#include <linux/obd_class.h>
#include <linux/lustre_mds.h>
#include <linux/lustre_dlm.h>
+#include <linux/lustre_sec.h>
#include <linux/lprocfs_status.h>
+#include <linux/lustre_acl.h>
+#include <linux/lustre_gs.h>
+#include <linux/lustre_lite.h>
#include "mdc_internal.h"
int it_disposition(struct lookup_intent *it, int flag)
{
- return it->d.lustre.it_disposition & flag;
+ return LUSTRE_IT(it)->it_disposition & flag;
}
EXPORT_SYMBOL(it_disposition);
void it_set_disposition(struct lookup_intent *it, int flag)
{
- it->d.lustre.it_disposition |= flag;
+ LUSTRE_IT(it)->it_disposition |= flag;
}
EXPORT_SYMBOL(it_set_disposition);
-static void mdc_id2mdc_op_data(struct mdc_op_data *data,
- struct lustre_id *f1,
- struct lustre_id *f2,
- const char *name,
- int namelen, int mode)
+static void mdc_id2mdc_data(struct mdc_op_data *data,
+ struct lustre_id *f1,
+ struct lustre_id *f2,
+ const char *name,
+ int namelen, int mode)
{
LASSERT(data);
LASSERT(f1);
{
if (it_disposition(it, DISP_OPEN_OPEN)) {
if (phase == DISP_OPEN_OPEN)
- return it->d.lustre.it_status;
+ return LUSTRE_IT(it)->it_status;
else
return 0;
}
if (it_disposition(it, DISP_OPEN_CREATE)) {
if (phase == DISP_OPEN_CREATE)
- return it->d.lustre.it_status;
+ return LUSTRE_IT(it)->it_status;
else
return 0;
}
if (it_disposition(it, DISP_LOOKUP_EXECD)) {
if (phase == DISP_LOOKUP_EXECD)
- return it->d.lustre.it_status;
+ return LUSTRE_IT(it)->it_status;
else
return 0;
}
if (it_disposition(it, DISP_IT_EXECD)) {
if (phase == DISP_IT_EXECD)
- return it->d.lustre.it_status;
+ return LUSTRE_IT(it)->it_status;
else
return 0;
}
- CERROR("it disp: %X, status: %d\n", it->d.lustre.it_disposition,
- it->d.lustre.it_status);
+ CERROR("it disp: %X, status: %d\n", LUSTRE_IT(it)->it_disposition,
+ LUSTRE_IT(it)->it_status);
LBUG();
return 0;
}
lock = ldlm_handle2lock(lockh);
LASSERT(lock != NULL);
- l_lock(&lock->l_resource->lr_namespace->ns_lock);
+ lock_res_and_lock(lock);
#ifdef __KERNEL__
if (lock->l_ast_data && lock->l_ast_data != data) {
- struct inode *new_inode = data;
+ struct ll_inode_info *lli = ll_i2info(data);
struct inode *old_inode = lock->l_ast_data;
- LASSERTF(old_inode->i_state & I_FREEING,
- "Found existing inode %p/%lu/%u state %lu in lock: "
- "setting data to %p/%lu/%u\n", old_inode,
- old_inode->i_ino, old_inode->i_generation,
- old_inode->i_state, new_inode, new_inode->i_ino,
- new_inode->i_generation);
+ struct inode *new_inode = data;
+ LASSERT(lli->lli_inode_magic == LLI_INODE_MAGIC);
+ if (!(old_inode->i_state & I_FREEING)) {
+ CERROR("Found existing inode %p/%lu/%u state %lu "
+ "in lock: setting data to %p/%lu/%u\n",
+ old_inode, old_inode->i_ino,
+ old_inode->i_generation, old_inode->i_state,
+ new_inode, new_inode->i_ino,
+ new_inode->i_generation);
+ unlock_res_and_lock(lock);
+ LBUG();
+ }
}
#endif
lock->l_ast_data = data;
- l_unlock(&lock->l_resource->lr_namespace->ns_lock);
+ unlock_res_and_lock(lock);
LDLM_LOCK_PUT(lock);
EXIT;
return 0;
}
+static inline void
+mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
+{
+ /* Don't hold error requests for replay. */
+ if (req->rq_replay) {
+ unsigned long irqflags;
+ spin_lock_irqsave(&req->rq_lock, irqflags);
+ req->rq_replay = 0;
+ spin_unlock_irqrestore(&req->rq_lock, irqflags);
+ }
+ if (rc && req->rq_transno != 0) {
+ DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
+ LBUG();
+ }
+}
+
/* We always reserve enough space in the reply packet for a stripe MD, because
* we don't know in advance the file type. */
int mdc_enqueue(struct obd_export *exp,
ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
struct ldlm_intent *lit;
struct ldlm_request *lockreq;
- struct ldlm_reply *dlm_rep;
int reqsize[6] = {[MDS_REQ_SECDESC_OFF] = 0,
[MDS_REQ_INTENT_LOCKREQ_OFF] = sizeof(*lockreq),
[MDS_REQ_INTENT_IT_OFF] = sizeof(*lit)};
- int repsize[4] = {sizeof(struct ldlm_reply),
+ int repsize[8] = {sizeof(struct ldlm_reply),
sizeof(struct mds_body),
- obddev->u.cli.cl_max_mds_easize,
- obddev->u.cli.cl_max_mds_cookiesize};
+ obddev->u.cli.cl_max_mds_easize};
int req_buffers = 3, reply_buffers = 0;
int rc, flags = LDLM_FL_HAS_INTENT;
+ struct ldlm_reply *dlm_rep = NULL;
void *eadata;
unsigned long irqflags;
ENTRY;
// LDLM_DEBUG_NOLOCK("mdsintent=%s,name=%s,dir=%lu",
// ldlm_it2str(it->it_op), it_name, it_inode->i_ino);
- reqsize[0] = mdc_get_secdesc_size();
+ reqsize[0] = lustre_secdesc_size();
if (it->it_op & IT_OPEN) {
+ struct lustre_intent_data *lustre_data =
+ (struct lustre_intent_data *) it->d.fs_data;
it->it_create_mode |= S_IFREG;
it->it_create_mode &= ~current->fs->umask;
reqsize[req_buffers++] = sizeof(struct mds_rec_create);
reqsize[req_buffers++] = data->namelen + 1;
reqsize[req_buffers++] = obddev->u.cli.cl_max_mds_easize;
+ /*pack the lustre key*/
+ if (lustre_data->it_key_size > 0 && lustre_data->it_key)
+ reqsize[req_buffers++] = lustre_data->it_key_size;
+
req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
LDLM_ENQUEUE, req_buffers, reqsize, NULL);
if (!req)
/* pack the intended request */
mdc_open_pack(req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF, data,
- it->it_create_mode, 0, it->it_flags,
- lmm, lmmsize);
+ it->it_create_mode, 0, it->it_flags, lmm, lmmsize,
+ lustre_data->it_key, lustre_data->it_key_size);
/* get ready for the reply */
reply_buffers = 3;
- req->rq_replen = lustre_msg_size(3, repsize);
+ repsize[reply_buffers++] = sizeof(int);
+ repsize[reply_buffers++] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
+ repsize[reply_buffers++] = sizeof(int);
+ repsize[reply_buffers++] = sizeof(struct crypto_key);
+ repsize[reply_buffers++] = sizeof(struct lustre_capa);
+ req->rq_replen = lustre_msg_size(reply_buffers, repsize);
} else if (it->it_op & (IT_GETATTR | IT_LOOKUP | IT_CHDIR)) {
- int valid = data->valid | OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
+ __u64 valid = data->valid | OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE |
+ OBD_MD_FLACL | OBD_MD_FLKEY;
+
+ /* we don't expect xattr retrieve could reach here */
+ LASSERT(!(valid & (OBD_MD_FLXATTR | OBD_MD_FLXATTRLIST)));
reqsize[req_buffers++] = sizeof(struct mds_body);
reqsize[req_buffers++] = data->namelen + 1;
/* get ready for the reply */
reply_buffers = 3;
- req->rq_replen = lustre_msg_size(3, repsize);
+ repsize[reply_buffers++] = sizeof(int);
+ repsize[reply_buffers++] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
+ repsize[reply_buffers++] = sizeof(int);
+ repsize[reply_buffers++] = sizeof(struct crypto_key);
+ req->rq_replen = lustre_msg_size(reply_buffers, repsize);
} else if (it->it_op == IT_READDIR) {
policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
RETURN(-EINVAL);
}
- mdc_pack_secdesc(req, reqsize[0]);
+ lustre_pack_secdesc(req, reqsize[0]);
mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
rc = ldlm_cli_enqueue(exp, req, obddev->obd_namespace, res_id,
/* This can go when we're sure that this can never happen */
LASSERT(rc != -ENOENT);
+ /* We need dlm_rep to be assigned this early, to check lock mode of
+ returned lock from request to avoid possible race with lock
+ conversion */
+ if (rc == ELDLM_LOCK_ABORTED || !rc) {
+ dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
+ LASSERT(dlm_rep != NULL); /* checked by ldlm_cli_enqueue() */
+ }
if (rc == ELDLM_LOCK_ABORTED) {
lock_mode = 0;
memset(lockh, 0, sizeof(*lockh));
rc = 0;
} else if (rc != 0) {
CERROR("ldlm_cli_enqueue: %d\n", rc);
- LASSERT (rc < 0);
+ LASSERTF(rc < 0, "rc = %d\n", rc);
+ mdc_clear_replay_flag(req, rc);
ptlrpc_req_finished(req);
RETURN(rc);
} else { /* rc = 0 */
/* If the server gave us back a different lock mode, we should
* fix up our variables. */
- if (lock->l_req_mode != lock_mode) {
- ldlm_lock_addref(lockh, lock->l_req_mode);
+ if (dlm_rep->lock_desc.l_req_mode != lock_mode) {
+ ldlm_lock_addref(lockh, dlm_rep->lock_desc.l_req_mode);
ldlm_lock_decref(lockh, lock_mode);
- lock_mode = lock->l_req_mode;
+ lock_mode = dlm_rep->lock_desc.l_req_mode;
}
ldlm_lock_allow_match(lock);
LDLM_LOCK_PUT(lock);
}
- dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
- LASSERT(dlm_rep != NULL); /* checked by ldlm_cli_enqueue() */
LASSERT_REPSWABBED(req, 0); /* swabbed by ldlm_cli_enqueue() */
- it->d.lustre.it_disposition = (int) dlm_rep->lock_policy_res1;
- it->d.lustre.it_status = (int) dlm_rep->lock_policy_res2;
- it->d.lustre.it_lock_mode = lock_mode;
- it->d.lustre.it_data = req;
+ LUSTRE_IT(it)->it_disposition = (int) dlm_rep->lock_policy_res1;
+ LUSTRE_IT(it)->it_status = (int) dlm_rep->lock_policy_res2;
+ LUSTRE_IT(it)->it_lock_mode = lock_mode;
+ LUSTRE_IT(it)->it_data = req;
- if (it->d.lustre.it_status < 0 && req->rq_replay) {
- LASSERT(req->rq_transno == 0);
- /* Don't hold error requests for replay. */
- spin_lock(&req->rq_lock);
- req->rq_replay = 0;
- spin_unlock(&req->rq_lock);
- }
+ if (LUSTRE_IT(it)->it_status < 0 && req->rq_replay)
+ mdc_clear_replay_flag(req, LUSTRE_IT(it)->it_status);
- DEBUG_REQ(D_RPCTRACE, req, "disposition: %x, status: %d",
- it->d.lustre.it_disposition, it->d.lustre.it_status);
+ DEBUG_REQ(D_DLMTRACE, req, "disposition: %x, status: %d",
+ LUSTRE_IT(it)->it_disposition, LUSTRE_IT(it)->it_status);
/* We know what to expect, so we do any byte flipping required here */
- LASSERT(reply_buffers == 4 || reply_buffers == 3 || reply_buffers == 1);
+ LASSERT(reply_buffers == 1 || reply_buffers == 3 ||
+ reply_buffers == 5 || reply_buffers == 7 ||
+ reply_buffers == 8);
if (reply_buffers >= 3) {
struct mds_body *body;
body = lustre_swab_repbuf(req, 1, sizeof (*body),
- lustre_swab_mds_body);
+ lustre_swab_mds_body);
if (body == NULL) {
CERROR ("Can't swab mds_body\n");
RETURN (-EPROTO);
LASSERT(replayea);
memcpy(replayea, eadata, body->eadatasize);
- LASSERT(req->rq_reqmsg->bufcount == 6);
+ LASSERT(req->rq_reqmsg->bufcount == 6 ||
+ req->rq_reqmsg->bufcount == 7);
req->rq_reqmsg->buflens[5] = body->eadatasize;
/* If this isn't the last buffer, we might
* have to shift other data around. */
}
}
+
+ /* just swab out capa to check here, and for future use */
+ if (body->valid & OBD_MD_CAPA) {
+ struct lustre_capa *capa;
+
+ LASSERT(it->it_op & IT_OPEN);
+ capa = lustre_swab_repbuf(req, 7, sizeof(*capa),
+ lustre_swab_lustre_capa);
+ LASSERT(capa);
+ }
}
RETURN(rc);
* ll_create/ll_open gets called.
*
* The server will return to us, in it_disposition, an indication of
- * exactly what d.lustre.it_status refers to.
+ * exactly what d.lustre->it_status refers to.
*
- * If DISP_OPEN_OPEN is set, then d.lustre.it_status refers to the open() call,
+ * If DISP_OPEN_OPEN is set, then d.lustre->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 d.lustre.it_status is the rc of the
+ * Else, if DISP_LOOKUP_EXECD then d.lustre->it_status is the rc of the
* child lookup.
*/
int mdc_intent_lock(struct obd_export *exp, struct lustre_id *pid,
id_group(cid)}};
struct lustre_handle lockh;
ldlm_policy_data_t policy;
- int mode = LCK_PR;
+ int mode;
/* For the GETATTR case, ll_revalidate_it issues two separate
queries - for LOOKUP and for UPDATE lock because it cannot
check them together - we might have those two bits to be
present in two separate granted locks */
policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
- MDS_INODELOCK_UPDATE: MDS_INODELOCK_LOOKUP;
+ MDS_INODELOCK_UPDATE : MDS_INODELOCK_LOOKUP;
mode = LCK_PR;
rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
LDLM_FL_BLOCK_GRANTED, &res_id,
- LDLM_IBITS, &policy, LCK_PR, &lockh);
+ LDLM_IBITS, &policy, mode,
+ &lockh);
+
+ if (!rc) {
+ mode = LCK_CR;
+ rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
+ LDLM_FL_BLOCK_GRANTED, &res_id,
+ LDLM_IBITS, &policy, mode,
+ &lockh);
+ }
if (!rc) {
mode = LCK_PW;
rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
LDLM_FL_BLOCK_GRANTED, &res_id,
- LDLM_IBITS, &policy, LCK_PW,
+ LDLM_IBITS, &policy, mode,
+ &lockh);
+ }
+ if (!rc) {
+ mode = LCK_CW;
+ rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
+ LDLM_FL_BLOCK_GRANTED, &res_id,
+ LDLM_IBITS, &policy, mode,
&lockh);
}
if (rc) {
- memcpy(&it->d.lustre.it_lock_handle, &lockh,
+ if (ptlrpcs_check_cred(exp->exp_obd->u.cli.cl_import)) {
+ /* return immediately if no credential held */
+ ldlm_lock_decref(&lockh, mode);
+ RETURN(-EACCES);
+ }
+ memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh,
sizeof(lockh));
- it->d.lustre.it_lock_mode = mode;
+ LUSTRE_IT(it)->it_lock_mode = mode;
}
- /* Only return failure if it was not GETATTR by cfid
- (from inode_revalidate) */
+ /* Only return failure if it was not GETATTR by cid (from
+ inode_revalidate) */
if (rc || name)
RETURN(rc);
}
* this and use the request from revalidate. In this case, revalidate
* never dropped its reference, so the refcounts are all OK */
if (!it_disposition(it, DISP_ENQ_COMPLETE)) {
- struct mdc_op_data op_data;
+ struct mdc_op_data *op_data;
+
+ OBD_ALLOC(op_data, sizeof(*op_data));
+ if (op_data == NULL)
+ RETURN(-ENOMEM);
- mdc_id2mdc_op_data(&op_data, pid, cid, name, len, 0);
+ mdc_id2mdc_data(op_data, pid, cid, name, len, 0);
- /*
- * if we getting inode by name (ll_lookup_it() case), we always
- * should ask for fid, as later we will not be able to take
- * locks, revalidate dentry, etc. with invalid fid in inode.
- */
- if (cid == NULL && name != NULL)
- op_data.valid |= OBD_MD_FID;
+ if (name != NULL)
+ op_data->valid |= OBD_MD_FID;
rc = mdc_enqueue(exp, LDLM_IBITS, it, it_to_lock_mode(it),
- &op_data, &lockh, lmm, lmmsize,
+ op_data, &lockh, lmm, lmmsize,
ldlm_completion_ast, cb_blocking, NULL);
+ OBD_FREE(op_data, sizeof(*op_data));
if (rc < 0)
RETURN(rc);
- memcpy(&it->d.lustre.it_lock_handle, &lockh, sizeof(lockh));
+ memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh, sizeof(lockh));
}
- request = *reqp = it->d.lustre.it_data;
+ request = *reqp = LUSTRE_IT(it)->it_data;
LASSERT(request != NULL);
/* 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.
+ * successful open, then we need to remove the bit which saves this
+ * request for unconditional replay.
*
* It's important that we do this first! Otherwise we might exit the
- * function without doing so, and try to replay a failed create
- * (bug 3440) */
- if (it->it_op & IT_OPEN) {
- if (!it_disposition(it, DISP_OPEN_OPEN) ||
- it->d.lustre.it_status != 0) {
- unsigned long irqflags;
-
- spin_lock_irqsave(&request->rq_lock, irqflags);
- request->rq_replay = 0;
- spin_unlock_irqrestore(&request->rq_lock, irqflags);
- }
- }
+ * function without doing so, and try to replay a failed create (bug
+ * 3440) */
+ if (it->it_op & IT_OPEN && request->rq_replay &&
+ (!it_disposition(it, DISP_OPEN_OPEN) || LUSTRE_IT(it)->it_status != 0))
+ mdc_clear_replay_flag(request, LUSTRE_IT(it)->it_status);
+
if (!it_disposition(it, DISP_IT_EXECD)) {
/* The server failed before it even started executing the
* intent, i.e. because it couldn't unpack the request. */
- LASSERT(it->d.lustre.it_status != 0);
- RETURN(it->d.lustre.it_status);
+ LASSERT(LUSTRE_IT(it)->it_status != 0);
+ RETURN(LUSTRE_IT(it)->it_status);
}
rc = it_open_error(DISP_IT_EXECD, it);
if (rc)
/* we have to compare all the fields but type, because MDS can
* return fid/mds/ino/gen if inode lives on another MDS -bzzz */
- if (id_gen(cid) != id_gen(&mds_body->id1) ||
- id_ino(cid) != id_ino(&mds_body->id1) ||
- id_fid(cid) != id_fid(&mds_body->id1) ||
- id_group(cid) != id_group(&mds_body->id1))
+ if (!(lookup_flags & LOOKUP_COBD) && !id_equal(cid, &mds_body->id1))
RETURN(-ESTALE);
}
if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
LDLM_IBITS, &policy, LCK_NL, &old_lock)) {
ldlm_lock_decref_and_cancel(&lockh,
- it->d.lustre.it_lock_mode);
+ LUSTRE_IT(it)->it_lock_mode);
memcpy(&lockh, &old_lock, sizeof(old_lock));
- memcpy(&it->d.lustre.it_lock_handle, &lockh,
+ memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh,
sizeof(lockh));
}
}
CDEBUG(D_DENTRY, "D_IT dentry %*s intent: %s status %d disp %x rc %d\n",
- len, name, ldlm_it2str(it->it_op), it->d.lustre.it_status,
- it->d.lustre.it_disposition, rc);
+ len, name, ldlm_it2str(it->it_op), LUSTRE_IT(it)->it_status,
+ LUSTRE_IT(it)->it_disposition, rc);
RETURN(rc);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff