* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2016, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <lustre_net.h>
#include <lustre_req_layout.h>
#include <lustre_swab.h>
+#include <lustre_acl.h>
#include "mdc_internal.h"
static struct ptlrpc_request *
mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
- struct md_op_data *op_data)
+ struct md_op_data *op_data, __u32 acl_bufsize)
{
struct ptlrpc_request *req;
struct obd_device *obddev = class_exp2obd(exp);
int count = 0;
enum ldlm_mode mode;
int rc;
+ int repsize, repsize_estimate;
+
ENTRY;
it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
/* If inode is known, cancel conflicting OPEN locks. */
if (fid_is_sane(&op_data->op_fid2)) {
if (it->it_flags & MDS_OPEN_LEASE) { /* try to get lease */
- if (it->it_flags & FMODE_WRITE)
+ if (it->it_flags & MDS_FMODE_WRITE)
mode = LCK_EX;
else
mode = LCK_PR;
} else {
- if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
+ if (it->it_flags & (MDS_FMODE_WRITE | MDS_OPEN_TRUNC))
mode = LCK_CW;
#ifdef FMODE_EXEC
else if (it->it_flags & FMODE_EXEC)
req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
op_data->op_namelen + 1);
- req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
+ if (cl_is_lov_delay_create(it->it_flags)) {
+ /* open(O_LOV_DELAY_CREATE) won't pack lmm */
+ LASSERT(lmmsize == 0);
+ req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
+ } else {
+ req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
max(lmmsize, obddev->u.cli.cl_default_mds_easize));
+ }
req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
RCL_CLIENT, op_data->op_file_secctx_name != NULL ?
req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
obddev->u.cli.cl_max_mds_easize);
- ptlrpc_request_set_replen(req);
- return req;
+ req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
+
+ /**
+ * Inline buffer for possible data from Data-on-MDT files.
+ */
+ req_capsule_set_size(&req->rq_pill, &RMF_NIOBUF_INLINE, RCL_SERVER,
+ sizeof(struct niobuf_remote));
+ ptlrpc_request_set_replen(req);
+
+ /* Get real repbuf allocated size as rounded up power of 2 */
+ repsize = size_roundup_power2(req->rq_replen +
+ lustre_msg_early_size());
+ /* Estimate free space for DoM files in repbuf */
+ repsize_estimate = repsize - (req->rq_replen -
+ obddev->u.cli.cl_max_mds_easize +
+ sizeof(struct lov_comp_md_v1) +
+ sizeof(struct lov_comp_md_entry_v1) +
+ lov_mds_md_size(0, LOV_MAGIC_V3));
+
+ if (repsize_estimate < obddev->u.cli.cl_dom_min_inline_repsize) {
+ repsize = obddev->u.cli.cl_dom_min_inline_repsize -
+ repsize_estimate + sizeof(struct niobuf_remote);
+ req_capsule_set_size(&req->rq_pill, &RMF_NIOBUF_INLINE,
+ RCL_SERVER,
+ sizeof(struct niobuf_remote) + repsize);
+ ptlrpc_request_set_replen(req);
+ CDEBUG(D_INFO, "Increase repbuf by %d bytes, total: %d\n",
+ repsize, req->rq_replen);
+ repsize = size_roundup_power2(req->rq_replen +
+ lustre_msg_early_size());
+ }
+ /* The only way to report real allocated repbuf size to the server
+ * is the lm_repsize but it must be set prior buffer allocation itself
+ * due to security reasons - it is part of buffer used in signature
+ * calculation (see LU-11414). Therefore the saved size is predicted
+ * value as rq_replen rounded to the next higher power of 2.
+ * Such estimation is safe. Though the final allocated buffer might
+ * be even larger, it is not possible to know that at this point.
