#include <lprocfs_status.h>
#include <lustre_disk.h>
#include <lustre_param.h>
+#include <lustre_log.h>
+#include <obd_cksum.h>
#include "llite_internal.h"
-kmem_cache_t *ll_file_data_slab;
+cfs_mem_cache_t *ll_file_data_slab;
LIST_HEAD(ll_super_blocks);
spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
spin_lock_init(&sbi->ll_lock);
spin_lock_init(&sbi->ll_lco.lco_lock);
+ spin_lock_init(&sbi->ll_pp_extent_lock);
+ spin_lock_init(&sbi->ll_process_lock);
+ sbi->ll_rw_stats_on = 0;
INIT_LIST_HEAD(&sbi->ll_pglist);
if (num_physpages >> (20 - CFS_PAGE_SHIFT) < 512)
sbi->ll_async_page_max = num_physpages / 2;
SBI_DEFAULT_READAHEAD_MAX);
sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
SBI_DEFAULT_READAHEAD_WHOLE_MAX;
-
+ sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
+ sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
INIT_LIST_HEAD(&sbi->ll_conn_chain);
INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
- class_generate_random_uuid(uuid);
+ ll_generate_random_uuid(uuid);
class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
- CDEBUG(D_HA, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
+ CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
spin_lock(&ll_sb_lock);
list_add_tail(&sbi->ll_list, &ll_super_blocks);
spin_unlock(&ll_sb_lock);
+#ifdef ENABLE_LLITE_CHECKSUM
+ sbi->ll_flags |= LL_SBI_CHECKSUM;
+#endif
+
+#ifdef HAVE_LRU_RESIZE_SUPPORT
+ sbi->ll_flags |= LL_SBI_LRU_RESIZE;
+#endif
+
+#ifdef HAVE_EXPORT___IGET
INIT_LIST_HEAD(&sbi->ll_deathrow);
spin_lock_init(&sbi->ll_deathrow_lock);
- for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
+#endif
+ for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
}
+ /* metadata statahead is enabled by default */
+ sbi->ll_sa_max = LL_SA_RPC_DEF;
+
RETURN(sbi);
}
}
static struct dentry_operations ll_d_root_ops = {
-#ifdef LUSTRE_KERNEL_VERSION
+#ifdef DCACHE_LUSTRE_INVALID
.d_compare = ll_dcompare,
#endif
};
-static int client_common_fill_super(struct super_block *sb,
- char *mdc, char *osc)
+/* Initialize the default and maximum LOV EA and cookie sizes. This allows
+ * us to make MDS RPCs with large enough reply buffers to hold the
+ * maximum-sized (= maximum striped) EA and cookie without having to
+ * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
+static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
+{
+ struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
+ __u32 valsize = sizeof(struct lov_desc);
+ int rc, easize, def_easize, cookiesize;
+ struct lov_desc desc;
+ __u32 stripes;
+ ENTRY;
+
+ rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
+ &valsize, &desc);
+ if (rc)
+ RETURN(rc);
+
+ stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
+ lsm.lsm_stripe_count = stripes;
+ easize = obd_size_diskmd(dt_exp, &lsm);
+
+ lsm.lsm_stripe_count = desc.ld_default_stripe_count;
+ def_easize = obd_size_diskmd(dt_exp, &lsm);
+
+ cookiesize = stripes * sizeof(struct llog_cookie);
+
+ CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
+ easize, cookiesize);
+
+ rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
+ RETURN(rc);
+}
+
+static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
{
struct inode *root = 0;
struct ll_sb_info *sbi = ll_s2sbi(sb);
struct obd_device *obd;
- struct ll_fid rootfid;
+ struct lu_fid rootfid;
+ struct obd_capa *oc = NULL;
struct obd_statfs osfs;
struct ptlrpc_request *request = NULL;
- struct lustre_handle osc_conn = {0, };
- struct lustre_handle mdc_conn = {0, };
- struct lustre_md md;
+ struct lustre_handle dt_conn = {0, };
+ struct lustre_handle md_conn = {0, };
struct obd_connect_data *data = NULL;
- int err;
+ struct lustre_md lmd;
+ obd_valid valid;
+ int size, err, checksum;
ENTRY;
- obd = class_name2obd(mdc);
+ obd = class_name2obd(md);
if (!obd) {
- CERROR("MDC %s: not setup or attached\n", mdc);
+ CERROR("MD %s: not setup or attached\n", md);
RETURN(-EINVAL);
}
- OBD_ALLOC(data, sizeof(*data));
+ OBD_ALLOC_PTR(data);
if (data == NULL)
RETURN(-ENOMEM);
if (proc_lustre_fs_root) {
err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
- osc, mdc);
+ dt, md);
if (err < 0)
- CERROR("could not register mount in /proc/lustre");
+ CERROR("could not register mount in /proc/fs/lustre\n");
}
/* indicate the features supported by this client */
- data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
- OBD_CONNECT_JOIN |
- OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION;
+ data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
+ OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
+ OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
+ OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
+ OBD_CONNECT_FID;
+
+#ifdef HAVE_LRU_RESIZE_SUPPORT
+ if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
+ data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
+#endif
#ifdef CONFIG_FS_POSIX_ACL
data->ocd_connect_flags |= OBD_CONNECT_ACL;
#endif
if (sbi->ll_flags & LL_SBI_FLOCK)
sbi->ll_fop = &ll_file_operations_flock;
- else
+ else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
sbi->ll_fop = &ll_file_operations;
+ else
+ sbi->ll_fop = &ll_file_operations_noflock;
- err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, data);
+ /* real client */
+ data->ocd_connect_flags |= OBD_CONNECT_REAL;
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
+ data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
+ data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
+ } else {
+ data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
+ data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
+ }
+
+ err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
if (err == -EBUSY) {
- LCONSOLE_ERROR("An MDT (mdc %s) is performing recovery, of "
- "which this client is not a part. Please wait "
- "for recovery to complete, abort, or "
- "time out.\n", mdc);
+ LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
+ "recovery, of which this client is not a "
+ "part. Please wait for recovery to complete,"
+ " abort, or time out.\n", md);
GOTO(out, err);
} else if (err) {
- CERROR("cannot connect to %s: rc = %d\n", mdc, err);
+ CERROR("cannot connect to %s: rc = %d\n", md, err);
GOTO(out, err);
}
- sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
+ sbi->ll_md_exp = class_conn2export(&md_conn);
- err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
+ err = obd_fid_init(sbi->ll_md_exp);
+ if (err) {
+ CERROR("Can't init metadata layer FID infrastructure, "
+ "rc %d\n", err);
+ GOTO(out_md, err);
+ }
+
+ err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
if (err)
- GOTO(out_mdc, err);
+ GOTO(out_md_fid, err);
- /* MDC connect is surely finished by now because we actually sent
- * a statfs RPC, otherwise obd_connect() is asynchronous. */
- *data = class_exp2cliimp(sbi->ll_mdc_exp)->imp_connect_data;
+ size = sizeof(*data);
+ err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
+ KEY_CONN_DATA, &size, data);
+ if (err) {
+ CERROR("Get connect data failed: %d \n", err);
+ GOTO(out_md, err);
+ }
LASSERT(osfs.os_bsize);
sb->s_blocksize = osfs.os_bsize;
sb->s_flags |= MS_POSIXACL;
#endif
sbi->ll_flags |= LL_SBI_ACL;
- } else
+ } else {
+ LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
+#ifdef MS_POSIXACL
+ sb->s_flags &= ~MS_POSIXACL;
+#endif
sbi->ll_flags &= ~LL_SBI_ACL;
+ }
if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
sbi->ll_flags |= LL_SBI_JOIN;
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
+ if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
+ /* sometimes local client claims to be remote, but mdt
+ * will disagree when client gss not applied. */
