#define DEBUG_SUBSYSTEM S_FILTER
+#include <linux/init.h>
+#include <linux/module.h>
#include <linux/fs.h>
#include <linux/jbd.h>
#include <linux/slab.h>
#include <linux/ext3_fs.h>
#include <linux/ext3_jbd.h>
#include <linux/version.h>
-/* XXX ugh */
+#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>
+#include <linux/ext3_xattr.h>
#else
- #include <linux/../../fs/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/module.h>
+#include <linux/lustre_quota.h>
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
+#include <linux/iobuf.h>
+#endif
+
+#ifdef EXT3_MULTIBLOCK_ALLOCATOR
+#include <linux/ext3_extents.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 fsfilt_cb_data {
struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
#endif
#define XATTR_LUSTRE_MDS_LOV_EA "lov"
-#define EXT3_XATTR_INDEX_LUSTRE 5 /* old */
-#define XATTR_LUSTRE_MDS_OBJID "system.lustre_mds_objid" /* old */
-
/*
* We don't currently need any additional blocks for rmdir and
* unlink transactions because we are storing the OST oa_id inside
* the inode (which we will be changing anyways as part of this
* transaction).
*/
-static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private)
+static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
+ int logs)
{
- /* For updates to the last recieved file */
- int nblocks = EXT3_DATA_TRANS_BLOCKS;
- int blocksize, block_count = 0;
+ /* For updates to the last received file */
+ int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS;
+ journal_t *journal;
void *handle;
if (current->journal_info) {
- CDEBUG(D_INODE, "increasing refcount on %p\n", current->journal_info);
+ CDEBUG(D_INODE, "increasing refcount on %p\n",
+ current->journal_info);
goto journal_start;
}
switch(op) {
- case FSFILT_OP_CREATE_LOG:
- nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
- op = FSFILT_OP_CREATE;
- break;
- case FSFILT_OP_UNLINK_LOG:
- nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
- op = FSFILT_OP_UNLINK;
- break;
- }
-
- 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;
break;
case FSFILT_OP_RENAME:
/* modify additional directory */
- nblocks += EXT3_DATA_TRANS_BLOCKS;
+ nblocks += EXT3_SINGLEDATA_TRANS_BLOCKS;
/* no break */
case FSFILT_OP_SYMLINK:
/* additional block + block bitmap + GDT for long symlink */
nblocks += 3;
/* no break */
- case FSFILT_OP_CREATE:
+ 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 */
/* no break */
case FSFILT_OP_LINK:
/* modify parent directory */
- nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS+EXT3_DATA_TRANS_BLOCKS;
+ nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
+ EXT3_DATA_TRANS_BLOCKS;
+ /* 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 +
+ EXT3_DATA_TRANS_BLOCKS;
+ /* quota chown log for each stripe */
+ nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
+ EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
break;
- case FSFILT_OP_CANCEL_UNLINK_LOG:
- blocksize = 1 << inode->i_blkbits;
- block_count = (blocksize - 1) + LLOG_CHUNK_SIZE;
- block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
- block_count = block_count * EXT3_DATA_TRANS_BLOCKS + 2;
- nblocks = 2 * 2 * block_count;
+ 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;
break;
default: CERROR("unknown transaction start op %d\n", op);
- LBUG();
+ 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);
+ nblocks = journal->j_max_transaction_buffers;
+ }
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);
+ else
+ CERROR("error starting handle for op %u (%u credits): rc %ld\n",
+ op, nblocks, PTR_ERR(handle));
return handle;
}
/* last_rcvd update */
needed += EXT3_DATA_TRANS_BLOCKS;
-#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.
