X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdclass%2Fdt_object.c;h=3bec5f72da952fe32caa3966cf825bc91888615e;hp=a4bbb9a61e8e609d5901e507b5a55ad23604fea5;hb=907a82654245c20c83c4291ab1617c9c64340c30;hpb=90d8e7fd28746a572c8de488222f5312fe927fc3 diff --git a/lustre/obdclass/dt_object.c b/lustre/obdclass/dt_object.c index a4bbb9a..3bec5f7 100644 --- a/lustre/obdclass/dt_object.c +++ b/lustre/obdclass/dt_object.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -17,17 +15,15 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -42,238 +38,238 @@ */ #define DEBUG_SUBSYSTEM S_CLASS -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif -#include +#include +#include #include -#include /* fid_be_to_cpu() */ #include - -struct dt_find_hint { - struct lu_fid *dfh_fid; - struct dt_device *dfh_dt; - struct dt_object *dfh_o; -}; - -struct dt_thread_info { - char dti_buf[DT_MAX_PATH]; - struct lu_fid_pack dti_pack; - struct dt_find_hint dti_dfh; -}; +#include +#include +#include /* context key constructor/destructor: dt_global_key_init, dt_global_key_fini */ LU_KEY_INIT(dt_global, struct dt_thread_info); LU_KEY_FINI(dt_global, struct dt_thread_info); -static struct lu_context_key dt_key = { - .lct_tags = LCT_MD_THREAD|LCT_DT_THREAD, - .lct_init = dt_global_key_init, - .lct_fini = dt_global_key_fini +struct lu_context_key dt_key = { + .lct_tags = LCT_MD_THREAD | LCT_DT_THREAD | LCT_MG_THREAD | LCT_LOCAL, + .lct_init = dt_global_key_init, + .lct_fini = dt_global_key_fini }; -/* no lock is necessary to protect the list, because call-backs +/* + * no lock is necessary to protect the list, because call-backs * are added during system startup. Please refer to "struct dt_device". */ void dt_txn_callback_add(struct dt_device *dev, struct dt_txn_callback *cb) { - list_add(&cb->dtc_linkage, &dev->dd_txn_callbacks); + list_add(&cb->dtc_linkage, &dev->dd_txn_callbacks); } EXPORT_SYMBOL(dt_txn_callback_add); void dt_txn_callback_del(struct dt_device *dev, struct dt_txn_callback *cb) { - list_del_init(&cb->dtc_linkage); + list_del_init(&cb->dtc_linkage); } EXPORT_SYMBOL(dt_txn_callback_del); int dt_txn_hook_start(const struct lu_env *env, - struct dt_device *dev, struct txn_param *param) + struct dt_device *dev, struct thandle *th) { - int result; - struct dt_txn_callback *cb; + int rc = 0; + struct dt_txn_callback *cb; - result = 0; - list_for_each_entry(cb, &dev->dd_txn_callbacks, dtc_linkage) { - if (cb->dtc_txn_start == NULL) - continue; - result = cb->dtc_txn_start(env, param, cb->dtc_cookie); - if (result < 0) - break; - } - return result; + if (th->th_local) + return 0; + + list_for_each_entry(cb, &dev->dd_txn_callbacks, dtc_linkage) { + struct thandle *dtc_th = th; + + if (cb->dtc_txn_start == NULL || + !(cb->dtc_tag & env->le_ctx.lc_tags)) + continue; + + /* + * Usually dt_txn_hook_start is called from bottom device, + * and if the thandle has th_top, then we need use top + * thandle for the callback in the top thandle layer + */ + if (th->th_top != NULL) + dtc_th = th->th_top; + + rc = cb->dtc_txn_start(env, dtc_th, cb->dtc_cookie); + if (rc < 0) + break; + } + return rc; } EXPORT_SYMBOL(dt_txn_hook_start); -int dt_txn_hook_stop(const struct lu_env *env, struct thandle *txn) +int dt_txn_hook_stop(const struct lu_env *env, struct thandle *th) { - struct dt_device *dev = txn->th_dev; - struct dt_txn_callback *cb; - int result; + struct dt_device *dev = th->th_dev; + struct dt_txn_callback *cb; + int rc = 0; - result = 0; - list_for_each_entry(cb, &dev->dd_txn_callbacks, dtc_linkage) { - if (cb->dtc_txn_stop == NULL) - continue; - result = cb->dtc_txn_stop(env, txn, cb->dtc_cookie); - if (result < 0) - break; - } - return result; -} -EXPORT_SYMBOL(dt_txn_hook_stop); + if (th->th_local) + return 0; -int dt_txn_hook_commit(const struct lu_env *env, struct thandle *txn) -{ - struct dt_device *dev = txn->th_dev; - struct dt_txn_callback *cb; - int result; + if (OBD_FAIL_CHECK(OBD_FAIL_DT_TXN_STOP)) + return -EIO; + + list_for_each_entry(cb, &dev->dd_txn_callbacks, dtc_linkage) { + struct thandle *dtc_th = th; - result = 0; - list_for_each_entry(cb, &dev->dd_txn_callbacks, dtc_linkage) { - if (cb->dtc_txn_commit == NULL) - continue; - result = cb->dtc_txn_commit(env, txn, cb->dtc_cookie); - if (result < 0) - break; - } - return result; + if (cb->dtc_txn_stop == NULL || + !