+ LCONSOLE_INFO("client claims to be remote, but server "
+ "rejected, forced to be local.\n");
+ sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
+ }
+ } else {
+ if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
+ /* with gss applied, remote client can not claim to be
+ * local, so mdt maybe force client to be remote. */
+ LCONSOLE_INFO("client claims to be local, but server "
+ "rejected, forced to be remote.\n");
+ sbi->ll_flags |= LL_SBI_RMT_CLIENT;
+ }
+ }
+
+ if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
+ LCONSOLE_INFO("client enabled MDS capability!\n");
+ sbi->ll_flags |= LL_SBI_MDS_CAPA;
+ }
+
+ if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
+ LCONSOLE_INFO("client enabled OSS capability!\n");
+ sbi->ll_flags |= LL_SBI_OSS_CAPA;
+ }
+
+ sbi->ll_sdev_orig = sb->s_dev;
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
/* We set sb->s_dev equal on all lustre clients in order to support
* NFS export clustering. NFSD requires that the FSID be the same
* on all clients. */
/* s_dev is also used in lt_compare() to compare two fs, but that is
* only a node-local comparison. */
+
+ /* XXX: this will not work with LMV */
sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
#endif
- obd = class_name2obd(osc);
+ obd = class_name2obd(dt);
if (!obd) {
- CERROR("OSC %s: not setup or attached\n", osc);
- GOTO(out_mdc, err = -ENODEV);
+ CERROR("DT %s: not setup or attached\n", dt);
+ GOTO(out_md_fid, err = -ENODEV);
+ }
+
+ data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
+ OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
+ OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
+ OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK;
+ if (sbi->ll_flags & LL_SBI_OSS_CAPA)
+ data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
+
+ if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
+ /* OBD_CONNECT_CKSUM should always be set, even if checksums are
+ * disabled by default, because it can still be enabled on the
+ * fly via /proc. As a consequence, we still need to come to an
+ * agreement on the supported algorithms at connect time */
+ data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
+ data->ocd_cksum_types = OBD_CKSUM_ADLER;
+ else
+ /* send the list of supported checksum types */
+ data->ocd_cksum_types = OBD_CKSUM_ALL;
}
- data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
- OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE;
-
+#ifdef HAVE_LRU_RESIZE_SUPPORT
+ data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
+#endif
CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
"ocd_grant: %d\n", data->ocd_connect_flags,
data->ocd_version, data->ocd_grant);
obd->obd_upcall.onu_upcall = ll_ocd_update;
data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
-
- err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data);
- if (err == -EBUSY) {
- LCONSOLE_ERROR("An OST (osc %s) is performing recovery, of "
- "which this client is not a part. Please wait "
- "for recovery to complete, abort, or "
- "time out.\n", osc);
- GOTO(out, err);
+ err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
+ if (err == -EBUSY) {
+ LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
+ "recovery, of which this client is not a "
+ "part. Please wait for recovery to "
+ "complete, abort, or time out.\n", dt);
+ GOTO(out_md_fid, err);
} else if (err) {
- CERROR("cannot connect to %s: rc = %d\n", osc, err);
- GOTO(out_mdc, err);
+ CERROR("Cannot connect to %s: rc = %d\n", dt, err);
+ GOTO(out_md_fid, err);
}
- sbi->ll_osc_exp = class_conn2export(&osc_conn);
+
+ sbi->ll_dt_exp = class_conn2export(&dt_conn);
+
+ err = obd_fid_init(sbi->ll_dt_exp);
+ if (err) {
+ CERROR("Can't init data layer FID infrastructure, "
+ "rc %d\n", err);
+ GOTO(out_dt, err);
+ }
+
spin_lock(&sbi->ll_lco.lco_lock);
sbi->ll_lco.lco_flags = data->ocd_connect_flags;
spin_unlock(&sbi->ll_lco.lco_lock);
- mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
+ ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
- err = obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
+ err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
0, NULL, NULL, NULL);
if (err < 0) {
- LCONSOLE_ERROR("There are no OST's in this filesystem. "
- "There must be at least one active OST for "
- "a client to start.\n");
- GOTO(out_osc, err);
+ LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
+ "filesystem. There must be at least one "
+ "active OST for a client to start.\n");
+ GOTO(out_dt_fid, err);
}
if (!ll_async_page_slab) {
ll_async_page_slab_size =
size_round(sizeof(struct ll_async_page)) + err;
- ll_async_page_slab = kmem_cache_create("ll_async_page",
- ll_async_page_slab_size,
- 0, 0, NULL, NULL);
+ ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
+ ll_async_page_slab_size,
+ 0, 0);
if (!ll_async_page_slab)
- GOTO(out_osc, -ENOMEM);
+ GOTO(out_dt_fid, err = -ENOMEM);
}
- err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
+ err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
if (err) {
CERROR("cannot mds_connect: rc = %d\n", err);
- GOTO(out_osc, err);
+ GOTO(out_dt_fid, err);
}
- CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
- sbi->ll_rootino = rootfid.id;
+ CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
+ sbi->ll_root_fid = rootfid;
sb->s_op = &lustre_super_operations;
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
sb->s_export_op = &lustre_export_operations;
-#endif
/* make root inode
* XXX: move this to after cbd setup? */
- err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
- OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
- (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
- 0, &request);
+ valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
+ valid |= OBD_MD_FLRMTPERM;
+ else if (sbi->ll_flags & LL_SBI_ACL)
+ valid |= OBD_MD_FLACL;
+
+ err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
+ if (oc)
+ free_capa(oc);
if (err) {
- CERROR("mdc_getattr failed for root: rc = %d\n", err);
- GOTO(out_osc, err);
+ CERROR("md_getattr failed for root: rc = %d\n", err);
+ GOTO(out_dt_fid, err);
}
-
- err = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
+ memset(&lmd, 0, sizeof(lmd));
+ err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
+ sbi->ll_md_exp, &lmd);
if (err) {
- CERROR("failed to understand root inode md: rc = %d\n",err);
+ CERROR("failed to understand root inode md: rc = %d\n", err);
ptlrpc_req_finished (request);
- GOTO(out_osc, err);
+ GOTO(out_dt_fid, err);
}
- LASSERT(sbi->ll_rootino != 0);
- root = ll_iget(sb, sbi->ll_rootino, &md);
-
+ LASSERT(fid_is_sane(&sbi->ll_root_fid));
+ root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
+ md_free_lustre_md(sbi->ll_md_exp, &lmd);
ptlrpc_req_finished(request);
if (root == NULL || is_bad_inode(root)) {
- mdc_free_lustre_md(sbi->ll_osc_exp, &md);
+ if (lmd.lsm)
+ obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
+#ifdef CONFIG_FS_POSIX_ACL
+ if (lmd.posix_acl) {
+ posix_acl_release(lmd.posix_acl);
+ lmd.posix_acl = NULL;
+ }
+#endif
CERROR("lustre_lite: bad iget4 for root\n");
GOTO(out_root, err = -EBADF);
}
GOTO(out_root, err);
}
- /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
- backing dev info assigned to inode mapping is used for
- determining maximal readahead. */
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
- !defined(KERNEL_HAS_AS_MAX_READAHEAD)
- /* bug 2805 - set VM readahead to zero */
- vm_max_readahead = vm_min_readahead = 0;
+#ifdef CONFIG_FS_POSIX_ACL
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
+ rct_init(&sbi->ll_rct);
+ et_init(&sbi->ll_et);
+ }
#endif
+ checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
+ err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
+ sizeof(checksum), &checksum, NULL);
+
sb->s_root = d_alloc_root(root);
if (data != NULL)
OBD_FREE(data, sizeof(*data));
sb->s_root->d_op = &ll_d_root_ops;
RETURN(err);
-
out_root:
if (root)
iput(root);
-out_osc:
- obd_disconnect(sbi->ll_osc_exp);
- sbi->ll_osc_exp = NULL;
-out_mdc:
- obd_disconnect(sbi->ll_mdc_exp);
- sbi->ll_mdc_exp = NULL;
+out_dt_fid:
+ obd_fid_fini(sbi->ll_dt_exp);
+out_dt:
+ obd_disconnect(sbi->ll_dt_exp);
+ sbi->ll_dt_exp = NULL;
+out_md_fid:
+ obd_fid_fini(sbi->ll_md_exp);
+out_md:
+ obd_disconnect(sbi->ll_md_exp);
+ sbi->ll_md_exp = NULL;
out:
if (data != NULL)
- OBD_FREE(data, sizeof(*data));
+ OBD_FREE_PTR(data);
lprocfs_unregister_mountpoint(sbi);
- RETURN(err);
+ return err;
}
int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
{
int size, rc;
- *lmmsize = obd_size_diskmd(sbi->ll_osc_exp, NULL);
+ *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
size = sizeof(int);
- rc = obd_get_info(sbi->ll_mdc_exp, strlen("max_easize"), "max_easize",
+ rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
&size, lmmsize);
if (rc)
CERROR("Get max mdsize error rc %d \n", rc);
dentry_count++;
CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
- inode, ll_i2mdcexp(inode)->exp_obd->obd_name, inode->i_ino,
+ inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
inode->i_mode, atomic_read(&inode->i_count), dentry_count);
}
}
}
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
-void lustre_throw_orphan_dentries(struct super_block *sb)
-{
- struct dentry *dentry, *next;
- struct ll_sb_info *sbi = ll_s2sbi(sb);
-
- /* Do this to get rid of orphaned dentries. That is not really trw. */
- list_for_each_entry_safe(dentry, next, &sbi->ll_orphan_dentry_list,
- d_hash) {
- CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
- "before and after shrink_dcache_parent\n",
- dentry->d_name.len, dentry->d_name.name, dentry, next);
- lustre_dump_dentry(dentry, 1);
- shrink_dcache_parent(dentry);
- lustre_dump_dentry(dentry, 1);
- }
-}
-#else
-#define lustre_throw_orphan_dentries(sb)
-#endif
-
+#ifdef HAVE_EXPORT___IGET
static void prune_dir_dentries(struct inode *inode)
{
struct dentry *dentry, *prev = NULL;
} while (empty == 0);
}
+#else /* !HAVE_EXPORT___IGET */
+#define prune_deathrow(sbi, try) do {} while (0)
+#endif /* HAVE_EXPORT___IGET */
void client_common_put_super(struct super_block *sb)
{
struct ll_sb_info *sbi = ll_s2sbi(sb);
ENTRY;
- ll_close_thread_shutdown(sbi->ll_lcq);
+#ifdef CONFIG_FS_POSIX_ACL
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
+ et_fini(&sbi->ll_et);
+ rct_fini(&sbi->ll_rct);
+ }
+#endif
- lprocfs_unregister_mountpoint(sbi);
+ obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
+
+ ll_close_thread_shutdown(sbi->ll_lcq);
/* destroy inodes in deathrow */
prune_deathrow(sbi, 0);
list_del(&sbi->ll_conn_chain);
- obd_disconnect(sbi->ll_osc_exp);
- sbi->ll_osc_exp = NULL;
- obd_disconnect(sbi->ll_mdc_exp);
- sbi->ll_mdc_exp = NULL;
+ obd_fid_fini(sbi->ll_dt_exp);
+ obd_disconnect(sbi->ll_dt_exp);
+ sbi->ll_dt_exp = NULL;
+
+ lprocfs_unregister_mountpoint(sbi);
+
+ obd_fid_fini(sbi->ll_md_exp);
+ obd_disconnect(sbi->ll_md_exp);
+ sbi->ll_md_exp = NULL;
- lustre_throw_orphan_dentries(sb);
+ EXIT;
+}
+
+void ll_kill_super(struct super_block *sb)
+{
+ struct ll_sb_info *sbi;
+
+ ENTRY;
+
+ /* not init sb ?*/
+ if (!(sb->s_flags & MS_ACTIVE))
+ return;
+
+ sbi = ll_s2sbi(sb);
+ /* we need restore s_dev from changed for clustred NFS before put_super
+ * because new kernels have cached s_dev and change sb->s_dev in
+ * put_super not affected real removing devices */
+ if (sbi)
+ sb->s_dev = sbi->ll_sdev_orig;
EXIT;
}
*flags |= tmp;
goto next;
}
- tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK);
+ tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
+ if (tmp) {
+ *flags |= tmp;
+ goto next;
+ }
+ tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
if (tmp) {
*flags &= ~tmp;
goto next;
if (tmp) {
goto next;
}
+ tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
+ if (tmp) {
+ *flags |= tmp;
+ goto next;
+ }
+
+ tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
+ if (tmp) {
+ *flags |= tmp;
+ goto next;
+ }
+ tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
+ if (tmp) {
+ *flags &= ~tmp;
+ goto next;
+ }
+ tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
+ if (tmp) {
+ *flags |= tmp;
+ goto next;
+ }
+ tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
+ if (tmp) {
+ *flags &= ~tmp;
+ goto next;
+ }
- LCONSOLE_ERROR("Unknown option '%s', won't mount.\n", s1);
+ LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
+ s1);
RETURN(-EINVAL);
next:
/* Find next opt */
s2 = strchr(s1, ',');
- if (s2 == NULL)
+ if (s2 == NULL)
break;
s1 = s2 + 1;
}
RETURN(0);
}
-
+
void ll_lli_init(struct ll_inode_info *lli)
{
- sema_init(&lli->lli_open_sem, 1);
+ lli->lli_inode_magic = LLI_INODE_MAGIC;
sema_init(&lli->lli_size_sem, 1);
sema_init(&lli->lli_write_sem, 1);
lli->lli_flags = 0;
lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
spin_lock_init(&lli->lli_lock);
INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
+ INIT_LIST_HEAD(&lli->lli_close_list);
lli->lli_inode_magic = LLI_INODE_MAGIC;
sema_init(&lli->lli_och_sem, 1);
lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
lli->lli_open_fd_exec_count = 0;
INIT_LIST_HEAD(&lli->lli_dead_list);
+ lli->lli_remote_perms = NULL;
+ lli->lli_rmtperm_utime = 0;
+ sema_init(&lli->lli_rmtperm_sem, 1);
+ INIT_LIST_HEAD(&lli->lli_oss_capas);
}
-/* COMPAT_146 */
-#define MDCDEV "mdc_dev"
-static int old_lustre_process_log(struct super_block *sb, char *newprofile,
- struct config_llog_instance *cfg)
-{
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct obd_device *obd;
- struct lustre_handle mdc_conn = {0, };
- struct obd_export *exp;
- char *ptr, *mdt, *profile;
- char niduuid[10] = "mdtnid0";
- class_uuid_t uuid;
- struct obd_uuid mdc_uuid;
- struct llog_ctxt *ctxt;
- struct obd_connect_data ocd = { 0 };
- lnet_nid_t nid;
- int i, rc = 0, recov_bk = 1, failnodes = 0;
- ENTRY;
-
- class_generate_random_uuid(uuid);
- class_uuid_unparse(uuid, &mdc_uuid);
- CDEBUG(D_HA, "generated uuid: %s\n", mdc_uuid.uuid);
-
- /* Figure out the old mdt and profile name from new-style profile
- ("lustre" from "mds/lustre-client") */
- mdt = newprofile;
- profile = strchr(mdt, '/');
- if (profile == NULL) {
- CDEBUG(D_CONFIG, "Can't find MDT name in %s\n", newprofile);
- GOTO(out, rc = -EINVAL);
- }
- *profile = '\0';
- profile++;
- ptr = strrchr(profile, '-');
- if (ptr == NULL) {
- CDEBUG(D_CONFIG, "Can't find client name in %s\n", newprofile);
- GOTO(out, rc = -EINVAL);
- }
- *ptr = '\0';
-
- LCONSOLE_WARN("This looks like an old mount command; I will try to "
- "contact MDT '%s' for profile '%s'\n", mdt, profile);
-
- /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
- i = 0;
- ptr = lsi->lsi_lmd->lmd_dev;
- while (class_parse_nid(ptr, &nid, &ptr) == 0) {
- rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid, 0,0,0);
- i++;
- /* Stop at the first failover nid */
- if (*ptr == ':')
- break;
- }
- if (i == 0) {
- CERROR("No valid MDT nids found.\n");
- GOTO(out, rc = -EINVAL);
- }
- failnodes++;
-
- rc = do_lcfg(MDCDEV, 0, LCFG_ATTACH, LUSTRE_MDC_NAME,mdc_uuid.uuid,0,0);
- if (rc < 0)
- GOTO(out_del_uuid, rc);
-
- rc = do_lcfg(MDCDEV, 0, LCFG_SETUP, mdt, niduuid, 0, 0);
- if (rc < 0) {
- LCONSOLE_ERROR("I couldn't establish a connection with the MDT."
- " Check that the MDT host NID is correct and the"
- " networks are up.\n");
- GOTO(out_detach, rc);
- }
-
- obd = class_name2obd(MDCDEV);
- if (obd == NULL)
- GOTO(out_cleanup, rc = -EINVAL);
-
- /* Add any failover nids */
- while (*ptr == ':') {
- /* New failover node */
- sprintf(niduuid, "mdtnid%d", failnodes);
- i = 0;
- while (class_parse_nid(ptr, &nid, &ptr) == 0) {
- i++;
- rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid,0,0,0);
- if (rc)
- CERROR("Add uuid for %s failed %d\n",
- libcfs_nid2str(nid), rc);
- if (*ptr == ':')
- break;
- }
- if (i > 0) {
- rc = do_lcfg(MDCDEV, 0, LCFG_ADD_CONN, niduuid, 0, 0,0);
- if (rc)
- CERROR("Add conn for %s failed %d\n",
- libcfs_nid2str(nid), rc);
- failnodes++;
- } else {
- /* at ":/fsname" */
- break;
- }
- }
-
- /* Try all connections, but only once. */
- rc = obd_set_info_async(obd->obd_self_export,
- strlen("init_recov_bk"), "init_recov_bk",
- sizeof(recov_bk), &recov_bk, NULL);
- if (rc)
- GOTO(out_cleanup, rc);
-
- /* If we don't have this then an ACL MDS will refuse the connection */
- ocd.ocd_connect_flags = OBD_CONNECT_ACL;
-
- rc = obd_connect(&mdc_conn, obd, &mdc_uuid, &ocd);
- if (rc) {
- CERROR("cannot connect to %s: rc = %d\n", mdt, rc);
- GOTO(out_cleanup, rc);
- }
-
- exp = class_conn2export(&mdc_conn);
-
- ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
-
- cfg->cfg_flags |= CFG_F_COMPAT146;
-
-#if 1
- rc = class_config_parse_llog(ctxt, profile, cfg);
-#else
- /*
- * For debugging, it's useful to just dump the log
- */
- rc = class_config_dump_llog(ctxt, profile, cfg);
-#endif
- switch (rc) {
- case 0: {
- /* Set the caller's profile name to the old-style */
- memcpy(newprofile, profile, strlen(profile) + 1);
- break;
- }
- case -EINVAL:
- LCONSOLE_ERROR("%s: The configuration '%s' could not be read "
- "from the MDT '%s'. Make sure this client and "
- "the MDT are running compatible versions of "
- "Lustre.\n",
- obd->obd_name, profile, mdt);
- /* fall through */
- default:
- LCONSOLE_ERROR("%s: The configuration '%s' could not be read "
- "from the MDT '%s'. This may be the result of "
- "communication errors between the client and "
- "the MDT, or if the MDT is not running.\n",
- obd->obd_name, profile, mdt);
- break;
- }
-
- /* We don't so much care about errors in cleaning up the config llog
- * connection, as we have already read the config by this point. */
- obd_disconnect(exp);
-
-out_cleanup:
- do_lcfg(MDCDEV, 0, LCFG_CLEANUP, 0, 0, 0, 0);
-
-out_detach:
- do_lcfg(MDCDEV, 0, LCFG_DETACH, 0, 0, 0, 0);
-
-out_del_uuid:
- /* class_add_uuid adds a nid even if the same uuid exists; we might
- delete any copy here. So they all better match. */
- for (i = 0; i < failnodes; i++) {
- sprintf(niduuid, "mdtnid%d", i);
- do_lcfg(MDCDEV, 0, LCFG_DEL_UUID, niduuid, 0, 0, 0);
- }
- /* class_import_put will get rid of the additional connections */
-out:
- RETURN(rc);
-}
-/* end COMPAT_146 */
-
int ll_fill_super(struct super_block *sb)
{
struct lustre_profile *lprof;
struct lustre_sb_info *lsi = s2lsi(sb);
struct ll_sb_info *sbi;
- char *osc = NULL, *mdc = NULL;
+ char *dt = NULL, *md = NULL;
char *profilenm = get_profile_name(sb);
- struct config_llog_instance cfg;
+ struct config_llog_instance cfg = {0, };
char ll_instance[sizeof(sb) * 2 + 1];
int err;
ENTRY;
cfs_module_get();
+ sb->s_type->fs_flags |= FS_RENAME_DOES_D_MOVE;
/* client additional sb info */
lsi->lsi_llsbi = sbi = ll_init_sbi();
if (!sbi) {
GOTO(out_free, err);
/* Generate a string unique to this super, in case some joker tries
- to mount the same fs at two mount points.
+ to mount the same fs at two mount points.
Use the address of the super itself.*/
sprintf(ll_instance, "%p", sb);
cfg.cfg_instance = ll_instance;
cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
- cfg.cfg_last_idx = 0;
/* set up client obds */
err = lustre_process_log(sb, profilenm, &cfg);
- /* COMPAT_146 */
- if (err < 0) {
- char *oldname;
- int rc, oldnamelen;
- oldnamelen = strlen(profilenm) + 1;
- /* Temp storage for 1.4.6 profile name */
- OBD_ALLOC(oldname, oldnamelen);
- if (oldname) {
- memcpy(oldname, profilenm, oldnamelen);
- rc = old_lustre_process_log(sb, oldname, &cfg);
- if (rc >= 0) {
- /* That worked - update the profile name
- permanently */
- err = rc;
- OBD_FREE(lsi->lsi_lmd->lmd_profile,
- strlen(lsi->lsi_lmd->lmd_profile) + 1);
- OBD_ALLOC(lsi->lsi_lmd->lmd_profile,
- strlen(oldname) + 1);
- if (!lsi->lsi_lmd->lmd_profile) {
- OBD_FREE(oldname, oldnamelen);
- GOTO(out_free, err = -ENOMEM);
- }
- memcpy(lsi->lsi_lmd->lmd_profile, oldname,
- strlen(oldname) + 1);
- profilenm = get_profile_name(sb);
- }
- OBD_FREE(oldname, oldnamelen);
- }
- }
- /* end COMPAT_146 */
if (err < 0) {
CERROR("Unable to process log: %d\n", err);
GOTO(out_free, err);
lprof = class_get_profile(profilenm);
if (lprof == NULL) {
- LCONSOLE_ERROR("The client profile '%s' could not be read "
- "from the MGS. Does that filesystem exist?\n",
- profilenm);
+ LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
+ " read from the MGS. Does that filesystem "
+ "exist?\n", profilenm);
GOTO(out_free, err = -EINVAL);
}
- CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
- lprof->lp_mdc, lprof->lp_osc);
+ CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
+ lprof->lp_md, lprof->lp_dt);
- OBD_ALLOC(osc, strlen(lprof->lp_osc) +
+ OBD_ALLOC(dt, strlen(lprof->lp_dt) +
strlen(ll_instance) + 2);
- if (!osc)
+ if (!dt)
GOTO(out_free, err = -ENOMEM);
- sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
+ sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
- OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
+ OBD_ALLOC(md, strlen(lprof->lp_md) +
strlen(ll_instance) + 2);
- if (!mdc)
+ if (!md)
GOTO(out_free, err = -ENOMEM);
- sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
-
+ sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
+
/* connections, registrations, sb setup */
- err = client_common_fill_super(sb, mdc, osc);
-
+ err = client_common_fill_super(sb, md, dt);
+
out_free:
- if (mdc)
- OBD_FREE(mdc, strlen(mdc) + 1);
- if (osc)
- OBD_FREE(osc, strlen(osc) + 1);
+ if (md)
+ OBD_FREE(md, strlen(md) + 1);
+ if (dt)
+ OBD_FREE(dt, strlen(dt) + 1);
if (err)
ll_put_super(sb);
else
ENTRY;
CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
-
+
+ ll_print_capa_stat(sbi);
+
sprintf(ll_instance, "%p", sb);
cfg.cfg_instance = ll_instance;
lustre_end_log(sb, NULL, &cfg);
- if (sbi->ll_mdc_exp) {
- obd = class_exp2obd(sbi->ll_mdc_exp);
+ if (sbi->ll_md_exp) {
+ obd = class_exp2obd(sbi->ll_md_exp);
if (obd)
- force = obd->obd_no_recov;
+ force = obd->obd_force;
}
/* We need to set force before the lov_disconnect in
/* Only if client_common_fill_super succeeded */
client_common_put_super(sb);
}
-
next = 0;
while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
class_manual_cleanup(obd);
- }
-
- if (profilenm)
+ }
+
+ if (profilenm)
class_del_profile(profilenm);
ll_free_sbi(sb);
void ll_clear_inode(struct inode *inode)
{
- struct ll_fid fid;
struct ll_inode_info *lli = ll_i2info(inode);
struct ll_sb_info *sbi = ll_i2sbi(inode);
ENTRY;
CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
inode->i_generation, inode);
- ll_inode2fid(&fid, inode);
- clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
- mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
+ if (S_ISDIR(inode->i_mode)) {
+ /* these should have been cleared in ll_file_release */
+ LASSERT(lli->lli_sai == NULL);
+ LASSERT(lli->lli_opendir_key == NULL);
+ LASSERT(lli->lli_opendir_pid == 0);
+ }
+
+ ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
+ md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
+ null_if_equal, inode);
LASSERT(!lli->lli_open_fd_write_count);
LASSERT(!lli->lli_open_fd_read_count);
LASSERT(!lli->lli_open_fd_exec_count);
if (lli->lli_mds_write_och)
- ll_mdc_real_close(inode, FMODE_WRITE);
- if (lli->lli_mds_exec_och) {
- if (!FMODE_EXEC)
- CERROR("No FMODE exec, bug exec och is present for "
- "inode %ld\n", inode->i_ino);
- ll_mdc_real_close(inode, FMODE_EXEC);
- }
+ ll_md_real_close(inode, FMODE_WRITE);
+ if (lli->lli_mds_exec_och)
+ ll_md_real_close(inode, FMODE_EXEC);
if (lli->lli_mds_read_och)
- ll_mdc_real_close(inode, FMODE_READ);
-
+ ll_md_real_close(inode, FMODE_READ);
if (lli->lli_smd) {
- obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
+ obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
null_if_equal, inode);
- obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
+ obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
lli->lli_smd = NULL;
}
lli->lli_symlink_name = NULL;
}
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
+ LASSERT(lli->lli_posix_acl == NULL);
+ if (lli->lli_remote_perms) {
+ free_rmtperm_hash(lli->lli_remote_perms);
+ lli->lli_remote_perms = NULL;
+ }
+ }
#ifdef CONFIG_FS_POSIX_ACL
- if (lli->lli_posix_acl) {
+ else if (lli->lli_posix_acl) {
LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
+ LASSERT(lli->lli_remote_perms == NULL);
posix_acl_release(lli->lli_posix_acl);
lli->lli_posix_acl = NULL;
}
#endif
-
lli->lli_inode_magic = LLI_INODE_DEAD;
+#ifdef HAVE_EXPORT___IGET
spin_lock(&sbi->ll_deathrow_lock);
list_del_init(&lli->lli_dead_list);
spin_unlock(&sbi->ll_deathrow_lock);
+#endif
+ ll_clear_inode_capas(inode);
EXIT;
}
+int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
+ struct md_open_data **mod)
+{
+ struct lustre_md md;
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+ struct ptlrpc_request *request = NULL;
+ int rc;
+ ENTRY;
+
+ op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
+ LUSTRE_OPC_ANY, NULL);
+ if (IS_ERR(op_data))
+ RETURN(PTR_ERR(op_data));
+
+ rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
+ &request, mod);
+ if (rc) {
+ ptlrpc_req_finished(request);
+ if (rc == -ENOENT) {
+ inode->i_nlink = 0;
+ /* Unlinked special device node? Or just a race?
+ * Pretend we done everything. */
+ if (!S_ISREG(inode->i_mode) &&
+ !S_ISDIR(inode->i_mode))
+ rc = inode_setattr(inode, &op_data->op_attr);
+ } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
+ CERROR("md_setattr fails: rc = %d\n", rc);
+ }
+ RETURN(rc);
+ }
+
+ rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
+ sbi->ll_md_exp, &md);
+ if (rc) {
+ ptlrpc_req_finished(request);
+ RETURN(rc);
+ }
+
+ /* We call inode_setattr to adjust timestamps.
+ * If there is at least some data in file, we cleared ATTR_SIZE
+ * above to avoid invoking vmtruncate, otherwise it is important
+ * to call vmtruncate in inode_setattr to update inode->i_size
+ * (bug 6196) */
+ rc = inode_setattr(inode, &op_data->op_attr);
+
+ /* Extract epoch data if obtained. */
+ op_data->op_handle = md.body->handle;
+ op_data->op_ioepoch = md.body->ioepoch;
+
+ ll_update_inode(inode, &md);
+ ptlrpc_req_finished(request);
+
+ RETURN(rc);
+}
+
+/* Close IO epoch and send Size-on-MDS attribute update. */
+static int ll_setattr_done_writing(struct inode *inode,
+ struct md_op_data *op_data,
+ struct md_open_data *mod)
+{
+ struct ll_inode_info *lli = ll_i2info(inode);
+ int rc = 0;
+ ENTRY;
+
+ LASSERT(op_data != NULL);
+ if (!S_ISREG(inode->i_mode))
+ RETURN(0);
+
+ CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
+ op_data->op_ioepoch, PFID(&lli->lli_fid));
+
+ op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
+ rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
+ if (rc == -EAGAIN) {
+ /* MDS has instructed us to obtain Size-on-MDS attribute
+ * from OSTs and send setattr to back to MDS. */
+ rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
+ op_data->op_ioepoch);
+ } else if (rc) {
+ CERROR("inode %lu mdc truncate failed: rc = %d\n",
+ inode->i_ino, rc);
+ }
+ RETURN(rc);
+}
+
+static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
+{
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+ struct ll_inode_info *lli = ll_i2info(inode);
+ struct lov_stripe_md *lsm = lli->lli_smd;
+ int rc;
+ ldlm_policy_data_t policy = { .l_extent = {new_size,
+ OBD_OBJECT_EOF } };
+ struct lustre_handle lockh = { 0 };
+ int local_lock = 0; /* 0 - no local lock;
+ * 1 - lock taken by lock_extent;
+ * 2 - by obd_match*/
+ int ast_flags;
+ int err;
+ ENTRY;
+
+ UNLOCK_INODE_MUTEX(inode);
+ UP_WRITE_I_ALLOC_SEM(inode);
+
+ if (sbi->ll_lockless_truncate_enable &&
+ (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
+ ast_flags = LDLM_FL_BLOCK_GRANTED;
+ rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
+ &policy, LCK_PW, &ast_flags, inode, &lockh);
+ if (rc > 0) {
+ local_lock = 2;
+ rc = 0;
+ } else if (rc == 0) {
+ rc = ll_file_punch(inode, new_size, 1);
+ }
+ } else {
+ /* XXX when we fix the AST intents to pass the discard-range
+ * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
+ * XXX here. */
+ ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
+ rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
+ &lockh, ast_flags);
+ if (likely(rc == 0))
+ local_lock = 1;
+ }
+
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
+ DOWN_WRITE_I_ALLOC_SEM(inode);
+ LOCK_INODE_MUTEX(inode);
+#else
+ LOCK_INODE_MUTEX(inode);
+ DOWN_WRITE_I_ALLOC_SEM(inode);
+#endif
+ if (likely(rc == 0)) {
+ /* Only ll_inode_size_lock is taken at this level.
+ * lov_stripe_lock() is grabbed by ll_truncate() only over
+ * call to obd_adjust_kms(). If vmtruncate returns 0, then
+ * ll_truncate dropped ll_inode_size_lock() */
+ ll_inode_size_lock(inode, 0);
+ if (!local_lock) {
+ spin_lock(&lli->lli_lock);
+ lli->lli_flags |= LLIF_SRVLOCK;
+ spin_unlock(&lli->lli_lock);
+ }
+ rc = vmtruncate(inode, new_size);
+ if (!local_lock) {
+ spin_lock(&lli->lli_lock);
+ lli->lli_flags &= ~LLIF_SRVLOCK;
+ spin_unlock(&lli->lli_lock);
+ }
+ if (rc != 0) {
+ LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
+ ll_inode_size_unlock(inode, 0);
+ }
+ }
+
+ if (local_lock) {
+ if (local_lock == 2)
+ err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
+ else
+ err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
+ if (unlikely(err != 0)){
+ CERROR("extent unlock failed: err=%d,"
+ " unlock method =%d\n", err, local_lock);
+ if (rc == 0)
+ rc = err;
+ }
+ }
+ RETURN(rc);
+}
+
/* If this inode has objects allocated to it (lsm != NULL), then the OST
* object(s) determine the file size and mtime. Otherwise, the MDS will
* keep these values until such a time that objects are allocated for it.
struct ll_inode_info *lli = ll_i2info(inode);
struct lov_stripe_md *lsm = lli->lli_smd;
struct ll_sb_info *sbi = ll_i2sbi(inode);
- struct ptlrpc_request *request = NULL;
- struct mdc_op_data op_data;
- struct lustre_md md;
+ struct md_op_data *op_data = NULL;
+ struct md_open_data *mod = NULL;
int ia_valid = attr->ia_valid;
- int rc = 0;
+ int rc = 0, rc1 = 0;
ENTRY;
CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
attr->ia_valid);
- lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
+ ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
if (ia_valid & ATTR_SIZE) {
if (attr->ia_size > ll_file_maxbytes(inode)) {
/* To avoid stale mtime on mds, obtain it from ost and send
to mds. */
rc = ll_glimpse_size(inode, 0);
- if (rc)
+ if (rc)
RETURN(rc);
-
+
attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
attr->ia_mtime = inode->i_mtime;
}
if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
- CURRENT_SECONDS);
+ cfs_time_current_sec());
/* NB: ATTR_SIZE will only be set after this point if the size
* resides on the MDS, ie, this file has no objects. */
/* We always do an MDS RPC, even if we're only changing the size;
* only the MDS knows whether truncate() should fail with -ETXTBUSY */
- ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
- rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
- attr, NULL, 0, NULL, 0, &request);
+ OBD_ALLOC_PTR(op_data);
+ if (op_data == NULL)
+ RETURN(-ENOMEM);
- if (rc) {
- ptlrpc_req_finished(request);
- if (rc == -ENOENT) {
- inode->i_nlink = 0;
- /* Unlinked special device node? Or just a race?
- * Pretend we done everything. */
- if (!S_ISREG(inode->i_mode) &&
- !S_ISDIR(inode->i_mode))
- rc = inode_setattr(inode, attr);
- } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY)
- CERROR("mdc_setattr fails: rc = %d\n", rc);
- RETURN(rc);
- }
+ memcpy(&op_data->op_attr, attr, sizeof(*attr));
- rc = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
- if (rc) {
- ptlrpc_req_finished(request);
- RETURN(rc);
- }
+ /* Open epoch for truncate. */
+ if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
+ (ia_valid & ATTR_SIZE))
+ op_data->op_flags = MF_EPOCH_OPEN;
- /* We call inode_setattr to adjust timestamps.
- * If there is at least some data in file, we cleared ATTR_SIZE above to
- * avoid invoking vmtruncate, otherwise it is important to call
- * vmtruncate in inode_setattr to update inode->i_size (bug 6196) */
- rc = inode_setattr(inode, attr);
+ rc = ll_md_setattr(inode, op_data, &mod);
+ if (rc)
+ GOTO(out, rc);
- ll_update_inode(inode, &md);
- ptlrpc_req_finished(request);
+ if (op_data->op_ioepoch)
+ CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
+ "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
if (!lsm || !S_ISREG(inode->i_mode)) {
CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
- RETURN(rc);
+ GOTO(out, rc = 0);
}
/* We really need to get our PW lock before we change inode->i_size.
* last one is especially bad for racing o_append users on other
* nodes. */
if (ia_valid & ATTR_SIZE) {
- ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
- OBD_OBJECT_EOF } };
- struct lustre_handle lockh = { 0 };
- int err, ast_flags = 0;
- /* XXX when we fix the AST intents to pass the discard-range
- * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
- * XXX here. */
- if (attr->ia_size == 0)
- ast_flags = LDLM_AST_DISCARD_DATA;
-
- UNLOCK_INODE_MUTEX(inode);
- UP_WRITE_I_ALLOC_SEM(inode);
- rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
- ast_flags);
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- DOWN_WRITE_I_ALLOC_SEM(inode);
- LOCK_INODE_MUTEX(inode);
-#else
- LOCK_INODE_MUTEX(inode);
- DOWN_WRITE_I_ALLOC_SEM(inode);
-#endif
- if (rc != 0)
- RETURN(rc);
-
- /* Only ll_inode_size_lock is taken at this level.
- * lov_stripe_lock() is grabbed by ll_truncate() only over
- * call to obd_adjust_kms(). If vmtruncate returns 0, then
- * ll_truncate dropped ll_inode_size_lock() */
- ll_inode_size_lock(inode, 0);
- rc = vmtruncate(inode, attr->ia_size);
- if (rc != 0) {
- LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
- ll_inode_size_unlock(inode, 0);
- }
-
- err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
- if (err) {
- CERROR("ll_extent_unlock failed: %d\n", err);
- if (!rc)
- rc = err;
- }
+ rc = ll_setattr_do_truncate(inode, attr->ia_size);
} else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
obd_flag flags;
struct obd_info oinfo = { { { 0 } } };
- struct obdo *oa = obdo_alloc();
+ struct obdo *oa;
CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
inode->i_ino, LTIME_S(attr->ia_mtime));
+ OBDO_ALLOC(oa);
if (oa) {
oa->o_id = lsm->lsm_object_id;
- oa->o_valid = OBD_MD_FLID;
+ oa->o_gr = lsm->lsm_object_gr;
+ oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
OBD_MD_FLMTIME | OBD_MD_FLCTIME |
- OBD_MD_FLFID | OBD_MD_FLGENER;
+ OBD_MD_FLFID | OBD_MD_FLGENER |
+ OBD_MD_FLGROUP;
obdo_from_inode(oa, inode, flags);
oinfo.oi_oa = oa;
oinfo.oi_md = lsm;
+ oinfo.oi_capa = ll_mdscapa_get(inode);
- rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
+ /* XXX: this looks unnecessary now. */
+ rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
+ capa_put(oinfo.oi_capa);
if (rc)
CERROR("obd_setattr_async fails: rc=%d\n", rc);
- obdo_free(oa);
+ OBDO_FREE(oa);
} else {
rc = -ENOMEM;
}
}
- RETURN(rc);
+ EXIT;
+out:
+ if (op_data) {
+ if (op_data->op_ioepoch)
+ rc1 = ll_setattr_done_writing(inode, op_data, mod);
+ ll_finish_md_op_data(op_data);
+ }
+ return rc ? rc : rc1;
}
int ll_setattr(struct dentry *de, struct iattr *attr)
{
- ll_vfs_ops_tally(ll_i2sbi(de->d_inode), VFS_OPS_SETATTR);
-
if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
(ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
}
int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
- __u64 max_age)
+ __u64 max_age, __u32 flags)
{
struct ll_sb_info *sbi = ll_s2sbi(sb);
struct obd_statfs obd_osfs;
int rc;
ENTRY;
- rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
+ rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
if (rc) {
- CERROR("mdc_statfs fails: rc = %d\n", rc);
+ CERROR("md_statfs fails: rc = %d\n", rc);
RETURN(rc);
}
CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
- rc = obd_statfs_rqset(class_exp2obd(sbi->ll_osc_exp),
- &obd_osfs, max_age);
+ rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
+ &obd_osfs, max_age, flags);
if (rc) {
CERROR("obd_statfs fails: rc = %d\n", rc);
RETURN(rc);
obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
obd_osfs.os_files);
+ osfs->os_bsize = obd_osfs.os_bsize;
osfs->os_blocks = obd_osfs.os_blocks;
osfs->os_bfree = obd_osfs.os_bfree;
osfs->os_bavail = obd_osfs.os_bavail;
int rc;
CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
- lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
+ ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
/* For now we will always get up-to-date statfs values, but in the
* future we may allow some amount of caching on the client (e.g.
* from QOS or lprocfs updates). */
- rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
+ rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
if (rc)
return rc;
dump_lsm(D_INODE, lli->lli_smd);
LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
"lsm must be joined lsm %p\n", lsm);
- obd_free_memmd(ll_i2obdexp(inode), &lli->lli_smd);
+ obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
lli->lli_smd = lsm;
void ll_update_inode(struct inode *inode, struct lustre_md *md)
{
struct ll_inode_info *lli = ll_i2info(inode);
- struct mds_body *body = md->body;
+ struct mdt_body *body = md->body;
struct lov_stripe_md *lsm = md->lsm;
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
if (lsm != NULL) {
ll_replace_lsm(inode, lsm);
}
if (lli->lli_smd != lsm)
- obd_free_memmd(ll_i2obdexp(inode), &lsm);
+ obd_free_memmd(ll_i2dtexp(inode), &lsm);
}
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
+ if (body->valid & OBD_MD_FLRMTPERM)
+ ll_update_remote_perm(inode, md->remote_perm);
+ }
#ifdef CONFIG_FS_POSIX_ACL
- LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
- if (body->valid & OBD_MD_FLACL) {
+ else if (body->valid & OBD_MD_FLACL) {
spin_lock(&lli->lli_lock);
if (lli->lli_posix_acl)
posix_acl_release(lli->lli_posix_acl);
spin_unlock(&lli->lli_lock);
}
#endif
+ inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
- if (body->valid & OBD_MD_FLID)
- inode->i_ino = body->ino;
if (body->valid & OBD_MD_FLATIME &&
body->atime > LTIME_S(inode->i_atime))
LTIME_S(inode->i_atime) = body->atime;
if (body->valid & OBD_MD_FLTYPE)
inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
if (S_ISREG(inode->i_mode)) {
- inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS+1, LL_MAX_BLKSIZE_BITS);
+ inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
} else {
inode->i_blkbits = inode->i_sb->s_blocksize_bits;
}
inode->i_flags = ll_ext_to_inode_flags(body->flags);
if (body->valid & OBD_MD_FLNLINK)
inode->i_nlink = body->nlink;
- if (body->valid & OBD_MD_FLGENER)
- inode->i_generation = body->generation;
if (body->valid & OBD_MD_FLRDEV)
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- inode->i_rdev = body->rdev;
-#else
inode->i_rdev = old_decode_dev(body->rdev);
-#endif
- if (body->valid & OBD_MD_FLSIZE)
- inode->i_size = body->size;
- if (body->valid & OBD_MD_FLBLOCKS)
- inode->i_blocks = body->blocks;
- if (body->valid & OBD_MD_FLSIZE)
- set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
+ if (body->valid & OBD_MD_FLID) {
+ /* FID shouldn't be changed! */
+ if (fid_is_sane(&lli->lli_fid)) {
+ LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
+ "Trying to change FID "DFID
+ " to the "DFID", inode %lu/%u(%p)\n",
+ PFID(&lli->lli_fid), PFID(&body->fid1),
+ inode->i_ino, inode->i_generation, inode);
+ } else
+ lli->lli_fid = body->fid1;
+ }
+
+ LASSERT(fid_seq(&lli->lli_fid) != 0);
+
+ if (body->valid & OBD_MD_FLSIZE) {
+ if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
+ S_ISREG(inode->i_mode) && lli->lli_smd) {
+ struct lustre_handle lockh;
+ ldlm_mode_t mode;
+
+ /* As it is possible a blocking ast has been processed
+ * by this time, we need to check there is an UPDATE
+ * lock on the client and set LLIF_MDS_SIZE_LOCK holding
+ * it. */
+ mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
+ &lockh);
+ if (mode) {
+ if (lli->lli_flags & (LLIF_DONE_WRITING |
+ LLIF_EPOCH_PENDING |
+ LLIF_SOM_DIRTY)) {
+ CERROR("ino %lu flags %lu still has "
+ "size authority! do not trust "
+ "the size got from MDS\n",
+ inode->i_ino, lli->lli_flags);
+ } else {
+ /* Use old size assignment to avoid
+ * deadlock bz14138 & bz14326 */
+ inode->i_size = body->size;
+ lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
+ }
+ ldlm_lock_decref(&lockh, mode);
+ }
+ } else {
+ /* Use old size assignment to avoid
+ * deadlock bz14138 & bz14326 */
+ inode->i_size = body->size;
+ }
+
+ if (body->valid & OBD_MD_FLBLOCKS)
+ inode->i_blocks = body->blocks;
+ }
+
+ if (body->valid & OBD_MD_FLMDSCAPA) {
+ LASSERT(md->mds_capa);
+ ll_add_capa(inode, md->mds_capa);
+ }
+ if (body->valid & OBD_MD_FLOSSCAPA) {
+ LASSERT(md->oss_capa);
+ ll_add_capa(inode, md->oss_capa);
+ }
}
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
static struct backing_dev_info ll_backing_dev_info = {
.ra_pages = 0, /* No readahead */
#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
.memory_backed = 0, /* Does contribute to dirty memory */
#endif
};
-#endif
void ll_read_inode2(struct inode *inode, void *opaque)
{
struct ll_inode_info *lli = ll_i2info(inode);
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
- inode->i_generation, inode);
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
+ inode->i_ino, inode->i_generation, inode);
ll_lli_init(lli);
} else {
inode->i_op = &ll_special_inode_operations;
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
init_special_inode(inode, inode->i_mode,
kdev_t_to_nr(inode->i_rdev));
/* initializing backing dev info. */
inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
-#else
- init_special_inode(inode, inode->i_mode, inode->i_rdev);
-#endif
+
EXIT;
}
}
+void ll_delete_inode(struct inode *inode)
+{
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+ int rc;
+ ENTRY;
+
+ rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
+ if (rc) {
+ CERROR("fid_delete() failed, rc %d\n", rc);
+ }
+ truncate_inode_pages(&inode->i_data, 0);
+ clear_inode(inode);
+
+ EXIT;
+}
+
int ll_iocontrol(struct inode *inode, struct file *file,
unsigned int cmd, unsigned long arg)
{
switch(cmd) {
case EXT3_IOC_GETFLAGS: {
- struct ll_fid fid;
- struct mds_body *body;
+ struct mdt_body *body;
+ struct obd_capa *oc;
- ll_inode2fid(&fid, inode);
- rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLFLAGS,0,&req);
+ oc = ll_mdscapa_get(inode);
+ rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
+ OBD_MD_FLFLAGS, 0, &req);
+ capa_put(oc);
if (rc) {
CERROR("failure %d inode %lu\n", rc, inode->i_ino);
RETURN(-abs(rc));
}
- body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
- sizeof(*body));
+ body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+
+ flags = body->flags;
- /* We want to return EXT3_*_FL flags to the caller via this
- * ioctl. An older MDS may be sending S_* flags, fix it up. */
- flags = ll_inode_to_ext_flags(body->flags, body->flags);
- ptlrpc_req_finished (req);
+ ptlrpc_req_finished(req);
RETURN(put_user(flags, (int *)arg));
}
case EXT3_IOC_SETFLAGS: {
- struct mdc_op_data op_data;
- struct ll_iattr_struct attr = { 0 };
- struct obd_info oinfo = { { { 0 } } };
struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
+ struct obd_info oinfo = { { { 0 } } };
+ struct md_op_data *op_data;
if (get_user(flags, (int *)arg))
RETURN(-EFAULT);
oinfo.oi_md = lsm;
- oinfo.oi_oa = obdo_alloc();
+ OBDO_ALLOC(oinfo.oi_oa);
if (!oinfo.oi_oa)
RETURN(-ENOMEM);
- ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
-
- attr.ia_attr_flags = flags;
- ((struct iattr *)&attr)->ia_valid |= ATTR_ATTR_FLAG;
+ op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
+ LUSTRE_OPC_ANY, NULL);
+ if (IS_ERR(op_data))
+ RETURN(PTR_ERR(op_data));
- rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
- (struct iattr *)&attr, NULL, 0, NULL, 0, &req);
+ ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
+ op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
+ rc = md_setattr(sbi->ll_md_exp, op_data,
+ NULL, 0, NULL, 0, &req, NULL);
+ ll_finish_md_op_data(op_data);
ptlrpc_req_finished(req);
if (rc || lsm == NULL) {
- obdo_free(oinfo.oi_oa);
+ OBDO_FREE(oinfo.oi_oa);
RETURN(rc);
}
oinfo.oi_oa->o_id = lsm->lsm_object_id;
+ oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
oinfo.oi_oa->o_flags = flags;
- oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
+ oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
+ OBD_MD_FLGROUP;
+ oinfo.oi_capa = ll_mdscapa_get(inode);
obdo_from_inode(oinfo.oi_oa, inode,
OBD_MD_FLFID | OBD_MD_FLGENER);
- rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
- obdo_free(oinfo.oi_oa);
+ rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
+ capa_put(oinfo.oi_capa);
+ OBDO_FREE(oinfo.oi_oa);
if (rc) {
if (rc != -EPERM && rc != -EACCES)
- CERROR("mdc_setattr_async fails: rc = %d\n", rc);
+ CERROR("md_setattr_async fails: rc = %d\n", rc);
RETURN(rc);
}
RETURN(0);
}
+int ll_flush_ctx(struct inode *inode)
+{
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+
+ CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
+
+ obd_set_info_async(sbi->ll_md_exp,
+ sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
+ 0, NULL, NULL);
+ obd_set_info_async(sbi->ll_dt_exp,
+ sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
+ 0, NULL, NULL);
+ return 0;
+}
+
/* umount -f client means force down, don't save state */
#ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
sb->s_count, atomic_read(&sb->s_active));
- obd = class_exp2obd(sbi->ll_mdc_exp);
+ obd = class_exp2obd(sbi->ll_md_exp);
if (obd == NULL) {
CERROR("Invalid MDC connection handle "LPX64"\n",
- sbi->ll_mdc_exp->exp_handle.h_cookie);
+ sbi->ll_md_exp->exp_handle.h_cookie);
EXIT;
return;
}
- obd->obd_no_recov = 1;
- obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
+ obd->obd_force = 1;
+ obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
&ioc_data, NULL);
- obd = class_exp2obd(sbi->ll_osc_exp);
+ obd = class_exp2obd(sbi->ll_dt_exp);
if (obd == NULL) {
CERROR("Invalid LOV connection handle "LPX64"\n",
- sbi->ll_osc_exp->exp_handle.h_cookie);
+ sbi->ll_dt_exp->exp_handle.h_cookie);
EXIT;
return;
}
- obd->obd_no_recov = 1;
- obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
+ obd->obd_force = 1;
+ obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
&ioc_data, NULL);
/* Really, we'd like to wait until there are no requests outstanding,
struct ll_sb_info *sbi = ll_s2sbi(sb);
int err;
__u32 read_only;
-
+
if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
read_only = *flags & MS_RDONLY;
- err = obd_set_info_async(sbi->ll_mdc_exp, strlen("read-only"),
- "read-only", sizeof(read_only),
+ err = obd_set_info_async(sbi->ll_md_exp,
+ sizeof(KEY_READ_ONLY) - 1,
+ KEY_READ_ONLY, sizeof(read_only),
&read_only, NULL);
if (err) {
CERROR("Failed to change the read-only flag during "
"remount: %d\n", err);
return err;
}
-
+
if (read_only)
sb->s_flags |= MS_RDONLY;
else
return 0;
}
-int ll_prep_inode(struct obd_export *exp, struct inode **inode,
- struct ptlrpc_request *req, int offset,struct super_block *sb)
+int ll_prep_inode(struct inode **inode,
+ struct ptlrpc_request *req,
+ struct super_block *sb)
{
- struct lustre_md md;
struct ll_sb_info *sbi = NULL;
+ struct lustre_md md;
int rc = 0;
ENTRY;
LASSERT(*inode || sb);
sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
prune_deathrow(sbi, 1);
+ memset(&md, 0, sizeof(struct lustre_md));
- rc = mdc_req2lustre_md(req, offset, exp, &md);
+ rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
+ sbi->ll_md_exp, &md);
if (rc)
RETURN(rc);
if (*inode) {
ll_update_inode(*inode, &md);
} else {
- LASSERT(sb);
- *inode = ll_iget(sb, md.body->ino, &md);
+ LASSERT(sb != NULL);
+
+ /*
+ * At this point server returns to client's same fid as client
+ * generated for creating. So using ->fid1 is okay here.
+ */
+ LASSERT(fid_is_sane(&md.body->fid1));
+
+ *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
if (*inode == NULL || is_bad_inode(*inode)) {
- mdc_free_lustre_md(exp, &md);
+ if (md.lsm)
+ obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
+#ifdef CONFIG_FS_POSIX_ACL
+ if (md.posix_acl) {
+ posix_acl_release(md.posix_acl);
+ md.posix_acl = NULL;
+ }
+#endif
rc = -ENOMEM;
CERROR("new_inode -fatal: rc %d\n", rc);
GOTO(out, rc);
}
}
- rc = obd_checkmd(exp, ll_i2mdcexp(*inode),
+ rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
ll_i2info(*inode)->lli_smd);
out:
+ md_free_lustre_md(sbi->ll_md_exp, &md);
RETURN(rc);
}
[LLAP_ORIGIN_READAHEAD] = "ra",
[LLAP_ORIGIN_COMMIT_WRITE] = "cw",
[LLAP_ORIGIN_WRITEPAGE] = "wp",
+ [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
};
struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
int ll_obd_statfs(struct inode *inode, void *arg)
{
struct ll_sb_info *sbi = NULL;
- struct obd_device *client_obd = NULL, *lov_obd = NULL;
- struct lov_obd *lov = NULL;
- struct obd_statfs stat_buf = {0};
+ struct obd_export *exp;
char *buf = NULL;
struct obd_ioctl_data *data = NULL;
- __u32 type, index;
- int len, rc;
+ __u32 type;
+ int len = 0, rc;
if (!inode || !(sbi = ll_i2sbi(inode)))
GOTO(out_statfs, rc = -EINVAL);
GOTO(out_statfs, rc = -EINVAL);
memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
- memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
-
- if (type == LL_STATFS_MDC) {
- if (index > 0)
- GOTO(out_statfs, rc = -ENODEV);
- client_obd = class_exp2obd(sbi->ll_mdc_exp);
- } else if (type == LL_STATFS_LOV) {
- lov_obd = class_exp2obd(sbi->ll_osc_exp);
- lov = &lov_obd->u.lov;
-
- if ((index >= lov->desc.ld_tgt_count) ||
- !lov->lov_tgts[index])
- GOTO(out_statfs, rc = -ENODEV);
-
- client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
- if (!lov->lov_tgts[index]->ltd_active)
- GOTO(out_uuid, rc = -ENODATA);
- }
-
- if (!client_obd)
- GOTO(out_statfs, rc = -EINVAL);
-
- rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
+ if (type == LL_STATFS_MDC)
+ exp = sbi->ll_md_exp;
+ else if (type == LL_STATFS_LOV)
+ exp = sbi->ll_dt_exp;
+ else
+ GOTO(out_statfs, rc = -ENODEV);
+
+ rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
if (rc)
GOTO(out_statfs, rc);
-
- if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
- GOTO(out_statfs, rc = -EFAULT);
-
-out_uuid:
- if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
- data->ioc_plen2))
- rc = -EFAULT;
-
out_statfs:
if (buf)
obd_ioctl_freedata(buf, len);
unsigned long x;
int rc = 0;
- lprocfs_init_vars(llite, &lvars);
+ lprocfs_llite_init_vars(&lvars);
/* The instance name contains the sb: lustre-client-aacfe000 */
ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
return(rc);
}
+/* this function prepares md_op_data hint for passing ot down to MD stack. */
+struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
+ struct inode *i1, struct inode *i2,
+ const char *name, int namelen,
+ int mode, __u32 opc, void *data)
+{
+ LASSERT(i1 != NULL);
+
+ if (namelen > ll_i2sbi(i1)->ll_namelen)
+ return ERR_PTR(-ENAMETOOLONG);
+
+ if (op_data == NULL)
+ OBD_ALLOC_PTR(op_data);
+
+ if (op_data == NULL)
+ return ERR_PTR(-ENOMEM);
+
+ ll_i2gids(op_data->op_suppgids, i1, i2);
+ op_data->op_fid1 = *ll_inode2fid(i1);
+ op_data->op_capa1 = ll_mdscapa_get(i1);
+
+ if (i2) {
+ op_data->op_fid2 = *ll_inode2fid(i2);
+ op_data->op_capa2 = ll_mdscapa_get(i2);
+ } else {
+ fid_zero(&op_data->op_fid2);
+ op_data->op_capa2 = NULL;
+ }
+
+ op_data->op_name = name;
+ op_data->op_namelen = namelen;
+ op_data->op_mode = mode;
+ op_data->op_mod_time = cfs_time_current_sec();
+ op_data->op_fsuid = current->fsuid;
+ op_data->op_fsgid = current->fsgid;
+ op_data->op_cap = current->cap_effective;
+ op_data->op_bias = MDS_CHECK_SPLIT;
+ op_data->op_opc = opc;
+ op_data->op_mds = 0;
+ op_data->op_data = data;
+
+ return op_data;
+}
+
+void ll_finish_md_op_data(struct md_op_data *op_data)
+{
+ capa_put(op_data->op_capa1);
+ capa_put(op_data->op_capa2);
+ OBD_FREE_PTR(op_data);
+}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff