#include <linux/quotaops.h>
#include <linux/ext3_fs.h>
#include <linux/ext3_jbd.h>
+#include <linux/ext3_extents.h>
#include <linux/version.h>
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
#include <linux/ext3_xattr.h>
#include <ext3/xattr.h>
#endif
-#include <linux/kp30.h>
+#include <libcfs/kp30.h>
#include <linux/lustre_fsfilt.h>
#include <linux/obd.h>
#include <linux/obd_class.h>
#include <linux/iobuf.h>
#endif
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,7))
+# define lock_24kernel() lock_kernel()
+# define unlock_24kernel() unlock_kernel()
+#else
+# define lock_24kernel() do {} while (0)
+# define unlock_24kernel() do {} while (0)
+#endif
+
static kmem_cache_t *fcb_cache;
static atomic_t fcb_cache_count = ATOMIC_INIT(0);
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_rcvd; /* MDS/OST last committed operation */
+ __u64 cb_last_num; /* MDS/OST last committed operation */
void *cb_data; /* MDS/OST completion function data */
};
#ifndef EXT3_XATTR_INDEX_TRUSTED /* temporary until we hit l28 kernel */
#define EXT3_XATTR_INDEX_TRUSTED 4
#endif
-#define XATTR_LUSTRE_MDS_LOV_EA "lov"
/*
* We don't currently need any additional blocks for rmdir and
goto journal_start;
}
- /* XXX BUG 3188 -- must return to one set of opcodes */
- /* FIXME - cache hook */
- if (op & 0x20) {
- nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
- op = op & ~0x20;
- }
-
- /* FIXME - kml */
- if (op & 0x10) {
- nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
- op = op & ~0x10;
- }
-
+ 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 += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
+ EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;*/
break;
case FSFILT_OP_RENAME:
/* modify additional directory */
/* no break */
case FSFILT_OP_CREATE:
/* create/update logs for each stripe */
- nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
- EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
+ /*nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
+ EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;*/
/* no break */
case FSFILT_OP_MKDIR:
case FSFILT_OP_MKNOD:
}
journal_start:
- lock_kernel();
+ LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
+ lock_24kernel();
handle = journal_start(EXT3_JOURNAL(inode), nblocks);
- unlock_kernel();
+ unlock_24kernel();
if (!IS_ERR(handle))
LASSERT(current->journal_info == handle);
needed = journal->j_max_transaction_buffers;
}
- lock_kernel();
+ LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
+ lock_24kernel();
handle = journal_start(journal, needed);
- unlock_kernel();
+ unlock_24kernel();
if (IS_ERR(handle)) {
CERROR("can't get handle for %d credits: rc = %ld\n", needed,
PTR_ERR(handle));
RETURN(handle);
}
-static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
+static int fsfilt_ext3_commit(struct super_block *sb, struct inode *inode,
+ void *h, int force_sync)
{
int rc;
handle_t *handle = h;
if (force_sync)
handle->h_sync = 1; /* recovery likes this */
- lock_kernel();
+ lock_24kernel();
rc = journal_stop(handle);
- unlock_kernel();
+ unlock_24kernel();
return rc;
}
LASSERT(current->journal_info == handle);
- lock_kernel();
+ lock_24kernel();
transaction = handle->h_transaction;
journal = transaction->t_journal;
tid = transaction->t_tid;
CERROR("strange race: %lu != %lu\n",
(unsigned long) tid, (unsigned long) rtid);
#else
- log_start_commit(journal, transaction->t_tid);
+ log_start_commit(journal, tid);
#endif
- unlock_kernel();
+ unlock_24kernel();
*wait_handle = (void *) tid;
CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
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 (is_journal_aborted(EXT3_JOURNAL(inode)))
+ if (unlikely(is_journal_aborted(journal)))
return -EIO;
log_wait_commit(EXT3_JOURNAL(inode), tid);
+ if (unlikely(is_journal_aborted(journal)))
+ return -EIO;
return 0;
}
struct inode *inode = dentry->d_inode;
int rc;
- lock_kernel();
+ lock_24kernel();
/* A _really_ horrible hack to avoid removing the data stored
* in the block pointers; this is really the "small" stripe MD data.
rc = inode_setattr(inode, iattr);
}
- unlock_kernel();
+ unlock_24kernel();
return rc;
}
RETURN(rc);
}
-static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
- void *lmm, int lmm_size)
+static int fsfilt_ext3_set_xattr(struct inode * inode, void *handle, char *name,
+ void *buffer, int buffer_size)
{
- int rc;
+ int rc = 0;
- /* keep this when we get rid of OLD_EA (too noisy during conversion) */
- if (EXT3_I(inode)->i_file_acl /* || large inode EA flag */)
- CWARN("setting EA on %lu/%u again... interesting\n",
- inode->i_ino, inode->i_generation);
+ lock_24kernel();
- lock_kernel();
rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
- XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size, 0);
-
- unlock_kernel();
-
+ name, buffer, buffer_size, 0);
+ unlock_24kernel();
if (rc)
- CERROR("error adding MD data to inode %lu: rc = %d\n",
- inode->i_ino, rc);
+ CERROR("set xattr %s from inode %lu: rc %d\n",
+ name, inode->i_ino, rc);
return rc;
}
-/* Must be called with i_sem held */
-static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size)
+static int fsfilt_ext3_get_xattr(struct inode *inode, char *name,
+ void *buffer, int buffer_size)
{
- int rc;
-
- LASSERT(down_trylock(&inode->i_sem) != 0);
- lock_kernel();
+ int rc = 0;
+
+ lock_24kernel();
rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
- XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size);
- unlock_kernel();
+ name, buffer, buffer_size);
+ unlock_24kernel();
- /* This gives us the MD size */
- if (lmm == NULL)
+ if (buffer == 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, XATTR_LUSTRE_MDS_LOV_EA,
- inode->i_ino, rc);
- memset(lmm, 0, lmm_size);
+ CDEBUG(D_INFO, "error getting EA %s from inode %lu: rc %d\n",
+ name, inode->i_ino, rc);
+ memset(buffer, 0, buffer_size);
return (rc == -ENODATA) ? 0 : rc;
}
return rc;
}
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
-static int fsfilt_ext3_send_bio(struct inode *inode, struct bio *bio)
+static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
+ void *lmm, int lmm_size,
+ enum ea_type type)
{
- submit_bio(WRITE, bio);
- return 0;
+ 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;
}
-#else
-static int fsfilt_ext3_send_bio(struct inode *inode, struct kiobuf *bio)
+
+static int fsfilt_ext3_get_md(struct inode *inode, void *lmm,
+ int lmm_size, enum ea_type type)
{
- int rc, blocks_per_page;
+ 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;
+}
- rc = brw_kiovec(WRITE, 1, &bio, inode->i_dev,
- bio->blocks, 1 << inode->i_blkbits);
+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) * bio->nr_pages *
- blocks_per_page) {
+ if (rc != (1 << inode->i_blkbits) * b->nr_pages * blocks_per_page) {
CERROR("short write? expected %d, wrote %d\n",
- (1 << inode->i_blkbits) * bio->nr_pages *
+ (1 << inode->i_blkbits) * b->nr_pages *
blocks_per_page, rc);
}
-
+#endif
return rc;
}
-#endif
-/* FIXME-UMKA: This should be used in 2.6.x io code later. */
static struct page *fsfilt_ext3_getpage(struct inode *inode, long int index)
{
int rc;
{
struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
- fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
+ fcb->cb_func(fcb->cb_obd, fcb->cb_last_num, 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_rcvd,
- void *handle, fsfilt_cb_t cb_func,
+ __u64 last_num, void *handle,
+ fsfilt_cb_t cb_func,
void *cb_data)
{
struct fsfilt_cb_data *fcb;
atomic_inc(&fcb_cache_count);
fcb->cb_func = cb_func;
fcb->cb_obd = obd;
- fcb->cb_last_rcvd = last_rcvd;
+ fcb->cb_last_num = last_num;
fcb->cb_data = cb_data;
- CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
- lock_kernel();
+ CDEBUG(D_EXT2, "set callback for last_num: "LPD64"\n", last_num);
+ lock_24kernel();
journal_callback_set(handle, fsfilt_ext3_cb_func,
(struct journal_callback *)fcb);
- unlock_kernel();
-
+ unlock_24kernel();
return 0;
}
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(2,5,0))
+#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
