lustre/llite/vvp_dev.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   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.
   9  *
  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).
  15  *
  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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2014, Intel Corporation.
  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  *   Author: Jinshan Xiong <jinshan.xiong@intel.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_LLITE
  43
  44
  45 #include <obd.h>
  46 #include "llite_internal.h"
  47 #include "vvp_internal.h"
  48
  49 /*****************************************************************************
  50  *
  51  * Vvp device and device type functions.
  52  *
  53  */
  54
  55 /*
  56  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  57  * "llite_" (var. "ll_") prefix.
  58  */
  59
  60 struct kmem_cache *vvp_lock_kmem;
  61 struct kmem_cache *vvp_object_kmem;
  62 struct kmem_cache *vvp_req_kmem;
  63 static struct kmem_cache *vvp_thread_kmem;
  64 static struct kmem_cache *vvp_session_kmem;
  65 static struct lu_kmem_descr vvp_caches[] = {
  66         {
  67                 .ckd_cache = &vvp_lock_kmem,
  68                 .ckd_name  = "vvp_lock_kmem",
  69                 .ckd_size  = sizeof(struct vvp_lock),
  70         },
  71         {
  72                 .ckd_cache = &vvp_object_kmem,
  73                 .ckd_name  = "vvp_object_kmem",
  74                 .ckd_size  = sizeof(struct vvp_object),
  75         },
  76         {
  77                 .ckd_cache = &vvp_req_kmem,
  78                 .ckd_name  = "vvp_req_kmem",
  79                 .ckd_size  = sizeof(struct vvp_req),
  80         },
  81         {
  82                 .ckd_cache = &vvp_thread_kmem,
  83                 .ckd_name  = "vvp_thread_kmem",
  84                 .ckd_size  = sizeof (struct vvp_thread_info),
  85         },
  86         {
  87                 .ckd_cache = &vvp_session_kmem,
  88                 .ckd_name  = "vvp_session_kmem",
  89                 .ckd_size  = sizeof (struct vvp_session)
  90         },
  91         {
  92                 .ckd_cache = NULL
  93         }
  94 };
  95
  96 static void *vvp_key_init(const struct lu_context *ctx,
  97                           struct lu_context_key *key)
  98 {
  99         struct vvp_thread_info *info;
 100
 101         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, GFP_NOFS);
 102         if (info == NULL)
 103                 info = ERR_PTR(-ENOMEM);
 104         return info;
 105 }
 106
 107 static void vvp_key_fini(const struct lu_context *ctx,
 108                          struct lu_context_key *key, void *data)
 109 {
 110         struct vvp_thread_info *info = data;
 111         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
 112 }
 113
 114 static void *vvp_session_key_init(const struct lu_context *ctx,
 115                                   struct lu_context_key *key)
 116 {
 117         struct vvp_session *session;
 118
 119         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
 120         if (session == NULL)
 121                 session = ERR_PTR(-ENOMEM);
 122         return session;
 123 }
 124
 125 static void vvp_session_key_fini(const struct lu_context *ctx,
 126                                  struct lu_context_key *key, void *data)
 127 {
 128         struct vvp_session *session = data;
 129         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 130 }
 131
 132
 133 struct lu_context_key vvp_key = {
 134         .lct_tags = LCT_CL_THREAD,
 135         .lct_init = vvp_key_init,
 136         .lct_fini = vvp_key_fini
 137 };
 138
 139 struct lu_context_key vvp_session_key = {
 140         .lct_tags = LCT_SESSION,
 141         .lct_init = vvp_session_key_init,
 142         .lct_fini = vvp_session_key_fini
 143 };
 144
 145 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 146 LU_TYPE_INIT_FINI(vvp, &ccc_key, &vvp_key, &vvp_session_key);
 147
 148 static const struct lu_device_operations vvp_lu_ops = {
 149         .ldo_object_alloc      = vvp_object_alloc
 150 };
 151
 152 static const struct cl_device_operations vvp_cl_ops = {
 153         .cdo_req_init = vvp_req_init,
 154 };
 155
 156 static struct lu_device *vvp_device_free(const struct lu_env *env,
 157                                          struct lu_device *d)
 158 {
 159         struct vvp_device *vdv  = lu2vvp_dev(d);
 160         struct cl_site    *site = lu2cl_site(d->ld_site);
 161         struct lu_device  *next = cl2lu_dev(vdv->vdv_next);
 162
 163         if (d->ld_site != NULL) {
 164                 cl_site_fini(site);
 165                 OBD_FREE_PTR(site);
 166         }
 167
 168         cl_device_fini(lu2cl_dev(d));
 169         OBD_FREE_PTR(vdv);
 170         return next;
 171 }
 172
 173 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 174                                           struct lu_device_type *t,
 175                                           struct lustre_cfg *cfg)
 176 {
 177         struct vvp_device *vdv;
 178         struct lu_device *lud;
 179         struct cl_site *site;
 180         int rc;
 181         ENTRY;
 182
 183         OBD_ALLOC_PTR(vdv);
 184         if (vdv == NULL)
 185                 RETURN(ERR_PTR(-ENOMEM));
 186
 187         lud = &vdv->vdv_cl.cd_lu_dev;
 188         cl_device_init(&vdv->vdv_cl, t);
 189         vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
 190         vdv->vdv_cl.cd_ops = &vvp_cl_ops;
 191
 192         OBD_ALLOC_PTR(site);
 193         if (site != NULL) {
 194                 rc = cl_site_init(site, &vdv->vdv_cl);
 195                 if (rc == 0)
 196                         rc = lu_site_init_finish(&site->cs_lu);
 197                 else {
 198                         LASSERT(lud->ld_site == NULL);
 199                         CERROR("Cannot init lu_site, rc %d.\n", rc);
 200                         OBD_FREE_PTR(site);
 201                 }
 202         } else
 203                 rc = -ENOMEM;
 204         if (rc != 0) {
 205                 vvp_device_free(env, lud);
 206                 lud = ERR_PTR(rc);
 207         }
 208         RETURN(lud);
 209 }
 210
 211 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
 212                            const char *name, struct lu_device *next)
 213 {
 214         struct vvp_device  *vdv;
 215         int rc;
 216         ENTRY;
 217
 218         vdv = lu2vvp_dev(d);
 219         vdv->vdv_next = lu2cl_dev(next);
 220
 221         LASSERT(d->ld_site != NULL && next->ld_type != NULL);
 222         next->ld_site = d->ld_site;
 223         rc = next->ld_type->ldt_ops->ldto_device_init(
 224                 env, next, next->ld_type->ldt_name, NULL);
 225         if (rc == 0) {
 226                 lu_device_get(next);
 227                 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
 228         }
 229         RETURN(rc);
 230 }
 231
 232 static struct lu_device *vvp_device_fini(const struct lu_env *env,
 233                                          struct lu_device *d)
 234 {
 235         return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
 236 }
 237
 238 static const struct lu_device_type_operations vvp_device_type_ops = {
 239         .ldto_init = vvp_type_init,
 240         .ldto_fini = vvp_type_fini,
 241
 242         .ldto_start = vvp_type_start,
 243         .ldto_stop  = vvp_type_stop,
 244
 245         .ldto_device_alloc      = vvp_device_alloc,
 246         .ldto_device_free       = vvp_device_free,
 247         .ldto_device_init       = vvp_device_init,
 248         .ldto_device_fini       = vvp_device_fini,
 249 };
 250
 251 struct lu_device_type vvp_device_type = {
 252         .ldt_tags     = LU_DEVICE_CL,
 253         .ldt_name     = LUSTRE_VVP_NAME,
 254         .ldt_ops      = &vvp_device_type_ops,
 255         .ldt_ctx_tags = LCT_CL_THREAD
 256 };
 257
 258 /**
 259  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 260  * pressure, when environments cannot be allocated.
 261  */
 262 int vvp_global_init(void)
 263 {
 264         int result;
 265
 266         result = lu_kmem_init(vvp_caches);
 267         if (result == 0) {
 268                 result = ccc_global_init(&vvp_device_type);
 269                 if (result != 0)
 270                         lu_kmem_fini(vvp_caches);
 271         }
 272         return result;
 273 }
 274
 275 void vvp_global_fini(void)
 276 {
 277         ccc_global_fini(&vvp_device_type);
 278         lu_kmem_fini(vvp_caches);
 279 }
 280
 281
 282 /*****************************************************************************
 283  *
 284  * mirror obd-devices into cl devices.
 285  *
 286  */
 287
 288 int cl_sb_init(struct super_block *sb)
 289 {
 290         struct ll_sb_info *sbi;
 291         struct cl_device  *cl;
 292         struct lu_env     *env;
 293         int rc = 0;
 294         int refcheck;
 295
 296         sbi  = ll_s2sbi(sb);
 297         env = cl_env_get(&refcheck);
 298         if (!IS_ERR(env)) {
 299                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 300                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 301                 if (!IS_ERR(cl)) {
 302                         cl2vvp_dev(cl)->vdv_sb = sb;
 303                         sbi->ll_cl = cl;
 304                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 305                 }
 306                 cl_env_put(env, &refcheck);
 307         } else
 308                 rc = PTR_ERR(env);
 309         RETURN(rc);
 310 }
 311
 312 int cl_sb_fini(struct super_block *sb)
 313 {
 314         struct ll_sb_info *sbi;
 315         struct lu_env     *env;
 316         struct cl_device  *cld;
 317         int                refcheck;
 318         int                result;
 319
 320         ENTRY;
 321         sbi = ll_s2sbi(sb);
 322         env = cl_env_get(&refcheck);
 323         if (!IS_ERR(env)) {
 324                 cld = sbi->ll_cl;
 325
 326                 if (cld != NULL) {
 327                         cl_stack_fini(env, cld);
 328                         sbi->ll_cl = NULL;
 329                         sbi->ll_site = NULL;
 330                 }
 331                 cl_env_put(env, &refcheck);
 332                 result = 0;
 333         } else {
 334                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 335                 result = PTR_ERR(env);
 336         }
 337
 338         RETURN(result);
 339 }
 340
 341 /****************************************************************************
 342  *
 343  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 344  *
 345  ****************************************************************************/
 346
 347 /*
 348  * To represent contents of a page cache as a byte stream, following
 349  * information if encoded in 64bit offset:
 350  *
 351  *       - file hash bucket in lu_site::ls_hash[]       28bits
 352  *
 353  *       - how far file is from bucket head              4bits
 354  *
 355  *       - page index                                   32bits
 356  *
 357  * First two data identify a file in the cache uniquely.
 358  */
 359
 360 #define PGC_OBJ_SHIFT (32 + 4)
 361 #define PGC_DEPTH_SHIFT (32)
 362
 363 struct vvp_pgcache_id {
 364         unsigned                 vpi_bucket;
 365         unsigned                 vpi_depth;
 366         uint32_t                 vpi_index;
 367
 368         unsigned                 vpi_curdep;
 369         struct lu_object_header *vpi_obj;
 370 };
 371
 372 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 373 {
 374         CLASSERT(sizeof(pos) == sizeof(__u64));
 375
 376         id->vpi_index  = pos & 0xffffffff;
 377         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 378         id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
 379 }
 380
 381 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 382 {
 383         return
 384                 ((__u64)id->vpi_index) |
 385                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 386                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 387 }
 388
 389 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 390                                struct hlist_node *hnode, void *data)
 391 {
 392         struct vvp_pgcache_id   *id  = data;
 393         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 394
 395         if (id->vpi_curdep-- > 0)
 396                 return 0; /* continue */
 397
 398         if (lu_object_is_dying(hdr))
 399                 return 1;
 400
 401         cfs_hash_get(hs, hnode);
 402         id->vpi_obj = hdr;
 403         return 1;
 404 }
 405
 406 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 407                                          struct lu_device *dev,
 408                                          struct vvp_pgcache_id *id)
 409 {
 410         LASSERT(lu_device_is_cl(dev));
 411
 412         id->vpi_depth &= 0xf;
 413         id->vpi_obj    = NULL;
 414         id->vpi_curdep = id->vpi_depth;
 415
 416         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 417                                 vvp_pgcache_obj_get, id);
 418         if (id->vpi_obj != NULL) {
 419                 struct lu_object *lu_obj;
 420
 421                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 422                 if (lu_obj != NULL) {
 423                         lu_object_ref_add(lu_obj, "dump", current);
 424                         return lu2cl(lu_obj);
 425                 }
 426                 lu_object_put(env, lu_object_top(id->vpi_obj));
 427
 428         } else if (id->vpi_curdep > 0) {
 429                 id->vpi_depth = 0xf;
 430         }
 431         return NULL;
 432 }
 433
 434 static loff_t vvp_pgcache_find(const struct lu_env *env,
 435                                struct lu_device *dev, loff_t pos)
 436 {
 437         struct cl_object     *clob;
 438         struct lu_site       *site;
 439         struct vvp_pgcache_id id;
 440
 441         site = dev->ld_site;
 442         vvp_pgcache_id_unpack(pos, &id);
 443
 444         while (1) {
 445                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 446                         return ~0ULL;
 447                 clob = vvp_pgcache_obj(env, dev, &id);
 448                 if (clob != NULL) {
 449                         struct inode *inode = vvp_object_inode(clob);
 450                         struct page *vmpage;
 451                         int nr;
 452
 453                         nr = find_get_pages_contig(inode->i_mapping,
 454                                                    id.vpi_index, 1, &vmpage);
 455                         if (nr > 0) {
 456                                 id.vpi_index = vmpage->index;
 457                                 /* Cant support over 16T file */
 458                                 nr = !(vmpage->index > 0xffffffff);
 459                                 page_cache_release(vmpage);
 460                         }
 461
 462                         lu_object_ref_del(&clob->co_lu, "dump", current);
 463                         cl_object_put(env, clob);
 464                         if (nr > 0)
 465                                 return vvp_pgcache_id_pack(&id);
 466                 }
 467                 /* to the next object. */
 468                 ++id.vpi_depth;
 469                 id.vpi_depth &= 0xf;
 470                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 471                         return ~0ULL;
 472                 id.vpi_index = 0;
 473         }
 474 }
 475
 476 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 477         if (test_bit(PG_##flag, &(page)->flags)) {                  \
 478                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 479                 has_flags = 1;                                          \
 480         }                                                               \
 481 } while(0)
 482
 483 static void vvp_pgcache_page_show(const struct lu_env *env,
 484                                   struct seq_file *seq, struct cl_page *page)
 485 {
 486         struct vvp_page *vpg;
 487         struct page      *vmpage;
 488         int              has_flags;
 489
 490         vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
 491         vmpage = vpg->vpg_page;
 492         seq_printf(seq, " %5i | %p %p %s %s %s | %p "DFID"(%p) %lu %u [",
 493                    0 /* gen */,
 494                    vpg, page,
 495                    "none",
 496                    vpg->vpg_defer_uptodate ? "du" : "- ",
 497                    PageWriteback(vmpage) ? "wb" : "-",
 498                    vmpage,
 499                    PFID(ll_inode2fid(vmpage->mapping->host)),
 500                    vmpage->mapping->host, vmpage->index,
 501                    page_count(vmpage));
 502         has_flags = 0;
 503         seq_page_flag(seq, vmpage, locked, has_flags);
 504         seq_page_flag(seq, vmpage, error, has_flags);
 505         seq_page_flag(seq, vmpage, referenced, has_flags);
 506         seq_page_flag(seq, vmpage, uptodate, has_flags);
 507         seq_page_flag(seq, vmpage, dirty, has_flags);
 508         seq_page_flag(seq, vmpage, writeback, has_flags);
 509         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 510 }
 511
 512 static int vvp_pgcache_show(struct seq_file *f, void *v)
 513 {
 514         loff_t                   pos;
 515         struct ll_sb_info       *sbi;
 516         struct cl_object        *clob;
 517         struct lu_env           *env;
 518         struct vvp_pgcache_id    id;
 519         int                      refcheck;
 520         int                      result;
 521
 522         env = cl_env_get(&refcheck);
 523         if (!IS_ERR(env)) {
 524                 pos = *(loff_t *) v;
 525                 vvp_pgcache_id_unpack(pos, &id);
 526                 sbi = f->private;
 527                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 528                 if (clob != NULL) {
 529                         struct inode *inode = vvp_object_inode(clob);
 530                         struct cl_page *page = NULL;
 531                         struct page *vmpage;
 532
 533                         result = find_get_pages_contig(inode->i_mapping,
 534                                                       id.vpi_index, 1, &vmpage);
 535                         if (result > 0) {
 536                                 lock_page(vmpage);
 537                                 page = cl_vmpage_page(vmpage, clob);
 538                                 unlock_page(vmpage);
 539
 540                                 page_cache_release(vmpage);
 541                         }
 542
 543                         seq_printf(f, "%8x@"DFID": ", id.vpi_index,
 544                                    PFID(lu_object_fid(&clob->co_lu)));
 545                         if (page != NULL) {
 546                                 vvp_pgcache_page_show(env, f, page);
 547                                 cl_page_put(env, page);
 548                         } else
 549                                 seq_puts(f, "missing\n");
 550                         lu_object_ref_del(&clob->co_lu, "dump", current);
 551                         cl_object_put(env, clob);
 552                 } else
 553                         seq_printf(f, "%llx missing\n", pos);
 554                 cl_env_put(env, &refcheck);
 555                 result = 0;
 556         } else
 557                 result = PTR_ERR(env);
 558         return result;
 559 }
 560
 561 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 562 {
 563         struct ll_sb_info *sbi;
 564         struct lu_env     *env;
 565         int                refcheck;
 566
 567         sbi = f->private;
 568
 569         env = cl_env_get(&refcheck);
 570         if (!IS_ERR(env)) {
 571                 sbi = f->private;
 572                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 573                         pos = ERR_PTR(-EFBIG);
 574                 else {
 575                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 576                                                 *pos);
 577                         if (*pos == ~0ULL)
 578                                 pos = NULL;
 579                 }
 580                 cl_env_put(env, &refcheck);
 581         }
 582         return pos;
 583 }
 584
 585 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 586 {
 587         struct ll_sb_info *sbi;
 588         struct lu_env     *env;
 589         int                refcheck;
 590
 591         env = cl_env_get(&refcheck);
 592         if (!IS_ERR(env)) {
 593                 sbi = f->private;
 594                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 595                 if (*pos == ~0ULL)
 596                         pos = NULL;
 597                 cl_env_put(env, &refcheck);
 598         }
 599         return pos;
 600 }
 601
 602 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 603 {
 604         /* Nothing to do */
 605 }
 606
 607 static struct seq_operations vvp_pgcache_ops = {
 608         .start = vvp_pgcache_start,
 609         .next  = vvp_pgcache_next,
 610         .stop  = vvp_pgcache_stop,
 611         .show  = vvp_pgcache_show
 612 };
 613
 614 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 615 {
 616         struct ll_sb_info       *sbi = PDE_DATA(inode);
 617         struct seq_file         *seq;
 618         int                     result;
 619
 620         result = seq_open(filp, &vvp_pgcache_ops);
 621         if (result == 0) {
 622                 seq = filp->private_data;
 623                 seq->private = sbi;
 624         }
 625         return result;
 626 }
 627
 628 const struct file_operations vvp_dump_pgcache_file_ops = {
 629         .owner   = THIS_MODULE,
 630         .open    = vvp_dump_pgcache_seq_open,
 631         .read    = seq_read,
 632         .llseek  = seq_lseek,
 633         .release = seq_release,
 634 };