4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * cl code shared between vvp and liblustre (and other Lustre clients in the
35 * Author: Nikita Danilov <nikita.danilov@sun.com>
38 #define DEBUG_SUBSYSTEM S_LLITE
40 #include "../../include/linux/libcfs/libcfs.h"
41 # include <linux/fs.h>
42 # include <linux/sched.h>
43 # include <linux/mm.h>
44 # include <linux/quotaops.h>
45 # include <linux/highmem.h>
46 # include <linux/pagemap.h>
47 # include <linux/rbtree.h>
49 #include "../include/obd.h"
50 #include "../include/obd_support.h"
51 #include "../include/lustre_fid.h"
52 #include "../include/lustre_lite.h"
53 #include "../include/lustre_dlm.h"
54 #include "../include/lustre_ver.h"
55 #include "../include/lustre_mdc.h"
56 #include "../include/cl_object.h"
58 #include "../llite/llite_internal.h"
61 * ccc_ prefix stands for "Common Client Code".
64 /*****************************************************************************
66 * Vvp device and device type functions.
71 * An `emergency' environment used by cl_inode_fini() when cl_env_get()
72 * fails. Access to this environment is serialized by cl_inode_fini_guard
75 struct lu_env *cl_inode_fini_env;
76 int cl_inode_fini_refcheck;
79 * A mutex serializing calls to slp_inode_fini() under extreme memory
80 * pressure, when environments cannot be allocated.
82 static DEFINE_MUTEX(cl_inode_fini_guard);
84 int cl_setattr_ost(struct inode *inode, const struct iattr *attr)
91 env = cl_env_get(&refcheck);
95 io = vvp_env_thread_io(env);
96 io->ci_obj = ll_i2info(inode)->lli_clob;
98 io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime);
99 io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime);
100 io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime);
101 io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size;
102 io->u.ci_setattr.sa_valid = attr->ia_valid;
105 if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0) {
106 struct vvp_io *vio = vvp_env_io(env);
108 if (attr->ia_valid & ATTR_FILE)
109 /* populate the file descriptor for ftruncate to honor
110 * group lock - see LU-787
112 vio->vui_fd = LUSTRE_FPRIVATE(attr->ia_file);
114 result = cl_io_loop(env, io);
116 result = io->ci_result;
119 if (unlikely(io->ci_need_restart))
121 /* HSM import case: file is released, cannot be restored
122 * no need to fail except if restore registration failed
125 if (result == -ENODATA && io->ci_restore_needed &&
126 io->ci_result != -ENODATA)
128 cl_env_put(env, &refcheck);
133 * Initialize or update CLIO structures for regular files when new
134 * meta-data arrives from the server.
136 * \param inode regular file inode
137 * \param md new file metadata from MDS
138 * - allocates cl_object if necessary,
139 * - updated layout, if object was already here.
141 int cl_file_inode_init(struct inode *inode, struct lustre_md *md)
144 struct ll_inode_info *lli;
145 struct cl_object *clob;
146 struct lu_site *site;
148 struct cl_object_conf conf = {
157 LASSERT(md->body->valid & OBD_MD_FLID);
158 LASSERT(S_ISREG(inode->i_mode));
160 env = cl_env_get(&refcheck);
164 site = ll_i2sbi(inode)->ll_site;
165 lli = ll_i2info(inode);
167 LASSERT(fid_is_sane(fid));
169 if (!lli->lli_clob) {
170 /* clob is slave of inode, empty lli_clob means for new inode,
171 * there is no clob in cache with the given fid, so it is
172 * unnecessary to perform lookup-alloc-lookup-insert, just
173 * alloc and insert directly.
175 LASSERT(inode->i_state & I_NEW);
176 conf.coc_lu.loc_flags = LOC_F_NEW;
177 clob = cl_object_find(env, lu2cl_dev(site->ls_top_dev),
181 * No locking is necessary, as new inode is
182 * locked by I_NEW bit.
184 lli->lli_clob = clob;
185 lli->lli_has_smd = lsm_has_objects(md->lsm);
186 lu_object_ref_add(&clob->co_lu, "inode", inode);
188 result = PTR_ERR(clob);
191 result = cl_conf_set(env, lli->lli_clob, &conf);
194 cl_env_put(env, &refcheck);
197 CERROR("Failure to initialize cl object " DFID ": %d\n",
203 * Wait for others drop their references of the object at first, then we drop
204 * the last one, which will lead to the object be destroyed immediately.
205 * Must be called after cl_object_kill() against this object.
207 * The reason we want to do this is: destroying top object will wait for sub
208 * objects being destroyed first, so we can't let bottom layer (e.g. from ASTs)
209 * to initiate top object destroying which may deadlock. See bz22520.
211 static void cl_object_put_last(struct lu_env *env, struct cl_object *obj)
213 struct lu_object_header *header = obj->co_lu.lo_header;
216 if (unlikely(atomic_read(&header->loh_ref) != 1)) {
217 struct lu_site *site = obj->co_lu.lo_dev->ld_site;
218 struct lu_site_bkt_data *bkt;
220 bkt = lu_site_bkt_from_fid(site, &header->loh_fid);
222 init_waitqueue_entry(&waiter, current);
223 add_wait_queue(&bkt->lsb_marche_funebre, &waiter);
226 set_current_state(TASK_UNINTERRUPTIBLE);
227 if (atomic_read(&header->loh_ref) == 1)
232 set_current_state(TASK_RUNNING);
233 remove_wait_queue(&bkt->lsb_marche_funebre, &waiter);
236 cl_object_put(env, obj);
239 void cl_inode_fini(struct inode *inode)
242 struct ll_inode_info *lli = ll_i2info(inode);
243 struct cl_object *clob = lli->lli_clob;
250 cookie = cl_env_reenter();
251 env = cl_env_get(&refcheck);
252 emergency = IS_ERR(env);
254 mutex_lock(&cl_inode_fini_guard);
255 LASSERT(cl_inode_fini_env);
256 cl_env_implant(cl_inode_fini_env, &refcheck);
257 env = cl_inode_fini_env;
260 * cl_object cache is a slave to inode cache (which, in turn
261 * is a slave to dentry cache), don't keep cl_object in memory
262 * when its master is evicted.
264 cl_object_kill(env, clob);
265 lu_object_ref_del(&clob->co_lu, "inode", inode);
266 cl_object_put_last(env, clob);
267 lli->lli_clob = NULL;
269 cl_env_unplant(cl_inode_fini_env, &refcheck);
270 mutex_unlock(&cl_inode_fini_guard);
272 cl_env_put(env, &refcheck);
274 cl_env_reexit(cookie);
279 * build inode number from passed @fid
281 __u64 cl_fid_build_ino(const struct lu_fid *fid, int api32)
283 if (BITS_PER_LONG == 32 || api32)
284 return fid_flatten32(fid);
286 return fid_flatten(fid);
290 * build inode generation from passed @fid. If our FID overflows the 32-bit
291 * inode number then return a non-zero generation to distinguish them.
293 __u32 cl_fid_build_gen(const struct lu_fid *fid)
297 if (fid_is_igif(fid)) {
298 gen = lu_igif_gen(fid);
302 gen = fid_flatten(fid) >> 32;
306 /* lsm is unreliable after hsm implementation as layout can be changed at
307 * any time. This is only to support old, non-clio-ized interfaces. It will
308 * cause deadlock if clio operations are called with this extra layout refcount
309 * because in case the layout changed during the IO, ll_layout_refresh() will
310 * have to wait for the refcount to become zero to destroy the older layout.
312 * Notice that the lsm returned by this function may not be valid unless called
313 * inside layout lock - MDS_INODELOCK_LAYOUT.
315 struct lov_stripe_md *ccc_inode_lsm_get(struct inode *inode)
317 return lov_lsm_get(ll_i2info(inode)->lli_clob);
320 inline void ccc_inode_lsm_put(struct inode *inode, struct lov_stripe_md *lsm)
322 lov_lsm_put(ll_i2info(inode)->lli_clob, lsm);