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 static struct kmem_cache *vvp_thread_kmem;
  61 static struct kmem_cache *vvp_session_kmem;
  62 static struct lu_kmem_descr vvp_caches[] = {
  63         {
  64                 .ckd_cache = &vvp_thread_kmem,
  65                 .ckd_name  = "vvp_thread_kmem",
  66                 .ckd_size  = sizeof (struct vvp_thread_info),
  67         },
  68         {
  69                 .ckd_cache = &vvp_session_kmem,
  70                 .ckd_name  = "vvp_session_kmem",
  71                 .ckd_size  = sizeof (struct vvp_session)
  72         },
  73         {
  74                 .ckd_cache = NULL
  75         }
  76 };
  77
  78 static void *vvp_key_init(const struct lu_context *ctx,
  79                           struct lu_context_key *key)
  80 {
  81         struct vvp_thread_info *info;
  82
  83         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, GFP_NOFS);
  84         if (info == NULL)
  85                 info = ERR_PTR(-ENOMEM);
  86         return info;
  87 }
  88
  89 static void vvp_key_fini(const struct lu_context *ctx,
  90                          struct lu_context_key *key, void *data)
  91 {
  92         struct vvp_thread_info *info = data;
  93         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
  94 }
  95
  96 static void *vvp_session_key_init(const struct lu_context *ctx,
  97                                   struct lu_context_key *key)
  98 {
  99         struct vvp_session *session;
 100
 101         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
 102         if (session == NULL)
 103                 session = ERR_PTR(-ENOMEM);
 104         return session;
 105 }
 106
 107 static void vvp_session_key_fini(const struct lu_context *ctx,
 108                                  struct lu_context_key *key, void *data)
 109 {
 110         struct vvp_session *session = data;
 111         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 112 }
 113
 114
 115 struct lu_context_key vvp_key = {
 116         .lct_tags = LCT_CL_THREAD,
 117         .lct_init = vvp_key_init,
 118         .lct_fini = vvp_key_fini
 119 };
 120
 121 struct lu_context_key vvp_session_key = {
 122         .lct_tags = LCT_SESSION,
 123         .lct_init = vvp_session_key_init,
 124         .lct_fini = vvp_session_key_fini
 125 };
 126
 127 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 128 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
 129
 130 static const struct lu_device_operations vvp_lu_ops = {
 131         .ldo_object_alloc      = vvp_object_alloc
 132 };
 133
 134 static const struct cl_device_operations vvp_cl_ops = {
 135         .cdo_req_init = ccc_req_init
 136 };
 137
 138 static struct lu_device *vvp_device_free(const struct lu_env *env,
 139                                          struct lu_device *d)
 140 {
 141         struct vvp_device *vdv  = lu2vvp_dev(d);
 142         struct cl_site    *site = lu2cl_site(d->ld_site);
 143         struct lu_device  *next = cl2lu_dev(vdv->vdv_next);
 144
 145         if (d->ld_site != NULL) {
 146                 cl_site_fini(site);
 147                 OBD_FREE_PTR(site);
 148         }
 149
 150         cl_device_fini(lu2cl_dev(d));
 151         OBD_FREE_PTR(vdv);
 152         return next;
 153 }
 154
 155 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 156                                           struct lu_device_type *t,
 157                                           struct lustre_cfg *cfg)
 158 {
 159         struct vvp_device *vdv;
 160         struct lu_device *lud;
 161         struct cl_site *site;
 162         int rc;
 163         ENTRY;
 164
 165         OBD_ALLOC_PTR(vdv);
 166         if (vdv == NULL)
 167                 RETURN(ERR_PTR(-ENOMEM));
 168
 169         lud = &vdv->vdv_cl.cd_lu_dev;
 170         cl_device_init(&vdv->vdv_cl, t);
 171         vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
 172         vdv->vdv_cl.cd_ops = &vvp_cl_ops;
 173
 174         OBD_ALLOC_PTR(site);
 175         if (site != NULL) {
 176                 rc = cl_site_init(site, &vdv->vdv_cl);
 177                 if (rc == 0)
 178                         rc = lu_site_init_finish(&site->cs_lu);
 179                 else {
 180                         LASSERT(lud->ld_site == NULL);
 181                         CERROR("Cannot init lu_site, rc %d.\n", rc);
 182                         OBD_FREE_PTR(site);
 183                 }
 184         } else
 185                 rc = -ENOMEM;
 186         if (rc != 0) {
 187                 vvp_device_free(env, lud);
 188                 lud = ERR_PTR(rc);
 189         }
 190         RETURN(lud);
 191 }
 192
 193 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
 194                            const char *name, struct lu_device *next)
 195 {
