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, 2013, 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 <lustre_lite.h>
  47 #include "llite_internal.h"
  48 #include "vvp_internal.h"
  49
  50 /*****************************************************************************
  51  *
  52  * Vvp device and device type functions.
  53  *
  54  */
  55
  56 /*
  57  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  58  * "llite_" (var. "ll_") prefix.
  59  */
  60
  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_thread_kmem,
  66                 .ckd_name  = "vvp_thread_kmem",
  67                 .ckd_size  = sizeof (struct vvp_thread_info),
  68         },
  69         {
  70                 .ckd_cache = &vvp_session_kmem,
  71                 .ckd_name  = "vvp_session_kmem",
  72                 .ckd_size  = sizeof (struct vvp_session)
  73         },
  74         {
  75                 .ckd_cache = NULL
  76         }
  77 };
  78
  79 static void *vvp_key_init(const struct lu_context *ctx,
  80                           struct lu_context_key *key)
  81 {
  82         struct vvp_thread_info *info;
  83
  84         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, GFP_NOFS);
  85         if (info == NULL)
  86                 info = ERR_PTR(-ENOMEM);
  87         return info;
  88 }
  89
  90 static void vvp_key_fini(const struct lu_context *ctx,
  91                          struct lu_context_key *key, void *data)
  92 {
  93         struct vvp_thread_info *info = data;
  94         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
  95 }
  96
  97 static void *vvp_session_key_init(const struct lu_context *ctx,
  98                                   struct lu_context_key *key)
  99 {
 100         struct vvp_session *session;
 101
 102         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
 103         if (session == NULL)
 104                 session = ERR_PTR(-ENOMEM);
 105         return session;
 106 }
 107
 108 static void vvp_session_key_fini(const struct lu_context *ctx,
 109                                  struct lu_context_key *key, void *data)
 110 {
 111         struct vvp_session *session = data;
 112         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 113 }
 114
 115
 116 struct lu_context_key vvp_key = {
 117         .lct_tags = LCT_CL_THREAD,
 118         .lct_init = vvp_key_init,
 119         .lct_fini = vvp_key_fini
 120 };
 121
 122 struct lu_context_key vvp_session_key = {
 123         .lct_tags = LCT_SESSION,
 124         .lct_init = vvp_session_key_init,
 125         .lct_fini = vvp_session_key_fini
 126 };
 127
 128 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 129 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
 130
 131 static const struct lu_device_operations vvp_lu_ops = {
 132         .ldo_object_alloc      = vvp_object_alloc
 133 };
 134
 135 static const struct cl_device_operations vvp_cl_ops = {
 136         .cdo_req_init = ccc_req_init
 137 };
 138
 139 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 140                                           struct lu_device_type *t,
 141                                           struct lustre_cfg *cfg)
 142 {
 143         return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
 144 }
 145
 146 static const struct lu_device_type_operations vvp_device_type_ops = {
 147         .ldto_init = vvp_type_init,
 148         .ldto_fini = vvp_type_fini,
 149
 150         .ldto_start = vvp_type_start,
 151         .ldto_stop  = vvp_type_stop,
 152
 153         .ldto_device_alloc = vvp_device_alloc,
 154         .ldto_device_free  = ccc_device_free,
 155         .ldto_device_init  = ccc_device_init,
 156         .ldto_device_fini  = ccc_device_fini
 157 };
 158
 159 struct lu_device_type vvp_device_type = {
 160         .ldt_tags     = LU_DEVICE_CL,
 161         .ldt_name     = LUSTRE_VVP_NAME,
 162         .ldt_ops      = &vvp_device_type_ops,
 163         .ldt_ctx_tags = LCT_CL_THREAD
 164 };
 165
 166 /**
 167  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 168  * pressure, when environments cannot be allocated.
