navidrome/persistence/sql_base_repository.go

package persistence

import (
	"context"
	"database/sql"
	"errors"
	"fmt"
	"iter"
	"reflect"
	"regexp"
	"strings"
	"time"

	. "github.com/Masterminds/squirrel"
	"github.com/deluan/rest"
	"github.com/navidrome/navidrome/conf"
	"github.com/navidrome/navidrome/db"
	"github.com/navidrome/navidrome/log"
	"github.com/navidrome/navidrome/model"
	id2 "github.com/navidrome/navidrome/model/id"
	"github.com/navidrome/navidrome/model/request"
	"github.com/navidrome/navidrome/utils/hasher"
	"github.com/navidrome/navidrome/utils/slice"
	"github.com/pocketbase/dbx"
	"github.com/zeebo/xxh3"
)

// sqlRepository is the base repository for all SQL repositories. It provides common functions to interact with the DB.
// When creating a new repository using this base, you must:
//
//   - Embed this struct.
//   - Set ctx and db fields. ctx should be the context passed to the constructor method, usually obtained from the request
//   - Call registerModel with the model instance and any possible filters.
//   - If the model has a different table name than the default (lowercase of the model name), it should be set manually
//     using the tableName field.
//   - Sort mappings must be set with setSortMappings method. If a sort field is not in the map, it will be used as the name of the column.
//
// All fields in filters and sortMappings must be in snake_case. Only sorts and filters based on real field names or
// defined in the mappings will be allowed.
type sqlRepository struct {
	ctx       context.Context
	tableName string
	db        dbx.Builder

	// Do not set these fields manually, they are set by the registerModel method
	filterMappings     map[string]filterFunc
	isFieldWhiteListed fieldWhiteListedFunc
	// Do not set this field manually, it is set by the setSortMappings method
	sortMappings map[string]string
}

const invalidUserId = "-1"

func loggedUser(ctx context.Context) *model.User {
	if user, ok := request.UserFrom(ctx); !ok {
		return &model.User{ID: invalidUserId}
	} else {
		return &user
	}
}

// ownerFilter returns the predicate restricting access to rows owned by the logged-in user, for
// tables with a user_id column. It returns nil for admins and for headless/system contexts (invalid
// user), meaning "no ownership restriction". Callers should skip the WHERE clause when it is nil.
//
// The predicate uses an unqualified user_id, so it only works on queries where that column is
// unambiguous (no join introducing a second user_id).
func (r sqlRepository) ownerFilter() Sqlizer {
	if usr := loggedUser(r.ctx); !usr.IsAdmin && usr.ID != invalidUserId {
		return Eq{"user_id": usr.ID}
	}
	return nil
}

// addRestriction combines an optional caller predicate with the ownership filter, producing the
// WHERE clause for owner-scoped reads. For admins and headless contexts ownerFilter() is nil and
// only the caller's predicate (if any) remains.
func (r sqlRepository) addRestriction(sql ...Sqlizer) Sqlizer {
	s := And{}
	if len(sql) > 0 {
		s = append(s, sql[0])
	}
	if owner := r.ownerFilter(); owner != nil {
		s = append(s, owner)
	}
	return s
}

func (r *sqlRepository) registerModel(instance any, filters map[string]filterFunc) {
	if r.tableName == "" {
		r.tableName = strings.TrimPrefix(reflect.TypeOf(instance).String(), "*model.")
		r.tableName = toSnakeCase(r.tableName)
	}
	r.tableName = strings.ToLower(r.tableName)
	r.isFieldWhiteListed = registerModelWhiteList(instance)
	r.filterMappings = filters
}

