* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*/
/*
#include <linux/lustre_compat25.h>
#include <linux/lprocfs_status.h>
-#ifdef EXT3_MULTIBLOCK_ALLOCATOR
#ifdef HAVE_EXT4_LDISKFS
#include <ext4/ext4_extents.h>
#else
#include <linux/ext3_extents.h>
#endif
-#endif
#include "lustre_quota_fmt.h"
#define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
#endif
+#ifdef EXT_INSERT_EXTENT_WITH_5ARGS
+#define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
+ ext3_ext_insert_extent(handle, inode, path, newext, flag)
+#else
+#define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
+ ext3_ext_insert_extent(handle, inode, path, newext)
+#endif
+
+
static cfs_mem_cache_t *fcb_cache;
struct fsfilt_cb_data {
#define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
#endif
+#ifndef ext3_find_next_bit
+#define ext3_find_next_bit ext2_find_next_bit
+#endif
+
+#ifndef ext2_find_next_bit
+#ifdef __LITTLE_ENDIAN
+#define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
+#else
+error "Need implementation of find_next_bit on big-endian systems"
+#endif /* __LITTLE_ENDIAN */
+#endif /* !ext2_find_next_le_bit */
+
static char *fsfilt_ext3_get_label(struct super_block *sb)
{
return EXT3_SB(sb)->s_es->s_volume_name;
nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
break;
- case FSFILT_OP_JOIN:
- /* delete 2 file(file + array id) + create 1 file (array id)
- * create/update logs for each stripe */
- nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
-
- /*create array log for head file*/
- nblocks += 3;
- nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
- FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
- /*update head file array */
- nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
- FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
- break;
default: CERROR("unknown transaction start op %d\n", op);
LBUG();
}
if (iattr->ia_valid & ATTR_MODE) {
inode->i_mode = iattr->ia_mode;
- if (!in_group_p(inode->i_gid) &&
+ if (!cfs_curproc_is_in_groups(inode->i_gid) &&
!cfs_capable(CFS_CAP_FSETID))
inode->i_mode &= ~S_ISGID;
}
ENTRY;
/* FIXME: Can't do this because of nested transaction deadlock */
- if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
- CERROR("can't set data journal flag on file\n");
- RETURN(-EPERM);
+ if (cmd == EXT3_IOC_SETFLAGS) {
+ /* We can't enable data journaling on OST objects, because
+ * this forces the transaction to be closed in order to
+ * flush the journal, but the caller will already have a
+ * compound transaction open to update the last_rcvd file,
+ * and this thread would deadlock trying to set the flag. */
+ if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
+ CERROR("can't set data journal flag on file\n");
+ RETURN(-EPERM);
+ }
+ /* Because the MDS does not see the EXTENTS_FL set on the
+ * OST objects, mask this flag into all set flags. It is
+ * not legal to clear this flag in any case, so we are not
+ * changing the functionality by doing this. b=22911 */
+ *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
}
#ifdef HAVE_EXT4_LDISKFS
return ext3_force_commit(sb);
}
-#if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
-#warning "kernel code has old extents/mballoc patch, disabling"
-#undef EXT3_MULTIBLOCK_ALLOCATOR
-#endif
#ifndef EXT3_EXTENTS_FL
#define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
#endif
-#ifdef EXT3_MULTIBLOCK_ALLOCATOR
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
-#define fsfilt_up_truncate_sem(inode) up(&EXT3_I(inode)->truncate_sem);
-#define fsfilt_down_truncate_sem(inode) down(&EXT3_I(inode)->truncate_sem);
-#else
-#ifdef HAVE_EXT4_LDISKFS
-#define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem));
-#define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem));
+# define fsfilt_up_truncate_sem(inode) up(&LDISKFS_I(inode)->truncate_sem);
+# define fsfilt_down_truncate_sem(inode) down(&LDISKFS_I(inode)->truncate_sem);
#else
-#define fsfilt_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex);
-#define fsfilt_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex);
-#endif
+# ifdef HAVE_EXT4_LDISKFS
+# ifdef WALK_SPACE_HAS_DATA_SEM /* We only use it in fsfilt_map_nblocks() for now */
+# define fsfilt_up_truncate_sem(inode) do{ }while(0)
+# define fsfilt_down_truncate_sem(inode) do{ }while(0)
+# else
+# define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem))
+# define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem))
+# endif
+# else
+# define fsfilt_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex)
+# define fsfilt_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex)
+# endif
#endif
#ifndef EXT_ASSERT
nex.ee_block = cpu_to_le32(cex->ec_block);
ext3_ext_store_pblock(&nex, pblock);
nex.ee_len = cpu_to_le16(count);
- err = ext3_ext_insert_extent(handle, base, path, &nex);
+ err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
if (err) {
/* free data blocks we just allocated */
/* not a good idea to call discard here directly,
cleanup:
return rc;
}
-#endif /* EXT3_MULTIBLOCK_ALLOCATOR */
extern int ext3_map_inode_page(struct inode *inode, struct page *page,
unsigned long *blocks, int *created, int create);
int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
int pages, unsigned long *blocks,
int *created, int create,
- struct semaphore *optional_sem)
+ cfs_semaphore_t *optional_sem)
{
int rc;
-#ifdef EXT3_MULTIBLOCK_ALLOCATOR
+
if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
blocks, created, create);
return rc;
}
-#endif
if (optional_sem != NULL)
- down(optional_sem);
+ cfs_down(optional_sem);
rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
created, create);
if (optional_sem != NULL)
- up(optional_sem);
+ cfs_up(optional_sem);
return rc;
}
int err, blocksize, csize, boffs, osize = size;
/* prevent reading after eof */
- lock_kernel();
+ cfs_lock_kernel();
if (i_size_read(inode) < *offs + size) {
size = i_size_read(inode) - *offs;
- unlock_kernel();
+ cfs_unlock_kernel();
if (size < 0) {
- CERROR("size %llu is too short for read %u@%llu\n",
- i_size_read(inode), size, *offs);
- return -EIO;
+ CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
+ i_size_read(inode), *offs);
+ return -EBADR;
} else if (size == 0) {
return 0;
}
} else {
- unlock_kernel();
+ cfs_unlock_kernel();
}
blocksize = 1 << inode->i_blkbits;
/* correct in-core and on-disk sizes */
if (new_size > i_size_read(inode)) {
- lock_kernel();
+ cfs_lock_kernel();
if (new_size > i_size_read(inode))
i_size_write(inode, new_size);
if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
EXT3_I(inode)->i_disksize = i_size_read(inode);
if (i_size_read(inode) > old_size)
mark_inode_dirty(inode);
- unlock_kernel();
+ cfs_unlock_kernel();
}
if (err == 0)
}
struct chk_dqblk{
- struct hlist_node dqb_hash; /** quotacheck hash */
- struct list_head dqb_list; /** in list also */
+ cfs_hlist_node_t dqb_hash; /** quotacheck hash */
+ cfs_list_t dqb_list; /** in list also */
qid_t dqb_id; /** uid/gid */
short dqb_type; /** USRQUOTA/GRPQUOTA */
qsize_t dqb_bhardlimit; /** block hard limit */
}
static inline struct chk_dqblk *
-find_chkquot(struct hlist_head *head, qid_t id, int type)
+find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
{
- struct hlist_node *node;
+ cfs_hlist_node_t *node;
struct chk_dqblk *cdqb;
- hlist_for_each(node, head) {
- cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
+ cfs_hlist_for_each(node, head) {
+ cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
if (cdqb->dqb_id == id && cdqb->dqb_type == type)
return cdqb;
}
OBD_ALLOC_PTR(cdqb);
if (cdqb) {
- INIT_HLIST_NODE(&cdqb->dqb_hash);
- INIT_LIST_HEAD(&cdqb->dqb_list);
+ CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
+ CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
cdqb->dqb_id = id;
cdqb->dqb_type = type;
}
}
static struct chk_dqblk *
-cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
- qid_t id, int type, int first_check)
+cqget(struct super_block *sb, cfs_hlist_head_t *hash,
+ cfs_list_t *list, qid_t id, int type, int first_check)
{
- struct hlist_head *head = hash + chkquot_hash(id, type);
+ cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
struct if_dqblk dqb;
struct chk_dqblk *cdqb;
int rc;
}
}
- hlist_add_head(&cdqb->dqb_hash, head);
- list_add_tail(&cdqb->dqb_list, list);
+ cfs_hlist_add_head(&cdqb->dqb_hash, head);
+ cfs_list_add_tail(&cdqb->dqb_list, list);
return cdqb;
}
RETURN(rc);
}
+#ifndef HAVE_EXT4_LDISKFS
static inline struct ext3_group_desc *
-get_group_desc(struct super_block *sb, int group)
+get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
{
unsigned long desc_block, desc;
struct ext3_group_desc *gdp;
return gdp + desc;
}
-
-#ifndef HAVE_EXT4_LDISKFS
static inline struct buffer_head *
ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
{
struct ext3_group_desc *desc;
struct buffer_head *bh;
- desc = get_group_desc(sb, group);
+ desc = get_group_desc(sb, group, NULL);
bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
return bh;
}
-#endif
-
-static inline struct inode *ext3_iget_inuse(struct super_block *sb,
- struct buffer_head *bitmap_bh,
- int index, unsigned long ino)
-{
- struct inode *inode = NULL;
-
- if (ext3_test_bit(index, bitmap_bh->b_data))
-#ifdef HAVE_EXT4_LDISKFS
- inode = ext4_iget(sb, ino);
- if (IS_ERR(inode))
- /* Newer kernels return an error instead of a NULL pointer */
- inode = NULL;
+static __u32 ext3_itable_unused_count(struct super_block *sb,
+ struct ext3_group_desc *bg) {
+ return le16_to_cpu(bg->bg_itable_unused);
+}
#else
- inode = iget(sb, ino);
+#define get_group_desc ext3_get_group_desc
#endif
- return inode;
-}
struct qchk_ctxt {
- struct hlist_head qckt_hash[NR_DQHASH]; /* quotacheck hash */
- struct list_head qckt_list; /* quotacheck list */
+ cfs_hlist_head_t qckt_hash[NR_DQHASH]; /* quotacheck hash */
+ cfs_list_t qckt_list; /* quotacheck list */
int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
- struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
+ struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
};
static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
struct chk_dqblk *cdqb, *tmp;
int rc;
- list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
+ cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
if (!error) {
rc = commit_chkquot(sb, qctxt, cdqb);
if (rc)
error = rc;
}
- hlist_del_init(&cdqb->dqb_hash);
- list_del(&cdqb->dqb_list);
+ cfs_hlist_del_init(&cdqb->dqb_hash);
+ cfs_list_del(&cdqb->dqb_list);
OBD_FREE_PTR(cdqb);
}
return error;
}
+#ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
+#define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
+#endif
+
static int fsfilt_ext3_quotacheck(struct super_block *sb,
struct obd_quotactl *oqc)
{
struct ext3_sb_info *sbi = EXT3_SB(sb);
- int i, group;
+ int i, group, uninit_feat = 0;
struct qchk_ctxt *qctxt;
struct buffer_head *bitmap_bh = NULL;
- unsigned long ino;
+ unsigned long ino, inode_inuse;
struct inode *inode;
int rc = 0;
ENTRY;
}
for (i = 0; i < NR_DQHASH; i++)
- INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
- INIT_LIST_HEAD(&qctxt->qckt_list);
+ CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
+ CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
for (i = 0; i < MAXQUOTAS; i++) {
if (!Q_TYPESET(oqc, i))
GOTO(out, rc);
}
}
+ if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
+ /* This filesystem supports the uninit group feature */
+ uninit_feat = 1;
- /* check quota and update in hash */
- for (group = 0; group < sbi->s_groups_count; group++) {
- struct ext3_group_desc *desc;
- desc = get_group_desc(sb, group);
- if (!desc)
- GOTO(out, -EIO);
+ /* number of inodes that have been allocated */
+ inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
+ percpu_counter_sum(&sbi->s_freeinodes_counter);
- spin_lock(sb_bgl_lock(sbi, group));
- if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
- /* no inode in use in this group, just skip it */
+ /* check quota and update in hash */
+ for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
+ group++) {
+ unsigned long used_count = sbi->s_inodes_per_group;
+
+ if (uninit_feat) {
+ struct ext3_group_desc *desc;
+ desc = get_group_desc(sb, group, NULL);
+ if (!desc)
+ GOTO(out, -EIO);
+
+ /* we don't really need to take the group lock here,
+ * but it may be useful if one day we support online
+ * quotacheck */
+ spin_lock(sb_bgl_lock(sbi, group));
+ if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
+ /* no inode in use in this group, just skip it */
+ spin_unlock(sb_bgl_lock(sbi, group));
+ continue;
+ }
+ used_count -= ext3_itable_unused_count(sb, desc);
