From e05c378f4973674a16d5b9636f2310cf88aca5f2 Mon Sep 17 00:00:00 2001 From: Joern Engel Date: Tue, 30 Mar 2010 18:25:17 +0200 Subject: [LogFS] Remove unused method All callers are long gone. Signed-off-by: Joern Engel --- fs/logfs/readwrite.c | 21 --------------------- 1 file changed, 21 deletions(-) (limited to 'fs/logfs/readwrite.c') diff --git a/fs/logfs/readwrite.c b/fs/logfs/readwrite.c index c3a3a6814b84..3659c37fbd72 100644 --- a/fs/logfs/readwrite.c +++ b/fs/logfs/readwrite.c @@ -429,25 +429,6 @@ static void inode_write_block(struct logfs_block *block) } } -static gc_level_t inode_block_level(struct logfs_block *block) -{ - BUG_ON(block->inode->i_ino == LOGFS_INO_MASTER); - return GC_LEVEL(LOGFS_MAX_LEVELS); -} - -static gc_level_t indirect_block_level(struct logfs_block *block) -{ - struct page *page; - struct inode *inode; - u64 bix; - level_t level; - - page = block->page; - inode = page->mapping->host; - logfs_unpack_index(page->index, &bix, &level); - return expand_level(inode->i_ino, level); -} - /* * This silences a false, yet annoying gcc warning. I hate it when my editor * jumps into bitops.h each time I recompile this file. @@ -586,14 +567,12 @@ static void indirect_free_block(struct super_block *sb, static struct logfs_block_ops inode_block_ops = { .write_block = inode_write_block, - .block_level = inode_block_level, .free_block = inode_free_block, .write_alias = inode_write_alias, }; struct logfs_block_ops indirect_block_ops = { .write_block = indirect_write_block, - .block_level = indirect_block_level, .free_block = indirect_free_block, .write_alias = indirect_write_alias, }; -- cgit v1.2.3 From 032d8f7268444a0f5d4ee02d9513d682d5b8edfc Mon Sep 17 00:00:00 2001 From: Joern Engel Date: Tue, 13 Apr 2010 17:46:37 +0200 Subject: [LogFS] Prevent memory corruption on large deletes Removing sufficiently large files would create aliases for a large number of segments. This in turn results in a large number of journal entries and an overflow of s_je_array. Cheap fix is to add a BUG_ON, turning memory corruption into something annoying, but less dangerous. Real fix is to count the number of affected segments and prevent the problem completely. Signed-off-by: Joern Engel --- fs/logfs/gc.c | 8 ++++++++ fs/logfs/journal.c | 3 +++ fs/logfs/logfs.h | 8 +++++++- fs/logfs/readwrite.c | 14 ++++++++++++++ fs/logfs/super.c | 2 ++ 5 files changed, 34 insertions(+), 1 deletion(-) (limited to 'fs/logfs/readwrite.c') diff --git a/fs/logfs/gc.c b/fs/logfs/gc.c index 92949f95a901..e8253e7fb6b2 100644 --- a/fs/logfs/gc.c +++ b/fs/logfs/gc.c @@ -458,6 +458,14 @@ static void __logfs_gc_pass(struct super_block *sb, int target) struct logfs_block *block; int round, progress, last_progress = 0; + /* + * Doing too many changes to the segfile at once would result + * in a large number of aliases. Write the journal before + * things get out of hand. + */ + if (super->s_shadow_tree.no_shadowed_segments >= MAX_OBJ_ALIASES) + logfs_write_anchor(sb); + if (no_free_segments(sb) >= target && super->s_no_object_aliases < MAX_OBJ_ALIASES) return; diff --git a/fs/logfs/journal.c b/fs/logfs/journal.c index d57c7b07b60b..2c22a4ad5329 100644 --- a/fs/logfs/journal.c +++ b/fs/logfs/journal.c @@ -493,6 +493,8 @@ static void account_shadows(struct super_block *sb) btree_grim_visitor64(&tree->new, (unsigned long)sb, account_shadow); btree_grim_visitor64(&tree->old, (unsigned long)sb, account_shadow); + btree_grim_visitor32(&tree->segment_map, 0, NULL); + tree->no_shadowed_segments = 0; if (li->li_block) { /* @@ -660,6 +662,7 @@ static int logfs_write_je_buf(struct super_block *sb, void *buf, u16 type, if (ofs < 0) return ofs; logfs_buf_write(area, ofs, super->s_compressed_je, len); + BUG_ON(super->s_no_je >= MAX_JOURNAL_ENTRIES); super->s_je_array[super->s_no_je++] = cpu_to_be64(ofs); return 0; } diff --git a/fs/logfs/logfs.h b/fs/logfs/logfs.h index 97195b9e93a5..c9929eed80b1 100644 --- a/fs/logfs/logfs.h +++ b/fs/logfs/logfs.h @@ -257,10 +257,14 @@ struct logfs_shadow { * struct shadow_tree * @new: shadows where old_ofs==0, indexed by new_ofs * @old: shadows where old_ofs!=0, indexed by old_ofs + * @segment_map: bitfield of segments containing shadows + * @no_shadowed_segment: number of segments containing shadows */ struct shadow_tree { struct btree_head64 new; struct btree_head64 old; + struct btree_head32 segment_map; + int no_shadowed_segments; }; struct object_alias_item { @@ -311,6 +315,8 @@ struct logfs_block_ops { write_alias_t *write_one_alias); }; +#define MAX_JOURNAL_ENTRIES 256 + struct logfs_super { struct mtd_info *s_mtd; /* underlying device */ struct block_device *s_bdev; /* underlying device */ @@ -377,7 +383,7 @@ struct logfs_super { u32 s_journal_ec[LOGFS_JOURNAL_SEGS]; /* journal erasecounts */ u64 s_last_version; struct logfs_area *s_journal_area; /* open journal segment */ - __be64 s_je_array[64]; + __be64 s_je_array[MAX_JOURNAL_ENTRIES]; int s_no_je; int s_sum_index; /* for the 12 summaries */ diff --git a/fs/logfs/readwrite.c b/fs/logfs/readwrite.c index 3659c37fbd72..7e0c39c49719 100644 --- a/fs/logfs/readwrite.c +++ b/fs/logfs/readwrite.c @@ -1219,6 +1219,18 @@ static void free_shadow(struct inode *inode, struct logfs_shadow *shadow) mempool_free(shadow, super->s_shadow_pool); } +static void mark_segment(struct shadow_tree *tree, u32 segno) +{ + int err; + + if (!btree_lookup32(&tree->segment_map, segno)) { + err = btree_insert32(&tree->segment_map, segno, (void *)1, + GFP_NOFS); + BUG_ON(err); + tree->no_shadowed_segments++; + } +} + /** * fill_shadow_tree - Propagate shadow tree changes due to a write * @inode: Inode owning the page @@ -1266,6 +1278,8 @@ static void fill_shadow_tree(struct inode *inode, struct page *page, super->s_dirty_used_bytes += shadow->new_len; super->s_dirty_free_bytes += shadow->old_len; + mark_segment(tree, shadow->old_ofs >> super->s_segshift); + mark_segment(tree, shadow->new_ofs >> super->s_segshift); } } diff --git a/fs/logfs/super.c b/fs/logfs/super.c index 9d856c49afc5..d6e1f4fc3115 100644 --- a/fs/logfs/super.c +++ b/fs/logfs/super.c @@ -451,6 +451,8 @@ static int logfs_read_sb(struct super_block *sb, int read_only) btree_init_mempool64(&super->s_shadow_tree.new, super->s_btree_pool); btree_init_mempool64(&super->s_shadow_tree.old, super->s_btree_pool); + btree_init_mempool32(&super->s_shadow_tree.segment_map, + super->s_btree_pool); ret = logfs_init_mapping(sb); if (ret) -- cgit v1.2.3 From 1f1b0008e8dd1930d6e89522c70f4a438374302a Mon Sep 17 00:00:00 2001 From: Joern Engel Date: Thu, 15 Apr 2010 08:03:57 +0200 Subject: [LogFS] Prevent mempool_destroy NULL pointer dereference It would probably be better to just accept NULL pointers in mempool_destroy(). But for the current -rc series let's keep things simple. This patch was lost in the cracks for a while. Kevin Cernekee had to rediscover the problem and send a similar patch because of it. :( Signed-off-by: Joern Engel --- fs/logfs/logfs.h | 6 ++++++ fs/logfs/readwrite.c | 6 ++---- fs/logfs/segment.c | 2 +- fs/logfs/super.c | 4 ++-- 4 files changed, 11 insertions(+), 7 deletions(-) (limited to 'fs/logfs/readwrite.c') diff --git a/fs/logfs/logfs.h b/fs/logfs/logfs.h index c9929eed80b1..0a3df1a0c936 100644 --- a/fs/logfs/logfs.h +++ b/fs/logfs/logfs.h @@ -727,4 +727,10 @@ static inline struct logfs_area *get_area(struct super_block *sb, return logfs_super(sb)->s_area[(__force u8)gc_level]; } +static inline void logfs_mempool_destroy(mempool_t *pool) +{ + if (pool) + mempool_destroy(pool); +} + #endif diff --git a/fs/logfs/readwrite.c b/fs/logfs/readwrite.c index 7e0c39c49719..aca6c56a107a 100644 --- a/fs/logfs/readwrite.c +++ b/fs/logfs/readwrite.c @@ -2243,8 +2243,6 @@ void logfs_cleanup_rw(struct super_block *sb) struct logfs_super *super = logfs_super(sb); destroy_meta_inode(super->s_segfile_inode); - if (super->s_block_pool) - mempool_destroy(super->s_block_pool); - if (super->s_shadow_pool) - mempool_destroy(super->s_shadow_pool); + logfs_mempool_destroy(super->s_block_pool); + logfs_mempool_destroy(super->s_shadow_pool); } diff --git a/fs/logfs/segment.c b/fs/logfs/segment.c index 02db22ebbf13..8c82fe05d3e1 100644 --- a/fs/logfs/segment.c +++ b/fs/logfs/segment.c @@ -912,7 +912,7 @@ err: for (i--; i >= 0; i--) free_area(super->s_area[i]); free_area(super->s_journal_area); - mempool_destroy(super->s_alias_pool); + logfs_mempool_destroy(super->s_alias_pool); return -ENOMEM; } diff --git a/fs/logfs/super.c b/fs/logfs/super.c index d6e1f4fc3115..d4531eb46d0a 100644 --- a/fs/logfs/super.c +++ b/fs/logfs/super.c @@ -517,8 +517,8 @@ static void logfs_kill_sb(struct super_block *sb) if (super->s_erase_page) __free_page(super->s_erase_page); super->s_devops->put_device(sb); - mempool_destroy(super->s_btree_pool); - mempool_destroy(super->s_alias_pool); + logfs_mempool_destroy(super->s_btree_pool); + logfs_mempool_destroy(super->s_alias_pool); kfree(super); log_super("LogFS: Finished unmounting\n"); } -- cgit v1.2.3 From b6349ac89eacb813f6963f7263da05bc3f483351 Mon Sep 17 00:00:00 2001 From: Joern Engel Date: Tue, 20 Apr 2010 21:44:10 +0200 Subject: [LogFS] Split large truncated into smaller chunks Truncate would do an almost limitless amount of work without invoking the garbage collector in between. Split it up into more manageable, though still large, chunks. Signed-off-by: Joern Engel --- fs/logfs/readwrite.c | 34 ++++++++++++++++++++++++++-------- 1 file changed, 26 insertions(+), 8 deletions(-) (limited to 'fs/logfs/readwrite.c') diff --git a/fs/logfs/readwrite.c b/fs/logfs/readwrite.c index aca6c56a107a..7e3a1e5fd76d 100644 --- a/fs/logfs/readwrite.c +++ b/fs/logfs/readwrite.c @@ -1837,19 +1837,37 @@ static int __logfs_truncate(struct inode *inode, u64 size) return logfs_truncate_direct(inode, size); } -int logfs_truncate(struct inode *inode, u64 size) +/* + * Truncate, by changing the segment file, can consume a fair amount + * of resources. So back off from time to time and do some GC. + * 8 or 2048 blocks should be well within safety limits even if + * every single block resided in a different segment. + */ +#define TRUNCATE_STEP (8 * 1024 * 1024) +int logfs_truncate(struct inode *inode, u64 target) { struct super_block *sb = inode->i_sb; - int err; + u64 size = i_size_read(inode); + int err = 0; - logfs_get_wblocks(sb, NULL, 1); - err = __logfs_truncate(inode, size); - if (!err) - err = __logfs_write_inode(inode, 0); - logfs_put_wblocks(sb, NULL, 1); + size = ALIGN(size, TRUNCATE_STEP); + while (size > target) { + if (size > TRUNCATE_STEP) + size -= TRUNCATE_STEP; + else + size = 0; + if (size < target) + size = target; + + logfs_get_wblocks(sb, NULL, 1); + err = __logfs_truncate(inode, target); + if (!err) + err = __logfs_write_inode(inode, 0); + logfs_put_wblocks(sb, NULL, 1); + } if (!err) - err = vmtruncate(inode, size); + err = vmtruncate(inode, target); /* I don't trust error recovery yet. */ WARN_ON(err); -- cgit v1.2.3 From 2e531fa0d0868f5114c2b3a782ab02eb9d6f914d Mon Sep 17 00:00:00 2001 From: Joern Engel Date: Thu, 29 Apr 2010 14:56:37 +0200 Subject: LogFS: Fix typo in b6349ac8 Signed-off-by: Joern Engel --- fs/logfs/readwrite.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'fs/logfs/readwrite.c') diff --git a/fs/logfs/readwrite.c b/fs/logfs/readwrite.c index 3159db6958e5..8c663a55b726 100644 --- a/fs/logfs/readwrite.c +++ b/fs/logfs/readwrite.c @@ -1861,7 +1861,7 @@ int logfs_truncate(struct inode *inode, u64 target) size = target; logfs_get_wblocks(sb, NULL, 1); - err = __logfs_truncate(inode, target); + err = __logfs_truncate(inode, size); if (!err) err = __logfs_write_inode(inode, 0); logfs_put_wblocks(sb, NULL, 1); -- cgit v1.2.3 From bd2b3f29594c50d7c5bd864d9af05d440394ee82 Mon Sep 17 00:00:00 2001 From: Joern Engel Date: Sat, 1 May 2010 17:33:06 +0200 Subject: logfs: fix logfs_seek_hole() logfs_seek_hole(inode, 0x200) would crap itself if the inode contained just 0x1ff (or fewer) blocks. Signed-off-by: Joern Engel --- fs/logfs/readwrite.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'fs/logfs/readwrite.c') diff --git a/fs/logfs/readwrite.c b/fs/logfs/readwrite.c index 8c663a55b726..e37cee3b1007 100644 --- a/fs/logfs/readwrite.c +++ b/fs/logfs/readwrite.c @@ -892,6 +892,8 @@ u64 logfs_seek_hole(struct inode *inode, u64 bix) return bix; else if (li->li_data[INDIRECT_INDEX] & LOGFS_FULLY_POPULATED) bix = maxbix(li->li_height); + else if (bix >= maxbix(li->li_height)) + return bix; else { bix = seek_holedata_loop(inode, bix, 0); if (bix < maxbix(li->li_height)) -- cgit v1.2.3 From 05ebad852901cf9127a743df6ea10c0e8b1590c3 Mon Sep 17 00:00:00 2001 From: Joern Engel Date: Tue, 4 May 2010 19:41:09 +0200 Subject: logfs: commit reservations under space pressure Ensures we only return -ENOSPC when there really is no space. Signed-off-by: Joern Engel --- fs/logfs/file.c | 12 +++++++++++- fs/logfs/logfs.h | 1 + fs/logfs/readwrite.c | 15 ++++++++++++--- 3 files changed, 24 insertions(+), 4 deletions(-) (limited to 'fs/logfs/readwrite.c') diff --git a/fs/logfs/file.c b/fs/logfs/file.c index 370f367a933e..bf9b1cf953a6 100644 --- a/fs/logfs/file.c +++ b/fs/logfs/file.c @@ -161,7 +161,17 @@ static int logfs_writepage(struct page *page, struct writeback_control *wbc) static void logfs_invalidatepage(struct page *page, unsigned long offset) { - move_page_to_btree(page); + struct logfs_block *block = logfs_block(page); + + if (block->reserved_bytes) { + struct super_block *sb = page->mapping->host->i_sb; + struct logfs_super *super = logfs_super(sb); + + super->s_dirty_pages -= block->reserved_bytes; + block->ops->free_block(sb, block); + BUG_ON(bitmap_weight(block->alias_map, LOGFS_BLOCK_FACTOR)); + } else + move_page_to_btree(page); BUG_ON(PagePrivate(page) || page->private); } diff --git a/fs/logfs/logfs.h b/fs/logfs/logfs.h index 32bf55616e56..26a9458e6b13 100644 --- a/fs/logfs/logfs.h +++ b/fs/logfs/logfs.h @@ -394,6 +394,7 @@ struct logfs_super { int s_lock_count; mempool_t *s_block_pool; /* struct logfs_block pool */ mempool_t *s_shadow_pool; /* struct logfs_shadow pool */ + struct list_head s_writeback_list; /* writeback pages */ /* * Space accounting: * - s_used_bytes specifies space used to store valid data objects. diff --git a/fs/logfs/readwrite.c b/fs/logfs/readwrite.c index e37cee3b1007..0718d112a1a5 100644 --- a/fs/logfs/readwrite.c +++ b/fs/logfs/readwrite.c @@ -1095,17 +1095,25 @@ static int logfs_reserve_bytes(struct inode *inode, int bytes) int get_page_reserve(struct inode *inode, struct page *page) { struct logfs_super *super = logfs_super(inode->i_sb); + struct logfs_block *block = logfs_block(page); int ret; - if (logfs_block(page) && logfs_block(page)->reserved_bytes) + if (block && block->reserved_bytes) return 0; logfs_get_wblocks(inode->i_sb, page, WF_LOCK); - ret = logfs_reserve_bytes(inode, 6 * LOGFS_MAX_OBJECTSIZE); + while ((ret = logfs_reserve_bytes(inode, 6 * LOGFS_MAX_OBJECTSIZE)) && + !list_empty(&super->s_writeback_list)) { + block = list_entry(super->s_writeback_list.next, + struct logfs_block, alias_list); + block->ops->write_block(block); + } if (!ret) { alloc_data_block(inode, page); - logfs_block(page)->reserved_bytes += 6 * LOGFS_MAX_OBJECTSIZE; + block = logfs_block(page); + block->reserved_bytes += 6 * LOGFS_MAX_OBJECTSIZE; super->s_dirty_pages += 6 * LOGFS_MAX_OBJECTSIZE; + list_move_tail(&block->alias_list, &super->s_writeback_list); } logfs_put_wblocks(inode->i_sb, page, WF_LOCK); return ret; @@ -2251,6 +2259,7 @@ int logfs_init_rw(struct super_block *sb) int min_fill = 3 * super->s_no_blocks; INIT_LIST_HEAD(&super->s_object_alias); + INIT_LIST_HEAD(&super->s_writeback_list); mutex_init(&super->s_write_mutex); super->s_block_pool = mempool_create_kmalloc_pool(min_fill, sizeof(struct logfs_block)); -- cgit v1.2.3