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