lustre/llite/llite_capa.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2004, 2005 Cluster File Systems, Inc.
   5  *
   6  * Author: Lai Siyao <lsy@clusterfs.com>
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <linux/fs.h>
  27 #include <linux/version.h>
  28 #include <asm/uaccess.h>
  29 #include <linux/file.h>
  30 #include <linux/kmod.h>
  31
  32 #include <linux/lustre_lite.h>
  33 #include "llite_internal.h"
  34
  35 static struct list_head *ll_capa_list = &capa_list[CLIENT_CAPA];
  36 static struct ptlrpc_thread capa_thread;
  37
  38 static struct thread_ctl {
  39         struct completion ctl_starting;
  40         struct completion ctl_finishing;
  41 } ll_capa_ctl;
  42
  43 static inline int have_expired_capa(void)
  44 {
  45         struct obd_capa *ocapa;
  46         struct lustre_capa *capa;
  47         int expired = 0;
  48         unsigned long expiry;
  49         struct timeval tv;
  50
  51         ENTRY;
  52
  53         do_gettimeofday(&tv);
  54         spin_lock(&capa_lock);
  55         if (!list_empty(ll_capa_list)) {
  56                 ocapa = list_entry(ll_capa_list->next, struct obd_capa, c_list);
  57                 expired = __capa_is_to_expire(ocapa, &tv);
  58                 if (!expired) {
  59                         capa = &ocapa->c_capa;
  60                         expiry = expiry_to_jiffies(capa->lc_expiry -
  61                                                    capa_pre_expiry(capa));
  62                         if (time_before(expiry, ll_capa_timer.expires) ||
  63                             !timer_pending(&ll_capa_timer)) {
  64                                 mod_timer(&ll_capa_timer, expiry);
  65                                 CDEBUG(D_INFO,"ll_capa_timer new expiry: %lu\n",
  66                                        expiry);
  67                         }
  68                 }
  69         }
  70         spin_unlock(&capa_lock);
  71
  72         RETURN(expired);
  73 }
  74
  75 static int inline ll_capa_check_stop(void)
  76 {
  77         return (capa_thread.t_flags & SVC_STOPPING) ? 1: 0;
  78 }
  79
  80 static int ll_renew_capa(struct obd_capa *ocapa)
  81 {
  82         struct ptlrpc_request *req = NULL;
  83         /* no need to lock, no one else will touch it */
  84         struct inode *inode = ocapa->c_inode;
  85         struct obd_export *md_exp = ll_i2mdexp(inode);
  86         struct ll_inode_info *lli = ll_i2info(inode);
  87         __u64 valid = OBD_MD_CAPA;
  88         int rc;
  89         ENTRY;
  90
  91         rc = md_getattr(md_exp, &lli->lli_id, valid, NULL, NULL, 0,
  92                         0, ocapa, &req);
  93         RETURN(rc);
  94 }
  95
  96 static int ll_capa_thread(void *arg)
  97 {
  98         struct thread_ctl *ctl = arg;
  99         unsigned long flags;
 100         int rc;
 101         ENTRY;
 102
 103         {
 104                 char name[sizeof(current->comm)];
 105                 snprintf(name, sizeof(name) - 1, "ll_capa");
 106                 kportal_daemonize(name);
 107         }
 108
 109         SIGNAL_MASK_LOCK(current, flags);
 110         sigfillset(&current->blocked);
 111         RECALC_SIGPENDING;
 112         SIGNAL_MASK_UNLOCK(current, flags);
 113
 114         /*
 115          * letting starting function know, that we are ready and control may be
 116          * returned.
 117          */
 118         capa_thread.t_flags = SVC_RUNNING;
 119         complete(&ctl->ctl_starting);
 120
 121         while (1) {
 122                 struct l_wait_info lwi = { 0 };
 123                 struct obd_capa *ocapa, tcapa, *tmp, *next = NULL;
 124                 unsigned long expiry, sleep = CAPA_PRE_EXPIRY;
 125                 struct inode *inode;
 126                 struct timeval tv;
 127
 128                 l_wait_event(capa_thread.t_ctl_waitq,
 129                              (have_expired_capa() || ll_capa_check_stop()),
 130                              &lwi);
 131
 132                 if (ll_capa_check_stop())
 133                         break;
 134
 135                 do_gettimeofday(&tv);
 136                 spin_lock(&capa_lock);
 137                 list_for_each_entry_safe(ocapa, tmp, ll_capa_list, c_list) {
 138                         if (ocapa->c_capa.lc_flags & CAPA_FL_SHORT)
 139                                 sleep = CAPA_PRE_EXPIRY_SHORT;
 140
 141                         if (ocapa->c_capa.lc_op == CAPA_TRUNC)
 142                                 continue;
 143
 144                         if (capa_expired(&ocapa->c_capa)) {
 145                                 capa_put_nolock(ocapa);
 146                                 continue;
 147                         }
 148
 149                         if (__capa_is_to_expire(ocapa, &tv)) {
 150                                 inode = igrab(ocapa->c_inode);
 151                                 if (inode == NULL) {
 152                                         DEBUG_CAPA(D_ERROR, &ocapa->c_capa,
 153                                                    "igrab failed for");
 154                                         continue;
 155                                 }
 156
 157                                 tcapa = *ocapa;
 158                                 spin_unlock(&capa_lock);
 159
 160                                 rc = ll_renew_capa(&tcapa);
 161                                 iput(inode);
 162
 163                                 spin_lock(&capa_lock);
 164                         } else {
 165                                 next = ocapa;
 166                                 break;
 167                         }
 168                 }
 169
 170                 if (next) {
 171                         struct lustre_capa *capa = &next->c_capa;
 172
 173                         expiry = expiry_to_jiffies(capa->lc_expiry -
 174                                                    capa_pre_expiry(capa));
 175                         if (time_before(expiry, ll_capa_timer.expires) ||
 176                             !timer_pending(&ll_capa_timer)) {
 177                                 mod_timer(&ll_capa_timer, expiry);
 178                                 CDEBUG(D_INFO,"ll_capa_timer new expiry: %lu\n",
 179                                        expiry);
 180                         }
 181                 }
 182                 spin_unlock(&capa_lock);
 183
 184                 /* wait ll_renew_capa finish */
 185                 set_current_state(TASK_INTERRUPTIBLE);
 186                 schedule_timeout(sleep * HZ);
 187         }
 188
 189         capa_thread.t_flags = SVC_STOPPED;
 190
 191         /* this is SMP-safe way to finish thread. */
 192         complete_and_exit(&ctl->ctl_finishing, 0);
 193         EXIT;
 194 }
 195
 196 /* just wake up, others are handled by ll_capa_thread */
 197 void ll_capa_timer_callback(unsigned long unused)
 198 {
 199         ENTRY;
 200         wake_up(&capa_thread.t_ctl_waitq);
 201         EXIT;
 202 }
 203
 204 int ll_capa_thread_start(void)
 205 {
 206         int rc;
 207         ENTRY;
 208
 209         LASSERT(capa_thread.t_flags == 0);
 210         init_completion(&ll_capa_ctl.ctl_starting);
 211         init_completion(&ll_capa_ctl.ctl_finishing);
 212         init_waitqueue_head(&capa_thread.t_ctl_waitq);
 213
 214         rc = kernel_thread(ll_capa_thread, &ll_capa_ctl,
 215                            (CLONE_VM | CLONE_FILES));
 216         if (rc < 0) {
 217                 CERROR("cannot start expired capa thread, "
 218                        "err = %d\n", rc);
 219                 RETURN(rc);
 220         }
 221         wait_for_completion(&ll_capa_ctl.ctl_starting);
 222         LASSERT(capa_thread.t_flags == SVC_RUNNING);
 223         RETURN(0);
 224 }
 225
 226 void ll_capa_thread_stop(void)
 227 {
 228         ENTRY;
 229
 230         capa_thread.t_flags = SVC_STOPPING;
 231         wake_up(&capa_thread.t_ctl_waitq);
 232         wait_for_completion(&ll_capa_ctl.ctl_finishing);
 233         LASSERT(capa_thread.t_flags == SVC_STOPPED);
 234         capa_thread.t_flags = 0;
 235
 236         EXIT;
 237 }
 238
 239 int ll_set_capa(struct inode *inode, struct lookup_intent *it,
 240                 struct obd_client_handle *och)
 241 {
 242         struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
 243         struct mds_body *body;
 244         struct lustre_capa *capa;
 245         struct obd_capa *ocapa;
 246         struct ll_inode_info *lli = ll_i2info(inode);
 247         unsigned long expiry;
 248
 249         if (!S_ISREG(inode->i_mode))
 250                 return 0;
 251
 252         /* GNS code path will have no req */
 253         if (!req)
 254                 return 0;
 255
 256         body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
 257         LASSERT(body != NULL);          /* reply already checked out */
 258         LASSERT_REPSWABBED(req, 1);     /* and swabbed down */
 259
 260         if (!(body->valid & OBD_MD_CAPA)) {
 261                 if (atomic_read(&ll_capa_stat))
 262                         DEBUG_REQ(D_ERROR, req,
 263                                   "no capa for (uid %u, op %d, "DLID4"\n",
 264                                    (unsigned)current->uid, it->it_flags,
 265                                    OLID4(&lli->lli_id));
 266
 267                 return 0;
 268         }
 269
 270         ENTRY;
 271
 272         capa = lustre_msg_buf(req->rq_repmsg, 7, sizeof (*capa));
 273         LASSERT(capa != NULL);          /* reply already checked out */
 274         LASSERT_REPSWABBED(req, 7);     /* and swabbed down */
 275
 276         ocapa = capa_renew(capa, CLIENT_CAPA);
 277         if (!ocapa)
 278                 RETURN(-ENOMEM);
 279
 280         spin_lock(&capa_lock);
 281         ocapa->c_inode = inode;
 282         ocapa->c_handle = och->och_fh;
 283         spin_unlock(&capa_lock);
 284
 285         spin_lock(&lli->lli_lock);
 286         /* in case it was linked to lli_capas already */
 287         if (list_empty(&ocapa->c_lli_list))
 288                 list_add(&ocapa->c_lli_list, &lli->lli_capas);
 289         spin_unlock(&lli->lli_lock);
 290
 291         expiry = expiry_to_jiffies(capa->lc_expiry - capa_pre_expiry(capa));
 292
 293         spin_lock(&capa_lock);
 294         if (time_before(expiry, ll_capa_timer.expires) ||
 295             !timer_pending(&ll_capa_timer)) {
 296                 mod_timer(&ll_capa_timer, expiry);
 297                 CDEBUG(D_INFO, "ll_capa_timer new expiry: %lu\n", expiry);
 298         }
 299         spin_unlock(&capa_lock);
 300
 301         RETURN(0);
 302 }
 303
 304 int ll_set_trunc_capa(struct ptlrpc_request *req, int offset, struct inode *inode)
 305 {
 306         struct mds_body *body;
 307         struct obd_capa *ocapa;
 308         struct lustre_capa *capa;
 309         struct ll_inode_info *lli = ll_i2info(inode);
 310
 311         body = lustre_msg_buf(req->rq_repmsg, offset, sizeof(*body));
 312         if (!body)
 313                 return -ENOMEM;
 314
 315         if (!(body->valid & OBD_MD_CAPA))
 316                 return 0;
 317
 318         ENTRY;
 319         capa = (struct lustre_capa *)lustre_swab_repbuf(req, offset + 1,
 320                                         sizeof(*capa), lustre_swab_lustre_capa);
 321         if (!capa)
 322                 RETURN(-ENOMEM);
 323
 324         ocapa = capa_renew(capa, CLIENT_CAPA);
 325         if (!ocapa)
 326                 RETURN(-ENOMEM);
 327
 328         spin_lock(&lli->lli_lock);
 329         /* in case it was linked to lli_capas already */
 330         if (list_empty(&ocapa->c_lli_list))
 331                 list_add(&ocapa->c_lli_list, &lli->lli_capas);
 332         spin_unlock(&lli->lli_lock);
 333
 334         RETURN(0);
 335 }
 336
 337 struct obd_capa *ll_get_capa(struct inode *inode, uid_t uid, int op)
 338 {
 339         struct ll_inode_info *lli = ll_i2info(inode);
 340         struct obd_capa *ocapa, *tmp;
 341         ENTRY;
 342
 343         list_for_each_entry_safe(ocapa, tmp, &lli->lli_capas, c_lli_list) {
 344                 if (ocapa->c_capa.lc_ruid != uid)
 345                         continue;
 346                 if (ocapa->c_capa.lc_op != op)
 347                         continue;
 348
 349                 RETURN(ocapa);
 350         }
 351
 352         if (atomic_read(&ll_capa_stat)) {
 353                 CDEBUG(D_ERROR, "can't find capa for (uid %u, op %d, mdsid "
 354                        LPU64", ino %u igen %u) failed.\n",
 355                        (unsigned)uid, op, id_group(&lli->lli_id),
 356                        (unsigned)id_ino(&lli->lli_id), id_gen(&lli->lli_id));
 357                 atomic_set(&ll_capa_stat, 0);
 358         }
 359
 360         RETURN(NULL);
 361 }