lustre/llite/dir.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/dir.c
  12  *  linux/fs/ext2/dir.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  ext2 directory handling functions
  17  *
  18  *  Big-endian to little-endian byte-swapping/bitmaps by
  19  *        David S. Miller (davem@caip.rutgers.edu), 1995
  20  *
  21  *  All code that works with directory layout had been switched to pagecache
  22  *  and moved here. AV
  23  *
  24  *  Adapted for Lustre Light
  25  *  Copyright (C) 2002-2003, Cluster File Systems, Inc.
  26  *
  27  */
  28
  29 #include <linux/fs.h>
  30 #include <linux/ext2_fs.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/mm.h>
  33 #include <linux/version.h>
  34 #include <linux/smp_lock.h>
  35 #include <asm/uaccess.h>
  36 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  37 # include <linux/locks.h>   // for wait_on_buffer
  38 #else
  39 # include <linux/buffer_head.h>   // for wait_on_buffer
  40 #endif
  41
  42 #define DEBUG_SUBSYSTEM S_LLITE
  43
  44 #include <linux/obd_support.h>
  45 #include <linux/obd_class.h>
  46 #include <linux/lustre_lib.h>
  47 #include <linux/lustre_idl.h>
  48 #include <linux/lustre_mds.h>
  49 #include <linux/lustre_lite.h>
  50 #include <linux/lustre_dlm.h>
  51 #include "llite_internal.h"
  52
  53 typedef struct ext2_dir_entry_2 ext2_dirent;
  54
  55 #define PageChecked(page)        test_bit(PG_checked, &(page)->flags)
  56 #define SetPageChecked(page)     set_bit(PG_checked, &(page)->flags)
  57
  58 /* returns the page unlocked, but with a reference */
  59 static int ll_dir_readpage(struct file *file, struct page *page)
  60 {
  61         struct inode *inode = page->mapping->host;
  62         struct ll_sb_info *sbi = ll_i2sbi(inode);
  63         struct ll_fid mdc_fid;
  64         __u64 offset;
  65         int rc = 0;
  66         struct ptlrpc_request *request;
  67         struct lustre_handle lockh;
  68         struct mds_body *body;
  69         struct lookup_intent it = { .it_op = IT_READDIR };
  70         struct mdc_op_data data;
  71         struct obd_device *obddev = class_exp2obd(sbi->ll_mdc_exp);
  72         struct ldlm_res_id res_id =
  73                 { .name = {inode->i_ino, (__u64)inode->i_generation} };
  74         ENTRY;
  75
  76         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
  77                inode->i_generation, inode);
  78         if ((inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_SHIFT <= page->index){
  79                 /* XXX why do we need this exactly, and why do we think that
  80                  *     an all-zero directory page is useful?
  81                  */
  82                 CERROR("memsetting dir page %lu to zero (size %lld)\n",
  83                        page->index, inode->i_size);
  84                 memset(kmap(page), 0, PAGE_CACHE_SIZE);
  85                 kunmap(page);
  86                 GOTO(readpage_out, rc);
  87         }
  88
  89         rc = ldlm_lock_match(obddev->obd_namespace, LDLM_FL_BLOCK_GRANTED,
  90                              &res_id, LDLM_PLAIN, NULL, 0, LCK_PR, &lockh);
  91         if (!rc) {
  92                 ll_prepare_mdc_op_data(&data, inode, NULL, NULL, 0, 0);
  93
  94                 rc = mdc_enqueue(sbi->ll_mdc_exp, LDLM_PLAIN, &it, LCK_PR,
  95                                  &data, &lockh, NULL, 0,
  96                                  ldlm_completion_ast, ll_mdc_blocking_ast,
  97                                  inode);
  98                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
  99                 if (request)
 100                         ptlrpc_req_finished(request);
 101                 if (rc < 0) {
 102                         CERROR("lock enqueue: err: %d\n", rc);
 103                         unlock_page(page);
 104                         RETURN(rc);
 105                 }
 106         }
 107         ldlm_lock_dump_handle(D_OTHER, &lockh);
 108
 109         if (PageUptodate(page)) {
 110                 CERROR("Explain this please?\n");
 111                 GOTO(readpage_out, rc);
 112         }
 113
 114         mdc_pack_fid(&mdc_fid, inode->i_ino, inode->i_generation, S_IFDIR);
 115
 116         offset = page->index << PAGE_SHIFT;
 117         rc = mdc_readpage(sbi->ll_mdc_exp, &mdc_fid,
 118                           offset, page, &request);
 119         if (!rc) {
 120                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
 121                 LASSERT (body != NULL);         /* checked by mdc_readpage() */
