lustre/llite/vvp_dev.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * cl_device and cl_device_type implementation for VVP layer.
  37  *
  38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #ifndef __KERNEL__
  44 # error This file is kernel only.
  45 #endif
  46
  47 #include <obd.h>
  48 #include <lustre_lite.h>
  49
  50 #include "vvp_internal.h"
  51
  52 /*****************************************************************************
  53  *
  54  * Vvp device and device type functions.
  55  *
  56  */
  57
  58 /*
  59  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  60  * "llite_" (var. "ll_") prefix.
  61  */
  62
  63 cfs_mem_cache_t *vvp_page_kmem;
  64 cfs_mem_cache_t *vvp_thread_kmem;
  65 static cfs_mem_cache_t *vvp_session_kmem;
  66 static struct lu_kmem_descr vvp_caches[] = {
  67         {
  68                 .ckd_cache = &vvp_page_kmem,
  69                 .ckd_name  = "vvp_page_kmem",
  70                 .ckd_size  = sizeof (struct ccc_page)
  71         },
  72         {
  73                 .ckd_cache = &vvp_thread_kmem,
  74                 .ckd_name  = "vvp_thread_kmem",
  75                 .ckd_size  = sizeof (struct vvp_thread_info),
  76         },
  77         {
  78                 .ckd_cache = &vvp_session_kmem,
  79                 .ckd_name  = "vvp_session_kmem",
  80                 .ckd_size  = sizeof (struct vvp_session)
  81         },
  82         {
  83                 .ckd_cache = NULL
  84         }
  85 };
  86
  87 static void *vvp_key_init(const struct lu_context *ctx,
  88                           struct lu_context_key *key)
  89 {
  90         struct vvp_thread_info *info;
  91
  92         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, CFS_ALLOC_IO);
  93         if (info == NULL)
  94                 info = ERR_PTR(-ENOMEM);
  95         return info;
  96 }
  97
  98 static void vvp_key_fini(const struct lu_context *ctx,
  99                          struct lu_context_key *key, void *data)
 100 {
 101         struct vvp_thread_info *info = data;
 102         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
 103 }
 104
 105 static void *vvp_session_key_init(const struct lu_context *ctx,
 106                                   struct lu_context_key *key)
 107 {
 108         struct vvp_session *session;
 109
 110         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, CFS_ALLOC_IO);
 111         if (session == NULL)
 112                 session = ERR_PTR(-ENOMEM);
 113         return session;
 114 }
 115
 116 static void vvp_session_key_fini(const struct lu_context *ctx,
 117                                  struct lu_context_key *key, void *data)
 118 {
 119         struct vvp_session *session = data;
 120         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 121 }
 122
 123
 124 struct lu_context_key vvp_key = {
 125         .lct_tags = LCT_CL_THREAD,
 126         .lct_init = vvp_key_init,
 127         .lct_fini = vvp_key_fini
 128 };
 129
 130 struct lu_context_key vvp_session_key = {
 131         .lct_tags = LCT_SESSION,
 132         .lct_init = vvp_session_key_init,
 133         .lct_fini = vvp_session_key_fini
 134 };
 135
 136 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 137 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
 138
 139 static const struct lu_device_operations vvp_lu_ops = {
 140         .ldo_object_alloc      = vvp_object_alloc
 141 };
 142
 143 static const struct cl_device_operations vvp_cl_ops = {
 144         .cdo_req_init = ccc_req_init
 145 };
 146
 147 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 148                                           struct lu_device_type *t,
 149                                           struct lustre_cfg *cfg)
 150 {
 151         return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
 152 }
 153
 154 static const struct lu_device_type_operations vvp_device_type_ops = {
 155         .ldto_init = vvp_type_init,
 156         .ldto_fini = vvp_type_fini,
 157
 158         .ldto_start = vvp_type_start,
 159         .ldto_stop  = vvp_type_stop,
 160
 161         .ldto_device_alloc = vvp_device_alloc,
 162         .ldto_device_free  = ccc_device_free,
 163         .ldto_device_init  = ccc_device_init,
 164         .ldto_device_fini  = ccc_device_fini
 165 };
 166
 167 struct lu_device_type vvp_device_type = {
 168         .ldt_tags     = LU_DEVICE_CL,
 169         .ldt_name     = LUSTRE_VVP_NAME,
 170         .ldt_ops      = &vvp_device_type_ops,
 171         .ldt_ctx_tags = LCT_CL_THREAD
 172 };
 173
 174 /**
 175  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 176  * pressure, when environments cannot be allocated.
 177  */
 178 int vvp_global_init(void)
 179 {
 180         int result;
 181
 182         result = lu_kmem_init(vvp_caches);
 183         if (result == 0) {
 184                 result = ccc_global_init(&vvp_device_type);
 185                 if (result != 0)
 186                         lu_kmem_fini(vvp_caches);
 187         }
 188         return result;
 189 }
 190
 191 void vvp_global_fini(void)
 192 {
 193         ccc_global_fini(&vvp_device_type);
 194         lu_kmem_fini(vvp_caches);
 195 }
 196
 197
 198 /*****************************************************************************
 199  *
 200  * mirror obd-devices into cl devices.
 201  *
 202  */
 203
 204 int cl_sb_init(struct super_block *sb)
 205 {
 206         struct ll_sb_info *sbi;
 207         struct cl_device  *cl;
 208         struct lu_env     *env;
 209         int rc = 0;
 210         int refcheck;
 211
 212         sbi  = ll_s2sbi(sb);
 213         env = cl_env_get(&refcheck);
 214         if (!IS_ERR(env)) {
 215                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 216                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 217                 if (!IS_ERR(cl)) {
 218                         cl2ccc_dev(cl)->cdv_sb = sb;
 219                         sbi->ll_cl = cl;
 220                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 221                 }
 222                 cl_env_put(env, &refcheck);
 223         } else
 224                 rc = PTR_ERR(env);
 225         RETURN(rc);
 226 }
 227
 228 int cl_sb_fini(struct super_block *sb)
 229 {
 230         struct ll_sb_info *sbi;
 231         struct lu_env     *env;
 232         struct cl_device  *cld;
 233         int                refcheck;
 234         int                result;
 235
 236         ENTRY;
 237         sbi = ll_s2sbi(sb);
 238         env = cl_env_get(&refcheck);
 239         if (!IS_ERR(env)) {
 240                 cld = sbi->ll_cl;
 241
 242                 if (cld != NULL) {
 243                         cl_stack_fini(env, cld);
 244                         sbi->ll_cl = NULL;
 245                         sbi->ll_site = NULL;
 246                 }
 247                 cl_env_put(env, &refcheck);
 248                 result = 0;
 249         } else {
 250                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 251                 result = PTR_ERR(env);
 252         }
 253         /*
 254          * If mount failed (sbi->ll_cl == NULL), and this there are no other
 255          * mounts, stop device types manually (this usually happens
 256          * automatically when last device is destroyed).
 257          */
 258         lu_types_stop();
 259         RETURN(result);
 260 }
 261
 262 /****************************************************************************
 263  *
 264  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 265  *
 266  ****************************************************************************/
 267
 268 /*
 269  * To represent contents of a page cache as a byte stream, following
 270  * information if encoded in 64bit offset:
 271  *
 272  *       - file hash bucket in lu_site::ls_hash[]       28bits
 273  *
 274  *       - how far file is from bucket head              4bits
 275  *
 276  *       - page index                                   32bits
 277  *
 278  * First two data identify a file in the cache uniquely.
 279  */
 280
 281 #define PGC_OBJ_SHIFT (32 + 4)
 282 #define PGC_DEPTH_SHIFT (32)
 283
 284 struct vvp_pgcache_id {
 285         unsigned vpi_bucket;
 286         unsigned vpi_depth;
 287         uint32_t vpi_index;
 288 };
 289
 290 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 291 {
 292         CLASSERT(sizeof(pos) == sizeof(__u64));
 293
 294         id->vpi_index  = pos & 0xffffffff;
 295         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 296         id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
 297 }
 298
 299 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 300 {
 301         return
 302                 ((__u64)id->vpi_index) |
 303                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 304                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 305 }
 306
 307 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 308                                          struct lu_device *dev,
 309                                          struct vvp_pgcache_id *id)
 310 {
 311         struct hlist_head       *bucket;
 312         struct lu_object_header *hdr;
 313         struct lu_site          *site;
 314         struct hlist_node       *scan;
 315         struct lu_object_header *found;
 316         struct cl_object        *clob;
 317         unsigned                 depth;
 318
 319         LASSERT(lu_device_is_cl(dev));
 320
 321         site   = dev->ld_site;
 322         bucket = site->ls_hash + (id->vpi_bucket & site->ls_hash_mask);
 323         depth  = id->vpi_depth & 0xf;
 324         found  = NULL;
 325         clob   = NULL;
 326
 327         /* XXX copy of lu_object.c:htable_lookup() */
 328         read_lock(&site->ls_guard);
 329         hlist_for_each_entry(hdr, scan, bucket, loh_hash) {
 330                 if (depth-- == 0) {
 331                         if (!lu_object_is_dying(hdr)) {
 332                                 if (atomic_add_return(1, &hdr->loh_ref) == 1)
 333                                         ++ site->ls_busy;
 334                                 found = hdr;
 335                         }
 336                         break;
 337                 }
 338         }
 339         read_unlock(&site->ls_guard);
 340
 341         if (found != NULL) {
 342                 struct lu_object *lu_obj;
 343
 344                 lu_obj = lu_object_locate(found, dev->ld_type);
 345                 if (lu_obj != NULL) {
 346                         lu_object_ref_add(lu_obj, "dump", cfs_current());
 347                         clob = lu2cl(lu_obj);
 348                 } else
 349                         lu_object_put(env, lu_object_top(found));
 350         } else if (depth > 0)
 351                 id->vpi_depth = 0xf;
 352         return clob;
 353 }
 354
 355 static loff_t vvp_pgcache_find(const struct lu_env *env,
 356                                struct lu_device *dev, loff_t pos)
 357 {
 358         struct cl_object     *clob;
 359         struct lu_site       *site;
 360         struct vvp_pgcache_id id;
 361
 362         site = dev->ld_site;
 363         vvp_pgcache_id_unpack(pos, &id);
 364
 365         while (1) {
 366                 if (id.vpi_bucket >= site->ls_hash_size)
 367                         return ~0ULL;
 368                 clob = vvp_pgcache_obj(env, dev, &id);
 369                 if (clob != NULL) {
 370                         struct cl_object_header *hdr;
 371                         int                      nr;
 372                         struct cl_page          *pg;
 373
 374                         /* got an object. Find next page. */
 375                         hdr = cl_object_header(clob);
 376
 377                         spin_lock(&hdr->coh_page_guard);
 378                         nr = radix_tree_gang_lookup(&hdr->coh_tree,
 379                                                     (void **)&pg,
 380                                                     id.vpi_index, 1);
 381                         if (nr > 0) {
 382                                 id.vpi_index = pg->cp_index;
 383                                 /* Cant support over 16T file */
 384                                 nr = !(pg->cp_index > 0xffffffff);
 385                         }
 386                         spin_unlock(&hdr->coh_page_guard);
 387
 388                         lu_object_ref_del(&clob->co_lu, "dump", cfs_current());
 389                         cl_object_put(env, clob);
 390                         if (nr > 0)
 391                                 return vvp_pgcache_id_pack(&id);
 392                 }
 393                 /* to the next object. */
 394                 ++id.vpi_depth;
 395                 id.vpi_depth &= 0xf;
 396                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 397                         return ~0ULL;
 398                 id.vpi_index = 0;
 399         }
 400 }
 401
 402 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 403         if (test_bit(PG_##flag, &(page)->flags)) {                      \
 404                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 405                 has_flags = 1;                                          \
 406         }                                                               \
 407 } while(0);
 408
 409 static void vvp_pgcache_page_show(const struct lu_env *env,
 410                                   struct seq_file *seq, struct cl_page *page)
 411 {
 412         struct ccc_page *cpg;
 413         cfs_page_t      *vmpage;
 414         int              has_flags;
 415
 416         cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
 417         vmpage = cpg->cpg_page;
 418         seq_printf(seq," %5i | %p %p %s %s %s %s | %p %lu/%u(%p) %lu %u [",
 419                    0 /* gen */,
 420                    cpg, page,
 421                    "none",
 422                    cpg->cpg_write_queued ? "wq" : "- ",
 423                    cpg->cpg_defer_uptodate ? "du" : "- ",
 424                    PageWriteback(vmpage) ? "wb" : "-",
 425                    vmpage, vmpage->mapping->host->i_ino,
 426                    vmpage->mapping->host->i_generation,
 427                    vmpage->mapping->host, vmpage->index,
 428                    page_count(vmpage));
 429         has_flags = 0;
 430         seq_page_flag(seq, vmpage, locked, has_flags);
 431         seq_page_flag(seq, vmpage, error, has_flags);
 432         seq_page_flag(seq, vmpage, referenced, has_flags);
 433         seq_page_flag(seq, vmpage, uptodate, has_flags);
 434         seq_page_flag(seq, vmpage, dirty, has_flags);
 435 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
 436         seq_page_flag(seq, vmpage, highmem, has_flags);
 437 #endif
 438         seq_page_flag(seq, vmpage, writeback, has_flags);
 439         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 440 }
 441
 442 static int vvp_pgcache_show(struct seq_file *f, void *v)
 443 {
 444         loff_t                   pos;
 445         struct ll_sb_info       *sbi;
 446         struct cl_object        *clob;
 447         struct lu_env           *env;
 448         struct cl_page          *page;
 449         struct cl_object_header *hdr;
 450         struct vvp_pgcache_id    id;
 451         int                      refcheck;
 452         int                      result;
 453
 454         env = cl_env_get(&refcheck);
 455         if (!IS_ERR(env)) {
 456                 pos = *(loff_t *) v;
 457                 vvp_pgcache_id_unpack(pos, &id);
 458                 sbi = f->private;
 459                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 460                 if (clob != NULL) {
 461                         hdr = cl_object_header(clob);
 462
 463                         spin_lock(&hdr->coh_page_guard);
 464                         page = cl_page_lookup(hdr, id.vpi_index);
 465                         spin_unlock(&hdr->coh_page_guard);
 466
 467                         seq_printf(f, "%8x@"DFID": ",
 468                                    id.vpi_index, PFID(&hdr->coh_lu.loh_fid));
 469                         if (page != NULL) {
 470                                 vvp_pgcache_page_show(env, f, page);
 471                                 cl_page_put(env, page);
 472                         } else
 473                                 seq_puts(f, "missing\n");
 474                         lu_object_ref_del(&clob->co_lu, "dump", cfs_current());
 475                         cl_object_put(env, clob);
 476                 } else
 477                         seq_printf(f, "%llx missing\n", pos);
 478                 cl_env_put(env, &refcheck);
 479                 result = 0;
 480         } else
 481                 result = PTR_ERR(env);
 482         return result;
 483 }
 484
 485 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 486 {
 487         struct ll_sb_info *sbi;
 488         struct lu_env     *env;
 489         int                refcheck;
 490
 491         sbi = f->private;
 492
 493         env = cl_env_get(&refcheck);
 494         if (!IS_ERR(env)) {
 495                 sbi = f->private;
 496                 if (sbi->ll_site->ls_hash_bits > 64 - PGC_OBJ_SHIFT)
 497                         pos = ERR_PTR(-EFBIG);
 498                 else {
 499                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 500                                                 *pos);
 501                         if (*pos == ~0ULL)
 502                                 pos = NULL;
 503                 }
 504                 cl_env_put(env, &refcheck);
 505         }
 506         return pos;
 507 }
 508
 509 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 510 {
 511         struct ll_sb_info *sbi;
 512         struct lu_env     *env;
 513         int                refcheck;
 514
 515         env = cl_env_get(&refcheck);
 516         if (!IS_ERR(env)) {
 517                 sbi = f->private;
 518                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 519                 if (*pos == ~0ULL)
 520                         pos = NULL;
 521                 cl_env_put(env, &refcheck);
 522         }
 523         return pos;
 524 }
 525
 526 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 527 {
 528         /* Nothing to do */
 529 }
 530
 531 static struct seq_operations vvp_pgcache_ops = {
 532         .start = vvp_pgcache_start,
 533         .next  = vvp_pgcache_next,
 534         .stop  = vvp_pgcache_stop,
 535         .show  = vvp_pgcache_show
 536 };
 537
 538 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 539 {
 540         struct proc_dir_entry *dp  = PDE(inode);
 541         struct ll_sb_info     *sbi = dp->data;
 542         struct seq_file       *seq;
 543         int                    result;
 544
 545         result = seq_open(filp, &vvp_pgcache_ops);
 546         if (result == 0) {
 547                 seq = filp->private_data;
 548                 seq->private = sbi;
 549         }
 550         return result;
 551 }
 552
 553 struct file_operations vvp_dump_pgcache_file_ops = {
 554         .owner   = THIS_MODULE,
 555         .open    = vvp_dump_pgcache_seq_open,
 556         .read    = seq_read,
 557         .llseek  = seq_lseek,
 558         .release = seq_release,
 559 };