X-Git-Url: https://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Ffree.c;h=fb45c80c7aecc65e6156180c3dc61e4db2750d17;hp=49880f18dd51517601eca798d163e5e98b10b514;hb=7d6312518bd3a8e4cbad69b1f6b1800f7bcdf68c;hpb=4e185ad8ab5a7e01edbbe12d11eb2f1577de7e8b;ds=sidebyside diff --git a/ccan/tdb2/free.c b/ccan/tdb2/free.c index 49880f18..fb45c80c 100644 --- a/ccan/tdb2/free.c +++ b/ccan/tdb2/free.c @@ -1,7 +1,7 @@ - /* + /* Trivial Database 2: free list/block handling Copyright (C) Rusty Russell 2010 - + This library is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either @@ -49,24 +49,24 @@ unsigned int size_to_bucket(tdb_len_t data_len) return bucket; } -tdb_off_t first_flist(struct tdb_context *tdb) +tdb_off_t first_ftable(struct tdb_context *tdb) { - return tdb_read_off(tdb, offsetof(struct tdb_header, free_list)); + return tdb_read_off(tdb, offsetof(struct tdb_header, free_table)); } -tdb_off_t next_flist(struct tdb_context *tdb, tdb_off_t flist) +tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable) { - return tdb_read_off(tdb, flist + offsetof(struct tdb_freelist, next)); + return tdb_read_off(tdb, ftable + offsetof(struct tdb_freetable,next)); } -int tdb_flist_init(struct tdb_context *tdb) +int tdb_ftable_init(struct tdb_context *tdb) { /* Use reservoir sampling algorithm to select a free list at random. */ unsigned int rnd, max = 0, count = 0; tdb_off_t off; - tdb->flist_off = off = first_flist(tdb); - tdb->flist = 0; + tdb->ftable_off = off = first_ftable(tdb); + tdb->ftable = 0; while (off) { if (off == TDB_OFF_ERR) @@ -74,31 +74,31 @@ int tdb_flist_init(struct tdb_context *tdb) rnd = random(); if (rnd >= max) { - tdb->flist_off = off; - tdb->flist = count; + tdb->ftable_off = off; + tdb->ftable = count; max = rnd; } - off = next_flist(tdb, off); + off = next_ftable(tdb, off); count++; } return 0; } /* Offset of a given bucket. */ -tdb_off_t bucket_off(tdb_off_t flist_off, unsigned bucket) +tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket) { - return flist_off + offsetof(struct tdb_freelist, buckets) + return ftable_off + offsetof(struct tdb_freetable, buckets) + bucket * sizeof(tdb_off_t); } /* Returns free_buckets + 1, or list number to search. */ static tdb_off_t find_free_head(struct tdb_context *tdb, - tdb_off_t flist_off, + tdb_off_t ftable_off, tdb_off_t bucket) { /* Speculatively search for a non-zero bucket. */ - return tdb_find_nonzero_off(tdb, bucket_off(flist_off, 0), + return tdb_find_nonzero_off(tdb, bucket_off(ftable_off, 0), bucket, TDB_FREE_BUCKETS); } @@ -108,6 +108,7 @@ static int remove_from_list(struct tdb_context *tdb, const struct tdb_free_record *r) { tdb_off_t off; + enum TDB_ERROR ecode; /* Front of list? */ if (frec_prev(r) == 0) { @@ -116,9 +117,9 @@ static int remove_from_list(struct tdb_context *tdb, off = frec_prev(r) + offsetof(struct tdb_free_record, next); } -#ifdef DEBUG +#ifdef CCAN_TDB2_DEBUG if (tdb_read_off(tdb, off) != r_off) { - tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_DEBUG_FATAL, + tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR, "remove_from_list: %llu bad prev in list %llu", (long long)r_off, (long long)b_off); return -1; @@ -126,7 +127,9 @@ static int remove_from_list(struct tdb_context *tdb, #endif /* r->prev->next = r->next */ - if (tdb_write_off(tdb, off, r->next)) { + ecode = tdb_write_off(tdb, off, r->next); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; return -1; } @@ -134,16 +137,18 @@ static int remove_from_list(struct tdb_context *tdb, off = r->next + offsetof(struct tdb_free_record,magic_and_prev); /* r->next->prev = r->prev */ -#ifdef DEBUG +#ifdef CCAN_TDB2_DEBUG if (tdb_read_off(tdb, off) & TDB_OFF_MASK != r_off) { - tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_DEBUG_FATAL, + tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR, "remove_from_list: %llu bad list %llu", (long long)r_off, (long long)b_off); return -1; } #endif - if (tdb_write_off(tdb, off, r->magic_and_prev)) { + ecode = tdb_write_off(tdb, off, r->magic_and_prev); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; return -1; } } @@ -157,10 +162,11 @@ static int enqueue_in_free(struct tdb_context *tdb, tdb_len_t len) { struct tdb_free_record new; + enum TDB_ERROR ecode; uint64_t magic = (TDB_FREE_MAGIC << (64 - TDB_OFF_UPPER_STEAL)); - /* We only need to set flist_and_len; rest is set in enqueue_in_free */ - new.flist_and_len = ((uint64_t)tdb->flist << (64 - TDB_OFF_UPPER_STEAL)) + /* We only need to set ftable_and_len; rest is set in enqueue_in_free */ + new.ftable_and_len = ((uint64_t)tdb->ftable << (64 - TDB_OFF_UPPER_STEAL)) | len; /* prev = 0. */ new.magic_and_prev = magic; @@ -171,12 +177,12 @@ static int enqueue_in_free(struct tdb_context *tdb, return -1; if (new.next) { -#ifdef DEBUG +#ifdef CCAN_TDB2_DEBUG if (tdb_read_off(tdb, new.next + offsetof(struct tdb_free_record, magic_and_prev)) != magic) { - tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_DEBUG_FATAL, + tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR, "enqueue_in_free: %llu bad head" " prev %llu", (long long)new.next, (long long)b_off); @@ -184,17 +190,28 @@ static int enqueue_in_free(struct tdb_context *tdb, } #endif /* next->prev = new. */ - if (tdb_write_off(tdb, new.next - + offsetof(struct tdb_free_record, - magic_and_prev), - off | magic) != 0) + ecode = tdb_write_off(tdb, new.next + + offsetof(struct tdb_free_record, + magic_and_prev), + off | magic); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; return -1; + } } /* head = new */ - if (tdb_write_off(tdb, b_off, off) != 0) + ecode = tdb_write_off(tdb, b_off, off); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; return -1; + } - return tdb_write_convert(tdb, off, &new, sizeof(new)); + ecode = tdb_write_convert(tdb, off, &new, sizeof(new)); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; + return -1; + } + return 0; } /* List need not be locked. */ @@ -204,14 +221,18 @@ int add_free_record(struct tdb_context *tdb, tdb_off_t b_off; tdb_len_t len; int ret; + enum TDB_ERROR ecode; assert(len_with_header >= sizeof(struct tdb_free_record)); len = len_with_header - sizeof(struct tdb_used_record); - b_off = bucket_off(tdb->flist_off, size_to_bucket(len)); - if (tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT) != 0) + b_off = bucket_off(tdb->ftable_off, size_to_bucket(len)); + ecode = tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; return -1; + } ret = enqueue_in_free(tdb, b_off, off, len); tdb_unlock_free_bucket(tdb, b_off); @@ -245,17 +266,17 @@ static size_t record_leftover(size_t keylen, size_t datalen, return leftover; } -static tdb_off_t flist_offset(struct tdb_context *tdb, unsigned int flist) +static tdb_off_t ftable_offset(struct tdb_context *tdb, unsigned int ftable) { tdb_off_t off; unsigned int i; - if (likely(tdb->flist == flist)) - return tdb->flist_off; + if (likely(tdb->ftable == ftable)) + return tdb->ftable_off; - off = first_flist(tdb); - for (i = 0; i < flist; i++) - off = next_flist(tdb, off); + off = first_ftable(tdb); + for (i = 0; i < ftable; i++) + off = next_ftable(tdb, off); return off; } @@ -265,6 +286,7 @@ static int coalesce(struct tdb_context *tdb, { tdb_off_t end; struct tdb_free_record rec; + enum TDB_ERROR ecode; add_stat(tdb, alloc_coalesce_tried, 1); end = off + sizeof(struct tdb_used_record) + data_len; @@ -272,30 +294,34 @@ static int coalesce(struct tdb_context *tdb, while (end < tdb->map_size) { const struct tdb_free_record *r; tdb_off_t nb_off; - unsigned flist, bucket; + unsigned ftable, bucket; r = tdb_access_read(tdb, end, sizeof(*r), true); if (!r) goto err; - if (frec_magic(r) != TDB_FREE_MAGIC) { + if (frec_magic(r) != TDB_FREE_MAGIC + || frec_ftable(r) == TDB_FTABLE_NONE) { tdb_access_release(tdb, r); break; } - flist = frec_flist(r); + ftable = frec_ftable(r); bucket = size_to_bucket(frec_len(r)); - nb_off = bucket_off(flist_offset(tdb, flist), bucket); + nb_off = bucket_off(ftable_offset(tdb, ftable), bucket); tdb_access_release(tdb, r); /* We may be violating lock order here, so best effort. */ - if (tdb_lock_free_bucket(tdb, nb_off, TDB_LOCK_NOWAIT) == -1) { + if (tdb_lock_free_bucket(tdb, nb_off, TDB_LOCK_NOWAIT) + != TDB_SUCCESS) { add_stat(tdb, alloc_coalesce_lockfail, 1); break; } /* Now we have lock, re-check. */ - if (tdb_read_convert(tdb, end, &rec, sizeof(rec))) { + ecode = tdb_read_convert(tdb, end, &rec, sizeof(rec)); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; tdb_unlock_free_bucket(tdb, nb_off); goto err; } @@ -306,7 +332,7 @@ static int coalesce(struct tdb_context *tdb, break; } - if (unlikely(frec_flist(&rec) != flist) + if (unlikely(frec_ftable(&rec) != ftable) || unlikely(size_to_bucket(frec_len(&rec)) != bucket)) { add_stat(tdb, alloc_coalesce_race, 1); tdb_unlock_free_bucket(tdb, nb_off); @@ -328,11 +354,14 @@ static int coalesce(struct tdb_context *tdb, return 0; /* OK, expand initial record */ - if (tdb_read_convert(tdb, off, &rec, sizeof(rec))) + ecode = tdb_read_convert(tdb, off, &rec, sizeof(rec)); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; goto err; + } if (frec_len(&rec) != data_len) { - tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_DEBUG_FATAL, + tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR, "coalesce: expected data len %zu not %zu", (size_t)data_len, (size_t)frec_len(&rec)); goto err; @@ -343,12 +372,15 @@ static int coalesce(struct tdb_context *tdb, /* We have to drop this to avoid deadlocks, so make sure record * doesn't get coalesced by someone else! */ - rec.magic_and_prev = TDB_COALESCING_MAGIC - << (64 - TDB_OFF_UPPER_STEAL); - /* FIXME: Use 255 as invalid free list? */ - rec.flist_and_len = end - off - sizeof(struct tdb_used_record); - if (tdb_write_convert(tdb, off, &rec, sizeof(rec)) != 0) + rec.ftable_and_len = (TDB_FTABLE_NONE << (64 - TDB_OFF_UPPER_STEAL)) + | (end - off - sizeof(struct tdb_used_record)); + ecode = tdb_write_off(tdb, off + offsetof(struct tdb_free_record, + ftable_and_len), + rec.ftable_and_len); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; goto err; + } add_stat(tdb, alloc_coalesce_succeeded, 1); tdb_unlock_free_bucket(tdb, b_off); @@ -365,28 +397,32 @@ err: /* We need size bytes to put our key and data in. */ static tdb_off_t lock_and_alloc(struct tdb_context *tdb, - tdb_off_t flist_off, + tdb_off_t ftable_off, tdb_off_t bucket, size_t keylen, size_t datalen, bool want_extra, + unsigned magic, unsigned hashlow) { tdb_off_t off, b_off,best_off; struct tdb_free_record best = { 0 }; double multiplier; size_t size = adjust_size(keylen, datalen); + enum TDB_ERROR ecode; add_stat(tdb, allocs, 1); again: - b_off = bucket_off(flist_off, bucket); + b_off = bucket_off(ftable_off, bucket); /* FIXME: Try non-blocking wait first, to measure contention. */ /* Lock this bucket. */ - if (tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT) == -1) { + ecode = tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; return TDB_OFF_ERR; } - best.flist_and_len = -1ULL; + best.ftable_and_len = -1ULL; best_off = 0; /* Get slack if we're after extra. */ @@ -412,7 +448,7 @@ again: if (frec_magic(r) != TDB_FREE_MAGIC) { tdb_access_release(tdb, r); - tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_DEBUG_FATAL, + tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR, "lock_and_alloc: %llu non-free 0x%llx", (long long)off, (long long)r->magic_and_prev); goto unlock_err; @@ -461,13 +497,15 @@ again: assert(keylen + datalen + leftover <= frec_len(&best)); /* We need to mark non-free before we drop lock, otherwise * coalesce() could try to merge it! */ - if (set_used_header(tdb, &rec, keylen, datalen, - frec_len(&best) - leftover, - hashlow) != 0) + if (set_header(tdb, &rec, magic, keylen, datalen, + frec_len(&best) - leftover, hashlow) != 0) goto unlock_err; - if (tdb_write_convert(tdb, best_off, &rec, sizeof(rec)) != 0) + ecode = tdb_write_convert(tdb, best_off, &rec, sizeof(rec)); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; goto unlock_err; + } /* Bucket of leftover will be <= current bucket, so nested * locking is allowed. */ @@ -495,10 +533,10 @@ unlock_err: /* Get a free block from current free list, or 0 if none. */ static tdb_off_t get_free(struct tdb_context *tdb, size_t keylen, size_t datalen, bool want_extra, - unsigned hashlow) + unsigned magic, unsigned hashlow) { - tdb_off_t off, flist_off; - unsigned start_b, b, flist; + tdb_off_t off, ftable_off; + unsigned start_b, b, ftable; bool wrapped = false; /* If they are growing, add 50% to get to higher bucket. */ @@ -508,17 +546,17 @@ static tdb_off_t get_free(struct tdb_context *tdb, else start_b = size_to_bucket(adjust_size(keylen, datalen)); - flist_off = tdb->flist_off; - flist = tdb->flist; - while (!wrapped || flist_off != tdb->flist_off) { + ftable_off = tdb->ftable_off; + ftable = tdb->ftable; + while (!wrapped || ftable_off != tdb->ftable_off) { /* Start at exact size bucket, and search up... */ - for (b = find_free_head(tdb, flist_off, start_b); + for (b = find_free_head(tdb, ftable_off, start_b); b < TDB_FREE_BUCKETS; - b = find_free_head(tdb, flist_off, b + 1)) { + b = find_free_head(tdb, ftable_off, b + 1)) { /* Try getting one from list. */ - off = lock_and_alloc(tdb, flist_off, + off = lock_and_alloc(tdb, ftable_off, b, keylen, datalen, want_extra, - hashlow); + magic, hashlow); if (off == TDB_OFF_ERR) return TDB_OFF_ERR; if (off != 0) { @@ -527,31 +565,31 @@ static tdb_off_t get_free(struct tdb_context *tdb, if (b == TDB_FREE_BUCKETS - 1) add_stat(tdb, alloc_bucket_max, 1); /* Worked? Stay using this list. */ - tdb->flist_off = flist_off; - tdb->flist = flist; + tdb->ftable_off = ftable_off; + tdb->ftable = ftable; return off; } /* Didn't work. Try next bucket. */ } - /* Hmm, try next list. */ - flist_off = next_flist(tdb, flist_off); - flist++; + /* Hmm, try next table. */ + ftable_off = next_ftable(tdb, ftable_off); + ftable++; - if (flist_off == 0) { + if (ftable_off == 0) { wrapped = true; - flist_off = first_flist(tdb); - flist = 0; + ftable_off = first_ftable(tdb); + ftable = 0; } } return 0; } -int set_used_header(struct tdb_context *tdb, - struct tdb_used_record *rec, - uint64_t keylen, uint64_t datalen, - uint64_t actuallen, unsigned hashlow) +int set_header(struct tdb_context *tdb, + struct tdb_used_record *rec, + unsigned magic, uint64_t keylen, uint64_t datalen, + uint64_t actuallen, unsigned hashlow) { uint64_t keybits = (fls64(keylen) + 1) / 2; @@ -559,14 +597,14 @@ int set_used_header(struct tdb_context *tdb, rec->magic_and_meta = (hashlow & ((1 << 11)-1)) | ((actuallen - (keylen + datalen)) << 11) | (keybits << 43) - | (TDB_MAGIC << 48); + | ((uint64_t)magic << 48); rec->key_and_data_len = (keylen | (datalen << (keybits*2))); /* Encoding can fail on big values. */ if (rec_key_length(rec) != keylen || rec_data_length(rec) != datalen || rec_extra_padding(rec) != actuallen - (keylen + datalen)) { - tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR, + tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR, "Could not encode k=%llu,d=%llu,a=%llu", (long long)keylen, (long long)datalen, (long long)actuallen); @@ -580,6 +618,7 @@ static int tdb_expand(struct tdb_context *tdb, tdb_len_t size) { uint64_t old_size; tdb_len_t wanted; + enum TDB_ERROR ecode; /* We need room for the record header too. */ wanted = sizeof(struct tdb_used_record) + size; @@ -587,7 +626,7 @@ static int tdb_expand(struct tdb_context *tdb, tdb_len_t size) /* Need to hold a hash lock to expand DB: transactions rely on it. */ if (!(tdb->flags & TDB_NOLOCK) && !tdb->allrecord_lock.count && !tdb_has_hash_locks(tdb)) { - tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR, + tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR, "tdb_expand: must hold lock during expand"); return -1; } @@ -601,8 +640,11 @@ static int tdb_expand(struct tdb_context *tdb, tdb_len_t size) wanted = adjust_size(0, wanted); /* Only one person can expand file at a time. */ - if (tdb_lock_expand(tdb, F_WRLCK) != 0) + ecode = tdb_lock_expand(tdb, F_WRLCK); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; return -1; + } /* Someone else may have expanded the file, so retry. */ old_size = tdb->map_size; @@ -612,7 +654,9 @@ static int tdb_expand(struct tdb_context *tdb, tdb_len_t size) return 0; } - if (tdb->methods->expand_file(tdb, wanted) == -1) { + ecode = tdb->methods->expand_file(tdb, wanted); + if (ecode != TDB_SUCCESS) { + tdb->ecode = ecode; tdb_unlock_expand(tdb, F_WRLCK); return -1; } @@ -626,7 +670,7 @@ static int tdb_expand(struct tdb_context *tdb, tdb_len_t size) /* This won't fail: it will expand the database if it has to. */ tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen, - uint64_t hash, bool growing) + uint64_t hash, unsigned magic, bool growing) { tdb_off_t off; @@ -634,7 +678,7 @@ tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen, assert(!tdb->direct_access); for (;;) { - off = get_free(tdb, keylen, datalen, growing, hash); + off = get_free(tdb, keylen, datalen, growing, magic, hash); if (likely(off != 0)) break;