+ */
+ req->rq_reqmsg->lm_repsize = repsize;
+ RETURN(req);
}
+#define GA_DEFAULT_EA_NAME_LEN 20
+#define GA_DEFAULT_EA_VAL_LEN 250
+#define GA_DEFAULT_EA_NUM 10
+
static struct ptlrpc_request *
mdc_intent_getxattr_pack(struct obd_export *exp,
struct lookup_intent *it,
struct ptlrpc_request *req;
struct ldlm_intent *lit;
int rc, count = 0;
- __u32 maxdata;
struct list_head cancels = LIST_HEAD_INIT(cancels);
+ u32 ea_vals_buf_size = GA_DEFAULT_EA_VAL_LEN * GA_DEFAULT_EA_NUM;
ENTRY;
lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
lit->opc = IT_GETXATTR;
- maxdata = class_exp2cliimp(exp)->imp_connect_data.ocd_max_easize;
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
+ /* If the supplied buffer is too small then the server will
+ * return -ERANGE and llite will fallback to using non cached
+ * xattr operations. On servers before 2.10.1 a (non-cached)
+ * listxattr RPC for an orphan or dead file causes an oops. So
+ * let's try to avoid sending too small a buffer to too old a
+ * server. This is effectively undoing the memory conservation
+ * of LU-9417 when it would be *more* likely to crash the
+ * server. See LU-9856. */
+ if (exp->exp_connect_data.ocd_version < OBD_OCD_VERSION(2, 10, 1, 0))
+ ea_vals_buf_size = max_t(u32, ea_vals_buf_size,
+ exp->exp_connect_data.ocd_max_easize);
+#endif
/* pack the intended request */
- mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid, maxdata, -1,
- 0);
-
- req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
- RCL_SERVER, maxdata);
+ mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
+ ea_vals_buf_size, -1, 0);
- req_capsule_set_size(&req->rq_pill, &RMF_EAVALS,
- RCL_SERVER, maxdata);
+ req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_SERVER,
+ GA_DEFAULT_EA_NAME_LEN * GA_DEFAULT_EA_NUM);
- req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS,
- RCL_SERVER, maxdata);
+ req_capsule_set_size(&req->rq_pill, &RMF_EAVALS, RCL_SERVER,
+ ea_vals_buf_size);
- ptlrpc_request_set_replen(req);
-
- RETURN(req);
-}
+ req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS, RCL_SERVER,
+ sizeof(u32) * GA_DEFAULT_EA_NUM);
-static struct ptlrpc_request *mdc_intent_unlink_pack(struct obd_export *exp,
- struct lookup_intent *it,
- struct md_op_data *op_data)
-{
- struct ptlrpc_request *req;
- struct obd_device *obddev = class_exp2obd(exp);
- struct ldlm_intent *lit;
- int rc;
- ENTRY;
+ req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
- req = ptlrpc_request_alloc(class_exp2cliimp(exp),
- &RQF_LDLM_INTENT_UNLINK);
- if (req == NULL)
- RETURN(ERR_PTR(-ENOMEM));
-
- req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
- op_data->op_namelen + 1);
-
- rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
- if (rc) {
- ptlrpc_request_free(req);
- RETURN(ERR_PTR(rc));
- }
-
- /* pack the intent */
- lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
- lit->opc = (__u64)it->it_op;
-
- /* pack the intended request */
- mdc_unlink_pack(req, op_data);
-
- req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
- obddev->u.cli.cl_default_mds_easize);
ptlrpc_request_set_replen(req);
+
RETURN(req);
}
-static struct ptlrpc_request *mdc_intent_getattr_pack(struct obd_export *exp,
- struct lookup_intent *it,
- struct md_op_data *op_data)
+static struct ptlrpc_request *
+mdc_intent_getattr_pack(struct obd_export *exp, struct lookup_intent *it,
+ struct md_op_data *op_data, __u32 acl_bufsize)
{
struct ptlrpc_request *req;
struct obd_device *obddev = class_exp2obd(exp);
mdc_getattr_pack(req, valid, it->it_flags, op_data, easize);
req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, easize);
+ req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
ptlrpc_request_set_replen(req);
RETURN(req);
}
struct ldlm_request *lockreq;
struct ldlm_reply *lockrep;
struct ldlm_lock *lock;
+ struct mdt_body *body = NULL;
void *lvb_data = NULL;
__u32 lvb_len = 0;
+
ENTRY;
LASSERT(rc >= 0);
* 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 && req->rq_replay &&
+ if (it->it_op & IT_OPEN && req->rq_replay &&
(!it_disposition(it, DISP_OPEN_OPEN) || it->it_status != 0))
mdc_clear_replay_flag(req, it->it_status);
- DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
+ DEBUG_REQ(D_RPCTRACE, req, "op: %x disposition: %x, status: %d",
it->it_op, it->it_disposition, it->it_status);
- /* We know what to expect, so we do any byte flipping required here */
+ /* We know what to expect, so we do any byte flipping required here */
if (it_has_reply_body(it)) {
- struct mdt_body *body;
-
body = req_capsule_server_get(pill, &RMF_MDT_BODY);
if (body == NULL) {
CERROR ("Can't swab mdt_body\n");
/* maybe the lock was granted right away and layout
* is packed into RMF_DLM_LVB of req */
lvb_len = req_capsule_get_size(pill, &RMF_DLM_LVB, RCL_SERVER);
+ CDEBUG(D_INFO, "%s: layout return lvb %d transno %lld\n",
+ class_exp2obd(exp)->obd_name, lvb_len, req->rq_transno);
if (lvb_len > 0) {
lvb_data = req_capsule_server_sized_get(pill,
&RMF_DLM_LVB, lvb_len);
* client still does this checking in case it's talking with an old
* server. - Jinshan */
lock = ldlm_handle2lock(lockh);
- if (lock != NULL && ldlm_has_layout(lock) && lvb_data != NULL &&
+ if (lock == NULL)
+ RETURN(rc);
+
+ if (ldlm_has_layout(lock) && lvb_data != NULL &&
!(lockrep->lock_flags & LDLM_FL_BLOCKED_MASK)) {
void *lmm;
ldlm_it2str(it->it_op), lvb_len);
OBD_ALLOC_LARGE(lmm, lvb_len);
- if (lmm == NULL) {
- LDLM_LOCK_PUT(lock);
- RETURN(-ENOMEM);
- }
+ if (lmm == NULL)
+ GOTO(out_lock, rc = -ENOMEM);
+
memcpy(lmm, lvb_data, lvb_len);
/* install lvb_data */
if (lmm != NULL)
OBD_FREE_LARGE(lmm, lvb_len);
}
- if (lock != NULL)
- LDLM_LOCK_PUT(lock);
+
+ if (ldlm_has_dom(lock)) {
+ LASSERT(lock->l_glimpse_ast == mdc_ldlm_glimpse_ast);
+
+ body = req_capsule_server_get(pill, &RMF_MDT_BODY);
+ if (!(body->mbo_valid & OBD_MD_DOM_SIZE)) {
+ LDLM_ERROR(lock, "%s: DoM lock without size.\n",
+ exp->exp_obd->obd_name);
+ GOTO(out_lock, rc = -EPROTO);
+ }
+
+ LDLM_DEBUG(lock, "DoM lock is returned by: %s, size: %llu",
+ ldlm_it2str(it->it_op), body->mbo_dom_size);
+
+ rc = mdc_fill_lvb(req, &lock->l_ost_lvb);
+ }
+out_lock:
+ LDLM_LOCK_PUT(lock);
RETURN(rc);
}
.l_inodebits = { MDS_INODELOCK_XATTR } };
int generation, resends = 0;
struct ldlm_reply *lockrep;
+ struct obd_import *imp = class_exp2cliimp(exp);
+ __u32 acl_bufsize;
enum lvb_type lvb_type = 0;
int rc;
ENTRY;
LASSERT(policy == NULL);
saved_flags |= LDLM_FL_HAS_INTENT;
- if (it->it_op & (IT_UNLINK | IT_GETATTR | IT_READDIR))
+ if (it->it_op & (IT_GETATTR | IT_READDIR))
policy = &update_policy;
else if (it->it_op & IT_LAYOUT)
policy = &layout_policy;
- else if (it->it_op & (IT_GETXATTR | IT_SETXATTR))
+ else if (it->it_op & IT_GETXATTR)
policy = &getxattr_policy;
else
policy = &lookup_policy;
}
- generation = obddev->u.cli.cl_import->imp_generation;
+ generation = obddev->u.cli.cl_import->imp_generation;
+ if (!it || (it->it_op & (IT_OPEN | IT_CREAT)))
+ acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
+ XATTR_SIZE_MAX);
+ else
+ acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
+
resend:
- flags = saved_flags;
+ flags = saved_flags;
if (it == NULL) {
/* The only way right now is FLOCK. */
LASSERTF(einfo->ei_type == LDLM_FLOCK, "lock type %d\n",
einfo->ei_type);
res_id.name[3] = LDLM_FLOCK;
} else if (it->it_op & IT_OPEN) {
- req = mdc_intent_open_pack(exp, it, op_data);
- } else if (it->it_op & IT_UNLINK) {
- req = mdc_intent_unlink_pack(exp, it, op_data);
+ req = mdc_intent_open_pack(exp, it, op_data, acl_bufsize);
} else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
- req = mdc_intent_getattr_pack(exp, it, op_data);
+ req = mdc_intent_getattr_pack(exp, it, op_data, acl_bufsize);
} else if (it->it_op & IT_READDIR) {
req = mdc_enqueue_pack(exp, 0);
} else if (it->it_op & IT_LAYOUT) {
- if (!imp_connect_lvb_type(class_exp2cliimp(exp)))
+ if (!imp_connect_lvb_type(imp))
RETURN(-EOPNOTSUPP);
req = mdc_intent_layout_pack(exp, it, op_data);
lvb_type = LVB_T_LAYOUT;
rc = obd_get_request_slot(&obddev->u.cli);
if (rc != 0) {
mdc_put_mod_rpc_slot(req, it);
- mdc_clear_replay_flag(req, 0);
- ptlrpc_req_finished(req);
- RETURN(rc);
- }
- }
+ mdc_clear_replay_flag(req, 0);
+ ptlrpc_req_finished(req);
+ RETURN(rc);
+ }
+ }
+
+ /* With Data-on-MDT the glimpse callback is needed too.
+ * It is set here in advance but not in mdc_finish_enqueue()
+ * to avoid possible races. It is safe to have glimpse handler
+ * for non-DOM locks and costs nothing.*/
+ if (einfo->ei_cb_gl == NULL)
+ einfo->ei_cb_gl = mdc_ldlm_glimpse_ast;
- rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
+ rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
0, lvb_type, lockh, 0);
- if (!it) {
- /* For flock requests we immediatelly return without further
- delay and let caller deal with the rest, since rest of
- this function metadata processing makes no sense for flock
+ if (!it) {
+ /* For flock requests we immediatelly return without further
+ delay and let caller deal with the rest, since rest of
+ this function metadata processing makes no sense for flock
requests anyway. But in case of problem during comms with
Server (ETIMEDOUT) or any signal/kill attempt (EINTR), we
can not rely on caller and this mainly for F_UNLCKs
mdc_put_mod_rpc_slot(req, it);
if (rc < 0) {
- CDEBUG(D_INFO, "%s: ldlm_cli_enqueue failed: rc = %d\n",
- obddev->obd_name, rc);
+ CDEBUG(D_INFO,
+ "%s: ldlm_cli_enqueue "DFID":"DFID"=%s failed: rc = %d\n",
+ obddev->obd_name, PFID(&op_data->op_fid1),
+ PFID(&op_data->op_fid2), op_data->op_name ?: "", rc);
mdc_clear_replay_flag(req, rc);
ptlrpc_req_finished(req);
* intent operation, when server returns -EINPROGRESS for acquiring
* intent lock, we'll retry in after_reply(). */
if (it && (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
- mdc_clear_replay_flag(req, rc);
- ptlrpc_req_finished(req);
- resends++;
-
- CDEBUG(D_HA, "%s: resend:%d op:%d "DFID"/"DFID"\n",
- obddev->obd_name, resends, it->it_op,
- PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
-
- if (generation == obddev->u.cli.cl_import->imp_generation) {
- goto resend;
- } else {
+ mdc_clear_replay_flag(req, rc);
+ ptlrpc_req_finished(req);
+ if (generation == obddev->u.cli.cl_import->imp_generation) {
+ if (signal_pending(current))
+ RETURN(-EINTR);
+
+ resends++;
+ CDEBUG(D_HA, "%s: resend:%d op:%d "DFID"/"DFID"\n",
+ obddev->obd_name, resends, it->it_op,
+ PFID(&op_data->op_fid1),
+ PFID(&op_data->op_fid2));
+ goto resend;
+ } else {
CDEBUG(D_HA, "resend cross eviction\n");
- RETURN(-EIO);
- }
- }
+ RETURN(-EIO);
+ }
+ }
+
+ if ((int)lockrep->lock_policy_res2 == -ERANGE &&
+ it->it_op & (IT_OPEN | IT_GETATTR | IT_LOOKUP) &&
+ acl_bufsize == LUSTRE_POSIX_ACL_MAX_SIZE_OLD) {
+ mdc_clear_replay_flag(req, -ERANGE);
+ ptlrpc_req_finished(req);
+ acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
+ XATTR_SIZE_MAX);
+ goto resend;
+ }
rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
if (rc < 0) {
.ei_mode = it_to_lock_mode(it),
.ei_cb_bl = cb_blocking,
.ei_cb_cp = ldlm_completion_ast,
+ .ei_cb_gl = mdc_ldlm_glimpse_ast,
};
struct lustre_handle lockh;
int rc = 0;
}
static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
- struct ptlrpc_request *req,
- void *args, int rc)
+ struct ptlrpc_request *req,
+ void *args, int rc)
{
struct mdc_getattr_args *ga = args;
- struct obd_export *exp = ga->ga_exp;
- struct md_enqueue_info *minfo = ga->ga_minfo;
+ struct obd_export *exp = ga->ga_exp;
+ struct md_enqueue_info *minfo = ga->ga_minfo;
struct ldlm_enqueue_info *einfo = &minfo->mi_einfo;
- struct lookup_intent *it;
- struct lustre_handle *lockh;
- struct obd_device *obddev;
- struct ldlm_reply *lockrep;
- __u64 flags = LDLM_FL_HAS_INTENT;
+ struct lookup_intent *it;
+ struct lustre_handle *lockh;
+ struct obd_device *obddev;
+ struct ldlm_reply *lockrep;
+ __u64 flags = LDLM_FL_HAS_INTENT;
ENTRY;
it = &minfo->mi_it;
PFID(&op_data->op_fid1), ldlm_it2str(it->it_op), it->it_flags);
fid_build_reg_res_name(&op_data->op_fid1, &res_id);
- req = mdc_intent_getattr_pack(exp, it, op_data);
+ /* If the MDT return -ERANGE because of large ACL, then the sponsor
+ * of the async getattr RPC will handle that by itself. */
+ req = mdc_intent_getattr_pack(exp, it, op_data,
+ LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
if (IS_ERR(req))
RETURN(PTR_ERR(req));
RETURN(rc);
}
+ /* With Data-on-MDT the glimpse callback is needed too.
+ * It is set here in advance but not in mdc_finish_enqueue()
+ * to avoid possible races. It is safe to have glimpse handler
+ * for non-DOM locks and costs nothing.*/
+ if (minfo->mi_einfo.ei_cb_gl == NULL)
+ minfo->mi_einfo.ei_cb_gl = mdc_ldlm_glimpse_ast;
+
rc = ldlm_cli_enqueue(exp, &req, &minfo->mi_einfo, &res_id, &policy,
&flags, NULL, 0, LVB_T_NONE, &minfo->mi_lockh, 1);
if (rc < 0) {
git://git.whamcloud.com / fs / lustre-release.git / blobdiff