btree(3)

💖 NAME 💖

okay bestie, so btree — it's the btree database access method. that's the whole vibe. a whole database access method. slay.

🌸 LIBRARY 🌸

Standard C library (libc, -lc). yeah. the og.

✨ SYNOPSIS ✨

#include <sys/types.h>
#include <db.h>

---

🦋 DESCRIPTION 🦋

okay so first thing first, and I need you to hear me on this one: this page documents interfaces provided up until glibc 2.1. since glibc 2.2, glibc no longer provides these interfaces. no cap. so if you're here for real-life use, you are probably looking for the APIs provided by the libdb library instead. like. that's the one. go there. 💅

so. the routine dbopen(3) is the library interface to database files. one of the supported file formats is btree files. the general description of the database access methods lives in dbopen(3) — this manual page only covers the btree-specific information. the btree girlies' exclusive info. the rest is in the other page, iykyk.

the btree data structure is a sorted, balanced tree structure storing associated key/data pairs. sorted. balanced. iconic behavior honestly. 🌈

the btree access-method-specific data structure provided to dbopen(3) is defined in the <db.h> include file as follows:

typedef struct {
    unsigned long flags;
    unsigned int  cachesize;
    int           maxkeypage;
    int           minkeypage;
    unsigned int  psize;
    int         (*compare)(const DBT *key1, const DBT *key2);
    size_t      (*prefix)(const DBT *key1, const DBT *key2);
    int           lorder;
} BTREEINFO;

and the elements of this structure are as follows, one by one, no skipping, we're getting through all of them together bestie 🫶

• flags — the flag value is specified by ORing any of the following values:
  • R_DUP — permit duplicate keys in the tree, that is, permit insertion if the key to be inserted already exists in the tree. the default behavior, as described in dbopen(3), is to overwrite a matching key when inserting a new key, or to fail if the R_NOOVERWRITE flag is specified. heads up: the R_DUP flag is overridden by the R_NOOVERWRITE flag, and if the R_NOOVERWRITE flag is specified, attempts to insert duplicate keys into the tree will fail. 💀

and if the database contains duplicate keys, the order of retrieval of key/data pairs is undefined if the get routine is used — like, genuinely undefined, don't @ me. however, seq routine calls with the R_CURSOR flag set will always return the logical "first" of any group of duplicate keys. so there's that. 🙈

• cachesize — a suggested maximum size (in bytes) of the memory cache. key word: suggested. this value is only advisory, and the access method will allocate more memory rather than fail. queen behavior honestly, she doesn't give up. since every search examines the root page of the tree, caching the most recently used pages substantially improves access time. in addition, physical writes are delayed as long as possible, so a moderate cache can reduce the number of I/O operations significantly. 🔥

obviously, using a cache increases (but only increases) the likelihood of corruption or lost data if the system crashes while a tree is being modified. literally only increases. that's the whole direction. 😭

if cachesize is 0 (no size is specified), a default cache is used.

• maxkeypage — the maximum number of keys which will be stored on any single page. not currently implemented. that's it. that's the whole thing. girl is not implemented. 💀
• minkeypage — the minimum number of keys which will be stored on any single page. this value is used to determine which keys will be stored on overflow pages, that is, if a key or data item is longer than the pagesize divided by the minkeypage value, it will be stored on overflow pages instead of in the page itself. if minkeypage is 0 (no minimum number of keys is specified), a value of 2 is used. simple as that bestie.
• psize — page size is the size (in bytes) of the pages used for nodes in the tree. the minimum page size is 512 bytes and the maximum page size is 64 KiB. if psize is 0 (no page size is specified), a page size is chosen based on the underlying filesystem I/O block size. she'll pick for you. 💅
• compare — compare is the key comparison function. it must return an integer less than, equal to, or greater than zero if the first key argument is considered to be respectively less than, equal to, or greater than the second key argument. the same comparison function must be used on a given tree every time it is opened — like, consistency is the whole point, don't switch up on her. if compare is NULL (no comparison function is specified), the keys are compared lexically, with shorter keys considered less than longer keys.
• prefix — prefix is the prefix comparison function. if specified, this routine must return the number of bytes of the second key argument which are necessary to determine that it is greater than the first key argument. if the keys are equal, the key length should be returned. note, the usefulness of this routine is very data-dependent, but, in some data sets it can produce significantly reduced tree sizes and search times. ✨ tiny trees, fast searches, we love to see it. if prefix is NULL (no prefix function is specified), and no comparison function is specified, a default lexical comparison routine is used. if prefix is NULL and a comparison routine is specified, no prefix comparison is done. okay? okay!
• lorder — the byte order for integers in the stored database metadata. the number should represent the order as an integer; for example, big endian order would be the number 4,321. if lorder is 0 (no order is specified), the current host order is used. so like, whatever the host woke up feeling today. 🎀

---

if the file already exists (and the O_TRUNC flag is not specified), the values specified for the arguments flags, lorder, and psize are ignored in favor of the values used when the tree was created. like, the tree already decided. respect her choices tbh.

forward sequential scans of a tree are from the least key to the greatest. that's the direction. always. 👑

space freed up by deleting key/data pairs from the tree is never reclaimed, although it is normally made available for reuse. this means that the btree storage structure is grow-only. she only goes up. the only solutions are to avoid excessive deletions, or to create a fresh tree periodically from a scan of an existing one. like a little glow-up ritual. 🧚

searches, insertions, and deletions in a btree will all complete in O lg base N where base is the average fill factor. often, inserting ordered data into btrees results in a low fill factor. but this implementation has been modified to make ordered insertion the best case, resulting in a much better than normal page fill factor. ate and left no crumbs honestly. 🔥

---

😭 ERRORS 😭

the btree access method routines may fail and set errno for any of the errors specified for the library routine dbopen(3). so like, it's not being mysterious, it's just redirecting you. go read dbopen(3). 💜

💀 BUGS 💀

only big and little endian byte order is supported. that's the bug. that's it. only those two. ngl kind of a limitation but also kind of a whole mood.

⭐ SEE ALSO ⭐

• dbopen(3)
• hash(3)
• mpool(3)
• recno(3)

The Ubiquitous B-tree, Douglas Comer, ACM Comput. Surv. 11, 2 (June 1979), 121-138.

Prefix B-trees, Bayer and Unterauer, ACM Transactions on Database Systems, Vol. 2, 1 (March 1977), 11-26.

The Art of Computer Programming Vol. 3: Sorting and Searching, D.E. Knuth, 1968, pp 471-480.

---

and that's the page bestie, fr fr — this one's from Linux man-pages 6.20, dated 2025-09-21, and it's btree(3) if you somehow forgot. go forth and be sorted and balanced. mwah. 💖💅🦋