*/
*/
static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
int niocount, struct niobuf_local *nb,
- void *desc_private)
+ void *desc_private, int logs)
{
journal_t *journal;
handle_t *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));
if (force_sync)
handle->h_sync = 1; /* recovery likes this */
- lock_kernel();
+ lock_24kernel();
rc = journal_stop(handle);
- unlock_kernel();
+ unlock_24kernel();
- // LASSERT(current->journal_info == NULL);
return rc;
}
static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
- void **wait_handle)
+ void **wait_handle)
{
+ unsigned long tid;
transaction_t *transaction;
- unsigned long tid, rtid;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
+ unsigned long rtid;
+#endif
handle_t *handle = h;
journal_t *journal;
int rc;
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;
}
RETURN(rc);
}
-#undef INLINE_EA
-#undef OLD_EA
static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
void *lmm, int lmm_size)
{
- int rc, old_ea = 0;
+ int rc;
-#ifdef INLINE_EA /* can go away before 1.0 - just for testing bug 2097 now */
- /* Nasty hack city - store stripe MD data in the block pointers if
- * it will fit, because putting it in an EA currently kills the MDS
- * performance. We'll fix this with "fast EAs" in the future.
- */
- if (inode->i_blocks == 0 && lmm_size <= sizeof(EXT3_I(inode)->i_data) -
- sizeof(EXT3_I(inode)->i_data[0])) {
- unsigned old_size = EXT3_I(inode)->i_data[0];
- if (old_size != 0) {
- LASSERT(old_size < sizeof(EXT3_I(inode)->i_data));
- CERROR("setting EA on %lu/%u again... interesting\n",
- inode->i_ino, inode->i_generation);
- }
+ LASSERT(down_trylock(&inode->i_sem) != 0);
- EXT3_I(inode)->i_data[0] = cpu_to_le32(lmm_size);
- memcpy(&EXT3_I(inode)->i_data[1], lmm, lmm_size);
- mark_inode_dirty(inode);
- return 0;
- }
-#endif
-#ifdef OLD_EA
- /* keep this when we get rid of OLD_EA (too noisy during conversion) */
- if (EXT3_I(inode)->i_file_acl /* || large inode EA flag */) {
+ 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);
- old_ea = 1;
- }
lock_kernel();
- /* this can go away before 1.0. For bug 2097 testing only. */
- rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_LUSTRE,
- XATTR_LUSTRE_MDS_OBJID, lmm, lmm_size, 0);
-#else
- lock_kernel();
rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size, 0);
- /* This tries to delete the old-format LOV EA, but only as long as we
- * have successfully saved the new-format LOV EA (we can always try
- * the conversion again the next time the file is accessed). It is
- * possible (although unlikely) that the new-format LOV EA couldn't be
- * saved because it ran out of space but we would need a file striped
- * over least 123 OSTs before the two EAs filled a 4kB block.
- *
- * This can be removed when all filesystems have converted to the
- * new EA format, but otherwise adds little if any overhead. If we
- * wanted backward compatibility for existing files, we could keep
- * the old EA around for a while but we'd have to clean it up later. */
- if (rc >= 0 && old_ea) {
- int err = ext3_xattr_set_handle(handle, inode,
- EXT3_XATTR_INDEX_LUSTRE,
- XATTR_LUSTRE_MDS_OBJID,
- NULL, 0, 0);
- if (err)
- CERROR("error deleting old LOV EA on %lu/%u: rc %d\n",
- inode->i_ino, inode->i_generation, err);
- }
-#endif
unlock_kernel();
if (rc)
LASSERT(down_trylock(&inode->i_sem) != 0);
lock_kernel();
- /* Keep support for reading "inline EAs" until we convert
- * users over to new format entirely. See bug 841/2097. */
- if (inode->i_blocks == 0 && EXT3_I(inode)->i_data[0]) {
- unsigned size = le32_to_cpu(EXT3_I(inode)->i_data[0]);
- void *handle;
-
- LASSERT(size < sizeof(EXT3_I(inode)->i_data));
- if (lmm) {
- if (size > lmm_size) {
- CERROR("inline EA on %lu/%u bad size %u > %u\n",
- inode->i_ino, inode->i_generation,
- size, lmm_size);
- return -ERANGE;
- }
- memcpy(lmm, &EXT3_I(inode)->i_data[1], size);
- }
-
-#ifndef INLINE_EA
- /* migrate LOV EA data to external block - keep same format */
- CWARN("DEBUG: migrate inline EA for inode %lu/%u to block\n",
- inode->i_ino, inode->i_generation);
-
- handle = journal_start(EXT3_JOURNAL(inode),
- EXT3_XATTR_TRANS_BLOCKS);
- if (!IS_ERR(handle)) {
- int err;
- rc = fsfilt_ext3_set_md(inode, handle,
- &EXT3_I(inode)->i_data[1],size);
- if (rc == 0) {
- memset(EXT3_I(inode)->i_data, 0,
- sizeof(EXT3_I(inode)->i_data));
- mark_inode_dirty(inode);
- }
- err = journal_stop(handle);
- if (err && rc == 0)
- rc = err;
- } else {
- rc = PTR_ERR(handle);
- }
-#endif
- unlock_kernel();
- return size;
- }
rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
XATTR_LUSTRE_MDS_LOV_EA, lmm, lmm_size);
- /* try old EA type if new one failed - MDS will convert it for us */
- if (rc == -ENODATA) {
- CDEBUG(D_INFO,"failed new LOV EA %d/%s from inode %lu: rc %d\n",
- EXT3_XATTR_INDEX_TRUSTED, XATTR_LUSTRE_MDS_LOV_EA,
- inode->i_ino, rc);
-
- rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_LUSTRE,
- XATTR_LUSTRE_MDS_OBJID, lmm, lmm_size);
- }
unlock_kernel();
/* This gives us the MD size */
if (rc < 0) {
CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
- EXT3_XATTR_INDEX_LUSTRE, XATTR_LUSTRE_MDS_OBJID,
+ EXT3_XATTR_INDEX_TRUSTED, XATTR_LUSTRE_MDS_LOV_EA,
inode->i_ino, rc);
memset(lmm, 0, lmm_size);
return (rc == -ENODATA) ? 0 : rc;
return rc;
}
+#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
+static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
+{
+ submit_bio(rw, bio);
+ return 0;
+}
+#else
+static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct kiobuf *bio)
+{
+ 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.)
+ */
+
+ blk_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
+
+ 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);
+ }
+ 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 rc;
+}
+#endif
+
static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
loff_t *off)
{
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, __u64 last_rcvd,
if (fcb == NULL)
RETURN(-ENOMEM);
- atomic_inc(&fcb_cache_count);
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);
- lock_kernel();
+ 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;
+ unsigned long tgen;
+ 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;
+ }
+
+ 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)) {
+ 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);
+ aflags |= 2; /* block have been already reserved */
+ pblock = ext3_mb_new_blocks(handle, inode, goal, &count, aflags, &err);
+ if (!pblock)
+ goto out;
+ EXT_ASSERT(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) {
+ /* 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++) {
+ if (cex->ec_type == EXT3_EXT_CACHE_EXTENT)
+ *(bp->created) = 0;
+ else
+ *(bp->created) = 1;
+ bp->created++;
+ *(bp->blocks) = cex->ec_start + i;
+ 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;
+
+ 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 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)
+{
+ 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)
{
- return ext3_map_inode_page(inode, page, blocks, created, create);
+ int rc;
+#ifdef EXT3_MULTIBLOCK_ALLOCATOR
+ if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
+ rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
+ blocks, created, create);
+ return rc;
+ }
+#endif
+ if (optional_sem != NULL)
+ down(optional_sem);
+ 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 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 << (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;
+ }
+ return needed + 3 * logs;
+ }
+
+ return 0;
+}
+
+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;
+}
+
+#ifdef HAVE_QUOTA_SUPPORT
+# include "fsfilt_ext3_quota.h"
+#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_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_send_bio = fsfilt_ext3_send_bio,
+ .fs_get_op_len = fsfilt_ext3_get_op_len,
+#ifdef HAVE_QUOTA_SUPPORT
+ .fs_quotactl = fsfilt_ext3_quotactl,
+ .fs_quotacheck = fsfilt_ext3_quotacheck,
+ .fs_quotainfo = fsfilt_ext3_quotainfo,
+ .fs_dquot = fsfilt_ext3_dquot,
+#endif
};
static int __init fsfilt_ext3_init(void)
{
int rc;
- //rc = ext3_xattr_register();
fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
sizeof(struct fsfilt_cb_data), 0,
0, NULL, NULL);
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);
- }
-
- //rc = ext3_xattr_unregister();
+ LASSERT(kmem_cache_destroy(fcb_cache) == 0);
}
module_init(fsfilt_ext3_init);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff