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
  52 typedef struct ext2_dir_entry_2 ext2_dirent;
  53
  54 #define PageChecked(page)        test_bit(PG_checked, &(page)->flags)
  55 #define SetPageChecked(page)     set_bit(PG_checked, &(page)->flags)
  56
  57 /* returns the page unlocked, but with a reference */
  58 static int ll_dir_readpage(struct file *file, struct page *page)
  59 {
  60         struct inode *inode = page->mapping->host;
  61         struct ll_sb_info *sbi = ll_i2sbi(inode);
  62         __u64 offset;
  63         int rc = 0;
  64         struct ptlrpc_request *request;
  65         struct lustre_handle lockh;
  66         struct mds_body *body;
  67         struct lookup_intent it = { .it_op = IT_READDIR };
  68         struct mdc_op_data data;
  69         struct obd_device *obddev = class_conn2obd(&sbi->ll_mdc_conn);
  70         struct ldlm_res_id res_id =
  71                 { .name = {inode->i_ino, (__u64)inode->i_generation} };
  72         int flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_MATCH_DATA;
  73         ENTRY;
  74
  75         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
  76                inode->i_generation, inode);
  77         if ((inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_SHIFT <= page->index){
  78                 /* XXX why do we need this exactly, and why do we think that
  79                  *     an all-zero directory page is useful?
  80                  */
  81                 CERROR("memsetting dir page %lu to zero (size %lld)\n",
  82                        page->index, inode->i_size);
  83                 memset(kmap(page), 0, PAGE_CACHE_SIZE);
  84                 kunmap(page);
  85                 GOTO(readpage_out, rc);
  86         }
  87
  88         rc = ldlm_lock_match(obddev->obd_namespace, flags, &res_id,
  89                              LDLM_PLAIN, NULL, 0, LCK_PR, inode,
  90                              &lockh);
  91         if (!rc) {
  92                 ll_prepare_mdc_op_data(&data, inode, NULL, NULL, 0, 0);
  93
  94                 rc = mdc_enqueue(&sbi->ll_mdc_conn, 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.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         offset = page->index << PAGE_SHIFT;
 115         rc = mdc_readpage(&sbi->ll_mdc_conn, inode->i_ino,
 116                           S_IFDIR, offset, page, &request);
 117         if (!rc) {
 118                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
 119                 LASSERT (body != NULL);         /* checked by mdc_readpage() */
 120                 LASSERT_REPSWABBED (request, 0); /* swabbed by mdc_readpage() */
 121
 122                 inode->i_size = body->size;
 123         }
 124         ptlrpc_req_finished(request);
 125         EXIT;
 126
 127  readpage_out:
 128         if (!rc)
 129                 SetPageUptodate(page);
 130
 131         unlock_page(page);
 132         ldlm_lock_decref(&lockh, LCK_PR);
 133         return rc;
 134 }
 135
 136 struct address_space_operations ll_dir_aops = {
 137         readpage: ll_dir_readpage,
 138 };
 139
 140 /*
 141  * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
 142  * more robust, but we have what we have
 143  */
 144 static inline unsigned ext2_chunk_size(struct inode *inode)
 145 {
 146         return inode->i_sb->s_blocksize;
 147 }
 148
 149 static inline void ext2_put_page(struct page *page)
 150 {
 151         kunmap(page);
 152         page_cache_release(page);
 153 }
 154
 155 static inline unsigned long dir_pages(struct inode *inode)
 156 {
 157         return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
 158 }
 159
 160
 161 static void ext2_check_page(struct page *page)
 162 {
 163         struct inode *dir = page->mapping->host;
 164         unsigned chunk_size = ext2_chunk_size(dir);
 165         char *kaddr = page_address(page);
 166         //      u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
 167         unsigned offs, rec_len;
 168         unsigned limit = PAGE_CACHE_SIZE;
 169         ext2_dirent *p;
 170         char *error;
 171
 172         if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
 173                 limit = dir->i_size & ~PAGE_CACHE_MASK;
 174                 if (limit & (chunk_size - 1)) {
 175                         CERROR("limit %d dir size %lld index %ld\n",
 176                                limit, dir->i_size, page->index);
 177                         goto Ebadsize;
