lustre/liblustre/rw.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Light Super operations
   5  *
   6  *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include <assert.h>
  29 #include <time.h>
  30 #include <sys/types.h>
  31 #include <sys/queue.h>
  32 #include <fcntl.h>
  33
  34 #include <sysio.h>
  35 #include <fs.h>
  36 #include <mount.h>
  37 #include <inode.h>
  38 #include <file.h>
  39
  40 #undef LIST_HEAD
  41
  42 #include "llite_lib.h"
  43
  44 static int llu_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock)
  45 {
  46         struct llu_inode_info *lli = llu_i2info(inode);
  47         struct lov_stripe_md *lsm = lli->lli_smd;
  48         struct obd_export *exp = llu_i2obdexp(inode);
  49         struct {
  50                 char name[16];
  51                 struct ldlm_lock *lock;
  52                 struct lov_stripe_md *lsm;
  53         } key = { .name = "lock_to_stripe", .lock = lock, .lsm = lsm };
  54         __u32 stripe, vallen = sizeof(stripe);
  55         int rc;
  56         ENTRY;
  57
  58         if (lsm->lsm_stripe_count == 1)
  59                 RETURN(0);
  60
  61         /* get our offset in the lov */
  62         rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe);
  63         if (rc != 0) {
  64                 CERROR("obd_get_info: rc = %d\n", rc);
  65                 LBUG();
  66         }
  67         LASSERT(stripe < lsm->lsm_stripe_count);
  68         RETURN(stripe);
  69 }
  70
  71 static int llu_extent_lock_callback(struct ldlm_lock *lock,
  72                                     struct ldlm_lock_desc *new, void *data,
  73                                     int flag)
  74 {
  75         struct lustre_handle lockh = { 0 };
  76         int rc;
  77         ENTRY;
  78
  79
  80         if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) {
  81                 LDLM_ERROR(lock, "cancelling lock with bad data %p", data);
  82                 LBUG();
  83         }
  84
  85         switch (flag) {
  86         case LDLM_CB_BLOCKING:
  87                 ldlm_lock2handle(lock, &lockh);
  88                 rc = ldlm_cli_cancel(&lockh);
  89                 if (rc != ELDLM_OK)
  90                         CERROR("ldlm_cli_cancel failed: %d\n", rc);
  91                 break;
  92         case LDLM_CB_CANCELING: {
  93                 struct inode *inode;
  94                 struct llu_inode_info *lli;
  95                 struct lov_stripe_md *lsm;
  96                 __u32 stripe;
  97                 __u64 kms;
  98
  99                 /* This lock wasn't granted, don't try to evict pages */
 100                 if (lock->l_req_mode != lock->l_granted_mode)
 101                         RETURN(0);
 102
 103                 inode = llu_inode_from_lock(lock);
 104                 if (!inode)
 105                         RETURN(0);
 106                 lli= llu_i2info(inode);
 107                 if (!lli)
 108                         goto iput;
 109                 if (!lli->lli_smd)
 110                         goto iput;
 111                 lsm = lli->lli_smd;
 112
 113                 stripe = llu_lock_to_stripe_offset(inode, lock);
 114                 kms = ldlm_extent_shift_kms(lock,
 115                                             lsm->lsm_oinfo[stripe].loi_kms);
 116                 if (lsm->lsm_oinfo[stripe].loi_kms != kms)
 117                         LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
 118                                    lsm->lsm_oinfo[stripe].loi_kms, kms);
 119                 lsm->lsm_oinfo[stripe].loi_kms = kms;
 120 iput:
 121                 I_RELE(inode);
 122                 break;
 123         }
 124         default:
 125                 LBUG();
 126         }
 127
 128         RETURN(0);
 129 }
 130
 131 static int llu_glimpse_callback(struct ldlm_lock *lock, void *reqp)
 132 {
 133         struct ptlrpc_request *req = reqp;
 134         struct inode *inode = llu_inode_from_lock(lock);
 135         struct llu_inode_info *lli;
 136         struct ost_lvb *lvb;
 137         int rc, size = sizeof(*lvb), stripe = 0;
 138         ENTRY;
 139
 140         if (inode == NULL)
 141                 GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
 142         lli = llu_i2info(inode);
 143         if (lli == NULL)
 144                 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
 145         if (lli->lli_smd == NULL)
 146                 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
 147
 148         /* First, find out which stripe index this lock corresponds to. */
 149         if (lli->lli_smd->lsm_stripe_count > 1)
 150                 stripe = llu_lock_to_stripe_offset(inode, lock);
 151
 152         rc = lustre_pack_reply(req, 1, &size, NULL);
 153         if (rc) {
 154                 CERROR("lustre_pack_reply: %d\n", rc);
 155                 GOTO(iput, rc);
 156         }
 157
 158         lvb = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*lvb));