(cb->dtc_tag & env->le_ctx.lc_tags)) + continue; + + /* + * Usually dt_txn_hook_stop is called from bottom device, + * and if the thandle has th_top, then we need use top + * thandle for the callback in the top thandle layer + */ + if (th->th_top != NULL) + dtc_th = th->th_top; + + rc = cb->dtc_txn_stop(env, dtc_th, cb->dtc_cookie); + if (rc < 0) + break; + } + return rc; } -EXPORT_SYMBOL(dt_txn_hook_commit); +EXPORT_SYMBOL(dt_txn_hook_stop); int dt_device_init(struct dt_device *dev, struct lu_device_type *t) { - - CFS_INIT_LIST_HEAD(&dev->dd_txn_callbacks); - return lu_device_init(&dev->dd_lu_dev, t); + INIT_LIST_HEAD(&dev->dd_txn_callbacks); + return lu_device_init(&dev->dd_lu_dev, t); } EXPORT_SYMBOL(dt_device_init); void dt_device_fini(struct dt_device *dev) { - lu_device_fini(&dev->dd_lu_dev); + lu_device_fini(&dev->dd_lu_dev); } EXPORT_SYMBOL(dt_device_fini); int dt_object_init(struct dt_object *obj, - struct lu_object_header *h, struct lu_device *d) + struct lu_object_header *h, struct lu_device *d) { - return lu_object_init(&obj->do_lu, h, d); + return lu_object_init(&obj->do_lu, h, d); } EXPORT_SYMBOL(dt_object_init); void dt_object_fini(struct dt_object *obj) { - lu_object_fini(&obj->do_lu); + lu_object_fini(&obj->do_lu); } EXPORT_SYMBOL(dt_object_fini); int dt_try_as_dir(const struct lu_env *env, struct dt_object *obj) { - if (obj->do_index_ops == NULL) - obj->do_ops->do_index_try(env, obj, &dt_directory_features); - return obj->do_index_ops != NULL; + if (obj->do_index_ops == NULL) + obj->do_ops->do_index_try(env, obj, &dt_directory_features); + return obj->do_index_ops != NULL; } EXPORT_SYMBOL(dt_try_as_dir); enum dt_format_type dt_mode_to_dft(__u32 mode) { - enum dt_format_type result; - - switch (mode & S_IFMT) { - case S_IFDIR: - result = DFT_DIR; - break; - case S_IFREG: - result = DFT_REGULAR; - break; - case S_IFLNK: - result = DFT_SYM; - break; - case S_IFCHR: - case S_IFBLK: - case S_IFIFO: - case S_IFSOCK: - result = DFT_NODE; - break; - default: - LBUG(); - break; - } - return result; -} + enum dt_format_type result; + switch (mode & S_IFMT) { + case S_IFDIR: + result = DFT_DIR; + break; + case S_IFREG: + result = DFT_REGULAR; + break; + case S_IFLNK: + result = DFT_SYM; + break; + case S_IFCHR: + case S_IFBLK: + case S_IFIFO: + case S_IFSOCK: + result = DFT_NODE; + break; + default: + LASSERTF(0, "invalid mode %o\n", mode); + result = 0; /* Just for satisfying compiler. */ + break; + } + return result; +} EXPORT_SYMBOL(dt_mode_to_dft); + /** * lookup fid for object named \a name in directory \a dir. */ -static int dt_lookup(const struct lu_env *env, struct dt_object *dir, - const char *name, struct lu_fid *fid) +int dt_lookup_dir(const struct lu_env *env, struct dt_object *dir, + const char *name, struct lu_fid *fid) { - struct dt_thread_info *info = lu_context_key_get(&env->le_ctx, - &dt_key); - struct lu_fid_pack *pack = &info->dti_pack; - struct dt_rec *rec = (struct dt_rec *)pack; - const struct dt_key *key = (const struct dt_key *)name; - int result; - - if (dt_try_as_dir(env, dir)) { - result = dir->do_index_ops->dio_lookup(env, dir, rec, key, - BYPASS_CAPA); - if (result > 0) - result = fid_unpack(pack, fid); - else if (result == 0) - result = -ENOENT; - } else - result = -ENOTDIR; - return result; + if (dt_try_as_dir(env, dir)) + return dt_lookup(env, dir, (struct dt_rec *)fid, + (const struct dt_key *)name); + return -ENOTDIR; } +EXPORT_SYMBOL(dt_lookup_dir); -/** - * get object for given \a fid. +/* + * this differs from dt_locate by top_dev as parameter + * but not one from lu_site */ -struct dt_object *dt_locate(const struct lu_env *env, - struct dt_device *dev, - const struct lu_fid *fid) +struct dt_object *dt_locate_at(const struct lu_env *env, + struct dt_device *dev, + const struct lu_fid *fid, + struct lu_device *top_dev, + const struct lu_object_conf *conf) { - struct lu_object *obj; - struct dt_object *dt; + struct lu_object *lo; + struct lu_object *n; + + lo = lu_object_find_at(env, top_dev, fid, conf); + if (IS_ERR(lo)) + return ERR_PTR(PTR_ERR(lo)); + + LASSERT(lo != NULL); + + list_for_each_entry(n, &lo->lo_header->loh_layers, lo_linkage) { + if (n->lo_dev == &dev->dd_lu_dev) + return container_of0(n, struct dt_object, do_lu); + } - obj = lu_object_find(env, &dev->dd_lu_dev, fid, NULL); - if (!IS_ERR(obj)) { - obj = lu_object_locate(obj->lo_header, dev->dd_lu_dev.ld_type); - LASSERT(obj != NULL); - dt = container_of(obj, struct dt_object, do_lu); - } else - dt = (struct dt_object *)obj; - return dt; + lu_object_put(env, lo); + return ERR_PTR(-ENOENT); } -EXPORT_SYMBOL(dt_locate); +EXPORT_SYMBOL(dt_locate_at); /** - * find a object named \a entry in given \a dfh->dfh_o directory. + * find an object named \a entry in given \a dfh->dfh_o directory. */ -static int dt_find_entry(const struct lu_env *env, const char *entry, void *data) +static int dt_find_entry(const struct lu_env *env, const char *entry, + void *data) { - struct dt_find_hint *dfh = data; - struct dt_device *dt = dfh->dfh_dt; - struct lu_fid *fid = dfh->dfh_fid; - struct dt_object *obj = dfh->dfh_o; - int result; + struct dt_find_hint *dfh = data; + struct dt_device *dt = dfh->dfh_dt; + struct lu_fid *fid = dfh->dfh_fid; + struct dt_object *obj = dfh->dfh_o; + int rc; - result = dt_lookup(env, obj, entry, fid); - lu_object_put(env, &obj->do_lu); - if (result == 0) { - obj = dt_locate(env, dt, fid); - if (IS_ERR(obj)) - result = PTR_ERR(obj); - } - dfh->dfh_o = obj; - return result; + rc = dt_lookup_dir(env, obj, entry, fid); + dt_object_put(env, obj); + if (rc == 0) { + obj = dt_locate(env, dt, fid); + if (IS_ERR(obj)) + rc = PTR_ERR(obj); + } + dfh->dfh_o = obj; + + return rc; } /** @@ -281,80 +277,79 @@ static int dt_find_entry(const struct lu_env *env, const char *entry, void *data * path component to \a entry_func. */ int dt_path_parser(const struct lu_env *env, - char *path, dt_entry_func_t entry_func, - void *data) -{ - char *e; - int rc = 0; - - while (1) { - e = strsep(&path, "/"); - if (e == NULL) - break; - - if (e[0] == 0) { - if (!path || path[0] == '\0') - break; - continue; - } - rc = entry_func(env, e, data); - if (rc) - break; - } - - return rc; -} - -static struct dt_object *dt_store_resolve(const struct lu_env *env, - struct dt_device *dt, - const char *path, - struct lu_fid *fid) -{ - struct dt_thread_info *info = lu_context_key_get(&env->le_ctx, - &dt_key); - struct dt_find_hint *dfh = &info->dti_dfh; - struct dt_object *obj = dfh->dfh_o; - char *local = info->dti_buf; - int result; - - dfh->dfh_dt = dt; - dfh->dfh_fid = fid; - - strncpy(local, path, DT_MAX_PATH); - local[DT_MAX_PATH - 1] = '\0'; - - result = dt->dd_ops->dt_root_get(env, dt, fid); - if (result == 0) { - obj = dt_locate(env, dt, fid); - if (!IS_ERR(obj)) { - dfh->dfh_o = obj; - result = dt_path_parser(env, local, dt_find_entry, dfh); - if (result != 0) - obj = ERR_PTR(result); - } - } else { - obj = ERR_PTR(result); - } - return obj; + char *path, dt_entry_func_t entry_func, + void *data) +{ + char *e; + int rc = 0; + + while (1) { + e = strsep(&path, "/"); + if (e == NULL) + break; + + if (e[0] == 0) { + if (!path || path[0] == '\0') + break; + continue; + } + rc = entry_func(env, e, data); + if (rc) + break; + } + + return rc; +} + +struct dt_object * +dt_store_resolve(const struct lu_env *env, struct dt_device *dt, + const char *path, struct lu_fid *fid) +{ + struct dt_thread_info *info = dt_info(env); + struct dt_find_hint *dfh = &info->dti_dfh; + struct dt_object *obj; + int result; + + + dfh->dfh_dt = dt; + dfh->dfh_fid = fid; + + strlcpy(info->dti_buf, path, sizeof(info->dti_buf)); + + result = dt->dd_ops->dt_root_get(env, dt, fid); + if (result == 0) { + obj = dt_locate(env, dt, fid); + if (!IS_ERR(obj)) { + dfh->dfh_o = obj; + result = dt_path_parser(env, info->dti_buf, + dt_find_entry, dfh); + if (result != 0) + obj = ERR_PTR(result); + else + obj = dfh->dfh_o; + } + } else { + obj = ERR_PTR(result); + } + return obj; } static struct dt_object *dt_reg_open(const struct lu_env *env, - struct dt_device *dt, - struct dt_object *p, - const char *name, - struct lu_fid *fid) + struct dt_device *dt, + struct dt_object *p, + const char *name, + struct lu_fid *fid) { - struct dt_object *o; - int result; + struct dt_object *o; + int result; - result = dt_lookup(env, p, name, fid); - if (result == 0){ - o = dt_locate(env, dt, fid); - } - else - o = ERR_PTR(result); + result = dt_lookup_dir(env, p, name, fid); + if (result == 0) + o = dt_locate(env, dt, fid); + else + o = ERR_PTR(result); - return o; + return o; } /** @@ -362,41 +357,948 @@ static struct dt_object *dt_reg_open(const struct lu_env *env, * \param dt dt device * \param fid on success, object fid is stored in *fid */ -struct dt_object *dt_store_open(const struct lu_env *env, - struct dt_device *dt, - const char *dirname, - const char *filename, - struct lu_fid *fid) -{ - struct dt_object *file; - struct dt_object *dir; - - dir = dt_store_resolve(env, dt, dirname, fid); - if (!IS_ERR(dir)) { - file = dt_reg_open(env, dt, dir, - filename, fid); - lu_object_put(env, &dir->do_lu); - } else { - file = dir; - } - return file; -} -EXPORT_SYMBOL(dt_store_open); +struct dt_object *dt_store_open(const struct lu_env *env, struct dt_device *dt, + const char *dirname, const char *filename, + struct lu_fid *fid) +{ + struct dt_object *file; + struct dt_object *dir; + + dir = dt_store_resolve(env, dt, dirname, fid); + if (!IS_ERR(dir)) { + file = dt_reg_open(env, dt, dir, filename, fid); + dt_object_put(env, dir); + } else { + file = dir; + } + + return file; +} + +struct dt_object *dt_find_or_create(const struct lu_env *env, + struct dt_device *dt, + const struct lu_fid *fid, + struct dt_object_format *dof, + struct lu_attr *at) +{ + struct dt_object *dto; + struct thandle *th; + int rc; + + ENTRY; + + dto = dt_locate(env, dt, fid); + if (IS_ERR(dto)) + RETURN(dto); + + LASSERT(dto != NULL); + if (dt_object_exists(dto)) + RETURN(dto); + + th = dt_trans_create(env, dt); + if (IS_ERR(th)) + GOTO(out, rc = PTR_ERR(th)); + + rc = dt_declare_create(env, dto, at, NULL, dof, th); + if (rc) + GOTO(trans_stop, rc); + + rc = dt_trans_start_local(env, dt, th); + if (rc) + GOTO(trans_stop, rc); + + dt_write_lock(env, dto, 0); + if (dt_object_exists(dto)) + GOTO(unlock, rc = 0); + + CDEBUG(D_OTHER, "create new object "DFID"\n", PFID(fid)); + + rc = dt_create(env, dto, at, NULL, dof, th); + if (rc) + GOTO(unlock, rc); + LASSERT(dt_object_exists(dto)); +unlock: + dt_write_unlock(env, dto); +trans_stop: + dt_trans_stop(env, dt, th); +out: + if (rc) { + dt_object_put(env, dto); + dto = ERR_PTR(rc); + } + + RETURN(dto); +} +EXPORT_SYMBOL(dt_find_or_create); /* dt class init function. */ int dt_global_init(void) { - int result; + int result; - LU_CONTEXT_KEY_INIT(&dt_key); - result = lu_context_key_register(&dt_key); - return result; + LU_CONTEXT_KEY_INIT(&dt_key); + result = lu_context_key_register(&dt_key); + return result; } void dt_global_fini(void) { - lu_context_key_degister(&dt_key); + lu_context_key_degister(&dt_key); } +/** + * Generic read helper. May return an error for partial reads. + * + * \param env lustre environment + * \param dt object to be read + * \param buf lu_buf to be filled, with buffer pointer and length + * \param pos position to start reading, updated as data is read + * + * \retval real size of data read + * \retval -ve errno on failure + */ +int dt_read(const struct lu_env *env, struct dt_object *dt, + struct lu_buf *buf, loff_t *pos) +{ + LASSERTF(dt != NULL, "dt is NULL when we want to read record\n"); + return dt->do_body_ops->dbo_read(env, dt, buf, pos); +} +EXPORT_SYMBOL(dt_read); + +/** + * Read structures of fixed size from storage. Unlike dt_read(), using + * dt_record_read() will return an error for partial reads. + * + * \param env lustre environment + * \param dt object to be read + * \param buf lu_buf to be filled, with buffer pointer and length + * \param pos position to start reading, updated as data is read + * + * \retval 0 on successfully reading full buffer + * \retval -EFAULT on short read + * \retval -ve errno on failure + */ +int dt_record_read(const struct lu_env *env, struct dt_object *dt, + struct lu_buf *buf, loff_t *pos) +{ + ssize_t size; + + LASSERTF(dt != NULL, "dt is NULL when we want to read record\n"); + + size = dt->do_body_ops->dbo_read(env, dt, buf, pos); + if (size < 0) + return size; + return (size == (ssize_t)buf->lb_len) ? 0 : -EFAULT; +} +EXPORT_SYMBOL(dt_record_read); + +int dt_record_write(const struct lu_env *env, struct dt_object *dt, + const struct lu_buf *buf, loff_t *pos, struct thandle *th) +{ + ssize_t size; + + LASSERTF(dt != NULL, "dt is NULL when we want to write record\n"); + LASSERT(th != NULL); + LASSERT(dt->do_body_ops); + LASSERT(dt->do_body_ops->dbo_write); + + size = dt->do_body_ops->dbo_write(env, dt, buf, pos, th); + if (size < 0) + return size; + return (size == (ssize_t)buf->lb_len) ? 0 : -EFAULT; +} +EXPORT_SYMBOL(dt_record_write); + +int dt_declare_version_set(const struct lu_env *env, struct dt_object *o, + struct thandle *th) +{ + struct lu_buf vbuf; + char *xname = XATTR_NAME_VERSION; + + LASSERT(o); + vbuf.lb_buf = NULL; + vbuf.lb_len = sizeof(dt_obj_version_t); + return dt_declare_xattr_set(env, o, &vbuf, xname, 0, th); + +} +EXPORT_SYMBOL(dt_declare_version_set); + +void dt_version_set(const struct lu_env *env, struct dt_object *o, + dt_obj_version_t version, struct thandle *th) +{ + struct lu_buf vbuf; + char *xname = XATTR_NAME_VERSION; + int rc; + + LASSERT(o); + vbuf.lb_buf = &version; + vbuf.lb_len = sizeof(version); + + rc = dt_xattr_set(env, o, &vbuf, xname, 0, th); + if (rc < 0) + CDEBUG(D_INODE, "Can't set version, rc %d\n", rc); + return; +} +EXPORT_SYMBOL(dt_version_set); + +dt_obj_version_t dt_version_get(const struct lu_env *env, struct dt_object *o) +{ + struct lu_buf vbuf; + char *xname = XATTR_NAME_VERSION; + dt_obj_version_t version; + int rc; + + LASSERT(o); + vbuf.lb_buf = &version; + vbuf.lb_len = sizeof(version); + rc = dt_xattr_get(env, o, &vbuf, xname); + if (rc != sizeof(version)) { + CDEBUG(D_INODE, "Can't get version, rc %d\n", rc); + version = 0; + } + return version; +} +EXPORT_SYMBOL(dt_version_get); + +/* list of all supported index types */ + +/* directories */ const struct dt_index_features dt_directory_features; EXPORT_SYMBOL(dt_directory_features); + +/* scrub iterator */ +const struct dt_index_features dt_otable_features; +EXPORT_SYMBOL(dt_otable_features); + +/* lfsck layout orphan */ +const struct dt_index_features dt_lfsck_layout_orphan_features = { + .dif_flags = 0, + .dif_keysize_min = sizeof(struct lu_fid), + .dif_keysize_max = sizeof(struct lu_fid), + .dif_recsize_min = sizeof(struct lu_orphan_rec_v3), + .dif_recsize_max = sizeof(struct lu_orphan_rec_v3), + .dif_ptrsize = 4 +}; +EXPORT_SYMBOL(dt_lfsck_layout_orphan_features); + +/* lfsck layout dangling */ +const struct dt_index_features dt_lfsck_layout_dangling_features = { + .dif_flags = DT_IND_UPDATE, + .dif_keysize_min = sizeof(struct lfsck_layout_dangling_key), + .dif_keysize_max = sizeof(struct lfsck_layout_dangling_key), + .dif_recsize_min = sizeof(struct lu_fid), + .dif_recsize_max = sizeof(struct lu_fid), + .dif_ptrsize = 4 +}; +EXPORT_SYMBOL(dt_lfsck_layout_dangling_features); + +/* lfsck namespace */ +const struct dt_index_features dt_lfsck_namespace_features = { + .dif_flags = DT_IND_UPDATE, + .dif_keysize_min = sizeof(struct lu_fid), + .dif_keysize_max = sizeof(struct lu_fid), + .dif_recsize_min = sizeof(__u8), + .dif_recsize_max = sizeof(__u8), + .dif_ptrsize = 4 +}; +EXPORT_SYMBOL(dt_lfsck_namespace_features); + +/* accounting indexes */ +const struct dt_index_features dt_acct_features = { + .dif_flags = DT_IND_UPDATE, + .dif_keysize_min = sizeof(__u64), /* 64-bit uid/gid */ + .dif_keysize_max = sizeof(__u64), /* 64-bit uid/gid */ + .dif_recsize_min = sizeof(struct lquota_acct_rec), /* 16 bytes */ + .dif_recsize_max = sizeof(struct lquota_acct_rec), /* 16 bytes */ + .dif_ptrsize = 4 +}; +EXPORT_SYMBOL(dt_acct_features); + +/* global quota files */ +const struct dt_index_features dt_quota_glb_features = { + .dif_flags = DT_IND_UPDATE, + /* a different key would have to be used for per-directory quota */ + .dif_keysize_min = sizeof(__u64), /* 64-bit uid/gid */ + .dif_keysize_max = sizeof(__u64), /* 64-bit uid/gid */ + .dif_recsize_min = sizeof(struct lquota_glb_rec), /* 32 bytes */ + .dif_recsize_max = sizeof(struct lquota_glb_rec), /* 32 bytes */ + .dif_ptrsize = 4 +}; +EXPORT_SYMBOL(dt_quota_glb_features); + +/* slave quota files */ +const struct dt_index_features dt_quota_slv_features = { + .dif_flags = DT_IND_UPDATE, + /* a different key would have to be used for per-directory quota */ + .dif_keysize_min = sizeof(__u64), /* 64-bit uid/gid */ + .dif_keysize_max = sizeof(__u64), /* 64-bit uid/gid */ + .dif_recsize_min = sizeof(struct lquota_slv_rec), /* 8 bytes */ + .dif_recsize_max = sizeof(struct lquota_slv_rec), /* 8 bytes */ + .dif_ptrsize = 4 +}; +EXPORT_SYMBOL(dt_quota_slv_features); + +/* nodemap files, nodemap_rec size asserted in nodemap_storage.c */ +const struct dt_index_features dt_nodemap_features = { + .dif_flags = DT_IND_UPDATE, + .dif_keysize_min = sizeof(__u64), /* 64-bit nodemap/record id */ + .dif_keysize_max = sizeof(__u64), /* 64-bit nodemap/record id */ + .dif_recsize_min = sizeof(union nodemap_rec), /* 32 bytes */ + .dif_recsize_max = sizeof(union nodemap_rec), /* 32 bytes */ + .dif_ptrsize = 4 +}; +EXPORT_SYMBOL(dt_nodemap_features); + +/* + * helper function returning what dt_index_features structure should be used + * based on the FID sequence. This is used by OBD_IDX_READ RPC + */ +static inline const struct dt_index_features *dt_index_feat_select(__u64 seq, + __u32 mode) +{ + if (seq == FID_SEQ_QUOTA_GLB) { + /* global quota index */ + if (!S_ISREG(mode)) + /* global quota index should be a regular file */ + return ERR_PTR(-ENOENT); + return &dt_quota_glb_features; + } else if (seq == FID_SEQ_QUOTA) { + /* quota slave index */ + if (!S_ISREG(mode)) + /* slave index should be a regular file */ + return ERR_PTR(-ENOENT); + return &dt_quota_slv_features; + } else if (seq == FID_SEQ_LAYOUT_RBTREE){ + return &dt_lfsck_layout_orphan_features; + } else if (seq >= FID_SEQ_NORMAL) { + /* object is part of the namespace, verify that it is a + * directory */ + if (!S_ISDIR(mode)) + /* sorry, we can only deal with directory */ + return ERR_PTR(-ENOTDIR); + return &dt_directory_features; + } + + return ERR_PTR(-EOPNOTSUPP); +} + +/* + * Fill a lu_idxpage with key/record pairs read for transfer via OBD_IDX_READ + * RPC + * + * \param env - is the environment passed by the caller + * \param lp - is a pointer to the lu_page to fill + * \param nob - is the maximum number of bytes that should be copied + * \param iops - is the index operation vector associated with the index object + * \param it - is a pointer to the current iterator + * \param attr - is the index attribute to pass to iops->rec() + * \param arg - is a pointer to the idx_info structure + */ +static int dt_index_page_build(const struct lu_env *env, union lu_page *lp, + size_t nob, const struct dt_it_ops *iops, + struct dt_it *it, __u32 attr, void *arg) +{ + struct idx_info *ii = (struct idx_info *)arg; + struct lu_idxpage *lip = &lp->lp_idx; + char *entry; + __u64 hash; + __u16 hashsize = 0; + __u16 keysize = 0; + __u16 recsize; + int rc; + + ENTRY; + + if (nob < LIP_HDR_SIZE) + return -EINVAL; + + /* initialize the header of the new container */ + memset(lip, 0, LIP_HDR_SIZE); + lip->lip_magic = LIP_MAGIC; + nob -= LIP_HDR_SIZE; + + /* client wants to the 64-bit hash value associated with each record */ + if (!(ii->ii_flags & II_FL_NOHASH)) + hashsize = sizeof(hash); + + entry = lip->lip_entries; + do { + /* fetch 64-bit hash value */ + hash = iops->store(env, it); + ii->ii_hash_end = hash; + + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_IDX_READ_BREAK)) { + if (lip->lip_nr != 0) + GOTO(out, rc = 0); + } + + if (!(ii->ii_flags & II_FL_NOKEY)) { + keysize = iops->key_size(env, it); + if (!(ii->ii_flags & II_FL_VARKEY) && + keysize != ii->ii_keysize) { + CERROR("keysize mismatch %hu != %hu.\n", + keysize, ii->ii_keysize); + GOTO(out, rc = -EINVAL); + } + } + + /* and finally the record */ + if (ii->ii_flags & II_FL_VARREC) + recsize = iops->rec_size(env, it, attr); + else + recsize = ii->ii_recsize; + + if (nob < hashsize + keysize + recsize) { + if (lip->lip_nr == 0) + GOTO(out, rc = -E2BIG); + GOTO(out, rc = 0); + } + + rc = iops->rec(env, it, + (struct dt_rec *)(entry + hashsize + keysize), + attr); + if (!rc) { + if (hashsize) + memcpy(entry, &hash, hashsize); + if (keysize) { + struct dt_key *key; + + key = iops->key(env, it); + memcpy(entry + hashsize, key, keysize); + } + /* hash/key/record successfully copied! */ + lip->lip_nr++; + if (unlikely(lip->lip_nr == 1 && ii->ii_count == 0)) + ii->ii_hash_start = hash; + entry += hashsize + keysize + recsize; + nob -= hashsize + keysize + recsize; + } else if (rc != -ESTALE) { + GOTO(out, rc); + } + + /* move on to the next record */ + do { + rc = iops->next(env, it); + } while (rc == -ESTALE); + } while (rc == 0); + + GOTO(out, rc); +out: + if (rc >= 0 && lip->lip_nr > 0) + /* one more container */ + ii->ii_count++; + if (rc > 0) + /* no more entries */ + ii->ii_hash_end = II_END_OFF; + return rc; +} + + +/* + * Walk index and fill lu_page containers with key/record pairs + * + * \param env - is the environment passed by the caller + * \param obj - is the index object to parse + * \param rdpg - is the lu_rdpg descriptor associated with the transfer + * \param filler - is the callback function responsible for filling a lu_page + * with key/record pairs in the format wanted by the caller. + * If NULL, uses dt_index_page_build + * \param arg - is an opaq argument passed to the filler function + * + * \retval sum (in bytes) of all filled lu_pages + * \retval -ve errno on failure + */ +int dt_index_walk(const struct lu_env *env, struct dt_object *obj, + const struct lu_rdpg *rdpg, dt_index_page_build_t filler, + void *arg) +{ + struct dt_it *it; + const struct dt_it_ops *iops; + size_t pageidx, nob, nlupgs = 0; + int rc; + ENTRY; + + LASSERT(rdpg->rp_pages != NULL); + LASSERT(obj->do_index_ops != NULL); + + if (filler == NULL) + filler = dt_index_page_build; + + nob = rdpg->rp_count; + if (nob == 0) + RETURN(-EFAULT); + + /* Iterate through index and fill containers from @rdpg */ + iops = &obj->do_index_ops->dio_it; + LASSERT(iops != NULL); + it = iops->init(env, obj, rdpg->rp_attrs); + if (IS_ERR(it)) + RETURN(PTR_ERR(it)); + + rc = iops->load(env, it, rdpg->rp_hash); + if (rc == 0) { + /* + * Iterator didn't find record with exactly the key requested. + * + * It is currently either + * + * - positioned above record with key less than + * requested---skip it. + * - or not positioned at all (is in IAM_IT_SKEWED + * state)---position it on the next item. + */ + rc = iops->next(env, it); + } else if (rc > 0) { + rc = 0; + } else { + if (rc == -ENODATA) + rc = 0; + GOTO(out, rc); + } + + /* + * Fill containers one after the other. There might be multiple + * containers per physical page. + * + * At this point and across for-loop: + * rc == 0 -> ok, proceed. + * rc > 0 -> end of index. + * rc < 0 -> error. + */ + for (pageidx = 0; rc == 0 && nob > 0; pageidx++) { + union lu_page *lp; + int i; + + LASSERT(pageidx < rdpg->rp_npages); + lp = kmap(rdpg->rp_pages[pageidx]); + + /* fill lu pages */ + for (i = 0; i < LU_PAGE_COUNT; i++, lp++, nob -= LU_PAGE_SIZE) { + rc = filler(env, lp, min_t(size_t, nob, LU_PAGE_SIZE), + iops, it, rdpg->rp_attrs, arg); + if (rc < 0) + break; + /* one more lu_page */ + nlupgs++; + if (rc > 0) + /* end of index */ + break; + } + kunmap(rdpg->rp_pages[i]); + } + +out: + iops->put(env, it); + iops->fini(env, it); + + if (rc >= 0) + rc = min_t(size_t, nlupgs * LU_PAGE_SIZE, rdpg->rp_count); + + RETURN(rc); +} +EXPORT_SYMBOL(dt_index_walk); + +/** + * Walk key/record pairs of an index and copy them into 4KB containers to be + * transferred over the network. This is the common handler for OBD_IDX_READ + * RPC processing. + * + * \param env - is the environment passed by the caller + * \param dev - is the dt_device storing the index + * \param ii - is the idx_info structure packed by the client in the + * OBD_IDX_READ request + * \param rdpg - is the lu_rdpg descriptor + * + * \retval on success, return sum (in bytes) of all filled containers + * \retval appropriate error otherwise. + */ +int dt_index_read(const struct lu_env *env, struct dt_device *dev, + struct idx_info *ii, const struct lu_rdpg *rdpg) +{ + const struct dt_index_features *feat; + struct dt_object *obj; + int rc; + ENTRY; + + /* + * rp_count shouldn't be null and should be a multiple of the container + * size + */ + if (rdpg->rp_count == 0 || (rdpg->rp_count & (LU_PAGE_SIZE - 1)) != 0) + RETURN(-EFAULT); + + if (!fid_is_quota(&ii->ii_fid) && !fid_is_layout_rbtree(&ii->ii_fid) && + !fid_is_norm(&ii->ii_fid)) + RETURN(-EOPNOTSUPP); + + /* lookup index object subject to the transfer */ + obj = dt_locate(env, dev, &ii->ii_fid); + if (IS_ERR(obj)) + RETURN(PTR_ERR(obj)); + if (dt_object_exists(obj) == 0) + GOTO(out, rc = -ENOENT); + + /* fetch index features associated with index object */ + feat = dt_index_feat_select(fid_seq(&ii->ii_fid), + lu_object_attr(&obj->do_lu)); + if (IS_ERR(feat)) + GOTO(out, rc = PTR_ERR(feat)); + + /* load index feature if not done already */ + if (obj->do_index_ops == NULL) { + rc = obj->do_ops->do_index_try(env, obj, feat); + if (rc) + GOTO(out, rc); + } + + /* fill ii_flags with supported index features */ + ii->ii_flags &= (II_FL_NOHASH | II_FL_NOKEY | II_FL_VARKEY | + II_FL_VARREC); + + if (!(feat->dif_flags & DT_IND_VARKEY)) + ii->ii_keysize = feat->dif_keysize_max; + + if (!(feat->dif_flags & DT_IND_VARREC)) + ii->ii_recsize = feat->dif_recsize_max; + + if (feat->dif_flags & DT_IND_NONUNQ) + /* key isn't necessarily unique */ + ii->ii_flags |= II_FL_NONUNQ; + + if (!fid_is_layout_rbtree(&ii->ii_fid)) { + dt_read_lock(env, obj, 0); + /* fetch object version before walking the index */ + ii->ii_version = dt_version_get(env, obj); + } + + /* walk the index and fill lu_idxpages with key/record pairs */ + rc = dt_index_walk(env, obj, rdpg, dt_index_page_build, ii); + if (!fid_is_layout_rbtree(&ii->ii_fid)) + dt_read_unlock(env, obj); + + if (rc == 0) { + /* index is empty */ + LASSERT(ii->ii_count == 0); + ii->ii_hash_end = II_END_OFF; + } + + GOTO(out, rc); +out: + dt_object_put(env, obj); + return rc; +} +EXPORT_SYMBOL(dt_index_read); + +#ifdef CONFIG_PROC_FS +int lprocfs_dt_blksize_seq_show(struct seq_file *m, void *v) +{ + struct dt_device *dt = m->private; + struct obd_statfs osfs; + + int rc = dt_statfs(NULL, dt, &osfs); + if (rc == 0) + seq_printf(m, "%u\n", (unsigned) osfs.os_bsize); + return rc; +} +EXPORT_SYMBOL(lprocfs_dt_blksize_seq_show); + +int lprocfs_dt_kbytestotal_seq_show(struct seq_file *m, void *v) +{ + struct dt_device *dt = m->private; + struct obd_statfs osfs; + + int rc = dt_statfs(NULL, dt, &osfs); + if (rc == 0) { + __u32 blk_size = osfs.os_bsize >> 10; + __u64 result = osfs.os_blocks; + + while (blk_size >>= 1) + result <<= 1; + + seq_printf(m, "%llu\n", result); + } + return rc; +} +EXPORT_SYMBOL(lprocfs_dt_kbytestotal_seq_show); + +int lprocfs_dt_kbytesfree_seq_show(struct seq_file *m, void *v) +{ + struct dt_device *dt = m->private; + struct obd_statfs osfs; + + int rc = dt_statfs(NULL, dt, &osfs); + if (rc == 0) { + __u32 blk_size = osfs.os_bsize >> 10; + __u64 result = osfs.os_bfree; + + while (blk_size >>= 1) + result <<= 1; + + seq_printf(m, "%llu\n", result); + } + return rc; +} +EXPORT_SYMBOL(lprocfs_dt_kbytesfree_seq_show); + +int lprocfs_dt_kbytesavail_seq_show(struct seq_file *m, void *v) +{ + struct dt_device *dt = m->private; + struct obd_statfs osfs; + + int rc = dt_statfs(NULL, dt, &osfs); + if (rc == 0) { + __u32 blk_size = osfs.os_bsize >> 10; + __u64 result = osfs.os_bavail; + + while (blk_size >>= 1) + result <<= 1; + + seq_printf(m, "%llu\n", result); + } + return rc; +} +EXPORT_SYMBOL(lprocfs_dt_kbytesavail_seq_show); + +int lprocfs_dt_filestotal_seq_show(struct seq_file *m, void *v) +{ + struct dt_device *dt = m->private; + struct obd_statfs osfs; + + int rc = dt_statfs(NULL, dt, &osfs); + if (rc == 0) + seq_printf(m, "%llu\n", osfs.os_files); + return rc; +} +EXPORT_SYMBOL(lprocfs_dt_filestotal_seq_show); + +int lprocfs_dt_filesfree_seq_show(struct seq_file *m, void *v) +{ + struct dt_device *dt = m->private; + struct obd_statfs osfs; + + int rc = dt_statfs(NULL, dt, &osfs); + if (rc == 0) + seq_printf(m, "%llu\n", osfs.os_ffree); + return rc; +} +EXPORT_SYMBOL(lprocfs_dt_filesfree_seq_show); + +#endif /* CONFIG_PROC_FS */ + +static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr, + char *buf) +{ + struct dt_device *dt = container_of(kobj, struct dt_device, + dd_kobj); + struct lu_device *lu = dt2lu_dev(dt); + + if (!lu->ld_obd) + return -ENODEV; + + return sprintf(buf, "%s\n", lu->ld_obd->obd_uuid.uuid); +} +LUSTRE_RO_ATTR(uuid); + +static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr, + char *buf) +{ + struct dt_device *dt = container_of(kobj, struct dt_device, + dd_kobj); + struct obd_statfs osfs; + int rc; + + rc = dt_statfs(NULL, dt, &osfs); + if (rc) + return rc; + + return sprintf(buf, "%u\n", (unsigned) osfs.os_bsize); +} +LUSTRE_RO_ATTR(blocksize); + +static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr, + char *buf) +{ + struct dt_device *dt = container_of(kobj, struct dt_device, + dd_kobj); + struct obd_statfs osfs; + u32 blk_size; + u64 result; + int rc; + + rc = dt_statfs(NULL, dt, &osfs); + if (rc) + return rc; + + blk_size = osfs.os_bsize >> 10; + result = osfs.os_blocks; + + while (blk_size >>= 1) + result <<= 1; + + return sprintf(buf, "%llu\n", result); +} +LUSTRE_RO_ATTR(kbytestotal); + +static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr, + char *buf) +{ + struct dt_device *dt = container_of(kobj, struct dt_device, + dd_kobj); + struct obd_statfs osfs; + u32 blk_size; + u64 result; + int rc; + + rc = dt_statfs(NULL, dt, &osfs); + if (rc) + return rc; + + blk_size = osfs.os_bsize >> 10; + result = osfs.os_bfree; + + while (blk_size >>= 1) + result <<= 1; + + return sprintf(buf, "%llu\n", result); +} +LUSTRE_RO_ATTR(kbytesfree); + +static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr, + char *buf) +{ + struct dt_device *dt = container_of(kobj, struct dt_device, + dd_kobj); + struct obd_statfs osfs; + u32 blk_size; + u64 result; + int rc; + + rc = dt_statfs(NULL, dt, &osfs); + if (rc) + return rc; + + blk_size = osfs.os_bsize >> 10; + result = osfs.os_bavail; + + while (blk_size >>= 1) + result <<= 1; + + return sprintf(buf, "%llu\n", result); +} +LUSTRE_RO_ATTR(kbytesavail); + +static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr, + char *buf) +{ + struct dt_device *dt = container_of(kobj, struct dt_device, + dd_kobj); + struct obd_statfs osfs; + int rc; + + rc = dt_statfs(NULL, dt, &osfs); + if (rc) + return rc; + + return sprintf(buf, "%llu\n", osfs.os_files); +} +LUSTRE_RO_ATTR(filestotal); + +static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr, + char *buf) +{ + struct dt_device *dt = container_of(kobj, struct dt_device, + dd_kobj); + struct obd_statfs osfs; + int rc; + + rc = dt_statfs(NULL, dt, &osfs); + if (rc) + return rc; + + return sprintf(buf, "%llu\n", osfs.os_ffree); +} +LUSTRE_RO_ATTR(filesfree); + +static const struct attribute *dt_def_attrs[] = { + &lustre_attr_uuid.attr, + &lustre_attr_blocksize.attr, + &lustre_attr_kbytestotal.attr, + &lustre_attr_kbytesfree.attr, + &lustre_attr_kbytesavail.attr, + &lustre_attr_filestotal.attr, + &lustre_attr_filesfree.attr, + NULL, +}; + +static void dt_sysfs_release(struct kobject *kobj) +{ + struct dt_device *dt = container_of(kobj, struct dt_device, + dd_kobj); + + debugfs_remove_recursive(dt->dd_debugfs_entry); + dt->dd_debugfs_entry = NULL; + + complete(&dt->dd_kobj_unregister); +} + +int dt_tunables_fini(struct dt_device *dt) +{ + if (!dt) + return -EINVAL; + + if (dt->dd_def_attrs) + sysfs_remove_files(&dt->dd_kobj, dt->dd_def_attrs); + + kobject_put(&dt->dd_kobj); + wait_for_completion(&dt->dd_kobj_unregister); + + return 0; +} +EXPORT_SYMBOL(dt_tunables_fini); + +int dt_tunables_init(struct dt_device *dt, struct obd_type *type, + const char *name, struct lprocfs_vars *list) +{ + int rc; + + dt->dd_ktype.sysfs_ops = &lustre_sysfs_ops; + dt->dd_ktype.release = dt_sysfs_release; + + init_completion(&dt->dd_kobj_unregister); + rc = kobject_init_and_add(&dt->dd_kobj, &dt->dd_ktype, &type->typ_kobj, + "%s", name); + if (rc) + return rc; + + dt->dd_def_attrs = dt_def_attrs; + + rc = sysfs_create_files(&dt->dd_kobj, dt->dd_def_attrs); + if (rc) { + kobject_put(&dt->dd_kobj); + return rc; + } + + /* + * No need to register debugfs if no enteries. This allows us to + * choose between using dt_device or obd_device for debugfs. + */ + if (!list) + return rc; + + dt->dd_debugfs_entry = ldebugfs_register(name, + type->typ_debugfs_entry, + list, dt); + if (IS_ERR_OR_NULL(dt->dd_debugfs_entry)) { + rc = dt->dd_debugfs_entry ? PTR_ERR(dt->dd_debugfs_entry) + : -ENOMEM; + CERROR("%s: error %d setting up debugfs\n", + name, rc); + dt->dd_debugfs_entry = NULL; + sysfs_remove_files(&dt->dd_kobj, dt->dd_def_attrs); + kobject_put(&dt->dd_kobj); + return rc; + } + + return rc; +} +EXPORT_SYMBOL(dt_tunables_init);