 169  */
 170 int vvp_global_init(void)
 171 {
 172         int result;
 173
 174         result = lu_kmem_init(vvp_caches);
 175         if (result == 0) {
 176                 result = ccc_global_init(&vvp_device_type);
 177                 if (result != 0)
 178                         lu_kmem_fini(vvp_caches);
 179         }
 180         return result;
 181 }
 182
 183 void vvp_global_fini(void)
 184 {
 185         ccc_global_fini(&vvp_device_type);
 186         lu_kmem_fini(vvp_caches);
 187 }
 188
 189
 190 /*****************************************************************************
 191  *
 192  * mirror obd-devices into cl devices.
 193  *
 194  */
 195
 196 int cl_sb_init(struct super_block *sb)
 197 {
 198         struct ll_sb_info *sbi;
 199         struct cl_device  *cl;
 200         struct lu_env     *env;
 201         int rc = 0;
 202         int refcheck;
 203
 204         sbi  = ll_s2sbi(sb);
 205         env = cl_env_get(&refcheck);
 206         if (!IS_ERR(env)) {
 207                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 208                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 209                 if (!IS_ERR(cl)) {
 210                         cl2ccc_dev(cl)->cdv_sb = sb;
 211                         sbi->ll_cl = cl;
 212                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 213                 }
 214                 cl_env_put(env, &refcheck);
 215         } else
 216                 rc = PTR_ERR(env);
 217         RETURN(rc);
 218 }
 219
 220 int cl_sb_fini(struct super_block *sb)
 221 {
 222         struct ll_sb_info *sbi;
 223         struct lu_env     *env;
 224         struct cl_device  *cld;
 225         int                refcheck;
 226         int                result;
 227
 228         ENTRY;
 229         sbi = ll_s2sbi(sb);
 230         env = cl_env_get(&refcheck);
 231         if (!IS_ERR(env)) {
 232                 cld = sbi->ll_cl;
 233
 234                 if (cld != NULL) {
 235                         cl_stack_fini(env, cld);
 236                         sbi->ll_cl = NULL;
 237                         sbi->ll_site = NULL;
 238                 }
 239                 cl_env_put(env, &refcheck);
 240                 result = 0;
 241         } else {
 242                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 243                 result = PTR_ERR(env);
 244         }
 245
 246         RETURN(result);
 247 }
 248
 249 /****************************************************************************
 250  *
 251  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 252  *
 253  ****************************************************************************/
 254
 255 /*
 256  * To represent contents of a page cache as a byte stream, following
 257  * information if encoded in 64bit offset:
 258  *
 259  *       - file hash bucket in lu_site::ls_hash[]       28bits
 260  *
 261  *       - how far file is from bucket head              4bits
 262  *
 263  *       - page index                                   32bits
 264  *
 265  * First two data identify a file in the cache uniquely.
 266  */
 267
 268 #define PGC_OBJ_SHIFT (32 + 4)
 269 #define PGC_DEPTH_SHIFT (32)
 270
 271 struct vvp_pgcache_id {
 272         unsigned                 vpi_bucket;
 273         unsigned                 vpi_depth;
 274         uint32_t                 vpi_index;
 275
 276         unsigned                 vpi_curdep;
 277         struct lu_object_header *vpi_obj;
 278 };
 279
 280 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 281 {
 282         CLASSERT(sizeof(pos) == sizeof(__u64));
 283
 284         id->vpi_index  = pos & 0xffffffff;
 285         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 286         id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
 287 }
 288
 289 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 290 {
 291         return
 292                 ((__u64)id->vpi_index) |
 293                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 294                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 295 }
 296
 297 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 298                                struct hlist_node *hnode, void *data)
 299 {
 300         struct vvp_pgcache_id   *id  = data;
 301         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 302
 303         if (id->vpi_curdep-- > 0)
 304                 return 0; /* continue */
 305
 306         if (lu_object_is_dying(hdr))
 307                 return 1;
 308
 309         cfs_hash_get(hs, hnode);
 310         id->vpi_obj = hdr;
 311         return 1;
 312 }
 313