 159         lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe].loi_kms;
 160
 161         LDLM_DEBUG(lock, "i_size: %llu -> stripe number %u -> kms "LPU64,
 162                    lli->lli_st_size, stripe, lvb->lvb_size);
 163  iput:
 164         I_RELE(inode);
 165  out:
 166         /* These errors are normal races, so we don't want to fill the console
 167          * with messages by calling ptlrpc_error() */
 168         if (rc == -ELDLM_NO_LOCK_DATA)
 169                 lustre_pack_reply(req, 0, NULL, NULL);
 170
 171         req->rq_status = rc;
 172         return rc;
 173 }
 174
 175 __u64 lov_merge_size(struct lov_stripe_md *lsm, int kms);
 176 __u64 lov_merge_mtime(struct lov_stripe_md *lsm, __u64 current_time);
 177
 178 /* NB: lov_merge_size will prefer locally cached writes if they extend the
 179  * file (because it prefers KMS over RSS when larger) */
 180 int llu_glimpse_size(struct inode *inode, struct ost_lvb *lvb)
 181 {
 182         struct llu_inode_info *lli = llu_i2info(inode);
 183         struct llu_sb_info *sbi = llu_i2sbi(inode);
 184         ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
 185         struct lustre_handle lockh;
 186         int rc, flags = LDLM_FL_HAS_INTENT;
 187         ENTRY;
 188
 189         CDEBUG(D_DLMTRACE, "Glimpsing inode %lu\n", lli->lli_st_ino);
 190
 191         rc = obd_enqueue(sbi->ll_osc_exp, lli->lli_smd, LDLM_EXTENT, &policy,
 192                          LCK_PR, &flags, llu_extent_lock_callback,
 193                          ldlm_completion_ast, llu_glimpse_callback, inode,
 194                          sizeof(*lvb), lustre_swab_ost_lvb, &lockh);
 195         if (rc > 0)
 196                 RETURN(-EIO);
 197
 198         lvb->lvb_size = lov_merge_size(lli->lli_smd, 0);
 199         //inode->i_mtime = lov_merge_mtime(lli->lli_smd, inode->i_mtime);
 200
 201         CDEBUG(D_DLMTRACE, "glimpse: size: "LPU64"\n", lvb->lvb_size);
 202
 203         obd_cancel(sbi->ll_osc_exp, lli->lli_smd, LCK_PR, &lockh);
 204
 205         RETURN(rc);
 206 }
 207
 208 int llu_extent_lock(struct ll_file_data *fd, struct inode *inode,
 209                     struct lov_stripe_md *lsm, int mode,
 210                     ldlm_policy_data_t *policy, struct lustre_handle *lockh,
 211                     int ast_flags)
 212 {
 213         struct llu_sb_info *sbi = llu_i2sbi(inode);
 214         struct llu_inode_info *lli = llu_i2info(inode);
 215         int rc;
 216         ENTRY;
 217
 218         LASSERT(lockh->cookie == 0);
 219
 220         /* XXX phil: can we do this?  won't it screw the file size up? */
 221         if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
 222             (sbi->ll_flags & LL_SBI_NOLCK))
 223                 RETURN(0);
 224
 225         CDEBUG(D_DLMTRACE, "Locking inode %lu, start "LPU64" end "LPU64"\n",
 226                lli->lli_st_ino, policy->l_extent.start, policy->l_extent.end);
 227
 228         rc = obd_enqueue(sbi->ll_osc_exp, lsm, LDLM_EXTENT, policy, mode,
 229                          &ast_flags, llu_extent_lock_callback,
 230                          ldlm_completion_ast, llu_glimpse_callback, inode,
 231                          sizeof(struct ost_lvb), lustre_swab_ost_lvb, lockh);
 232         if (rc > 0)
 233                 rc = -EIO;
 234
 235         if (policy->l_extent.start == 0 &&
 236             policy->l_extent.end == OBD_OBJECT_EOF)
 237                 lli->lli_st_size = lov_merge_size(lsm, 1);
 238
 239         //inode->i_mtime = lov_merge_mtime(lsm, inode->i_mtime);
 240
 241         RETURN(rc);
 242 }
 243
 244 #if 0
 245 int llu_extent_lock_no_validate(struct ll_file_data *fd,
 246                                 struct inode *inode,
 247                                 struct lov_stripe_md *lsm,
 248                                 int mode,
 249                                 struct ldlm_extent *extent,
 250                                 struct lustre_handle *lockh,
 251                                 int ast_flags)
 252 {
 253         struct llu_sb_info *sbi = llu_i2sbi(inode);
 254         struct llu_inode_info *lli = llu_i2info(inode);
 255         int rc;
 256         ENTRY;
 257
 258         LASSERT(lockh->cookie == 0);
 259
 260         /* XXX phil: can we do this?  won't it screw the file size up? */
 261         if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
 262             (sbi->ll_flags & LL_SBI_NOLCK))
 263                 RETURN(0);
 264
 265         CDEBUG(D_DLMTRACE, "Locking inode %lu, start "LPU64" end "LPU64"\n",
 266                lli->lli_st_ino, extent->start, extent->end);
 267
 268         rc = obd_enqueue(sbi->ll_osc_exp, lsm, NULL, LDLM_EXTENT, extent,
 269                          sizeof(extent), mode, &ast_flags,
 270                          llu_extent_lock_callback, inode, lockh);
 271
 272         RETURN(rc);
 273 }
 274
 275 /*
 276  * this grabs a lock and manually implements behaviour that makes it look like
 277  * the OST is returning the file size with each lock acquisition.