// setSortMappings sets the mappings for the sort fields. If the sort field is not in the map, it will be used as is.
// This applies per comma-separated part, so a key added here also defines that bare name wherever a
// caller uses it inside a sort list.
//
// If PreferSortTags is enabled, it will map the order fields to the corresponding sort expression,
// which gives precedence to sort tags.
// Ex: order_title => (coalesce(nullif(sort_title,”),order_title) collate nocase)
// To avoid performance issues, indexes should be created for these sort expressions
//
// NOTE: if an individual item has spaces, it should be wrapped in parentheses. For example,
// you should write "(lyrics != '[]')". This prevents the item being split unexpectedly.
// Without parentheses, "lyrics != '[]'" would be mapped as simply "lyrics"
func (r *sqlRepository) setSortMappings(mappings map[string]string, tableName ...string) {
	tn := r.tableName
	if len(tableName) > 0 {
		tn = tableName[0]
	}
	if conf.Server.PreferSortTags || conf.Server.EnableNaturalSorting {
		for k, v := range mappings {
			mappings[k] = mapSortOrder(tn, v)
		}
	}
	r.sortMappings = mappings
}

func (r sqlRepository) newSelect(options ...model.QueryOptions) SelectBuilder {
	sq := Select().From(r.tableName)
	if len(options) > 0 {
		r.resetSeededRandom(options)
		sq = r.applyOptions(sq, options...)
		sq = r.applyFilters(sq, options...)
	}
	return sq
}

func (r sqlRepository) applyOptions(sq SelectBuilder, options ...model.QueryOptions) SelectBuilder {
	if len(options) > 0 {
		if options[0].Max > 0 {
			sq = sq.Limit(uint64(options[0].Max))
		}
		if options[0].Offset > 0 {
			sq = sq.Offset(uint64(options[0].Offset))
		}
		if options[0].Sort != "" {
			sq = sq.OrderBy(r.buildSortOrder(options[0].Sort, options[0].Order))
		}
	}
	return sq
}

// TODO Change all sortMappings to have a consistent case
func (r sqlRepository) sortMapping(sort string) string {
	if mapping, _, ok := r.lookupSortMapping(sort); ok {
		return mapping
	}
	// Each part of a comma list is resolved on its own, so a mix of mapped keys and plain columns
	// keeps the mappings the recognized parts have.
	parts := strings.FieldsFunc(sort, splitFunc(','))
	mapped := make([]string, 0, len(parts))
	for _, part := range parts {
		part = strings.TrimSpace(part)
		if partMapping, _, ok := r.lookupSortMapping(part); ok {
			part = partMapping
		} else {
			part = toSnakeCase(part)
		}
		mapped = append(mapped, part)
	}
	return strings.Join(mapped, ", ")
}

// lookupSortMapping also returns the snake_case form when it had to derive one, so a caller's
// fallback doesn't recompute it: toSnakeCase runs two regexps.
func (r sqlRepository) lookupSortMapping(sort string) (mapping, snakeCased string, ok bool) {
	if mapping, ok = r.sortMappings[sort]; ok {
		return mapping, sort, true
	}
	if mapping, ok = r.sortMappings[toCamelCase(sort)]; ok {
		return mapping, "", true
	}
	snakeCased = toSnakeCase(sort)
	mapping, ok = r.sortMappings[snakeCased]
	return mapping, snakeCased, ok
}

func (r sqlRepository) buildSortOrder(sort, order string) string {
	sort = r.sortMapping(sort)
	order = strings.ToLower(strings.TrimSpace(order))
	var reverseOrder string
	if order == "desc" {
		reverseOrder = "asc"
	} else {
		order = "asc"
		reverseOrder = "desc"
	}

	parts := strings.FieldsFunc(sort, splitFunc(','))
	newSort := make([]string, 0, len(parts))
	for _, p := range parts {
		f := strings.FieldsFunc(p, splitFunc(' '))
		newField := make([]string, 1, len(f))
		newField[0] = f[0]
		if len(f) == 1 {
			newField = append(newField, order)
		} else {
			if f[1] == "asc" {
				newField = append(newField, order)
			} else {
				newField = append(newField, reverseOrder)
			}
		}
		newSort = append(newSort, strings.Join(newField, " "))
	}
	return strings.Join(newSort, ", ")
}

func splitFunc(delimiter rune) func(c rune) bool {
	open := 0
	return func(c rune) bool {
		if c == '(' {
			open++
			return false
		}
		if open > 0 {
			if c == ')' {
				open--
			}
			return false
		}
		return c == delimiter
	}
}

func (r sqlRepository) applyFilters(sq SelectBuilder, options ...model.QueryOptions) SelectBuilder {
	if len(options) > 0 && options[0].Filters != nil {
		sq = sq.Where(options[0].Filters)
	}
	return sq
}