 178                 }
 179                 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
 180                         ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
 181                         p->rec_len = cpu_to_le16(chunk_size);
 182                         p->name_len = 0;
 183                         p->inode = 0;
 184                 }
 185                 if (!limit)
 186                         goto out;
 187         }
 188         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
 189                 p = (ext2_dirent *)(kaddr + offs);
 190                 rec_len = le16_to_cpu(p->rec_len);
 191
 192                 if (rec_len < EXT2_DIR_REC_LEN(1))
 193                         goto Eshort;
 194                 if (rec_len & 3)
 195                         goto Ealign;
 196                 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
 197                         goto Enamelen;
 198                 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
 199                         goto Espan;
 200                 //              if (le32_to_cpu(p->inode) > max_inumber)
 201                 //goto Einumber;
 202         }
 203         if (offs != limit)
 204                 goto Eend;
 205 out:
 206         SetPageChecked(page);
 207         return;
 208
 209         /* Too bad, we had an error */
 210
 211 Ebadsize:
 212         CERROR("ext2_check_page"
 213                 "size of directory #%lu is not a multiple of chunk size\n",
 214                 dir->i_ino
 215         );
 216         goto fail;
 217 Eshort:
 218         error = "rec_len is smaller than minimal";
 219         goto bad_entry;
 220 Ealign:
 221         error = "unaligned directory entry";
 222         goto bad_entry;
 223 Enamelen:
 224         error = "rec_len is too small for name_len";
 225         goto bad_entry;
 226 Espan:
 227         error = "directory entry across blocks";
 228         goto bad_entry;
 229         //Einumber:
 230         // error = "inode out of bounds";
 231 bad_entry:
 232         CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
 233                 "offset=%lu+%u, inode=%lu, rec_len=%d, name_len=%d",
 234                 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT), offs,
 235                 (unsigned long) le32_to_cpu(p->inode),
 236                 rec_len, p->name_len);
 237         goto fail;
 238 Eend:
 239         p = (ext2_dirent *)(kaddr + offs);
 240         CERROR("ext2_check_page"
 241                 "entry in directory #%lu spans the page boundary"
 242                 "offset=%lu, inode=%lu",
 243                 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
 244                 (unsigned long) le32_to_cpu(p->inode));
 245 fail:
 246         SetPageChecked(page);
 247         SetPageError(page);
 248         LBUG();
 249 }
 250
 251 static struct page *ll_get_dir_page(struct inode *dir, unsigned long n)
 252 {
 253         struct address_space *mapping = dir->i_mapping;
 254         struct page *page = read_cache_page(mapping, n,
 255                                 (filler_t*)mapping->a_ops->readpage, NULL);
 256         if (!IS_ERR(page)) {
 257                 wait_on_page(page);
 258                 kmap(page);
 259                 if (!PageUptodate(page))
 260                         goto fail;
 261                 if (!PageChecked(page))
 262                         ext2_check_page(page);
 263                 if (PageError(page))
 264                         goto fail;
 265         }
 266         return page;
 267
 268 fail:
 269         ext2_put_page(page);
 270         return ERR_PTR(-EIO);
 271 }
 272
 273
 274 /*
 275  * p is at least 6 bytes before the end of page
 276  */
 277 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
 278 {
 279         return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
 280 }
 281
 282 static inline unsigned
 283 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
 284 {
 285         ext2_dirent *de = (ext2_dirent*)(base + offset);
 286         ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
 287         while ((char*)p < (char*)de)
 288                 p = ext2_next_entry(p);
 289         return (char *)p - base;
 290 }
 291
 292 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
 293         [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
 294         [EXT2_FT_REG_FILE]      DT_REG,
 295         [EXT2_FT_DIR]           DT_DIR,
 296         [EXT2_FT_CHRDEV]        DT_CHR,
 297         [EXT2_FT_BLKDEV]        DT_BLK,
 298         [EXT2_FT_FIFO]          DT_FIFO,
 299         [EXT2_FT_SOCK]          DT_SOCK,
 300         [EXT2_FT_SYMLINK]       DT_LNK,
 301 };
 302
 303
 304 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
 305 {
 306         loff_t pos = filp->f_pos;
 307         struct inode *inode = filp->f_dentry->d_inode;
 308         // XXX struct super_block *sb = inode->i_sb;
 309         unsigned offset = pos & ~PAGE_CACHE_MASK;
 310         unsigned long n = pos >> PAGE_CACHE_SHIFT;
 311         unsigned long npages = dir_pages(inode);
 312         unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
 313         unsigned char *types = NULL;
 314         int need_revalidate = (filp->f_version != inode->i_version);
 315         ENTRY;
 316
 317         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 318                inode->i_generation, inode);
 319         if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
 320                 GOTO(done, 0);
 321
 322         types = ext2_filetype_table;
 323
 324         for ( ; n < npages; n++, offset = 0) {
 325                 char *kaddr, *limit;
 326                 ext2_dirent *de;
 327                 struct page *page;
 328
 329                 CDEBUG(D_EXT2, "reading %lu of dir %lu page %lu, size %llu\n",
 330                        PAGE_CACHE_SIZE, inode->i_ino, n, inode->i_size);
 331                 page = ll_get_dir_page(inode, n);
 332
 333                 /* size might have been updated by mdc_readpage */
 334                 npages = dir_pages(inode);
 335
 336                 if (IS_ERR(page))
 337                         continue;
 338                 kaddr = page_address(page);
 339                 if (need_revalidate) {
 340                         offset = ext2_validate_entry(kaddr, offset, chunk_mask);
 341                         need_revalidate = 0;
 342                 }
 343                 de = (ext2_dirent *)(kaddr+offset);
 344                 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
 345                 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
 346                         if (de->inode) {
 347                                 int over;
 348                                 unsigned char d_type = DT_UNKNOWN;
 349
 350                                 if (types && de->file_type < EXT2_FT_MAX)
 351                                         d_type = types[de->file_type];
 352
 353                                 offset = (char *)de - kaddr;
 354                                 over = filldir(dirent, de->name, de->name_len,
 355                                                (n<<PAGE_CACHE_SHIFT) | offset,
 356                                                le32_to_cpu(de->inode), d_type);
 357                                 if (over) {
 358                                         ext2_put_page(page);
 359                                         GOTO(done,0);
 360                                 }
 361                         }
 362                 }
 363                 ext2_put_page(page);
 364         }
 365
 366 done:
 367         filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
 368         filp->f_version = inode->i_version;
 369         update_atime(inode);
 370         RETURN(0);
 371 }
 372
 373 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 374                         unsigned int cmd, unsigned long arg)
 375 {
 376         struct ll_sb_info *sbi = ll_i2sbi(inode);
 377         struct obd_ioctl_data *data;
 378         ENTRY;
 379
 380         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%u\n", inode->i_ino,
 381                inode->i_generation, inode, cmd);
 382
 383         if (_IOC_TYPE(cmd) == 'T') /* tty ioctls */
 384                 return -ENOTTY;
 385
 386         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_IOCTL);
 387         switch(cmd) {
 388         case IOC_MDC_LOOKUP: {
 389                 struct ptlrpc_request *request = NULL;
 390                 struct ll_fid fid;
 391                 char *buf = NULL;
 392                 struct mds_body *body;
 393                 char *filename;
 394                 int namelen, rc, err, len = 0;
 395                 unsigned long valid;
 396
 397                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 398                 if (rc)
 399                         RETURN(rc);
 400                 data = (void *)buf;
 401
 402                 filename = data->ioc_inlbuf1;
 403                 namelen = data->ioc_inllen1;
 404
 405                 if (namelen < 1) {
 406                         CERROR("IOC_MDC_LOOKUP missing filename\n");
 407                         GOTO(out, rc = -EINVAL);
 408                 }
 409
 410                 valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE;
 411                 ll_inode2fid(&fid, inode);
 412                 rc = mdc_getattr_name(&sbi->ll_mdc_conn, &fid,
 413                                       filename, namelen, valid, 0, &request);
 414                 if (rc < 0) {
 415                         CERROR("mdc_getattr_name: %d\n", rc);
 416                         GOTO(out, rc);
 417                 }
 418
 419                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
 420                 LASSERT(body != NULL);         /* checked by mdc_getattr_name */
 421                 LASSERT_REPSWABBED(request, 0);/* swabbed by mdc_getattr_name */
 422
 423                 /* surely there's a better way -phik */
 424                 data->ioc_obdo1.o_mode = body->mode;
 425                 data->ioc_obdo1.o_uid = body->uid;
 426                 data->ioc_obdo1.o_gid = body->gid;
 427
 428                 ptlrpc_req_finished(request);
 429
 430                 err = copy_to_user((void *)arg, buf, len);
 431                 if (err)
 432                         GOTO(out, rc = -EFAULT);
 433
 434                 EXIT;
 435         out:
 436                 obd_ioctl_freedata(buf, len);
 437                 return rc;
 438         }
 439         case LL_IOC_LOV_SETSTRIPE:
 440         case LL_IOC_LOV_GETSTRIPE:
 441                 RETURN(-ENOTTY);
 442         case IOC_MDC_GETSTRIPE: {
 443                 struct ptlrpc_request *request = NULL;
 444                 struct ll_fid fid;
 445                 struct mds_body *body;
 446                 struct lov_mds_md *lmm;
 447                 char *filename;
 448                 int rc, lmmsize;
 449
 450                 filename = getname((const char *)arg);
 451                 if (IS_ERR(filename))
 452                         RETURN(PTR_ERR(filename));
 453
 454                 ll_inode2fid(&fid, inode);
 455                 rc = mdc_getattr_name(&sbi->ll_mdc_conn, &fid, filename,
 456                                       strlen(filename)+1, OBD_MD_FLEASIZE,
 457                                       obd_size_diskmd(&sbi->ll_osc_conn, NULL),
 458                                       &request);
 459                 if (rc < 0) {
 460                         CERROR("mdc_getattr_name: failed on %s: rc %d\n",
 461                                filename, rc);
 462                         GOTO(out_name, rc);
 463                 }
 464
 465                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
 466                 LASSERT(body != NULL);         /* checked by mdc_getattr_name */
 467                 LASSERT_REPSWABBED(request, 0);/* swabbed by mdc_getattr_name */
 468
 469                 lmmsize = body->eadatasize;
 470
 471                 if (!(body->valid & OBD_MD_FLEASIZE) || lmmsize == 0)
 472                         GOTO(out_req, rc = -ENODATA);
 473
 474                 if (lmmsize > 4096)
 475                         GOTO(out_req, rc = -EFBIG);
 476
 477                 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
 478                 LASSERT(lmm != NULL);
 479                 LASSERT_REPSWABBED(request, 1);
 480
 481                 rc = copy_to_user((struct lov_mds_md *)arg, lmm, lmmsize);
 482                 if (rc)
 483                         GOTO(out_req, rc = -EFAULT);
 484
 485                 EXIT;
 486         out_req:
 487                 ptlrpc_req_finished(request);
 488         out_name:
 489                 putname(filename);
 490                 return rc;
 491         }
 492         default:
 493                 return obd_iocontrol(cmd,&sbi->ll_osc_conn,0,NULL,(void *)arg);
 494         }
 495 }
 496
 497 int ll_dir_open(struct inode *inode, struct file *file)
 498 {
 499         return ll_file_open(inode, file);
 500 }
 501
 502 int ll_dir_release(struct inode *inode, struct file *file)
 503 {
 504         return ll_file_release(inode, file);
 505 }
 506
 507 struct file_operations ll_dir_operations = {
 508         open: ll_dir_open,
 509         release: ll_dir_release,
 510         read: generic_read_dir,
 511         readdir: ll_readdir,
 512         ioctl: ll_dir_ioctl
 513 };
 514