 278  */
 279 int llu_extent_lock(struct ll_file_data *fd, struct inode *inode,
 280                     struct lov_stripe_md *lsm, int mode,
 281                     struct ldlm_extent *extent, struct lustre_handle *lockh,
 282                     int nonblock)
 283 {
 284         struct llu_inode_info *lli = llu_i2info(inode);
 285         struct obd_export *exp = llu_i2obdexp(inode);
 286         struct ldlm_extent size_lock;
 287         struct lustre_handle match_lockh = {0};
 288         int flags, rc, matched;
 289         int astflags = nonblock ? LDLM_FL_BLOCK_NOWAIT : 0;
 290         ENTRY;
 291
 292         rc = llu_extent_lock_no_validate(fd, inode, lsm, mode, extent,
 293                                          lockh, astflags);
 294         if (rc != ELDLM_OK)
 295                 RETURN(rc);
 296
 297         if (test_bit(LLI_F_HAVE_OST_SIZE_LOCK, &lli->lli_flags))
 298                 RETURN(0);
 299
 300         rc = llu_inode_getattr(inode, lsm);
 301         if (rc) {
 302                 llu_extent_unlock(fd, inode, lsm, mode, lockh);
 303                 RETURN(rc);
 304         }
 305
 306         size_lock.start = lli->lli_st_size;
 307         size_lock.end = OBD_OBJECT_EOF;
 308
 309         /* XXX I bet we should be checking the lock ignore flags.. */
 310         /* FIXME use LDLM_FL_TEST_LOCK instead */
 311         flags = LDLM_FL_CBPENDING | LDLM_FL_BLOCK_GRANTED;
 312         matched = obd_match(exp, lsm, LDLM_EXTENT, &size_lock,
 313                             sizeof(size_lock), LCK_PR, &flags, inode,
 314                             &match_lockh);
 315
 316         /* hey, alright, we hold a size lock that covers the size we
 317          * just found, its not going to change for a while.. */
 318         if (matched == 1) {
 319                 set_bit(LLI_F_HAVE_OST_SIZE_LOCK, &lli->lli_flags);
 320                 obd_cancel(exp, lsm, LCK_PR, &match_lockh);
 321         }
 322
 323         RETURN(0);
 324 }
 325 #endif
 326
 327 int llu_extent_unlock(struct ll_file_data *fd, struct inode *inode,
 328                 struct lov_stripe_md *lsm, int mode,
 329                 struct lustre_handle *lockh)
 330 {
 331         struct llu_sb_info *sbi = llu_i2sbi(inode);
 332         int rc;
 333         ENTRY;
 334
 335         /* XXX phil: can we do this?  won't it screw the file size up? */
 336         if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
 337             (sbi->ll_flags & LL_SBI_NOLCK))
 338                 RETURN(0);
 339
 340         rc = obd_cancel(sbi->ll_osc_exp, lsm, mode, lockh);
 341
 342         RETURN(rc);
 343 }
 344
 345 #define LLAP_MAGIC 12346789
 346
 347 struct ll_async_page {
 348         int             llap_magic;
 349         void           *llap_cookie;
 350         int             llap_queued;
 351         struct page    *llap_page;
 352         struct inode   *llap_inode;
 353 };
 354
 355 static struct ll_async_page *llap_from_cookie(void *cookie)
 356 {
 357         struct ll_async_page *llap = cookie;
 358         if (llap->llap_magic != LLAP_MAGIC)
 359                 return ERR_PTR(-EINVAL);
 360         return llap;
 361 };
 362
 363 static void llu_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
 364 {
 365         struct ll_async_page *llap;
 366         struct inode *inode;
 367         struct lov_stripe_md *lsm;
 368         obd_flag valid_flags;
 369         ENTRY;
 370
 371         llap = llap_from_cookie(data);
 372         if (IS_ERR(llap)) {
 373                 EXIT;
 374                 return;
 375         }
 376
 377         inode = llap->llap_inode;
 378         lsm = llu_i2info(inode)->lli_smd;
 379
 380         oa->o_id = lsm->lsm_object_id;
 381         oa->o_valid = OBD_MD_FLID;
 382         valid_flags = OBD_MD_FLTYPE | OBD_MD_FLATIME;
 383         if (cmd == OBD_BRW_WRITE)
 384                 valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME;
 385
