* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*/
/*
#include <linux/module.h>
#include <linux/types.h>
-#include <linux/random.h>
#include <linux/version.h>
#include <linux/mm.h>
#include <lustre_disk.h>
#include <lustre_param.h>
#include <lustre_log.h>
+#include <cl_object.h>
#include <obd_cksum.h>
-#include <lustre_cache.h>
#include "llite_internal.h"
cfs_mem_cache_t *ll_file_data_slab;
-LIST_HEAD(ll_super_blocks);
-spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
+CFS_LIST_HEAD(ll_super_blocks);
+cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
extern struct address_space_operations ll_aops;
extern struct address_space_operations ll_dir_aops;
#ifndef log2
-#define log2(n) ffz(~(n))
+#define log2(n) cfs_ffz(~(n))
#endif
-static inline void ll_pglist_fini(struct ll_sb_info *sbi)
-{
- struct page *page;
- int i;
-
- if (sbi->ll_pglist == NULL)
- return;
-
- for_each_possible_cpu(i) {
- page = sbi->ll_pglist[i]->llpd_page;
- if (page) {
- sbi->ll_pglist[i] = NULL;
- __free_page(page);
- }
- }
-
- OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
- sbi->ll_pglist = NULL;
-}
-
-static inline int ll_pglist_init(struct ll_sb_info *sbi)
-{
- struct ll_pglist_data *pd;
- unsigned long budget;
- int i, color = 0;
- ENTRY;
-
- OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
- if (sbi->ll_pglist == NULL)
- RETURN(-ENOMEM);
-
- budget = sbi->ll_async_page_max / num_online_cpus();
- for_each_possible_cpu(i) {
- struct page *page = alloc_pages_node(cpu_to_node(i),
- GFP_KERNEL, 0);
- if (page == NULL) {
- ll_pglist_fini(sbi);
- RETURN(-ENOMEM);
- }
-
- if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
- color = 0;
-
- pd = (struct ll_pglist_data *)(page_address(page) + color);
- memset(pd, 0, sizeof(*pd));
- spin_lock_init(&pd->llpd_lock);
- INIT_LIST_HEAD(&pd->llpd_list);
- if (cpu_online(i))
- pd->llpd_budget = budget;
- pd->llpd_cpu = i;
- pd->llpd_page = page;
- atomic_set(&pd->llpd_sample_count, 0);
- sbi->ll_pglist[i] = pd;
- color += L1_CACHE_ALIGN(sizeof(*pd));
- }
-
- RETURN(0);
-}
-
static struct ll_sb_info *ll_init_sbi(void)
{
struct ll_sb_info *sbi = NULL;
if (!sbi)
RETURN(NULL);
- OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
- if (sbi->ll_async_page_sample == NULL)
- GOTO(out, 0);
-
- 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);
+ cfs_spin_lock_init(&sbi->ll_lock);
+ cfs_init_mutex(&sbi->ll_lco.lco_lock);
+ cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
+ cfs_spin_lock_init(&sbi->ll_process_lock);
sbi->ll_rw_stats_on = 0;
si_meminfo(&si);
} else {
sbi->ll_async_page_max = (pages / 4) * 3;
}
- lcounter_init(&sbi->ll_async_page_count);
- spin_lock_init(&sbi->ll_async_page_reblnc_lock);
- sbi->ll_async_page_sample_max = 64 * num_online_cpus();
- sbi->ll_async_page_reblnc_count = 0;
- sbi->ll_async_page_clock_hand = 0;
- if (ll_pglist_init(sbi))
- GOTO(out, 0);
- sbi->ll_ra_info.ra_max_pages = min(pages / 32,
+ sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
SBI_DEFAULT_READAHEAD_MAX);
+ sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
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);
+ CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
+ CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
ll_generate_random_uuid(uuid);
class_uuid_unparse(uuid, &sbi->ll_sb_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);
+ cfs_spin_lock(&ll_sb_lock);
+ cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
+ cfs_spin_unlock(&ll_sb_lock);
-#ifdef ENABLE_LLITE_CHECKSUM
+#ifdef ENABLE_CHECKSUM
sbi->ll_flags |= LL_SBI_CHECKSUM;
#endif
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);
-#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);
+ cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
+ pp_r_hist.oh_lock);
+ cfs_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;
+ atomic_set(&sbi->ll_sa_total, 0);
+ atomic_set(&sbi->ll_sa_wrong, 0);
RETURN(sbi);
-
-out:
- if (sbi->ll_async_page_sample)
- OBD_FREE(sbi->ll_async_page_sample,
- sizeof(long) * num_possible_cpus());
- ll_pglist_fini(sbi);
- OBD_FREE(sbi, sizeof(*sbi));
- RETURN(NULL);
}
void ll_free_sbi(struct super_block *sb)
ENTRY;
if (sbi != NULL) {
- ll_pglist_fini(sbi);
- spin_lock(&ll_sb_lock);
- list_del(&sbi->ll_list);
- spin_unlock(&ll_sb_lock);
- lcounter_destroy(&sbi->ll_async_page_count);
- OBD_FREE(sbi->ll_async_page_sample,
- sizeof(long) * num_possible_cpus());
+ cfs_spin_lock(&ll_sb_lock);
+ cfs_list_del(&sbi->ll_list);
+ cfs_spin_unlock(&ll_sb_lock);
OBD_FREE(sbi, sizeof(*sbi));
}
EXIT;
}
static struct dentry_operations ll_d_root_ops = {
-#ifdef DCACHE_LUSTRE_INVALID
.d_compare = ll_dcompare,
-#endif
+ .d_revalidate = ll_revalidate_nd,
};
-/* 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, sizeof(KEY_LOVDESC), KEY_LOVDESC,
- &valsize, &desc, NULL);
- 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 lu_fid rootfid;
struct obd_capa *oc = NULL;
struct obd_statfs osfs;
struct ptlrpc_request *request = NULL;
- struct lustre_handle dt_conn = {0, };
- struct lustre_handle md_conn = {0, };
struct obd_connect_data *data = NULL;
+ struct obd_uuid *uuid;
+ struct md_op_data *op_data;
struct lustre_md lmd;
obd_valid valid;
int size, err, checksum;
OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
- OBD_CONNECT_FID | OBD_CONNECT_AT;
+ OBD_CONNECT_FID | OBD_CONNECT_AT |
+ OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
+ OBD_CONNECT_VBR | OBD_CONNECT_FULL20;
+
+ if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
+ data->ocd_connect_flags |= OBD_CONNECT_SOM;
#ifdef HAVE_LRU_RESIZE_SUPPORT
if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
/* 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;
- }
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
+ data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
- err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
+ err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
if (err == -EBUSY) {
LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
"recovery, of which this client is not a "
CERROR("cannot connect to %s: rc = %d\n", md, err);
GOTO(out, err);
}
- sbi->ll_md_exp = class_conn2export(&md_conn);
err = obd_fid_init(sbi->ll_md_exp);
if (err) {
GOTO(out_md, err);
}
- err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
+ err = obd_statfs(obd, &osfs,
+ cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
if (err)
GOTO(out_md_fid, err);
*
* it will check if *ppos is greater than max. However, max equals to
* s_maxbytes, which is a negative integer in a x86_64 box since loff_t
- * has been defined as a signed long long ineger in linux kernel. */
+ * has been defined as a signed long long integer in linux kernel. */
#if BITS_PER_LONG == 64
sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
#else
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;
+ if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
+ if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
+ sbi->ll_flags |= LL_SBI_RMT_CLIENT;
+ LCONSOLE_INFO("client is set as remote by default.\n");
}
} 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 (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
+ sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
+ LCONSOLE_INFO("client claims to be remote, but server "
+ "rejected, forced to be local.\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(dt);
if (!obd) {
CERROR("DT %s: not setup or attached\n", dt);
OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
- OBD_CONNECT_AT;
- if (sbi->ll_flags & LL_SBI_OSS_CAPA)
- data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
+ OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
+ OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
+ OBD_CONNECT_FULL20;
+
+ if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
+ data->ocd_connect_flags |= OBD_CONNECT_SOM;
if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
/* OBD_CONNECT_CKSUM should always be set, even if checksums are
#ifdef HAVE_LRU_RESIZE_SUPPORT
data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
#endif
+ if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
+ data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
+
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_owner = &sbi->ll_lco;
- obd->obd_upcall.onu_upcall = ll_ocd_update;
+ obd->obd_upcall.onu_upcall = cl_ocd_update;
data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
- err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
+ err = obd_connect(NULL, &sbi->ll_dt_exp, 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 "
GOTO(out_md_fid, err);
}
- 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, "
GOTO(out_dt, err);
}
- spin_lock(&sbi->ll_lco.lco_lock);
+ cfs_mutex_down(&sbi->ll_lco.lco_lock);
sbi->ll_lco.lco_flags = data->ocd_connect_flags;
- spin_unlock(&sbi->ll_lco.lco_lock);
-
- err = obd_register_page_removal_cb(sbi->ll_dt_exp,
- ll_page_removal_cb,
- ll_pin_extent_cb);
- if (err) {
- CERROR("cannot register page removal callback: rc = %d\n",err);
- GOTO(out_dt, err);
- }
- err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
- ll_extent_lock_cancel_cb);
- if (err) {
- CERROR("cannot register lock cancel callback: rc = %d\n", err);
- GOTO(out_page_rm_cb, err);
- }
-
- err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
- if (err) {
- CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
- GOTO(out_lock_cn_cb, err);
- }
+ sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
+ sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
+ cfs_mutex_up(&sbi->ll_lco.lco_lock);
- err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
- 0, NULL, NULL, NULL, 0, NULL);
- if (err < 0) {
- 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_lock_cn_cb, err);
- }
-
- if (!ll_async_page_slab) {
- ll_async_page_slab_size =
- size_round(sizeof(struct ll_async_page)) + err;
- 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_lock_cn_cb, err = -ENOMEM);
- }
-
- err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
+ fid_zero(&sbi->ll_root_fid);
+ err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
if (err) {
CERROR("cannot mds_connect: rc = %d\n", err);
GOTO(out_lock_cn_cb, err);
}
- CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
- sbi->ll_root_fid = rootfid;
+ if (!fid_is_sane(&sbi->ll_root_fid)) {
+ CERROR("Invalid root fid during mount\n");
+ GOTO(out_lock_cn_cb, err = -EINVAL);
+ }
+ CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
sb->s_op = &lustre_super_operations;
+#if THREAD_SIZE >= 8192
sb->s_export_op = &lustre_export_operations;
+#endif
/* make root inode
* XXX: move this to after cbd setup? */
else if (sbi->ll_flags & LL_SBI_ACL)
valid |= OBD_MD_FLACL;
- err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
+ OBD_ALLOC_PTR(op_data);
+ if (op_data == NULL)
+ GOTO(out_lock_cn_cb, err = -ENOMEM);
+
+ op_data->op_fid1 = sbi->ll_root_fid;
+ op_data->op_mode = 0;
+ op_data->op_capa1 = oc;
+ op_data->op_valid = valid;
+
+ err = md_getattr(sbi->ll_md_exp, op_data, &request);
if (oc)
- free_capa(oc);
+ capa_put(oc);
+ OBD_FREE_PTR(op_data);
if (err) {
CERROR("md_getattr failed for root: rc = %d\n", err);
GOTO(out_lock_cn_cb, err);
}
LASSERT(fid_is_sane(&sbi->ll_root_fid));
- root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
+ root = ll_iget(sb, cl_fid_build_ino(&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)) {
+ if (root == NULL || IS_ERR(root)) {
if (lmd.lsm)
obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
#ifdef CONFIG_FS_POSIX_ACL
lmd.posix_acl = NULL;
}
#endif
+ err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
+ root = NULL;
CERROR("lustre_lite: bad iget4 for root\n");
- GOTO(out_root, err = -EBADF);
+ GOTO(out_root, err);
}
err = ll_close_thread_start(&sbi->ll_lcq);
err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
KEY_CHECKSUM, sizeof(checksum), &checksum,
NULL);
+ cl_sb_init(sb);
sb->s_root = d_alloc_root(root);
if (data != NULL)
OBD_FREE(data, sizeof(*data));
+
sb->s_root->d_op = &ll_d_root_ops;
+
+ sbi->ll_sdev_orig = sb->s_dev;
+
+ /* 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. */
+ uuid = obd_get_uuid(sbi->ll_md_exp);
+ if (uuid != NULL)
+ sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
+
RETURN(err);
out_root:
if (root)
iput(root);
out_lock_cn_cb:
- obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
- ll_extent_lock_cancel_cb);
-out_page_rm_cb:
- obd_unregister_page_removal_cb(sbi->ll_dt_exp,
- ll_page_removal_cb);
obd_fid_fini(sbi->ll_dt_exp);
out_dt:
obd_disconnect(sbi->ll_dt_exp);
if (recur == 0)
return;
- list_for_each(tmp, &dentry->d_subdirs) {
+ list_for_each(tmp, &dentry->d_subdirs) {
struct dentry *d = list_entry(tmp, struct dentry, d_child);
lustre_dump_dentry(d, recur - 1);
}
}
-#ifdef HAVE_EXPORT___IGET
-static void prune_dir_dentries(struct inode *inode)
-{
- struct dentry *dentry, *prev = NULL;
-
- /* due to lustre specific logic, a directory
- * can have few dentries - a bug from VFS POV */
-restart:
- spin_lock(&dcache_lock);
- if (!list_empty(&inode->i_dentry)) {
- dentry = list_entry(inode->i_dentry.prev,
- struct dentry, d_alias);
- /* in order to prevent infinite loops we
- * break if previous dentry is busy */
- if (dentry != prev) {
- prev = dentry;
- dget_locked(dentry);
- spin_unlock(&dcache_lock);
-
- /* try to kill all child dentries */
- lock_dentry(dentry);
- shrink_dcache_parent(dentry);
- unlock_dentry(dentry);
- dput(dentry);
-
- /* now try to get rid of current dentry */
- d_prune_aliases(inode);
- goto restart;
- }
- }
- spin_unlock(&dcache_lock);
-}
-
-static void prune_deathrow_one(struct ll_inode_info *lli)
-{
- struct inode *inode = ll_info2i(lli);
-
- /* first, try to drop any dentries - they hold a ref on the inode */
- if (S_ISDIR(inode->i_mode))
- prune_dir_dentries(inode);
- else
- d_prune_aliases(inode);
-
-
- /* if somebody still uses it, leave it */
- LASSERT(atomic_read(&inode->i_count) > 0);
- if (atomic_read(&inode->i_count) > 1)
- goto out;
-
- CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
- inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
-
- /* seems nobody uses it anymore */
- inode->i_nlink = 0;
-
-out:
- iput(inode);
- return;
-}
-
-static void prune_deathrow(struct ll_sb_info *sbi, int try)
-{
- struct ll_inode_info *lli;
- int empty;
-
- do {
- if (need_resched() && try)
- break;
-
- if (try) {
- if (!spin_trylock(&sbi->ll_deathrow_lock))
- break;
- } else {
- spin_lock(&sbi->ll_deathrow_lock);
- }
-
- empty = 1;
- lli = NULL;
- if (!list_empty(&sbi->ll_deathrow)) {
- lli = list_entry(sbi->ll_deathrow.next,
- struct ll_inode_info,
- lli_dead_list);
- list_del_init(&lli->lli_dead_list);
- if (!list_empty(&sbi->ll_deathrow))
- empty = 0;
- }
- spin_unlock(&sbi->ll_deathrow_lock);
-
- if (lli)
- prune_deathrow_one(lli);
-
- } 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);
ll_close_thread_shutdown(sbi->ll_lcq);
- /* destroy inodes in deathrow */
- prune_deathrow(sbi, 0);
+ cl_sb_fini(sb);
- list_del(&sbi->ll_conn_chain);
+ cfs_list_del(&sbi->ll_conn_chain);
obd_fid_fini(sbi->ll_dt_exp);
obd_disconnect(sbi->ll_dt_exp);
*flags &= ~tmp;
goto next;
}
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
if (tmp) {
/* Ignore deprecated mount option. The client will
* always try to mount with ACL support, whether this
* is used depends on whether server supports it. */
+ LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
+ "mount option 'acl'.\n");
goto next;
}
tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
if (tmp) {
+ LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
+ "mount option 'noacl'.\n");
goto next;
}
+#else
+#warning "{no}acl options have been deprecated since 1.8, please remove them"
+#endif
tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
if (tmp) {
*flags |= tmp;
*flags &= ~tmp;
goto next;
}
+ tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
+ if (tmp) {
+ *flags |= tmp;
+ goto next;
+ }
+ tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
+ if (tmp) {
+ *flags &= ~tmp;
+ goto next;
+ }
+ tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
+ if (tmp) {
+ *flags |= tmp;
+ goto next;
+ }
+ tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
+ if (tmp) {
+ *flags |= tmp;
+ goto next;
+ }
LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
s1);
void ll_lli_init(struct ll_inode_info *lli)
{
lli->lli_inode_magic = LLI_INODE_MAGIC;
- sema_init(&lli->lli_size_sem, 1);
- sema_init(&lli->lli_write_sem, 1);
+ cfs_sema_init(&lli->lli_size_sem, 1);
+ cfs_sema_init(&lli->lli_write_sem, 1);
+ cfs_init_rwsem(&lli->lli_trunc_sem);
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);
+ cfs_spin_lock_init(&lli->lli_lock);
+ CFS_INIT_LIST_HEAD(&lli->lli_close_list);
lli->lli_inode_magic = LLI_INODE_MAGIC;
- sema_init(&lli->lli_och_sem, 1);
+ cfs_sema_init(&lli->lli_och_sem, 1);
lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
lli->lli_mds_exec_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);
+ CFS_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);
+ cfs_sema_init(&lli->lli_rmtperm_sem, 1);
+ CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
+ cfs_spin_lock_init(&lli->lli_sa_lock);
}
+#ifdef HAVE_NEW_BACKING_DEV_INFO
+static atomic_t ll_bdi_num = ATOMIC_INIT(0);
+#endif
+
int ll_fill_super(struct super_block *sb)
{
struct lustre_profile *lprof;
/* client additional sb info */
lsi->lsi_llsbi = sbi = ll_init_sbi();
if (!sbi) {
- cfs_module_put();
+ cfs_module_put(THIS_MODULE);
RETURN(-ENOMEM);
}
if (err)
GOTO(out_free, err);
+ err = ll_bdi_init(&lsi->bdi);
+ if (err)
+ GOTO(out_free, err);
+
+#ifdef HAVE_NEW_BACKING_DEV_INFO
+ lsi->bdi.name = "lustre";
+ lsi->bdi.capabilities = BDI_CAP_MAP_COPY;
+ err = bdi_register(&lsi->bdi, NULL, "lustre-%d",
+ atomic_inc_return(&ll_bdi_num));
+ sb->s_bdi = &lsi->bdi;
+#endif
+
/* Generate a string unique to this super, in case some joker tries
to mount the same fs at two mount points.
Use the address of the super itself.*/
GOTO(out_free, err);
}
+ /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
lprof = class_get_profile(profilenm);
if (lprof == NULL) {
LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
} /* ll_fill_super */
+void lu_context_keys_dump(void);
+
void ll_put_super(struct super_block *sb)
{
struct config_llog_instance cfg;
/* 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)
class_del_profile(profilenm);
+ if (ll_bdi_wb_cnt(lsi->bdi) > 0)
+ ll_bdi_destroy(&lsi->bdi);
+
ll_free_sbi(sb);
lsi->lsi_llsbi = NULL;
lustre_common_put_super(sb);
+ cl_env_cache_purge(~0);
+
LCONSOLE_WARN("client %s umount complete\n", ll_instance);
- cfs_module_put();
+ cfs_module_put(THIS_MODULE);
EXIT;
} /* client_put_super */
-#if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
-
-#if defined(HAVE_CACHE_RETURN_INT)
-static int
-#else
-static void
-#endif
-ll_shrink_cache(int priority, unsigned int gfp_mask)
-{
- struct ll_sb_info *sbi;
- int count = 0;
-
- list_for_each_entry(sbi, &ll_super_blocks, ll_list)
- count += llap_shrink_cache(sbi, priority);
-
-#if defined(HAVE_CACHE_RETURN_INT)
- return count;
-#endif
-}
-
-struct cache_definition ll_cache_definition = {
- .name = "llap_cache",
- .shrink = ll_shrink_cache
-};
-#endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
-
struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
{
struct inode *inode = NULL;
inode = igrab(lock->l_ast_data);
} else {
inode = lock->l_ast_data;
- ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
- D_INFO : D_WARNING,
- lock, __FILE__, __func__, __LINE__,
- "l_ast_data %p is bogus: magic %08x",
- lock->l_ast_data, lli->lli_inode_magic);
+ LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
+ D_WARNING, lock, "l_ast_data %p is "
+ "bogus: magic %08x", lock->l_ast_data,
+ lli->lli_inode_magic);
inode = NULL;
}
}
if (lli->lli_mds_read_och)
ll_md_real_close(inode, FMODE_READ);
- if (lli->lli_smd) {
- obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
- null_if_equal, inode);
-
- obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
- lli->lli_smd = NULL;
- }
-
if (lli->lli_symlink_name) {
OBD_FREE(lli->lli_symlink_name,
strlen(lli->lli_symlink_name) + 1);
}
#ifdef CONFIG_FS_POSIX_ACL
else if (lli->lli_posix_acl) {
- LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
+ LASSERT(cfs_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);
+ /*
+ * XXX This has to be done before lsm is freed below, because
+ * cl_object still uses inode lsm.
+ */
+ cl_inode_fini(inode);
+
+ if (lli->lli_smd) {
+ obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
+ lli->lli_smd = NULL;
+ }
+
EXIT;
}
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;
+ op_data->op_flags = MF_EPOCH_CLOSE;
+ ll_done_writing_attr(inode, op_data);
+ ll_pack_inode2opdata(inode, op_data, NULL);
+
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);
+ rc = ll_som_update(inode, op_data);
} 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)
+static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
{
- struct ll_sb_info *sbi = ll_i2sbi(inode);
- struct ll_inode_info *lli = ll_i2info(inode);
- struct lov_stripe_md *lsm = lli->lli_smd;
+ struct obd_capa *capa;
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 (attr->ia_valid & ATTR_SIZE)
+ capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
+ else
+ capa = ll_mdscapa_get(inode);
- LOCK_INODE_MUTEX(inode);
- DOWN_WRITE_I_ALLOC_SEM(inode);
-
- 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);
- }
- }
+ rc = cl_setattr_ost(inode, attr, capa);
- 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 (attr->ia_valid & ATTR_SIZE)
+ ll_truncate_free_capa(capa);
+ else
+ capa_put(capa);
+
+ return rc;
}
/* If this inode has objects allocated to it (lsm != NULL), then the OST
{
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 md_op_data *op_data = NULL;
struct md_open_data *mod = NULL;
int ia_valid = attr->ia_valid;
/* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
- if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
+ if (cfs_curproc_fsuid() != inode->i_uid &&
+ !cfs_capable(CFS_CAP_FOWNER))
RETURN(-EPERM);
}
/* We mark all of the fields "set" so MDS/OST does not re-set them */
if (attr->ia_valid & ATTR_CTIME) {
- attr->ia_ctime = CURRENT_TIME;
+ attr->ia_ctime = CFS_CURRENT_TIME;
attr->ia_valid |= ATTR_CTIME_SET;
}
if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
- attr->ia_atime = CURRENT_TIME;
+ attr->ia_atime = CFS_CURRENT_TIME;
attr->ia_valid |= ATTR_ATIME_SET;
}
if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
- attr->ia_mtime = CURRENT_TIME;
+ attr->ia_mtime = CFS_CURRENT_TIME;
attr->ia_valid |= ATTR_MTIME_SET;
}
- if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
- /* To avoid stale mtime on mds, obtain it from ost and send
- to mds. */
- rc = ll_glimpse_size(inode, 0);
- 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",
if (op_data == NULL)
RETURN(-ENOMEM);
+ UNLOCK_INODE_MUTEX(inode);
+ if (ia_valid & ATTR_SIZE)
+ UP_WRITE_I_ALLOC_SEM(inode);
+ cfs_down_write(&lli->lli_trunc_sem);
+ LOCK_INODE_MUTEX(inode);
+ if (ia_valid & ATTR_SIZE)
+ DOWN_WRITE_I_ALLOC_SEM(inode);
+
memcpy(&op_data->op_attr, attr, sizeof(*attr));
/* Open epoch for truncate. */
- if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
- (ia_valid & ATTR_SIZE))
+ if (exp_connect_som(ll_i2mdexp(inode)) &&
+ (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
op_data->op_flags = MF_EPOCH_OPEN;
rc = ll_md_setattr(inode, op_data, &mod);
if (rc)
GOTO(out, rc);
- if (op_data->op_ioepoch)
- CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
- "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
-
+ ll_ioepoch_open(lli, op_data->op_ioepoch);
if (!lsm || !S_ISREG(inode->i_mode)) {
CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
GOTO(out, rc = 0);
}
- /* We really need to get our PW lock before we change inode->i_size.
- * If we don't we can race with other i_size updaters on our node, like
- * ll_file_read. We can also race with i_size propogation to other
- * nodes through dirtying and writeback of final cached pages. This
- * last one is especially bad for racing o_append users on other
- * nodes. */
- if (ia_valid & ATTR_SIZE) {
- 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;
-
- 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_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_FLGROUP;
-
- obdo_from_inode(oa, inode, flags);
-
- oinfo.oi_oa = oa;
- oinfo.oi_md = lsm;
- oinfo.oi_capa = ll_mdscapa_get(inode);
-
- /* 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);
- } else {
- rc = -ENOMEM;
- }
- }
+ if (ia_valid & ATTR_SIZE)
+ attr->ia_valid |= ATTR_SIZE;
+ if (ia_valid & (ATTR_SIZE |
+ ATTR_ATIME | ATTR_ATIME_SET |
+ ATTR_MTIME | ATTR_MTIME_SET))
+ /* on truncate and utimes send attributes to osts, setting
+ * mtime/atime to past will be performed under PW 0:EOF extent
+ * lock (new_size:EOF for truncate)
+ * it may seem excessive to send mtime/atime updates to osts
+ * when not setting times to past, but it is necessary due to
+ * possible time de-synchronization */
+ rc = ll_setattr_ost(inode, attr);
EXIT;
out:
if (op_data) {
rc1 = ll_setattr_done_writing(inode, op_data, mod);
ll_finish_md_op_data(op_data);
}
+ cfs_up_write(&lli->lli_trunc_sem);
return rc ? rc : rc1;
}
(ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
+ if ((de->d_inode->i_mode & S_ISUID) &&
+ !(attr->ia_mode & S_ISUID) &&
+ !(attr->ia_valid & ATTR_KILL_SUID))
+ attr->ia_valid |= ATTR_KILL_SUID;
+
+ if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
+ !(attr->ia_mode & S_ISGID) &&
+ !(attr->ia_valid & ATTR_KILL_SGID))
+ attr->ia_valid |= ATTR_KILL_SGID;
+
return ll_setattr_raw(de->d_inode, attr);
}
CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
+ if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
+ flags |= OBD_STATFS_NODELAY;
+
rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
&obd_osfs, max_age, flags);
if (rc) {
CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
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, 0);
+ /* Some amount of caching on the client is allowed */
+ rc = ll_statfs_internal(sb, &osfs,
+ cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
+ 0);
if (rc)
return rc;
lli = ll_i2info(inode);
LASSERT(lli->lli_size_sem_owner != current);
- down(&lli->lli_size_sem);
+ cfs_down(&lli->lli_size_sem);
LASSERT(lli->lli_size_sem_owner == NULL);
lli->lli_size_sem_owner = current;
lsm = lli->lli_smd;
lov_stripe_unlock(lsm);
LASSERT(lli->lli_size_sem_owner == current);
lli->lli_size_sem_owner = NULL;
- up(&lli->lli_size_sem);
-}
-
-static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
-{
- struct ll_inode_info *lli = ll_i2info(inode);
-
- dump_lsm(D_INODE, lsm);
- 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_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;
- lli->lli_maxbytes = lsm->lsm_maxbytes;
- if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
- lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
+ cfs_up(&lli->lli_size_sem);
}
void ll_update_inode(struct inode *inode, struct lustre_md *md)
LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
if (lsm != NULL) {
+ cfs_down(&lli->lli_och_sem);
if (lli->lli_smd == NULL) {
- if (lsm->lsm_magic != LOV_MAGIC &&
- lsm->lsm_magic != LOV_MAGIC_JOIN) {
+ if (lsm->lsm_magic != LOV_MAGIC_V1 &&
+ lsm->lsm_magic != LOV_MAGIC_V3) {
dump_lsm(D_ERROR, lsm);
LBUG();
}
CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
lsm, inode->i_ino, inode->i_generation, inode);
- /* ll_inode_size_lock() requires it is only called
- * with lli_smd != NULL or lock_lsm == 0 or we can
- * race between lock/unlock. bug 9547 */
+ /* cl_inode_init must go before lli_smd or a race is
+ * possible where client thinks the file has stripes,
+ * but lov raid0 is not setup yet and parallel e.g.
+ * glimpse would try to use uninitialized lov */
+ cl_inode_init(inode, md);
lli->lli_smd = lsm;
+ cfs_up(&lli->lli_och_sem);
lli->lli_maxbytes = lsm->lsm_maxbytes;
if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
} else {
- if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
- lli->lli_smd->lsm_stripe_count ==
- lsm->lsm_stripe_count) {
- if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
- CERROR("lsm mismatch for inode %ld\n",
- inode->i_ino);
- CERROR("lli_smd:\n");
- dump_lsm(D_ERROR, lli->lli_smd);
- CERROR("lsm:\n");
- dump_lsm(D_ERROR, lsm);
- LBUG();
- }
- } else
- ll_replace_lsm(inode, lsm);
+ cfs_up(&lli->lli_och_sem);
+ LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
+ lli->lli_smd->lsm_stripe_count ==
+ lsm->lsm_stripe_count);
+ if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
+ CERROR("lsm mismatch for inode %ld\n",
+ inode->i_ino);
+ CERROR("lli_smd:\n");
+ dump_lsm(D_ERROR, lli->lli_smd);
+ CERROR("lsm:\n");
+ dump_lsm(D_ERROR, lsm);
+ LBUG();
+ }
}
if (lli->lli_smd != lsm)
obd_free_memmd(ll_i2dtexp(inode), &lsm);
}
#ifdef CONFIG_FS_POSIX_ACL
else if (body->valid & OBD_MD_FLACL) {
- spin_lock(&lli->lli_lock);
+ cfs_spin_lock(&lli->lli_lock);
if (lli->lli_posix_acl)
posix_acl_release(lli->lli_posix_acl);
lli->lli_posix_acl = md->posix_acl;
- spin_unlock(&lli->lli_lock);
+ cfs_spin_unlock(&lli->lli_lock);
}
#endif
- inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
-
- if (body->valid & OBD_MD_FLATIME &&
- body->atime > LTIME_S(inode->i_atime))
- LTIME_S(inode->i_atime) = body->atime;
-
- /* mtime is always updated with ctime, but can be set in past.
- As write and utime(2) may happen within 1 second, and utime's
- mtime has a priority over write's one, so take mtime from mds
- for the same ctimes. */
- if (body->valid & OBD_MD_FLCTIME &&
- body->ctime >= LTIME_S(inode->i_ctime)) {
- LTIME_S(inode->i_ctime) = body->ctime;
- if (body->valid & OBD_MD_FLMTIME) {
- CDEBUG(D_INODE, "setting ino %lu mtime "
- "from %lu to "LPU64"\n", inode->i_ino,
+ inode->i_ino = cl_fid_build_ino(&body->fid1);
+ inode->i_generation = cl_fid_build_gen(&body->fid1);
+
+ if (body->valid & OBD_MD_FLATIME) {
+ if (body->atime > LTIME_S(inode->i_atime))
+ LTIME_S(inode->i_atime) = body->atime;
+ lli->lli_lvb.lvb_atime = body->atime;
+ }
+ if (body->valid & OBD_MD_FLMTIME) {
+ if (body->mtime > LTIME_S(inode->i_mtime)) {
+ CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
+ "to "LPU64"\n", inode->i_ino,
LTIME_S(inode->i_mtime), body->mtime);
LTIME_S(inode->i_mtime) = body->mtime;
}
+ lli->lli_lvb.lvb_mtime = body->mtime;
+ }
+ if (body->valid & OBD_MD_FLCTIME) {
+ if (body->ctime > LTIME_S(inode->i_ctime))
+ LTIME_S(inode->i_ctime) = body->ctime;
+ lli->lli_lvb.lvb_ctime = body->ctime;
}
if (body->valid & OBD_MD_FLMODE)
inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
if (body->valid & OBD_MD_FLTYPE)
inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
+ LASSERT(inode->i_mode != 0);
if (S_ISREG(inode->i_mode)) {
inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
} else {
LASSERT(fid_seq(&lli->lli_fid) != 0);
if (body->valid & OBD_MD_FLSIZE) {
- if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
+ if (exp_connect_som(ll_i2mdexp(inode)) &&
S_ISREG(inode->i_mode) && lli->lli_smd) {
struct lustre_handle lockh;
ldlm_mode_t mode;
/* Use old size assignment to avoid
* deadlock bz14138 & bz14326 */
inode->i_size = body->size;
+
+ CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
+ inode->i_ino, (unsigned long long)body->size);
}
if (body->valid & OBD_MD_FLBLOCKS)
}
}
-static struct backing_dev_info ll_backing_dev_info = {
- .ra_pages = 0, /* No readahead */
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
- .capabilities = 0, /* Does contribute to dirty memory */
-#else
- .memory_backed = 0, /* Does contribute to dirty memory */
-#endif
-};
-
void ll_read_inode2(struct inode *inode, void *opaque)
{
struct lustre_md *md = opaque;
/* OIDEBUG(inode); */
+ /* initializing backing dev info. */
+ inode->i_mapping->backing_dev_info = &(s2lsi(inode->i_sb)->bdi);
+
+
if (S_ISREG(inode->i_mode)) {
struct ll_sb_info *sbi = ll_i2sbi(inode);
inode->i_op = &ll_file_inode_operations;
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;
-
EXIT;
}
}
ENTRY;
switch(cmd) {
- case EXT3_IOC_GETFLAGS: {
+ case FSFILT_IOC_GETFLAGS: {
struct mdt_body *body;
- struct obd_capa *oc;
+ struct md_op_data *op_data;
- oc = ll_mdscapa_get(inode);
- rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
- OBD_MD_FLFLAGS, 0, &req);
- capa_put(oc);
+ op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
+ 0, 0, LUSTRE_OPC_ANY,
+ NULL);
+ if (op_data == NULL)
+ RETURN(-ENOMEM);
+
+ op_data->op_valid = OBD_MD_FLFLAGS;
+ rc = md_getattr(sbi->ll_md_exp, op_data, &req);
+ ll_finish_md_op_data(op_data);
if (rc) {
CERROR("failure %d inode %lu\n", rc, inode->i_ino);
RETURN(-abs(rc));
RETURN(put_user(flags, (int *)arg));
}
- case EXT3_IOC_SETFLAGS: {
+ case FSFILT_IOC_SETFLAGS: {
struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
struct obd_info oinfo = { { { 0 } } };
struct md_op_data *op_data;
NULL, 0, NULL, 0, &req, NULL);
ll_finish_md_op_data(op_data);
ptlrpc_req_finished(req);
- if (rc || lsm == NULL) {
+ if (rc) {
OBDO_FREE(oinfo.oi_oa);
RETURN(rc);
}
+ if (lsm == NULL) {
+ OBDO_FREE(oinfo.oi_oa);
+ GOTO(update_cache, rc);
+ }
+
oinfo.oi_oa->o_id = lsm->lsm_object_id;
- oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
+ oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
oinfo.oi_oa->o_flags = flags;
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);
+ &ll_i2info(inode)->lli_fid, 0);
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("md_setattr_async fails: rc = %d\n", rc);
+ CERROR("osc_setattr_async fails: rc = %d\n",rc);
RETURN(rc);
}
- inode->i_flags = ll_ext_to_inode_flags(flags |
- MDS_BFLAG_EXT_FLAGS);
- RETURN(0);
+ EXIT;
+update_cache:
+ inode->i_flags = ll_ext_to_inode_flags(flags);
+ return 0;
}
default:
RETURN(-ENOSYS);
{
struct ll_sb_info *sbi = ll_i2sbi(inode);
- CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
+ CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
obd_set_info_async(sbi->ll_md_exp,
sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
struct lustre_sb_info *lsi = s2lsi(sb);
struct ll_sb_info *sbi = ll_s2sbi(sb);
struct obd_device *obd;
- struct obd_ioctl_data ioc_data = { 0 };
+ struct obd_ioctl_data *ioc_data;
ENTRY;
#ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
return;
}
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_dt_exp);
if (obd == NULL) {
EXIT;
return;
}
-
obd->obd_force = 1;
- obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
- &ioc_data, NULL);
+
+ OBD_ALLOC_PTR(ioc_data);
+ if (ioc_data) {
+ obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
+ sizeof ioc_data, ioc_data, NULL);
+
+ obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
+ sizeof ioc_data, ioc_data, NULL);
+
+ OBD_FREE_PTR(ioc_data);
+ }
+
/* Really, we'd like to wait until there are no requests outstanding,
* and then continue. For now, we just invalidate the requests,
- * schedule, and hope.
+ * schedule() and sleep one second if needed, and hope.
*/
- schedule();
+ cfs_schedule();
+#ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
+ if (atomic_read(&vfsmnt->mnt_count) > 2) {
+ cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
+ cfs_time_seconds(1));
+ if (atomic_read(&vfsmnt->mnt_count) > 2)
+ LCONSOLE_WARN("Mount still busy with %d refs! You "
+ "may try to umount it a bit later\n",
+ atomic_read(&vfsmnt->mnt_count));
+ }
+#endif
EXIT;
}
{
struct ll_sb_info *sbi = NULL;
struct lustre_md md;
- int rc = 0;
+ int rc;
ENTRY;
LASSERT(*inode || sb);
sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
- prune_deathrow(sbi, 1);
memset(&md, 0, sizeof(struct lustre_md));
rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
*/
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)) {
+ *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
+ if (*inode == NULL || IS_ERR(*inode)) {
if (md.lsm)
obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
#ifdef CONFIG_FS_POSIX_ACL
md.posix_acl = NULL;
}
#endif
- rc = -ENOMEM;
+ rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
+ *inode = NULL;
CERROR("new_inode -fatal: rc %d\n", rc);
GOTO(out, rc);
}
}
- 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);
}
-char *llap_origins[] = {
- [LLAP_ORIGIN_UNKNOWN] = "--",
- [LLAP_ORIGIN_READPAGE] = "rp",
- [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 list_head *head,
- struct list_head *list)
-{
- struct ll_async_page *llap;
- struct list_head *pos;
-
- list_for_each(pos, list) {
- if (pos == head)
- return NULL;
- llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
- if (llap->llap_page == NULL)
- continue;
- return llap;
- }
- LBUG();
- return NULL;
-}
-
int ll_obd_statfs(struct inode *inode, void *arg)
{
struct ll_sb_info *sbi = NULL;
!data->ioc_pbuf1 || !data->ioc_pbuf2)
GOTO(out_statfs, rc = -EINVAL);
+ if (data->ioc_inllen1 != sizeof(__u32) ||
+ data->ioc_inllen2 != sizeof(__u32) ||
+ data->ioc_plen1 != sizeof(struct obd_statfs) ||
+ data->ioc_plen2 != sizeof(struct obd_uuid))
+ GOTO(out_statfs, rc = -EINVAL);
+
memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
if (type == LL_STATFS_MDC)
exp = sbi->ll_md_exp;
proc fns must be able to handle that! */
rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
lcfg, sb);
+ if (rc > 0)
+ rc = 0;
return(rc);
}
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_fsuid = cfs_curproc_fsuid();
+ op_data->op_fsgid = cfs_curproc_fsgid();
+ op_data->op_cap = cfs_curproc_cap_pack();
op_data->op_bias = MDS_CHECK_SPLIT;
op_data->op_opc = opc;
op_data->op_mds = 0;
if (sbi->ll_flags & LL_SBI_USER_XATTR)
seq_puts(seq, ",user_xattr");
- if (sbi->ll_flags & LL_SBI_ACL)
- seq_puts(seq, ",acl");
+ if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
+ seq_puts(seq, ",lazystatfs");
RETURN(0);
}