typedef void (*fsfilt_cb_t)(struct obd_device *obd, __u64 last_rcvd,
void *data, int error);
-struct fsfilt_objinfo {
- struct dentry *fso_dentry;
- int fso_bufcnt;
-};
-
-struct fsfilt_fid {
- __u32 ino;
- __u32 gen;
-};
-
-struct lustre_dquot;
struct fsfilt_operations {
cfs_list_t fs_list;
cfs_module_t *fs_owner;
char *fs_type;
char *(* fs_getlabel)(struct super_block *sb);
- int (* fs_setlabel)(struct super_block *sb, char *label);
- char *(* fs_uuid)(struct super_block *sb);
void *(* fs_start)(struct inode *inode, int op, void *desc_private,
int logs);
- void *(* fs_brw_start)(int objcount, struct fsfilt_objinfo *fso,
- int niocount, struct niobuf_local *nb,
- void *desc_private, int logs);
- int (* fs_extend)(struct inode *inode, unsigned nblocks, void *h);
int (* fs_commit)(struct inode *inode, void *handle,int force_sync);
- int (* fs_commit_async)(struct inode *inode, void *handle,
- void **wait_handle);
- int (* fs_commit_wait)(struct inode *inode, void *handle);
- int (* fs_setattr)(struct dentry *dentry, void *handle,
- struct iattr *iattr, int do_trunc);
- int (* fs_iocontrol)(struct inode *inode, struct file *file,
- unsigned int cmd, unsigned long arg);
- int (* fs_set_md)(struct inode *inode, void *handle, void *md,
- int size, const char *name);
- int (* fs_get_md)(struct inode *inode, void *md, int size,
- const char *name);
- /*
- * this method is needed to make IO operation fsfilt nature depend.
- *
- * This operation maybe synchronous or asynchronous.
- *
- * Return convention: positive number of bytes written (synchronously)
- * on success. Negative errno value on failure. Zero if asynchronous
- * IO was submitted successfully.
- *
- */
- int (* fs_send_bio)(int rw, struct inode *inode, struct bio *bio);
- ssize_t (* fs_readpage)(struct file *file, char *buf, size_t count,
- loff_t *offset);
- int (* fs_add_journal_cb)(struct obd_device *obd, __u64 last_rcvd,
- void *handle, fsfilt_cb_t cb_func,
- void *cb_data);
- int (* fs_statfs)(struct super_block *sb, struct obd_statfs *osfs);
- int (* fs_sync)(struct super_block *sb);
int (* fs_map_inode_pages)(struct inode *inode, struct page **page,
int pages, unsigned long *blocks,
int *created, int create,
int force_sync);
int (* fs_read_record)(struct file *, void *, int size, loff_t *);
int (* fs_setup)(struct super_block *sb);
- int (* fs_get_op_len)(int, struct fsfilt_objinfo *, int);
- __u64 (* fs_get_version) (struct inode *inode);
- __u64 (* fs_set_version) (struct inode *inode, __u64 new_version);
- lvfs_sbdev_type (* fs_journal_sbdev)(struct super_block *sb);
- struct dentry *(* fs_fid2dentry)(struct vfsmount *mnt,
- struct fsfilt_fid *fid,
- int ignore_gen);
};
extern int fsfilt_register_ops(struct fsfilt_operations *fs_ops);
return obd->obd_fsops->fs_getlabel(sb);
}
-static inline int fsfilt_set_label(struct obd_device *obd,
- struct super_block *sb, char *label)
-{
- if (obd->obd_fsops->fs_setlabel == NULL)
- return -ENOSYS;
- return (obd->obd_fsops->fs_setlabel(sb, label));
-}
-
-static inline __u8 *fsfilt_uuid(struct obd_device *obd, struct super_block *sb)
-{
- if (obd->obd_fsops->fs_uuid == NULL)
- return NULL;
-
- return obd->obd_fsops->fs_uuid(sb);
-}
-
-static inline lvfs_sbdev_type fsfilt_journal_sbdev(struct obd_device *obd,
- struct super_block *sb)
-{
- if (obd && obd->obd_fsops && obd->obd_fsops->fs_journal_sbdev)
- return obd->obd_fsops->fs_journal_sbdev(sb);
- return (lvfs_sbdev_type)0;
-}
-
#define FSFILT_OP_UNLINK 1
#define FSFILT_OP_RMDIR 2
#define FSFILT_OP_RENAME 3
#define FSFILT_OP_SETATTR 8
#define FSFILT_OP_LINK 9
#define FSFILT_OP_CANCEL_UNLINK 10
-#define FSFILT_OP_NOOP 15
-#define FSFILT_OP_UNLINK_PARTIAL_CHILD 21
-#define FSFILT_OP_UNLINK_PARTIAL_PARENT 22
-#define FSFILT_OP_CREATE_PARTIAL_CHILD 23
#define __fsfilt_check_slow(obd, start, msg) \
do { \
return handle;
}
-static inline void *fsfilt_start(struct obd_device *obd, struct inode *inode,
- int op, struct obd_trans_info *oti)
-{
- return fsfilt_start_log(obd, inode, op, oti, 0);
-}
-
-static inline void *fsfilt_brw_start_log(struct obd_device *obd, int objcount,
- struct fsfilt_objinfo *fso,
- int niocount, struct niobuf_local *nb,
- struct obd_trans_info *oti, int logs)
-{
- unsigned long now = jiffies;
- void *parent_handle = oti ? oti->oti_handle : NULL;
- void *handle;
-
- handle = obd->obd_fsops->fs_brw_start(objcount, fso, niocount, nb,
- parent_handle, logs);
- CDEBUG(D_INFO, "started handle %p (%p)\n", handle, parent_handle);
-
- if (oti != NULL) {
- if (parent_handle == NULL) {
- oti->oti_handle = handle;
- } else if (handle != parent_handle) {
- CERROR("mismatch: parent %p, handle %p, oti %p\n",
- parent_handle, handle, oti);
- LBUG();
- }
- }
- fsfilt_check_slow(obd, now, "journal start");
-
- return handle;
-}
-
-static inline void *fsfilt_brw_start(struct obd_device *obd, int objcount,
- struct fsfilt_objinfo *fso, int niocount,
- struct niobuf_local *nb,
- struct obd_trans_info *oti)
-{
- return fsfilt_brw_start_log(obd, objcount, fso, niocount, nb, oti, 0);
-}
-
-static inline int fsfilt_extend(struct obd_device *obd, struct inode *inode,
- unsigned int nblocks, void *handle)
-{
- unsigned long now = jiffies;
- int rc = obd->obd_fsops->fs_extend(inode, nblocks, handle);
- CDEBUG(D_INFO, "extending handle %p with %u blocks\n", handle, nblocks);
-
- fsfilt_check_slow(obd, now, "journal extend");
-
- return rc;
-}
-
static inline int fsfilt_commit(struct obd_device *obd, struct inode *inode,
void *handle, int force_sync)
{
return rc;
}
-static inline int fsfilt_commit_async(struct obd_device *obd,
- struct inode *inode, void *handle,
- void **wait_handle)
-{
- unsigned long now = jiffies;
- int rc = obd->obd_fsops->fs_commit_async(inode, handle, wait_handle);
-
- CDEBUG(D_INFO, "committing handle %p (async)\n", *wait_handle);
- fsfilt_check_slow(obd, now, "journal start");
-
- return rc;
-}
-
-static inline int fsfilt_commit_wait(struct obd_device *obd,
- struct inode *inode, void *handle)
-{
- unsigned long now = jiffies;
- int rc = obd->obd_fsops->fs_commit_wait(inode, handle);
- CDEBUG(D_INFO, "waiting for completion %p\n", handle);
- fsfilt_check_slow(obd, now, "journal start");
- return rc;
-}
-
-static inline int fsfilt_setattr(struct obd_device *obd, struct dentry *dentry,
- void *handle, struct iattr *iattr,int do_trunc)
-{
- unsigned long now = jiffies;
- int rc;
- rc = obd->obd_fsops->fs_setattr(dentry, handle, iattr, do_trunc);
- fsfilt_check_slow(obd, now, "setattr");
- return rc;
-}
-
-static inline int fsfilt_iocontrol(struct obd_device *obd, struct dentry *dentry,
- unsigned int cmd, unsigned long arg)
-{
- struct file *dummy_file = NULL;
- int ret;
-
- OBD_ALLOC_PTR(dummy_file);
- if (!dummy_file)
- return(-ENOMEM);
-
- dummy_file->f_dentry = dentry;
- dummy_file->f_vfsmnt = obd->u.obt.obt_vfsmnt;
-
- ret = obd->obd_fsops->fs_iocontrol(dentry->d_inode, dummy_file, cmd,
- arg);
-
- OBD_FREE_PTR(dummy_file);
- return ret;
-}
-
-static inline int fsfilt_set_md(struct obd_device *obd, struct inode *inode,
- void *handle, void *md, int size,
- const char *name)
-{
- return obd->obd_fsops->fs_set_md(inode, handle, md, size, name);
-}
-
-static inline int fsfilt_get_md(struct obd_device *obd, struct inode *inode,
- void *md, int size, const char *name)
-{
- return obd->obd_fsops->fs_get_md(inode, md, size, name);
-}
-
-static inline int fsfilt_send_bio(int rw, struct obd_device *obd,
- struct inode *inode, void *bio)
-{
- LASSERTF(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ, "%x\n", rw);
-
- if (rw == OBD_BRW_READ)
- return obd->obd_fsops->fs_send_bio(READ, inode, bio);
- return obd->obd_fsops->fs_send_bio(WRITE, inode, bio);
-}
-
-static inline ssize_t fsfilt_readpage(struct obd_device *obd,
- struct file *file, char *buf,
- size_t count, loff_t *offset)
-{
- return obd->obd_fsops->fs_readpage(file, buf, count, offset);
-}
-
-static inline int fsfilt_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
- void *handle, fsfilt_cb_t cb_func,
- void *cb_data)
-{
- return obd->obd_fsops->fs_add_journal_cb(obd, last_rcvd,
- handle, cb_func, cb_data);
-}
-
-/* very similar to obd_statfs(), but caller already holds obd_osfs_lock */
-static inline int fsfilt_statfs(struct obd_device *obd, struct super_block *sb,
- __u64 max_age)
-{
- int rc = 0;
-
- CDEBUG(D_SUPER, "osfs "LPU64", max_age "LPU64"\n",
- obd->obd_osfs_age, max_age);
- if (cfs_time_before_64(obd->obd_osfs_age, max_age)) {
- rc = obd->obd_fsops->fs_statfs(sb, &obd->obd_osfs);
- if (rc == 0) /* N.B. statfs can't really fail */
- obd->obd_osfs_age = cfs_time_current_64();
- } else {
- CDEBUG(D_SUPER, "using cached obd_statfs data\n");
- }
-
- return rc;
-}
-
-static inline int fsfilt_sync(struct obd_device *obd, struct super_block *sb)
-{
- return obd->obd_fsops->fs_sync(sb);
-}
-
static inline int fsfilt_map_inode_pages(struct obd_device *obd,
struct inode *inode,
struct page **page, int pages,
return 0;
}
-static inline __u64 fsfilt_set_version(struct obd_device *obd,
- struct inode *inode, __u64 new_version)
-{
- if (obd->obd_fsops->fs_set_version)
- return obd->obd_fsops->fs_set_version(inode, new_version);
- return -EOPNOTSUPP;
-}
-
-static inline __u64 fsfilt_get_version(struct obd_device *obd,
- struct inode *inode)
-{
- if (obd->obd_fsops->fs_get_version)
- return obd->obd_fsops->fs_get_version(inode);
- return -EOPNOTSUPP;
-}
-static inline struct dentry *fsfilt_fid2dentry(struct obd_device *obd,
- struct vfsmount *mnt,
- struct fsfilt_fid *fid,
- int ignore_gen)
-{
- if (obd->obd_fsops->fs_fid2dentry)
- return obd->obd_fsops->fs_fid2dentry(mnt, fid, ignore_gen);
- return ERR_PTR(-EOPNOTSUPP);
-}
#endif /* __KERNEL__ */
#include <linux/fs.h>
#include <linux/slab.h>
#include <linux/pagemap.h>
-#include <linux/quotaops.h>
-#ifdef HAVE_LINUX_EXPORTFS_H
-#include <linux/exportfs.h>
-#endif
#include <ext4/ext4.h>
#include <ext4/ext4_jbd2.h>
#include <linux/version.h>
#include <linux/bitops.h>
#include <linux/quota.h>
-#if defined(HAVE_EXT3_XATTR_H)
-# include <ext3/xattr.h>
-#elif !defined(EXT3_XATTR_INDEX_TRUSTED)
-/* ext3 xattr.h not available in rh style kernel-devel rpm */
-/* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
-# define EXT3_XATTR_INDEX_TRUSTED 4
-extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
-extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
-#endif
-
#include <libcfs/libcfs.h>
#include <lustre_fsfilt.h>
#include <obd.h>
void *cb_data; /* MDS/OST completion function data */
};
-#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;
}
-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;
- handle = ext3_journal_start_sb(sb, 1);
- 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:
- ext3_journal_stop(handle);
-
- return(err);
-}
-
-static char *fsfilt_ext3_uuid(struct super_block *sb)
-{
- return EXT3_SB(sb)->s_es->s_uuid;
-}
-
-#ifdef HAVE_DISK_INODE_VERSION
-
-static __u64 get_i_version(struct inode *inode)
-{
- return EXT3_I(inode)->i_fs_version;
-}
-
-static void set_i_version(struct inode *inode, __u64 new_version)
-{
- (EXT3_I(inode))->i_fs_version = new_version;
-}
-
-/*
- * Get the 64-bit version for an inode.
- */
-static __u64 fsfilt_ext3_get_version(struct inode *inode)
-{
- CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
- get_i_version(inode), inode->i_ino);
- return get_i_version(inode);
-}
-
-/*
- * Set the 64-bit version and return the old version.
- */
-static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
-{
- __u64 old_version = get_i_version(inode);
-
- CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
- new_version, old_version, inode->i_ino);
- set_i_version(inode, new_version);
- /* version is set after all inode operations are finished, so we should
- * mark it dirty here */
- inode->i_sb->s_op->dirty_inode(inode);
- return old_version;
-}
-
-#endif
-
/* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
#ifdef HAVE_BLOCKS_FOR_TRUNCATE
# include <ext4/truncate.h>
return handle;
}
-/*
- * Calculate the number of buffer credits needed to write multiple pages in
- * a single ext3 transaction. No, this shouldn't be here, but as yet ext3
- * doesn't have a nice API for calculating this sort of thing in advance.
- *
- * See comment above ext3_writepage_trans_blocks for details. We assume
- * no data journaling is being done, but it does allow for all of the pages
- * being non-contiguous. If we are guaranteed contiguous pages we could
- * reduce the number of (d)indirect blocks a lot.
- *
- * With N blocks per page and P pages, for each inode we have at most:
- * N*P indirect
- * min(N*P, blocksize/4 + 1) dindirect blocks
- * niocount tindirect
- *
- * For the entire filesystem, we have at most:
- * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
- * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
- * objcount inode blocks
- * 1 superblock
- * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
- *
- * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
- */
-static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
- int niocount, struct niobuf_local *nb)
-{
- struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
- __u64 next_indir;
- const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
- int nbitmaps = 0, ngdblocks;
- int needed = objcount + 1; /* inodes + superblock */
- int i, j;
-
- for (i = 0, j = 0; i < objcount; i++, fso++) {
- /* two or more dindirect blocks in case we cross boundary */
- int ndind =
- (long)((nb[j + fso->fso_bufcnt - 1].lnb_file_offset -
- nb[j].lnb_file_offset) >>
- sb->s_blocksize_bits) /
- (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
- nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
-
- /* leaf, indirect, tindirect blocks for first block */
- nbitmaps += blockpp + 2;
-
- j += fso->fso_bufcnt;
- }
-
- next_indir = nb[0].lnb_file_offset +
- (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
- for (i = 1; i < niocount; i++) {
- if (nb[i].lnb_file_offset >= next_indir) {
- nbitmaps++; /* additional indirect */
- next_indir = nb[i].lnb_file_offset +
- (EXT3_ADDR_PER_BLOCK(sb) <<
- sb->s_blocksize_bits);
- } else if (nb[i].lnb_file_offset !=
- nb[i - 1].lnb_file_offset + sb->s_blocksize) {
- nbitmaps++; /* additional indirect */
- }
- nbitmaps += blockpp; /* each leaf in different group? */
- }
-
- ngdblocks = nbitmaps;
- if (nbitmaps > EXT3_SB(sb)->s_groups_count)
- nbitmaps = EXT3_SB(sb)->s_groups_count;
- if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
- ngdblocks = EXT3_SB(sb)->s_gdb_count;
-
- needed += nbitmaps + ngdblocks;
-
- /* last_rcvd update */
- needed += EXT3_DATA_TRANS_BLOCKS(sb);
-
-#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.
- */
- needed += hweight32(sb_any_quota_loaded(sb)) *
- FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
-#endif
-
- return needed;
-}
-
-/* We have to start a huge journal transaction here to hold all of the
- * metadata for the pages being written here. This is necessitated by
- * the fact that we do lots of prepare_write operations before we do
- * any of the matching commit_write operations, so even if we split
- * up to use "smaller" transactions none of them could complete until
- * all of them were opened. By having a single journal transaction,
- * we eliminate duplicate reservations for common blocks like the
- * superblock and group descriptors or bitmaps.
- *
- * We will start the transaction here, but each prepare_write will
- * add a refcount to the transaction, and each commit_write will
- * remove a refcount. The transaction will be closed when all of
- * the pages have been written.
- */
-static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
- int niocount, struct niobuf_local *nb,
- void *desc_private, int logs)
-{
- journal_t *journal;
- handle_t *handle;
- int needed;
- ENTRY;
-
- LASSERT(current->journal_info == desc_private);
- journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
- needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
-
- /* The number of blocks we could _possibly_ dirty can very large.
- * We reduce our request if it is absurd (and we couldn't get that
- * many credits for a single handle anyways).
- *
- * At some point we have to limit the size of I/Os sent at one time,
- * increase the size of the journal, or we have to calculate the
- * actual journal requirements more carefully by checking all of
- * the blocks instead of being maximally pessimistic. It remains to
- * be seen if this is a real problem or not.
- */
- if (needed > journal->j_max_transaction_buffers) {
- CERROR("want too many journal credits (%d) using %d instead\n",
- needed, journal->j_max_transaction_buffers);
- needed = journal->j_max_transaction_buffers;
- }
-
- LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
- handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
- if (IS_ERR(handle)) {
- CERROR("can't get handle for %d credits: rc = %ld\n", needed,
- PTR_ERR(handle));
- } else {
- LASSERT(handle->h_buffer_credits >= needed);
- LASSERT(current->journal_info == handle);
- }
-
- RETURN(handle);
-}
-
-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 (ext3_journal_extend(handle, nblocks) == 0)
- return 0;
-
- ext3_mark_inode_dirty(handle, inode);
- return ext3_journal_restart(handle, nblocks);
-}
-
static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
{
int rc;
return rc;
}
-static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
- void **wait_handle)
-{
- unsigned long tid;
- transaction_t *transaction;
- handle_t *handle = h;
- journal_t *journal;
- int rc;
-
- LASSERT(current->journal_info == handle);
-
- transaction = handle->h_transaction;
- journal = transaction->t_journal;
- tid = transaction->t_tid;
- /* we don't want to be blocked */
- handle->h_sync = 0;
- rc = ext3_journal_stop(handle);
- if (rc) {
- CERROR("error while stopping transaction: %d\n", rc);
- return rc;
- }
- fsfilt_log_start_commit(journal, tid);
-
- *wait_handle = (void *) tid;
- CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
- return 0;
-}
-
-static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
-{
- journal_t *journal = EXT3_JOURNAL(inode);
- tid_t tid = (tid_t)(long)h;
-
- CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
- if (unlikely(is_journal_aborted(journal)))
- return -EIO;
-
- fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
-
- if (unlikely(is_journal_aborted(journal)))
- return -EIO;
- return 0;
-}
-
-static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
- struct iattr *iattr, int do_trunc)
-{
- struct inode *inode = dentry->d_inode;
- int rc = 0;
-
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
- /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
- * timestamps to appear to be in the far future, due old timestamp
- * being stored on disk as an unsigned value. This fixes up any
- * bad values held by the client before storing them on disk,
- * and ensures any timestamp updates are correct. LU-1042 */
- if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
- !(iattr->ia_valid & ATTR_ATIME))) {
- iattr->ia_valid |= ATTR_ATIME;
- LTIME_S(iattr->ia_atime) = 0;
- }
- if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
- !(iattr->ia_valid & ATTR_MTIME))) {
- iattr->ia_valid |= ATTR_MTIME;
- LTIME_S(iattr->ia_mtime) = 0;
- }
- if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
- LTIME_S(inode->i_ctime) == 0) &&
- !(iattr->ia_valid & ATTR_CTIME))) {
- iattr->ia_valid |= ATTR_CTIME;
- LTIME_S(iattr->ia_ctime) = 0;
- }
-#else
-#warning "remove old LU-221/LU-1042 workaround code"
-#endif
-
- /* When initializating timestamps for new inodes, use the filesystem
- * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
- * that this is potentially an invalid inode. Files with an old ctime
- * migrated to a newly-formatted OST with a newer s_mkfs_time will not
- * hit this check, since it is only for ctime == 0. LU-1010/LU-1042 */
- if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
- LTIME_S(iattr->ia_ctime) =
- EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
-
- /* 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 = iattr->ia_size;
- i_size_write(inode, iattr->ia_size);
-
- 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 (!cfs_curproc_is_in_groups(inode->i_gid) &&
- !cfs_capable(CFS_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 */
- 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. */
- iattr->ia_valid &= ~TIMES_SET_FLAGS;
-
- if (inode->i_op->setattr) {
- rc = inode->i_op->setattr(dentry, iattr);
- } else {
-#ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
- rc = inode_change_ok(inode, iattr);
- if (!rc)
-#endif
- rc = simple_setattr(dentry, iattr);
- }
-
- out:
- RETURN(rc);
-}
-
-static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
- unsigned int cmd, unsigned long arg)
-{
- int rc = 0;
- ENTRY;
-
- /* FIXME: Can't do this because of nested transaction deadlock */
- 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;
- }
-
- /* ext4_ioctl does not have a inode argument */
- if (inode->i_fop->unlocked_ioctl)
- rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
- else
- RETURN(-ENOTTY);
-
- RETURN(rc);
-}
-
-static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
- void *lmm, int lmm_size, const char *name)
-{
- int rc;
-
- LASSERT(mutex_trylock(&inode->i_mutex) == 0);
-
- rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
- name, lmm, lmm_size, XATTR_NO_CTIME);
-
- if (rc && rc != -EROFS)
- CERROR("error adding MD data to inode %lu: rc = %d\n",
- inode->i_ino, rc);
- return rc;
-}
-
-/* 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;
-
- LASSERT(mutex_trylock(&inode->i_mutex) == 0);
-
- rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
- name, lmm, lmm_size);
-
- /* This gives us the MD size */
- if (lmm == NULL)
- return (rc == -ENODATA) ? 0 : rc;
-
- if (rc < 0) {
- 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_send_bio(int rw, struct inode *inode, struct bio *bio)
-{
- submit_bio(rw, bio);
- return 0;
-}
-
-static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
- loff_t *off)
-{
- struct inode *inode = file->f_dentry->d_inode;
- int rc = 0;
-
- if (S_ISREG(inode->i_mode))
- rc = file->f_op->read(file, buf, count, off);
- else {
- const int blkbits = inode->i_sb->s_blocksize_bits;
- const int blksize = inode->i_sb->s_blocksize;
-
- CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
- (unsigned long)count, inode->i_ino, *off);
- while (count > 0) {
- struct buffer_head *bh;
-
- bh = NULL;
- if (*off < i_size_read(inode)) {
- int err = 0;
-
- bh = ext3_bread(NULL, inode, *off >> blkbits,
- 0, &err);
-
- CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
-
- if (bh) {
- memcpy(buf, bh->b_data, blksize);
- brelse(bh);
- } else if (err) {
- /* XXX in theory we should just fake
- * this buffer and continue like ext3,
- * especially if this is a partial read
- */
- CERROR("error read dir %lu+%llu: %d\n",
- inode->i_ino, *off, err);
- RETURN(err);
- }
- }
- if (!bh) {
- struct ext3_dir_entry_2 *fake = (void *)buf;
-
- CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
- memset(fake, 0, sizeof(*fake));
- fake->rec_len = cpu_to_le16(blksize);
- }
- count -= blksize;
- buf += blksize;
- *off += blksize;
- rc += blksize;
- }
- }
-
- return rc;
-}
-
-#ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
-static void fsfilt_ext3_cb_func(struct super_block *sb,
- struct journal_callback *jcb, int error)
-#else
-static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
-#endif
-{
- struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
-
- fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
-
- OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
-}
-
-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;
-
- OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
- if (fcb == NULL)
- RETURN(-ENOMEM);
-
- fcb->cb_func = cb_func;
- fcb->cb_obd = obd;
- fcb->cb_last_rcvd = last_rcvd;
- fcb->cb_data = cb_data;
-
- CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
- fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
-
- return 0;
-}
-
-static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
-{
- struct kstatfs sfs;
- int rc;
-
- memset(&sfs, 0, sizeof(sfs));
- rc = sb->s_op->statfs(sb->s_root, &sfs);
- statfs_pack(osfs, &sfs);
- return rc;
-}
-
-static int fsfilt_ext3_sync(struct super_block *sb)
-{
- return ext3_force_commit(sb);
-}
-
#ifndef EXT3_EXTENTS_FL
#define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
#endif
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;
- }
- } 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;
-}
-
-lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
-{
- return (EXT3_SB(sb)->journal_bdev);
-}
-EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
-
-static int ll_decode_fh_accept(void *context, struct dentry *de)
-{
- return 1;
-}
-
-#ifdef HAVE_EXPORTFS_DECODE_FH
-# define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
- exportfs_decode_fh(mnt, (struct fid*)(fid), len, type, \
- acceptable, context)
-#else
-# define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
- export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len, \
- type, acceptable, context)
-# define FILEID_INO32_GEN 1
-extern struct export_operations export_op_default;
-#endif
-
-struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
- struct fsfilt_fid *fid, int ignore_gen)
-{
- struct inode *inode;
- struct dentry *result;
-
- result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
- ll_decode_fh_accept, NULL);
- if (IS_ERR(result)) {
- CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
- fid->ino, fid->gen, PTR_ERR(result));
- return result;
- }
-
- CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
- fid->ino, fid->gen);
- inode = result->d_inode;
- if (inode == NULL)
- goto err_out;
-
- if (inode->i_nlink == 0 &&
- inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
- LCONSOLE_WARN("Found inode with zero nlink, mode and"
- " ctime -- this may indicate disk "
- "corruption (inode: %lu, link: %lu, "
- "count: %d)\n", inode->i_ino,
- (unsigned long)inode->i_nlink,
- atomic_read(&inode->i_count));
- goto err_out;
- }
- if (fid->gen && inode->i_generation != fid->gen) {
- /* we didn't find the right inode.. */
- CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
- "count: %d, generation %u/%u\n",
- inode->i_ino, (unsigned long)inode->i_nlink,
- atomic_read(&inode->i_count), inode->i_generation,
- fid->gen);
- goto err_out;
- }
-
- return result;
-err_out:
- l_dput(result);
- return ERR_PTR(-ENOENT);
-}
-
static struct fsfilt_operations fsfilt_ext3_ops = {
.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_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,
-#ifdef HAVE_DISK_INODE_VERSION
- .fs_get_version = fsfilt_ext3_get_version,
- .fs_set_version = fsfilt_ext3_set_version,
-#endif
- .fs_journal_sbdev = fsfilt_ext3_journal_sbdev,
- .fs_fid2dentry = fsfilt_ext3_fid2dentry,
};
static int __init fsfilt_ext3_init(void)
git://git.whamcloud.com / fs / lustre-release.git / commitdiff