// SPDX-License-Identifier: GPL-2.0-only /* * This file is part of UBIFS. * * Copyright (C) 2006-2008 Nokia Corporation. * * Authors: Adrian Hunter * Artem Bityutskiy (Битюцкий Артём) */ /* * This file implements garbage collection. The procedure for garbage collection * is different depending on whether a LEB as an index LEB (contains index * nodes) or not. For non-index LEBs, garbage collection finds a LEB which * contains a lot of dirty space (obsolete nodes), and copies the non-obsolete * nodes to the journal, at which point the garbage-collected LEB is free to be * reused. For index LEBs, garbage collection marks the non-obsolete index nodes * dirty in the TNC, and after the next commit, the garbage-collected LEB is * to be reused. Garbage collection will cause the number of dirty index nodes * to grow, however sufficient space is reserved for the index to ensure the * commit will never run out of space. * * Notes about dead watermark. At current UBIFS implementation we assume that * LEBs which have less than @c->dead_wm bytes of free + dirty space are full * and not worth garbage-collecting. The dead watermark is one min. I/O unit * size, or min. UBIFS node size, depending on what is greater. Indeed, UBIFS * Garbage Collector has to synchronize the GC head's write buffer before * returning, so this is about wasting one min. I/O unit. However, UBIFS GC can * actually reclaim even very small pieces of dirty space by garbage collecting * enough dirty LEBs, but we do not bother doing this at this implementation. * * Notes about dark watermark. The results of GC work depends on how big are * the UBIFS nodes GC deals with. Large nodes make GC waste more space. Indeed, * if GC move data from LEB A to LEB B and nodes in LEB A are large, GC would * have to waste large pieces of free space at the end of LEB B, because nodes * from LEB A would not fit. And the worst situation is when all nodes are of * maximum size. So dark watermark is the amount of free + dirty space in LEB * which are guaranteed to be reclaimable. If LEB has less space, the GC might * be unable to reclaim it. So, LEBs with free + dirty greater than dark * watermark are "good" LEBs from GC's point of view. The other LEBs are not so * good, and GC takes extra care when moving them. */ #include <linux/slab.h> #include <linux/pagemap.h> #include <linux/list_sort.h> #include "ubifs.h" /* * GC may need to move more than one LEB to make progress. The below constants * define "soft" and "hard" limits on the number of LEBs the garbage collector * may move. */ #define SOFT_LEBS_LIMIT … #define HARD_LEBS_LIMIT … /** * switch_gc_head - switch the garbage collection journal head. * @c: UBIFS file-system description object * * This function switch the GC head to the next LEB which is reserved in * @c->gc_lnum. Returns %0 in case of success, %-EAGAIN if commit is required, * and other negative error code in case of failures. */ static int switch_gc_head(struct ubifs_info *c) { … } /** * data_nodes_cmp - compare 2 data nodes. * @priv: UBIFS file-system description object * @a: first data node * @b: second data node * * This function compares data nodes @a and @b. Returns %1 if @a has greater * inode or block number, and %-1 otherwise. */ static int data_nodes_cmp(void *priv, const struct list_head *a, const struct list_head *b) { … } /* * nondata_nodes_cmp - compare 2 non-data nodes. * @priv: UBIFS file-system description object * @a: first node * @a: second node * * This function compares nodes @a and @b. It makes sure that inode nodes go * first and sorted by length in descending order. Directory entry nodes go * after inode nodes and are sorted in ascending hash valuer order. */ static int nondata_nodes_cmp(void *priv, const struct list_head *a, const struct list_head *b) { … } /** * sort_nodes - sort nodes for GC. * @c: UBIFS file-system description object * @sleb: describes nodes to sort and contains the result on exit * @nondata: contains non-data nodes on exit * @min: minimum node size is returned here * * This function sorts the list of inodes to garbage collect. First of all, it * kills obsolete nodes and separates data and non-data nodes to the * @sleb->nodes and @nondata lists correspondingly. * * Data nodes are then sorted in block number order - this is important for * bulk-read; data nodes with lower inode number go before data nodes with * higher inode number, and data nodes with lower block number go before data * nodes with higher block number; * * Non-data nodes are sorted as follows. * o First go inode nodes - they are sorted in descending length order. * o Then go directory entry nodes - they are sorted in hash order, which * should supposedly optimize 'readdir()'. Direntry nodes with lower parent * inode number go before direntry nodes with higher parent inode number, * and direntry nodes with lower name hash values go before direntry nodes * with higher name hash values. * * This function returns zero in case of success and a negative error code in * case of failure. */ static int sort_nodes(struct ubifs_info *c, struct ubifs_scan_leb *sleb, struct list_head *nondata, int *min) { … } /** * move_node - move a node. * @c: UBIFS file-system description object * @sleb: describes the