// libraryIdFilter is a filter function to be added to resources that have a library_id column.
func libraryIdFilter(_ string, value any) Sqlizer {
	return Eq{"library_id": value}
}

// applyLibraryFilter adds library filtering to queries for tables that have a library_id column
// This ensures users only see content from libraries they have access to
func (r sqlRepository) applyLibraryFilter(sq SelectBuilder, tableName ...string) SelectBuilder {
	user := loggedUser(r.ctx)

	// If the user is an admin, or the user ID is invalid (e.g., when no user is logged in), skip the library filter
	if user.IsAdmin || user.ID == invalidUserId {
		return sq
	}

	// A non-admin granted every library sees everything the subquery would return, so applying it is
	// pure overhead. Skip it in that case (same fast path admins get).
	if visible, err := r.visibleLibraryIDs(); err == nil && r.userSeesAllLibraries(visible) {
		return sq
	}

	table := r.tableName
	if len(tableName) > 0 {
		table = tableName[0]
	}

	// Get user's accessible library IDs
	// Use subquery to filter by user's library access
	return sq.Where(Expr(table+".library_id IN ("+
		"SELECT ul.library_id FROM user_library ul WHERE ul.user_id = ?)", user.ID))
}

// userSeesAllLibraries reports whether the visible set already covers every library, so a
// library filter would exclude nothing.
func (r sqlRepository) userSeesAllLibraries(visible []int) bool {
	user := loggedUser(r.ctx)
	if user.IsAdmin || user.ID == invalidUserId {
		return true // visible is the whole library table
	}
	total, err := NewLibraryRepository(r.ctx, r.db).CountAll()
	if err != nil || total == 0 {
		return false
	}
	return int64(len(visible)) == total
}

// visibleLibraryIDs returns the libraries the current user can see: all libraries for admin and
// headless processes, otherwise the user's granted libraries.
func (r sqlRepository) visibleLibraryIDs() ([]int, error) {
	user := loggedUser(r.ctx)
	if user.IsAdmin || user.ID == invalidUserId {
		var ids []int
		err := r.queryAllSlice(Select("id").From("library"), &ids)
		return ids, err
	}
	return slice.Map(user.Libraries, func(lib model.Library) int { return lib.ID }), nil
}

func (r sqlRepository) seedKey() string {
	// Seed keys must be all lowercase, or else SQLite3 will encode it, making it not match the seed
	// used in the query. Hashing the user ID and converting it to a hex string will do the trick
	userIDHash := xxh3.Hash([]byte(loggedUser(r.ctx).ID))
	return fmt.Sprintf("%s|%016x", r.tableName, userIDHash)
}

func (r sqlRepository) resetSeededRandom(options []model.QueryOptions) {
	if len(options) == 0 || options[0].Sort != "random" {
		return
	}
	// CAST: playlist_tracks.id is an INTEGER (unlike other tables' TEXT ids); passing it to
	// SEEDEDRAND's string param uncast silently drops every row (go-sqlite3 binding gotcha).
	options[0].Sort = fmt.Sprintf("SEEDEDRAND('%s', CAST(%s.id AS TEXT))", r.seedKey(), r.tableName)
	if options[0].Seed != "" {
		hasher.SetSeed(r.seedKey(), options[0].Seed)
		return
	}
	if options[0].Offset == 0 {
		hasher.Reseed(r.seedKey())
	}
}

func (r sqlRepository) executeSQL(sq Sqlizer) (int64, error) {
	query, args, err := r.toSQL(sq)
	if err != nil {
		return 0, err
	}
	start := time.Now()
	var c int64
	res, err := r.db.NewQuery(query).Bind(args).WithContext(r.ctx).Execute()
	if res != nil {
		c, _ = res.RowsAffected()
	}
	r.logSQL(query, args, err, c, start)
	if err != nil {
		if err.Error() != "LastInsertId is not supported by this driver" {
			return 0, err
		}
	}
	return c, err
}

var placeholderRegex = regexp.MustCompile(`\?`)

func (r sqlRepository) toSQL(sq Sqlizer) (string, dbx.Params, error) {
	query, args, err := sq.ToSql()
	if err != nil {
		return "", nil, err
	}
	// Replace query placeholders with named params
	params := make(dbx.Params, len(args))
	counter := 0
	result := placeholderRegex.ReplaceAllStringFunc(query, func(_ string) string {
		p := fmt.Sprintf("p%d", counter)
		params[p] = args[counter]
		counter++
		return "{:" + p + "}"
	})
	return result, params, nil
}

func (r sqlRepository) queryOne(sq Sqlizer, response any) error {
	query, args, err := r.toSQL(sq)
	if err != nil {
		return err
	}
	start := time.Now()
	err = r.db.NewQuery(query).Bind(args).WithContext(r.ctx).One(response)
	if errors.Is(err, sql.ErrNoRows) {
		r.logSQL(query, args, nil, 0, start)
		return model.ErrNotFound
	}
	r.logSQL(query, args, err, 1, start)
	return err
}

// wrapCursor adapts a cursor over db rows into one over their models. toModel pulls out the row's
// embedded model, which a type parameter can't reach on its own.
func wrapCursor[D, T any](cursor iter.Seq2[D, error], toModel func(D) *T) iter.Seq2[T, error] {
	return func(yield func(T, error) bool) {
		for row, err := range cursor {
			m := toModel(row)
			if m == nil {
				var zero T
				yield(zero, fmt.Errorf("unexpected nil %T (%v): %w", zero, row, err))
				return
			}
			if !yield(*m, err) || err != nil {
				return
			}
		}
	}
}

// queryWithStableResults is a helper function to execute a query and return an iterator that will yield its results
// from a cursor, guaranteeing that the results will be stable, even if the underlying data changes.
func queryWithStableResults[T any](r sqlRepository, sq SelectBuilder, options ...model.QueryOptions) (iter.Seq2[T, error], error) {
	if len(options) > 0 && options[0].Offset > 0 {
		sq = r.optimizePagination(sq, options[0])
	}
	query, args, err := r.toSQL(sq)
	if err != nil {
		return nil, err
	}
	start := time.Now()
	rows, err := r.db.NewQuery(query).Bind(args).WithContext(r.ctx).Rows()
	r.logSQL(query, args, err, -1, start)
	if err != nil {
		return nil, err
	}
	return func(yield func(T, error) bool) {
		defer rows.Close()
		for rows.Next() {
			var row T
			err := rows.ScanStruct(&row)
			if !yield(row, err) || err != nil {
				return
			}
		}
		if err := rows.Err(); err != nil {
			var empty T
			yield(empty, err)
		}
	}, nil
}

func (r sqlRepository) queryAll(sq SelectBuilder, response any, options ...model.QueryOptions) error {
	if len(options) > 0 && options[0].Offset > 0 {
		sq = r.optimizePagination(sq, options[0])
	}
	query, args, err := r.toSQL(sq)
	if err != nil {
		return err
	}
	start := time.Now()
	err = r.db.NewQuery(query).Bind(args).WithContext(r.ctx).All(response)
	if errors.Is(err, sql.ErrNoRows) {
		r.logSQL(query, args, nil, -1, start)
		return model.ErrNotFound
	}
	r.logSQL(query, args, err, int64(reflect.ValueOf(response).Elem().Len()), start)
	return err
}

// queryAllSlice is a helper function to query a single column and return the result in a slice
func (r sqlRepository) queryAllSlice(sq SelectBuilder, response any) error {
	query, args, err := r.toSQL(sq)
	if err != nil {
		return err
	}
	start := time.Now()
	err = r.db.NewQuery(query).Bind(args).WithContext(r.ctx).Column(response)
	if errors.Is(err, sql.ErrNoRows) {
		r.logSQL(query, args, nil, -1, start)
		return model.ErrNotFound
	}
	r.logSQL(query, args, err, int64(reflect.ValueOf(response).Elem().Len()), start)
	return err
}

// optimizePagination uses a less inefficient pagination, by not using OFFSET.
// See https://gist.github.com/ssokolow/262503
func (r sqlRepository) optimizePagination(sq SelectBuilder, options model.QueryOptions) SelectBuilder {
	if options.Offset > conf.Server.DevOffsetOptimize {
		sq = sq.RemoveOffset()
		rowidSq := sq.RemoveColumns().Columns(r.tableName + ".rowid")
		rowidSq = rowidSq.Limit(uint64(options.Offset))
		rowidSql, args, _ := rowidSq.ToSql()
		sq = sq.Where(r.tableName+".rowid not in ("+rowidSql+")", args...)
	}
	return sq
}

func (r sqlRepository) exists(cond Sqlizer) (bool, error) {
	existsQuery := Select("count(*) as exist").From(r.tableName).Where(cond)
	var res struct{ Exist int64 }
	err := r.queryOne(existsQuery, &res)
	return res.Exist > 0, err
}

// updateOwned performs an atomic, ownership-restricted update of the row identified by id, for
// repositories whose table has a user_id column. Non-admins can only update rows they own: the
// ownership predicate is part of the UPDATE's WHERE clause, so a row owned by another user simply
// does not match and no write happens. Ownership itself is immutable here: user_id is never written,
// so no caller (admin included) can reassign a row to a different owner via an update. Unlike put,
// it never falls through to an INSERT, so a non-matching id never creates a row.
//
// When the update matches no row it classifies the failure: if the row exists but is owned by
// another user it returns rest.ErrPermissionDenied, otherwise rest.ErrNotFound. The write itself is
// still atomic; the extra lookup happens only on the failure path (count == 0), where no write
// occurred, so there is no TOCTOU on the update.
func (r sqlRepository) updateOwned(id string, m any, colsToUpdate ...string) error {
	values, err := toSQLArgs(m)
	if err != nil {
		return fmt.Errorf("error preparing values to write to DB: %w", err)
	}
	updateValues := filterUpdateValues(values, id, colsToUpdate...)
	delete(updateValues, "user_id") // ownership is immutable on update
	update := Update(r.tableName).Where(r.addRestriction(Eq{"id": id})).SetMap(updateValues)
	count, err := r.executeSQL(update)
	if err != nil {
		return err
	}
	if count == 0 {
		return r.classifyOwnedWriteMiss(id)
	}
	return nil
}

// deleteOwned performs an atomic, ownership-restricted delete of the row identified by id, for
// repositories whose table has a user_id column. Non-admins can only delete rows they own: the
// ownership predicate is part of the DELETE's WHERE clause, so a row owned by another user simply
// does not match and is left untouched. The failure path mirrors updateOwned (see
// classifyOwnedWriteMiss), so there is no TOCTOU on the delete.
func (r sqlRepository) deleteOwned(id string) error {
	count, err := r.executeSQL(Delete(r.tableName).Where(r.addRestriction(Eq{"id": id})))
	if err != nil {
		return err
	}
	if count == 0 {
		return r.classifyOwnedWriteMiss(id)
	}
	return nil
}