+#define ext3_up_truncate_sem(inode) up(&EXT3_I(inode)->truncate_sem);
+#define ext3_down_truncate_sem(inode) down(&EXT3_I(inode)->truncate_sem);
+#endif
+
+#include <linux/lustre_version.h>
+#if EXT3_EXT_MAGIC == 0xf301
+#define ee_start e_start
+#define ee_block e_block
+#define ee_len e_num
+#endif
+#ifndef EXT3_BB_MAX_BLOCKS
+#define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
+ ext3_new_blocks(handle, inode, count, goal, err)
+#endif
+
+struct bpointers {
+ unsigned long *blocks;
+ int *created;
+ unsigned long start;
+ int num;
+ int init_num;
+ int create;
+};
+
+static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
+ unsigned long block, int *aflags)
+{
+ struct ext3_inode_info *ei = EXT3_I(inode);
+ unsigned long bg_start;
+ unsigned long colour;
+ int depth;
+
+ if (path) {
+ struct ext3_extent *ex;
+ depth = path->p_depth;
+
+ /* try to predict block placement */
+ if ((ex = path[depth].p_ext)) {
+#if 0
+ /* This prefers to eat into a contiguous extent
+ * rather than find an extent that the whole
+ * request will fit into. This can fragment data
+ * block allocation and prevents our lovely 1M I/Os
+ * from reaching the disk intact. */
+ if (ex->ee_block + ex->ee_len == block)
+ *aflags |= 1;
+#endif
+ return ex->ee_start + (block - ex->ee_block);
+ }
+
+ /* it looks index is empty
+ * try to find starting from index itself */
+ if (path[depth].p_bh)
+ return path[depth].p_bh->b_blocknr;
+ }
+
+ /* OK. use inode's group */
+ bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
+ le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
+ colour = (current->pid % 16) *
+ (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
+ return bg_start + colour + block;
+}
+
+static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
+ struct ext3_ext_path *path,
+ struct ext3_ext_cache *cex)
+{
+ struct inode *inode = tree->inode;
+ struct bpointers *bp = tree->private;
+ struct ext3_extent nex;
+ int count, err, goal;
+ unsigned long pblock;
+ loff_t new_i_size;
+ handle_t *handle;
+ int i, aflags = 0;
+
+ i = EXT_DEPTH(tree);
+ EXT_ASSERT(i == path->p_depth);
+ EXT_ASSERT(path[i].p_hdr);
+
+ if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
+ err = EXT_CONTINUE;
+ goto map;
+ }
+
+ if (bp->create == 0) {
+ i = 0;
+ if (cex->ec_block < bp->start)
+ i = bp->start - cex->ec_block;
+ if (i >= cex->ec_len)
+ CERROR("nothing to do?! i = %d, e_num = %u\n",
+ i, cex->ec_len);
+ for (; i < cex->ec_len && bp->num; i++) {
+ *(bp->created) = 0;
+ bp->created++;
+ *(bp->blocks) = 0;
+ bp->blocks++;
+ bp->num--;
+ bp->start++;
+ }
+
+ return EXT_CONTINUE;
+ }
+
+ count = ext3_ext_calc_credits_for_insert(tree, path);
+
+ lock_24kernel();
+ handle = journal_start(EXT3_JOURNAL(inode), count+EXT3_ALLOC_NEEDED+1);
+ unlock_24kernel();
+ if (IS_ERR(handle))
+ return PTR_ERR(handle);
+
+ count = cex->ec_len;
+ goal = ext3_ext_find_goal(inode, path, cex->ec_block, &aflags);
+ aflags |= 2; /* block have been already reserved */
+ lock_24kernel();
+ pblock = ext3_mb_new_blocks(handle, inode, goal, &count, aflags, &err);
+ unlock_24kernel();
+ 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;
+ nex.ee_start = pblock;
+ nex.ee_len = count;
+ err = ext3_ext_insert_extent(handle, tree, path, &nex);
+ if (err)
+ goto out;
+
+ /*
+ * Putting len of the actual extent we just inserted,
+ * we are asking ext3_ext_walk_space() to continue
+ * scaning after that block
+ */
+ cex->ec_len = nex.ee_len;
+ cex->ec_start = nex.ee_start;
+ 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");
+ CERROR("initial space: %lu:%u\n",
+ bp->start, bp->init_num);
+ CERROR("current extent: %u/%u/%u %d\n",
+ cex->ec_block, cex->ec_len,
+ cex->ec_start, cex->ec_type);
+ }
+ i = 0;
+ if (cex->ec_block < bp->start)
+ i = bp->start - cex->ec_block;
+ if (i >= cex->ec_len)
+ CERROR("nothing to do?! i = %d, e_num = %u\n",
+ i, cex->ec_len);
+ for (; i < cex->ec_len && bp->num; i++) {