 386         obdo_from_inode(oa, inode, valid_flags);
 387         EXIT;
 388 }
 389
 390 /* called for each page in a completed rpc.*/
 391 static void llu_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
 392 {
 393         struct ll_async_page *llap;
 394         struct page *page;
 395
 396         llap = llap_from_cookie(data);
 397         if (IS_ERR(llap)) {
 398                 EXIT;
 399                 return;
 400         }
 401
 402         llap->llap_queued = 0;
 403         page = llap->llap_page;
 404
 405         if (rc != 0) {
 406                 if (cmd == OBD_BRW_WRITE)
 407                         CERROR("writeback error on page %p index %ld: %d\n",
 408                                page, page->index, rc);
 409         }
 410         EXIT;
 411 }
 412
 413 static struct obd_async_page_ops llu_async_page_ops = {
 414         .ap_make_ready =        NULL,
 415         .ap_refresh_count =     NULL,
 416         .ap_fill_obdo =         llu_ap_fill_obdo,
 417         .ap_completion =        llu_ap_completion,
 418 };
 419
 420 static
 421 struct llu_sysio_cookie* get_sysio_cookie(struct inode *inode, int maxpages)
 422 {
 423         struct llu_sysio_cookie *cookie;
 424         int rc;
 425
 426         OBD_ALLOC(cookie, LLU_SYSIO_COOKIE_SIZE(maxpages));
 427         if (cookie == NULL)
 428                 goto out;
 429
 430         I_REF(inode);
 431         cookie->lsc_inode = inode;
 432         cookie->lsc_maxpages = maxpages;
 433         cookie->lsc_llap = (struct ll_async_page *)(cookie + 1);
 434         cookie->lsc_pages = (struct page *) (cookie->lsc_llap + maxpages);
 435
 436         rc = oig_init(&cookie->lsc_oig);
 437         if (rc) {
 438                 OBD_FREE(cookie, LLU_SYSIO_COOKIE_SIZE(maxpages));
 439                 cookie = NULL;
 440         }
 441
 442 out:
 443         return cookie;
 444 }
 445
 446 static
 447 void put_sysio_cookie(struct llu_sysio_cookie *cookie)
 448 {
 449         struct lov_stripe_md *lsm = llu_i2info(cookie->lsc_inode)->lli_smd;
 450         struct obd_export *exp = llu_i2obdexp(cookie->lsc_inode);
 451         struct ll_async_page *llap = cookie->lsc_llap;
 452 #ifdef LIBLUSTRE_HANDLE_UNALIGNED_PAGE
 453         struct page *pages = cookie->lsc_pages;
 454 #endif
 455         int i;
 456
 457         for (i = 0; i< cookie->lsc_maxpages; i++) {
 458                 if (llap[i].llap_cookie)
 459                         obd_teardown_async_page(exp, lsm, NULL,
 460                                                 llap[i].llap_cookie);
 461 #ifdef LIBLUSTRE_HANDLE_UNALIGNED_PAGE
 462                 if (pages[i]._managed) {
 463                         free(pages[i].addr);
 464                         pages[i]._managed = 0;
 465                 }
 466 #endif
 467         }
 468
 469         I_RELE(cookie->lsc_inode);
 470
 471         oig_release(cookie->lsc_oig);
 472         OBD_FREE(cookie, LLU_SYSIO_COOKIE_SIZE(cookie->lsc_maxpages));
 473 }
 474
 475 #ifdef LIBLUSTRE_HANDLE_UNALIGNED_PAGE
 476 /* Note: these code should be removed finally, don't need
 477  * more cleanup
 478  */
 479 static
 480 int prepare_unaligned_write(struct llu_sysio_cookie *cookie)
 481 {
 482         struct inode *inode = cookie->lsc_inode;
 483         struct llu_inode_info *lli = llu_i2info(inode);
 484         struct lov_stripe_md *lsm = lli->lli_smd;
 485         struct obdo oa;
 486         struct page *pages = cookie->lsc_pages;
 487         int i, pgidx[2] = {0, cookie->lsc_npages-1};
 488         int rc;
 489         ENTRY;
 490
 491         for (i = 0; i < 2; i++) {
 492                 struct page *oldpage = &pages[pgidx[i]];
 493                 struct page newpage;
 494                 struct brw_page pg;
 495                 char *newbuf;
 496
 497                 if (i == 0 && pgidx[0] == pgidx[1])
 498                         continue;
 499
 500                 LASSERT(oldpage->_offset + oldpage->_count <= PAGE_CACHE_SIZE);
