linux_dsm_epyc7002/fs/f2fs/dir.c

811 lines
19 KiB
C
Raw Normal View History

/*
* fs/f2fs/dir.c
*
* Copyright (c) 2012 Samsung Electronics Co., Ltd.
* http://www.samsung.com/
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
#include <linux/fs.h>
#include <linux/f2fs_fs.h>
#include "f2fs.h"
#include "node.h"
#include "acl.h"
#include "xattr.h"
static unsigned long dir_blocks(struct inode *inode)
{
return ((unsigned long long) (i_size_read(inode) + PAGE_CACHE_SIZE - 1))
>> PAGE_CACHE_SHIFT;
}
f2fs: introduce large directory support This patch introduces an i_dir_level field to support large directory. Previously, f2fs maintains multi-level hash tables to find a dentry quickly from a bunch of chiild dentries in a directory, and the hash tables consist of the following tree structure as below. In Documentation/filesystems/f2fs.txt, ---------------------- A : bucket B : block N : MAX_DIR_HASH_DEPTH ---------------------- level #0 | A(2B) | level #1 | A(2B) - A(2B) | level #2 | A(2B) - A(2B) - A(2B) - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) But, if we can guess that a directory will handle a number of child files, we don't need to traverse the tree from level #0 to #N all the time. Since the lower level tables contain relatively small number of dentries, the miss ratio of the target dentry is likely to be high. In order to avoid that, we can configure the hash tables sparsely from level #0 like this. level #0 | A(2B) - A(2B) - A(2B) - A(2B) level #1 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) With this structure, we can skip the ineffective tree searches in lower level hash tables. This patch adds just a facility for this by introducing i_dir_level in f2fs_inode. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-27 16:20:00 +07:00
static unsigned int dir_buckets(unsigned int level, int dir_level)
{
if (level + dir_level < MAX_DIR_HASH_DEPTH / 2)
f2fs: introduce large directory support This patch introduces an i_dir_level field to support large directory. Previously, f2fs maintains multi-level hash tables to find a dentry quickly from a bunch of chiild dentries in a directory, and the hash tables consist of the following tree structure as below. In Documentation/filesystems/f2fs.txt, ---------------------- A : bucket B : block N : MAX_DIR_HASH_DEPTH ---------------------- level #0 | A(2B) | level #1 | A(2B) - A(2B) | level #2 | A(2B) - A(2B) - A(2B) - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) But, if we can guess that a directory will handle a number of child files, we don't need to traverse the tree from level #0 to #N all the time. Since the lower level tables contain relatively small number of dentries, the miss ratio of the target dentry is likely to be high. In order to avoid that, we can configure the hash tables sparsely from level #0 like this. level #0 | A(2B) - A(2B) - A(2B) - A(2B) level #1 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) With this structure, we can skip the ineffective tree searches in lower level hash tables. This patch adds just a facility for this by introducing i_dir_level in f2fs_inode. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-27 16:20:00 +07:00
return 1 << (level + dir_level);
else
return MAX_DIR_BUCKETS;
}
static unsigned int bucket_blocks(unsigned int level)
{
if (level < MAX_DIR_HASH_DEPTH / 2)
return 2;
else
return 4;
}
unsigned char f2fs_filetype_table[F2FS_FT_MAX] = {
[F2FS_FT_UNKNOWN] = DT_UNKNOWN,
[F2FS_FT_REG_FILE] = DT_REG,
[F2FS_FT_DIR] = DT_DIR,
[F2FS_FT_CHRDEV] = DT_CHR,
[F2FS_FT_BLKDEV] = DT_BLK,
[F2FS_FT_FIFO] = DT_FIFO,
[F2FS_FT_SOCK] = DT_SOCK,
[F2FS_FT_SYMLINK] = DT_LNK,
};
#define S_SHIFT 12
static unsigned char f2fs_type_by_mode[S_IFMT >> S_SHIFT] = {
[S_IFREG >> S_SHIFT] = F2FS_FT_REG_FILE,
[S_IFDIR >> S_SHIFT] = F2FS_FT_DIR,
[S_IFCHR >> S_SHIFT] = F2FS_FT_CHRDEV,
[S_IFBLK >> S_SHIFT] = F2FS_FT_BLKDEV,
[S_IFIFO >> S_SHIFT] = F2FS_FT_FIFO,
[S_IFSOCK >> S_SHIFT] = F2FS_FT_SOCK,
[S_IFLNK >> S_SHIFT] = F2FS_FT_SYMLINK,
};
void set_de_type(struct f2fs_dir_entry *de, struct inode *inode)
{
umode_t mode = inode->i_mode;
de->file_type = f2fs_type_by_mode[(mode & S_IFMT) >> S_SHIFT];
}
f2fs: introduce large directory support This patch introduces an i_dir_level field to support large directory. Previously, f2fs maintains multi-level hash tables to find a dentry quickly from a bunch of chiild dentries in a directory, and the hash tables consist of the following tree structure as below. In Documentation/filesystems/f2fs.txt, ---------------------- A : bucket B : block N : MAX_DIR_HASH_DEPTH ---------------------- level #0 | A(2B) | level #1 | A(2B) - A(2B) | level #2 | A(2B) - A(2B) - A(2B) - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) But, if we can guess that a directory will handle a number of child files, we don't need to traverse the tree from level #0 to #N all the time. Since the lower level tables contain relatively small number of dentries, the miss ratio of the target dentry is likely to be high. In order to avoid that, we can configure the hash tables sparsely from level #0 like this. level #0 | A(2B) - A(2B) - A(2B) - A(2B) level #1 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) With this structure, we can skip the ineffective tree searches in lower level hash tables. This patch adds just a facility for this by introducing i_dir_level in f2fs_inode. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-27 16:20:00 +07:00
static unsigned long dir_block_index(unsigned int level,
int dir_level, unsigned int idx)
{
unsigned long i;
unsigned long bidx = 0;
for (i = 0; i < level; i++)
f2fs: introduce large directory support This patch introduces an i_dir_level field to support large directory. Previously, f2fs maintains multi-level hash tables to find a dentry quickly from a bunch of chiild dentries in a directory, and the hash tables consist of the following tree structure as below. In Documentation/filesystems/f2fs.txt, ---------------------- A : bucket B : block N : MAX_DIR_HASH_DEPTH ---------------------- level #0 | A(2B) | level #1 | A(2B) - A(2B) | level #2 | A(2B) - A(2B) - A(2B) - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) But, if we can guess that a directory will handle a number of child files, we don't need to traverse the tree from level #0 to #N all the time. Since the lower level tables contain relatively small number of dentries, the miss ratio of the target dentry is likely to be high. In order to avoid that, we can configure the hash tables sparsely from level #0 like this. level #0 | A(2B) - A(2B) - A(2B) - A(2B) level #1 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) With this structure, we can skip the ineffective tree searches in lower level hash tables. This patch adds just a facility for this by introducing i_dir_level in f2fs_inode. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-27 16:20:00 +07:00
bidx += dir_buckets(i, dir_level) * bucket_blocks(i);
bidx += idx * bucket_blocks(level);
return bidx;
}
static bool early_match_name(size_t namelen, f2fs_hash_t namehash,
struct f2fs_dir_entry *de)
{
if (le16_to_cpu(de->name_len) != namelen)
return false;
if (de->hash_code != namehash)
return false;
return true;
}
static struct f2fs_dir_entry *find_in_block(struct page *dentry_page,
struct qstr *name, int *max_slots,
struct page **res_page)
{
struct f2fs_dentry_block *dentry_blk;
struct f2fs_dir_entry *de;
struct f2fs_dentry_ptr d;
dentry_blk = (struct f2fs_dentry_block *)kmap(dentry_page);
make_dentry_ptr(&d, (void *)dentry_blk, 1);
de = find_target_dentry(name, max_slots, &d);
if (de)
*res_page = dentry_page;
else
kunmap(dentry_page);
/*
* For the most part, it should be a bug when name_len is zero.
* We stop here for figuring out where the bugs has occurred.
*/
f2fs_bug_on(F2FS_P_SB(dentry_page), d.max < 0);
return de;
}
struct f2fs_dir_entry *find_target_dentry(struct qstr *name, int *max_slots,
struct f2fs_dentry_ptr *d)
{
struct f2fs_dir_entry *de;
unsigned long bit_pos = 0;
f2fs_hash_t namehash = f2fs_dentry_hash(name);
int max_len = 0;
if (max_slots)
*max_slots = 0;
while (bit_pos < d->max) {
if (!test_bit_le(bit_pos, d->bitmap)) {
if (bit_pos == 0)
max_len = 1;
else if (!test_bit_le(bit_pos - 1, d->bitmap))
max_len++;
bit_pos++;
continue;
}
de = &d->dentry[bit_pos];
if (early_match_name(name->len, namehash, de) &&
!memcmp(d->filename[bit_pos], name->name, name->len))
goto found;
if (max_slots && *max_slots >= 0 && max_len > *max_slots) {
*max_slots = max_len;
max_len = 0;
}
/* remain bug on condition */
if (unlikely(!de->name_len))
d->max = -1;
bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
}
de = NULL;
found:
if (max_slots && max_len > *max_slots)
*max_slots = max_len;
return de;
}
static struct f2fs_dir_entry *find_in_level(struct inode *dir,
unsigned int level, struct qstr *name,
f2fs_hash_t namehash, struct page **res_page)
{
int s = GET_DENTRY_SLOTS(name->len);
unsigned int nbucket, nblock;
unsigned int bidx, end_block;
struct page *dentry_page;
struct f2fs_dir_entry *de = NULL;
bool room = false;
int max_slots;
f2fs_bug_on(F2FS_I_SB(dir), level > MAX_DIR_HASH_DEPTH);
f2fs: introduce large directory support This patch introduces an i_dir_level field to support large directory. Previously, f2fs maintains multi-level hash tables to find a dentry quickly from a bunch of chiild dentries in a directory, and the hash tables consist of the following tree structure as below. In Documentation/filesystems/f2fs.txt, ---------------------- A : bucket B : block N : MAX_DIR_HASH_DEPTH ---------------------- level #0 | A(2B) | level #1 | A(2B) - A(2B) | level #2 | A(2B) - A(2B) - A(2B) - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) But, if we can guess that a directory will handle a number of child files, we don't need to traverse the tree from level #0 to #N all the time. Since the lower level tables contain relatively small number of dentries, the miss ratio of the target dentry is likely to be high. In order to avoid that, we can configure the hash tables sparsely from level #0 like this. level #0 | A(2B) - A(2B) - A(2B) - A(2B) level #1 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) With this structure, we can skip the ineffective tree searches in lower level hash tables. This patch adds just a facility for this by introducing i_dir_level in f2fs_inode. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-27 16:20:00 +07:00
nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
nblock = bucket_blocks(level);
f2fs: introduce large directory support This patch introduces an i_dir_level field to support large directory. Previously, f2fs maintains multi-level hash tables to find a dentry quickly from a bunch of chiild dentries in a directory, and the hash tables consist of the following tree structure as below. In Documentation/filesystems/f2fs.txt, ---------------------- A : bucket B : block N : MAX_DIR_HASH_DEPTH ---------------------- level #0 | A(2B) | level #1 | A(2B) - A(2B) | level #2 | A(2B) - A(2B) - A(2B) - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) But, if we can guess that a directory will handle a number of child files, we don't need to traverse the tree from level #0 to #N all the time. Since the lower level tables contain relatively small number of dentries, the miss ratio of the target dentry is likely to be high. In order to avoid that, we can configure the hash tables sparsely from level #0 like this. level #0 | A(2B) - A(2B) - A(2B) - A(2B) level #1 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) With this structure, we can skip the ineffective tree searches in lower level hash tables. This patch adds just a facility for this by introducing i_dir_level in f2fs_inode. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-27 16:20:00 +07:00
bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
le32_to_cpu(namehash) % nbucket);
end_block = bidx + nblock;
for (; bidx < end_block; bidx++) {
/* no need to allocate new dentry pages to all the indices */
f2fs: give a chance to merge IOs by IO scheduler Previously, background GC submits many 4KB read requests to load victim blocks and/or its (i)node blocks. ... f2fs_gc : f2fs_readpage: ino = 1, page_index = 0xb61, blkaddr = 0x3b964ed f2fs_gc : block_rq_complete: 8,16 R () 499854968 + 8 [0] f2fs_gc : f2fs_readpage: ino = 1, page_index = 0xb6f, blkaddr = 0x3b964ee f2fs_gc : block_rq_complete: 8,16 R () 499854976 + 8 [0] f2fs_gc : f2fs_readpage: ino = 1, page_index = 0xb79, blkaddr = 0x3b964ef f2fs_gc : block_rq_complete: 8,16 R () 499854984 + 8 [0] ... However, by the fact that many IOs are sequential, we can give a chance to merge the IOs by IO scheduler. In order to do that, let's use blk_plug. ... f2fs_gc : f2fs_iget: ino = 143 f2fs_gc : f2fs_readpage: ino = 143, page_index = 0x1c6, blkaddr = 0x2e6ee f2fs_gc : f2fs_iget: ino = 143 f2fs_gc : f2fs_readpage: ino = 143, page_index = 0x1c7, blkaddr = 0x2e6ef <idle> : block_rq_complete: 8,16 R () 1519616 + 8 [0] <idle> : block_rq_complete: 8,16 R () 1519848 + 8 [0] <idle> : block_rq_complete: 8,16 R () 1520432 + 96 [0] <idle> : block_rq_complete: 8,16 R () 1520536 + 104 [0] <idle> : block_rq_complete: 8,16 R () 1521008 + 112 [0] <idle> : block_rq_complete: 8,16 R () 1521440 + 152 [0] <idle> : block_rq_complete: 8,16 R () 1521688 + 144 [0] <idle> : block_rq_complete: 8,16 R () 1522128 + 192 [0] <idle> : block_rq_complete: 8,16 R () 1523256 + 328 [0] ... Note that this issue should be addressed in checkpoint, and some readahead flows too. Reviewed-by: Namjae Jeon <namjae.jeon@samsung.com> Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2013-04-24 11:19:56 +07:00
dentry_page = find_data_page(dir, bidx, true);
if (IS_ERR(dentry_page)) {
room = true;
continue;
}
de = find_in_block(dentry_page, name, &max_slots, res_page);
if (de)
break;
if (max_slots >= s)
room = true;
f2fs_put_page(dentry_page, 0);
}
if (!de && room && F2FS_I(dir)->chash != namehash) {
F2FS_I(dir)->chash = namehash;
F2FS_I(dir)->clevel = level;
}
return de;
}
/*
* Find an entry in the specified directory with the wanted name.
* It returns the page where the entry was found (as a parameter - res_page),
* and the entry itself. Page is returned mapped and unlocked.
* Entry is guaranteed to be valid.
*/
struct f2fs_dir_entry *f2fs_find_entry(struct inode *dir,
struct qstr *child, struct page **res_page)
{
unsigned long npages = dir_blocks(dir);
struct f2fs_dir_entry *de = NULL;
f2fs_hash_t name_hash;
unsigned int max_depth;
unsigned int level;
if (f2fs_has_inline_dentry(dir))
return find_in_inline_dir(dir, child, res_page);
if (npages == 0)
return NULL;
*res_page = NULL;
name_hash = f2fs_dentry_hash(child);
max_depth = F2FS_I(dir)->i_current_depth;
for (level = 0; level < max_depth; level++) {
de = find_in_level(dir, level, child, name_hash, res_page);
if (de)
break;
}
if (!de && F2FS_I(dir)->chash != name_hash) {
F2FS_I(dir)->chash = name_hash;
F2FS_I(dir)->clevel = level - 1;
}
return de;
}
struct f2fs_dir_entry *f2fs_parent_dir(struct inode *dir, struct page **p)
{
struct page *page;
struct f2fs_dir_entry *de;
struct f2fs_dentry_block *dentry_blk;
if (f2fs_has_inline_dentry(dir))
return f2fs_parent_inline_dir(dir, p);
page = get_lock_data_page(dir, 0);
if (IS_ERR(page))
return NULL;
dentry_blk = kmap(page);
de = &dentry_blk->dentry[1];
*p = page;
unlock_page(page);
return de;
}
ino_t f2fs_inode_by_name(struct inode *dir, struct qstr *qstr)
{
ino_t res = 0;
struct f2fs_dir_entry *de;
struct page *page;
de = f2fs_find_entry(dir, qstr, &page);
if (de) {
res = le32_to_cpu(de->ino);
f2fs_dentry_kunmap(dir, page);
f2fs_put_page(page, 0);
}
return res;
}
void f2fs_set_link(struct inode *dir, struct f2fs_dir_entry *de,
struct page *page, struct inode *inode)
{
enum page_type type = f2fs_has_inline_dentry(dir) ? NODE : DATA;
lock_page(page);
f2fs_wait_on_page_writeback(page, type);
de->ino = cpu_to_le32(inode->i_ino);
set_de_type(de, inode);
f2fs_dentry_kunmap(dir, page);
set_page_dirty(page);
dir->i_mtime = dir->i_ctime = CURRENT_TIME;
mark_inode_dirty(dir);
f2fs_put_page(page, 1);
}
static void init_dent_inode(const struct qstr *name, struct page *ipage)
{
struct f2fs_inode *ri;
f2fs_wait_on_page_writeback(ipage, NODE);
/* copy name info. to this inode page */
ri = F2FS_INODE(ipage);
ri->i_namelen = cpu_to_le32(name->len);
memcpy(ri->i_name, name->name, name->len);
set_page_dirty(ipage);
}
int update_dent_inode(struct inode *inode, const struct qstr *name)
{
struct page *page;
page = get_node_page(F2FS_I_SB(inode), inode->i_ino);
if (IS_ERR(page))
return PTR_ERR(page);
init_dent_inode(name, page);
f2fs_put_page(page, 1);
return 0;
}
void do_make_empty_dir(struct inode *inode, struct inode *parent,
struct f2fs_dentry_ptr *d)
{
struct f2fs_dir_entry *de;
de = &d->dentry[0];
de->name_len = cpu_to_le16(1);
de->hash_code = 0;
de->ino = cpu_to_le32(inode->i_ino);
memcpy(d->filename[0], ".", 1);
set_de_type(de, inode);
de = &d->dentry[1];
de->hash_code = 0;
de->name_len = cpu_to_le16(2);
de->ino = cpu_to_le32(parent->i_ino);
memcpy(d->filename[1], "..", 2);
set_de_type(de, inode);
test_and_set_bit_le(0, (void *)d->bitmap);
test_and_set_bit_le(1, (void *)d->bitmap);
}
static int make_empty_dir(struct inode *inode,
struct inode *parent, struct page *page)
f2fs: introduce a new global lock scheme In the previous version, f2fs uses global locks according to the usage types, such as directory operations, block allocation, block write, and so on. Reference the following lock types in f2fs.h. enum lock_type { RENAME, /* for renaming operations */ DENTRY_OPS, /* for directory operations */ DATA_WRITE, /* for data write */ DATA_NEW, /* for data allocation */ DATA_TRUNC, /* for data truncate */ NODE_NEW, /* for node allocation */ NODE_TRUNC, /* for node truncate */ NODE_WRITE, /* for node write */ NR_LOCK_TYPE, }; In that case, we lose the performance under the multi-threading environment, since every types of operations must be conducted one at a time. In order to address the problem, let's share the locks globally with a mutex array regardless of any types. So, let users grab a mutex and perform their jobs in parallel as much as possbile. For this, I propose a new global lock scheme as follows. 0. Data structure - f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS] - f2fs_sb_info -> node_write 1. mutex_lock_op(sbi) - try to get an avaiable lock from the array. - returns the index of the gottern lock variable. 2. mutex_unlock_op(sbi, index of the lock) - unlock the given index of the lock. 3. mutex_lock_all(sbi) - grab all the locks in the array before the checkpoint. 4. mutex_unlock_all(sbi) - release all the locks in the array after checkpoint. 5. block_operations() - call mutex_lock_all() - sync_dirty_dir_inodes() - grab node_write - sync_node_pages() Note that, the pairs of mutex_lock_op()/mutex_unlock_op() and mutex_lock_all()/mutex_unlock_all() should be used together. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2012-11-22 14:21:29 +07:00
{
struct page *dentry_page;
struct f2fs_dentry_block *dentry_blk;
struct f2fs_dentry_ptr d;
f2fs: introduce a new global lock scheme In the previous version, f2fs uses global locks according to the usage types, such as directory operations, block allocation, block write, and so on. Reference the following lock types in f2fs.h. enum lock_type { RENAME, /* for renaming operations */ DENTRY_OPS, /* for directory operations */ DATA_WRITE, /* for data write */ DATA_NEW, /* for data allocation */ DATA_TRUNC, /* for data truncate */ NODE_NEW, /* for node allocation */ NODE_TRUNC, /* for node truncate */ NODE_WRITE, /* for node write */ NR_LOCK_TYPE, }; In that case, we lose the performance under the multi-threading environment, since every types of operations must be conducted one at a time. In order to address the problem, let's share the locks globally with a mutex array regardless of any types. So, let users grab a mutex and perform their jobs in parallel as much as possbile. For this, I propose a new global lock scheme as follows. 0. Data structure - f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS] - f2fs_sb_info -> node_write 1. mutex_lock_op(sbi) - try to get an avaiable lock from the array. - returns the index of the gottern lock variable. 2. mutex_unlock_op(sbi, index of the lock) - unlock the given index of the lock. 3. mutex_lock_all(sbi) - grab all the locks in the array before the checkpoint. 4. mutex_unlock_all(sbi) - release all the locks in the array after checkpoint. 5. block_operations() - call mutex_lock_all() - sync_dirty_dir_inodes() - grab node_write - sync_node_pages() Note that, the pairs of mutex_lock_op()/mutex_unlock_op() and mutex_lock_all()/mutex_unlock_all() should be used together. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2012-11-22 14:21:29 +07:00
if (f2fs_has_inline_dentry(inode))
return make_empty_inline_dir(inode, parent, page);
dentry_page = get_new_data_page(inode, page, 0, true);
f2fs: introduce a new global lock scheme In the previous version, f2fs uses global locks according to the usage types, such as directory operations, block allocation, block write, and so on. Reference the following lock types in f2fs.h. enum lock_type { RENAME, /* for renaming operations */ DENTRY_OPS, /* for directory operations */ DATA_WRITE, /* for data write */ DATA_NEW, /* for data allocation */ DATA_TRUNC, /* for data truncate */ NODE_NEW, /* for node allocation */ NODE_TRUNC, /* for node truncate */ NODE_WRITE, /* for node write */ NR_LOCK_TYPE, }; In that case, we lose the performance under the multi-threading environment, since every types of operations must be conducted one at a time. In order to address the problem, let's share the locks globally with a mutex array regardless of any types. So, let users grab a mutex and perform their jobs in parallel as much as possbile. For this, I propose a new global lock scheme as follows. 0. Data structure - f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS] - f2fs_sb_info -> node_write 1. mutex_lock_op(sbi) - try to get an avaiable lock from the array. - returns the index of the gottern lock variable. 2. mutex_unlock_op(sbi, index of the lock) - unlock the given index of the lock. 3. mutex_lock_all(sbi) - grab all the locks in the array before the checkpoint. 4. mutex_unlock_all(sbi) - release all the locks in the array after checkpoint. 5. block_operations() - call mutex_lock_all() - sync_dirty_dir_inodes() - grab node_write - sync_node_pages() Note that, the pairs of mutex_lock_op()/mutex_unlock_op() and mutex_lock_all()/mutex_unlock_all() should be used together. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2012-11-22 14:21:29 +07:00
if (IS_ERR(dentry_page))
return PTR_ERR(dentry_page);
dentry_blk = kmap_atomic(dentry_page);
f2fs: introduce a new global lock scheme In the previous version, f2fs uses global locks according to the usage types, such as directory operations, block allocation, block write, and so on. Reference the following lock types in f2fs.h. enum lock_type { RENAME, /* for renaming operations */ DENTRY_OPS, /* for directory operations */ DATA_WRITE, /* for data write */ DATA_NEW, /* for data allocation */ DATA_TRUNC, /* for data truncate */ NODE_NEW, /* for node allocation */ NODE_TRUNC, /* for node truncate */ NODE_WRITE, /* for node write */ NR_LOCK_TYPE, }; In that case, we lose the performance under the multi-threading environment, since every types of operations must be conducted one at a time. In order to address the problem, let's share the locks globally with a mutex array regardless of any types. So, let users grab a mutex and perform their jobs in parallel as much as possbile. For this, I propose a new global lock scheme as follows. 0. Data structure - f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS] - f2fs_sb_info -> node_write 1. mutex_lock_op(sbi) - try to get an avaiable lock from the array. - returns the index of the gottern lock variable. 2. mutex_unlock_op(sbi, index of the lock) - unlock the given index of the lock. 3. mutex_lock_all(sbi) - grab all the locks in the array before the checkpoint. 4. mutex_unlock_all(sbi) - release all the locks in the array after checkpoint. 5. block_operations() - call mutex_lock_all() - sync_dirty_dir_inodes() - grab node_write - sync_node_pages() Note that, the pairs of mutex_lock_op()/mutex_unlock_op() and mutex_lock_all()/mutex_unlock_all() should be used together. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2012-11-22 14:21:29 +07:00
make_dentry_ptr(&d, (void *)dentry_blk, 1);
do_make_empty_dir(inode, parent, &d);
f2fs: introduce a new global lock scheme In the previous version, f2fs uses global locks according to the usage types, such as directory operations, block allocation, block write, and so on. Reference the following lock types in f2fs.h. enum lock_type { RENAME, /* for renaming operations */ DENTRY_OPS, /* for directory operations */ DATA_WRITE, /* for data write */ DATA_NEW, /* for data allocation */ DATA_TRUNC, /* for data truncate */ NODE_NEW, /* for node allocation */ NODE_TRUNC, /* for node truncate */ NODE_WRITE, /* for node write */ NR_LOCK_TYPE, }; In that case, we lose the performance under the multi-threading environment, since every types of operations must be conducted one at a time. In order to address the problem, let's share the locks globally with a mutex array regardless of any types. So, let users grab a mutex and perform their jobs in parallel as much as possbile. For this, I propose a new global lock scheme as follows. 0. Data structure - f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS] - f2fs_sb_info -> node_write 1. mutex_lock_op(sbi) - try to get an avaiable lock from the array. - returns the index of the gottern lock variable. 2. mutex_unlock_op(sbi, index of the lock) - unlock the given index of the lock. 3. mutex_lock_all(sbi) - grab all the locks in the array before the checkpoint. 4. mutex_unlock_all(sbi) - release all the locks in the array after checkpoint. 5. block_operations() - call mutex_lock_all() - sync_dirty_dir_inodes() - grab node_write - sync_node_pages() Note that, the pairs of mutex_lock_op()/mutex_unlock_op() and mutex_lock_all()/mutex_unlock_all() should be used together. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2012-11-22 14:21:29 +07:00
kunmap_atomic(dentry_blk);
f2fs: introduce a new global lock scheme In the previous version, f2fs uses global locks according to the usage types, such as directory operations, block allocation, block write, and so on. Reference the following lock types in f2fs.h. enum lock_type { RENAME, /* for renaming operations */ DENTRY_OPS, /* for directory operations */ DATA_WRITE, /* for data write */ DATA_NEW, /* for data allocation */ DATA_TRUNC, /* for data truncate */ NODE_NEW, /* for node allocation */ NODE_TRUNC, /* for node truncate */ NODE_WRITE, /* for node write */ NR_LOCK_TYPE, }; In that case, we lose the performance under the multi-threading environment, since every types of operations must be conducted one at a time. In order to address the problem, let's share the locks globally with a mutex array regardless of any types. So, let users grab a mutex and perform their jobs in parallel as much as possbile. For this, I propose a new global lock scheme as follows. 0. Data structure - f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS] - f2fs_sb_info -> node_write 1. mutex_lock_op(sbi) - try to get an avaiable lock from the array. - returns the index of the gottern lock variable. 2. mutex_unlock_op(sbi, index of the lock) - unlock the given index of the lock. 3. mutex_lock_all(sbi) - grab all the locks in the array before the checkpoint. 4. mutex_unlock_all(sbi) - release all the locks in the array after checkpoint. 5. block_operations() - call mutex_lock_all() - sync_dirty_dir_inodes() - grab node_write - sync_node_pages() Note that, the pairs of mutex_lock_op()/mutex_unlock_op() and mutex_lock_all()/mutex_unlock_all() should be used together. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2012-11-22 14:21:29 +07:00
set_page_dirty(dentry_page);
f2fs_put_page(dentry_page, 1);
return 0;
}
f2fs: avoid deadlock on init_inode_metadata Previously, init_inode_metadata does not hold any parent directory's inode page. So, f2fs_init_acl can grab its parent inode page without any problem. But, when we use inline_dentry, that page is grabbed during f2fs_add_link, so that we can fall into deadlock condition like below. INFO: task mknod:11006 blocked for more than 120 seconds. Tainted: G OE 3.17.0-rc1+ #13 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. mknod D ffff88003fc94580 0 11006 11004 0x00000000 ffff880007717b10 0000000000000002 ffff88003c323220 ffff880007717fd8 0000000000014580 0000000000014580 ffff88003daecb30 ffff88003c323220 ffff88003fc94e80 ffff88003ffbb4e8 ffff880007717ba0 0000000000000002 Call Trace: [<ffffffff8173dc40>] ? bit_wait+0x50/0x50 [<ffffffff8173d4cd>] io_schedule+0x9d/0x130 [<ffffffff8173dc6c>] bit_wait_io+0x2c/0x50 [<ffffffff8173da3b>] __wait_on_bit_lock+0x4b/0xb0 [<ffffffff811640a7>] __lock_page+0x67/0x70 [<ffffffff810acf50>] ? autoremove_wake_function+0x40/0x40 [<ffffffff811652cc>] pagecache_get_page+0x14c/0x1e0 [<ffffffffa029afa9>] get_node_page+0x59/0x130 [f2fs] [<ffffffffa02a63ad>] read_all_xattrs+0x24d/0x430 [f2fs] [<ffffffffa02a6ca2>] f2fs_getxattr+0x52/0xe0 [f2fs] [<ffffffffa02a7481>] f2fs_get_acl+0x41/0x2d0 [f2fs] [<ffffffff8122d847>] get_acl+0x47/0x70 [<ffffffff8122db5a>] posix_acl_create+0x5a/0x150 [<ffffffffa02a7759>] f2fs_init_acl+0x29/0xcb [f2fs] [<ffffffffa0286a8d>] init_inode_metadata+0x5d/0x340 [f2fs] [<ffffffffa029253a>] f2fs_add_inline_entry+0x12a/0x2e0 [f2fs] [<ffffffffa0286ea5>] __f2fs_add_link+0x45/0x4a0 [f2fs] [<ffffffffa028b5b6>] ? f2fs_new_inode+0x146/0x220 [f2fs] [<ffffffffa028b816>] f2fs_mknod+0x86/0xf0 [f2fs] [<ffffffff811e3ec1>] vfs_mknod+0xe1/0x160 [<ffffffff811e4b26>] SyS_mknod+0x1f6/0x200 [<ffffffff81741d7f>] tracesys+0xe1/0xe6 Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-10-14 09:42:53 +07:00
struct page *init_inode_metadata(struct inode *inode, struct inode *dir,
const struct qstr *name, struct page *dpage)
{
struct page *page;
int err;
if (is_inode_flag_set(F2FS_I(inode), FI_NEW_INODE)) {
page = new_inode_page(inode);
if (IS_ERR(page))
return page;
if (S_ISDIR(inode->i_mode)) {
err = make_empty_dir(inode, dir, page);
if (err)
goto error;
}
f2fs: avoid deadlock on init_inode_metadata Previously, init_inode_metadata does not hold any parent directory's inode page. So, f2fs_init_acl can grab its parent inode page without any problem. But, when we use inline_dentry, that page is grabbed during f2fs_add_link, so that we can fall into deadlock condition like below. INFO: task mknod:11006 blocked for more than 120 seconds. Tainted: G OE 3.17.0-rc1+ #13 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. mknod D ffff88003fc94580 0 11006 11004 0x00000000 ffff880007717b10 0000000000000002 ffff88003c323220 ffff880007717fd8 0000000000014580 0000000000014580 ffff88003daecb30 ffff88003c323220 ffff88003fc94e80 ffff88003ffbb4e8 ffff880007717ba0 0000000000000002 Call Trace: [<ffffffff8173dc40>] ? bit_wait+0x50/0x50 [<ffffffff8173d4cd>] io_schedule+0x9d/0x130 [<ffffffff8173dc6c>] bit_wait_io+0x2c/0x50 [<ffffffff8173da3b>] __wait_on_bit_lock+0x4b/0xb0 [<ffffffff811640a7>] __lock_page+0x67/0x70 [<ffffffff810acf50>] ? autoremove_wake_function+0x40/0x40 [<ffffffff811652cc>] pagecache_get_page+0x14c/0x1e0 [<ffffffffa029afa9>] get_node_page+0x59/0x130 [f2fs] [<ffffffffa02a63ad>] read_all_xattrs+0x24d/0x430 [f2fs] [<ffffffffa02a6ca2>] f2fs_getxattr+0x52/0xe0 [f2fs] [<ffffffffa02a7481>] f2fs_get_acl+0x41/0x2d0 [f2fs] [<ffffffff8122d847>] get_acl+0x47/0x70 [<ffffffff8122db5a>] posix_acl_create+0x5a/0x150 [<ffffffffa02a7759>] f2fs_init_acl+0x29/0xcb [f2fs] [<ffffffffa0286a8d>] init_inode_metadata+0x5d/0x340 [f2fs] [<ffffffffa029253a>] f2fs_add_inline_entry+0x12a/0x2e0 [f2fs] [<ffffffffa0286ea5>] __f2fs_add_link+0x45/0x4a0 [f2fs] [<ffffffffa028b5b6>] ? f2fs_new_inode+0x146/0x220 [f2fs] [<ffffffffa028b816>] f2fs_mknod+0x86/0xf0 [f2fs] [<ffffffff811e3ec1>] vfs_mknod+0xe1/0x160 [<ffffffff811e4b26>] SyS_mknod+0x1f6/0x200 [<ffffffff81741d7f>] tracesys+0xe1/0xe6 Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-10-14 09:42:53 +07:00
err = f2fs_init_acl(inode, dir, page, dpage);
if (err)
goto put_error;
err = f2fs_init_security(inode, dir, name, page);
if (err)
goto put_error;
} else {
page = get_node_page(F2FS_I_SB(dir), inode->i_ino);
if (IS_ERR(page))
return page;
set_cold_node(inode, page);
}
if (name)
init_dent_inode(name, page);
/*
* This file should be checkpointed during fsync.
* We lost i_pino from now on.
*/
if (is_inode_flag_set(F2FS_I(inode), FI_INC_LINK)) {
file_lost_pino(inode);
/*
* If link the tmpfile to alias through linkat path,
* we should remove this inode from orphan list.
*/
if (inode->i_nlink == 0)
remove_orphan_inode(F2FS_I_SB(dir), inode->i_ino);
inc_nlink(inode);
}
return page;
put_error:
f2fs_put_page(page, 1);
error:
f2fs: fix to truncate dentry pages in the error case When a new directory is allocated, if an error is occurred, we should truncate preallocated dentry pages too. This bug was reported by Andrey Tsyvarev after a while as follows. mkdir()-> f2fs_add_link()-> init_inode_metadata()-> f2fs_init_acl()-> f2fs_get_acl()-> f2fs_getxattr()-> read_all_xattrs() fails. Also there was a BUG_ON triggered after the fault in mkdir()-> f2fs_add_link()-> init_inode_metadata()-> remove_inode_page() -> f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1); But, previous patch wasn't perfect to resolve that bug, so the following bug report was also submitted. kernel BUG at fs/f2fs/inode.c:274! Call Trace: [<ffffffff811fde03>] evict+0xa3/0x1a0 [<ffffffff811fe615>] iput+0xf5/0x180 [<ffffffffa01c7f63>] f2fs_mkdir+0xf3/0x150 [f2fs] [<ffffffff811f2a77>] vfs_mkdir+0xb7/0x160 [<ffffffff811f36bf>] SyS_mkdir+0x5f/0xc0 [<ffffffff81680769>] system_call_fastpath+0x16/0x1b Finally, this patch resolves all the issues like below. If an error is occurred after make_empty_dir(), 1. truncate_inode_pages() The make_bad_inode() prior to iput() will change i_mode to S_IFREG, which means that f2fs will not decrement fi->dirty_dents during f2fs_evict_inode. But, by calling it here, we can do that. 2. truncate_blocks() Preallocated dentry pages are trucated here to sync i_blocks. 3. remove_dirty_dir_inode() Remove this directory inode from the list. Reported-and-Tested-by: Andrey Tsyvarev <tsyvarev@ispras.ru> Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-05 09:16:39 +07:00
/* once the failed inode becomes a bad inode, i_mode is S_IFREG */
truncate_inode_pages(&inode->i_data, 0);
truncate_blocks(inode, 0, false);
f2fs: fix to truncate dentry pages in the error case When a new directory is allocated, if an error is occurred, we should truncate preallocated dentry pages too. This bug was reported by Andrey Tsyvarev after a while as follows. mkdir()-> f2fs_add_link()-> init_inode_metadata()-> f2fs_init_acl()-> f2fs_get_acl()-> f2fs_getxattr()-> read_all_xattrs() fails. Also there was a BUG_ON triggered after the fault in mkdir()-> f2fs_add_link()-> init_inode_metadata()-> remove_inode_page() -> f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1); But, previous patch wasn't perfect to resolve that bug, so the following bug report was also submitted. kernel BUG at fs/f2fs/inode.c:274! Call Trace: [<ffffffff811fde03>] evict+0xa3/0x1a0 [<ffffffff811fe615>] iput+0xf5/0x180 [<ffffffffa01c7f63>] f2fs_mkdir+0xf3/0x150 [f2fs] [<ffffffff811f2a77>] vfs_mkdir+0xb7/0x160 [<ffffffff811f36bf>] SyS_mkdir+0x5f/0xc0 [<ffffffff81680769>] system_call_fastpath+0x16/0x1b Finally, this patch resolves all the issues like below. If an error is occurred after make_empty_dir(), 1. truncate_inode_pages() The make_bad_inode() prior to iput() will change i_mode to S_IFREG, which means that f2fs will not decrement fi->dirty_dents during f2fs_evict_inode. But, by calling it here, we can do that. 2. truncate_blocks() Preallocated dentry pages are trucated here to sync i_blocks. 3. remove_dirty_dir_inode() Remove this directory inode from the list. Reported-and-Tested-by: Andrey Tsyvarev <tsyvarev@ispras.ru> Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-05 09:16:39 +07:00
remove_dirty_dir_inode(inode);
remove_inode_page(inode);
return ERR_PTR(err);
}
void update_parent_metadata(struct inode *dir, struct inode *inode,
unsigned int current_depth)
{
if (is_inode_flag_set(F2FS_I(inode), FI_NEW_INODE)) {
if (S_ISDIR(inode->i_mode)) {
inc_nlink(dir);
set_inode_flag(F2FS_I(dir), FI_UPDATE_DIR);
}
clear_inode_flag(F2FS_I(inode), FI_NEW_INODE);
}
dir->i_mtime = dir->i_ctime = CURRENT_TIME;
mark_inode_dirty(dir);
if (F2FS_I(dir)->i_current_depth != current_depth) {
F2FS_I(dir)->i_current_depth = current_depth;
set_inode_flag(F2FS_I(dir), FI_UPDATE_DIR);
}
if (is_inode_flag_set(F2FS_I(inode), FI_INC_LINK))
clear_inode_flag(F2FS_I(inode), FI_INC_LINK);
}
int room_for_filename(const void *bitmap, int slots, int max_slots)
{
int bit_start = 0;
int zero_start, zero_end;
next:
zero_start = find_next_zero_bit_le(bitmap, max_slots, bit_start);
if (zero_start >= max_slots)
return max_slots;
zero_end = find_next_bit_le(bitmap, max_slots, zero_start);
if (zero_end - zero_start >= slots)
return zero_start;
bit_start = zero_end + 1;
if (zero_end + 1 >= max_slots)
return max_slots;
goto next;
}
void f2fs_update_dentry(struct inode *inode, struct f2fs_dentry_ptr *d,
const struct qstr *name, f2fs_hash_t name_hash,
unsigned int bit_pos)
{
struct f2fs_dir_entry *de;
int slots = GET_DENTRY_SLOTS(name->len);
int i;
de = &d->dentry[bit_pos];
de->hash_code = name_hash;
de->name_len = cpu_to_le16(name->len);
memcpy(d->filename[bit_pos], name->name, name->len);
de->ino = cpu_to_le32(inode->i_ino);
set_de_type(de, inode);
for (i = 0; i < slots; i++)
test_and_set_bit_le(bit_pos + i, (void *)d->bitmap);
}
f2fs: introduce a new global lock scheme In the previous version, f2fs uses global locks according to the usage types, such as directory operations, block allocation, block write, and so on. Reference the following lock types in f2fs.h. enum lock_type { RENAME, /* for renaming operations */ DENTRY_OPS, /* for directory operations */ DATA_WRITE, /* for data write */ DATA_NEW, /* for data allocation */ DATA_TRUNC, /* for data truncate */ NODE_NEW, /* for node allocation */ NODE_TRUNC, /* for node truncate */ NODE_WRITE, /* for node write */ NR_LOCK_TYPE, }; In that case, we lose the performance under the multi-threading environment, since every types of operations must be conducted one at a time. In order to address the problem, let's share the locks globally with a mutex array regardless of any types. So, let users grab a mutex and perform their jobs in parallel as much as possbile. For this, I propose a new global lock scheme as follows. 0. Data structure - f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS] - f2fs_sb_info -> node_write 1. mutex_lock_op(sbi) - try to get an avaiable lock from the array. - returns the index of the gottern lock variable. 2. mutex_unlock_op(sbi, index of the lock) - unlock the given index of the lock. 3. mutex_lock_all(sbi) - grab all the locks in the array before the checkpoint. 4. mutex_unlock_all(sbi) - release all the locks in the array after checkpoint. 5. block_operations() - call mutex_lock_all() - sync_dirty_dir_inodes() - grab node_write - sync_node_pages() Note that, the pairs of mutex_lock_op()/mutex_unlock_op() and mutex_lock_all()/mutex_unlock_all() should be used together. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2012-11-22 14:21:29 +07:00
/*
* Caller should grab and release a rwsem by calling f2fs_lock_op() and
* f2fs_unlock_op().
f2fs: introduce a new global lock scheme In the previous version, f2fs uses global locks according to the usage types, such as directory operations, block allocation, block write, and so on. Reference the following lock types in f2fs.h. enum lock_type { RENAME, /* for renaming operations */ DENTRY_OPS, /* for directory operations */ DATA_WRITE, /* for data write */ DATA_NEW, /* for data allocation */ DATA_TRUNC, /* for data truncate */ NODE_NEW, /* for node allocation */ NODE_TRUNC, /* for node truncate */ NODE_WRITE, /* for node write */ NR_LOCK_TYPE, }; In that case, we lose the performance under the multi-threading environment, since every types of operations must be conducted one at a time. In order to address the problem, let's share the locks globally with a mutex array regardless of any types. So, let users grab a mutex and perform their jobs in parallel as much as possbile. For this, I propose a new global lock scheme as follows. 0. Data structure - f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS] - f2fs_sb_info -> node_write 1. mutex_lock_op(sbi) - try to get an avaiable lock from the array. - returns the index of the gottern lock variable. 2. mutex_unlock_op(sbi, index of the lock) - unlock the given index of the lock. 3. mutex_lock_all(sbi) - grab all the locks in the array before the checkpoint. 4. mutex_unlock_all(sbi) - release all the locks in the array after checkpoint. 5. block_operations() - call mutex_lock_all() - sync_dirty_dir_inodes() - grab node_write - sync_node_pages() Note that, the pairs of mutex_lock_op()/mutex_unlock_op() and mutex_lock_all()/mutex_unlock_all() should be used together. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2012-11-22 14:21:29 +07:00
*/
int __f2fs_add_link(struct inode *dir, const struct qstr *name,
struct inode *inode)
{
unsigned int bit_pos;
unsigned int level;
unsigned int current_depth;
unsigned long bidx, block;
f2fs_hash_t dentry_hash;
unsigned int nbucket, nblock;
size_t namelen = name->len;
struct page *dentry_page = NULL;
struct f2fs_dentry_block *dentry_blk = NULL;
struct f2fs_dentry_ptr d;
int slots = GET_DENTRY_SLOTS(namelen);
struct page *page;
int err = 0;
if (f2fs_has_inline_dentry(dir)) {
err = f2fs_add_inline_entry(dir, name, inode);
if (!err || err != -EAGAIN)
return err;
else
err = 0;
}
dentry_hash = f2fs_dentry_hash(name);
level = 0;
current_depth = F2FS_I(dir)->i_current_depth;
if (F2FS_I(dir)->chash == dentry_hash) {
level = F2FS_I(dir)->clevel;
F2FS_I(dir)->chash = 0;
}
start:
if (unlikely(current_depth == MAX_DIR_HASH_DEPTH))
return -ENOSPC;
/* Increase the depth, if required */
if (level == current_depth)
++current_depth;
f2fs: introduce large directory support This patch introduces an i_dir_level field to support large directory. Previously, f2fs maintains multi-level hash tables to find a dentry quickly from a bunch of chiild dentries in a directory, and the hash tables consist of the following tree structure as below. In Documentation/filesystems/f2fs.txt, ---------------------- A : bucket B : block N : MAX_DIR_HASH_DEPTH ---------------------- level #0 | A(2B) | level #1 | A(2B) - A(2B) | level #2 | A(2B) - A(2B) - A(2B) - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) But, if we can guess that a directory will handle a number of child files, we don't need to traverse the tree from level #0 to #N all the time. Since the lower level tables contain relatively small number of dentries, the miss ratio of the target dentry is likely to be high. In order to avoid that, we can configure the hash tables sparsely from level #0 like this. level #0 | A(2B) - A(2B) - A(2B) - A(2B) level #1 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) With this structure, we can skip the ineffective tree searches in lower level hash tables. This patch adds just a facility for this by introducing i_dir_level in f2fs_inode. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-27 16:20:00 +07:00
nbucket = dir_buckets(level, F2FS_I(dir)->i_dir_level);
nblock = bucket_blocks(level);
f2fs: introduce large directory support This patch introduces an i_dir_level field to support large directory. Previously, f2fs maintains multi-level hash tables to find a dentry quickly from a bunch of chiild dentries in a directory, and the hash tables consist of the following tree structure as below. In Documentation/filesystems/f2fs.txt, ---------------------- A : bucket B : block N : MAX_DIR_HASH_DEPTH ---------------------- level #0 | A(2B) | level #1 | A(2B) - A(2B) | level #2 | A(2B) - A(2B) - A(2B) - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) But, if we can guess that a directory will handle a number of child files, we don't need to traverse the tree from level #0 to #N all the time. Since the lower level tables contain relatively small number of dentries, the miss ratio of the target dentry is likely to be high. In order to avoid that, we can configure the hash tables sparsely from level #0 like this. level #0 | A(2B) - A(2B) - A(2B) - A(2B) level #1 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N/2 | A(2B) - A(2B) - A(2B) - A(2B) - A(2B) - ... - A(2B) . | . . . . level #N | A(4B) - A(4B) - A(4B) - A(4B) - A(4B) - ... - A(4B) With this structure, we can skip the ineffective tree searches in lower level hash tables. This patch adds just a facility for this by introducing i_dir_level in f2fs_inode. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2014-02-27 16:20:00 +07:00
bidx = dir_block_index(level, F2FS_I(dir)->i_dir_level,
(le32_to_cpu(dentry_hash) % nbucket));
for (block = bidx; block <= (bidx + nblock - 1); block++) {
dentry_page = get_new_data_page(dir, NULL, block, true);
f2fs: introduce a new global lock scheme In the previous version, f2fs uses global locks according to the usage types, such as directory operations, block allocation, block write, and so on. Reference the following lock types in f2fs.h. enum lock_type { RENAME, /* for renaming operations */ DENTRY_OPS, /* for directory operations */ DATA_WRITE, /* for data write */ DATA_NEW, /* for data allocation */ DATA_TRUNC, /* for data truncate */ NODE_NEW, /* for node allocation */ NODE_TRUNC, /* for node truncate */ NODE_WRITE, /* for node write */ NR_LOCK_TYPE, }; In that case, we lose the performance under the multi-threading environment, since every types of operations must be conducted one at a time. In order to address the problem, let's share the locks globally with a mutex array regardless of any types. So, let users grab a mutex and perform their jobs in parallel as much as possbile. For this, I propose a new global lock scheme as follows. 0. Data structure - f2fs_sb_info -> mutex_lock[NR_GLOBAL_LOCKS] - f2fs_sb_info -> node_write 1. mutex_lock_op(sbi) - try to get an avaiable lock from the array. - returns the index of the gottern lock variable. 2. mutex_unlock_op(sbi, index of the lock) - unlock the given index of the lock. 3. mutex_lock_all(sbi) - grab all the locks in the array before the checkpoint. 4. mutex_unlock_all(sbi) - release all the locks in the array after checkpoint. 5. block_operations() - call mutex_lock_all() - sync_dirty_dir_inodes() - grab node_write - sync_node_pages() Note that, the pairs of mutex_lock_op()/mutex_unlock_op() and mutex_lock_all()/mutex_unlock_all() should be used together. Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
2012-11-22 14:21:29 +07:00
if (IS_ERR(dentry_page))
return PTR_ERR(dentry_page);
dentry_blk = kmap(dentry_page);
bit_pos = room_for_filename(&dentry_blk->dentry_bitmap,
slots, NR_DENTRY_IN_BLOCK);
if (bit_pos < NR_DENTRY_IN_BLOCK)
goto add_dentry;
kunmap(dentry_page);
f2fs_put_page(dentry_page, 1);
}
/* Move to next level to find the empty slot for new dentry */
++level;
goto start;
add_dentry:
f2fs_wait_on_page_writeback(dentry_page, DATA);
down_write(&F2FS_I(inode)->i_sem);
f2fs: avoid deadlock on init_inode_metadata Previously, init_inode_metadata does not hold any parent directory's inode page. So, f2fs_init_acl can grab its parent inode page without any problem. But, when we use inline_dentry, that page is grabbed during f2fs_add_link, so that we can fall into deadlock condition like below. INFO: task mknod:11006 blocked for more than 120 seconds. Tainted: G OE 3.17.0-rc1+ #13 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. mknod D ffff88003fc94580 0 11006 11004 0x00000000 ffff880007717b10 0000000000000002 ffff88003c323220 ffff880007717fd8 0000000000014580 0000000000014580 ffff88003daecb30 ffff88003c323220 ffff88003fc94e80 ffff88003ffbb4e8 ffff880007717ba0 0000000000000002 Call Trace: [<ffffffff8173dc40>] ? bit_wait+0x50/0x50 [<ffffffff8173d4cd>] io_schedule+0x9d/0x130 [<ffffffff8173dc6c>] bit_wait_io+0x2c/0x50 [<ffffffff8173da3b>] __wait_on_bit_lock+0x4b/0xb0 [<ffffffff811640a7>] __lock_page+0x67/0x70 [<ffffffff810acf50>] ? autoremove_wake_function+0x40/0x40 [<ffffffff811652cc>] pagecache_get_page+0x14c/0x1e0 [<ffffffffa029afa9>] get_node_page+0x59/0x130 [f2fs] [<ffffffffa02a63ad>] read_all_xattrs+0x24d/0x430 [f2fs] [<ffffffffa02a6ca2>] f2fs_getxattr+0x52/0xe0 [f2fs] [<ffffffffa02a7481>] f2fs_get_acl+0x41/0x2d0 [f2fs] [<ffffffff8122d847>] get_acl+0x47/0x70 [<ffffffff8122db5a>] posix_acl_create+0x5a/0x150 [<ffffffffa02a7759>] f2fs_init_acl+0x29/0xcb [f2fs] [<ffffffffa0286a8d>] init_inode_metadata+0x5d/0x340 [f2fs] [<ffffffffa029253a>] f2fs_add_inline_entry+0x12a/0x2e0 [f2fs] [<ffffffffa0286ea5>] __f2fs_add_link+0x45/0x4a0 [f2fs] [<ffffffffa028b5b6>] ? f2fs_new_inode+0x146/0x220 [f2fs] [<ffffffffa028b816>] f2fs_mknod+0x86/0xf0 [f2fs] [<ffffffff811e3ec1>] vfs_mknod+0xe1/0x160 [<ffffffff811e4b26>] SyS_mknod+0x1f6/0x200 [<ffffffff81741d7f>] tracesys+0xe1/0xe6 Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-10-14 09:42:53 +07:00
page = init_inode_metadata(inode, dir, name, NULL);
if (IS_ERR(page)) {
err = PTR_ERR(page);
goto fail;
}
make_dentry_ptr(&d, (void *)dentry_blk, 1);
f2fs_update_dentry(inode, &d, name, dentry_hash, bit_pos);
set_page_dirty(dentry_page);
/* we don't need to mark_inode_dirty now */
F2FS_I(inode)->i_pino = dir->i_ino;
update_inode(inode, page);
f2fs_put_page(page, 1);
update_parent_metadata(dir, inode, current_depth);
fail:
up_write(&F2FS_I(inode)->i_sem);
if (is_inode_flag_set(F2FS_I(dir), FI_UPDATE_DIR)) {
update_inode_page(dir);
clear_inode_flag(F2FS_I(dir), FI_UPDATE_DIR);
}
kunmap(dentry_page);
f2fs_put_page(dentry_page, 1);
return err;
}
int f2fs_do_tmpfile(struct inode *inode, struct inode *dir)
{
struct page *page;
int err = 0;
down_write(&F2FS_I(inode)->i_sem);
f2fs: avoid deadlock on init_inode_metadata Previously, init_inode_metadata does not hold any parent directory's inode page. So, f2fs_init_acl can grab its parent inode page without any problem. But, when we use inline_dentry, that page is grabbed during f2fs_add_link, so that we can fall into deadlock condition like below. INFO: task mknod:11006 blocked for more than 120 seconds. Tainted: G OE 3.17.0-rc1+ #13 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. mknod D ffff88003fc94580 0 11006 11004 0x00000000 ffff880007717b10 0000000000000002 ffff88003c323220 ffff880007717fd8 0000000000014580 0000000000014580 ffff88003daecb30 ffff88003c323220 ffff88003fc94e80 ffff88003ffbb4e8 ffff880007717ba0 0000000000000002 Call Trace: [<ffffffff8173dc40>] ? bit_wait+0x50/0x50 [<ffffffff8173d4cd>] io_schedule+0x9d/0x130 [<ffffffff8173dc6c>] bit_wait_io+0x2c/0x50 [<ffffffff8173da3b>] __wait_on_bit_lock+0x4b/0xb0 [<ffffffff811640a7>] __lock_page+0x67/0x70 [<ffffffff810acf50>] ? autoremove_wake_function+0x40/0x40 [<ffffffff811652cc>] pagecache_get_page+0x14c/0x1e0 [<ffffffffa029afa9>] get_node_page+0x59/0x130 [f2fs] [<ffffffffa02a63ad>] read_all_xattrs+0x24d/0x430 [f2fs] [<ffffffffa02a6ca2>] f2fs_getxattr+0x52/0xe0 [f2fs] [<ffffffffa02a7481>] f2fs_get_acl+0x41/0x2d0 [f2fs] [<ffffffff8122d847>] get_acl+0x47/0x70 [<ffffffff8122db5a>] posix_acl_create+0x5a/0x150 [<ffffffffa02a7759>] f2fs_init_acl+0x29/0xcb [f2fs] [<ffffffffa0286a8d>] init_inode_metadata+0x5d/0x340 [f2fs] [<ffffffffa029253a>] f2fs_add_inline_entry+0x12a/0x2e0 [f2fs] [<ffffffffa0286ea5>] __f2fs_add_link+0x45/0x4a0 [f2fs] [<ffffffffa028b5b6>] ? f2fs_new_inode+0x146/0x220 [f2fs] [<ffffffffa028b816>] f2fs_mknod+0x86/0xf0 [f2fs] [<ffffffff811e3ec1>] vfs_mknod+0xe1/0x160 [<ffffffff811e4b26>] SyS_mknod+0x1f6/0x200 [<ffffffff81741d7f>] tracesys+0xe1/0xe6 Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
2014-10-14 09:42:53 +07:00
page = init_inode_metadata(inode, dir, NULL, NULL);
if (IS_ERR(page)) {
err = PTR_ERR(page);
goto fail;
}
/* we don't need to mark_inode_dirty now */
update_inode(inode, page);
f2fs_put_page(page, 1);
clear_inode_flag(F2FS_I(inode), FI_NEW_INODE);
fail:
up_write(&F2FS_I(inode)->i_sem);
return err;
}
void f2fs_drop_nlink(struct inode *dir, struct inode *inode, struct page *page)
{
struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
down_write(&F2FS_I(inode)->i_sem);
if (S_ISDIR(inode->i_mode)) {
drop_nlink(dir);
if (page)
update_inode(dir, page);
else
update_inode_page(dir);
}
inode->i_ctime = CURRENT_TIME;
drop_nlink(inode);
if (S_ISDIR(inode->i_mode)) {
drop_nlink(inode);
i_size_write(inode, 0);
}
up_write(&F2FS_I(inode)->i_sem);
update_inode_page(inode);
if (inode->i_nlink == 0)
add_orphan_inode(sbi, inode->i_ino);
else
release_orphan_inode(sbi);
}
/*
* It only removes the dentry from the dentry page, corresponding name
* entry in name page does not need to be touched during deletion.
*/
void f2fs_delete_entry(struct f2fs_dir_entry *dentry, struct page *page,
struct inode *dir, struct inode *inode)
{
struct f2fs_dentry_block *dentry_blk;
unsigned int bit_pos;
int slots = GET_DENTRY_SLOTS(le16_to_cpu(dentry->name_len));
int i;
if (f2fs_has_inline_dentry(dir))
return f2fs_delete_inline_entry(dentry, page, dir, inode);
lock_page(page);
f2fs_wait_on_page_writeback(page, DATA);
dentry_blk = page_address(page);
bit_pos = dentry - dentry_blk->dentry;
for (i = 0; i < slots; i++)
clear_bit_le(bit_pos + i, &dentry_blk->dentry_bitmap);
/* Let's check and deallocate this dentry page */
bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
NR_DENTRY_IN_BLOCK,
0);
kunmap(page); /* kunmap - pair of f2fs_find_entry */
set_page_dirty(page);
dir->i_ctime = dir->i_mtime = CURRENT_TIME;
if (inode)
f2fs_drop_nlink(dir, inode, NULL);
if (bit_pos == NR_DENTRY_IN_BLOCK) {
truncate_hole(dir, page->index, page->index + 1);
clear_page_dirty_for_io(page);
ClearPagePrivate(page);
ClearPageUptodate(page);
inode_dec_dirty_pages(dir);
}
f2fs_put_page(page, 1);
}
bool f2fs_empty_dir(struct inode *dir)
{
unsigned long bidx;
struct page *dentry_page;
unsigned int bit_pos;
struct f2fs_dentry_block *dentry_blk;
unsigned long nblock = dir_blocks(dir);
if (f2fs_has_inline_dentry(dir))
return f2fs_empty_inline_dir(dir);
for (bidx = 0; bidx < nblock; bidx++) {
dentry_page = get_lock_data_page(dir, bidx);
if (IS_ERR(dentry_page)) {
if (PTR_ERR(dentry_page) == -ENOENT)
continue;
else
return false;
}
dentry_blk = kmap_atomic(dentry_page);
if (bidx == 0)
bit_pos = 2;
else
bit_pos = 0;
bit_pos = find_next_bit_le(&dentry_blk->dentry_bitmap,
NR_DENTRY_IN_BLOCK,
bit_pos);
kunmap_atomic(dentry_blk);
f2fs_put_page(dentry_page, 1);
if (bit_pos < NR_DENTRY_IN_BLOCK)
return false;
}
return true;
}
bool f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
unsigned int start_pos)
{
unsigned char d_type = DT_UNKNOWN;
unsigned int bit_pos;
struct f2fs_dir_entry *de = NULL;
bit_pos = ((unsigned long)ctx->pos % d->max);
while (bit_pos < d->max) {
bit_pos = find_next_bit_le(d->bitmap, d->max, bit_pos);
if (bit_pos >= d->max)
break;
de = &d->dentry[bit_pos];
if (de->file_type < F2FS_FT_MAX)
d_type = f2fs_filetype_table[de->file_type];
else
d_type = DT_UNKNOWN;
if (!dir_emit(ctx, d->filename[bit_pos],
le16_to_cpu(de->name_len),
le32_to_cpu(de->ino), d_type))
return true;
bit_pos += GET_DENTRY_SLOTS(le16_to_cpu(de->name_len));
ctx->pos = start_pos + bit_pos;
}
return false;
}
static int f2fs_readdir(struct file *file, struct dir_context *ctx)
{
struct inode *inode = file_inode(file);
unsigned long npages = dir_blocks(inode);
struct f2fs_dentry_block *dentry_blk = NULL;
struct page *dentry_page = NULL;
struct file_ra_state *ra = &file->f_ra;
unsigned int n = ((unsigned long)ctx->pos / NR_DENTRY_IN_BLOCK);
struct f2fs_dentry_ptr d;
if (f2fs_has_inline_dentry(inode))
return f2fs_read_inline_dir(file, ctx);
/* readahead for multi pages of dir */
if (npages - n > 1 && !ra_has_index(ra, n))
page_cache_sync_readahead(inode->i_mapping, ra, file, n,
min(npages - n, (pgoff_t)MAX_DIR_RA_PAGES));
for (; n < npages; n++) {
dentry_page = get_lock_data_page(inode, n);
if (IS_ERR(dentry_page))
continue;
dentry_blk = kmap(dentry_page);
make_dentry_ptr(&d, (void *)dentry_blk, 1);
if (f2fs_fill_dentries(ctx, &d, n * NR_DENTRY_IN_BLOCK))
goto stop;
ctx->pos = (n + 1) * NR_DENTRY_IN_BLOCK;
kunmap(dentry_page);
f2fs_put_page(dentry_page, 1);
dentry_page = NULL;
}
stop:
if (dentry_page && !IS_ERR(dentry_page)) {
kunmap(dentry_page);
f2fs_put_page(dentry_page, 1);
}
return 0;
}
const struct file_operations f2fs_dir_operations = {
.llseek = generic_file_llseek,
.read = generic_read_dir,
.iterate = f2fs_readdir,
.fsync = f2fs_sync_file,
.unlocked_ioctl = f2fs_ioctl,
};