strerror(errno)); return 0; } if (!get_dev_size(fd, dev, &ldsize)) { @@ -2060,76 +2396,16 @@ int validate_geometry_ddf_container(struct supertype *st, } close(fd); - *freesize = avail_size_ddf(st, ldsize); + *freesize = avail_size_ddf(st, ldsize >> 9); return 1; } -struct extent { - unsigned long long start, size; -}; -int cmp_extent(const void *av, const void *bv) -{ - const struct extent *a = av; - const struct extent *b = bv; - if (a->start < b->start) - return -1; - if (a->start > b->start) - return 1; - return 0; -} - -struct extent *get_extents(struct ddf_super *ddf, struct dl *dl) -{ - /* find a list of used extents on the give physical device - * (dnum) or the given ddf. - * Return a malloced array of 'struct extent' - -FIXME ignore DDF_Legacy devices? - - */ - struct extent *rv; - int n = 0; - int dnum; - int i, j; - - for (dnum = 0; dnum < ddf->phys->used_pdes; dnum++) - if (memcmp(dl->disk.guid, - ddf->phys->entries[dnum].guid, - DDF_GUID_LEN) == 0) - break; - - if (dnum == ddf->phys->used_pdes) - return NULL; - - rv = malloc(sizeof(struct extent) * (ddf->max_part + 2)); - if (!rv) - return NULL; - - for (i = 0; i < ddf->max_part+1; i++) { - struct vcl *v = dl->vlist[i]; - if (v == NULL) - continue; - for (j=0; j < v->conf.prim_elmnt_count; j++) - if (v->conf.phys_refnum[j] == dl->disk.refnum) { - /* This device plays role 'j' in 'v'. */ - rv[n].start = __be64_to_cpu(v->lba_offset[j]); - rv[n].size = __be64_to_cpu(v->conf.blocks); - n++; - break; - } - } - qsort(rv, n, sizeof(*rv), cmp_extent); - - rv[n].start = __be64_to_cpu(ddf->phys->entries[dnum].config_size); - rv[n].size = 0; - return rv; -} - -int validate_geometry_ddf_bvd(struct supertype *st, - int level, int layout, int raiddisks, - int chunk, unsigned long long size, - char *dev, unsigned long long *freesize) +static int validate_geometry_ddf_bvd(struct supertype *st, + int level, int layout, int raiddisks, + int chunk, unsigned long long size, + char *dev, unsigned long long *freesize, + int verbose) { struct stat stb; struct ddf_super *ddf = st->sb; @@ -2156,6 +2432,7 @@ int validate_geometry_ddf_bvd(struct supertype *st, for (dl = ddf->dlist; dl ; dl = dl->next) { int found = 0; + pos = 0; i = 0; e = get_extents(ddf, dl); @@ -2173,9 +2450,11 @@ int validate_geometry_ddf_bvd(struct supertype *st, free(e); } if (dcnt < raiddisks) { - fprintf(stderr, Name ": Not enough devices with space " - "for this array (%d < %d)\n", - dcnt, raiddisks); + if (verbose) + fprintf(stderr, + Name ": ddf: Not enough devices with " + "space for this array (%d < %d)\n", + dcnt, raiddisks); return 0; } return 1; @@ -2191,8 +2470,10 @@ int validate_geometry_ddf_bvd(struct supertype *st, break; } if (!dl) { - fprintf(stderr, Name ": %s is not in the same DDF set\n", - dev); + if (verbose) + fprintf(stderr, Name ": ddf: %s is not in the " + "same DDF set\n", + dev); return 0; } e = get_extents(ddf, dl); @@ -2211,19 +2492,6 @@ int validate_geometry_ddf_bvd(struct supertype *st, return 1; } -int validate_geometry_ddf_svd(struct supertype *st, - int level, int layout, int raiddisks, - int chunk, unsigned long long size, - char *dev, unsigned long long *freesize) -{ - /* dd/svd only supports striped, mirrored, concat, spanned... */ - if (level != LEVEL_LINEAR && - level != 0 && - level != 1) - return 0; - return 1; -} - static int load_super_ddf_all(struct supertype *st, int fd, void **sbp, char *devname, int keep_fd) @@ -2244,19 +2512,19 @@ static int load_super_ddf_all(struct supertype *st, int fd, strcmp(sra->text_version, "ddf") != 0) return 1; - super = malloc(sizeof(*super)); - if (!super) + if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0) return 1; + memset(super, 0, sizeof(*super)); /* first, try each device, and choose the best ddf */ for (sd = sra->devs ; sd ; sd = sd->next) { int rv; sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); - dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY); - if (!dfd) + dfd = dev_open(nm, O_RDONLY); + if (dfd < 0) return 2; rv = load_ddf_headers(dfd, super, NULL); - if (!keep_fd) close(dfd); + close(dfd); if (rv == 0) { seq = __be32_to_cpu(super->active->seq); if (super->active->openflag) @@ -2272,7 +2540,7 @@ static int load_super_ddf_all(struct supertype *st, int fd, /* OK, load this ddf */ sprintf(nm, "%d:%d", best->disk.major, best->disk.minor); dfd = dev_open(nm, O_RDONLY); - if (!dfd) + if (dfd < 0) return 1; load_ddf_headers(dfd, super, NULL); load_ddf_global(dfd, super, NULL); @@ -2281,22 +2549,32 @@ static int load_super_ddf_all(struct supertype *st, int fd, for (sd = sra->devs ; sd ; sd = sd->next) { sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY); - if (!dfd) + if (dfd < 0) return 2; + load_ddf_headers(dfd, super, NULL); seq = load_ddf_local(dfd, super, NULL, keep_fd); if (!keep_fd) close(dfd); } + if (st->subarray[0]) { + struct vcl *v; + + for (v = super->conflist; v; v = v->next) + if (v->vcnum == atoi(st->subarray)) + super->currentconf = v; + if (!super->currentconf) + return 1; + } *sbp = super; if (st->ss == NULL) { - st->ss = &super_ddf_container; + st->ss = &super_ddf; st->minor_version = 0; st->max_devs = 512; + st->container_dev = fd2devnum(fd); } + st->loaded_container = 1; return 0; } -#endif - - +#endif /* MDASSEMBLE */ static struct mdinfo *container_content_ddf(struct supertype *st) { @@ -2314,7 +2592,6 @@ static struct mdinfo *container_content_ddf(struct supertype *st) for (vc = ddf->conflist ; vc ; vc=vc->next) { - int mppe; int i; struct mdinfo *this; this = malloc(sizeof(*this)); @@ -2322,38 +2599,47 @@ static struct mdinfo *container_content_ddf(struct supertype *st) this->next = rest; rest = this; - this->array.major_version = 1000; - this->array.minor_version = 0; - this->array.patch_version = 0; this->array.level = map_num1(ddf_level_num, vc->conf.prl); this->array.raid_disks = __be16_to_cpu(vc->conf.prim_elmnt_count); - /* FIXME this should be mapped */ - this->array.layout = vc->conf.rlq; + this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl, + this->array.raid_disks); this->array.md_minor = -1; + this->array.major_version = -1; + this->array.minor_version = -2; this->array.ctime = DECADE + __be32_to_cpu(*(__u32*)(vc->conf.guid+16)); this->array.utime = DECADE + __be32_to_cpu(vc->conf.timestamp); this->array.chunk_size = 512 << vc->conf.chunk_shift; - for (i=0; i < __be16_to_cpu(ddf->virt->populated_vdes); i++) - if (memcmp(ddf->virt->entries[i].guid, - vc->conf.guid, DDF_GUID_LEN) == 0) - break; - if (ddf->virt->entries[i].state & DDF_state_inconsistent) + i = vc->vcnum; + if ((ddf->virt->entries[i].state & DDF_state_inconsistent) || + (ddf->virt->entries[i].init_state & DDF_initstate_mask) != + DDF_init_full) { this->array.state = 0; - else + this->resync_start = 0; + } else { this->array.state = 1; + this->resync_start = ~0ULL; + } memcpy(this->name, ddf->virt->entries[i].name, 32); - this->name[33]=0; + this->name[32]=0; memset(this->uuid, 0, sizeof(this->uuid)); this->component_size = __be64_to_cpu(vc->conf.blocks); this->array.size = this->component_size / 2; + this->container_member = i; - mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries); - for (i=0 ; i < mppe ; i++) { + ddf->currentconf = vc; + uuid_from_super_ddf(st, this->uuid); + ddf->currentconf = NULL; + + sprintf(this->text_version, "/%s/%d", + devnum2devname(st->container_dev), + this->container_member); + + for (i=0 ; i < ddf->mppe ; i++) { struct mdinfo *dev; struct dl *d; @@ -2379,8 +2665,8 @@ static struct mdinfo *container_content_ddf(struct supertype *st) dev->disk.raid_disk = i; dev->disk.state = (1<events = __le32_to_cpu(ddf->primary.seq); - dev->data_offset = vc->lba_offset[i]; + dev->events = __be32_to_cpu(ddf->primary.seq); + dev->data_offset = __be64_to_cpu(vc->lba_offset[i]); dev->component_size = __be64_to_cpu(vc->conf.blocks); if (d->devname) strcpy(dev->name, d->devname); @@ -2389,27 +2675,20 @@ static struct mdinfo *container_content_ddf(struct supertype *st) return rest; } -static int init_zero_ddf(struct supertype *st, - mdu_array_info_t *info, - unsigned long long size, char *name, - char *homehost, int *uuid) -{ - st->sb = NULL; - return 0; -} - static int store_zero_ddf(struct supertype *st, int fd) { unsigned long long dsize; - char buf[512]; - memset(buf, 0, 512); - + void *buf; if (!get_dev_size(fd, NULL, &dsize)) return 1; + posix_memalign(&buf, 512, 512); + memset(buf, 0, 512); + lseek64(fd, dsize-512, 0); write(fd, buf, 512); + free(buf); return 0; } @@ -2438,6 +2717,558 @@ static int compare_super_ddf(struct supertype *st, struct supertype *tst) return 0; } +#ifndef MDASSEMBLE +/* + * A new array 'a' has been started which claims to be instance 'inst' + * within container 'c'. + * We need to confirm that the array matches the metadata in 'c' so + * that we don't corrupt any metadata. + */ +static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst) +{ + dprintf("ddf: open_new %s\n", inst); + a->info.container_member = atoi(inst); + return 0; +} + +/* + * The array 'a' is to be marked clean in the metadata. + * If '->resync_start' is not ~(unsigned long long)0, then the array is only + * clean up to the point (in sectors). If that cannot be recorded in the + * metadata, then leave it as dirty. + * + * For DDF, we need to clear the DDF_state_inconsistent bit in the + * !global! virtual_disk.virtual_entry structure. + */ +static int ddf_set_array_state(struct active_array *a, int consistent) +{ + struct ddf_super *ddf = a->container->sb; + int inst = a->info.container_member; + int old = ddf->virt->entries[inst].state; + if (consistent == 2) { + /* Should check if a recovery should be started FIXME */ + consistent = 1; + if (a->resync_start != ~0ULL) + consistent = 0; + } + if (consistent) + ddf->virt->entries[inst].state &= ~DDF_state_inconsistent; + else + ddf->virt->entries[inst].state |= DDF_state_inconsistent; + if (old != ddf->virt->entries[inst].state) + ddf->updates_pending = 1; + + old = ddf->virt->entries[inst].init_state; + ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask; + if (a->resync_start == ~0ULL) + ddf->virt->entries[inst].init_state |= DDF_init_full; + else if (a->resync_start == 0) + ddf->virt->entries[inst].init_state |= DDF_init_not; + else + ddf->virt->entries[inst].init_state |= DDF_init_quick; + if (old != ddf->virt->entries[inst].init_state) + ddf->updates_pending = 1; + + dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty", + a->resync_start); + return consistent; +} + +/* + * The state of each disk is stored in the global phys_disk structure + * in phys_disk.entries[n].state. + * This makes various combinations awkward. + * - When a device fails in any array, it must be failed in all arrays + * that include a part of this device. + * - When a component is rebuilding, we cannot include it officially in the + * array unless this is the only array that uses the device. + * + * So: when transitioning: + * Online -> failed, just set failed flag. monitor will propagate + * spare -> online, the device might need to be added to the array. + * spare -> failed, just set failed. Don't worry if in array or not. + */ +static void ddf_set_disk(struct active_array *a, int n, int state) +{ + struct ddf_super *ddf = a->container->sb; + int inst = a->info.container_member; + struct vd_config *vc = find_vdcr(ddf, inst); + int pd = find_phys(ddf, vc->phys_refnum[n]); + int i, st, working; + + if (vc == NULL) { + dprintf("ddf: cannot find instance %d!!\n", inst); + return; + } + if (pd < 0) { + /* disk doesn't currently exist. If it is now in_sync, + * insert it. */ + if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) { + /* Find dev 'n' in a->info->devs, determine the + * ddf refnum, and set vc->phys_refnum and update + * phys->entries[] + */ + /* FIXME */ + } + } else { + int old = ddf->phys->entries[pd].state; + if (state & DS_FAULTY) + ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Failed); + if (state & DS_INSYNC) { + ddf->phys->entries[pd].state |= __cpu_to_be16(DDF_Online); + ddf->phys->entries[pd].state &= __cpu_to_be16(~DDF_Rebuilding); + } + if (old != ddf->phys->entries[pd].state) + ddf->updates_pending = 1; + } + + dprintf("ddf: set_disk %d to %x\n", n, state); + + /* Now we need to check the state of the array and update + * virtual_disk.entries[n].state. + * It needs to be one of "optimal", "degraded", "failed". + * I don't understand 'deleted' or 'missing'. + */ + working = 0; + for (i=0; i < a->info.array.raid_disks; i++) { + pd = find_phys(ddf, vc->phys_refnum[i]); + if (pd < 0) + continue; + st = __be16_to_cpu(ddf->phys->entries[pd].state); + if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding)) + == DDF_Online) + working++; + } + state = DDF_state_degraded; + if (working == a->info.array.raid_disks) + state = DDF_state_optimal; + else switch(vc->prl) { + case DDF_RAID0: + case DDF_CONCAT: + case DDF_JBOD: + state = DDF_state_failed; + break; + case DDF_RAID1: + if (working == 0) + state = DDF_state_failed; + break; + case DDF_RAID4: + case DDF_RAID5: + if (working < a->info.array.raid_disks-1) + state = DDF_state_failed; + break; + case DDF_RAID6: + if (working < a->info.array.raid_disks-2) + state = DDF_state_failed; + else if (working == a->info.array.raid_disks-1) + state = DDF_state_part_optimal; + break; + } + + if (ddf->virt->entries[inst].state != + ((ddf->virt->entries[inst].state & ~DDF_state_mask) + | state)) { + + ddf->virt->entries[inst].state = + (ddf->virt->entries[inst].state & ~DDF_state_mask) + | state; + ddf->updates_pending = 1; + } + +} + +static