LEB to move nodes from * @snod: the mode to move * @wbuf: write-buffer to move node to * * This function moves node @snod to @wbuf, changes TNC correspondingly, and * destroys @snod. Returns zero in case of success and a negative error code in * case of failure. */ static int move_node(struct ubifs_info *c, struct ubifs_scan_leb *sleb, struct ubifs_scan_node *snod, struct ubifs_wbuf *wbuf) { … } /** * move_nodes - move nodes. * @c: UBIFS file-system description object * @sleb: describes the LEB to move nodes from * * This function moves valid nodes from data LEB described by @sleb to the GC * journal head. This function returns zero in case of success, %-EAGAIN if * commit is required, and other negative error codes in case of other * failures. */ static int move_nodes(struct ubifs_info *c, struct ubifs_scan_leb *sleb) { … } /** * gc_sync_wbufs - sync write-buffers for GC. * @c: UBIFS file-system description object * * We must guarantee that obsoleting nodes are on flash. Unfortunately they may * be in a write-buffer instead. That is, a node could be written to a * write-buffer, obsoleting another node in a LEB that is GC'd. If that LEB is * erased before the write-buffer is sync'd and then there is an unclean * unmount, then an existing node is lost. To avoid this, we sync all * write-buffers. * * This function returns %0 on success or a negative error code on failure. */ static int gc_sync_wbufs(struct ubifs_info *c) { … } /** * ubifs_garbage_collect_leb - garbage-collect a logical eraseblock. * @c: UBIFS file-system description object * @lp: describes the LEB to garbage collect * * This function garbage-collects an LEB and returns one of the @LEB_FREED, * @LEB_RETAINED, etc positive codes in case of success, %-EAGAIN if commit is * required, and other negative error codes in case of failures. */ int ubifs_garbage_collect_leb(struct ubifs_info *c, struct ubifs_lprops *lp) { … } /** * ubifs_garbage_collect - UBIFS garbage collector. * @c: UBIFS file-system description object * @anyway: do GC even if there are free LEBs * * This function does out-of-place garbage collection. The return codes are: * o positive LEB number if the LEB has been freed and may be used; * o %-EAGAIN if the caller has to run commit; * o %-ENOSPC if GC failed to make any progress; * o other negative error codes in case of other errors. * * Garbage collector writes data to the journal when GC'ing data LEBs, and just * marking indexing nodes dirty when GC'ing indexing LEBs. Thus, at some point * commit may be required. But commit cannot be run from inside GC, because the * caller might be holding the commit lock, so %-EAGAIN is returned instead; * And this error code means that the caller has to run commit, and re-run GC * if there is still no free space. * * There are many reasons why this function may return %-EAGAIN: * o the log is full and there is no space to write an LEB reference for * @c->gc_lnum; * o the journal is too large and exceeds size limitations; * o GC moved indexing LEBs, but they can be used only after the commit; * o the shrinker fails to find clean znodes to free and requests the commit; * o etc. * * Note, if the file-system is close to be full, this function may return * %-EAGAIN infinitely, so the caller has to limit amount of re-invocations of * the function. E.g., this happens if the limits on the journal size are too * tough and GC writes too much to the journal before an LEB is freed. This * might also mean that the journal is too large, and the TNC becomes to big, * so that the shrinker is constantly called, finds not clean znodes to free, * and requests commit. Well, this may also happen if the journal is all right, * but another kernel process consumes too much memory. Anyway, infinite * %-EAGAIN may happen, but in some extreme/misconfiguration cases. */ int ubifs_garbage_collect(struct ubifs_info *c, int anyway) { … } /** * ubifs_gc_start_commit - garbage collection at start of commit. * @c: UBIFS file-system description object * * If a LEB has only dirty and free space, then we may safely unmap it and make * it free. Note, we cannot do this with indexing LEBs because dirty space may * correspond index nodes that are required for recovery. In that case, the * LEB cannot be unmapped until after the next commit. * * This function returns %0 upon success and a negative error code upon failure. */ int ubifs_gc_start_commit(struct ubifs_info *c) { … } /** * ubifs_gc_end_commit - garbage collection at end of commit. * @c: UBIFS file-system description object * * This function completes out-of-place garbage collection of index LEBs. */ int ubifs_gc_end_commit(struct ubifs_info *c) { … } /** * ubifs_destroy_idx_gc - destroy idx_gc list. * @c: UBIFS file-system description object * * This function destroys the @c->idx_gc list. It is called when unmounting * so locks are not needed. Returns zero in case of success and a negative * error code in case of failure. */ void ubifs_destroy_idx_gc(struct ubifs_info *c) { … } /** * ubifs_get_idx_gc_leb - get a LEB from GC'd index LEB list. * @c: UBIFS file-system description object * * Called during start commit so locks are not needed. */ int ubifs_get_idx_gc_leb(struct ubifs_info *c) { … }
Codebase Browser
linux