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_object_kmem;
  61 static struct kmem_cache *vvp_thread_kmem;
  62 static struct kmem_cache *vvp_session_kmem;
  63 static struct lu_kmem_descr vvp_caches[] = {
  64         {
  65                 .ckd_cache = &vvp_object_kmem,
  66                 .ckd_name  = "vvp_object_kmem",
  67                 .ckd_size  = sizeof(struct vvp_object),
  68         },
  69         {
  70                 .ckd_cache = &vvp_thread_kmem,
  71                 .ckd_name  = "vvp_thread_kmem",
  72                 .ckd_size  = sizeof (struct vvp_thread_info),
  73         },
  74         {
  75                 .ckd_cache = &vvp_session_kmem,
  76                 .ckd_name  = "vvp_session_kmem",
  77                 .ckd_size  = sizeof (struct vvp_session)
  78         },
  79         {
  80                 .ckd_cache = NULL
  81         }
  82 };
  83
  84 static void *vvp_key_init(const struct lu_context *ctx,
  85                           struct lu_context_key *key)
  86 {
  87         struct vvp_thread_info *info;
  88
  89         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, GFP_NOFS);
  90         if (info == NULL)
  91                 info = ERR_PTR(-ENOMEM);
  92         return info;
  93 }
  94
  95 static void vvp_key_fini(const struct lu_context *ctx,
  96                          struct lu_context_key *key, void *data)
  97 {
  98         struct vvp_thread_info *info = data;
  99         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
 100 }
 101
 102 static void *vvp_session_key_init(const struct lu_context *ctx,
 103                                   struct lu_context_key *key)
 104 {
 105         struct vvp_session *session;
 106
 107         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
 108         if (session == NULL)
 109                 session = ERR_PTR(-ENOMEM);
 110         return session;
 111 }
 112
 113 static void vvp_session_key_fini(const struct lu_context *ctx,
 114                                  struct lu_context_key *key, void *data)
 115 {
 116         struct vvp_session *session = data;
 117         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 118 }
 119
 120
 121 struct lu_context_key vvp_key = {
 122         .lct_tags = LCT_CL_THREAD,
 123         .lct_init = vvp_key_init,
 124         .lct_fini = vvp_key_fini
 125 };
 126
 127 struct lu_context_key vvp_session_key = {
 128         .lct_tags = LCT_SESSION,
 129         .lct_init = vvp_session_key_init,
 130         .lct_fini = vvp_session_key_fini
 131 };
 132
 133 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 134 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
 135
 136 static const struct lu_device_operations vvp_lu_ops = {
 137         .ldo_object_alloc      = vvp_object_alloc
 138 };
 139
 140 static const struct cl_device_operations vvp_cl_ops = {
 141         .cdo_req_init = ccc_req_init
 142 };
 143
 144 static struct lu_device *vvp_device_free(const struct lu_env *env,
 145                                          struct lu_device *d)
 146 {
 147         struct vvp_device *vdv  = lu2vvp_dev(d);
 148         struct cl_site    *site = lu2cl_site(d->ld_site);
 149         struct lu_device  *next = cl2lu_dev(vdv->vdv_next);
 150
 151         if (d->ld_site != NULL) {
 152                 cl_site_fini(site);
 153                 OBD_FREE_PTR(site);
 154         }
 155
 156         cl_device_fini(lu2cl_dev(d));
 157         OBD_FREE_PTR(vdv);
 158         return next;
 159 }
 160
 161 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 162                                           struct lu_device_type *t,
 163                                           struct lustre_cfg *cfg)
 164 {
 165         struct vvp_device *vdv;
 166         struct lu_device *lud;
 167         struct cl_site *site;
 168         int rc;
 169         ENTRY;
 170
 171         OBD_ALLOC_PTR(vdv);
 172         if (vdv == NULL)
 173                 RETURN(ERR_PTR(-ENOMEM));
 174
 175         lud = &vdv->vdv_cl.cd_lu_dev;
 176         cl_device_init(&vdv->vdv_cl, t);
 177         vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
 178         vdv->vdv_cl.cd_ops = &vvp_cl_ops;
 179
 180         OBD_ALLOC_PTR(site);
 181         if (site != NULL) {
 182                 rc = cl_site_init(site, &vdv->vdv_cl);
 183                 if (rc == 0)
 184                         rc = lu_site_init_finish(&site->cs_lu);
 185                 else {
 186                         LASSERT(lud->ld_site == NULL);
 187                         CERROR("Cannot init lu_site, rc %d.\n", rc);
 188                         OBD_FREE_PTR(site);
 189                 }
 190         } else
 191                 rc = -ENOMEM;
 192         if (rc != 0) {
 193                 vvp_device_free(env, lud);
 194                 lud = ERR_PTR(rc);
 195         }
 196         RETURN(lud);
 197 }
 198
 199 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
 200                            const char *name, struct lu_device *next)
 201 {
 202         struct vvp_device  *vdv;
 203         int rc;
 204         ENTRY;
 205
 206         vdv = lu2vvp_dev(d);
 207         vdv->vdv_next = lu2cl_dev(next);
 208
 209         LASSERT(d->ld_site != NULL && next->ld_type != NULL);
 210         next->ld_site = d->ld_site;
 211         rc = next->ld_type->ldt_ops->ldto_device_init(
 212                 env, next, next->ld_type->ldt_name, NULL);
 213         if (rc == 0) {
 214                 lu_device_get(next);
 215                 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
 216         }
 217         RETURN(rc);
 218 }
 219
 220 static struct lu_device *vvp_device_fini(const struct lu_env *env,
 221                                          struct lu_device *d)
 222 {
 223         return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
 224 }
 225
 226 static const struct lu_device_type_operations vvp_device_type_ops = {
 227         .ldto_init = vvp_type_init,
 228         .ldto_fini = vvp_type_fini,
 229
 230         .ldto_start = vvp_type_start,
 231         .ldto_stop  = vvp_type_stop,
 232
 233         .ldto_device_alloc      = vvp_device_alloc,
 234         .ldto_device_free       = vvp_device_free,
 235         .ldto_device_init       = vvp_device_init,
 236         .ldto_device_fini       = vvp_device_fini,
 237 };
 238
 239 struct lu_device_type vvp_device_type = {
 240         .ldt_tags     = LU_DEVICE_CL,
 241         .ldt_name     = LUSTRE_VVP_NAME,
 242         .ldt_ops      = &vvp_device_type_ops,
 243         .ldt_ctx_tags = LCT_CL_THREAD
 244 };
 245
 246 /**
 247  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 248  * pressure, when environments cannot be allocated.
 249  */
 250 int vvp_global_init(void)
 251 {
 252         int result;
 253
 254         result = lu_kmem_init(vvp_caches);
 255         if (result == 0) {
 256                 result = ccc_global_init(&vvp_device_type);
 257                 if (result != 0)
 258                         lu_kmem_fini(vvp_caches);
 259         }
 260         return result;
 261 }
 262
 263 void vvp_global_fini(void)
 264 {
 265         ccc_global_fini(&vvp_device_type);
 266         lu_kmem_fini(vvp_caches);
 267 }
 268
 269
 270 /*****************************************************************************
 271  *
 272  * mirror obd-devices into cl devices.
 273  *
 274  */
 275
 276 int cl_sb_init(struct super_block *sb)
 277 {
 278         struct ll_sb_info *sbi;
 279         struct cl_device  *cl;
 280         struct lu_env     *env;
 281         int rc = 0;
 282         int refcheck;
 283
 284         sbi  = ll_s2sbi(sb);
 285         env = cl_env_get(&refcheck);
 286         if (!IS_ERR(env)) {
 287                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 288                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 289                 if (!IS_ERR(cl)) {
 290                         cl2vvp_dev(cl)->vdv_sb = sb;
 291                         sbi->ll_cl = cl;
 292                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 293                 }
 294                 cl_env_put(env, &refcheck);
 295         } else
 296                 rc = PTR_ERR(env);
 297         RETURN(rc);
 298 }
 299
 300 int cl_sb_fini(struct super_block *sb)
 301 {
 302         struct ll_sb_info *sbi;
 303         struct lu_env     *env;
 304         struct cl_device  *cld;
 305         int                refcheck;
 306         int                result;
 307
 308         ENTRY;
 309         sbi = ll_s2sbi(sb);
 310         env = cl_env_get(&refcheck);
 311         if (!IS_ERR(env)) {
 312                 cld = sbi->ll_cl;
 313
 314                 if (cld != NULL) {
 315                         cl_stack_fini(env, cld);
 316                         sbi->ll_cl = NULL;
 317                         sbi->ll_site = NULL;
 318                 }
 319                 cl_env_put(env, &refcheck);
 320                 result = 0;
 321         } else {
 322                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 323                 result = PTR_ERR(env);
 324         }
 325
 326         RETURN(result);
 327 }
 328
 329 /****************************************************************************
 330  *
 331  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 332  *
 333  ****************************************************************************/
 334
 335 /*
 336  * To represent contents of a page cache as a byte stream, following
 337  * information if encoded in 64bit offset:
 338  *
 339  *       - file hash bucket in lu_site::ls_hash[]       28bits
 340  *
 341  *       - how far file is from bucket head              4bits
 342  *
 343  *       - page index                                   32bits
 344  *
 345  * First two data identify a file in the cache uniquely.
 346  */
 347
 348 #define PGC_OBJ_SHIFT (32 + 4)
 349 #define PGC_DEPTH_SHIFT (32)
 350
 351 struct vvp_pgcache_id {
 352         unsigned                 vpi_bucket;
 353         unsigned                 vpi_depth;
 354         uint32_t                 vpi_index;
 355
 356         unsigned                 vpi_curdep;
 357         struct lu_object_header *vpi_obj;
 358 };
 359
 360 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 361 {
 362         CLASSERT(sizeof(pos) == sizeof(__u64));
 363
 364         id->vpi_index  = pos & 0xffffffff;
 365         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 366         id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
 367 }
 368
 369 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 370 {
 371         return
 372                 ((__u64)id->vpi_index) |
 373                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 374                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 375 }
 376
 377 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 378                                struct hlist_node *hnode, void *data)
 379 {
 380         struct vvp_pgcache_id   *id  = data;
 381         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 382
 383         if (id->vpi_curdep-- > 0)
 384                 return 0; /* continue */
 385
 386         if (lu_object_is_dying(hdr))
 387                 return 1;
 388
 389         cfs_hash_get(hs, hnode);
 390         id->vpi_obj = hdr;
 391         return 1;
 392 }
 393
 394 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 395                                          struct lu_device *dev,
 396                                          struct vvp_pgcache_id *id)
 397 {
 398         LASSERT(lu_device_is_cl(dev));
 399
 400         id->vpi_depth &= 0xf;
 401         id->vpi_obj    = NULL;
 402         id->vpi_curdep = id->vpi_depth;
 403
 404         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 405                                 vvp_pgcache_obj_get, id);
 406         if (id->vpi_obj != NULL) {
 407                 struct lu_object *lu_obj;
 408
 409                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 410                 if (lu_obj != NULL) {
 411                         lu_object_ref_add(lu_obj, "dump", current);
 412                         return lu2cl(lu_obj);
 413                 }
 414                 lu_object_put(env, lu_object_top(id->vpi_obj));
 415
 416         } else if (id->vpi_curdep > 0) {
 417                 id->vpi_depth = 0xf;
 418         }
 419         return NULL;
 420 }
 421
 422 static loff_t vvp_pgcache_find(const struct lu_env *env,
 423                                struct lu_device *dev, loff_t pos)
 424 {
 425         struct cl_object     *clob;
 426         struct lu_site       *site;
 427         struct vvp_pgcache_id id;
 428
 429         site = dev->ld_site;
 430         vvp_pgcache_id_unpack(pos, &id);
 431
 432         while (1) {
 433                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 434                         return ~0ULL;
 435                 clob = vvp_pgcache_obj(env, dev, &id);
 436                 if (clob != NULL) {
 437                         struct inode *inode = vvp_object_inode(clob);
 438                         struct page *vmpage;
 439                         int nr;
 440
 441                         nr = find_get_pages_contig(inode->i_mapping,
 442                                                    id.vpi_index, 1, &vmpage);
 443                         if (nr > 0) {
 444                                 id.vpi_index = vmpage->index;
 445                                 /* Cant support over 16T file */
 446                                 nr = !(vmpage->index > 0xffffffff);
 447                                 page_cache_release(vmpage);
 448                         }
 449
 450                         lu_object_ref_del(&clob->co_lu, "dump", current);
 451                         cl_object_put(env, clob);
 452                         if (nr > 0)
 453                                 return vvp_pgcache_id_pack(&id);
 454                 }
 455                 /* to the next object. */
 456                 ++id.vpi_depth;
 457                 id.vpi_depth &= 0xf;
 458                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 459                         return ~0ULL;
 460                 id.vpi_index = 0;
 461         }
 462 }
 463
 464 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 465         if (test_bit(PG_##flag, &(page)->flags)) {                  \
 466                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 467                 has_flags = 1;                                          \
 468         }                                                               \
 469 } while(0)
 470
 471 static void vvp_pgcache_page_show(const struct lu_env *env,
 472                                   struct seq_file *seq, struct cl_page *page)
 473 {
 474         struct ccc_page *cpg;
 475         struct page      *vmpage;
 476         int              has_flags;
 477
 478         cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
 479         vmpage = cpg->cpg_page;
 480         seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
 481                    0 /* gen */,
 482                    cpg, page,
 483                    "none",
 484                    cpg->cpg_write_queued ? "wq" : "- ",
 485                    cpg->cpg_defer_uptodate ? "du" : "- ",
 486                    PageWriteback(vmpage) ? "wb" : "-",
 487                    vmpage,
 488                    PFID(ll_inode2fid(vmpage->mapping->host)),
 489                    vmpage->mapping->host, vmpage->index,
 490                    page_count(vmpage));
 491         has_flags = 0;
 492         seq_page_flag(seq, vmpage, locked, has_flags);
 493         seq_page_flag(seq, vmpage, error, has_flags);
 494         seq_page_flag(seq, vmpage, referenced, has_flags);
 495         seq_page_flag(seq, vmpage, uptodate, has_flags);
 496         seq_page_flag(seq, vmpage, dirty, has_flags);
 497         seq_page_flag(seq, vmpage, writeback, has_flags);
 498         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 499 }
 500
 501 static int vvp_pgcache_show(struct seq_file *f, void *v)
 502 {
 503         loff_t                   pos;
 504         struct ll_sb_info       *sbi;
 505         struct cl_object        *clob;
 506         struct lu_env           *env;
 507         struct vvp_pgcache_id    id;
 508         int                      refcheck;
 509         int                      result;
 510
 511         env = cl_env_get(&refcheck);
 512         if (!IS_ERR(env)) {
 513                 pos = *(loff_t *) v;
 514                 vvp_pgcache_id_unpack(pos, &id);
 515                 sbi = f->private;
 516                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 517                 if (clob != NULL) {
 518                         struct inode *inode = vvp_object_inode(clob);
 519                         struct cl_page *page = NULL;
 520                         struct page *vmpage;
 521
 522                         result = find_get_pages_contig(inode->i_mapping,
 523                                                       id.vpi_index, 1, &vmpage);
 524                         if (result > 0) {
 525                                 lock_page(vmpage);
 526                                 page = cl_vmpage_page(vmpage, clob);
 527                                 unlock_page(vmpage);
 528
 529                                 page_cache_release(vmpage);
 530                         }
 531
 532                         seq_printf(f, "%8x@"DFID": ", id.vpi_index,
 533                                    PFID(lu_object_fid(&clob->co_lu)));
 534                         if (page != NULL) {
 535                                 vvp_pgcache_page_show(env, f, page);
 536                                 cl_page_put(env, page);
 537                         } else
 538                                 seq_puts(f, "missing\n");
 539                         lu_object_ref_del(&clob->co_lu, "dump", current);
 540                         cl_object_put(env, clob);
 541                 } else
 542                         seq_printf(f, "%llx missing\n", pos);
 543                 cl_env_put(env, &refcheck);
 544                 result = 0;
 545         } else
 546                 result = PTR_ERR(env);
 547         return result;
 548 }
 549
 550 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 551 {
 552         struct ll_sb_info *sbi;
 553         struct lu_env     *env;
 554         int                refcheck;
 555
 556         sbi = f->private;
 557
 558         env = cl_env_get(&refcheck);
 559         if (!IS_ERR(env)) {
 560                 sbi = f->private;
 561                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 562                         pos = ERR_PTR(-EFBIG);
 563                 else {
 564                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 565                                                 *pos);
 566                         if (*pos == ~0ULL)
 567                                 pos = NULL;
 568                 }
 569                 cl_env_put(env, &refcheck);
 570         }
 571         return pos;
 572 }
 573
 574 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 575 {
 576         struct ll_sb_info *sbi;
 577         struct lu_env     *env;
 578         int                refcheck;
 579
 580         env = cl_env_get(&refcheck);
 581         if (!IS_ERR(env)) {
 582                 sbi = f->private;
 583                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 584                 if (*pos == ~0ULL)
 585                         pos = NULL;
 586                 cl_env_put(env, &refcheck);
 587         }
 588         return pos;
 589 }
 590
 591 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 592 {
 593         /* Nothing to do */
 594 }
 595
 596 static struct seq_operations vvp_pgcache_ops = {
 597         .start = vvp_pgcache_start,
 598         .next  = vvp_pgcache_next,
 599         .stop  = vvp_pgcache_stop,
 600         .show  = vvp_pgcache_show
 601 };
 602
 603 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 604 {
 605         struct ll_sb_info       *sbi = PDE_DATA(inode);
 606         struct seq_file         *seq;
 607         int                     result;
 608
 609         result = seq_open(filp, &vvp_pgcache_ops);
 610         if (result == 0) {
 611                 seq = filp->private_data;
 612                 seq->private = sbi;
 613         }
 614         return result;
 615 }
 616
 617 const struct file_operations vvp_dump_pgcache_file_ops = {
 618         .owner   = THIS_MODULE,
 619         .open    = vvp_dump_pgcache_seq_open,
 620         .read    = seq_read,
 621         .llseek  = seq_lseek,
 622         .release = seq_release,
 623 };