void ddf_sync_metadata(struct supertype *st) +{ + + /* + * Write all data to all devices. + * Later, we might be able to track whether only local changes + * have been made, or whether any global data has been changed, + * but ddf is sufficiently weird that it probably always + * changes global data .... + */ + struct ddf_super *ddf = st->sb; + if (!ddf->updates_pending) + return; + ddf->updates_pending = 0; + __write_init_super_ddf(st, 0); + dprintf("ddf: sync_metadata\n"); +} + +static void ddf_process_update(struct supertype *st, + struct metadata_update *update) +{ + /* Apply this update to the metadata. + * The first 4 bytes are a DDF_*_MAGIC which guides + * our actions. + * Possible update are: + * DDF_PHYS_RECORDS_MAGIC + * Add a new physical device. Changes to this record + * only happen implicitly. + * used_pdes is the device number. + * DDF_VIRT_RECORDS_MAGIC + * Add a new VD. Possibly also change the 'access' bits. + * populated_vdes is the entry number. + * DDF_VD_CONF_MAGIC + * New or updated VD. the VIRT_RECORD must already + * exist. For an update, phys_refnum and lba_offset + * (at least) are updated, and the VD_CONF must + * be written to precisely those devices listed with + * a phys_refnum. + * DDF_SPARE_ASSIGN_MAGIC + * replacement Spare Assignment Record... but for which device? + * + * So, e.g.: + * - to create a new array, we send a VIRT_RECORD and + * a VD_CONF. Then assemble and start the array. + * - to activate a spare we send a VD_CONF to add the phys_refnum + * and offset. This will also mark the spare as active with + * a spare-assignment record. + */ + struct ddf_super *ddf = st->sb; + __u32 *magic = (__u32*)update->buf; + struct phys_disk *pd; + struct virtual_disk *vd; + struct vd_config *vc; + struct vcl *vcl; + struct dl *dl; + int mppe; + int ent; + + dprintf("Process update %x\n", *magic); + + switch (*magic) { + case DDF_PHYS_RECORDS_MAGIC: + + if (update->len != (sizeof(struct phys_disk) + + sizeof(struct phys_disk_entry))) + return; + pd = (struct phys_disk*)update->buf; + + ent = __be16_to_cpu(pd->used_pdes); + if (ent >= __be16_to_cpu(ddf->phys->max_pdes)) + return; + if (!all_ff(ddf->phys->entries[ent].guid)) + return; + ddf->phys->entries[ent] = pd->entries[0]; + ddf->phys->used_pdes = __cpu_to_be16(1 + + __be16_to_cpu(ddf->phys->used_pdes)); + ddf->updates_pending = 1; + if (ddf->add_list) { + struct active_array *a; + struct dl *al = ddf->add_list; + ddf->add_list = al->next; + + al->next = ddf->dlist; + ddf->dlist = al; + + /* As a device has been added, we should check + * for any degraded devices that might make + * use of this spare */ + for (a = st->arrays ; a; a=a->next) + a->check_degraded = 1; + } + break; + + case DDF_VIRT_RECORDS_MAGIC: + + if (update->len != (sizeof(struct virtual_disk) + + sizeof(struct virtual_entry))) + return; + vd = (struct virtual_disk*)update->buf; + + ent = __be16_to_cpu(vd->populated_vdes); + if (ent >= __be16_to_cpu(ddf->virt->max_vdes)) + return; + if (!all_ff(ddf->virt->entries[ent].guid)) + return; + ddf->virt->entries[ent] = vd->entries[0]; + ddf->virt->populated_vdes = __cpu_to_be16(1 + + __be16_to_cpu(ddf->virt->populated_vdes)); + ddf->updates_pending = 1; + break; + + case DDF_VD_CONF_MAGIC: + dprintf("len %d %d\n", update->len, ddf->conf_rec_len); + + mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries); + if (update->len != ddf->conf_rec_len * 512) + return; + vc = (struct vd_config*)update->buf; + for (vcl = ddf->conflist; vcl ; vcl = vcl->next) + if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0) + break; + dprintf("vcl = %p\n", vcl); + if (vcl) { + /* An update, just copy the phys_refnum and lba_offset + * fields + */ + memcpy(vcl->conf.phys_refnum, vc->phys_refnum, + mppe * (sizeof(__u32) + sizeof(__u64))); + } else { + /* A new VD_CONF */ + vcl = update->space; + update->space = NULL; + vcl->next = ddf->conflist; + memcpy(&vcl->conf, vc, update->len); + vcl->lba_offset = (__u64*) + &vcl->conf.phys_refnum[mppe]; + ddf->conflist = vcl; + } + /* Now make sure vlist is correct for each dl. */ + for (dl = ddf->dlist; dl; dl = dl->next) { + int dn; + int vn = 0; + for (vcl = ddf->conflist; vcl ; vcl = vcl->next) + for (dn=0; dn < ddf->mppe ; dn++) + if (vcl->conf.phys_refnum[dn] == + dl->disk.refnum) { + dprintf("dev %d has %p at %d\n", + dl->pdnum, vcl, vn); + dl->vlist[vn++] = vcl; + break; + } + while (vn < ddf->max_part) + dl->vlist[vn++] = NULL; + if (dl->vlist[0]) { + ddf->phys->entries[dl->pdnum].type &= + ~__cpu_to_be16(DDF_Global_Spare); + ddf->phys->entries[dl->pdnum].type |= + __cpu_to_be16(DDF_Active_in_VD); + } + if (dl->spare) { + ddf->phys->entries[dl->pdnum].type &= + ~__cpu_to_be16(DDF_Global_Spare); + ddf->phys->entries[dl->pdnum].type |= + __cpu_to_be16(DDF_Spare); + } + if (!dl->vlist[0] && !dl->spare) { + ddf->phys->entries[dl->pdnum].type |= + __cpu_to_be16(DDF_Global_Spare); + ddf->phys->entries[dl->pdnum].type &= + ~__cpu_to_be16(DDF_Spare | + DDF_Active_in_VD); + } + } + ddf->updates_pending = 1; + break; + case DDF_SPARE_ASSIGN_MAGIC: + default: break; + } +} + +static void ddf_prepare_update(struct supertype *st, + struct metadata_update *update) +{ + /* This update arrived at managemon. + * We are about to pass it to monitor. + * If a malloc is needed, do it here. + */ + struct ddf_super *ddf = st->sb; + __u32 *magic = (__u32*)update->buf; + if (*magic == DDF_VD_CONF_MAGIC) + posix_memalign(&update->space, 512, + offsetof(struct vcl, conf) + + ddf->conf_rec_len * 512); +} + +/* + * Check if the array 'a' is degraded but not failed. + * If it is, find as many spares as are available and needed and + * arrange for their inclusion. + * We only choose devices which are not already in the array, + * and prefer those with a spare-assignment to this array. + * otherwise we choose global spares - assuming always that + * there is enough room. + * For each spare that we assign, we return an 'mdinfo' which + * describes the position for the device in the array. + * We also add to 'updates' a DDF_VD_CONF_MAGIC update with + * the new phys_refnum and lba_offset values. + * + * Only worry about BVDs at the moment. + */ +static struct mdinfo *ddf_activate_spare(struct active_array *a, + struct metadata_update **updates) +{ + int working = 0; + struct mdinfo *d; + struct ddf_super *ddf = a->container->sb; + int global_ok = 0; + struct mdinfo *rv = NULL; + struct mdinfo *di; + struct metadata_update *mu; + struct dl *dl; + int i; + struct vd_config *vc; + __u64 *lba; + + for (d = a->info.devs ; d ; d = d->next) { + if ((d->curr_state & DS_FAULTY) && + d->state_fd >= 0) + /* wait for Removal to happen */ + return NULL; + if (d->state_fd >= 0) + working ++; + } + + dprintf("ddf_activate: working=%d (%d) level=%d\n", working, a->info.array.raid_disks, + a->info.array.level); + if (working == a->info.array.raid_disks) + return NULL; /* array not degraded */ + switch (a->info.array.level) { + case 1: + if (working == 0) + return NULL; /* failed */ + break; + case 4: + case 5: + if (working < a->info.array.raid_disks - 1) + return NULL; /* failed */ + break; + case 6: + if (working < a->info.array.raid_disks - 2) + return NULL; /* failed */ + break; + default: /* concat or stripe */ + return NULL; /* failed */ + } + + /* For each slot, if it is not working, find a spare */ + dl = ddf->dlist; + for (i = 0; i < a->info.array.raid_disks; i++) { + for (d = a->info.devs ; d ; d = d->next) + if (d->disk.raid_disk == i) + break; + dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0); + if (d && (d->state_fd >= 0)) + continue; + + /* OK, this device needs recovery. Find a spare */ + again: + for ( ; dl ; dl = dl->next) { + unsigned long long esize; + unsigned long long pos; + struct mdinfo *d2; + int is_global = 0; + int is_dedicated = 0; + struct extent *ex; + int j; + /* If in this array, skip */ + for (d2 = a->info.devs ; d2 ; d2 = d2->next) + if (d2->disk.major == dl->major && + d2->disk.minor == dl->minor) { + dprintf("%x:%x already in array\n", dl->major, dl->minor); + break; + } + if (d2) + continue; + if (ddf->phys->entries[dl->pdnum].type & + __cpu_to_be16(DDF_Spare)) { + /* Check spare assign record */ + if (dl->spare) { + if (dl->spare->type & DDF_spare_dedicated) { + /* check spare_ents for guid */ + for (j = 0 ; + j < __be16_to_cpu(dl->spare->populated); + j++) { + if (memcmp(dl->spare->spare_ents[j].guid, + ddf->virt->entries[a->info.container_member].guid, + DDF_GUID_LEN) == 0) + is_dedicated = 1; + } + } else + is_global = 1; + } + } else if (ddf->phys->entries[dl->pdnum].type & + __cpu_to_be16(DDF_Global_Spare)) { + is_global = 1; + } + if ( ! (is_dedicated || + (is_global && global_ok))) { + dprintf("%x:%x not suitable: %d %d\n", dl->major, dl->minor, + is_dedicated, is_global); + continue; + } + + /* We are allowed to use this device - is there space? + * We need a->info.component_size sectors */ + ex = get_extents(ddf, dl); + if (!ex) { + dprintf("cannot get extents\n"); + continue; + } + j = 0; pos = 0; + esize = 0; + + do { + esize = ex[j].start - pos; + if (esize >= a->info.component_size) + break; + pos = ex[i].start + ex[i].size; + i++; + } while (ex[i-1].size); + + free(ex); + if (esize < a->info.component_size) { + dprintf("%x:%x has no room: %llu %llu\n", dl->major, dl->minor, + esize, a->info.component_size); + /* No room */ + continue; + } + + /* Cool, we have a device with some space at pos */ + di = malloc(sizeof(*di)); + memset(di, 0, sizeof(*di)); + di->disk.number = i; + di->disk.raid_disk = i; + di->disk.major = dl->major; + di->disk.minor = dl->minor; + di->disk.state = 0; + di->data_offset = pos; + di->component_size = a->info.component_size; + di->container_member = dl->pdnum; + di->next = rv; + rv = di; + dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor, + i, pos); + + break; + } + if (!dl && ! global_ok) { + /* not enough dedicated spares, try global */ + global_ok = 1; + dl = ddf->dlist; + goto again; + } + } + + if (!rv) + /* No spares found */ + return rv; + /* Now 'rv' has a list of devices to return. + * Create a metadata_update record to update the + * phys_refnum and lba_offset values + */ + mu = malloc(sizeof(*mu)); + mu->buf = malloc(ddf->conf_rec_len * 512); + posix_memalign(&mu->space, 512, sizeof(struct vcl)); + mu->len = ddf->conf_rec_len; + mu->next = *updates; + vc = find_vdcr(ddf, a->info.container_member); + memcpy(mu->buf, vc, ddf->conf_rec_len * 512); + + vc = (struct vd_config*)mu->buf; + lba = (__u64*)&vc->phys_refnum[ddf->mppe]; + for (di = rv ; di ; di = di->next) { + vc->phys_refnum[di->disk.raid_disk] = + ddf->phys->entries[dl->pdnum].refnum; + lba[di->disk.raid_disk] = di->data_offset; + } + *updates = mu; + return rv; +} +#endif /* MDASSEMBLE */ + struct superswitch super_ddf = { #ifndef MDASSEMBLE .examine_super = examine_super_ddf, @@ -2445,6 +3276,8 @@ struct superswitch super_ddf = { .detail_super = detail_super_ddf, .brief_detail_super = brief_detail_super_ddf, .validate_geometry = validate_geometry_ddf, + .write_init_super = write_init_super_ddf, + .add_to_super = add_to_super_ddf, #endif .match_home = match_home_ddf, .uuid_from_super= uuid_from_super_ddf, @@ -2456,80 +3289,22 @@ struct superswitch super_ddf = { .compare_super = compare_super_ddf, .load_super = load_super_ddf, - .init_super = init_zero_ddf, + .init_super = init_super_ddf, .store_super = store_zero_ddf, .free_super = free_super_ddf, .match_metadata_desc = match_metadata_desc_ddf, - .getinfo_super_n = getinfo_super_n_container, - - - .major = 1000, - .swapuuid = 0, - .external = 1, - .text_version = "ddf", -}; - -/* Super_ddf_container is set by validate_geometry_ddf when given a - * device that is not part of any array - */ -struct superswitch super_ddf_container = { -#ifndef MDASSEMBLE - .validate_geometry = validate_geometry_ddf_container, - .write_init_super = write_init_super_ddf, -#endif - - .init_super = init_super_ddf, - .add_to_super = add_to_super_ddf, - - .free_super = free_super_ddf, - .container_content = container_content_ddf, - .major = 1000, - .swapuuid = 0, .external = 1, - .text_version = "ddf", -}; - -struct superswitch super_ddf_bvd = { -#ifndef MDASSEMBLE -// .detail_super = detail_super_ddf_bvd, -// .brief_detail_super = brief_detail_super_ddf_bvd, - .validate_geometry = validate_geometry_ddf_bvd, - .write_init_super = write_init_super_ddf, -#endif - .update_super = update_super_ddf, - .init_super = init_super_ddf_bvd, - .add_to_super = add_to_super_ddf_bvd, - .getinfo_super = getinfo_super_ddf_bvd, - .getinfo_super_n = getinfo_super_n_bvd, - - .load_super = load_super_ddf, - .free_super = free_super_ddf, - .match_metadata_desc = match_metadata_desc_ddf_bvd, - - .major = 1001, - .swapuuid = 0, - .external = 2, - .text_version = "ddf", -}; - -struct superswitch super_ddf_svd = { -#ifndef MDASSEMBLE -// .detail_super = detail_super_ddf_svd, -// .brief_detail_super = brief_detail_super_ddf_svd, - .validate_geometry = validate_geometry_ddf_svd, +#ifndef MDASSEMBLE +/* for mdmon */ + .open_new = ddf_open_new, + .set_array_state= ddf_set_array_state, + .set_disk = ddf_set_disk, + .sync_metadata = ddf_sync_metadata, + .process_update = ddf_process_update, + .prepare_update = ddf_prepare_update, + .activate_spare = ddf_activate_spare, #endif - .update_super = update_super_ddf, - .init_super = init_super_ddf, - - .load_super = load_super_ddf, - .free_super = free_super_ddf, - .match_metadata_desc = match_metadata_desc_ddf_svd, - - .major = 1002, - .swapuuid = 0, - .external = 2, - .text_version = "ddf", };