// classifyOwnedWriteMiss explains why an ownership-filtered write (updateOwned/deleteOwned) matched
// no row: rest.ErrPermissionDenied if the row exists but is owned by another user, otherwise
// rest.ErrNotFound. It runs only on the failure path (count == 0), where no write occurred.
func (r sqlRepository) classifyOwnedWriteMiss(id string) error {
	exists, err := r.exists(Eq{"id": id})
	if err != nil {
		return err
	}
	if exists {
		return rest.ErrPermissionDenied
	}
	return rest.ErrNotFound
}

func (r sqlRepository) count(countQuery SelectBuilder, options ...model.QueryOptions) (int64, error) {
	countQuery = countQuery.
		RemoveColumns().Columns("count(distinct " + r.tableName + ".id) as count").
		RemoveOffset().RemoveLimit().
		OrderBy(r.tableName + ".id"). // To remove any ORDER BY clause that could slow down the query
		From(r.tableName)
	countQuery = r.applyFilters(countQuery, options...)
	var res struct{ Count int64 }
	err := r.queryOne(countQuery, &res)
	return res.Count, err
}

func (r sqlRepository) putByMatch(filter Sqlizer, id string, m any, colsToUpdate ...string) (string, error) {
	if id != "" {
		return r.put(id, m, colsToUpdate...)
	}
	existsQuery := r.newSelect().Columns("id").From(r.tableName).Where(filter)

	var res struct{ ID string }
	err := r.queryOne(existsQuery, &res)
	if err != nil && !errors.Is(err, model.ErrNotFound) {
		return "", err
	}
	return r.put(res.ID, m, colsToUpdate...)
}

// filterUpdateValues selects, from a marshaled column map, the values to write in an UPDATE on the
// row identified by id: only the requested colsToUpdate (or all columns when none are specified),
// dropping columns that must never be overwritten on update (created_at, birth_time).
func filterUpdateValues(values map[string]any, id string, colsToUpdate ...string) map[string]any {
	updateValues := map[string]any{}

	// This is a map of the columns that need to be updated, if specified
	c2upd := slice.ToMap(colsToUpdate, func(s string) (string, struct{}) {
		return toSnakeCase(s), struct{}{}
	})
	for k, v := range values {
		if _, found := c2upd[k]; len(c2upd) == 0 || found {
			updateValues[k] = v
		}
	}

	updateValues["id"] = id
	delete(updateValues, "created_at")
	// To avoid updating the media_file birth_time on each scan. Not the best solution, but it works for now
	// TODO move to mediafile_repository when each repo has its own upsert method
	delete(updateValues, "birth_time")
	return updateValues
}

func (r sqlRepository) put(id string, m any, colsToUpdate ...string) (newId string, err error) {
	values, err := toSQLArgs(m)
	if err != nil {
		return "", fmt.Errorf("error preparing values to write to DB: %w", err)
	}
	// If there's an ID, try to update first
	if id != "" {
		update := Update(r.tableName).Where(Eq{"id": id}).SetMap(filterUpdateValues(values, id, colsToUpdate...))
		count, err := r.executeSQL(update)
		if err != nil {
			return "", err
		}
		if count > 0 {
			return id, nil
		}
	}
	// If it does not have an ID OR the ID was not found (when it is a new record with predefined id)
	if id == "" {
		id = id2.NewRandom()
		values["id"] = id
	}
	insert := Insert(r.tableName).SetMap(values)
	_, err = r.executeSQL(insert)
	return id, err
}

func (r sqlRepository) delete(cond Sqlizer) error {
	del := Delete(r.tableName).Where(cond)
	_, err := r.executeSQL(del)
	if errors.Is(err, sql.ErrNoRows) {
		return model.ErrNotFound
	}
	return err
}

func (r sqlRepository) logSQL(sql string, args dbx.Params, err error, rowsAffected int64, start time.Time) {
	elapsed := time.Since(start)
	fields := []any{r.ctx, "SQL: `" + sql + "`", "args", args, "rowsAffected", rowsAffected, "elapsedTime", elapsed}
	if err == nil || errors.Is(err, context.Canceled) {
		log.Trace(append(fields, err)...)
		return
	}
	// The result codes separate errors that share a message, notably SQLITE_BUSY from
	// SQLITE_BUSY_SNAPSHOT, which no busy_timeout can retry.
	if code, extended, ok := db.ErrorCodes(err); ok {
		fields = append(fields, "sqliteCode", code, "sqliteExtended", extended)
	}
	log.Error(append(fields, err)...)
}