+ *(bp->blocks) = cex->ec_start + i;
+ if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
+ *(bp->created) = 0;
+ } 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));
+ }
+ bp->created++;
+ bp->blocks++;
+ bp->num--;
+ bp->start++;
+ }
+ }
+ return err;
+}
+
+int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
+ unsigned long num, unsigned long *blocks,
+ int *created, int create)
+{
+ struct ext3_extents_tree tree;
+ struct bpointers bp;
+ int err;
+
+ CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
+ block, block + num, (unsigned) inode->i_ino);
+
+ ext3_init_tree_desc(&tree, inode);
+ tree.private = &bp;
+ bp.blocks = blocks;
+ bp.created = created;
+ bp.start = block;
+ bp.init_num = bp.num = num;
+ bp.create = create;
+
+ err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
+ ext3_ext_invalidate_cache(&tree);
+
+ return err;
+}
+
+int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
+ int pages, unsigned long *blocks,
+ int *created, int create)
+{
+ int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
+ int rc = 0, i = 0;
+ struct page *fp = NULL;
+ int clen = 0;
+
+ CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
+ inode->i_ino, pages, (*page)->index);
+
+ /* pages are sorted already. so, we just have to find
+ * contig. space and process them properly */
+ while (i < pages) {
+ if (fp == NULL) {
+ /* start new extent */
+ fp = *page++;
+ clen = 1;
+ i++;
+ continue;
+ } else if (fp->index + clen == (*page)->index) {
+ /* continue the extent */
+ page++;
+ clen++;
+ i++;
+ continue;
+ }
+
+ /* process found extent */
+ rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
+ clen * blocks_per_page, blocks,
+ created, create);
+ if (rc)
+ GOTO(cleanup, rc);
+
+ /* look for next extent */
+ fp = NULL;
+ blocks += blocks_per_page * clen;
+ created += blocks_per_page * clen;
+ }
+
+ if (fp)
+ rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
+ clen * blocks_per_page, blocks,
+ created, create);
+cleanup:
+ return rc;
+}
+#endif
+
extern int ext3_map_inode_page(struct inode *inode, struct page *page,
unsigned long *blocks, int *created, int create);
-int fsfilt_ext3_map_inode_page(struct inode *inode, struct page *page,
- unsigned long *blocks, int *created, int create)
+int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
+ int pages, unsigned long *blocks,
+ int *created, int create)
{
- return ext3_map_inode_page(inode, page, blocks, created, create);
+ int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
+ unsigned long *b;
+ int rc = 0, i, *cr;
+
+ for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
+ rc = ext3_map_inode_page(inode, *page, b, cr, create);
+ if (rc) {
+ CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
+ inode->i_ino, *b, *cr, create, rc);
+ break;
+ }
+
+ b += blocks_per_page;
+ cr += blocks_per_page;
+ }
+ return rc;
+}
+
+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)
+{
+ int rc = 0;
+ if (optional_sem != NULL)
+ down(optional_sem);
+#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);
+ else
+#endif
+ rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
+ created, create);
+ if (optional_sem != NULL)
+ up(optional_sem);
+
+ return rc;
}
extern int ext3_prep_san_write(struct inode *inode, long *blocks,
block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
journal = EXT3_SB(inode->i_sb)->s_journal;
- lock_kernel();
+ lock_24kernel();
handle = journal_start(journal,
block_count * EXT3_DATA_TRANS_BLOCKS + 2);
- unlock_kernel();
+ unlock_24kernel();
if (IS_ERR(handle)) {
CERROR("can't start transaction\n");
return PTR_ERR(handle);
unlock_kernel();
}
- lock_kernel();
+ lock_24kernel();
journal_stop(handle);
- unlock_kernel();
+ unlock_24kernel();
if (err == 0)
*offs = offset;
return err;
}
-static int fsfilt_ext3_setup(struct super_block *sb)
+static int fsfilt_ext3_setup(struct obd_device *obd, 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);
+ }
+ 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);
+ }
return 0;
}
-static int fsfilt_ext3_set_xattr(struct inode * inode, void *handle, char *name,
- void *buffer, int buffer_size)
+static inline int ext3_match (int len, const char * const name,
+ struct ext3_dir_entry_2 * de)
{
- int rc = 0;
+ if (len != de->name_len)
+ return 0;
+ if (!de->inode)
+ return 0;
+ return !memcmp(name, de->name, len);
+}
+static int fsfilt_ext3_scan_for_entry(struct dentry *dentry)
+{
+ 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;
+ }
- lock_kernel();
+ 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);
+ }
- rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
- name, buffer, buffer_size, 0);
- unlock_kernel();
- if (rc)
- CERROR("set xattr %s from inode %lu: rc %d\n",
- name, inode->i_ino, rc);
- return rc;
+ count -= blksize;
+ brelse(bh);
+ }
+
+ return num;
}
-static int fsfilt_ext3_get_xattr(struct inode *inode, char *name,
- void *buffer, int buffer_size)
-{
- int rc = 0;
- lock_kernel();
+extern int ext3_add_dir_entry(struct dentry *dentry);
+extern int ext3_del_dir_entry(struct dentry *dentry);
- rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
- name, buffer, buffer_size);
- unlock_kernel();
+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)
+{
+#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));
+ }
+ 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);
+ }
- if (buffer == 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);
- return (rc == -ENODATA) ? 0 : rc;
+ /* 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);
}
- return rc;
+ 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(err);
+#else
+#error "rebuild kernel and lustre with ext3-mds-num patch!"
+ LASSERT(0);
+#endif
+}
+
+static int fsfilt_ext3_del_dir_entry(struct obd_device *obd,
+ struct dentry *dentry)
+{
+#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);
+ }
+ if (err == 0)
+ d_drop(dentry);
+ return err;
+#else
+#error "rebuild kernel and lustre with ext3-mds-num patch!"
+ LASSERT(0);
+#endif
}
/* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
return 0;
}
+
+#if 0
+#define EXTENTS_EA "write_extents"
+#define EXTENTS_EA_SIZE 64
+
+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 *);
+
+static int fsfilt_ext3_insert_extents_ea(struct inode *inode,
+ unsigned long from,
+ unsigned long num)
+{
+ int rc = 0;
+
+ rc = ext3_ext_in_ea_alloc_space(inode, EXT3_XATTR_INDEX_TRUSTED,
+ EXTENTS_EA, from, num);
+ return rc;
+}
+
+static int fsfilt_ext3_remove_extents_ea(struct inode *inode,
+ unsigned long from,
+ unsigned long num)
+{
+ int rc = 0;
+
+ rc = ext3_ext_in_ea_remove_space(inode, EXT3_XATTR_INDEX_TRUSTED,
+ EXTENTS_EA, from, num);
+ 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)
+{
+ int rc = 0;
+
+ rc = ext3_init_tree_in_ea(inode, EXT3_XATTR_INDEX_TRUSTED,
+ EXTENTS_EA, 64);
+ return rc;
+}
+
+static int fsfilt_ext3_get_inode_write_extents(struct inode *inode,
+ char **pbuf, int *size)
+{
+ int rc = 0;
+
+ rc = ext3_ext_in_ea_get_extents(inode, EXT3_XATTR_INDEX_TRUSTED,
+ EXTENTS_EA, pbuf, size);
+ return rc;
+}
+
+static int fsfilt_ext3_get_write_extents_num(struct inode *inode, int *size)
+{
+ int rc = 0;
+
+ rc = ext3_ext_in_ea_get_extents_num(inode, EXT3_XATTR_INDEX_TRUSTED,
+ EXTENTS_EA, size);
+ 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_page = fsfilt_ext3_map_inode_page,
- .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_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,
+#endif
};
static int __init fsfilt_ext3_init(void)
static void __exit fsfilt_ext3_exit(void)
{
- int rc;
-
fsfilt_unregister_ops(&fsfilt_ext3_ops);
- rc = kmem_cache_destroy(fcb_cache);
-
- if (rc || atomic_read(&fcb_cache_count)) {
- CERROR("can't free fsfilt callback cache: count %d, rc = %d\n",
- atomic_read(&fcb_cache_count), rc);
- }
+ LASSERTF(kmem_cache_destroy(fcb_cache) == 0,
+ "can't free fsfilt callback cache: count %d\n",
+ atomic_read(&fcb_cache_count));
}
module_init(fsfilt_ext3_init);