Merge git://git.kernel.org/pub/scm/linux/kernel/git/joern/logfs

* git://git.kernel.org/pub/scm/linux/kernel/git/joern/logfs:
  [LogFS] Split large truncated into smaller chunks
  [LogFS] Set s_bdi
  [LogFS] Prevent mempool_destroy NULL pointer dereference
  [LogFS] Move assertion
  [LogFS] Plug 8 byte information leak
  [LogFS] Prevent memory corruption on large deletes
  [LogFS] Remove unused method

Fix trivial conflict with added header includes in fs/logfs/super.c

diff --git a/fs/logfs/gc.c b/fs/logfs/gc.c
index 84e36f5..76c242f 100644 (file)
--- a/fs/logfs/gc.c
+++ b/fs/logfs/gc.c
@@ -459,6 +459,14 @@ 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 33bd260..fb0a613 100644 (file)
--- a/fs/logfs/journal.c
+++ b/fs/logfs/journal.c
@@ -389,7 +389,10 @@ static void journal_get_erase_count(struct logfs_area *area)
 static int journal_erase_segment(struct logfs_area *area)
 {
        struct super_block *sb = area->a_sb;
-       struct logfs_segment_header sh;
+       union {
+               struct logfs_segment_header sh;
+               unsigned char c[ALIGN(sizeof(struct logfs_segment_header), 16)];
+       } u;
        u64 ofs;
        int err;
 
@@ -397,20 +400,21 @@ static int journal_erase_segment(struct logfs_area *area)
        if (err)
                return err;
 
-       sh.pad = 0;
-       sh.type = SEG_JOURNAL;
-       sh.level = 0;
-       sh.segno = cpu_to_be32(area->a_segno);
-       sh.ec = cpu_to_be32(area->a_erase_count);
-       sh.gec = cpu_to_be64(logfs_super(sb)->s_gec);
-       sh.crc = logfs_crc32(&sh, sizeof(sh), 4);
+       memset(&u, 0, sizeof(u));
+       u.sh.pad = 0;
+       u.sh.type = SEG_JOURNAL;
+       u.sh.level = 0;
+       u.sh.segno = cpu_to_be32(area->a_segno);
+       u.sh.ec = cpu_to_be32(area->a_erase_count);
+       u.sh.gec = cpu_to_be64(logfs_super(sb)->s_gec);
+       u.sh.crc = logfs_crc32(&u.sh, sizeof(u.sh), 4);
 
        /* This causes a bug in segment.c.  Not yet. */
        //logfs_set_segment_erased(sb, area->a_segno, area->a_erase_count, 0);
 
        ofs = dev_ofs(sb, area->a_segno, 0);
-       area->a_used_bytes = ALIGN(sizeof(sh), 16);
-       logfs_buf_write(area, ofs, &sh, sizeof(sh));
+       area->a_used_bytes = sizeof(u);
+       logfs_buf_write(area, ofs, &u, sizeof(u));
        return 0;
 }
 
@@ -494,6 +498,8 @@ 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) {
                /*
@@ -607,9 +613,9 @@ static size_t __logfs_write_je(struct super_block *sb, void *buf, u16 type,
        if (len == 0)
                return logfs_write_header(super, header, 0, type);
 
+       BUG_ON(len > sb->s_blocksize);
        compr_len = logfs_compress(buf, data, len, sb->s_blocksize);
        if (compr_len < 0 || type == JE_ANCHOR) {
-               BUG_ON(len > sb->s_blocksize);
                memcpy(data, buf, len);
                compr_len = len;
                compr = COMPR_NONE;
@@ -661,6 +667,7 @@ 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 b84b0ee..0a3df1a 100644 (file)
--- 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 {
@@ -305,13 +309,14 @@ typedef int write_alias_t(struct super_block *sb, u64 ino, u64 bix,
                level_t level, int child_no, __be64 val);
 struct logfs_block_ops {
        void    (*write_block)(struct logfs_block *block);
-       gc_level_t      (*block_level)(struct logfs_block *block);
        void    (*free_block)(struct super_block *sb, struct logfs_block*block);
        int     (*write_alias)(struct super_block *sb,
                        struct logfs_block *block,
                        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 */
@@ -378,7 +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 */
@@ -722,4 +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 bff4025..3159db6 100644 (file)
--- a/fs/logfs/readwrite.c
+++ b/fs/logfs/readwrite.c
@@ -430,25 +430,6 @@ 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.
@@ -587,14 +568,12 @@ 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,
 };
@@ -1241,6 +1220,18 @@ 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
@@ -1288,6 +1279,8 @@ 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);
        }
 }
 
@@ -1845,19 +1838,37 @@ 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);
@@ -2251,8 +2262,6 @@ 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 801a3a1..f77ce2b 100644 (file)
--- a/fs/logfs/segment.c
+++ b/fs/logfs/segment.c
@@ -183,14 +183,8 @@ static int btree_write_alias(struct super_block *sb, struct logfs_block *block,
        return 0;
 }
 
-static gc_level_t btree_block_level(struct logfs_block *block)
-{
-       return expand_level(block->ino, block->level);
-}
-
 static struct logfs_block_ops btree_block_ops = {
        .write_block    = btree_write_block,
-       .block_level    = btree_block_level,
        .free_block     = __free_block,
        .write_alias    = btree_write_alias,
 };
@@ -919,7 +913,7 @@ 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 b60bfac..5866ee6 100644 (file)
--- a/fs/logfs/super.c
+++ b/fs/logfs/super.c
@@ -12,6 +12,7 @@
 #include "logfs.h"
 #include <linux/bio.h>
 #include <linux/slab.h>
+#include <linux/blkdev.h>
 #include <linux/mtd/mtd.h>
 #include <linux/statfs.h>
 #include <linux/buffer_head.h>
@@ -137,6 +138,10 @@ static int logfs_sb_set(struct super_block *sb, void *_super)
        sb->s_fs_info = super;
        sb->s_mtd = super->s_mtd;
        sb->s_bdev = super->s_bdev;
+       if (sb->s_bdev)
+               sb->s_bdi = &bdev_get_queue(sb->s_bdev)->backing_dev_info;
+       if (sb->s_mtd)
+               sb->s_bdi = sb->s_mtd->backing_dev_info;
        return 0;
 }
 
@@ -452,6 +457,8 @@ 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)
@@ -516,8 +523,8 @@ 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");
 }