#include <linux/obd_support.h>
#include <linux/lustre_lite.h>
#include <linux/lustre_dlm.h>
+#include <linux/lustre_version.h>
#include "llite_internal.h"
/* methods */
static int ll_test_inode(struct inode *inode, void *opaque)
#endif
{
+ static int last_ino, last_gen, last_count;
struct lustre_md *md = opaque;
- if (!(md->body->valid & (OBD_MD_FLGENER | OBD_MD_FLID)))
- CERROR("invalid generation\n");
- CDEBUG(D_VFSTRACE, "comparing inode %p ino %lu/%u to body %u/%u\n",
- inode, inode->i_ino, inode->i_generation,
- md->body->ino, md->body->generation);
+ if (!(md->body->valid & (OBD_MD_FLGENER | OBD_MD_FLID))) {
+ CERROR("MDS body missing inum or generation\n");
+ return 0;
+ }
+
+ if (last_ino == md->body->ino && last_gen == md->body->generation &&
+ last_count < 500) {
+ last_count++;
+ } else {
+ if (last_count > 1)
+ CDEBUG(D_VFSTRACE, "compared %u/%u %u times\n",
+ last_ino, last_gen, last_count);
+ last_count = 0;
+ last_ino = md->body->ino;
+ last_gen = md->body->generation;
+ CDEBUG(D_VFSTRACE,
+ "comparing inode %p ino %lu/%u/%u to body %u/%u/%u\n",
+ inode, inode->i_ino, inode->i_generation,
+ ll_i2info(inode)->lli_mds,
+ md->body->ino, md->body->generation,
+ md->body->mds);
+ }
+#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
+ if (inode->i_ino != md->body->ino)
+ return 0;
+#endif
if (inode->i_generation != md->body->generation)
return 0;
+ if (ll_i2info(inode)->lli_mds != md->body->mds)
+ return 0;
+
/* Apply the attributes in 'opaque' to this inode */
- ll_update_inode(inode, md->body, md->lsm);
+ ll_update_inode(inode, md);
return 1;
}
ll_read_inode2(inode, opaque);
return 0;
}
+
struct inode *ll_iget(struct super_block *sb, ino_t hash,
struct lustre_md *md)
{
LASSERT(hash != 0);
inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
- if (!inode)
- return (NULL); /* removed ERR_PTR(-ENOMEM) -eeb */
-
- if (inode->i_state & I_NEW)
- unlock_new_inode(inode);
+ if (inode) {
+ if (inode->i_state & I_NEW)
+ unlock_new_inode(inode);
+ CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
+ inode->i_generation, inode);
+ }
- // XXX Coda always fills inodes, should Lustre?
return inode;
}
#else
struct inode *inode;
LASSERT(hash != 0);
inode = iget4(sb, hash, ll_test_inode, md);
- CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
- inode->i_generation, inode);
+ if (inode)
+ CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
+ inode->i_generation, inode);
return inode;
}
#endif
-static int ll_intent_to_lock_mode(struct lookup_intent *it)
-{
- /* CREAT needs to be tested before open (both could be set) */
- if (it->it_op & IT_CREAT)
- return LCK_PW;
- else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP))
- return LCK_PR;
-
- LBUG();
- RETURN(-EINVAL);
-}
-
-int ll_it_open_error(int phase, struct lookup_intent *it)
-{
- if (it_disposition(it, DISP_OPEN_OPEN)) {
- if (phase == DISP_OPEN_OPEN)
- return it->it_status;
- else
- return 0;
- }
-
- if (it_disposition(it, DISP_OPEN_CREATE)) {
- if (phase == DISP_OPEN_CREATE)
- return it->it_status;
- else
- return 0;
- }
-
- if (it_disposition(it, DISP_LOOKUP_EXECD)) {
- if (phase == DISP_LOOKUP_EXECD)
- return it->it_status;
- else
- return 0;
- }
- CERROR("it disp: %X, status: %d\n", it->it_disposition, it->it_status);
- LBUG();
- return 0;
-}
-
int ll_mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
void *data, int flag)
{
int rc;
struct lustre_handle lockh;
- struct inode *inode = lock->l_data;
ENTRY;
switch (flag) {
}
break;
case LDLM_CB_CANCELING: {
- /* Invalidate all dentries associated with this inode */
+ struct inode *inode = ll_inode_from_lock(lock);
+ __u64 bits = lock->l_policy_data.l_inodebits.bits;
+
+ /* For lookup locks: Invalidate all dentries associated with
+ this inode, for UPDATE locks - invalidate directory pages */
if (inode == NULL)
break;
+
+ if (bits & MDS_INODELOCK_UPDATE)
+ clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
+ &(ll_i2info(inode)->lli_flags));
+
+
if (lock->l_resource->lr_name.name[0] != inode->i_ino ||
lock->l_resource->lr_name.name[1] != inode->i_generation) {
LDLM_ERROR(lock, "data mismatch with ino %lu/%u",
inode->i_ino, inode->i_generation);
}
- if (S_ISDIR(inode->i_mode)) {
+
+ /* If lookup lock is cancelled, we just drop the dentry and
+ this will cause us to reget data from MDS when we'd want to
+ access this dentry/inode again. If this is lock on
+ other parts of inode that is cancelled, we do not need to do
+ much (but need to discard data from readdir, if any), since
+ abscence of lock will cause ll_revalidate_it (called from
+ stat() and similar functions) to renew the data anyway */
+ if (S_ISDIR(inode->i_mode) &&
+ (bits & MDS_INODELOCK_UPDATE)) {
CDEBUG(D_INODE, "invalidating inode %lu\n",
inode->i_ino);
- ll_invalidate_inode_pages(inode);
+ truncate_inode_pages(inode->i_mapping, 0);
}
-#warning FIXME: we should probably free this inode if there are no aliases
if (inode->i_sb->s_root &&
- inode != inode->i_sb->s_root->d_inode)
+ inode != inode->i_sb->s_root->d_inode &&
+ (bits & MDS_INODELOCK_LOOKUP))
ll_unhash_aliases(inode);
+ iput(inode);
break;
}
default:
{ .name = {inode->i_ino, inode->i_generation} };
struct obd_device *obddev = class_conn2obd(conn);
ENTRY;
+
RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, &res_id, flags,
opaque));
}
-void ll_prepare_mdc_op_data(struct mdc_op_data *data,
- struct inode *i1,
- struct inode *i2,
- const char *name,
- int namelen,
- int mode)
-{
- LASSERT(i1);
-
- data->ino1 = i1->i_ino;
- data->gen1 = i1->i_generation;
- data->typ1 = i1->i_mode & S_IFMT;
- data->gid1 = i1->i_gid;
-
- if (i2) {
- data->ino2 = i2->i_ino;
- data->gen2 = i2->i_generation;
- data->typ2 = i2->i_mode & S_IFMT;
- data->gid2 = i2->i_gid;
- } else {
- data->ino2 = 0;
- }
-
- data->name = name;
- data->namelen = namelen;
- data->mode = mode;
-}
-
-/*
- *This long block is all about fixing up the local state so that it is
- *correct as of the moment _before_ the operation was applied; that
- *way, the VFS will think that everything is normal and call Lustre's
- *regular VFS methods.
- *
- * If we're performing a creation, that means that unless the creation
- * failed with EEXIST, we should fake up a negative dentry.
- *
- * For everything else, we want to lookup to succeed.
- *
- * One additional note: if CREATE or OPEN succeeded, we add an extra
- * reference to the request because we need to keep it around until
- * ll_create/ll_open gets called.
- *
- * The server will return to us, in it_disposition, an indication of
- * exactly what it_status refers to.
- *
- * If DISP_OPEN_OPEN is set, then it_status refers to the open() call,
- * otherwise if DISP_OPEN_CREATE is set, then it status is the
- * creation failure mode. In either case, one of DISP_LOOKUP_NEG or
- * DISP_LOOKUP_POS will be set, indicating whether the child lookup
- * was successful.
- *
- * Else, if DISP_LOOKUP_EXECD then it_status is the rc of the child
- * lookup.
- */
-int ll_intent_lock(struct inode *parent, struct dentry **de,
- struct lookup_intent *it, int flags, intent_finish_cb intent_finish)
-{
- struct dentry *dentry = *de;
- struct inode *inode = dentry->d_inode;
- struct ll_sb_info *sbi = ll_i2sbi(parent);
- struct lustre_handle lockh;
- struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
- struct ptlrpc_request *request;
- int rc = 0;
- struct mds_body *mds_body;
- int mode;
- obd_id ino = 0;
- ENTRY;
-
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
- if (it && it->it_magic != INTENT_MAGIC) {
- CERROR("WARNING: uninitialized intent\n");
- LBUG();
- intent_init(it, IT_LOOKUP, 0);
- }
- if (it->it_op == IT_GETATTR ||
- it->it_op == 0)
- it->it_op = IT_LOOKUP;
-
-#endif
- if (!it ||it->it_op == IT_GETXATTR)
- it = &lookup_it;
-
- it->it_op_release = ll_intent_release;
-
- CDEBUG(D_DLMTRACE, "name: %*s, intent: %s\n", dentry->d_name.len,
- dentry->d_name.name, ldlm_it2str(it->it_op));
-
- if (dentry->d_name.len > EXT2_NAME_LEN)
- RETURN(-ENAMETOOLONG);
-
- /* This function may be called twice, we only once want to
- execute the request associated with the intent. If it was
- done already, we skip past this and use the results. */
- if (!it_disposition(it, DISP_ENQ_COMPLETE)) {
- struct mdc_op_data op_data;
-
- ll_prepare_mdc_op_data(&op_data, parent, dentry->d_inode,
- dentry->d_name.name, dentry->d_name.len,
- 0);
-
- rc = mdc_enqueue(&sbi->ll_mdc_conn, LDLM_PLAIN, it,
- ll_intent_to_lock_mode(it), &op_data,
- &lockh, NULL, 0, ldlm_completion_ast,
- ll_mdc_blocking_ast, NULL);
- if (rc < 0)
- RETURN(rc);
- memcpy(it->it_lock_handle, &lockh, sizeof(lockh));
- }
- request = it->it_data;
- LASSERT(request != NULL);
-
- /* non-zero it_disposition indicates that the server performed the
- * intent on our behalf. */
- LASSERT(it_disposition(it, DISP_IT_EXECD));
-
-
- mds_body = lustre_msg_buf(request->rq_repmsg, 1, sizeof(*mds_body));
- LASSERT(mds_body != NULL); /* mdc_enqueue checked */
- LASSERT_REPSWABBED(request, 1); /* mdc_enqueue swabbed */
-
- /* XXX everything with fids please, no ino's inode's etc */
- ino = mds_body->fid1.id;
- mode = mds_body->mode;
-
- /*We were called from revalidate2: did we find the same inode?*/
- if (inode &&
- (ino != inode->i_ino ||
- mds_body->fid1.generation != inode->i_generation)) {
- it_set_disposition(it, DISP_ENQ_COMPLETE);
- RETURN(-ESTALE);
- }
-
- /* If we're doing an IT_OPEN which did not result in an actual
- * successful open, then we need to remove the bit which saves
- * this request for unconditional replay. */
- if (it->it_op & IT_OPEN) {
- if (!it_disposition(it, DISP_OPEN_OPEN) ||
- it->it_status != 0) {
- unsigned long flags;
-
- spin_lock_irqsave (&request->rq_lock, flags);
- request->rq_replay = 0;
- spin_unlock_irqrestore (&request->rq_lock, flags);
- }
- }
-
- rc = ll_it_open_error(DISP_LOOKUP_EXECD, it);
- if (rc)
- GOTO(drop_req, rc);
-
- /* keep requests around for the multiple phases of the call
- * this shows the DISP_XX must guarantee we make it into the call
- */
- if (it_disposition(it, DISP_OPEN_CREATE))
- ptlrpc_request_addref(request);
- if (it_disposition(it, DISP_OPEN_OPEN))
- ptlrpc_request_addref(request);
-
- if (it->it_op & IT_CREAT) {
- /* XXX this belongs in ll_create_iit */
- } else if (it->it_op == IT_OPEN) {
- LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
- } else
- LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
-
- if (intent_finish != NULL) {
- struct lustre_handle old_lock;
- struct ldlm_lock *lock;
-
- rc = intent_finish(request, parent, de, it, 1, ino);
- dentry = *de; /* intent_finish may change *de */
- inode = dentry->d_inode;
- if (rc != 0)
- GOTO(drop_lock, rc);
-
- /* The intent processing may well have given us a lock different
- * from the one we requested. If we already have a matching
- * lock, then cancel the new one. (We have to do this here,
- * instead of in mdc_enqueue, because we need to use the child's
- * inode as the l_data to match, and that's not available until
- * intent_finish has performed the iget().) */
- lock = ldlm_handle2lock(&lockh);
- if (lock) {
- LDLM_DEBUG(lock, "matching against this");
- LDLM_LOCK_PUT(lock);
- memcpy(&old_lock, &lockh, sizeof(lockh));
- if (ldlm_lock_match(NULL,
- LDLM_FL_BLOCK_GRANTED |
- LDLM_FL_MATCH_DATA,
- NULL, LDLM_PLAIN, NULL, 0, LCK_NL,
- inode, &old_lock)) {
- ldlm_lock_decref_and_cancel(&lockh,
- it->it_lock_mode);
- memcpy(&lockh, &old_lock, sizeof(old_lock));
- memcpy(it->it_lock_handle, &lockh,
- sizeof(lockh));
- }
- }
-
- }
- ptlrpc_req_finished(request);
-
- CDEBUG(D_DENTRY, "D_IT dentry %p intent: %s status %d disp %x\n",
- dentry, ldlm_it2str(it->it_op), it->it_status, it->it_disposition);
-
- /* drop IT_LOOKUP locks */
- if (it->it_op == IT_LOOKUP)
- ll_intent_release(it);
- RETURN(rc);
-
- drop_lock:
- ll_intent_release(it);
- drop_req:
- ptlrpc_req_finished(request);
- RETURN(rc);
-}
-
/* Search "inode"'s alias list for a dentry that has the same name and parent as
* de. If found, return it. If not found, return de. */
struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
atomic_inc(&dentry->d_count);
iput(inode);
dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
+ CDEBUG(D_DENTRY, "alias dentry %*s (%p) parent %p inode %p "
+ "refc %d\n", de->d_name.len, de->d_name.name, de,
+ de->d_parent, de->d_inode, atomic_read(&de->d_count));
return dentry;
}
return de;
}
-static int
-lookup2_finish(struct ptlrpc_request *request,
- struct inode *parent, struct dentry **de,
- struct lookup_intent *it, int offset, obd_id ino)
+static int lookup_it_finish(struct ptlrpc_request *request, int offset,
+ struct lookup_intent *it, void *data)
{
+ struct it_cb_data *icbd = data;
+ struct dentry **de = icbd->icbd_childp;
+ struct inode *parent = icbd->icbd_parent;
struct ll_sb_info *sbi = ll_i2sbi(parent);
struct dentry *dentry = *de, *saved = *de;
struct inode *inode = NULL;
/* NB 1 request reference will be taken away by ll_intent_lock()
* when I return */
if (!it_disposition(it, DISP_LOOKUP_NEG)) {
- struct lustre_md md;
ENTRY;
- rc =mdc_req2lustre_md(request, offset, &sbi->ll_osc_conn, &md);
- if (rc)
+ rc = ll_prep_inode(sbi->ll_osc_exp, sbi->ll_mdc_exp,
+ &inode, request, offset, dentry->d_sb);
+ if (rc)
RETURN(rc);
- inode = ll_iget(dentry->d_sb, ino, &md);
- if (!inode) {
- /* free the lsm if we allocated one above */
- if (md.lsm != NULL)
- obd_free_memmd(&sbi->ll_osc_conn, &md.lsm);
- RETURN(-ENOMEM);
- } else if (md.lsm != NULL &&
- ll_i2info(inode)->lli_smd != md.lsm) {
- obd_free_memmd(&sbi->ll_osc_conn, &md.lsm);
- }
+ CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
+ inode, inode->i_ino, inode->i_generation);
+ mdc_set_lock_data(NULL, &it->d.lustre.it_lock_handle, inode);
/* If this is a stat, get the authoritative file size */
if (it->it_op == IT_GETATTR && S_ISREG(inode->i_mode) &&
ll_i2info(inode)->lli_smd != NULL) {
- struct ldlm_extent extent = {0, OBD_OBJECT_EOF};
- struct lustre_handle lockh = {0};
struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
+ struct ost_lvb lvb;
ldlm_error_t rc;
LASSERT(lsm->lsm_object_id != 0);
- rc = ll_extent_lock(NULL, inode, lsm, LCK_PR, &extent,
- &lockh);
- if (rc != ELDLM_OK) {
+ /* bug 2334: drop MDS lock before acquiring OST lock */
+ ll_intent_drop_lock(it);
+
+ rc = ll_glimpse_size(inode, &lvb);
+ if (rc) {
iput(inode);
- RETURN(-EIO);
+ RETURN(rc);
}
- ll_extent_unlock(NULL, inode, lsm, LCK_PR, &lockh);
+ inode->i_size = lvb.lvb_size;
}
dentry = *de = ll_find_alias(inode, dentry);
-
- /* We asked for a lock on the directory, and may have been
- * granted a lock on the inode. Just in case, fixup the data
- * pointer. */
- CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
- inode, inode->i_ino, inode->i_generation);
- ldlm_lock_set_data((struct lustre_handle*)it->it_lock_handle,
- inode);
} else {
ENTRY;
}
dentry->d_op = &ll_d_ops;
ll_set_dd(dentry);
- if (dentry == saved)
+ if (dentry == saved) {
d_add(dentry, inode);
+ }
RETURN(0);
}
+
static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
+ struct nameidata *nd,
struct lookup_intent *it, int flags)
{
struct dentry *save = dentry, *retval;
+ struct ll_fid pfid;
+ struct ll_uctxt ctxt;
+ struct it_cb_data icbd;
+ struct ptlrpc_request *req = NULL;
+ struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
int rc;
ENTRY;
+ if (dentry->d_name.len > EXT3_NAME_LEN)
+ RETURN(ERR_PTR(-ENAMETOOLONG));
+
CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p),intent=%s\n",
dentry->d_name.name, parent->i_ino, parent->i_generation,
parent, LL_IT2STR(it));
- if (d_mountpoint(dentry)) {
+ if (d_mountpoint(dentry))
CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
- }
- rc = ll_intent_lock(parent, &dentry, it, flags, lookup2_finish);
- if (rc < 0) {
- CDEBUG(D_INFO, "ll_intent_lock: %d\n", rc);
+ if (nd != NULL)
+ nd->mnt->mnt_last_used = jiffies;
+
+ ll_frob_intent(&it, &lookup_it);
+
+ icbd.icbd_childp = &dentry;
+ icbd.icbd_parent = parent;
+ ll_inode2fid(&pfid, parent);
+ ll_i2uctxt(&ctxt, parent, NULL);
+
+ rc = md_intent_lock(ll_i2mdcexp(parent), &ctxt, &pfid,
+ dentry->d_name.name, dentry->d_name.len, NULL, 0,
+ NULL, it, flags, &req, ll_mdc_blocking_ast);
+ if (rc < 0)
+ GOTO(out, retval = ERR_PTR(rc));
+
+ rc = lookup_it_finish(req, 1, it, &icbd);
+ if (rc != 0) {
+ ll_intent_release(it);
GOTO(out, retval = ERR_PTR(rc));
}
+ ll_lookup_finish_locks(it, dentry);
+
+ if (nd &&
+ dentry->d_inode != NULL && dentry->d_inode->i_mode & S_ISUID &&
+ S_ISDIR(dentry->d_inode->i_mode) &&
+ (flags & LOOKUP_CONTINUE || (it->it_op & (IT_CHDIR | IT_OPEN))))
+ ll_dir_process_mount_object(dentry, nd->mnt);
+
if (dentry == save)
GOTO(out, retval = NULL);
else
GOTO(out, retval = dentry);
out:
+ if (req)
+ ptlrpc_req_finished(req);
+ if (dentry->d_inode)
+ CDEBUG(D_INODE, "lookup 0x%p in %lu/%lu: %*s -> %lu/%lu\n",
+ dentry,
+ (unsigned long) parent->i_ino,
+ (unsigned long) parent->i_generation,
+ dentry->d_name.len, dentry->d_name.name,
+ (unsigned long) dentry->d_inode->i_ino,
+ (unsigned long) dentry->d_inode->i_generation);
+ else
+ CDEBUG(D_INODE, "lookup 0x%p in %lu/%lu: %*s -> ??\n",
+ dentry,
+ (unsigned long) parent->i_ino,
+ (unsigned long) parent->i_generation,
+ dentry->d_name.len, dentry->d_name.name);
return retval;
}
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
-static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
+static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
struct nameidata *nd)
{
struct dentry *de;
ENTRY;
- if (nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
- de = ll_lookup_it(parent, dentry, &nd->it, nd->flags);
- else
- de = ll_lookup_it(parent, dentry, NULL, 0);
+ if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
+ de = ll_lookup_it(parent, dentry, nd, &nd->intent, nd->flags);
+ else
+ de = ll_lookup_it(parent, dentry, nd, NULL, 0);
RETURN(de);
}
#endif
-static int ll_mdc_unlink(struct inode *dir, struct inode *child, __u32 mode,
- const char *name, int len)
-{
- struct ptlrpc_request *request = NULL;
- struct mds_body *body;
- struct lov_mds_md *eadata;
- struct lov_stripe_md *lsm = NULL;
- struct obd_trans_info oti = { 0 };
- struct mdc_op_data op_data;
- struct obdo *oa;
- int rc;
- ENTRY;
-
- ll_prepare_mdc_op_data(&op_data, dir, child, name, len, mode);
- rc = mdc_unlink(&ll_i2sbi(dir)->ll_mdc_conn, &op_data, &request);
- if (rc)
- GOTO(out, rc);
- /* req is swabbed so this is safe */
- body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
-
- if (!(body->valid & OBD_MD_FLEASIZE))
- GOTO(out, rc = 0);
-
- if (body->eadatasize == 0) {
- CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
- GOTO(out, rc = -EPROTO);
- }
-
- /* The MDS sent back the EA because we unlinked the last reference
- * to this file. Use this EA to unlink the objects on the OST.
- * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
- * check it is complete and sensible. */
- eadata = lustre_swab_repbuf(request, 1, body->eadatasize, NULL);
- LASSERT(eadata != NULL);
- if (eadata == NULL) {
- CERROR("Can't unpack MDS EA data\n");
- GOTO(out, rc = -EPROTO);
- }
-
- rc = obd_unpackmd(ll_i2obdconn(dir), &lsm, eadata, body->eadatasize);
- if (rc < 0) {
- CERROR("obd_unpackmd: %d\n", rc);
- GOTO(out, rc);
- }
- LASSERT(rc >= sizeof(*lsm));
-
- oa = obdo_alloc();
- if (oa == NULL)
- GOTO(out_free_memmd, rc = -ENOMEM);
-
- oa->o_id = lsm->lsm_object_id;
- oa->o_mode = body->mode & S_IFMT;
- oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
-
- if (body->valid & OBD_MD_FLCOOKIE) {
- oa->o_valid |= OBD_MD_FLCOOKIE;
- oti.oti_logcookies = lustre_msg_buf(request->rq_repmsg, 3,
- body->eadatasize);
- }
-
- rc = obd_destroy(ll_i2obdconn(dir), oa, lsm, &oti);
- obdo_free(oa);
- if (rc)
- CERROR("obd destroy objid 0x"LPX64" error %d\n",
- lsm->lsm_object_id, rc);
- out_free_memmd:
- obd_free_memmd(ll_i2obdconn(dir), &lsm);
- out:
- ptlrpc_req_finished(request);
- return rc;
-}
-
/* We depend on "mode" being set with the proper file type/umask by now */
static struct inode *ll_create_node(struct inode *dir, const char *name,
int namelen, const void *data, int datalen,
int mode, __u64 extra,
struct lookup_intent *it)
{
- struct inode *inode;
+ struct inode *inode = NULL;
struct ptlrpc_request *request = NULL;
struct ll_sb_info *sbi = ll_i2sbi(dir);
- struct lustre_md md;
int rc;
ENTRY;
- LASSERT(it && it->it_disposition);
+ LASSERT(it && it->d.lustre.it_disposition);
- ll_invalidate_inode_pages(dir);
-
- request = it->it_data;
- rc = mdc_req2lustre_md(request, 1, &sbi->ll_osc_conn, &md);
- if (rc) {
+ request = it->d.lustre.it_data;
+ rc = ll_prep_inode(sbi->ll_osc_exp, sbi->ll_mdc_exp,
+ &inode, request, 1, dir->i_sb);
+ if (rc)
GOTO(out, inode = ERR_PTR(rc));
- }
- inode = ll_iget(dir->i_sb, md.body->ino, &md);
- if (!inode || is_bad_inode(inode)) {
- /* XXX might need iput() for bad inode */
- int rc = -EIO;
- CERROR("new_inode -fatal: rc %d\n", rc);
- LBUG();
- GOTO(out, rc);
- }
LASSERT(list_empty(&inode->i_dentry));
- CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
+ /* We asked for a lock on the directory, but were granted a
+ * lock on the inode. Since we finally have an inode pointer,
+ * stuff it in the lock. */
+ CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
inode, inode->i_ino, inode->i_generation);
- ldlm_lock_set_data((struct lustre_handle*)it->it_lock_handle,
- inode);
-
+ mdc_set_lock_data(NULL, &it->d.lustre.it_lock_handle, inode);
EXIT;
out:
ptlrpc_req_finished(request);
* If the create succeeds, we fill in the inode information
* with d_instantiate().
*/
-static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode, struct lookup_intent *it)
+static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
+ struct lookup_intent *it)
{
struct inode *inode;
- struct ptlrpc_request *request = it->it_data;
+ struct ptlrpc_request *request = it->d.lustre.it_data;
+ struct obd_export *mdc_exp = ll_i2mdcexp(dir);
int rc = 0;
ENTRY;
dentry->d_name.name, dir->i_ino, dir->i_generation, dir,
LL_IT2STR(it));
- rc = ll_it_open_error(DISP_OPEN_CREATE, it);
- if (rc) {
- ptlrpc_req_finished(request);
+ rc = it_open_error(DISP_OPEN_CREATE, it);
+ if (rc)
RETURN(rc);
- }
- mdc_store_inode_generation(request, 2, 1);
+ mdc_store_inode_generation(mdc_exp, request, 2, 1);
inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
NULL, 0, mode, 0, it);
if (IS_ERR(inode)) {
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
{
- return ll_create_it(dir, dentry, mode, &nd->it);
+ return ll_create_it(dir, dentry, mode, &nd->intent);
}
#endif
static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
{
+ struct ptlrpc_request *request = NULL;
struct inode *dir = nd->dentry->d_inode;
const char *name = nd->last.name;
int len = nd->last.len;
- struct ptlrpc_request *request = NULL;
- time_t time = LTIME_S(CURRENT_TIME);
struct ll_sb_info *sbi = ll_i2sbi(dir);
struct mdc_op_data op_data;
int err = -EMLINK;
CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
name, dir->i_ino, dir->i_generation, dir);
- if (dir->i_nlink >= EXT2_LINK_MAX)
+ if (dir->i_nlink >= EXT3_LINK_MAX)
RETURN(err);
mode &= ~current->fs->umask;
switch (mode & S_IFMT) {
- case 0:
+ case 0:
case S_IFREG:
mode |= S_IFREG; /* for mode = 0 case, fallthrough */
- case S_IFCHR:
+ case S_IFCHR:
case S_IFBLK:
- case S_IFIFO:
+ case S_IFIFO:
case S_IFSOCK:
ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
- err = mdc_create(&sbi->ll_mdc_conn, &op_data, NULL, 0, mode,
- current->fsuid, current->fsgid, time,
- rdev, &request);
+ err = md_create(sbi->ll_mdc_exp, &op_data, NULL, 0, mode,
+ current->fsuid, current->fsgid, rdev, &request);
ptlrpc_req_finished(request);
break;
case S_IFDIR:
RETURN(err);
}
+static int ll_mknod(struct inode *dir, struct dentry *child, int mode,
+ ll_dev_t rdev)
+{
+ struct ptlrpc_request *request = NULL;
+ struct inode *inode = NULL;
+ const char *name = child->d_name.name;
+ int len = child->d_name.len;
+ struct ll_sb_info *sbi = ll_i2sbi(dir);
+ struct mdc_op_data op_data;
+ int err = -EMLINK;
+ ENTRY;
+
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
+ name, dir->i_ino, dir->i_generation, dir);
+
+ if (dir->i_nlink >= EXT3_LINK_MAX)
+ RETURN(err);
+
+ mode &= ~current->fs->umask;
+
+ switch (mode & S_IFMT) {
+ case 0:
+ case S_IFREG:
+ mode |= S_IFREG; /* for mode = 0 case, fallthrough */
+ case S_IFCHR:
+ case S_IFBLK:
+ case S_IFIFO:
+ case S_IFSOCK:
+ ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
+ err = md_create(sbi->ll_mdc_exp, &op_data, NULL, 0, mode,
+ current->fsuid, current->fsgid, rdev, &request);
+ err = ll_prep_inode(sbi->ll_osc_exp, sbi->ll_mdc_exp,
+ &inode, request, 0, child->d_sb);
+ if (err)
+ GOTO(out_err, err);
+ break;
+ case S_IFDIR:
+ RETURN(-EPERM);
+ break;
+ default:
+ RETURN(-EINVAL);
+ }
+
+ d_instantiate(child, inode);
+ out_err:
+ ptlrpc_req_finished(request);
+ RETURN(err);
+}
+
static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
{
struct inode *dir = nd->dentry->d_inode;
const char *name = nd->last.name;
int len = nd->last.len;
struct ptlrpc_request *request = NULL;
- time_t time = LTIME_S(CURRENT_TIME);
struct ll_sb_info *sbi = ll_i2sbi(dir);
struct mdc_op_data op_data;
int err = -EMLINK;
CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p),target=%s\n",
name, dir->i_ino, dir->i_generation, dir, tgt);
- if (dir->i_nlink >= EXT2_LINK_MAX)
+ if (dir->i_nlink >= EXT3_LINK_MAX)
RETURN(err);
ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
- err = mdc_create(&sbi->ll_mdc_conn, &op_data,
- tgt, strlen(tgt) + 1, S_IFLNK | S_IRWXUGO,
- current->fsuid, current->fsgid, time, 0, &request);
+ err = md_create(sbi->ll_mdc_exp, &op_data,
+ tgt, strlen(tgt) + 1, S_IFLNK | S_IRWXUGO,
+ current->fsuid, current->fsgid, 0, &request);
ptlrpc_req_finished(request);
RETURN(err);
}
dir->i_ino, dir->i_generation, dir, name);
ll_prepare_mdc_op_data(&op_data, src, dir, name, len, 0);
- err = mdc_link(&sbi->ll_mdc_conn, &op_data, &request);
+ err = md_link(sbi->ll_mdc_exp, &op_data, &request);
ptlrpc_req_finished(request);
RETURN(err);
const char *name = nd->last.name;
int len = nd->last.len;
struct ptlrpc_request *request = NULL;
- time_t time = LTIME_S(CURRENT_TIME);
struct ll_sb_info *sbi = ll_i2sbi(dir);
struct mdc_op_data op_data;
int err = -EMLINK;
CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
name, dir->i_ino, dir->i_generation, dir);
- if (dir->i_nlink >= EXT2_LINK_MAX)
- RETURN(err);
-
mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
- err = mdc_create(&sbi->ll_mdc_conn, &op_data, NULL, 0, mode,
- current->fsuid, current->fsgid, time, 0, &request);
+ err = md_create(sbi->ll_mdc_exp, &op_data, NULL, 0, mode,
+ current->fsuid, current->fsgid, 0, &request);
ptlrpc_req_finished(request);
RETURN(err);
}
struct inode *dir = nd->dentry->d_inode;
const char *name = nd->last.name;
int len = nd->last.len;
+ struct ptlrpc_request *request = NULL;
+ struct mdc_op_data op_data;
int rc;
ENTRY;
CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
name, dir->i_ino, dir->i_generation, dir);
- rc = ll_mdc_unlink(dir, NULL, S_IFDIR, name, len);
+ ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, S_IFDIR);
+ rc = md_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
+ ptlrpc_req_finished(request);
RETURN(rc);
}
+int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
+{
+ struct mds_body *body;
+ struct lov_mds_md *eadata;
+ struct lov_stripe_md *lsm = NULL;
+ struct obd_trans_info oti = { 0 };
+ struct obdo *oa;
+ int rc;
+ ENTRY;
+
+ /* req is swabbed so this is safe */
+ body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
+
+ if (!(body->valid & OBD_MD_FLEASIZE))
+ RETURN(0);
+
+ if (body->eadatasize == 0) {
+ CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
+ GOTO(out, rc = -EPROTO);
+ }
+
+ /* The MDS sent back the EA because we unlinked the last reference
+ * to this file. Use this EA to unlink the objects on the OST.
+ * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
+ * check it is complete and sensible. */
+ eadata = lustre_swab_repbuf(request, 1, body->eadatasize, NULL);
+ LASSERT(eadata != NULL);
+ if (eadata == NULL) {
+ CERROR("Can't unpack MDS EA data\n");
+ GOTO(out, rc = -EPROTO);
+ }
+
+ rc = obd_unpackmd(ll_i2obdexp(dir), &lsm, eadata, body->eadatasize);
+ if (rc < 0) {
+ CERROR("obd_unpackmd: %d\n", rc);
+ GOTO(out, rc);
+ }
+ LASSERT(rc >= sizeof(*lsm));
+
+ oa = obdo_alloc();
+ if (oa == NULL)
+ GOTO(out_free_memmd, rc = -ENOMEM);
+
+ oa->o_id = lsm->lsm_object_id;
+ oa->o_gr = lsm->lsm_object_gr;
+ oa->o_mode = body->mode & S_IFMT;
+ oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
+
+ if (body->valid & OBD_MD_FLCOOKIE) {
+ oa->o_valid |= OBD_MD_FLCOOKIE;
+ oti.oti_logcookies =
+ lustre_msg_buf(request->rq_repmsg, 2,
+ sizeof(struct llog_cookie) *
+ lsm->lsm_stripe_count);
+ if (oti.oti_logcookies == NULL) {
+ oa->o_valid &= ~OBD_MD_FLCOOKIE;
+ body->valid &= ~OBD_MD_FLCOOKIE;
+ }
+ }
+
+ rc = obd_destroy(ll_i2obdexp(dir), oa, lsm, &oti);
+ obdo_free(oa);
+ if (rc)
+ CERROR("obd destroy objid "LPX64" error %d\n",
+ lsm->lsm_object_id, rc);
+ out_free_memmd:
+ obd_free_memmd(ll_i2obdexp(dir), &lsm);
+ out:
+ return rc;
+}
+
static int ll_unlink_raw(struct nameidata *nd)
{
struct inode *dir = nd->dentry->d_inode;
const char *name = nd->last.name;
int len = nd->last.len;
+ struct ptlrpc_request *request = NULL;
+ struct mdc_op_data op_data;
int rc;
ENTRY;
CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
name, dir->i_ino, dir->i_generation, dir);
- rc = ll_mdc_unlink(dir, NULL, S_IFREG, name, len);
+ ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
+ rc = md_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = ll_objects_destroy(request, dir);
+ out:
+ ptlrpc_req_finished(request);
RETURN(rc);
}
src, newname, tgt->i_ino, tgt->i_generation, tgt);
ll_prepare_mdc_op_data(&op_data, src, tgt, NULL, 0, 0);
- err = mdc_rename(&sbi->ll_mdc_conn, &op_data,
- oldname, oldlen, newname, newlen, &request);
+ err = md_rename(sbi->ll_mdc_exp, &op_data,
+ oldname, oldlen, newname, newlen, &request);
+ if (!err) {
+ err = ll_objects_destroy(request, src);
+ }
+
ptlrpc_req_finished(request);
RETURN(err);
}
struct inode_operations ll_dir_inode_operations = {
- link_raw: ll_link_raw,
- unlink_raw: ll_unlink_raw,
- symlink_raw: ll_symlink_raw,
- mkdir_raw: ll_mkdir_raw,
- rmdir_raw: ll_rmdir_raw,
- mknod_raw: ll_mknod_raw,
- rename_raw: ll_rename_raw,
- setattr: ll_setattr,
- setattr_raw: ll_setattr_raw,
+ .link_raw = ll_link_raw,
+ .unlink_raw = ll_unlink_raw,
+ .symlink_raw = ll_symlink_raw,
+ .mkdir_raw = ll_mkdir_raw,
+ .rmdir_raw = ll_rmdir_raw,
+ .mknod_raw = ll_mknod_raw,
+ .mknod = ll_mknod,
+ .rename_raw = ll_rename_raw,
+ .setattr = ll_setattr,
+ .setattr_raw = ll_setattr_raw,
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- create_it: ll_create_it,
- lookup_it: ll_lookup_it,
- revalidate_it: ll_inode_revalidate_it,
+ .create_it = ll_create_it,
+ .lookup_it = ll_lookup_it,
+ .revalidate_it = ll_inode_revalidate_it,
#else
- lookup_it: ll_lookup_nd,
- create_nd: ll_create_nd,
- getattr_it: ll_getattr,
+ .lookup = ll_lookup_nd,
+ .create = ll_create_nd,
+ .getattr_it = ll_getattr,
#endif
};