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, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * Author: Nikita Danilov <nikita.danilov@sun.com>
35 #define DEBUG_SUBSYSTEM S_LLITE
37 #include <libcfs/libcfs.h>
39 #include <linux/sched.h>
41 #include <linux/quotaops.h>
42 #include <linux/highmem.h>
43 #include <linux/pagemap.h>
44 #include <linux/rbtree.h>
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_dlm.h>
50 #include <lustre_mdc.h>
51 #include <cl_object.h>
53 #include "llite_internal.h"
54 #include "vvp_internal.h"
57 * An `emergency' environment used by cl_inode_fini() when cl_env_get()
58 * fails. Access to this environment is serialized by cl_inode_fini_guard
61 struct lu_env *cl_inode_fini_env;
62 __u16 cl_inode_fini_refcheck;
65 * A mutex serializing calls to slp_inode_fini() under extreme memory
66 * pressure, when environments cannot be allocated.
68 static DEFINE_MUTEX(cl_inode_fini_guard);
70 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
71 unsigned int attr_flags)
80 env = cl_env_get(&refcheck);
84 io = vvp_env_thread_io(env);
86 io->ci_verify_layout = 1;
88 io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime);
89 io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime);
90 io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime);
91 io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size;
92 io->u.ci_setattr.sa_attr_flags = attr_flags;
93 io->u.ci_setattr.sa_valid = attr->ia_valid;
94 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&obj->co_lu);
97 if (attr->ia_valid & ATTR_FILE)
98 ll_io_set_mirror(io, attr->ia_file);
100 if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0) {
101 struct vvp_io *vio = vvp_env_io(env);
103 if (attr->ia_valid & ATTR_FILE)
104 /* populate the file descriptor for ftruncate to honor
105 * group lock - see LU-787 */
106 vio->vui_fd = LUSTRE_FPRIVATE(attr->ia_file);
108 result = cl_io_loop(env, io);
110 result = io->ci_result;
113 if (unlikely(io->ci_need_restart))
116 cl_env_put(env, &refcheck);
121 * Initialize or update CLIO structures for regular files when new
122 * meta-data arrives from the server.
124 * \param inode regular file inode
125 * \param md new file metadata from MDS
126 * - allocates cl_object if necessary,
127 * - updated layout, if object was already here.
129 int cl_file_inode_init(struct inode *inode, struct lustre_md *md)
132 struct ll_inode_info *lli;
133 struct cl_object *clob;
134 struct lu_site *site;
136 struct cl_object_conf conf = {
139 .coc_layout = md->layout,
145 LASSERT(md->body->mbo_valid & OBD_MD_FLID);
146 LASSERT(S_ISREG(inode->i_mode));
148 env = cl_env_get(&refcheck);
152 site = ll_i2sbi(inode)->ll_site;
153 lli = ll_i2info(inode);
155 LASSERT(fid_is_sane(fid));
157 if (lli->lli_clob == NULL) {
158 /* clob is slave of inode, empty lli_clob means for new inode,
159 * there is no clob in cache with the given fid, so it is
160 * unnecessary to perform lookup-alloc-lookup-insert, just
161 * alloc and insert directly. */
162 LASSERT(inode->i_state & I_NEW);
163 conf.coc_lu.loc_flags = LOC_F_NEW;
164 clob = cl_object_find(env, lu2cl_dev(site->ls_top_dev),
168 * No locking is necessary, as new inode is
169 * locked by I_NEW bit.
171 lli->lli_clob = clob;
172 lu_object_ref_add(&clob->co_lu, "inode", inode);
174 result = PTR_ERR(clob);
176 result = cl_conf_set(env, lli->lli_clob, &conf);
177 if (result == -EBUSY) {
178 /* ignore the error since I/O will handle it later */
183 cl_env_put(env, &refcheck);
186 CERROR("Failure to initialize cl object "DFID": %d\n",
192 * Wait for others drop their references of the object at first, then we drop
193 * the last one, which will lead to the object be destroyed immediately.
194 * Must be called after cl_object_kill() against this object.
196 * The reason we want to do this is: destroying top object will wait for sub
197 * objects being destroyed first, so we can't let bottom layer (e.g. from ASTs)
198 * to initiate top object destroying which may deadlock. See bz22520.
200 static void cl_object_put_last(struct lu_env *env, struct cl_object *obj)
202 struct lu_object_header *header = obj->co_lu.lo_header;
203 wait_queue_entry_t waiter;
205 if (unlikely(atomic_read(&header->loh_ref) != 1)) {
206 struct lu_site *site = obj->co_lu.lo_dev->ld_site;
207 struct lu_site_bkt_data *bkt;
209 bkt = lu_site_bkt_from_fid(site, &header->loh_fid);
211 init_waitqueue_entry(&waiter, current);
212 add_wait_queue(&bkt->lsb_marche_funebre, &waiter);
215 set_current_state(TASK_UNINTERRUPTIBLE);
216 if (atomic_read(&header->loh_ref) == 1)
221 set_current_state(TASK_RUNNING);
222 remove_wait_queue(&bkt->lsb_marche_funebre, &waiter);
225 cl_object_put(env, obj);
228 void cl_inode_fini(struct inode *inode)
231 struct ll_inode_info *lli = ll_i2info(inode);
232 struct cl_object *clob = lli->lli_clob;
237 env = cl_env_get(&refcheck);
238 emergency = IS_ERR(env);
240 mutex_lock(&cl_inode_fini_guard);
241 LASSERT(cl_inode_fini_env != NULL);
242 env = cl_inode_fini_env;
246 * cl_object cache is a slave to inode cache (which, in turn
247 * is a slave to dentry cache), don't keep cl_object in memory
248 * when its master is evicted.
250 cl_object_kill(env, clob);
251 lu_object_ref_del(&clob->co_lu, "inode", inode);
252 cl_object_put_last(env, clob);
253 lli->lli_clob = NULL;
255 mutex_unlock(&cl_inode_fini_guard);
257 cl_env_put(env, &refcheck);
262 * build inode number from passed @fid.
264 * For 32-bit systems or syscalls limit the inode number to a 32-bit value
265 * to avoid EOVERFLOW errors. This will inevitably result in inode number
266 * collisions, but fid_flatten32() tries hard to avoid this if possible.
268 __u64 cl_fid_build_ino(const struct lu_fid *fid, int api32)
270 if (BITS_PER_LONG == 32 || api32)
271 RETURN(fid_flatten32(fid));
273 RETURN(fid_flatten(fid));
277 * build inode generation from passed @fid. If our FID overflows the 32-bit
278 * inode number then return a non-zero generation to distinguish them.
280 __u32 cl_fid_build_gen(const struct lu_fid *fid)
282 if (fid_is_igif(fid))
283 RETURN(lu_igif_gen(fid));
285 RETURN(fid_flatten(fid) >> 32);