 122                 LASSERT_REPSWABBED (request, 0); /* swabbed by mdc_readpage() */
 123
 124                 inode->i_size = body->size;
 125         }
 126         ptlrpc_req_finished(request);
 127         EXIT;
 128
 129  readpage_out:
 130         if (!rc)
 131                 SetPageUptodate(page);
 132
 133         unlock_page(page);
 134         ldlm_lock_decref(&lockh, LCK_PR);
 135         return rc;
 136 }
 137
 138 struct address_space_operations ll_dir_aops = {
 139         readpage: ll_dir_readpage,
 140 };
 141
 142 /*
 143  * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
 144  * more robust, but we have what we have
 145  */
 146 static inline unsigned ext2_chunk_size(struct inode *inode)
 147 {
 148         return inode->i_sb->s_blocksize;
 149 }
 150
 151 static inline void ext2_put_page(struct page *page)
 152 {
 153         kunmap(page);
 154         page_cache_release(page);
 155 }
 156
 157 static inline unsigned long dir_pages(struct inode *inode)
 158 {
 159         return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
 160 }
 161
 162
 163 static void ext2_check_page(struct page *page)
 164 {
 165         struct inode *dir = page->mapping->host;
 166         unsigned chunk_size = ext2_chunk_size(dir);
 167         char *kaddr = page_address(page);
 168         //      u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
 169         unsigned offs, rec_len;
 170         unsigned limit = PAGE_CACHE_SIZE;
 171         ext2_dirent *p;
 172         char *error;
 173
 174         if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
 175                 limit = dir->i_size & ~PAGE_CACHE_MASK;
 176                 if (limit & (chunk_size - 1)) {
 177                         CERROR("limit %d dir size %lld index %ld\n",
 178                                limit, dir->i_size, page->index);
 179                         goto Ebadsize;
 180                 }
 181                 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
 182                         ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
 183                         p->rec_len = cpu_to_le16(chunk_size);
 184                         p->name_len = 0;
 185                         p->inode = 0;
 186                 }
 187                 if (!limit)
 188                         goto out;
 189         }
 190         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
 191                 p = (ext2_dirent *)(kaddr + offs);
 192                 rec_len = le16_to_cpu(p->rec_len);
 193
 194                 if (rec_len < EXT2_DIR_REC_LEN(1))
 195                         goto Eshort;
 196                 if (rec_len & 3)
 197                         goto Ealign;
 198                 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
 199                         goto Enamelen;
 200                 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
 201                         goto Espan;
 202                 //              if (le32_to_cpu(p->inode) > max_inumber)
 203                 //goto Einumber;
 204         }
 205         if (offs != limit)
 206                 goto Eend;
 207 out:
 208         SetPageChecked(page);
 209         return;
 210
 211         /* Too bad, we had an error */
 212
 213 Ebadsize:
 214         CERROR("ext2_check_page"
 215                 "size of directory #%lu is not a multiple of chunk size\n",
 216                 dir->i_ino
 217         );
 218         goto fail;
 219 Eshort:
 220         error = "rec_len is smaller than minimal";
 221         goto bad_entry;
 222 Ealign:
 223         error = "unaligned directory entry";
 224         goto bad_entry;
 225 Enamelen:
 226         error = "rec_len is too small for name_len";
 227         goto bad_entry;
 228 Espan:
 229         error = "directory entry across blocks";
 230         goto bad_entry;
 231         //Einumber:
 232         // error = "inode out of bounds";
 233 bad_entry:
 234         CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
 235                 "offset=%lu+%u, inode=%lu, rec_len=%d, name_len=%d",
 236                 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT), offs,
 237                 (unsigned long) le32_to_cpu(p->inode),
 238                 rec_len, p->name_len);
 239         goto fail;
 240 Eend:
 241         p = (ext2_dirent *)(kaddr + offs);
 242         CERROR("ext2_check_page"
 243                 "entry in directory #%lu spans the page boundary"
 244                 "offset=%lu, inode=%lu",
 245                 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
 246                 (unsigned long) le32_to_cpu(p->inode));
 247 fail:
 248         SetPageChecked(page);
 249         SetPageError(page);
 250         LBUG();
 251 }
 252
 253 static struct page *ll_get_dir_page(struct inode *dir, unsigned long n)
 254 {
 255         struct address_space *mapping = dir->i_mapping;
 256         struct page *page = read_cache_page(mapping, n,
 257                                 (filler_t*)mapping->a_ops->readpage, NULL);
 258         if (!IS_ERR(page)) {
 259                 wait_on_page(page);
 260                 kmap(page);
 261                 if (!PageUptodate(page))
 262                         goto fail;
 263                 if (!PageChecked(page))
 264                         ext2_check_page(page);
 265                 if (PageError(page))
 266                         goto fail;
 267         }
 268         return page;
 269
 270 fail:
 271         ext2_put_page(page);
 272         return ERR_PTR(-EIO);
 273 }
 274
 275
 276 /*
 277  * p is at least 6 bytes before the end of page
 278  */
 279 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
 280 {
 281         return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
 282 }
 283
 284 static inline unsigned
 285 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
 286 {
 287         ext2_dirent *de = (ext2_dirent*)(base + offset);
 288         ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
 289         while ((char*)p < (char*)de)
 290                 p = ext2_next_entry(p);
 291         return (char *)p - base;
 292 }
 293
 294 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
 295         [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
 296         [EXT2_FT_REG_FILE]      DT_REG,
 297         [EXT2_FT_DIR]           DT_DIR,
 298         [EXT2_FT_CHRDEV]        DT_CHR,
 299         [EXT2_FT_BLKDEV]        DT_BLK,
 300         [EXT2_FT_FIFO]          DT_FIFO,
 301         [EXT2_FT_SOCK]          DT_SOCK,
 302         [EXT2_FT_SYMLINK]       DT_LNK,
 303 };
 304
 305
 306 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
 307 {
 308         loff_t pos = filp->f_pos;
 309         struct inode *inode = filp->f_dentry->d_inode;
 310         // XXX struct super_block *sb = inode->i_sb;
 311         unsigned offset = pos & ~PAGE_CACHE_MASK;
 312         unsigned long n = pos >> PAGE_CACHE_SHIFT;
 313         unsigned long npages = dir_pages(inode);
 314         unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
 315         unsigned char *types = NULL;
 316         int need_revalidate = (filp->f_version != inode->i_version);
 317         ENTRY;
 318
 319         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 320                inode->i_generation, inode);
 321         if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
 322                 GOTO(done, 0);
 323
 324         types = ext2_filetype_table;
 325
 326         for ( ; n < npages; n++, offset = 0) {
 327                 char *kaddr, *limit;
 328                 ext2_dirent *de;
 329                 struct page *page;
 330
 331                 CDEBUG(D_EXT2, "reading %lu of dir %lu page %lu, size %llu\n",
 332                        PAGE_CACHE_SIZE, inode->i_ino, n, inode->i_size);
 333                 page = ll_get_dir_page(inode, n);
 334
 335                 /* size might have been updated by mdc_readpage */
 336                 npages = dir_pages(inode);
 337
 338                 if (IS_ERR(page))
 339                         continue;
 340                 kaddr = page_address(page);
 341                 if (need_revalidate) {
 342                         offset = ext2_validate_entry(kaddr, offset, chunk_mask);
 343                         need_revalidate = 0;
 344                 }
 345                 de = (ext2_dirent *)(kaddr+offset);
 346                 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
 347                 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
 348                         if (de->inode) {
 349                                 int over;
 350                                 unsigned char d_type = DT_UNKNOWN;
 351
 352                                 if (types && de->file_type < EXT2_FT_MAX)
 353                                         d_type = types[de->file_type];
 354
 355                                 offset = (char *)de - kaddr;
 356                                 over = filldir(dirent, de->name, de->name_len,
 357                                                (n<<PAGE_CACHE_SHIFT) | offset,
 358                                                le32_to_cpu(de->inode), d_type);
 359                                 if (over) {
 360                                         ext2_put_page(page);
 361                                         GOTO(done,0);
 362                                 }
 363                         }
 364                 }
 365                 ext2_put_page(page);
 366         }
 367
 368 done:
 369         filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
 370         filp->f_version = inode->i_version;
 371         update_atime(inode);
 372         RETURN(0);
 373 }
 374
 375 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 376                         unsigned int cmd, unsigned long arg)
 377 {
 378         struct ll_sb_info *sbi = ll_i2sbi(inode);
 379         struct obd_ioctl_data *data;
 380         ENTRY;
 381
 382         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 383                inode->i_ino, inode->i_generation, inode, cmd);
 384
 385         if (_IOC_TYPE(cmd) == 'T') /* tty ioctls */
 386                 return -ENOTTY;
 387
 388         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_IOCTL);
 389         switch(cmd) {
 390         case EXT3_IOC_GETFLAGS:
 391         case EXT3_IOC_SETFLAGS:
 392                 RETURN( ll_iocontrol(inode, file, cmd, arg) );
 393         case IOC_MDC_LOOKUP: {
 394                 struct ptlrpc_request *request = NULL;
 395                 struct ll_fid fid;
 396                 char *buf = NULL;
 397                 char *filename;
 398                 int namelen, rc, len = 0;
 399                 unsigned long valid;
 400
 401                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 402                 if (rc)
 403                         RETURN(rc);
 404                 data = (void *)buf;
 405
 406                 filename = data->ioc_inlbuf1;
 407                 namelen = data->ioc_inllen1;
 408
 409                 if (namelen < 1) {
 410                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 411                         GOTO(out, rc = -EINVAL);
 412                 }
 413
 414                 valid = OBD_MD_FLID;
 415                 ll_inode2fid(&fid, inode);
 416                 rc = mdc_getattr_name(sbi->ll_mdc_exp, &fid,
 417                                       filename, namelen, valid, 0, &request);
 418                 if (rc < 0) {
 419                         CDEBUG(D_INFO, "mdc_getattr_name: %d\n", rc);
 420                         GOTO(out, rc);
 421                 }
 422
 423                 ptlrpc_req_finished(request);
 424
 425                 EXIT;
 426         out:
 427                 obd_ioctl_freedata(buf, len);
 428                 return rc;
 429         }
 430         case LL_IOC_LOV_SETSTRIPE:
 431         case LL_IOC_LOV_GETSTRIPE:
 432                 RETURN(-ENOTTY);
 433         case IOC_MDC_GETSTRIPE: {
 434                 struct ptlrpc_request *request = NULL;
 435                 struct ll_fid fid;
 436                 struct mds_body *body;
 437                 struct lov_user_md *lump = (struct lov_user_md *)arg;
 438                 struct lov_mds_md *lmm;
 439                 char *filename;
 440                 int rc, lmmsize;
 441
 442                 filename = getname((const char *)arg);
 443                 if (IS_ERR(filename))
 444                         RETURN(PTR_ERR(filename));
 445
 446                 ll_inode2fid(&fid, inode);
 447                 rc = mdc_getattr_name(sbi->ll_mdc_exp, &fid, filename,
 448                                       strlen(filename)+1, OBD_MD_FLEASIZE,
 449                                       obd_size_diskmd(sbi->ll_osc_exp, NULL),
 450                                       &request);
 451                 if (rc < 0) {
 452                         CDEBUG(D_INFO, "mdc_getattr_name failed on %s: rc %d\n",
 453                                filename, rc);
 454                         GOTO(out_name, rc);
 455                 }
 456
 457                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
 458                 LASSERT(body != NULL);         /* checked by mdc_getattr_name */
 459                 LASSERT_REPSWABBED(request, 0);/* swabbed by mdc_getattr_name */
 460
 461                 lmmsize = body->eadatasize;
 462
 463                 if (!(body->valid & OBD_MD_FLEASIZE) || lmmsize == 0)
 464                         GOTO(out_req, rc = -ENODATA);
 465
 466                 if (lmmsize > 4096)
 467                         GOTO(out_req, rc = -EFBIG);
 468
 469                 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
