drivers/staging/lustre/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, 2015, 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 #include "../include/obd.h"
  45 #include "../include/lustre_lite.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         info = kmem_cache_zalloc(vvp_thread_kmem, GFP_NOFS);
  96         if (!info)
  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
 106         kmem_cache_free(vvp_thread_kmem, info);
 107 }
 108
 109 static void *vvp_session_key_init(const struct lu_context *ctx,
 110                                   struct lu_context_key *key)
 111 {
 112         struct vvp_session *session;
 113
 114         session = kmem_cache_zalloc(vvp_session_kmem, GFP_NOFS);
 115         if (!session)
 116                 session = ERR_PTR(-ENOMEM);
 117         return session;
 118 }
 119
 120 static void vvp_session_key_fini(const struct lu_context *ctx,
 121                                  struct lu_context_key *key, void *data)
 122 {
 123         struct vvp_session *session = data;
 124
 125         kmem_cache_free(vvp_session_kmem, session);
 126 }
 127
 128 struct lu_context_key vvp_key = {
 129         .lct_tags = LCT_CL_THREAD,
 130         .lct_init = vvp_key_init,
 131         .lct_fini = vvp_key_fini
 132 };
 133
 134 struct lu_context_key vvp_session_key = {
 135         .lct_tags = LCT_SESSION,
 136         .lct_init = vvp_session_key_init,
 137         .lct_fini = vvp_session_key_fini
 138 };
 139
 140 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 141 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
 142
 143 static const struct lu_device_operations vvp_lu_ops = {
 144         .ldo_object_alloc      = vvp_object_alloc
 145 };
 146
 147 static const struct cl_device_operations vvp_cl_ops = {
 148         .cdo_req_init = ccc_req_init
 149 };
 150
 151 static struct lu_device *vvp_device_free(const struct lu_env *env,
 152                                          struct lu_device *d)
 153 {
 154         struct vvp_device *vdv  = lu2vvp_dev(d);
 155         struct cl_site    *site = lu2cl_site(d->ld_site);
 156         struct lu_device  *next = cl2lu_dev(vdv->vdv_next);
 157
 158         if (d->ld_site) {
 159                 cl_site_fini(site);
 160                 kfree(site);
 161         }
 162         cl_device_fini(lu2cl_dev(d));
 163         kfree(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
 176         vdv = kzalloc(sizeof(*vdv), GFP_NOFS);
 177         if (!vdv)
 178                 return ERR_PTR(-ENOMEM);
 179
 180         lud = &vdv->vdv_cl.cd_lu_dev;
 181         cl_device_init(&vdv->vdv_cl, t);
 182         vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
 183         vdv->vdv_cl.cd_ops = &vvp_cl_ops;
 184
 185         site = kzalloc(sizeof(*site), GFP_NOFS);
 186         if (site) {
 187                 rc = cl_site_init(site, &vdv->vdv_cl);
 188                 if (rc == 0) {
 189                         rc = lu_site_init_finish(&site->cs_lu);
 190                 } else {
 191                         LASSERT(!lud->ld_site);
 192                         CERROR("Cannot init lu_site, rc %d.\n", rc);
 193                         kfree(site);
 194                 }
 195         } else {
 196                 rc = -ENOMEM;
 197         }
 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
 211         vdv = lu2vvp_dev(d);
 212         vdv->vdv_next = lu2cl_dev(next);
 213
 214         LASSERT(d->ld_site && next->ld_type);
 215         next->ld_site = d->ld_site;
 216         rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
 217                                                       next->ld_type->ldt_name,
 218                                                       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  * mirror obd-devices into cl devices.
 278  *
 279  */
 280
 281 int cl_sb_init(struct super_block *sb)
 282 {
 283         struct ll_sb_info *sbi;
 284         struct cl_device  *cl;
 285         struct lu_env     *env;
 286         int rc = 0;
 287         int refcheck;
 288
 289         sbi  = ll_s2sbi(sb);
 290         env = cl_env_get(&refcheck);
 291         if (!IS_ERR(env)) {
 292                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 293                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 294                 if (!IS_ERR(cl)) {
 295                         cl2vvp_dev(cl)->vdv_sb = sb;
 296                         sbi->ll_cl = cl;
 297                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 298                 }
 299                 cl_env_put(env, &refcheck);
 300         } else
 301                 rc = PTR_ERR(env);
 302         return rc;
 303 }
 304
 305 int cl_sb_fini(struct super_block *sb)
 306 {
 307         struct ll_sb_info *sbi;
 308         struct lu_env     *env;
 309         struct cl_device  *cld;
 310         int             refcheck;
 311         int             result;
 312
 313         sbi = ll_s2sbi(sb);
 314         env = cl_env_get(&refcheck);
 315         if (!IS_ERR(env)) {
 316                 cld = sbi->ll_cl;
 317
 318                 if (cld) {
 319                         cl_stack_fini(env, cld);
 320                         sbi->ll_cl = NULL;
 321                         sbi->ll_site = NULL;
 322                 }
 323                 cl_env_put(env, &refcheck);
 324                 result = 0;
 325         } else {
 326                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 327                 result = PTR_ERR(env);
 328         }
 329         /*
 330          * If mount failed (sbi->ll_cl == NULL), and this there are no other
 331          * mounts, stop device types manually (this usually happens
 332          * automatically when last device is destroyed).
 333          */
 334         lu_types_stop();
 335         return result;
 336 }
 337
 338 /****************************************************************************
 339  *
 340  * debugfs/lustre/llite/$MNT/dump_page_cache
 341  *
 342  ****************************************************************************/
 343
 344 /*
 345  * To represent contents of a page cache as a byte stream, following
 346  * information if encoded in 64bit offset:
 347  *
 348  *       - file hash bucket in lu_site::ls_hash[]       28bits
 349  *
 350  *       - how far file is from bucket head           4bits
 351  *
 352  *       - page index                              32bits
 353  *
 354  * First two data identify a file in the cache uniquely.
 355  */
 356
 357 #define PGC_OBJ_SHIFT (32 + 4)
 358 #define PGC_DEPTH_SHIFT (32)
 359
 360 struct vvp_pgcache_id {
 361         unsigned                 vpi_bucket;
 362         unsigned                 vpi_depth;
 363         uint32_t                 vpi_index;
 364
 365         unsigned                 vpi_curdep;
 366         struct lu_object_header *vpi_obj;
 367 };
 368
 369 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 370 {
 371         CLASSERT(sizeof(pos) == sizeof(__u64));
 372
 373         id->vpi_index  = pos & 0xffffffff;
 374         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 375         id->vpi_bucket = (unsigned long long)pos >> PGC_OBJ_SHIFT;
 376 }
 377
 378 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 379 {
 380         return
 381                 ((__u64)id->vpi_index) |
 382                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 383                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 384 }
 385
 386 static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 387                                struct hlist_node *hnode, void *data)
 388 {
 389         struct vvp_pgcache_id   *id  = data;
 390         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 391
 392         if (id->vpi_curdep-- > 0)
 393                 return 0; /* continue */
 394
 395         if (lu_object_is_dying(hdr))
 396                 return 1;
 397
 398         cfs_hash_get(hs, hnode);
 399         id->vpi_obj = hdr;
 400         return 1;
 401 }
 402
 403 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 404                                          struct lu_device *dev,
 405                                          struct vvp_pgcache_id *id)
 406 {
 407         LASSERT(lu_device_is_cl(dev));
 408
 409         id->vpi_depth &= 0xf;
 410         id->vpi_obj    = NULL;
 411         id->vpi_curdep = id->vpi_depth;
 412
 413         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 414                                 vvp_pgcache_obj_get, id);
 415         if (id->vpi_obj) {
 416                 struct lu_object *lu_obj;
 417
 418                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 419                 if (lu_obj) {
 420                         lu_object_ref_add(lu_obj, "dump", current);
 421                         return lu2cl(lu_obj);
 422                 }
 423                 lu_object_put(env, lu_object_top(id->vpi_obj));
 424
 425         } else if (id->vpi_curdep > 0) {
 426                 id->vpi_depth = 0xf;
 427         }
 428         return NULL;
 429 }
 430
 431 static loff_t vvp_pgcache_find(const struct lu_env *env,
 432                                struct lu_device *dev, loff_t pos)
 433 {
 434         struct cl_object     *clob;
 435         struct lu_site       *site;
 436         struct vvp_pgcache_id id;
 437
 438         site = dev->ld_site;
 439         vvp_pgcache_id_unpack(pos, &id);
 440
 441         while (1) {
 442                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 443                         return ~0ULL;
 444                 clob = vvp_pgcache_obj(env, dev, &id);
 445                 if (clob) {
 446                         struct inode *inode = vvp_object_inode(clob);
 447                         struct page *vmpage;
 448                         int nr;
 449
 450                         nr = find_get_pages_contig(inode->i_mapping,
 451                                                    id.vpi_index, 1, &vmpage);
 452                         if (nr > 0) {
 453                                 id.vpi_index = vmpage->index;
 454                                 /* Cant support over 16T file */
 455                                 nr = !(vmpage->index > 0xffffffff);
 456                                 page_cache_release(vmpage);
 457                         }
 458
 459                         lu_object_ref_del(&clob->co_lu, "dump", current);
 460                         cl_object_put(env, clob);
 461                         if (nr > 0)
 462                                 return vvp_pgcache_id_pack(&id);
 463                 }
 464                 /* to the next object. */
 465                 ++id.vpi_depth;
 466                 id.vpi_depth &= 0xf;
 467                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 468                         return ~0ULL;
 469                 id.vpi_index = 0;
 470         }
 471 }
 472
 473 #define seq_page_flag(seq, page, flag, has_flags) do {            \
 474         if (test_bit(PG_##flag, &(page)->flags)) {                \
 475                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 476                 has_flags = 1;                                    \
 477         }                                                              \
 478 } while (0)
 479
 480 static void vvp_pgcache_page_show(const struct lu_env *env,
 481                                   struct seq_file *seq, struct cl_page *page)
 482 {
 483         struct vvp_page *vpg;
 484         struct page      *vmpage;
 485         int           has_flags;
 486
 487         vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
 488         vmpage = vpg->vpg_page;
 489         seq_printf(seq, " %5i | %p %p %s %s %s %s | %p %lu/%u(%p) %lu %u [",
 490                    0 /* gen */,
 491                    vpg, page,
 492                    "none",
 493                    vpg->vpg_write_queued ? "wq" : "- ",
 494                    vpg->vpg_defer_uptodate ? "du" : "- ",
 495                    PageWriteback(vmpage) ? "wb" : "-",
 496                    vmpage, vmpage->mapping->host->i_ino,
 497                    vmpage->mapping->host->i_generation,
 498                    vmpage->mapping->host, vmpage->index,
 499                    page_count(vmpage));
 500         has_flags = 0;
 501         seq_page_flag(seq, vmpage, locked, has_flags);
 502         seq_page_flag(seq, vmpage, error, has_flags);
 503         seq_page_flag(seq, vmpage, referenced, has_flags);
 504         seq_page_flag(seq, vmpage, uptodate, has_flags);
 505         seq_page_flag(seq, vmpage, dirty, has_flags);
 506         seq_page_flag(seq, vmpage, writeback, has_flags);
 507         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 508 }
 509
 510 static int vvp_pgcache_show(struct seq_file *f, void *v)
 511 {
 512         loff_t             pos;
 513         struct ll_sb_info       *sbi;
 514         struct cl_object        *clob;
 515         struct lu_env      *env;
 516         struct vvp_pgcache_id    id;
 517         int                   refcheck;
 518         int                   result;
 519
 520         env = cl_env_get(&refcheck);
 521         if (!IS_ERR(env)) {
 522                 pos = *(loff_t *) v;
 523                 vvp_pgcache_id_unpack(pos, &id);
 524                 sbi = f->private;
 525                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 526                 if (clob) {
 527                         struct inode *inode = vvp_object_inode(clob);
 528                         struct cl_page *page = NULL;
 529                         struct page *vmpage;
 530
 531                         result = find_get_pages_contig(inode->i_mapping,
 532                                                        id.vpi_index, 1,
 533                                                        &vmpage);
 534                         if (result > 0) {
 535                                 lock_page(vmpage);
 536                                 page = cl_vmpage_page(vmpage, clob);
 537                                 unlock_page(vmpage);
 538
 539                                 page_cache_release(vmpage);
 540                         }
 541
 542                         seq_printf(f, "%8x@" DFID ": ", id.vpi_index,
 543                                    PFID(lu_object_fid(&clob->co_lu)));
 544                         if (page) {
 545                                 vvp_pgcache_page_show(env, f, page);
 546                                 cl_page_put(env, page);
 547                         } else
 548                                 seq_puts(f, "missing\n");
 549                         lu_object_ref_del(&clob->co_lu, "dump", current);
 550                         cl_object_put(env, clob);
 551                 } else
 552                         seq_printf(f, "%llx missing\n", pos);
 553                 cl_env_put(env, &refcheck);
 554                 result = 0;
 555         } else
 556                 result = PTR_ERR(env);
 557         return result;
 558 }
 559
 560 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 561 {
 562         struct ll_sb_info *sbi;
 563         struct lu_env     *env;
 564         int             refcheck;
 565
 566         sbi = f->private;
 567
 568         env = cl_env_get(&refcheck);
 569         if (!IS_ERR(env)) {
 570                 sbi = f->private;
 571                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 572                         pos = ERR_PTR(-EFBIG);
 573                 else {
 574                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 575                                                 *pos);
 576                         if (*pos == ~0ULL)
 577                                 pos = NULL;
 578                 }
 579                 cl_env_put(env, &refcheck);
 580         }
 581         return pos;
 582 }
 583
 584 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 585 {
 586         struct ll_sb_info *sbi;
 587         struct lu_env     *env;
 588         int             refcheck;
 589
 590         env = cl_env_get(&refcheck);
 591         if (!IS_ERR(env)) {
 592                 sbi = f->private;
 593                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 594                 if (*pos == ~0ULL)
 595                         pos = NULL;
 596                 cl_env_put(env, &refcheck);
 597         }
 598         return pos;
 599 }
 600
 601 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 602 {
 603         /* Nothing to do */
 604 }
 605
 606 static const struct seq_operations vvp_pgcache_ops = {
 607         .start = vvp_pgcache_start,
 608         .next  = vvp_pgcache_next,
 609         .stop  = vvp_pgcache_stop,
 610         .show  = vvp_pgcache_show
 611 };
 612
 613 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 614 {
 615         struct seq_file *seq;
 616         int rc;
 617
 618         rc = seq_open(filp, &vvp_pgcache_ops);
 619         if (rc)
 620                 return rc;
 621
 622         seq = filp->private_data;
 623         seq->private = inode->i_private;
 624
 625         return 0;
 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 };