kernel/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, 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  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include "../include/obd.h"
  44 #include "../include/lustre_lite.h"
  45 #include "llite_internal.h"
  46 #include "vvp_internal.h"
  47
  48 /*****************************************************************************
  49  *
  50  * Vvp device and device type functions.
  51  *
  52  */
  53
  54 /*
  55  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  56  * "llite_" (var. "ll_") prefix.
  57  */
  58
  59 static struct kmem_cache *vvp_thread_kmem;
  60 static struct kmem_cache *vvp_session_kmem;
  61 static struct lu_kmem_descr vvp_caches[] = {
  62         {
  63                 .ckd_cache = &vvp_thread_kmem,
  64                 .ckd_name  = "vvp_thread_kmem",
  65                 .ckd_size  = sizeof(struct vvp_thread_info),
  66         },
  67         {
  68                 .ckd_cache = &vvp_session_kmem,
  69                 .ckd_name  = "vvp_session_kmem",
  70                 .ckd_size  = sizeof(struct vvp_session)
  71         },
  72         {
  73                 .ckd_cache = NULL
  74         }
  75 };
  76
  77 static void *vvp_key_init(const struct lu_context *ctx,
  78                           struct lu_context_key *key)
  79 {
  80         struct vvp_thread_info *info;
  81
  82         info = kmem_cache_alloc(vvp_thread_kmem, GFP_NOFS | __GFP_ZERO);
  83         if (info == NULL)
  84                 info = ERR_PTR(-ENOMEM);
  85         return info;
  86 }
  87
  88 static void vvp_key_fini(const struct lu_context *ctx,
  89                          struct lu_context_key *key, void *data)
  90 {
  91         struct vvp_thread_info *info = data;
  92
  93         kmem_cache_free(vvp_thread_kmem, info);
  94 }
  95
  96 static void *vvp_session_key_init(const struct lu_context *ctx,
  97                                   struct lu_context_key *key)
  98 {
  99         struct vvp_session *session;
 100
 101         session = kmem_cache_alloc(vvp_session_kmem, GFP_NOFS | __GFP_ZERO);
 102         if (session == NULL)
 103                 session = ERR_PTR(-ENOMEM);
 104         return session;
 105 }
 106
 107 static void vvp_session_key_fini(const struct lu_context *ctx,
 108                                  struct lu_context_key *key, void *data)
 109 {
 110         struct vvp_session *session = data;
 111
 112         kmem_cache_free(vvp_session_kmem, session);
 113 }
 114
 115 struct lu_context_key vvp_key = {
 116         .lct_tags = LCT_CL_THREAD,
 117         .lct_init = vvp_key_init,
 118         .lct_fini = vvp_key_fini
 119 };
 120
 121 struct lu_context_key vvp_session_key = {
 122         .lct_tags = LCT_SESSION,
 123         .lct_init = vvp_session_key_init,
 124         .lct_fini = vvp_session_key_fini
 125 };
 126
 127 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 128 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
 129
 130 static const struct lu_device_operations vvp_lu_ops = {
 131         .ldo_object_alloc      = vvp_object_alloc
 132 };
 133
 134 static const struct cl_device_operations vvp_cl_ops = {
 135         .cdo_req_init = ccc_req_init
 136 };
 137
 138 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 139                                           struct lu_device_type *t,
 140                                           struct lustre_cfg *cfg)
 141 {
 142         return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
 143 }
 144
 145 static const struct lu_device_type_operations vvp_device_type_ops = {
 146         .ldto_init = vvp_type_init,
 147         .ldto_fini = vvp_type_fini,
 148
 149         .ldto_start = vvp_type_start,
 150         .ldto_stop  = vvp_type_stop,
 151
 152         .ldto_device_alloc = vvp_device_alloc,
 153         .ldto_device_free  = ccc_device_free,
 154         .ldto_device_init  = ccc_device_init,
 155         .ldto_device_fini  = ccc_device_fini
 156 };
 157
 158 struct lu_device_type vvp_device_type = {
 159         .ldt_tags     = LU_DEVICE_CL,
 160         .ldt_name     = LUSTRE_VVP_NAME,
 161         .ldt_ops      = &vvp_device_type_ops,
 162         .ldt_ctx_tags = LCT_CL_THREAD
 163 };
 164
 165 /**
 166  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 167  * pressure, when environments cannot be allocated.
 168  */
 169 int vvp_global_init(void)
 170 {
 171         int result;
 172
 173         result = lu_kmem_init(vvp_caches);
 174         if (result == 0) {
 175                 result = ccc_global_init(&vvp_device_type);
 176                 if (result != 0)
 177                         lu_kmem_fini(vvp_caches);
 178         }
 179         return result;
 180 }
 181
 182 void vvp_global_fini(void)
 183 {
 184         ccc_global_fini(&vvp_device_type);
 185         lu_kmem_fini(vvp_caches);
 186 }
 187
 188 /*****************************************************************************
 189  *
 190  * mirror obd-devices into cl devices.
 191  *
 192  */
 193
 194 int cl_sb_init(struct super_block *sb)
 195 {
 196         struct ll_sb_info *sbi;
 197         struct cl_device  *cl;
 198         struct lu_env     *env;
 199         int rc = 0;
 200         int refcheck;
 201
 202         sbi  = ll_s2sbi(sb);
 203         env = cl_env_get(&refcheck);
 204         if (!IS_ERR(env)) {
 205                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 206                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 207                 if (!IS_ERR(cl)) {
 208                         cl2ccc_dev(cl)->cdv_sb = sb;
 209                         sbi->ll_cl = cl;
 210                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 211                 }
 212                 cl_env_put(env, &refcheck);
 213         } else
 214                 rc = PTR_ERR(env);
 215         return rc;
 216 }
 217
 218 int cl_sb_fini(struct super_block *sb)
 219 {
 220         struct ll_sb_info *sbi;
 221         struct lu_env     *env;
 222         struct cl_device  *cld;
 223         int             refcheck;
 224         int             result;
 225
 226         sbi = ll_s2sbi(sb);
 227         env = cl_env_get(&refcheck);
 228         if (!IS_ERR(env)) {
 229                 cld = sbi->ll_cl;
 230
 231                 if (cld != NULL) {
 232                         cl_stack_fini(env, cld);
 233                         sbi->ll_cl = NULL;
 234                         sbi->ll_site = NULL;
 235                 }
 236                 cl_env_put(env, &refcheck);
 237                 result = 0;
 238         } else {
 239                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 240                 result = PTR_ERR(env);
 241         }
 242         /*
 243          * If mount failed (sbi->ll_cl == NULL), and this there are no other
 244          * mounts, stop device types manually (this usually happens
 245          * automatically when last device is destroyed).
 246          */
 247         lu_types_stop();
 248         return result;
 249 }
 250
 251 /****************************************************************************
 252  *
 253  * debugfs/lustre/llite/$MNT/dump_page_cache
 254  *
 255  ****************************************************************************/
 256
 257 /*
 258  * To represent contents of a page cache as a byte stream, following
 259  * information if encoded in 64bit offset:
 260  *
 261  *       - file hash bucket in lu_site::ls_hash[]       28bits
 262  *
 263  *       - how far file is from bucket head           4bits
 264  *
 265  *       - page index                              32bits
 266  *
 267  * First two data identify a file in the cache uniquely.
 268  */
 269
 270 #define PGC_OBJ_SHIFT (32 + 4)
 271 #define PGC_DEPTH_SHIFT (32)
 272
 273 struct vvp_pgcache_id {
 274         unsigned                 vpi_bucket;
 275         unsigned                 vpi_depth;
 276         uint32_t                 vpi_index;
 277
 278         unsigned                 vpi_curdep;
 279         struct lu_object_header *vpi_obj;
 280 };
 281
 282 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 283 {
 284         CLASSERT(sizeof(pos) == sizeof(__u64));
 285
 286         id->vpi_index  = pos & 0xffffffff;
 287         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 288         id->vpi_bucket = (unsigned long long)pos >> PGC_OBJ_SHIFT;
 289 }
 290
 291 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 292 {
 293         return
 294                 ((__u64)id->vpi_index) |
 295                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 296                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 297 }
 298
 299 static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 300                                struct hlist_node *hnode, void *data)
 301 {
 302         struct vvp_pgcache_id   *id  = data;
 303         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 304
 305         if (id->vpi_curdep-- > 0)
 306                 return 0; /* continue */
 307
 308         if (lu_object_is_dying(hdr))
 309                 return 1;
 310
 311         cfs_hash_get(hs, hnode);
 312         id->vpi_obj = hdr;
 313         return 1;
 314 }
 315
 316 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 317                                          struct lu_device *dev,
 318                                          struct vvp_pgcache_id *id)
 319 {
 320         LASSERT(lu_device_is_cl(dev));
 321
 322         id->vpi_depth &= 0xf;
 323         id->vpi_obj    = NULL;
 324         id->vpi_curdep = id->vpi_depth;
 325
 326         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 327                                 vvp_pgcache_obj_get, id);
 328         if (id->vpi_obj != NULL) {
 329                 struct lu_object *lu_obj;
 330
 331                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 332                 if (lu_obj != NULL) {
 333                         lu_object_ref_add(lu_obj, "dump", current);
 334                         return lu2cl(lu_obj);
 335                 }
 336                 lu_object_put(env, lu_object_top(id->vpi_obj));
 337
 338         } else if (id->vpi_curdep > 0) {
 339                 id->vpi_depth = 0xf;
 340         }
 341         return NULL;
 342 }
 343
 344 static loff_t vvp_pgcache_find(const struct lu_env *env,
 345                                struct lu_device *dev, loff_t pos)
 346 {
 347         struct cl_object     *clob;
 348         struct lu_site       *site;
 349         struct vvp_pgcache_id id;
 350
 351         site = dev->ld_site;
 352         vvp_pgcache_id_unpack(pos, &id);
 353
 354         while (1) {
 355                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 356                         return ~0ULL;
 357                 clob = vvp_pgcache_obj(env, dev, &id);
 358                 if (clob != NULL) {
 359                         struct cl_object_header *hdr;
 360                         int                   nr;
 361                         struct cl_page    *pg;
 362
 363                         /* got an object. Find next page. */
 364                         hdr = cl_object_header(clob);
 365
 366                         spin_lock(&hdr->coh_page_guard);
 367                         nr = radix_tree_gang_lookup(&hdr->coh_tree,
 368                                                     (void **)&pg,
 369                                                     id.vpi_index, 1);
 370                         if (nr > 0) {
 371                                 id.vpi_index = pg->cp_index;
 372                                 /* Cant support over 16T file */
 373                                 nr = !(pg->cp_index > 0xffffffff);
 374                         }
 375                         spin_unlock(&hdr->coh_page_guard);
 376
 377                         lu_object_ref_del(&clob->co_lu, "dump", current);
 378                         cl_object_put(env, clob);
 379                         if (nr > 0)
 380                                 return vvp_pgcache_id_pack(&id);
 381                 }
 382                 /* to the next object. */
 383                 ++id.vpi_depth;
 384                 id.vpi_depth &= 0xf;
 385                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 386                         return ~0ULL;
 387                 id.vpi_index = 0;
 388         }
 389 }
 390
 391 #define seq_page_flag(seq, page, flag, has_flags) do {            \
 392         if (test_bit(PG_##flag, &(page)->flags)) {                \
 393                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 394                 has_flags = 1;                                    \
 395         }                                                              \
 396 } while (0)
 397
 398 static void vvp_pgcache_page_show(const struct lu_env *env,
 399                                   struct seq_file *seq, struct cl_page *page)
 400 {
 401         struct ccc_page *cpg;
 402         struct page      *vmpage;
 403         int           has_flags;
 404
 405         cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
 406         vmpage = cpg->cpg_page;
 407         seq_printf(seq, " %5i | %p %p %s %s %s %s | %p %lu/%u(%p) %lu %u [",
 408                    0 /* gen */,
 409                    cpg, page,
 410                    "none",
 411                    cpg->cpg_write_queued ? "wq" : "- ",
 412                    cpg->cpg_defer_uptodate ? "du" : "- ",
 413                    PageWriteback(vmpage) ? "wb" : "-",
 414                    vmpage, vmpage->mapping->host->i_ino,
 415                    vmpage->mapping->host->i_generation,
 416                    vmpage->mapping->host, vmpage->index,
 417                    page_count(vmpage));
 418         has_flags = 0;
 419         seq_page_flag(seq, vmpage, locked, has_flags);
 420         seq_page_flag(seq, vmpage, error, has_flags);
 421         seq_page_flag(seq, vmpage, referenced, has_flags);
 422         seq_page_flag(seq, vmpage, uptodate, has_flags);
 423         seq_page_flag(seq, vmpage, dirty, has_flags);
 424         seq_page_flag(seq, vmpage, writeback, has_flags);
 425         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 426 }
 427
 428 static int vvp_pgcache_show(struct seq_file *f, void *v)
 429 {
 430         loff_t             pos;
 431         struct ll_sb_info       *sbi;
 432         struct cl_object        *clob;
 433         struct lu_env      *env;
 434         struct cl_page    *page;
 435         struct cl_object_header *hdr;
 436         struct vvp_pgcache_id    id;
 437         int                   refcheck;
 438         int                   result;
 439
 440         env = cl_env_get(&refcheck);
 441         if (!IS_ERR(env)) {
 442                 pos = *(loff_t *) v;
 443                 vvp_pgcache_id_unpack(pos, &id);
 444                 sbi = f->private;
 445                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 446                 if (clob != NULL) {
 447                         hdr = cl_object_header(clob);
 448
 449                         spin_lock(&hdr->coh_page_guard);
 450                         page = cl_page_lookup(hdr, id.vpi_index);
 451                         spin_unlock(&hdr->coh_page_guard);
 452
 453                         seq_printf(f, "%8x@"DFID": ",
 454                                    id.vpi_index, PFID(&hdr->coh_lu.loh_fid));
 455                         if (page != NULL) {
 456                                 vvp_pgcache_page_show(env, f, page);
 457                                 cl_page_put(env, page);
 458                         } else
 459                                 seq_puts(f, "missing\n");
 460                         lu_object_ref_del(&clob->co_lu, "dump", current);
 461                         cl_object_put(env, clob);
 462                 } else
 463                         seq_printf(f, "%llx missing\n", pos);
 464                 cl_env_put(env, &refcheck);
 465                 result = 0;
 466         } else
 467                 result = PTR_ERR(env);
 468         return result;
 469 }
 470
 471 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 472 {
 473         struct ll_sb_info *sbi;
 474         struct lu_env     *env;
 475         int             refcheck;
 476
 477         sbi = f->private;
 478
 479         env = cl_env_get(&refcheck);
 480         if (!IS_ERR(env)) {
 481                 sbi = f->private;
 482                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 483                         pos = ERR_PTR(-EFBIG);
 484                 else {
 485                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 486                                                 *pos);
 487                         if (*pos == ~0ULL)
 488                                 pos = NULL;
 489                 }
 490                 cl_env_put(env, &refcheck);
 491         }
 492         return pos;
 493 }
 494
 495 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 496 {
 497         struct ll_sb_info *sbi;
 498         struct lu_env     *env;
 499         int             refcheck;
 500
 501         env = cl_env_get(&refcheck);
 502         if (!IS_ERR(env)) {
 503                 sbi = f->private;
 504                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 505                 if (*pos == ~0ULL)
 506                         pos = NULL;
 507                 cl_env_put(env, &refcheck);
 508         }
 509         return pos;
 510 }
 511
 512 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 513 {
 514         /* Nothing to do */
 515 }
 516
 517 static const struct seq_operations vvp_pgcache_ops = {
 518         .start = vvp_pgcache_start,
 519         .next  = vvp_pgcache_next,
 520         .stop  = vvp_pgcache_stop,
 521         .show  = vvp_pgcache_show
 522 };
 523
 524 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 525 {
 526         struct seq_file *seq;
 527         int rc;
 528
 529         rc = seq_open(filp, &vvp_pgcache_ops);
 530         if (rc)
 531                 return rc;
 532
 533         seq = filp->private_data;
 534         seq->private = inode->i_private;
 535
 536         return 0;
 537 }
 538
 539 const struct file_operations vvp_dump_pgcache_file_ops = {
 540         .owner   = THIS_MODULE,
 541         .open    = vvp_dump_pgcache_seq_open,
 542         .read    = seq_read,
 543         .llseek  = seq_lseek,
 544         .release = seq_release,
 545 };