 470                 LASSERT(lmm != NULL);
 471                 LASSERT_REPSWABBED(request, 1);
 472
 473                 rc = copy_to_user(lump, lmm, lmmsize);
 474                 if (rc)
 475                         GOTO(out_req, rc = -EFAULT);
 476
 477                 EXIT;
 478         out_req:
 479                 ptlrpc_req_finished(request);
 480         out_name:
 481                 putname(filename);
 482                 return rc;
 483         }
 484         case OBD_IOC_PING: {
 485                 struct ptlrpc_request *req = NULL;
 486                 char *buf = NULL;
 487                 int rc, len=0;
 488                 struct client_obd *cli;
 489                 struct obd_device *obd;
 490
 491                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 492                 if (rc)
 493                         RETURN(rc);
 494                 data = (void *)buf;
 495
 496                 obd = class_name2obd(data->ioc_inlbuf1);
 497
 498                 if (!obd )
 499                         GOTO(out_ping, rc = -ENODEV);
 500
 501                 if (!obd->obd_attached) {
 502                         CERROR("Device %d not attached\n", obd->obd_minor);
 503                         GOTO(out_ping, rc = -ENODEV);
 504                 }
 505                 if (!obd->obd_set_up) {
 506                         CERROR("Device %d still not setup\n", obd->obd_minor);
 507                         GOTO(out_ping, rc = -ENODEV);
 508                 }
 509                 cli = &obd->u.cli;
 510                 req = ptlrpc_prep_req(cli->cl_import, OBD_PING, 0, NULL, NULL);
 511                 if (!req)
 512                         GOTO(out_ping, rc = -ENOMEM);
 513
 514                 req->rq_replen = lustre_msg_size(0, NULL);
 515                 req->rq_send_state = LUSTRE_IMP_FULL;
 516
 517                 rc = ptlrpc_queue_wait(req);
 518
 519                 ptlrpc_req_finished(req);
 520         out_ping:
 521                 obd_ioctl_freedata(buf, len);
 522                 return rc;
 523         }
 524         case OBD_IOC_LLOG_CATINFO: {
 525                 struct ptlrpc_request *req = NULL;
 526                 char *buf = NULL;
 527                 int rc, len = 0;
 528                 char *bufs[2], *str;
 529                 int lens[2], size;
 530
 531                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 532                 if (rc)
 533                         RETURN(rc);
 534                 data = (void *)buf;
 535
 536                 if (!data->ioc_inlbuf1) {
 537                         obd_ioctl_freedata(buf, len);
 538                         RETURN(-EINVAL);
 539                 }
 540
 541                 lens[0] = data->ioc_inllen1;
 542                 bufs[0] = data->ioc_inlbuf1;
 543                 if (data->ioc_inllen2) {
 544                         lens[1] = data->ioc_inllen2;
 545                         bufs[1] = data->ioc_inlbuf2;
 546                 } else {
 547                         lens[1] = 0;
 548                         bufs[1] = NULL;
 549                 }
 550                 size = data->ioc_plen1;
 551                 req = ptlrpc_prep_req(sbi2mdc(sbi)->cl_import, LLOG_CATINFO,
 552                                       2, lens, bufs);
 553                 if (!req)
 554                         GOTO(out_catinfo, rc = -ENOMEM);
 555                 req->rq_replen = lustre_msg_size(1, &size);
 556
 557                 rc = ptlrpc_queue_wait(req);
 558                 str = lustre_msg_string(req->rq_repmsg, 0, data->ioc_plen1);
 559                 if (!rc)
 560                         rc = copy_to_user(data->ioc_pbuf1, str,
 561                                           data->ioc_plen1);
 562                 ptlrpc_req_finished(req);
 563         out_catinfo:
 564                 obd_ioctl_freedata(buf, len);
 565                 RETURN(rc);
 566         }
 567         default:
 568                 return obd_iocontrol(cmd, sbi->ll_osc_exp,0,NULL,(void *)arg);
 569         }
 570 }
 571
 572 int ll_dir_open(struct inode *inode, struct file *file)
 573 {
 574         return ll_file_open(inode, file);
 575 }
 576
 577 int ll_dir_release(struct inode *inode, struct file *file)
 578 {
 579         return ll_file_release(inode, file);
 580 }
 581
 582 struct file_operations ll_dir_operations = {
 583         open: ll_dir_open,
 584         release: ll_dir_release,
 585         read: generic_read_dir,
 586         readdir: ll_readdir,
 587         ioctl: ll_dir_ioctl
 588 };
 589