 501
 502                 if (oldpage->_count == PAGE_CACHE_SIZE)
 503                         continue;
 504
 505                 if (oldpage->index << PAGE_CACHE_SHIFT >=
 506                     lli->lli_st_size)
 507                         continue;
 508
 509                 newbuf = malloc(PAGE_CACHE_SIZE);
 510                 if (!newbuf)
 511                         return -ENOMEM;
 512
 513                 newpage.index = oldpage->index;
 514                 newpage.addr = newbuf;
 515
 516                 pg.pg = &newpage;
 517                 pg.off = ((obd_off)newpage.index << PAGE_CACHE_SHIFT);
 518                 if (pg.off + PAGE_CACHE_SIZE > lli->lli_st_size)
 519                         pg.count = lli->lli_st_size % PAGE_CACHE_SIZE;
 520                 else
 521                         pg.count = PAGE_CACHE_SIZE;
 522                 pg.flag = 0;
 523
 524                 oa.o_id = lsm->lsm_object_id;
 525                 oa.o_mode = lli->lli_st_mode;
 526                 oa.o_valid = OBD_MD_FLID | OBD_MD_FLMODE | OBD_MD_FLTYPE;
 527
 528                 /* issue read */
 529                 rc = obd_brw(OBD_BRW_READ, llu_i2obdexp(inode), &oa, lsm, 1, &pg, NULL);
 530                 if (rc) {
 531                         free(newbuf);
 532                         RETURN(rc);
 533                 }
 534
 535                 /* copy page content, and reset page params */
 536                 memcpy(newbuf + oldpage->_offset,
 537                        (char*)oldpage->addr + oldpage->_offset,
 538                        oldpage->_count);
 539
 540                 oldpage->addr = newbuf;
 541                 if ((((obd_off)oldpage->index << PAGE_CACHE_SHIFT) +
 542                     oldpage->_offset + oldpage->_count) > lli->lli_st_size)
 543                         oldpage->_count += oldpage->_offset;
 544                 else
 545                         oldpage->_count = PAGE_CACHE_SIZE;
 546                 oldpage->_offset = 0;
 547                 oldpage->_managed = 1;
 548         }
 549
 550         RETURN(0);
 551 }
 552 #endif
 553
 554 static
 555 int llu_prep_async_io(struct llu_sysio_cookie *cookie, int cmd,
 556                       char *buf, loff_t pos, size_t count)
 557 {
 558         struct llu_inode_info *lli = llu_i2info(cookie->lsc_inode);
 559         struct lov_stripe_md *lsm = lli->lli_smd;
 560         struct obd_export *exp = llu_i2obdexp(cookie->lsc_inode);
 561         struct page *pages = cookie->lsc_pages;
 562         struct ll_async_page *llap = cookie->lsc_llap;
 563         int i, rc, npages = 0;
 564         ENTRY;
 565
 566         if (!exp)
 567                 RETURN(-EINVAL);
 568
 569         /* prepare the pages array */
 570         do {
 571                 unsigned long index, offset, bytes;
 572
 573                 offset = (pos & ~PAGE_CACHE_MASK);
 574                 index = pos >> PAGE_CACHE_SHIFT;
 575                 bytes = PAGE_CACHE_SIZE - offset;
 576                 if (bytes > count)
 577                         bytes = count;
 578
 579                 /* prevent read beyond file range */
 580                 if ((cmd == OBD_BRW_READ) &&
 581                     (pos + bytes) >= lli->lli_st_size) {
 582                         if (pos >= lli->lli_st_size)
 583                                 break;
 584                         bytes = lli->lli_st_size - pos;
 585                 }
 586
 587                 /* prepare page for this index */
 588                 pages[npages].index = index;
 589                 pages[npages].addr = buf - offset;
 590
 591                 pages[npages]._offset = offset;
 592                 pages[npages]._count = bytes;
 593
 594                 npages++;
 595                 count -= bytes;
 596                 pos += bytes;
 597                 buf += bytes;
 598
 599                 cookie->lsc_rwcount += bytes;
 600         } while (count);
 601
 602         cookie->lsc_npages = npages;
 603
 604 #ifdef LIBLUSTRE_HANDLE_UNALIGNED_PAGE
 605         if (cmd == OBD_BRW_WRITE) {
 606                 rc = prepare_unaligned_write(cookie);
 607                 if (rc)
 608                         RETURN(rc);
 609         }
 610 #endif
 611
 612         for (i = 0; i < npages; i++) {
 613                 llap[i].llap_magic = LLAP_MAGIC;
 614                 rc = obd_prep_async_page(exp, lsm, NULL, &pages[i],
 615                                          (obd_off)pages[i].index << PAGE_SHIFT,
 616                                          &llu_async_page_ops,
 617                                          &llap[i], &llap[i].llap_cookie);
 618                 if (rc) {
 619                         llap[i].llap_cookie = NULL;
 620                         RETURN(rc);
 621                 }
 622                 CDEBUG(D_CACHE, "llap %p page %p cookie %p obj off "LPU64"\n",
 623                        &llap[i], &pages[i], llap[i].llap_cookie,
 624                        (obd_off)pages[i].index << PAGE_SHIFT);
 625                 pages[i].private = (unsigned long)&llap[i];
 626                 llap[i].llap_page = &pages[i];
 627                 llap[i].llap_inode = cookie->lsc_inode;
 628
 629                 rc = obd_queue_group_io(exp, lsm, NULL, cookie->lsc_oig,
 630                                         llap[i].llap_cookie, cmd,
 631                                         pages[i]._offset, pages[i]._count, 0,
 632                                         ASYNC_READY | ASYNC_URGENT |
 633                                         ASYNC_COUNT_STABLE | ASYNC_GROUP_SYNC);
 634                 if (rc)
 635                         RETURN(rc);
 636
 637                 llap[i].llap_queued = 1;
 638         }
 639
 640         RETURN(0);
 641 }
 642
 643 static
 644 int llu_start_async_io(struct llu_sysio_cookie *cookie)
 645 {
 646         struct lov_stripe_md *lsm = llu_i2info(cookie->lsc_inode)->lli_smd;
 647         struct obd_export *exp = llu_i2obdexp(cookie->lsc_inode);
 648
 649         return obd_trigger_group_io(exp, lsm, NULL, cookie->lsc_oig);
 650 }
 651
 652 /*
 653  * read/write a continuous buffer for an inode (zero-copy)
 654  */
 655 struct llu_sysio_cookie*
 656 llu_rw(int cmd, struct inode *inode, char *buf, size_t count, loff_t pos)
 657 {
 658         struct llu_sysio_cookie *cookie;
 659         int max_pages, rc;
 660         ENTRY;
 661
 662         max_pages = (count >> PAGE_SHIFT) + 2;
 663
 664         cookie = get_sysio_cookie(inode, max_pages);
 665         if (!cookie)
 666                 RETURN(ERR_PTR(-ENOMEM));
 667
 668         rc = llu_prep_async_io(cookie, cmd, buf, pos, count);
 669         if (rc)
 670                 GOTO(out_cleanup, rc);
 671
 672         rc = llu_start_async_io(cookie);
 673         if (rc)
 674                 GOTO(out_cleanup, rc);
 675
 676 /*
 677         rc = oig_wait(&oig);
 678         if (rc) {
 679                 CERROR("file i/o error!\n");
 680                 rw_count = rc;
 681         }
 682 */
 683         RETURN(cookie);
 684
 685 out_cleanup:
 686         put_sysio_cookie(cookie);
 687         RETURN(ERR_PTR(rc));
 688 }
 689
 690 void lov_increase_kms(struct obd_export *exp, struct lov_stripe_md *lsm,
 691                       obd_off size);
 692
 693 struct llu_sysio_callback_args*
 694 llu_file_write(struct inode *inode, const struct iovec *iovec,
 695                size_t iovlen, loff_t pos)
 696 {
 697         struct llu_inode_info *lli = llu_i2info(inode);
 698         struct ll_file_data *fd = lli->lli_file_data;
 699         struct lustre_handle lockh = {0};
 700         struct lov_stripe_md *lsm = lli->lli_smd;
 701         struct obd_export *exp = NULL;
 702         ldlm_policy_data_t policy;
 703         struct llu_sysio_callback_args *lsca;
 704         struct llu_sysio_cookie *cookie;
 705         int astflag = (lli->lli_open_flags & O_NONBLOCK) ?
 706                        LDLM_FL_BLOCK_NOWAIT : 0;
 707         ldlm_error_t err;
 708         int iovidx;
 709         ENTRY;
 710
 711         /* XXX consider other types later */
 712         if (!S_ISREG(lli->lli_st_mode))
 713                 LBUG();
 714
 715         LASSERT(iovlen <= MAX_IOVEC);
 716
 717         exp = llu_i2obdexp(inode);
 718         if (exp == NULL)
 719                 RETURN(ERR_PTR(-EINVAL));
 720
 721         OBD_ALLOC(lsca, sizeof(*lsca));
 722         if (!lsca)
 723                 RETURN(ERR_PTR(-ENOMEM));
 724
 725         /* FIXME optimize the following extent locking */
 726         for (iovidx = 0; iovidx < iovlen; iovidx++) {
 727                 char *buf = (char*)iovec[iovidx].iov_base;
 728                 size_t count = iovec[iovidx].iov_len;
 729
 730                 if (count == 0)
 731                         continue;
 732
 733                 if (pos + count > lli->lli_maxbytes)
 734                         GOTO(err_out, err = -ERANGE);
 735
 736                 /* FIXME libsysio haven't handle O_APPEND?? */
 737                 policy.l_extent.start = pos;
 738                 policy.l_extent.end = pos + count - 1;
 739
 740                 err = llu_extent_lock(fd, inode, lsm, LCK_PW, &policy,
 741                                       &lockh, astflag);
 742                 if (err != ELDLM_OK)
 743                         GOTO(err_out, err = -ENOLCK);
 744
 745                 CDEBUG(D_INFO, "Writing inode %lu, "LPSZ" bytes, offset %Lu\n",
 746                        lli->lli_st_ino, count, pos);
 747
 748                 cookie = llu_rw(OBD_BRW_WRITE, inode, buf, count, pos);
 749                 if (!IS_ERR(cookie)) {
 750                         /* save cookie */
 751                         lsca->cookies[lsca->ncookies++] = cookie;
 752                         pos += count;
 753                         lov_increase_kms(exp, lsm, pos);
 754                         /* file size grow */
 755                         if (pos > lli->lli_st_size)
 756                                 lli->lli_st_size = pos;
 757                 } else {
 758                         llu_extent_unlock(fd, inode, lsm, LCK_PW, &lockh);
 759                         GOTO(err_out, err = PTR_ERR(cookie));
 760                 }
 761
 762                 /* XXX errors? */
 763                 err = llu_extent_unlock(fd, inode, lsm, LCK_PW, &lockh);
 764                 if (err)
 765                         CERROR("extent unlock error %d\n", err);
 766         }
 767
 768         RETURN(lsca);
 769
 770 err_out:
 771         /* teardown all async stuff */
 772         while (lsca->ncookies--) {
 773                 put_sysio_cookie(lsca->cookies[lsca->ncookies]);
 774         }
 775         OBD_FREE(lsca, sizeof(*lsca));
 776
 777         RETURN(ERR_PTR(err));
 778 }
 779
 780 #if 0
 781 static void llu_update_atime(struct inode *inode)
 782 {
 783         struct llu_inode_info *lli = llu_i2info(inode);
 784
 785 #ifdef USE_ATIME
 786         struct iattr attr;
 787
 788         attr.ia_atime = LTIME_S(CURRENT_TIME);
 789         attr.ia_valid = ATTR_ATIME;
 790
 791         if (lli->lli_st_atime == attr.ia_atime) return;
 792         if (IS_RDONLY(inode)) return;
 793         if (IS_NOATIME(inode)) return;
 794
 795         /* ll_inode_setattr() sets inode->i_atime from attr.ia_atime */
 796         llu_inode_setattr(inode, &attr, 0);
 797 #else
 798         /* update atime, but don't explicitly write it out just this change */
 799         inode->i_atime = CURRENT_TIME;
 800 #endif
 801 }
 802 #endif
 803
 804 struct llu_sysio_callback_args*
 805 llu_file_read(struct inode *inode, const struct iovec *iovec,
 806               size_t iovlen, loff_t pos)
 807 {
 808         struct llu_inode_info *lli = llu_i2info(inode);
 809         struct ll_file_data *fd = lli->lli_file_data;
 810         struct lov_stripe_md *lsm = lli->lli_smd;
 811         struct lustre_handle lockh = { 0 };
 812         ldlm_policy_data_t policy;
 813         struct llu_sysio_callback_args *lsca;
 814         struct llu_sysio_cookie *cookie;
 815         int astflag = (lli->lli_open_flags & O_NONBLOCK) ?
 816                        LDLM_FL_BLOCK_NOWAIT : 0;
 817         __u64 kms;
 818         int iovidx;
 819
 820         ldlm_error_t err;
 821         ENTRY;
 822
 823         OBD_ALLOC(lsca, sizeof(*lsca));
 824         if (!lsca)
 825                 RETURN(ERR_PTR(-ENOMEM));
 826
 827         for (iovidx = 0; iovidx < iovlen; iovidx++) {
 828                 char *buf = iovec[iovidx].iov_base;
 829                 size_t count = iovec[iovidx].iov_len;
 830
 831                 /* "If nbyte is 0, read() will return 0 and have no other results."
 832                  *                      -- Single Unix Spec */
 833                 if (count == 0)
 834                         continue;
 835
 836                 policy.l_extent.start = pos;
 837                 policy.l_extent.end = pos + count - 1;
 838
 839                 err = llu_extent_lock(fd, inode, lsm, LCK_PR, &policy,
 840                                       &lockh, astflag);
 841                 if (err != ELDLM_OK)
 842                         GOTO(err_out, err = -ENOLCK);
 843
 844                 kms = lov_merge_size(lsm, 1);
 845                 if (policy.l_extent.end > kms) {
 846                         /* A glimpse is necessary to determine whether we
 847                          * return a short read or some zeroes at the end of
 848                          * the buffer */
 849                         struct ost_lvb lvb;
 850                         if (llu_glimpse_size(inode, &lvb)) {
 851                                 llu_extent_unlock(fd, inode, lsm, LCK_PR, &lockh);
 852                                 GOTO(err_out, err = -ENOLCK);
 853                         }
 854                         lli->lli_st_size = lvb.lvb_size;
 855                 } else {
 856                         lli->lli_st_size = kms;
 857                 }
 858
 859                 CDEBUG(D_INFO, "Reading inode %lu, "LPSZ" bytes, offset %Ld, "
 860                        "i_size "LPU64"\n", lli->lli_st_ino, count, pos,
 861                        lli->lli_st_size);
 862
 863                 if (pos >= lli->lli_st_size) {
 864                         llu_extent_unlock(fd, inode, lsm, LCK_PR, &lockh);
 865                         break;
 866                 }
 867
 868                 cookie = llu_rw(OBD_BRW_READ, inode, buf, count, pos);
 869                 if (!IS_ERR(cookie)) {
 870                         /* save cookie */
 871                         lsca->cookies[lsca->ncookies++] = cookie;
 872                         pos += count;
 873                 } else {
 874                         llu_extent_unlock(fd, inode, lsm, LCK_PR, &lockh);
 875                         GOTO(err_out, err = PTR_ERR(cookie));
 876                 }
 877
 878                 /* XXX errors? */
 879                 err = llu_extent_unlock(fd, inode, lsm, LCK_PR, &lockh);
 880                 if (err)
 881                         CERROR("extent_unlock fail: %d\n", err);
 882         }
 883 #if 0
 884         if (readed > 0)
 885                 llu_update_atime(inode);
 886 #endif
 887         RETURN(lsca);
 888
 889 err_out:
 890         /* teardown all async stuff */
 891         while (lsca->ncookies--) {
 892                 put_sysio_cookie(lsca->cookies[lsca->ncookies]);
 893         }
 894         OBD_FREE(lsca, sizeof(*lsca));
 895
 896         RETURN(ERR_PTR(err));
 897 }
 898
 899 int llu_iop_iodone(struct ioctx *ioctxp)
 900 {
 901         struct llu_sysio_callback_args *lsca = ioctxp->ioctx_private;
 902         struct llu_sysio_cookie *cookie;
 903         int i, err = 0, rc = 0;
 904         ENTRY;
 905
 906         /* write/read(fd, buf, 0) */
 907         if (!lsca) {
 908                 ioctxp->ioctx_cc = 0;
 909                 RETURN(1);
 910         }
 911
 912         LASSERT(!IS_ERR(lsca));
 913
 914         for (i = 0; i < lsca->ncookies; i++) {
 915                 cookie = lsca->cookies[i];
 916                 if (cookie) {
 917                         err = oig_wait(cookie->lsc_oig);
 918                         if (err && !rc)
 919                                 rc = err;
 920                         if (!rc)
 921                                 ioctxp->ioctx_cc += cookie->lsc_rwcount;
 922                         put_sysio_cookie(cookie);
 923                 }
 924         }
 925
 926         if (rc) {
 927                 LASSERT(rc < 0);
 928                 ioctxp->ioctx_cc = -1;
 929                 ioctxp->ioctx_errno = -rc;
 930         }
 931
 932         OBD_FREE(lsca, sizeof(*lsca));
 933         ioctxp->ioctx_private = NULL;
 934
 935         RETURN(1);
 936 }