spin_unlock(sb_bgl_lock(sbi, group));
- continue;
}
- spin_unlock(sb_bgl_lock(sbi, group));
ino = group * sbi->s_inodes_per_group + 1;
bitmap_bh = ext3_read_inode_bitmap(sb, group);
if (!bitmap_bh) {
- CERROR("ext3_read_inode_bitmap group %d failed", group);
- GOTO(out, rc = -EIO);
+ CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
+ sb->s_id, group);
+ GOTO(out, -EIO);
}
- for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
+ i = 0;
+ while (i < used_count &&
+ (i = ext3_find_next_bit(bitmap_bh->b_data,
+ used_count, i)) < used_count) {
+ inode_inuse--;
+ i++;
+ ino = i + group * sbi->s_inodes_per_group;
if (ino < sbi->s_first_ino)
continue;
+#if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
+ inode = ext3_iget(sb, ino);
+#else
+ inode = iget(sb, ino);
+#endif
+ if (!inode || IS_ERR(inode))
+ continue;
- inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
rc = add_inode_quota(inode, qctxt, oqc);
iput(inode);
if (rc) {
GOTO(out, rc);
}
}
-
brelse(bitmap_bh);
}
* has limits but hasn't file) */
#ifdef HAVE_QUOTA_SUPPORT
for (i = 0; i < MAXQUOTAS; i++) {
- struct list_head id_list;
+ cfs_list_t id_list;
struct dquot_id *dqid, *tmp;
if (!Q_TYPESET(oqc, i))
LASSERT(sb_dqopt(sb)->files[i] != NULL);
- INIT_LIST_HEAD(&id_list);
+ CFS_INIT_LIST_HEAD(&id_list);
#ifndef KERNEL_SUPPORTS_QUOTA_READ
rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
#else
if (rc)
CERROR("read old limits failed. (rc:%d)\n", rc);
- list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
- list_del_init(&dqid->di_link);
+ cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
+ cfs_list_del_init(&dqid->di_link);
if (!rc)
cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
}
static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
- struct list_head *list)
+ cfs_list_t *list)
{
return lustre_get_qids(file, inode, type, list);
}
dquot->dq_dqb.dqb_isoftlimit ||
dquot->dq_dqb.dqb_bhardlimit ||
dquot->dq_dqb.dqb_bsoftlimit)
- clear_bit(DQ_FAKE_B, &dquot->dq_flags);
+ cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
else
- set_bit(DQ_FAKE_B, &dquot->dq_flags);
+ cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
rc = lustre_commit_dquot(dquot);
if (rc >= 0)
}
EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
+ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
+ char *buf, int count, loff_t pos)
+{
+ loff_t p = pos;
+ int rc;
+
+ if (!f && !inode) {
+ CERROR("lustre_read_quota failed for no quota file!\n");
+ libcfs_debug_dumpstack(NULL);
+ return -EINVAL;
+ }
+
+ /* Support for both adm and op quota files must be provided */
+ if (f) {
+ rc = fsfilt_ext3_read_record(f, buf, count, &p);
+ rc = rc < 0 ? rc : p - pos;
+ } else {
+ struct super_block *sb = inode->i_sb;
+ rc = sb->s_op->quota_read(sb, type, buf, count, pos);
+ }
+ return rc;
+}
+
+ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
+{
+ loff_t p = pos;
+ int rc;
+
+ /* Only adm quota files are supported, op updates are handled by vfs */
+ rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
+ rc = rc < 0 ? rc : p - pos;
+
+ return rc;
+}
+
+void *lustre_quota_journal_start(struct inode *inode, int delete)
+{
+ handle_t *handle;
+ unsigned block_count;
+
+ if (delete) {
+ /* each indirect block (+4) may become free, attaching to the
+ * header list of free blocks (+1); the data block (+1) may
+ * become a free block (+0) or a block with free dqentries (+0) */
+ block_count = (4 + 1) + 1;
+ handle = ext3_journal_start(inode,
+ block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
+ } else {
+ /* indirect blocks are touched (+4), each causes file expansion (+0) or
+ * freeblk reusage with a header update (+1); dqentry is either reused
+ * causing update of the entry block (+1), prev (+1) and next (+1) or
+ * a new block allocation (+1) with a header update (+1) */
+ block_count = (4 + 1) + 3;
+ handle = ext3_journal_start(inode,
+ block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
+
+ }
+
+ return handle;
+}
+
+void lustre_quota_journal_stop(void *handle)
+{
+ ext3_journal_stop((handle_t *)handle);
+}
+
static struct fsfilt_operations fsfilt_ext3_ops = {
.fs_type = "ext3",
.fs_owner = THIS_MODULE,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff