#include <linux/quotaops.h>
#include <linux/ext3_fs.h>
#include <linux/ext3_jbd.h>
-#include <linux/ext3_extents.h>
#include <linux/version.h>
+#include <linux/bitops.h>
+#include <linux/quota.h>
+#include <linux/quotaio_v1.h>
+#include <linux/quotaio_v2.h>
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
#include <linux/ext3_xattr.h>
#else
#endif
#include <libcfs/kp30.h>
-#include <linux/lustre_fsfilt.h>
-#include <linux/obd.h>
-#include <linux/obd_class.h>
+#include <lustre_fsfilt.h>
+#include <obd.h>
+#include <obd_class.h>
+#include <lustre_quota.h>
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
-#include <linux/module.h>
#include <linux/iobuf.h>
#endif
+#include <linux/lustre_compat25.h>
+#include <linux/lprocfs_status.h>
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,7))
-# define lock_24kernel() lock_kernel()
-# define unlock_24kernel() unlock_kernel()
+#ifdef EXT3_MULTIBLOCK_ALLOCATOR
+#include <linux/ext3_extents.h>
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
+#define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS
+#define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS
#else
-# define lock_24kernel() do {} while (0)
-# define unlock_24kernel() do {} while (0)
+#define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS(sb)
+#define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS(sb)
#endif
static kmem_cache_t *fcb_cache;
-static atomic_t fcb_cache_count = ATOMIC_INIT(0);
struct fsfilt_cb_data {
struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
fsfilt_cb_t cb_func; /* MDS/OBD completion function */
struct obd_device *cb_obd; /* MDS/OBD completion device */
- __u64 cb_last_num; /* MDS/OST last committed operation */
+ __u64 cb_last_rcvd; /* MDS/OST last committed operation */
void *cb_data; /* MDS/OST completion function data */
};
#define EXT3_XATTR_INDEX_TRUSTED 4
#endif
+static char *fsfilt_ext3_get_label(struct super_block *sb)
+{
+ return EXT3_SB(sb)->s_es->s_volume_name;
+}
+
+static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
+{
+ /* see e.g. fsfilt_ext3_write_record() */
+ journal_t *journal;
+ handle_t *handle;
+ int err;
+
+ journal = EXT3_SB(sb)->s_journal;
+ lock_24kernel();
+ handle = journal_start(journal, 1);
+ unlock_24kernel();
+ if (IS_ERR(handle)) {
+ CERROR("can't start transaction\n");
+ return(PTR_ERR(handle));
+ }
+
+ err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
+ if (err)
+ goto out;
+
+ memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
+ sizeof(EXT3_SB(sb)->s_es->s_volume_name));
+
+ err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
+
+out:
+ lock_24kernel();
+ journal_stop(handle);
+ unlock_24kernel();
+
+ return(err);
+}
+
+static char *fsfilt_ext3_uuid(struct super_block *sb)
+{
+ return EXT3_SB(sb)->s_es->s_uuid;
+}
+
/*
* We don't currently need any additional blocks for rmdir and
* unlink transactions because we are storing the OST oa_id inside
static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
int logs)
{
- /* For updates to the last recieved file */
+ /* For updates to the last received file */
int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS;
journal_t *journal;
void *handle;
goto journal_start;
}
- if (logs)
- nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
- EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
-
switch(op) {
case FSFILT_OP_RMDIR:
case FSFILT_OP_UNLINK:
/* delete one file + create/update logs for each stripe */
- nblocks += EXT3_DELETE_TRANS_BLOCKS;
- /*nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
- EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;*/
+ nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
+ nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
+ EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
break;
case FSFILT_OP_RENAME:
/* modify additional directory */
/* additional block + block bitmap + GDT for long symlink */
nblocks += 3;
/* no break */
- case FSFILT_OP_CREATE:
- /* create/update logs for each stripe */
- /*nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
- EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;*/
+ case FSFILT_OP_CREATE: {
+#if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
+ static int warned;
+ if (!warned) {
+ if (!test_opt(inode->i_sb, EXTENTS)) {
+ warned = 1;
+ } else if (((EXT3_I(inode)->i_flags &
+ cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
+ cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
+ CWARN("extent-mapped directory found - contact "
+ "CFS: support@clusterfs.com\n");
+ warned = 1;
+ }
+ }
+#endif
/* no break */
+ }
case FSFILT_OP_MKDIR:
case FSFILT_OP_MKNOD:
/* modify one inode + block bitmap + GDT */
case FSFILT_OP_LINK:
/* modify parent directory */
nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
- EXT3_DATA_TRANS_BLOCKS;
+ FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
+ /* create/update logs for each stripe */
+ nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
+ EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
break;
case FSFILT_OP_SETATTR:
/* Setattr on inode */
nblocks += 1;
+ nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
+ FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
+ /* quota chown log for each stripe */
+ nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
+ EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
break;
case FSFILT_OP_CANCEL_UNLINK:
/* blocks for log header bitmap update OR
* blocks for catalog header bitmap update + unlink of logs */
nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
- EXT3_DELETE_TRANS_BLOCKS * logs;
+ FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
break;
- case FSFILT_OP_NOOP:
- nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
+ 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 +
+ EXT3_SINGLEDATA_TRANS_BLOCKS);
+ /*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();
LASSERT(current->journal_info == desc_private);
journal = EXT3_SB(inode->i_sb)->s_journal;
if (nblocks > journal->j_max_transaction_buffers) {
- CERROR("too many credits %d for op %ux%u using %d instead\n",
- nblocks, op, logs, journal->j_max_transaction_buffers);
+ CWARN("too many credits %d for op %ux%u using %d instead\n",
+ nblocks, op, logs, journal->j_max_transaction_buffers);
nblocks = journal->j_max_transaction_buffers;
}
{
struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
__u64 next_indir;
- const int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
+ const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
int nbitmaps = 0, ngdblocks;
int needed = objcount + 1; /* inodes + superblock */
int i, j;
needed += nbitmaps + ngdblocks;
/* last_rcvd update */
- needed += EXT3_DATA_TRANS_BLOCKS;
+ needed += FSFILT_DATA_TRANS_BLOCKS(sb);
-#if defined(CONFIG_QUOTA) && !defined(__x86_64__) /* XXX */
+#if defined(CONFIG_QUOTA)
/* We assume that there will be 1 bit set in s_dquot.flags for each
* quota file that is active. This is at least true for now.
*/
RETURN(handle);
}
-static int fsfilt_ext3_commit(struct super_block *sb, struct inode *inode,
- void *h, int force_sync)
+static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
+{
+ handle_t *handle = h;
+
+ /* fsfilt_extend called with nblocks = 0 for testing in special cases */
+ if (nblocks == 0) {
+ handle->h_buffer_credits = 0;
+ CWARN("setting credits of handle %p to zero by request\n", h);
+ }
+
+ if (handle->h_buffer_credits > nblocks)
+ return 0;
+ if (journal_extend(handle, nblocks) == 0)
+ return 0;
+
+ ext3_mark_inode_dirty(handle, inode);
+ return journal_restart(handle, nblocks);
+}
+
+static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
{
int rc;
handle_t *handle = h;
rc = journal_stop(handle);
if (rc) {
CERROR("error while stopping transaction: %d\n", rc);
- unlock_kernel();
+ unlock_24kernel();
return rc;
}
#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
struct iattr *iattr, int do_trunc)
{
struct inode *inode = dentry->d_inode;
- int rc;
+ int rc = 0;
- lock_24kernel();
+ lock_kernel();
- /* A _really_ horrible hack to avoid removing the data stored
- * in the block pointers; this is really the "small" stripe MD data.
- * We can avoid further hackery by virtue of the MDS file size being
- * zero all the time (which doesn't invoke block truncate at unlink
- * time), so we assert we never change the MDS file size from zero. */
+ /* Avoid marking the inode dirty on the superblock list unnecessarily.
+ * We are already writing the inode to disk as part of this
+ * transaction and want to avoid a lot of extra inode writeout
+ * later on. b=9828 */
if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
/* ATTR_SIZE would invoke truncate: clear it */
iattr->ia_valid &= ~ATTR_SIZE;
EXT3_I(inode)->i_disksize = inode->i_size = iattr->ia_size;
- /* make sure _something_ gets set - so new inode
- * goes to disk (probably won't work over XFS */
- if (!(iattr->ia_valid & (ATTR_MODE | ATTR_MTIME | ATTR_CTIME))){
- iattr->ia_valid |= ATTR_MODE;
- iattr->ia_mode = inode->i_mode;
+ if (iattr->ia_valid & ATTR_UID)
+ inode->i_uid = iattr->ia_uid;
+ if (iattr->ia_valid & ATTR_GID)
+ inode->i_gid = iattr->ia_gid;
+ if (iattr->ia_valid & ATTR_ATIME)
+ inode->i_atime = iattr->ia_atime;
+ if (iattr->ia_valid & ATTR_MTIME)
+ inode->i_mtime = iattr->ia_mtime;
+ if (iattr->ia_valid & ATTR_CTIME)
+ inode->i_ctime = iattr->ia_ctime;
+ if (iattr->ia_valid & ATTR_MODE) {
+ inode->i_mode = iattr->ia_mode;
+
+ if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
+ inode->i_mode &= ~S_ISGID;
}
+
+ inode->i_sb->s_op->dirty_inode(inode);
+
+ goto out;
}
/* Don't allow setattr to change file type */
- iattr->ia_mode = (inode->i_mode & S_IFMT)|(iattr->ia_mode & ~S_IFMT);
+ if (iattr->ia_valid & ATTR_MODE)
+ iattr->ia_mode = (inode->i_mode & S_IFMT) |
+ (iattr->ia_mode & ~S_IFMT);
/* We set these flags on the client, but have already checked perms
* so don't confuse inode_change_ok. */
rc = inode_setattr(inode, iattr);
}
- unlock_24kernel();
-
- return rc;
+ out:
+ unlock_kernel();
+ RETURN(rc);
}
static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
int rc = 0;
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 (inode->i_fop->ioctl)
rc = inode->i_fop->ioctl(inode, file, cmd, arg);
else
RETURN(rc);
}
-static int fsfilt_ext3_set_xattr(struct inode * inode, void *handle, char *name,
- void *buffer, int buffer_size)
+static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
+ void *lmm, int lmm_size, const char *name)
{
- int rc = 0;
+ int rc;
- lock_24kernel();
+ LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
+ lock_24kernel();
rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
- name, buffer, buffer_size, 0);
+ name, lmm, lmm_size, 0);
+
unlock_24kernel();
+
if (rc)
- CERROR("set xattr %s from inode %lu: rc %d\n",
- name, inode->i_ino, rc);
+ CERROR("error adding MD data to inode %lu: rc = %d\n",
+ inode->i_ino, rc);
return rc;
}
-static int fsfilt_ext3_get_xattr(struct inode *inode, char *name,
- void *buffer, int buffer_size)
+/* Must be called with i_mutex held */
+static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
+ const char *name)
{
- int rc = 0;
-
+ int rc;
+
+ LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
lock_24kernel();
rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
- name, buffer, buffer_size);
+ name, lmm, lmm_size);
unlock_24kernel();
- if (buffer == NULL)
+ /* This gives us the MD size */
+ if (lmm == NULL)
return (rc == -ENODATA) ? 0 : rc;
+
if (rc < 0) {
- CDEBUG(D_INFO, "error getting EA %s from inode %lu: rc %d\n",
- name, inode->i_ino, rc);
- memset(buffer, 0, buffer_size);
+ CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
+ EXT3_XATTR_INDEX_TRUSTED, name,
+ inode->i_ino, rc);
+ memset(lmm, 0, lmm_size);
return (rc == -ENODATA) ? 0 : rc;
}
return rc;
}
-static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
- void *lmm, int lmm_size,
- enum ea_type type)
-{
- int rc;
-
- switch(type) {
- case EA_LOV:
- rc = fsfilt_ext3_set_xattr(inode, handle,
- XATTR_LUSTRE_MDS_LOV_EA,
- lmm, lmm_size);
- break;
- case EA_MEA:
- rc = fsfilt_ext3_set_xattr(inode, handle,
- XATTR_LUSTRE_MDS_MEA_EA,
- lmm, lmm_size);
- break;
- case EA_SID:
- rc = fsfilt_ext3_set_xattr(inode, handle,
- XATTR_LUSTRE_MDS_SID_EA,
- lmm, lmm_size);
- break;
- case EA_PID:
- rc = fsfilt_ext3_set_xattr(inode, handle,
- XATTR_LUSTRE_MDS_PID_EA,
- lmm, lmm_size);
- break;
- case EA_KEY:
- rc = fsfilt_ext3_set_xattr(inode, handle,
- XATTR_LUSTRE_MDS_KEY_EA,
- lmm, lmm_size);
- break;
- default:
- return -EINVAL;
- }
-
- return rc;
-}
-
-static int fsfilt_ext3_get_md(struct inode *inode, void *lmm,
- int lmm_size, enum ea_type type)
+#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
+static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
{
- int rc;
-
- switch (type) {
- case EA_LOV:
- rc = fsfilt_ext3_get_xattr(inode,
- XATTR_LUSTRE_MDS_LOV_EA,
- lmm, lmm_size);
- break;
- case EA_MEA:
- rc = fsfilt_ext3_get_xattr(inode,
- XATTR_LUSTRE_MDS_MEA_EA,
- lmm, lmm_size);
- break;
- case EA_SID:
- rc = fsfilt_ext3_get_xattr(inode,
- XATTR_LUSTRE_MDS_SID_EA,
- lmm, lmm_size);
- break;
- case EA_PID:
- rc = fsfilt_ext3_get_xattr(inode,
- XATTR_LUSTRE_MDS_PID_EA,
- lmm, lmm_size);
- break;
- case EA_KEY:
- rc = fsfilt_ext3_get_xattr(inode,
- XATTR_LUSTRE_MDS_KEY_EA,
- lmm, lmm_size);
- break;
- default:
- return -EINVAL;
- }
-
- return rc;
+ submit_bio(rw, bio);
+ return 0;
}
-
-static int fsfilt_ext3_send_bio(int rw, struct inode *inode, void *bio)
-{
- int rc = 0;
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
- submit_bio(rw, (struct bio *)bio);
#else
- struct bio *b = (struct kiobuf *)bio;
- int blocks_per_page;
-
- rc = brw_kiovec(rw, 1, &b, inode->i_dev,
- b->blocks, 1 << inode->i_blkbits);
-
- blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
-
- if (rc != (1 << inode->i_blkbits) * b->nr_pages * blocks_per_page) {
- CERROR("short write? expected %d, wrote %d\n",
- (1 << inode->i_blkbits) * b->nr_pages *
- blocks_per_page, rc);
- }
-#endif
- return rc;
-}
-
-static struct page *fsfilt_ext3_getpage(struct inode *inode, long int index)
+static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct kiobuf *bio)
{
- int rc;
- struct page *page;
+ int rc, blk_per_page;
+
+ rc = brw_kiovec(rw, 1, &bio, inode->i_dev,
+ KIOBUF_GET_BLOCKS(bio), 1 << inode->i_blkbits);
+ /*
+ * brw_kiovec() returns number of bytes actually written. If error
+ * occurred after something was written, error code is returned though
+ * kiobuf->errno. (See bug 6854.)
+ */
- page = grab_cache_page(inode->i_mapping, index);
- if (page == NULL)
- return ERR_PTR(-ENOMEM);
+ blk_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
- if (PageUptodate(page)) {
- unlock_page(page);
- return page;
+ if (rc != (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page) {
+ CERROR("short write? expected %d, wrote %d (%d)\n",
+ (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
+ rc, bio->errno);
}
-
- rc = inode->i_mapping->a_ops->readpage(NULL, page);
- if (rc < 0) {
- page_cache_release(page);
- return ERR_PTR(rc);
+ if (bio->errno != 0) {
+ CERROR("IO error. Wrote %d of %d (%d)\n",
+ rc,
+ (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
+ bio->errno);
+ rc = bio->errno;
}
- return page;
+ return rc;
}
+#endif
static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
loff_t *off)
CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
memset(fake, 0, sizeof(*fake));
- fake->rec_len = cpu_to_le32(blksize);
+ fake->rec_len = cpu_to_le16(blksize);
}
count -= blksize;
buf += blksize;
{
struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
- fcb->cb_func(fcb->cb_obd, fcb->cb_last_num, fcb->cb_data, error);
+ fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
- atomic_dec(&fcb_cache_count);
}
-static int fsfilt_ext3_add_journal_cb(struct obd_device *obd,
- struct super_block *sb,
- __u64 last_num, void *handle,
- fsfilt_cb_t cb_func,
+static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
+ void *handle, fsfilt_cb_t cb_func,
void *cb_data)
{
struct fsfilt_cb_data *fcb;
if (fcb == NULL)
RETURN(-ENOMEM);
- atomic_inc(&fcb_cache_count);
fcb->cb_func = cb_func;
fcb->cb_obd = obd;
- fcb->cb_last_num = last_num;
+ fcb->cb_last_rcvd = last_rcvd;
fcb->cb_data = cb_data;
- CDEBUG(D_EXT2, "set callback for last_num: "LPD64"\n", last_num);
+ CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
lock_24kernel();
journal_callback_set(handle, fsfilt_ext3_cb_func,
(struct journal_callback *)fcb);
unlock_24kernel();
+
return 0;
}
int rc;
memset(&sfs, 0, sizeof(sfs));
-
- rc = sb->s_op->statfs(sb, &sfs);
+ rc = ll_do_statfs(sb,&sfs);
if (!rc && sfs.f_bfree < sfs.f_ffree) {
sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
#undef EXT3_MULTIBLOCK_ALLOCATOR
#endif
#ifndef EXT3_EXTENTS_FL
-#define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
+#define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
#endif
#ifdef EXT3_MULTIBLOCK_ALLOCATOR
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
#define ext3_up_truncate_sem(inode) up_write(&EXT3_I(inode)->truncate_sem);
#define ext3_down_truncate_sem(inode) down_write(&EXT3_I(inode)->truncate_sem);
-#else
+#elif (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
#define ext3_up_truncate_sem(inode) up(&EXT3_I(inode)->truncate_sem);
#define ext3_down_truncate_sem(inode) down(&EXT3_I(inode)->truncate_sem);
+#else
+#define ext3_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex);
+#define ext3_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex);
#endif
#include <linux/lustre_version.h>
return bg_start + colour + block;
}
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
+#include <linux/locks.h>
+static void ll_unmap_underlying_metadata(struct super_block *sb,
+ unsigned long blocknr)
+{
+ struct buffer_head *old_bh;
+
+ old_bh = get_hash_table(sb->s_dev, blocknr, sb->s_blocksize);
+ if (old_bh) {
+ mark_buffer_clean(old_bh);
+ wait_on_buffer(old_bh);
+ clear_bit(BH_Req, &old_bh->b_state);
+ __brelse(old_bh);
+ }
+}
+#else
+#define ll_unmap_underlying_metadata(sb, blocknr) \
+ unmap_underlying_metadata((sb)->s_bdev, blocknr)
+#endif
+
static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
struct ext3_ext_path *path,
struct ext3_ext_cache *cex)
struct ext3_extent nex;
int count, err, goal;
unsigned long pblock;
- loff_t new_i_size;
+ unsigned long tgen;
handle_t *handle;
int i, aflags = 0;
EXT_ASSERT(i == path->p_depth);
EXT_ASSERT(path[i].p_hdr);
- if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
+ if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
err = EXT_CONTINUE;
goto map;
}
return EXT_CONTINUE;
}
+ tgen = EXT_GENERATION(tree);
count = ext3_ext_calc_credits_for_insert(tree, path);
+ ext3_up_truncate_sem(inode);
lock_24kernel();
handle = journal_start(EXT3_JOURNAL(inode), count+EXT3_ALLOC_NEEDED+1);
unlock_24kernel();
- if (IS_ERR(handle))
+ if (IS_ERR(handle)) {
+ ext3_down_truncate_sem(inode);
return PTR_ERR(handle);
+ }
+
+ ext3_down_truncate_sem(inode);
+ if (tgen != EXT_GENERATION(tree)) {
+ /* the tree has changed. so path can be invalid at moment */
+ lock_24kernel();
+ journal_stop(handle);
+ unlock_24kernel();
+ return EXT_REPEAT;
+ }
count = cex->ec_len;
goal = ext3_ext_find_goal(inode, path, cex->ec_block, &aflags);
if (!pblock)
goto out;
EXT_ASSERT(count <= cex->ec_len);
- //if (count < cex->ec_len)
- // CDEBUG(D_ERROR, "%d < %lu\n", count, cex->ec_len);
/* insert new extent */
nex.ee_block = cex->ec_block;
BUG_ON(nex.ee_len == 0);
BUG_ON(nex.ee_block != cex->ec_block);
- /* correct on-disk inode size */
- if (nex.ee_len > 0) {
- new_i_size = (loff_t) nex.ee_block + nex.ee_len;
- new_i_size = new_i_size << inode->i_blkbits;
- if (new_i_size > EXT3_I(inode)->i_disksize) {
- EXT3_I(inode)->i_disksize = new_i_size;
- err = ext3_mark_inode_dirty(handle, inode);
- }
- }
-
out:
lock_24kernel();
journal_stop(handle);
unlock_24kernel();
map:
if (err >= 0) {
- struct block_device *bdev = inode->i_sb->s_bdev;
-
/* map blocks */
if (bp->num == 0) {
CERROR("hmm. why do we find this extent?\n");
} else {
*(bp->created) = 1;
/* unmap any possible underlying metadata from
- * the block device mapping. bug 6998.
- * This only compiles on 2.6, but there are
- * no users of mballoc on 2.4. */
- unmap_underlying_metadata(bdev, *(bp->blocks));
+ * the block device mapping. bug 6998. */
+ ll_unmap_underlying_metadata(inode->i_sb,
+ *(bp->blocks));
}
bp->created++;
bp->blocks++;
int err;
CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
- block, block + num, (unsigned) inode->i_ino);
+ block, block + num - 1, (unsigned) inode->i_ino);
ext3_init_tree_desc(&tree, inode);
tree.private = &bp;
bp.init_num = bp.num = num;
bp.create = create;
+ ext3_down_truncate_sem(inode);
err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
ext3_ext_invalidate_cache(&tree);
+ ext3_up_truncate_sem(inode);
return err;
}
int pages, unsigned long *blocks,
int *created, int create)
{
- int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
+ int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
int rc = 0, i = 0;
struct page *fp = NULL;
int clen = 0;
cleanup:
return rc;
}
-#endif
+#endif /* EXT3_MULTIBLOCK_ALLOCATOR */
extern int ext3_map_inode_page(struct inode *inode, struct page *page,
unsigned long *blocks, int *created, int create);
int pages, unsigned long *blocks,
int *created, int create)
{
- int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
+ int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
unsigned long *b;
int rc = 0, i, *cr;
int *created, int create,
struct semaphore *optional_sem)
{
- int rc = 0;
- if (optional_sem != NULL)
- down(optional_sem);
+ int rc;
#ifdef EXT3_MULTIBLOCK_ALLOCATOR
- if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL)
+ if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
blocks, created, create);
- else
+ return rc;
+ }
#endif
- rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
- created, create);
+ if (optional_sem != NULL)
+ down(optional_sem);
+ rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
+ created, create);
if (optional_sem != NULL)
up(optional_sem);
journal = EXT3_SB(inode->i_sb)->s_journal;
lock_24kernel();
handle = journal_start(journal,
- block_count * EXT3_DATA_TRANS_BLOCKS + 2);
+ block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
unlock_24kernel();
if (IS_ERR(handle)) {
- CERROR("can't start transaction\n");
+ CERROR("can't start transaction for %d blocks (%d bytes)\n",
+ block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
return PTR_ERR(handle);
}
return err;
}
-static int fsfilt_ext3_setup(struct obd_device *obd, struct super_block *sb)
+static int fsfilt_ext3_setup(struct super_block *sb)
{
-#ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
- struct mds_obd *mds = &obd->u.mds;
-#endif
#if 0
EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
sb->s_flags |= S_PDIROPS;
#endif
- /* setup mdsnum in underlying fs */
-#ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
- if (mds->mds_md_obd) {
- struct ext3_sb_info *sbi = EXT3_SB(sb);
- struct ext3_super_block *es = sbi->s_es;
- handle_t *handle;
- int err;
-
- if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_MDSNUM)) {
- CWARN("%s: set mdsnum %d in ext3\n",
- obd->obd_name, mds->mds_num);
- lock_24kernel();
- handle = journal_start(sbi->s_journal, 1);
- unlock_24kernel();
- LASSERT(!IS_ERR(handle));
- err = ext3_journal_get_write_access(handle, sbi->s_sbh);
- LASSERT(err == 0);
- EXT3_SET_INCOMPAT_FEATURE(sb,
- EXT3_FEATURE_INCOMPAT_MDSNUM);
- es->s_mdsnum = mds->mds_num;
- err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
- LASSERT(err == 0);
- lock_24kernel();
- journal_stop(handle);
- unlock_24kernel();
- } else {
- CWARN("%s: mdsnum initialized to %u in ext3fs\n",
- obd->obd_name, es->s_mdsnum);
+ if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
+ CWARN("filesystem doesn't have dir_index feature enabled\n");
+ return 0;
+}
+
+/* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
+ objects. Logs is number of logfiles to update */
+static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
+{
+ if ( !fso ) {
+ switch(op) {
+ case FSFILT_OP_CREATE:
+ /* directory leaf, index & indirect & EA*/
+ return 4 + 3 * logs;
+ case FSFILT_OP_UNLINK:
+ return 3 * logs;
}
- sbi->s_mdsnum = es->s_mdsnum;
- }
-#endif
- if (!strncmp(obd->obd_type->typ_name, "obdfilter", 9)) {
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
- request_queue_t *q = bdev_get_queue(sb->s_bdev);
- if (q->max_phys_segments * PAGE_SIZE < (1 << 20))
-#else
- if (MAX_SEGMENTS * PAGE_SIZE < (1 << 20))
-#endif
- CDEBUG(D_WARNING, "Lustre loves 1MB requests, "
- "but current max_phys_segments=%d\n",
- q->max_phys_segments);
+ } else {
+ int i;
+ int needed = 0;
+ struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
+ int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
+ int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
+ for (i = 0; i < op; i++, fso++) {
+ int nblocks = fso->fso_bufcnt * blockpp;
+ int ndindirect = min(nblocks, addrpp + 1);
+ int nindir = nblocks + ndindirect + 1;
+
+ needed += nindir;
+ }
+ return needed + 3 * logs;
}
+
return 0;
}
-static inline int ext3_match (int len, const char * const name,
- struct ext3_dir_entry_2 * de)
+static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
+
+#define DQINFO_COPY(out, in) \
+do { \
+ Q_COPY(out, in, dqi_bgrace); \
+ Q_COPY(out, in, dqi_igrace); \
+ Q_COPY(out, in, dqi_flags); \
+ Q_COPY(out, in, dqi_valid); \
+} while (0)
+
+#define DQBLK_COPY(out, in) \
+do { \
+ Q_COPY(out, in, dqb_bhardlimit); \
+ Q_COPY(out, in, dqb_bsoftlimit); \
+ Q_COPY(out, in, dqb_curspace); \
+ Q_COPY(out, in, dqb_ihardlimit); \
+ Q_COPY(out, in, dqb_isoftlimit); \
+ Q_COPY(out, in, dqb_curinodes); \
+ Q_COPY(out, in, dqb_btime); \
+ Q_COPY(out, in, dqb_itime); \
+ Q_COPY(out, in, dqb_valid); \
+} while (0)
+
+
+
+static int fsfilt_ext3_quotactl(struct super_block *sb,
+ struct obd_quotactl *oqc)
{
- if (len != de->name_len)
- return 0;
- if (!de->inode)
- return 0;
- return !memcmp(name, de->name, len);
+ int i, rc = 0, error = 0;
+ struct quotactl_ops *qcop;
+ struct if_dqinfo *info;
+ struct if_dqblk *dqblk;
+ ENTRY;
+
+ if (!sb->s_qcop)
+ RETURN(-ENOSYS);
+
+ OBD_ALLOC_PTR(info);
+ if (!info)
+ RETURN(-ENOMEM);
+ OBD_ALLOC_PTR(dqblk);
+ if (!dqblk) {
+ OBD_FREE_PTR(info);
+ RETURN(-ENOMEM);
+ }
+
+ DQINFO_COPY(info, &oqc->qc_dqinfo);
+ DQBLK_COPY(dqblk, &oqc->qc_dqblk);
+
+ qcop = sb->s_qcop;
+ if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
+ for (i = 0; i < MAXQUOTAS; i++) {
+ if (!Q_TYPESET(oqc, i))
+ continue;
+
+ if (oqc->qc_cmd == Q_QUOTAON) {
+ if (!qcop->quota_on)
+ GOTO(out, rc = -ENOSYS);
+ rc = qcop->quota_on(sb, i, oqc->qc_id,
+ (char *)op_quotafile[i]);
+ } else if (oqc->qc_cmd == Q_QUOTAOFF) {
+ if (!qcop->quota_off)
+ GOTO(out, rc = -ENOSYS);
+ rc = qcop->quota_off(sb, i);
+ }
+
+ if (rc == -EBUSY)
+ error = rc;
+ else if (rc)
+ GOTO(out, rc);
+ }
+ GOTO(out, rc ?: error);
+ }
+
+ switch (oqc->qc_cmd) {
+ case Q_GETOINFO:
+ case Q_GETINFO:
+ if (!qcop->get_info)
+ GOTO(out, rc = -ENOSYS);
+ rc = qcop->get_info(sb, oqc->qc_type, info);
+ break;
+ case Q_SETQUOTA:
+ case Q_INITQUOTA:
+ if (!qcop->set_dqblk)
+ GOTO(out, rc = -ENOSYS);
+ rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
+ break;
+ case Q_GETOQUOTA:
+ case Q_GETQUOTA:
+ if (!qcop->get_dqblk)
+ GOTO(out, rc = -ENOSYS);
+ rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
+ break;
+ case Q_SYNC:
+ if (!sb->s_qcop->quota_sync)
+ GOTO(out, rc = -ENOSYS);
+ qcop->quota_sync(sb, oqc->qc_type);
+ break;
+ default:
+ CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
+ LBUG();
+ }
+out:
+ DQINFO_COPY(&oqc->qc_dqinfo, info);
+ DQBLK_COPY(&oqc->qc_dqblk, dqblk);
+
+ OBD_FREE_PTR(info);
+ OBD_FREE_PTR(dqblk);
+
+ if (rc)
+ CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
+ "failed: %d\n",
+ oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
+ RETURN(rc);
}
-static int fsfilt_ext3_scan_for_entry(struct dentry *dentry)
+
+struct chk_dqblk{
+ struct hlist_node dqb_hash; /* quotacheck hash */
+ struct list_head dqb_list; /* in list also */
+ qid_t dqb_id; /* uid/gid */
+ short dqb_type; /* USRQUOTA/GRPQUOTA */
+ __u32 dqb_bhardlimit; /* block hard limit */
+ __u32 dqb_bsoftlimit; /* block soft limit */
+ qsize_t dqb_curspace; /* current space */
+ __u32 dqb_ihardlimit; /* inode hard limit */
+ __u32 dqb_isoftlimit; /* inode soft limit */
+ __u32 dqb_curinodes; /* current inodes */
+ __u64 dqb_btime; /* block grace time */
+ __u64 dqb_itime; /* inode grace time */
+ __u32 dqb_valid; /* flag for above fields */
+};
+
+static inline unsigned int chkquot_hash(qid_t id, int type)
+ __attribute__((__const__));
+
+static inline unsigned int chkquot_hash(qid_t id, int type)
{
- struct inode *dir = dentry->d_parent->d_inode;
- int blksize = dir->i_sb->s_blocksize;
- int blk = 0, err, num = 0, de_len;
- struct ext3_dir_entry_2 * de;
- int count = dir->i_size;
- struct qstr *name;
- char *dlimit;
-
- name = &dentry->d_name;
- while (count > 0) {
- struct buffer_head *bh;
-
- bh = ext3_bread(NULL, dir, blk++, 0, &err);
- if (bh == NULL) {
- CERROR("error read dir %lu+%d: %d\n",
- dir->i_ino, blk, err);
- break;
- }
+ return (id * (MAXQUOTAS - type)) % NR_DQHASH;
+}
- dlimit = bh->b_data + blksize;
- de = (struct ext3_dir_entry_2 *) bh->b_data;
- while ((char *) de < dlimit) {
- if ((char *) de + name->len <= dlimit &&
- ext3_match(name->len, name->name, de))
- num++;
- de_len = le16_to_cpu(de->rec_len);
- de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
- }
+static inline struct chk_dqblk *
+find_chkquot(struct hlist_head *head, qid_t id, int type)
+{
+ struct hlist_node *node;
+ struct chk_dqblk *cdqb;
- count -= blksize;
- brelse(bh);
+ hlist_for_each(node, head) {
+ cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
+ if (cdqb->dqb_id == id && cdqb->dqb_type == type)
+ return cdqb;
}
- return num;
+ return NULL;
}
-extern int ext3_add_dir_entry(struct dentry *dentry);
-extern int ext3_del_dir_entry(struct dentry *dentry);
+static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
+{
+ struct chk_dqblk *cdqb;
+
+ OBD_ALLOC_PTR(cdqb);
+ if (cdqb) {
+ INIT_HLIST_NODE(&cdqb->dqb_hash);
+ INIT_LIST_HEAD(&cdqb->dqb_list);
+ cdqb->dqb_id = id;
+ cdqb->dqb_type = type;
+ }
+
+ return cdqb;
+}
-static int fsfilt_ext3_add_dir_entry(struct obd_device *obd,
- struct dentry *parent,
- char *name, int namelen,
- unsigned long ino,
- unsigned long generation,
- unsigned long mds,
- unsigned long fid)
+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)
{
-#ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
- struct mds_obd *mdsobd = &obd->u.mds;
- struct dentry *dentry;
- int err;
- LASSERT(ino != 0);
- LASSERT(namelen != 0);
- dentry = ll_lookup_one_len(name, parent, namelen);
- if (IS_ERR(dentry)) {
- CERROR("can't lookup %*s in %lu/%lu: %d\n", dentry->d_name.len,
- dentry->d_name.name, dentry->d_inode->i_ino,
- (unsigned long) dentry->d_inode->i_generation,
- (int) PTR_ERR(dentry));
- RETURN(PTR_ERR(dentry));
+ struct hlist_head *head = hash + chkquot_hash(id, type);
+ struct if_dqblk dqb;
+ struct chk_dqblk *cdqb;
+ int rc;
+
+ cdqb = find_chkquot(head, id, type);
+ if (cdqb)
+ return cdqb;
+
+ cdqb = alloc_chkquot(id, type);
+ if (!cdqb)
+ return NULL;
+
+ if (!first_check) {
+ rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
+ if (rc) {
+ CERROR("get_dqblk of id %u, type %d failed: %d\n",
+ id, type, rc);
+ } else {
+ DQBLK_COPY(cdqb, &dqb);
+ cdqb->dqb_curspace = 0;
+ cdqb->dqb_curinodes = 0;
+ }
}
- if (dentry->d_inode != NULL || dentry->d_flags & DCACHE_CROSS_REF) {
- CERROR("dentry %*s(0x%p) found\n", dentry->d_name.len,
- dentry->d_name.name, dentry);
- l_dput(dentry);
- RETURN(-EEXIST);
+
+ hlist_add_head(&cdqb->dqb_hash, head);
+ list_add_tail(&cdqb->dqb_list, list);
+
+ return cdqb;
+}
+
+static inline int quota_onoff(struct super_block *sb, int cmd, int type)
+{
+ struct obd_quotactl *oqctl;
+ int rc;
+
+ OBD_ALLOC_PTR(oqctl);
+ if (!oqctl)
+ RETURN(-ENOMEM);
+
+ oqctl->qc_cmd = cmd;
+ oqctl->qc_id = QFMT_LDISKFS;
+ oqctl->qc_type = type;
+ rc = fsfilt_ext3_quotactl(sb, oqctl);
+
+ OBD_FREE_PTR(oqctl);
+ return rc;
+}
+
+static inline int read_old_dqinfo(struct super_block *sb, int type,
+ struct if_dqinfo *dqinfo)
+{
+ struct obd_quotactl *oqctl;
+ int rc;
+ ENTRY;
+
+ OBD_ALLOC_PTR(oqctl);
+ if (!oqctl)
+ RETURN(-ENOMEM);
+
+ oqctl->qc_cmd = Q_GETINFO;
+ oqctl->qc_type = type;
+ rc = fsfilt_ext3_quotactl(sb, oqctl);
+ if (!rc)
+ ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
+
+ OBD_FREE_PTR(oqctl);
+ RETURN(rc);
+}
+
+static inline struct ext3_group_desc *
+get_group_desc(struct super_block *sb, int group)
+{
+ unsigned long desc_block, desc;
+ struct ext3_group_desc *gdp;
+
+ desc_block = group / EXT3_DESC_PER_BLOCK(sb);
+ desc = group % EXT3_DESC_PER_BLOCK(sb);
+ gdp = (struct ext3_group_desc *)
+ EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
+
+ return gdp + desc;
+}
+
+static inline struct buffer_head *
+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);
+ bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
+
+ return bh;
+}
+
+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))
+ inode = iget(sb, ino);
+
+ return inode;
+}
+
+struct qchk_ctxt {
+ struct hlist_head qckt_hash[NR_DQHASH]; /* quotacheck hash */
+ struct list_head qckt_list; /* quotacheck list */
+ int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
+ struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
+};
+
+static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
+ struct obd_quotactl *oqc)
+{
+ struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
+ loff_t size = 0;
+ qid_t qid[MAXQUOTAS];
+ int cnt, i, rc = 0;
+
+ if (!inode)
+ return 0;
+
+ qid[USRQUOTA] = inode->i_uid;
+ qid[GRPQUOTA] = inode->i_gid;
+
+ if (S_ISDIR(inode->i_mode) ||
+ S_ISREG(inode->i_mode) ||
+ S_ISLNK(inode->i_mode))
+ size = inode_get_bytes(inode);
+
+ for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
+ if (!Q_TYPESET(oqc, cnt))
+ continue;
+
+ cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
+ &qctxt->qckt_list, qid[cnt], cnt,
+ qctxt->qckt_first_check[cnt]);
+ if (!cdqb[cnt]) {
+ rc = -ENOMEM;
+ break;
+ }
+
+ cdqb[cnt]->dqb_curspace += size;
+ cdqb[cnt]->dqb_curinodes++;
}
- /* mds_reint_rename() may use this method to add dir entry
- * that points onto local inode. and we don't want to find
- * it cross-ref by subsequent lookups */
- d_drop(dentry);
-
- if (parent->d_inode->i_ino == id_ino(&mdsobd->mds_rootid)) {
- err = fsfilt_ext3_scan_for_entry(dentry);
- if (err != 0)
- CDEBUG(D_ERROR, "existing %d entry %*s in %lu/%u\n",
- err, dentry->d_name.len, dentry->d_name.name,
- parent->d_inode->i_ino, parent->d_inode->i_generation);
+ if (rc) {
+ for (i = 0; i < cnt; i++) {
+ if (!Q_TYPESET(oqc, i))
+ continue;
+ LASSERT(cdqb[i]);
+ cdqb[i]->dqb_curspace -= size;
+ cdqb[i]->dqb_curinodes--;
+ }
}
- dentry->d_flags |= DCACHE_CROSS_REF;
- dentry->d_inum = ino;
- dentry->d_mdsnum = mds;
- dentry->d_generation = generation;
- dentry->d_fid = fid;
- lock_24kernel();
- err = ext3_add_dir_entry(dentry);
- unlock_24kernel();
-
- l_dput(dentry);
+ return rc;
+}
- RETURN(err);
-#else
-#error "rebuild kernel and lustre with ext3-mds-num patch!"
- LASSERT(0);
-#endif
+static int v2_write_dqheader(struct file *f, int type)
+{
+ static const __u32 quota_magics[] = V2_INITQMAGICS;
+ static const __u32 quota_versions[] = V2_INITQVERSIONS;
+ struct v2_disk_dqheader dqhead;
+ loff_t offset = 0;
+
+ CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
+ LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
+
+ dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
+ dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
+
+ return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
}
-static int fsfilt_ext3_del_dir_entry(struct obd_device *obd,
- struct dentry *dentry)
+/* write dqinfo struct in a new quota file */
+static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
{
-#ifdef EXT3_FEATURE_INCOMPAT_MDSNUM
- struct inode *dir = dentry->d_parent->d_inode;
- struct mds_obd *mdsobd = &obd->u.mds;
- int err;
- lock_24kernel();
- err = ext3_del_dir_entry(dentry);
- unlock_24kernel();
- if (!err && dir->i_ino == id_ino(&mdsobd->mds_rootid)) {
- int num = fsfilt_ext3_scan_for_entry(dentry);
- if (num != 0)
- CDEBUG(D_ERROR, "existing %d entry %*s in %lu/%u\n",
- err, dentry->d_name.len, dentry->d_name.name,
- dir->i_ino, dir->i_generation);
+ struct v2_disk_dqinfo dqinfo;
+ __u32 blocks = V2_DQTREEOFF + 1;
+ loff_t offset = V2_DQINFOOFF;
+
+ if (info) {
+ dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
+ dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
+ dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
+ ~DQF_INFO_DIRTY);
+ } else {
+ dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
+ dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
+ dqinfo.dqi_flags = 0;
}
- if (err == 0)
- d_drop(dentry);
- return err;
-#else
-#error "rebuild kernel and lustre with ext3-mds-num patch!"
- LASSERT(0);
-#endif
+
+ dqinfo.dqi_blocks = cpu_to_le32(blocks);
+ dqinfo.dqi_free_blk = 0;
+ dqinfo.dqi_free_entry = 0;
+
+ return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
}
-/* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
- objects. Logs is number of logfiles to update */
-static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
+static int create_new_quota_files(struct qchk_ctxt *qctxt,
+ struct obd_quotactl *oqc)
{
- if ( !fso ) {
- switch(op) {
- case FSFILT_OP_CREATE:
- /* directory leaf, index & indirect & EA*/
- return 4 + 3 * logs;
- case FSFILT_OP_UNLINK:
- return 3 * logs;
+ int i, rc = 0;
+ ENTRY;
+
+ for (i = 0; i < MAXQUOTAS; i++) {
+ struct if_dqinfo *info = qctxt->qckt_first_check[i]?
+ NULL : &qctxt->qckt_dqinfo[i];
+ struct file *file;
+
+ if (!Q_TYPESET(oqc, i))
+ continue;
+
+ file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
+ 0644);
+ if (IS_ERR(file)) {
+ rc = PTR_ERR(file);
+ CERROR("can't create %s file: rc = %d\n",
+ op_quotafile[i], rc);
+ GOTO(out, rc);
}
- } else {
- int i;
- int needed = 0;
- struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
- int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
- int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
- for (i = 0; i < op; i++, fso++) {
- int nblocks = fso->fso_bufcnt * blockpp;
- int ndindirect = min(nblocks, addrpp + 1);
- int nindir = nblocks + ndindirect + 1;
- needed += nindir;
+ if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
+ CERROR("file %s is not regular", op_quotafile[i]);
+ filp_close(file, 0);
+ GOTO(out, rc = -EINVAL);
}
- return needed + 3 * logs;
+
+ rc = v2_write_dqheader(file, i);
+ if (rc) {
+ filp_close(file, 0);
+ GOTO(out, rc);
+ }
+
+ rc = v2_write_dqinfo(file, i, info);
+ filp_close(file, 0);
+ if (rc)
+ GOTO(out, rc);
}
- return 0;
+out:
+ RETURN(rc);
}
-#if 0
-#define EXTENTS_EA "write_extents"
-#define EXTENTS_EA_SIZE 64
+static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
+ struct chk_dqblk *cdqb)
+{
+ struct obd_quotactl *oqc;
+ long now;
+ int rc;
+ ENTRY;
+
+ OBD_ALLOC_PTR(oqc);
+ if (!oqc)
+ RETURN(-ENOMEM);
+
+ now = CURRENT_SECONDS;
-int ext3_ext_in_ea_alloc_space(struct inode *, int, const char *, unsigned long, unsigned long);
-int ext3_ext_in_ea_remove_space(struct inode *, int, const char *, unsigned long, unsigned long);
-int ext3_ext_in_ea_get_extents(struct inode *, int, const char *, char **, int *);
-int ext3_ext_in_ea_get_extents_num(struct inode *, int, const char *, int *);
+ if (cdqb->dqb_bsoftlimit &&
+ toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
+ !cdqb->dqb_btime)
+ cdqb->dqb_btime =
+ now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
+
+ if (cdqb->dqb_isoftlimit &&
+ cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
+ !cdqb->dqb_itime)
+ cdqb->dqb_itime =
+ now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
+
+ cdqb->dqb_valid = QIF_ALL;
+
+ oqc->qc_cmd = Q_SETQUOTA;
+ oqc->qc_type = cdqb->dqb_type;
+ oqc->qc_id = cdqb->dqb_id;
+ DQBLK_COPY(&oqc->qc_dqblk, cdqb);
+
+ rc = fsfilt_ext3_quotactl(sb, oqc);
+ OBD_FREE_PTR(oqc);
+ RETURN(rc);
+}
-static int fsfilt_ext3_insert_extents_ea(struct inode *inode,
- unsigned long from,
- unsigned long num)
+static int prune_chkquots(struct super_block *sb,
+ struct qchk_ctxt *qctxt, int error)
{
- int rc = 0;
+ struct chk_dqblk *cdqb, *tmp;
+ int rc;
- rc = ext3_ext_in_ea_alloc_space(inode, EXT3_XATTR_INDEX_TRUSTED,
- EXTENTS_EA, from, num);
- return rc;
+ 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);
+ OBD_FREE_PTR(cdqb);
+ }
+
+ return error;
}
-static int fsfilt_ext3_remove_extents_ea(struct inode *inode,
- unsigned long from,
- unsigned long num)
+static int fsfilt_ext3_quotacheck(struct super_block *sb,
+ struct obd_quotactl *oqc)
{
+ struct ext3_sb_info *sbi = EXT3_SB(sb);
+ int i, group;
+ struct qchk_ctxt *qctxt;
+ struct buffer_head *bitmap_bh = NULL;
+ unsigned long ino;
+ struct inode *inode;
int rc = 0;
+ ENTRY;
- rc = ext3_ext_in_ea_remove_space(inode, EXT3_XATTR_INDEX_TRUSTED,
- EXTENTS_EA, from, num);
- return rc;
+ /* turn on quota and read dqinfo if existed */
+ OBD_ALLOC_PTR(qctxt);
+ if (!qctxt) {
+ oqc->qc_stat = -ENOMEM;
+ RETURN(-ENOMEM);
+ }
+
+ for (i = 0; i < NR_DQHASH; i++)
+ INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
+ INIT_LIST_HEAD(&qctxt->qckt_list);
+
+ for (i = 0; i < MAXQUOTAS; i++) {
+ if (!Q_TYPESET(oqc, i))
+ continue;
+
+ rc = quota_onoff(sb, Q_QUOTAON, i);
+ if (!rc || rc == -EBUSY) {
+ rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
+ if (rc)
+ GOTO(out, rc);
+ } else if (rc == -ENOENT) {
+ qctxt->qckt_first_check[i] = 1;
+ } else if (rc) {
+ GOTO(out, rc);
+ }
+ }
+
+ /* check quota and update in hash */
+ for (group = 0; group < sbi->s_groups_count; group++) {
+ ino = group * sbi->s_inodes_per_group + 1;
+ bitmap_bh = read_inode_bitmap(sb, group);
+ if (!bitmap_bh) {
+ CERROR("read_inode_bitmap group %d failed", group);
+ GOTO(out, -EIO);
+ }
+
+ for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
+ if (ino < sbi->s_first_ino)
+ continue;
+
+ inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
+ rc = add_inode_quota(inode, qctxt, oqc);
+ iput(inode);
+ if (rc) {
+ brelse(bitmap_bh);
+ GOTO(out, rc);
+ }
+ }
+
+ brelse(bitmap_bh);
+ }
+
+ /* read old quota limits from old quota file. (only for the user
+ * has limits but hasn't file) */
+#ifdef HAVE_QUOTA_SUPPORT
+ for (i = 0; i < MAXQUOTAS; i++) {
+ struct list_head id_list;
+ struct dquot_id *dqid, *tmp;
+
+ if (!Q_TYPESET(oqc, i))
+ continue;
+
+ if (qctxt->qckt_first_check[i])
+ continue;
+
+
+ LASSERT(sb_dqopt(sb)->files[i] != NULL);
+ INIT_LIST_HEAD(&id_list);
+#ifndef KERNEL_SUPPORTS_QUOTA_READ
+ rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
+#else
+ rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
+#endif
+ 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);
+
+ if (!rc)
+ cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
+ dqid->di_id, i,
+ qctxt->qckt_first_check[i]);
+ kfree(dqid);
+ }
+ }
+#endif
+ /* turn off quota cause we are to dump chk_dqblk to files */
+ quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
+
+ rc = create_new_quota_files(qctxt, oqc);
+ if (rc)
+ GOTO(out, rc);
+
+ /* we use vfs functions to set dqblk, so turn quota on */
+ rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
+out:
+ /* dump and free chk_dqblk */
+ rc = prune_chkquots(sb, qctxt, rc);
+ OBD_FREE_PTR(qctxt);
+
+ /* turn off quota, `lfs quotacheck` will turn on when all
+ * nodes quotacheck finish. */
+ quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
+
+ oqc->qc_stat = rc;
+ if (rc)
+ CERROR("quotacheck failed: rc = %d\n", rc);
+
+ RETURN(rc);
}
-extern int ext3_init_tree_in_ea(struct inode *inode, int name_index,
- const char *eaname, int size);
-
-static int fsfilt_ext3_init_extents_ea(struct inode *inode)
+#ifdef HAVE_QUOTA_SUPPORT
+static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
+ int cmd)
{
int rc = 0;
+ ENTRY;
- rc = ext3_init_tree_in_ea(inode, EXT3_XATTR_INDEX_TRUSTED,
- EXTENTS_EA, 64);
- return rc;
+ if (lqi->qi_files[type] == NULL) {
+ CERROR("operate qinfo before it's enabled!\n");
+ RETURN(-EIO);
+ }
+
+ switch (cmd) {
+ case QFILE_CHK:
+ rc = lustre_check_quota_file(lqi, type);
+ break;
+ case QFILE_RD_INFO:
+ rc = lustre_read_quota_info(lqi, type);
+ break;
+ case QFILE_WR_INFO:
+ rc = lustre_write_quota_info(lqi, type);
+ break;
+ case QFILE_INIT_INFO:
+ rc = lustre_init_quota_info(lqi, type);
+ break;
+ default:
+ CERROR("Unsupported admin quota file cmd %d\n", cmd);
+ LBUG();
+ break;
+ }
+ RETURN(rc);
}
-static int fsfilt_ext3_get_inode_write_extents(struct inode *inode,
- char **pbuf, int *size)
+static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
+ struct list_head *list)
{
- int rc = 0;
-
- rc = ext3_ext_in_ea_get_extents(inode, EXT3_XATTR_INDEX_TRUSTED,
- EXTENTS_EA, pbuf, size);
- return rc;
-}
+ return lustre_get_qids(file, inode, type, list);
+}
-static int fsfilt_ext3_get_write_extents_num(struct inode *inode, int *size)
+static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
{
int rc = 0;
-
- rc = ext3_ext_in_ea_get_extents_num(inode, EXT3_XATTR_INDEX_TRUSTED,
- EXTENTS_EA, size);
- return rc;
-}
+ ENTRY;
+
+ if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
+ CERROR("operate dquot before it's enabled!\n");
+ RETURN(-EIO);
+ }
+
+ switch (cmd) {
+ case QFILE_RD_DQUOT:
+ rc = lustre_read_dquot(dquot);
+ break;
+ case QFILE_WR_DQUOT:
+ if (dquot->dq_dqb.dqb_ihardlimit ||
+ dquot->dq_dqb.dqb_isoftlimit ||
+ dquot->dq_dqb.dqb_bhardlimit ||
+ dquot->dq_dqb.dqb_bsoftlimit)
+ clear_bit(DQ_FAKE_B, &dquot->dq_flags);
+ else
+ set_bit(DQ_FAKE_B, &dquot->dq_flags);
+
+ rc = lustre_commit_dquot(dquot);
+ if (rc >= 0)
+ rc = 0;
+ break;
+ default:
+ CERROR("Unsupported admin quota file cmd %d\n", cmd);
+ LBUG();
+ break;
+ }
+ RETURN(rc);
+}
#endif
static struct fsfilt_operations fsfilt_ext3_ops = {
- .fs_type = "ext3",
- .fs_owner = THIS_MODULE,
- .fs_start = fsfilt_ext3_start,
- .fs_brw_start = fsfilt_ext3_brw_start,
- .fs_commit = fsfilt_ext3_commit,
- .fs_commit_async = fsfilt_ext3_commit_async,
- .fs_commit_wait = fsfilt_ext3_commit_wait,
- .fs_setattr = fsfilt_ext3_setattr,
- .fs_iocontrol = fsfilt_ext3_iocontrol,
- .fs_set_md = fsfilt_ext3_set_md,
- .fs_get_md = fsfilt_ext3_get_md,
- .fs_readpage = fsfilt_ext3_readpage,
- .fs_add_journal_cb = fsfilt_ext3_add_journal_cb,
- .fs_statfs = fsfilt_ext3_statfs,
- .fs_sync = fsfilt_ext3_sync,
- .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
- .fs_prep_san_write = fsfilt_ext3_prep_san_write,
- .fs_write_record = fsfilt_ext3_write_record,
- .fs_read_record = fsfilt_ext3_read_record,
- .fs_setup = fsfilt_ext3_setup,
- .fs_getpage = fsfilt_ext3_getpage,
- .fs_send_bio = fsfilt_ext3_send_bio,
- .fs_set_xattr = fsfilt_ext3_set_xattr,
- .fs_get_xattr = fsfilt_ext3_get_xattr,
- .fs_get_op_len = fsfilt_ext3_get_op_len,
- .fs_add_dir_entry = fsfilt_ext3_add_dir_entry,
- .fs_del_dir_entry = fsfilt_ext3_del_dir_entry,
-#if 0
- .fs_init_extents_ea = fsfilt_ext3_init_extents_ea,
- .fs_insert_extents_ea = fsfilt_ext3_insert_extents_ea,
- .fs_remove_extents_ea = fsfilt_ext3_remove_extents_ea,
- .fs_get_inode_write_extents = fsfilt_ext3_get_inode_write_extents,
- .fs_get_write_extents_num = fsfilt_ext3_get_write_extents_num,
+ .fs_type = "ext3",
+ .fs_owner = THIS_MODULE,
+ .fs_getlabel = fsfilt_ext3_get_label,
+ .fs_setlabel = fsfilt_ext3_set_label,
+ .fs_uuid = fsfilt_ext3_uuid,
+ .fs_start = fsfilt_ext3_start,
+ .fs_brw_start = fsfilt_ext3_brw_start,
+ .fs_extend = fsfilt_ext3_extend,
+ .fs_commit = fsfilt_ext3_commit,
+ .fs_commit_async = fsfilt_ext3_commit_async,
+ .fs_commit_wait = fsfilt_ext3_commit_wait,
+ .fs_setattr = fsfilt_ext3_setattr,
+ .fs_iocontrol = fsfilt_ext3_iocontrol,
+ .fs_set_md = fsfilt_ext3_set_md,
+ .fs_get_md = fsfilt_ext3_get_md,
+ .fs_readpage = fsfilt_ext3_readpage,
+ .fs_add_journal_cb = fsfilt_ext3_add_journal_cb,
+ .fs_statfs = fsfilt_ext3_statfs,
+ .fs_sync = fsfilt_ext3_sync,
+ .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
+ .fs_prep_san_write = fsfilt_ext3_prep_san_write,
+ .fs_write_record = fsfilt_ext3_write_record,
+ .fs_read_record = fsfilt_ext3_read_record,
+ .fs_setup = fsfilt_ext3_setup,
+ .fs_send_bio = fsfilt_ext3_send_bio,
+ .fs_get_op_len = fsfilt_ext3_get_op_len,
+ .fs_quotactl = fsfilt_ext3_quotactl,
+ .fs_quotacheck = fsfilt_ext3_quotacheck,
+#ifdef HAVE_QUOTA_SUPPORT
+ .fs_quotainfo = fsfilt_ext3_quotainfo,
+ .fs_qids = fsfilt_ext3_qids,
+ .fs_dquot = fsfilt_ext3_dquot,
#endif
};
rc = fsfilt_register_ops(&fsfilt_ext3_ops);
- if (rc)
- kmem_cache_destroy(fcb_cache);
+ if (rc) {
+ int err = kmem_cache_destroy(fcb_cache);
+ LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
+ }
out:
return rc;
}
static void __exit fsfilt_ext3_exit(void)
{
+ int rc;
+
fsfilt_unregister_ops(&fsfilt_ext3_ops);
- LASSERTF(kmem_cache_destroy(fcb_cache) == 0,
- "can't free fsfilt callback cache: count %d\n",
- atomic_read(&fcb_cache_count));
+ rc = kmem_cache_destroy(fcb_cache);
+ LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
}
module_init(fsfilt_ext3_init);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff