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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * cl_device and cl_device_type implementation for VVP layer.
  33  *
  34  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  35  *   Author: Jinshan Xiong <jinshan.xiong@intel.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LLITE
  39
  40 #include <obd.h>
  41 #include "llite_internal.h"
  42 #include "vvp_internal.h"
  43 #include <linux/kallsyms.h>
  44
  45 /*****************************************************************************
  46  *
  47  * Vvp device and device type functions.
  48  *
  49  */
  50
  51 /*
  52  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  53  * "llite_" (var. "ll_") prefix.
  54  */
  55
  56 static struct kmem_cache *ll_thread_kmem;
  57 struct kmem_cache *vvp_object_kmem;
  58 static struct kmem_cache *vvp_session_kmem;
  59 static struct kmem_cache *vvp_thread_kmem;
  60
  61 static struct lu_kmem_descr vvp_caches[] = {
  62         {
  63                 .ckd_cache = &ll_thread_kmem,
  64                 .ckd_name  = "ll_thread_kmem",
  65                 .ckd_size  = sizeof(struct ll_thread_info),
  66         },
  67         {
  68                 .ckd_cache = &vvp_object_kmem,
  69                 .ckd_name  = "vvp_object_kmem",
  70                 .ckd_size  = sizeof(struct vvp_object),
  71         },
  72         {
  73                 .ckd_cache = &vvp_session_kmem,
  74                 .ckd_name  = "vvp_session_kmem",
  75                 .ckd_size  = sizeof (struct vvp_session)
  76         },
  77         {
  78                 .ckd_cache = &vvp_thread_kmem,
  79                 .ckd_name  = "vvp_thread_kmem",
  80                 .ckd_size  = sizeof(struct vvp_thread_info),
  81         },
  82         {
  83                 .ckd_cache = NULL
  84         }
  85 };
  86
  87 static void *ll_thread_key_init(const struct lu_context *ctx,
  88                                 struct lu_context_key *key)
  89 {
  90         struct ll_thread_info *lti;
  91
  92         OBD_SLAB_ALLOC_PTR_GFP(lti, ll_thread_kmem, GFP_NOFS);
  93         if (lti == NULL)
  94                 lti = ERR_PTR(-ENOMEM);
  95
  96         return lti;
  97 }
  98
  99 static void ll_thread_key_fini(const struct lu_context *ctx,
 100                                struct lu_context_key *key, void *data)
 101 {
 102         struct ll_thread_info *lti = data;
 103
 104         OBD_SLAB_FREE_PTR(lti, ll_thread_kmem);
 105 }
 106
 107 struct lu_context_key ll_thread_key = {
 108         .lct_tags = LCT_CL_THREAD,
 109         .lct_init = ll_thread_key_init,
 110         .lct_fini = ll_thread_key_fini,
 111 };
 112
 113 static void *vvp_session_key_init(const struct lu_context *ctx,
 114                                   struct lu_context_key *key)
 115 {
 116         struct vvp_session *session;
 117
 118         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
 119         if (session == NULL)
 120                 session = ERR_PTR(-ENOMEM);
 121         return session;
 122 }
 123
 124 static void vvp_session_key_fini(const struct lu_context *ctx,
 125                                  struct lu_context_key *key, void *data)
 126 {
 127         struct vvp_session *session = data;
 128         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 129 }
 130
 131 struct lu_context_key vvp_session_key = {
 132         .lct_tags = LCT_SESSION,
 133         .lct_init = vvp_session_key_init,
 134         .lct_fini = vvp_session_key_fini
 135 };
 136
 137 static void *vvp_thread_key_init(const struct lu_context *ctx,
 138                                  struct lu_context_key *key)
 139 {
 140         struct vvp_thread_info *vti;
 141
 142         OBD_SLAB_ALLOC_PTR_GFP(vti, vvp_thread_kmem, GFP_NOFS);
 143         if (vti == NULL)
 144                 vti = ERR_PTR(-ENOMEM);
 145         return vti;
 146 }
 147
 148 static void vvp_thread_key_fini(const struct lu_context *ctx,
 149                                 struct lu_context_key *key, void *data)
 150 {
 151         struct vvp_thread_info *vti = data;
 152         OBD_SLAB_FREE_PTR(vti, vvp_thread_kmem);
 153 }
 154