 196         struct vvp_device  *vdv;
 197         int rc;
 198         ENTRY;
 199
 200         vdv = lu2vvp_dev(d);
 201         vdv->vdv_next = lu2cl_dev(next);
 202
 203         LASSERT(d->ld_site != NULL && next->ld_type != NULL);
 204         next->ld_site = d->ld_site;
 205         rc = next->ld_type->ldt_ops->ldto_device_init(
 206                 env, next, next->ld_type->ldt_name, NULL);
 207         if (rc == 0) {
 208                 lu_device_get(next);
 209                 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
 210         }
 211         RETURN(rc);
 212 }
 213
 214 static struct lu_device *vvp_device_fini(const struct lu_env *env,
 215                                          struct lu_device *d)
 216 {
 217         return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
 218 }
 219
 220 static const struct lu_device_type_operations vvp_device_type_ops = {
 221         .ldto_init = vvp_type_init,
 222         .ldto_fini = vvp_type_fini,
 223
 224         .ldto_start = vvp_type_start,
 225         .ldto_stop  = vvp_type_stop,
 226
 227         .ldto_device_alloc      = vvp_device_alloc,
 228         .ldto_device_free       = vvp_device_free,
 229         .ldto_device_init       = vvp_device_init,
 230         .ldto_device_fini       = vvp_device_fini,
 231 };
 232
 233 struct lu_device_type vvp_device_type = {
 234         .ldt_tags     = LU_DEVICE_CL,
 235         .ldt_name     = LUSTRE_VVP_NAME,
 236         .ldt_ops      = &vvp_device_type_ops,
 237         .ldt_ctx_tags = LCT_CL_THREAD
 238 };
 239
 240 /**
 241  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 242  * pressure, when environments cannot be allocated.
 243  */
 244 int vvp_global_init(void)
 245 {
 246         int result;
 247
 248         result = lu_kmem_init(vvp_caches);
 249         if (result == 0) {
 250                 result = ccc_global_init(&vvp_device_type);
 251                 if (result != 0)
 252                         lu_kmem_fini(vvp_caches);
 253         }
 254         return result;
 255 }
 256
 257 void vvp_global_fini(void)
 258 {
 259         ccc_global_fini(&vvp_device_type);
 260         lu_kmem_fini(vvp_caches);
 261 }
 262
 263
 264 /*****************************************************************************
 265  *
 266  * mirror obd-devices into cl devices.
 267  *
 268  */
 269
 270 int cl_sb_init(struct super_block *sb)
 271 {
 272         struct ll_sb_info *sbi;
 273         struct cl_device  *cl;
 274         struct lu_env     *env;
 275         int rc = 0;
 276         int refcheck;
 277
 278         sbi  = ll_s2sbi(sb);
 279         env = cl_env_get(&refcheck);
 280         if (!IS_ERR(env)) {
 281                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 282                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 283                 if (!IS_ERR(cl)) {
 284                         cl2vvp_dev(cl)->vdv_sb = sb;
 285                         sbi->ll_cl = cl;
 286                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 287                 }
 288                 cl_env_put(env, &refcheck);
 289         } else
 290                 rc = PTR_ERR(env);
 291         RETURN(rc);
 292 }
 293
 294 int cl_sb_fini(struct super_block *sb)
 295 {
 296         struct ll_sb_info *sbi;
 297         struct lu_env     *env;
 298         struct cl_device  *cld;
 299         int                refcheck;
 300         int                result;
 301
 302         ENTRY;
 303         sbi = ll_s2sbi(sb);
 304         env = cl_env_get(&refcheck);
 305         if (!IS_ERR(env)) {
 306                 cld = sbi->ll_cl;
 307
 308                 if (cld != NULL) {
 309                         cl_stack_fini(env, cld);
 310                         sbi->ll_cl = NULL;
 311                         sbi->ll_site = NULL;
 312                 }
 313                 cl_env_put(env, &refcheck);
 314                 result = 0;
 315         } else {
 316                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 317                 result = PTR_ERR(env);
 318         }
 319
 320         RETURN(result);
 321 }
 322
 323 /****************************************************************************
 324  *
 325  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 326  *
 327  ****************************************************************************/
 328
 329 /*
 330  * To represent contents of a page cache as a byte stream, following
 331  * information if encoded in 64bit offset:
 332  *
 333  *       - file hash bucket in lu_site::ls_hash[]       28bits
 334  *
 335  *       - how far file is from bucket head              4bits
 336  *
 337  *       - page index                                   32bits
 338  *
 339  * First two data identify a file in the cache uniquely.
 340  */
 341
 342 #define PGC_OBJ_SHIFT (32 + 4)
 343 #define PGC_DEPTH_SHIFT (32)
 344
 345 struct vvp_pgcache_id {
 346         unsigned                 vpi_bucket;
 347         unsigned                 vpi_depth;
 348         uint32_t                 vpi_index;
 349
 350         unsigned                 vpi_curdep;
 351         struct lu_object_header *vpi_obj;
 352 };
 353
 354 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 355 {
 356         CLASSERT(sizeof(pos) == sizeof(__u64));
 357
 358         id->vpi_index  = pos & 0xffffffff;
 359         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 360         id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
 361 }
 362
 363 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 364 {
 365         return
 366                 ((__u64)id->vpi_index) |
 367                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 368                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 369 }
 370
 371 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 372                                struct hlist_node *hnode, void *data)
 373 {
 374         struct vvp_pgcache_id   *id  = data;
 375         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 376
 377         if (id->vpi_curdep-- > 0)
 378                 return 0; /* continue */
 379
 380         if (lu_object_is_dying(hdr))
 381                 return 1;
 382
 383         cfs_hash_get(hs, hnode);
 384         id->vpi_obj = hdr;
 385         return 1;
 386 }
 387
 388 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 389                                          struct lu_device *dev,
 390                                          struct vvp_pgcache_id *id)
 391 {
 392         LASSERT(lu_device_is_cl(dev));
 393
 394         id->vpi_depth &= 0xf;
 395         id->vpi_obj    = NULL;
 396         id->vpi_curdep = id->vpi_depth;
 397
 398         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 399                                 vvp_pgcache_obj_get, id);
 400         if (id->vpi_obj != NULL) {
 401                 struct lu_object *lu_obj;
 402
 403                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 404                 if (lu_obj != NULL) {
 405                         lu_object_ref_add(lu_obj, "dump", current);
 406                         return lu2cl(lu_obj);
 407                 }
 408                 lu_object_put(env, lu_object_top(id->vpi_obj));
 409
 410         } else if (id->vpi_curdep > 0) {
 411                 id->vpi_depth = 0xf;
 412         }
 413         return NULL;
 414 }
 415
 416 static loff_t vvp_pgcache_find(const struct lu_env *env,
 417                                struct lu_device *dev, loff_t pos)
 418 {
 419         struct cl_object     *clob;
 420         struct lu_site       *site;
 421         struct vvp_pgcache_id id;
 422
 423         site = dev->ld_site;
 424         vvp_pgcache_id_unpack(pos, &id);
 425
 426         while (1) {
 427                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 428                         return ~0ULL;
 429                 clob = vvp_pgcache_obj(env, dev, &id);
 430                 if (clob != NULL) {
 431                         struct inode *inode = ccc_object_inode(clob);
 432                         struct page *vmpage;
 433                         int nr;
 434
 435                         nr = find_get_pages_contig(inode->i_mapping,
 436                                                    id.vpi_index, 1, &vmpage);
 437                         if (nr > 0) {
 438                                 id.vpi_index = vmpage->index;
 439                                 /* Cant support over 16T file */
 440                                 nr = !(vmpage->index > 0xffffffff);
 441                                 page_cache_release(vmpage);
 442                         }
 443
 444                         lu_object_ref_del(&clob->co_lu, "dump", current);
 445                         cl_object_put(env, clob);
 446                         if (nr > 0)
 447                                 return vvp_pgcache_id_pack(&id);
 448                 }
 449                 /* to the next object. */
 450                 ++id.vpi_depth;
 451                 id.vpi_depth &= 0xf;
 452                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 453                         return ~0ULL;
 454                 id.vpi_index = 0;
 455         }
 456 }
 457
 458 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 459         if (test_bit(PG_##flag, &(page)->flags)) {                  \
 460                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 461                 has_flags = 1;                                          \
 462         }                                                               \
 463 } while(0)
 464
 465 static void vvp_pgcache_page_show(const struct lu_env *env,
 466                                   struct seq_file *seq, struct cl_page *page)
 467 {
 468         struct ccc_page *cpg;
 469         struct page      *vmpage;
 470         int              has_flags;
 471
 472         cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
 473         vmpage = cpg->cpg_page;
 474         seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
 475                    0 /* gen */,
 476                    cpg, page,
 477                    "none",
 478                    cpg->cpg_write_queued ? "wq" : "- ",
 479                    cpg->cpg_defer_uptodate ? "du" : "- ",
 480                    PageWriteback(vmpage) ? "wb" : "-",
 481                    vmpage,
 482                    PFID(ll_inode2fid(vmpage->mapping->host)),
 483                    vmpage->mapping->host, vmpage->index,
 484                    page_count(vmpage));
 485         has_flags = 0;
 486         seq_page_flag(seq, vmpage, locked, has_flags);
 487         seq_page_flag(seq, vmpage, error, has_flags);
 488         seq_page_flag(seq, vmpage, referenced, has_flags);
 489         seq_page_flag(seq, vmpage, uptodate, has_flags);
 490         seq_page_flag(seq, vmpage, dirty, has_flags);
 491         seq_page_flag(seq, vmpage, writeback, has_flags);
 492         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 493 }
 494
 495 static int vvp_pgcache_show(struct seq_file *f, void *v)
 496 {
 497         loff_t                   pos;
 498         struct ll_sb_info       *sbi;
 499         struct cl_object        *clob;
 500         struct lu_env           *env;
 501         struct vvp_pgcache_id    id;
 502         int                      refcheck;
 503         int                      result;
 504
 505         env = cl_env_get(&refcheck);
 506         if (!IS_ERR(env)) {
 507                 pos = *(loff_t *) v;
 508                 vvp_pgcache_id_unpack(pos, &id);
 509                 sbi = f->private;
 510                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 511                 if (clob != NULL) {
 512                         struct inode *inode = ccc_object_inode(clob);
 513                         struct cl_page *page = NULL;
 514                         struct page *vmpage;
 515
 516                         result = find_get_pages_contig(inode->i_mapping,
 517                                                       id.vpi_index, 1, &vmpage);
 518                         if (result > 0) {
 519                                 lock_page(vmpage);
 520                                 page = cl_vmpage_page(vmpage, clob);
 521                                 unlock_page(vmpage);
 522
 523                                 page_cache_release(vmpage);
 524                         }
 525
 526                         seq_printf(f, "%8x@"DFID": ", id.vpi_index,
 527                                    PFID(lu_object_fid(&clob->co_lu)));
 528                         if (page != NULL) {
 529                                 vvp_pgcache_page_show(env, f, page);
 530                                 cl_page_put(env, page);
 531                         } else
 532                                 seq_puts(f, "missing\n");
 533                         lu_object_ref_del(&clob->co_lu, "dump", current);
 534                         cl_object_put(env, clob);
 535                 } else
 536                         seq_printf(f, "%llx missing\n", pos);
 537                 cl_env_put(env, &refcheck);
 538                 result = 0;
 539         } else
 540                 result = PTR_ERR(env);
 541         return result;
 542 }
 543
 544 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 545 {
 546         struct ll_sb_info *sbi;
 547         struct lu_env     *env;
 548         int                refcheck;
 549
 550         sbi = f->private;
 551
 552         env = cl_env_get(&refcheck);
 553         if (!IS_ERR(env)) {
 554                 sbi = f->private;
 555                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 556                         pos = ERR_PTR(-EFBIG);
 557                 else {
 558                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 559                                                 *pos);
 560                         if (*pos == ~0ULL)
 561                                 pos = NULL;
 562                 }
 563                 cl_env_put(env, &refcheck);
 564         }
 565         return pos;
 566 }
 567
 568 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 569 {
 570         struct ll_sb_info *sbi;
 571         struct lu_env     *env;
 572         int                refcheck;
 573
 574         env = cl_env_get(&refcheck);
 575         if (!IS_ERR(env)) {
 576                 sbi = f->private;
 577                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 578                 if (*pos == ~0ULL)
 579                         pos = NULL;
 580                 cl_env_put(env, &refcheck);
 581         }
 582         return pos;
 583 }
 584
 585 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 586 {
 587         /* Nothing to do */
 588 }
 589
 590 static struct seq_operations vvp_pgcache_ops = {
 591         .start = vvp_pgcache_start,
 592         .next  = vvp_pgcache_next,
 593         .stop  = vvp_pgcache_stop,
 594         .show  = vvp_pgcache_show
 595 };
 596
 597 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 598 {
 599         struct ll_sb_info       *sbi = PDE_DATA(inode);
 600         struct seq_file         *seq;
 601         int                     result;
 602
 603         result = seq_open(filp, &vvp_pgcache_ops);
 604         if (result == 0) {
 605                 seq = filp->private_data;
 606                 seq->private = sbi;
 607         }
 608         return result;
 609 }
 610
 611 const struct file_operations vvp_dump_pgcache_file_ops = {
 612         .owner   = THIS_MODULE,
 613         .open    = vvp_dump_pgcache_seq_open,
 614         .read    = seq_read,
 615         .llseek  = seq_lseek,
 616         .release = seq_release,
 617 };