 314 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 315                                          struct lu_device *dev,
 316                                          struct vvp_pgcache_id *id)
 317 {
 318         LASSERT(lu_device_is_cl(dev));
 319
 320         id->vpi_depth &= 0xf;
 321         id->vpi_obj    = NULL;
 322         id->vpi_curdep = id->vpi_depth;
 323
 324         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 325                                 vvp_pgcache_obj_get, id);
 326         if (id->vpi_obj != NULL) {
 327                 struct lu_object *lu_obj;
 328
 329                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 330                 if (lu_obj != NULL) {
 331                         lu_object_ref_add(lu_obj, "dump", current);
 332                         return lu2cl(lu_obj);
 333                 }
 334                 lu_object_put(env, lu_object_top(id->vpi_obj));
 335
 336         } else if (id->vpi_curdep > 0) {
 337                 id->vpi_depth = 0xf;
 338         }
 339         return NULL;
 340 }
 341
 342 static loff_t vvp_pgcache_find(const struct lu_env *env,
 343                                struct lu_device *dev, loff_t pos)
 344 {
 345         struct cl_object     *clob;
 346         struct lu_site       *site;
 347         struct vvp_pgcache_id id;
 348
 349         site = dev->ld_site;
 350         vvp_pgcache_id_unpack(pos, &id);
 351
 352         while (1) {
 353                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 354                         return ~0ULL;
 355                 clob = vvp_pgcache_obj(env, dev, &id);
 356                 if (clob != NULL) {
 357                         struct inode *inode = ccc_object_inode(clob);
 358                         struct page *vmpage;
 359                         int nr;
 360
 361                         nr = find_get_pages_contig(inode->i_mapping,
 362                                                    id.vpi_index, 1, &vmpage);
 363                         if (nr > 0) {
 364                                 id.vpi_index = vmpage->index;
 365                                 /* Cant support over 16T file */
 366                                 nr = !(vmpage->index > 0xffffffff);
 367                                 page_cache_release(vmpage);
 368                         }
 369
 370                         lu_object_ref_del(&clob->co_lu, "dump", current);
 371                         cl_object_put(env, clob);
 372                         if (nr > 0)
 373                                 return vvp_pgcache_id_pack(&id);
 374                 }
 375                 /* to the next object. */
 376                 ++id.vpi_depth;
 377                 id.vpi_depth &= 0xf;
 378                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 379                         return ~0ULL;
 380                 id.vpi_index = 0;
 381         }
 382 }
 383
 384 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 385         if (test_bit(PG_##flag, &(page)->flags)) {                  \
 386                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 387                 has_flags = 1;                                          \
 388         }                                                               \
 389 } while(0)
 390
 391 static void vvp_pgcache_page_show(const struct lu_env *env,
 392                                   struct seq_file *seq, struct cl_page *page)
 393 {
 394         struct ccc_page *cpg;
 395         struct page      *vmpage;
 396         int              has_flags;
 397
 398         cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
 399         vmpage = cpg->cpg_page;
 400         seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
 401                    0 /* gen */,
 402                    cpg, page,
 403                    "none",
 404                    cpg->cpg_write_queued ? "wq" : "- ",
 405                    cpg->cpg_defer_uptodate ? "du" : "- ",
 406                    PageWriteback(vmpage) ? "wb" : "-",
 407                    vmpage,
 408                    PFID(ll_inode2fid(vmpage->mapping->host)),
 409                    vmpage->mapping->host, vmpage->index,
 410                    page_count(vmpage));
 411         has_flags = 0;
 412         seq_page_flag(seq, vmpage, locked, has_flags);
 413         seq_page_flag(seq, vmpage, error, has_flags);
 414         seq_page_flag(seq, vmpage, referenced, has_flags);
 415         seq_page_flag(seq, vmpage, uptodate, has_flags);
 416         seq_page_flag(seq, vmpage, dirty, has_flags);
 417         seq_page_flag(seq, vmpage, writeback, has_flags);
 418         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 419 }
 420
 421 static int vvp_pgcache_show(struct seq_file *f, void *v)
 422 {
 423         loff_t                   pos;
 424         struct ll_sb_info       *sbi;
 425         struct cl_object        *clob;
 426         struct lu_env           *env;
 427         struct vvp_pgcache_id    id;
 428         int                      refcheck;
 429         int                      result;
 430
 431         env = cl_env_get(&refcheck);
 432         if (!IS_ERR(env)) {
 433                 pos = *(loff_t *) v;
 434                 vvp_pgcache_id_unpack(pos, &id);
 435                 sbi = f->private;
 436                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 437                 if (clob != NULL) {
 438                         struct inode *inode = ccc_object_inode(clob);
 439                         struct cl_page *page = NULL;
 440                         struct page *vmpage;
 441
 442                         result = find_get_pages_contig(inode->i_mapping,
 443                                                       id.vpi_index, 1, &vmpage);
 444                         if (result > 0) {
 445                                 lock_page(vmpage);
 446                                 page = cl_vmpage_page(vmpage, clob);
 447                                 unlock_page(vmpage);
 448
 449                                 page_cache_release(vmpage);
 450                         }
 451
 452                         seq_printf(f, "%8x@"DFID": ", id.vpi_index,
 453                                    PFID(lu_object_fid(&clob->co_lu)));
 454                         if (page != NULL) {
 455                                 vvp_pgcache_page_show(env, f, page);
 456                                 cl_page_put(env, page);
 457                         } else
 458                                 seq_puts(f, "missing\n");
 459                         lu_object_ref_del(&clob->co_lu, "dump", current);
 460                         cl_object_put(env, clob);
 461                 } else
 462                         seq_printf(f, "%llx missing\n", pos);
 463                 cl_env_put(env, &refcheck);
 464                 result = 0;
 465         } else
 466                 result = PTR_ERR(env);
 467         return result;
 468 }
 469
 470 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 471 {
 472         struct ll_sb_info *sbi;
 473         struct lu_env     *env;
 474         int                refcheck;
 475
 476         sbi = f->private;
 477
 478         env = cl_env_get(&refcheck);
 479         if (!IS_ERR(env)) {
 480                 sbi = f->private;
 481                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 482                         pos = ERR_PTR(-EFBIG);
 483                 else {
 484                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 485                                                 *pos);
 486                         if (*pos == ~0ULL)
 487                                 pos = NULL;
 488                 }
 489                 cl_env_put(env, &refcheck);
 490         }
 491         return pos;
 492 }
 493
 494 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 495 {
 496         struct ll_sb_info *sbi;
 497         struct lu_env     *env;
 498         int                refcheck;
 499
 500         env = cl_env_get(&refcheck);
 501         if (!IS_ERR(env)) {
 502                 sbi = f->private;
 503                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 504                 if (*pos == ~0ULL)
 505                         pos = NULL;
 506                 cl_env_put(env, &refcheck);
 507         }
 508         return pos;
 509 }
 510
 511 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 512 {
 513         /* Nothing to do */
 514 }
 515
 516 static struct seq_operations vvp_pgcache_ops = {
 517         .start = vvp_pgcache_start,
 518         .next  = vvp_pgcache_next,
 519         .stop  = vvp_pgcache_stop,
 520         .show  = vvp_pgcache_show
 521 };
 522
 523 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 524 {
 525         struct ll_sb_info       *sbi = PDE_DATA(inode);
 526         struct seq_file         *seq;
 527         int                     result;
 528
 529         result = seq_open(filp, &vvp_pgcache_ops);
 530         if (result == 0) {
 531                 seq = filp->private_data;
 532                 seq->private = sbi;
 533         }
 534         return result;
 535 }
 536
 537 const struct file_operations vvp_dump_pgcache_file_ops = {
 538         .owner   = THIS_MODULE,
 539         .open    = vvp_dump_pgcache_seq_open,
 540         .read    = seq_read,
 541         .llseek  = seq_lseek,
 542         .release = seq_release,
 543 };