 155 struct lu_context_key vvp_thread_key = {
 156         .lct_tags = LCT_CL_THREAD,
 157         .lct_init = vvp_thread_key_init,
 158         .lct_fini = vvp_thread_key_fini,
 159 };
 160
 161 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 162 LU_TYPE_INIT_FINI(vvp, &ll_thread_key, &vvp_session_key, &vvp_thread_key);
 163
 164 static const struct lu_device_operations vvp_lu_ops = {
 165         .ldo_object_alloc      = vvp_object_alloc
 166 };
 167
 168 static struct lu_device *vvp_device_free(const struct lu_env *env,
 169                                          struct lu_device *d)
 170 {
 171         struct vvp_device *vdv  = lu2vvp_dev(d);
 172         struct cl_site    *site = lu2cl_site(d->ld_site);
 173         struct lu_device  *next = cl2lu_dev(vdv->vdv_next);
 174
 175         if (d->ld_site != NULL) {
 176                 cl_site_fini(site);
 177                 OBD_FREE_PTR(site);
 178         }
 179
 180         cl_device_fini(lu2cl_dev(d));
 181         OBD_FREE_PTR(vdv);
 182         return next;
 183 }
 184
 185 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 186                                           struct lu_device_type *t,
 187                                           struct lustre_cfg *cfg)
 188 {
 189         struct vvp_device *vdv;
 190         struct lu_device *lud;
 191         struct cl_site *site;
 192         int rc;
 193         ENTRY;
 194
 195         OBD_ALLOC_PTR(vdv);
 196         if (vdv == NULL)
 197                 RETURN(ERR_PTR(-ENOMEM));
 198
 199         lud = &vdv->vdv_cl.cd_lu_dev;
 200         cl_device_init(&vdv->vdv_cl, t);
 201         vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
 202
 203         OBD_ALLOC_PTR(site);
 204         if (site != NULL) {
 205                 rc = cl_site_init(site, &vdv->vdv_cl);
 206                 if (rc == 0)
 207                         rc = lu_site_init_finish(&site->cs_lu);
 208                 else {
 209                         LASSERT(lud->ld_site == NULL);
 210                         CERROR("Cannot init lu_site, rc %d.\n", rc);
 211                         OBD_FREE_PTR(site);
 212                 }
 213         } else
 214                 rc = -ENOMEM;
 215         if (rc != 0) {
 216                 vvp_device_free(env, lud);
 217                 lud = ERR_PTR(rc);
 218         }
 219         RETURN(lud);
 220 }
 221
 222 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
 223                            const char *name, struct lu_device *next)
 224 {
 225         struct vvp_device  *vdv;
 226         int rc;
 227         ENTRY;
 228
 229         vdv = lu2vvp_dev(d);
 230         vdv->vdv_next = lu2cl_dev(next);
 231
 232         LASSERT(d->ld_site != NULL && next->ld_type != NULL);
 233         next->ld_site = d->ld_site;
 234         rc = next->ld_type->ldt_ops->ldto_device_init(
 235                 env, next, next->ld_type->ldt_name, NULL);
 236         if (rc == 0) {
 237                 lu_device_get(next);
 238                 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
 239         }
 240         RETURN(rc);
 241 }
 242
 243 static struct lu_device *vvp_device_fini(const struct lu_env *env,
 244                                          struct lu_device *d)
 245 {
 246         return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
 247 }
 248
 249 static const struct lu_device_type_operations vvp_device_type_ops = {
 250         .ldto_init = vvp_type_init,
 251         .ldto_fini = vvp_type_fini,
 252
 253         .ldto_start = vvp_type_start,
 254         .ldto_stop  = vvp_type_stop,
 255
 256         .ldto_device_alloc      = vvp_device_alloc,
 257         .ldto_device_free       = vvp_device_free,
 258         .ldto_device_init       = vvp_device_init,
 259         .ldto_device_fini       = vvp_device_fini,
 260 };
 261
 262 struct lu_device_type vvp_device_type = {
 263         .ldt_tags     = LU_DEVICE_CL,
 264         .ldt_name     = LUSTRE_VVP_NAME,
 265         .ldt_ops      = &vvp_device_type_ops,
 266         .ldt_ctx_tags = LCT_CL_THREAD
 267 };
 268
 269 /**
 270  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 271  * pressure, when environments cannot be allocated.
 272  */
 273 int vvp_global_init(void)
 274 {
 275         int rc;
 276
 277         rc = lu_kmem_init(vvp_caches);
 278         if (rc != 0)
 279                 return rc;
 280
 281         rc = lu_device_type_init(&vvp_device_type);
 282         if (rc != 0)
 283                 goto out_kmem;
 284
 285 #ifndef HAVE_ACCOUNT_PAGE_DIRTIED_EXPORT
 286         /*
 287          * Kernel v5.2-5678-gac1c3e4 no longer exports account_page_dirtied
 288          */
 289         vvp_account_page_dirtied = (void *)
 290                 kallsyms_lookup_name("account_page_dirtied");
 291         BUG_ON(!vvp_account_page_dirtied);
 292 #endif
 293
 294         return 0;
 295
 296 out_kmem:
 297         lu_kmem_fini(vvp_caches);
 298
 299         return rc;
 300 }
 301
 302 void vvp_global_fini(void)
 303 {
 304         lu_device_type_fini(&vvp_device_type);
 305         lu_kmem_fini(vvp_caches);
 306 }
 307
 308 /*****************************************************************************
 309  *
 310  * mirror obd-devices into cl devices.
 311  *
 312  */
 313
 314 int cl_sb_init(struct super_block *sb)
 315 {
 316         struct ll_sb_info *sbi;
 317         struct cl_device  *cl;
 318         struct lu_env     *env;
 319         int rc = 0;
 320         __u16 refcheck;
 321
 322         sbi  = ll_s2sbi(sb);
 323         env = cl_env_get(&refcheck);
 324         if (!IS_ERR(env)) {
 325                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 326                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 327                 if (!IS_ERR(cl)) {
 328                         sbi->ll_cl = cl;
 329                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 330                 }
 331                 cl_env_put(env, &refcheck);
 332         } else
 333                 rc = PTR_ERR(env);
 334         RETURN(rc);
 335 }
 336
 337 int cl_sb_fini(struct super_block *sb)
 338 {
 339         struct ll_sb_info *sbi;
 340         struct lu_env     *env;
 341         struct cl_device  *cld;
 342         __u16              refcheck;
 343         int                result;
 344
 345         ENTRY;
 346         sbi = ll_s2sbi(sb);
 347         env = cl_env_get(&refcheck);
 348         if (!IS_ERR(env)) {
 349                 cld = sbi->ll_cl;
 350
 351                 if (cld != NULL) {
 352                         cl_stack_fini(env, cld);
 353                         sbi->ll_cl = NULL;
 354                         sbi->ll_site = NULL;
 355                 }
 356                 cl_env_put(env, &refcheck);
 357                 result = 0;
 358         } else {
 359                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 360                 result = PTR_ERR(env);
 361         }
 362
 363         RETURN(result);
 364 }
 365
 366 /****************************************************************************
 367  *
 368  * debugfs/lustre/llite/$MNT/dump_page_cache
 369  *
 370  ****************************************************************************/
 371
 372 struct vvp_pgcache_id {
 373         unsigned                 vpi_bucket;
 374         unsigned                 vpi_depth;
 375         uint32_t                 vpi_index;
 376
 377         unsigned                 vpi_curdep;
 378         struct lu_object_header *vpi_obj;
 379 };
 380
 381 struct vvp_seq_private {
 382         struct ll_sb_info       *vsp_sbi;
 383         struct lu_env           *vsp_env;
 384         u16                     vsp_refcheck;
 385         struct cl_object        *vsp_clob;
 386         struct vvp_pgcache_id   vvp_id;
 387         /*
 388          * prev_pos is the 'pos' of the last object returned
 389          * by ->start of ->next.
 390          */
 391         loff_t                  vvp_prev_pos;
 392 };
 393
 394 static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 395                                struct hlist_node *hnode, void *data)
 396 {
 397         struct vvp_pgcache_id   *id  = data;
 398         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 399
 400         if (lu_object_is_dying(hdr))
 401                 return 0;
 402
 403         if (id->vpi_curdep-- > 0)
 404                 return 0; /* continue */
 405
 406         cfs_hash_get(hs, hnode);
 407         id->vpi_obj = hdr;
 408         return 1;
 409 }
 410
 411 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 412                                          struct lu_device *dev,
 413                                          struct vvp_pgcache_id *id)
 414 {
 415         LASSERT(lu_device_is_cl(dev));
 416
 417         id->vpi_obj = NULL;
 418         id->vpi_curdep = id->vpi_depth;
 419
 420         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 421                                 vvp_pgcache_obj_get, id);
 422         if (id->vpi_obj != NULL) {
 423                 struct lu_object *lu_obj;
 424
 425                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 426                 if (lu_obj != NULL) {
 427                         lu_object_ref_add(lu_obj, "dump", current);
 428                         return lu2cl(lu_obj);
 429                 }
 430                 lu_object_put(env, lu_object_top(id->vpi_obj));
 431         }
 432         return NULL;
 433 }
 434
 435 static struct page *vvp_pgcache_current(struct vvp_seq_private *priv)
 436 {
 437         struct lu_device *dev = &priv->vsp_sbi->ll_cl->cd_lu_dev;
 438
 439         while (1) {
 440                 struct inode *inode;
 441                 struct page *vmpage;
 442                 int nr;
 443
 444                 if (!priv->vsp_clob) {
 445                         struct cl_object *clob;
 446
 447                         while ((clob = vvp_pgcache_obj(priv->vsp_env, dev, &priv->vvp_id)) == NULL &&
 448                                ++(priv->vvp_id.vpi_bucket) < CFS_HASH_NHLIST(dev->ld_site->ls_obj_hash))
 449                                 priv->vvp_id.vpi_depth = 0;
 450                         if (!clob)
 451                                 return NULL;
 452                         priv->vsp_clob = clob;
 453                         priv->vvp_id.vpi_index = 0;
 454                 }
 455
 456                 inode = vvp_object_inode(priv->vsp_clob);
 457                 nr = find_get_pages_contig(inode->i_mapping, priv->vvp_id.vpi_index, 1, &vmpage);
 458                 if (nr > 0) {
 459                         priv->vvp_id.vpi_index = vmpage->index;
 460                         return vmpage;
 461                 }
 462                 lu_object_ref_del(&priv->vsp_clob->co_lu, "dump", current);
 463                 cl_object_put(priv->vsp_env, priv->vsp_clob);
 464                 priv->vsp_clob = NULL;
 465                 priv->vvp_id.vpi_index = 0;
 466                 priv->vvp_id.vpi_depth++;
 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 | %p "DFID"(%p) %lu %u [",
 487                    0 /* gen */,
 488                    vpg, page,
 489                    "none",
 490                    vpg->vpg_defer_uptodate ? "du" : "- ",
 491                    PageWriteback(vmpage) ? "wb" : "-",
 492                    vmpage,
 493                    PFID(ll_inode2fid(vmpage->mapping->host)),
 494                    vmpage->mapping->host, vmpage->index,
 495                    page_count(vmpage));
 496         has_flags = 0;
 497         seq_page_flag(seq, vmpage, locked, has_flags);
 498         seq_page_flag(seq, vmpage, error, has_flags);
 499         seq_page_flag(seq, vmpage, referenced, has_flags);
 500         seq_page_flag(seq, vmpage, uptodate, has_flags);
 501         seq_page_flag(seq, vmpage, dirty, has_flags);
 502         seq_page_flag(seq, vmpage, writeback, has_flags);
 503         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 504 }
 505
 506 static int vvp_pgcache_show(struct seq_file *f, void *v)
 507 {
 508         struct vvp_seq_private *priv = f->private;
 509         struct page *vmpage = v;
 510         struct cl_page *page;
 511
 512         seq_printf(f, "%8lx@" DFID ": ", vmpage->index,
 513                    PFID(lu_object_fid(&priv->vsp_clob->co_lu)));
 514         lock_page(vmpage);
 515         page = cl_vmpage_page(vmpage, priv->vsp_clob);
 516         unlock_page(vmpage);
 517         put_page(vmpage);
 518
 519         if (page) {
 520                 vvp_pgcache_page_show(priv->vsp_env, f, page);
 521                 cl_page_put(priv->vsp_env, page);
 522         } else {
 523                 seq_puts(f, "missing\n");
 524         }
 525
 526         return 0;
 527 }
 528
 529 static void vvp_pgcache_rewind(struct vvp_seq_private *priv)
 530 {
 531         if (priv->vvp_prev_pos) {
 532                 memset(&priv->vvp_id, 0, sizeof(priv->vvp_id));
 533                 priv->vvp_prev_pos = 0;
 534                 if (priv->vsp_clob) {
 535                         lu_object_ref_del(&priv->vsp_clob->co_lu, "dump",
 536                                           current);
 537                         cl_object_put(priv->vsp_env, priv->vsp_clob);
 538                 }
 539                 priv->vsp_clob = NULL;
 540         }
 541 }
 542
 543 static struct page *vvp_pgcache_next_page(struct vvp_seq_private *priv)
 544 {
 545         priv->vvp_id.vpi_index += 1;
 546         return vvp_pgcache_current(priv);
 547 }
 548
 549 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 550 {
 551         struct vvp_seq_private *priv = f->private;
 552
 553         if (*pos == 0) {
 554                 vvp_pgcache_rewind(priv);
 555         } else if (*pos == priv->vvp_prev_pos) {
 556                 /* Return the current item */;
 557         } else {
 558                 WARN_ON(*pos != priv->vvp_prev_pos + 1);
 559                 priv->vvp_id.vpi_index += 1;
 560         }
 561
 562         priv->vvp_prev_pos = *pos;
 563         return vvp_pgcache_current(priv);
 564 }
 565
 566 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 567 {
 568         struct vvp_seq_private *priv = f->private;
 569
 570         WARN_ON(*pos != priv->vvp_prev_pos);
 571         *pos += 1;
 572         priv->vvp_prev_pos = *pos;
 573         return vvp_pgcache_next_page(priv);
 574 }
 575
 576 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 577 {
 578         /* Nothing to do */
 579 }
 580
 581 static struct seq_operations vvp_pgcache_ops = {
 582         .start = vvp_pgcache_start,
 583         .next  = vvp_pgcache_next,
 584         .stop  = vvp_pgcache_stop,
 585         .show  = vvp_pgcache_show
 586 };
 587
 588 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 589 {
 590         struct vvp_seq_private *priv;
 591
 592         priv = __seq_open_private(filp, &vvp_pgcache_ops, sizeof(*priv));
 593         if (!priv)
 594                 return -ENOMEM;
 595
 596         priv->vsp_sbi = inode->i_private;
 597         priv->vsp_env = cl_env_get(&priv->vsp_refcheck);
 598         priv->vsp_clob = NULL;
 599         memset(&priv->vvp_id, 0, sizeof(priv->vvp_id));
 600         if (IS_ERR(priv->vsp_env)) {
 601                 int err = PTR_ERR(priv->vsp_env);
 602
 603                 seq_release_private(inode, filp);
 604                 return err;
 605         }
 606
 607         return 0;
 608 }
 609
 610 static int vvp_dump_pgcache_seq_release(struct inode *inode, struct file *file)
 611 {
 612         struct seq_file *seq = file->private_data;
 613         struct vvp_seq_private *priv = seq->private;
 614
 615         if (priv->vsp_clob) {
 616                 lu_object_ref_del(&priv->vsp_clob->co_lu, "dump", current);
 617                 cl_object_put(priv->vsp_env, priv->vsp_clob);
 618         }
 619
 620         cl_env_put(priv->vsp_env, &priv->vsp_refcheck);
 621         return seq_release_private(inode, file);
 622 }
 623
 624 const struct file_operations vvp_dump_pgcache_file_ops = {
 625         .owner   = THIS_MODULE,
 626         .open    = vvp_dump_pgcache_seq_open,
 627         .read    = seq_read,
 628         .llseek  = seq_lseek,
 629         .release = vvp_dump_pgcache_seq_release,
 630 };