Whamcloud - gitweb
LU-10698 obdclass: allow specifying complex jobids
[fs/lustre-release.git] / lustre / llite / llite_lib.c
index c6374c3..ccd8863 100644 (file)
@@ -23,7 +23,7 @@
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2016, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #endif
 #include <linux/security.h>
 
-#include <uapi/linux/lustre_ioctl.h>
+#include <uapi/linux/lustre/lustre_ioctl.h>
 #include <lustre_ha.h>
 #include <lustre_dlm.h>
 #include <lprocfs_status.h>
 #include <lustre_disk.h>
-#include <uapi/linux/lustre_param.h>
+#include <uapi/linux/lustre/lustre_param.h>
 #include <lustre_log.h>
 #include <cl_object.h>
 #include <obd_cksum.h>
@@ -99,8 +99,7 @@ static struct ll_sb_info *ll_init_sbi(void)
        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_ra_info.ra_max_read_ahead_whole_pages = -1;
 
         ll_generate_random_uuid(uuid);
         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
@@ -131,6 +130,7 @@ static struct ll_sb_info *ll_init_sbi(void)
        atomic_set(&sbi->ll_agl_total, 0);
        sbi->ll_flags |= LL_SBI_AGL_ENABLED;
        sbi->ll_flags |= LL_SBI_FAST_READ;
+       sbi->ll_flags |= LL_SBI_TINY_WRITE;
 
        /* root squash */
        sbi->ll_squash.rsi_uid = 0;
@@ -196,13 +196,18 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                 RETURN(-ENOMEM);
         }
 
-        /* indicate the features supported by this client */
-        data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
-                                  OBD_CONNECT_ATTRFID  |
-                                  OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
-                                  OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
-                                  OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
-                                  OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
+       /* pass client page size via ocd_grant_blkbits, the server should report
+        * back its backend blocksize for grant calculation purpose */
+       data->ocd_grant_blkbits = PAGE_SHIFT;
+
+       /* indicate MDT features supported by this client */
+       data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
+                                 OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
+                                 OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
+                                 OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
+                                 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
+                                 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
+                                 OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
                                  OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
                                  OBD_CONNECT_64BITHASH |
                                  OBD_CONNECT_EINPROGRESS |
@@ -213,11 +218,13 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                                  OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
                                  OBD_CONNECT_OPEN_BY_FID |
                                  OBD_CONNECT_DIR_STRIPE |
-                                 OBD_CONNECT_BULK_MBITS |
+                                 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
                                  OBD_CONNECT_SUBTREE |
-                                 OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
+                                 OBD_CONNECT_MULTIMODRPCS |
+                                 OBD_CONNECT_GRANT_PARAM |
+                                 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 
-       data->ocd_connect_flags2 = 0;
+       data->ocd_connect_flags2 = OBD_CONNECT2_FLR;
 
 #ifdef HAVE_LRU_RESIZE_SUPPORT
         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
@@ -228,6 +235,8 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                                   OBD_CONNECT_LARGE_ACL;
 #endif
 
+       data->ocd_cksum_types = cksum_types_supported_client();
+
        if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
                /* flag mdc connection as lightweight, only used for test
                 * purpose, use with care */
@@ -259,7 +268,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
                data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 
-#ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
+#if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
        data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 
@@ -291,7 +300,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
         * can make sure the client can be mounted as long as MDT0 is
         * avaible */
        err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
-                       cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
+                       ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
                        OBD_STATFS_FOR_MDT0);
        if (err)
                GOTO(out_md_fid, err);
@@ -388,6 +397,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
         * back its backend blocksize for grant calculation purpose */
        data->ocd_grant_blkbits = PAGE_SHIFT;
 
+       /* indicate OST features supported by this client */
        data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
                                  OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
                                  OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
@@ -399,9 +409,26 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                                  OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
                                  OBD_CONNECT_LAYOUTLOCK |
                                  OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
-                                 OBD_CONNECT_BULK_MBITS;
+                                 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
+                                 OBD_CONNECT_FLAGS2;
+
+/* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
+ * can interoperate with an older version of lockahead which was released prior
+ * to landing in master. This support will be dropped when 2.13 development
+ * starts.  At the point, we should not just drop the connect flag (below), we
+ * should also remove the support in the code.
+ *
+ * Removing it means a few things:
+ * 1. Remove this section here
+ * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
+ * 3. Remove function exp_connect_lockahead_old
+ * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
+ * */
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
+       data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
+#endif
 
-       data->ocd_connect_flags2 = 0;
+       data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 
        if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
                data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
@@ -449,6 +476,12 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 
        sbi->ll_dt_exp->exp_connect_data = *data;
 
+       /* Don't change value if it was specified in the config log */
+       if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
+               sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
+                       max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
+                             (data->ocd_brw_size >> PAGE_SHIFT));
+
        err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
                           LUSTRE_SEQ_METADATA);
        if (err) {
@@ -916,29 +949,56 @@ void ll_lli_init(struct ll_inode_info *lli)
                init_rwsem(&lli->lli_trunc_sem);
                range_lock_tree_init(&lli->lli_write_tree);
                init_rwsem(&lli->lli_glimpse_sem);
-               lli->lli_glimpse_time = 0;
+               lli->lli_glimpse_time = ktime_set(0, 0);
                INIT_LIST_HEAD(&lli->lli_agl_list);
                lli->lli_agl_index = 0;
                lli->lli_async_rc = 0;
        }
        mutex_init(&lli->lli_layout_mutex);
-       memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
+       memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 }
 
-static inline int ll_bdi_register(struct backing_dev_info *bdi)
+#ifndef HAVE_SUPER_SETUP_BDI_NAME
+
+#define LSI_BDI_INITIALIZED    0x00400000
+
+#ifndef HAVE_BDI_CAP_MAP_COPY
+# define BDI_CAP_MAP_COPY      0
+#endif
+
+#define MAX_STRING_SIZE 128
+
+static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 {
-       static atomic_t ll_bdi_num = ATOMIC_INIT(0);
+       struct  lustre_sb_info *lsi = s2lsi(sb);
+       char buf[MAX_STRING_SIZE];
+       va_list args;
+       int err;
 
-       bdi->name = "lustre";
-       return bdi_register(bdi, NULL, "lustre-%d",
-                           atomic_inc_return(&ll_bdi_num));
+       err = bdi_init(&lsi->lsi_bdi);
+       if (err)
+               return err;
+
+       lsi->lsi_flags |= LSI_BDI_INITIALIZED;
+       lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
+       lsi->lsi_bdi.name = "lustre";
+       va_start(args, fmt);
+       vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
+       va_end(args);
+       err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
+       va_end(args);
+       if (!err)
+               sb->s_bdi = &lsi->lsi_bdi;
+
+       return err;
 }
+#endif /* !HAVE_SUPER_SETUP_BDI_NAME */
 
 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 {
        struct  lustre_profile *lprof = NULL;
        struct  lustre_sb_info *lsi = s2lsi(sb);
-       struct  ll_sb_info *sbi;
+       struct  ll_sb_info *sbi = NULL;
        char    *dt = NULL, *md = NULL;
        char    *profilenm = get_profile_name(sb);
        struct config_llog_instance *cfg;
@@ -951,38 +1011,25 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 
        CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 
+       try_module_get(THIS_MODULE);
+
        OBD_ALLOC_PTR(cfg);
        if (cfg == NULL)
-               RETURN(-ENOMEM);
-
-       try_module_get(THIS_MODULE);
+               GOTO(out_free, err = -ENOMEM);
 
        /* client additional sb info */
        lsi->lsi_llsbi = sbi = ll_init_sbi();
-       if (!sbi) {
-               module_put(THIS_MODULE);
-               OBD_FREE_PTR(cfg);
-               RETURN(-ENOMEM);
-       }
+       if (!sbi)
+               GOTO(out_free, err = -ENOMEM);
 
        err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
        if (err)
                GOTO(out_free, err);
 
-       err = bdi_init(&lsi->lsi_bdi);
-       if (err)
-               GOTO(out_free, err);
-       lsi->lsi_flags |= LSI_BDI_INITIALIZED;
-#ifdef HAVE_BDI_CAP_MAP_COPY
-       lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
-#else
-       lsi->lsi_bdi.capabilities = 0;
-#endif
-       err = ll_bdi_register(&lsi->lsi_bdi);
+       err = super_setup_bdi_name(sb, "lustre-%p", sb);
        if (err)
                GOTO(out_free, err);
 
-       sb->s_bdi = &lsi->lsi_bdi;
 #ifndef HAVE_DCACHE_LOCK
        /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
        sb->s_d_op = &ll_d_ops;
@@ -1051,12 +1098,12 @@ out_free:
                OBD_FREE(dt, dt_len);
        if (lprof != NULL)
                class_put_profile(lprof);
+       if (cfg)
+               OBD_FREE_PTR(cfg);
        if (err)
                ll_put_super(sb);
        else if (sbi->ll_flags & LL_SBI_VERBOSE)
                LCONSOLE_WARN("Mounted %s\n", profilenm);
-
-       OBD_FREE_PTR(cfg);
        RETURN(err);
 } /* ll_fill_super */
 
@@ -1071,6 +1118,9 @@ void ll_put_super(struct super_block *sb)
        int next, force = 1, rc = 0;
         ENTRY;
 
+       if (!sbi)
+               GOTO(out_no_sbi, 0);
+
         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 
         cfg.cfg_instance = sb;
@@ -1124,14 +1174,16 @@ void ll_put_super(struct super_block *sb)
         if (profilenm)
                 class_del_profile(profilenm);
 
+#ifndef HAVE_SUPER_SETUP_BDI_NAME
        if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
                bdi_destroy(&lsi->lsi_bdi);
                lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
        }
+#endif
 
         ll_free_sbi(sb);
         lsi->lsi_llsbi = NULL;
-
+out_no_sbi:
        lustre_common_put_super(sb);
 
        cl_env_cache_purge(~0);
@@ -1432,6 +1484,7 @@ void ll_clear_inode(struct inode *inode)
        ll_xattr_cache_destroy(inode);
 
 #ifdef CONFIG_FS_POSIX_ACL
+       forget_all_cached_acls(inode);
        if (lli->lli_posix_acl) {
                LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
                posix_acl_release(lli->lli_posix_acl);
@@ -1570,17 +1623,17 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
         /* We mark all of the fields "set" so MDS/OST does not re-set them */
        if (!(attr->ia_valid & ATTR_CTIME_SET) &&
            (attr->ia_valid & ATTR_CTIME)) {
-               attr->ia_ctime = CURRENT_TIME;
+               attr->ia_ctime = current_time(inode);
                 attr->ia_valid |= ATTR_CTIME_SET;
         }
        if (!(attr->ia_valid & ATTR_ATIME_SET) &&
            (attr->ia_valid & ATTR_ATIME)) {
-               attr->ia_atime = CURRENT_TIME;
+               attr->ia_atime = current_time(inode);
                 attr->ia_valid |= ATTR_ATIME_SET;
         }
        if (!(attr->ia_valid & ATTR_MTIME_SET) &&
            (attr->ia_valid & ATTR_MTIME)) {
-               attr->ia_mtime = CURRENT_TIME;
+               attr->ia_mtime = current_time(inode);
                 attr->ia_valid |= ATTR_MTIME_SET;
         }
 
@@ -1721,12 +1774,15 @@ int ll_setattr(struct dentry *de, struct iattr *attr)
 }
 
 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
-                       __u64 max_age, __u32 flags)
+                      u32 flags)
 {
-        struct ll_sb_info *sbi = ll_s2sbi(sb);
-        struct obd_statfs obd_osfs;
-        int rc;
-        ENTRY;
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       struct obd_statfs obd_osfs;
+       time64_t max_age;
+       int rc;
+
+       ENTRY;
+       max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
 
         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
         if (rc) {
@@ -1742,7 +1798,7 @@ int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
                 flags |= OBD_STATFS_NODELAY;
 
-        rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
+       rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
         if (rc) {
                 CERROR("obd_statfs fails: rc = %d\n", rc);
                 RETURN(rc);
@@ -1779,12 +1835,10 @@ int ll_statfs(struct dentry *de, struct kstatfs *sfs)
        CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
 
-        /* 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;
+       /* Some amount of caching on the client is allowed */
+       rc = ll_statfs_internal(sb, &osfs, 0);
+       if (rc)
+               return rc;
 
         statfs_unpack(sfs, &osfs);
 
@@ -2466,15 +2520,14 @@ struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
        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_mod_time = ktime_get_real_seconds();
        op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
        op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
        op_data->op_cap = cfs_curproc_cap_pack();
+       op_data->op_mds = 0;
        if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
             filename_is_volatile(name, namelen, &op_data->op_mds)) {
                op_data->op_bias |= MDS_CREATE_VOLATILE;
-       } else {
-               op_data->op_mds = 0;
        }
        op_data->op_data = data;