fs/ceph/dir.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/spinlock.h>
   4 #include <linux/fs_struct.h>
   5 #include <linux/namei.h>
   6 #include <linux/slab.h>
   7 #include <linux/sched.h>
   8
   9 #include "super.h"
  10 #include "mds_client.h"
  11
  12 /*
  13  * Directory operations: readdir, lookup, create, link, unlink,
  14  * rename, etc.
  15  */
  16
  17 /*
  18  * Ceph MDS operations are specified in terms of a base ino and
  19  * relative path.  Thus, the client can specify an operation on a
  20  * specific inode (e.g., a getattr due to fstat(2)), or as a path
  21  * relative to, say, the root directory.
  22  *
  23  * Normally, we limit ourselves to strict inode ops (no path component)
  24  * or dentry operations (a single path component relative to an ino).  The
  25  * exception to this is open_root_dentry(), which will open the mount
  26  * point by name.
  27  */
  28
  29 const struct dentry_operations ceph_dentry_ops;
  30
  31 /*
  32  * Initialize ceph dentry state.
  33  */
  34 int ceph_init_dentry(struct dentry *dentry)
  35 {
  36         struct ceph_dentry_info *di;
  37
  38         if (dentry->d_fsdata)
  39                 return 0;
  40
  41         di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
  42         if (!di)
  43                 return -ENOMEM;          /* oh well */
  44
  45         spin_lock(&dentry->d_lock);
  46         if (dentry->d_fsdata) {
  47                 /* lost a race */
  48                 kmem_cache_free(ceph_dentry_cachep, di);
  49                 goto out_unlock;
  50         }
  51
  52         if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP)
  53                 d_set_d_op(dentry, &ceph_dentry_ops);
  54         else if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
  55                 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
  56         else
  57                 d_set_d_op(dentry, &ceph_snap_dentry_ops);
  58
  59         di->dentry = dentry;
  60         di->lease_session = NULL;
  61         dentry->d_time = jiffies;
  62         /* avoid reordering d_fsdata setup so that the check above is safe */
  63         smp_mb();
  64         dentry->d_fsdata = di;
  65         ceph_dentry_lru_add(dentry);
  66 out_unlock:
  67         spin_unlock(&dentry->d_lock);
  68         return 0;
  69 }
  70
  71 /*
  72  * for readdir, we encode the directory frag and offset within that
  73  * frag into f_pos.
  74  */
  75 static unsigned fpos_frag(loff_t p)
  76 {
  77         return p >> 32;
  78 }
  79 static unsigned fpos_off(loff_t p)
  80 {
  81         return p & 0xffffffff;
  82 }
  83
  84 static int fpos_cmp(loff_t l, loff_t r)
  85 {
  86         int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
  87         if (v)
  88                 return v;
  89         return (int)(fpos_off(l) - fpos_off(r));
  90 }
  91
  92 /*
  93  * make note of the last dentry we read, so we can
  94  * continue at the same lexicographical point,
  95  * regardless of what dir changes take place on the
  96  * server.
  97  */
  98 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
  99                             int len, unsigned next_offset)
 100 {
 101         char *buf = kmalloc(len+1, GFP_KERNEL);
 102         if (!buf)
 103                 return -ENOMEM;
 104         kfree(fi->last_name);
 105         fi->last_name = buf;
 106         memcpy(fi->last_name, name, len);
 107         fi->last_name[len] = 0;
 108         fi->next_offset = next_offset;
 109         dout("note_last_dentry '%s'\n", fi->last_name);
 110         return 0;
 111 }
 112
 113
 114 static struct dentry *
 115 __dcache_find_get_entry(struct dentry *parent, u64 idx,
 116                         struct ceph_readdir_cache_control *cache_ctl)
 117 {
 118         struct inode *dir = d_inode(parent);
 119         struct dentry *dentry;
 120         unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
 121         loff_t ptr_pos = idx * sizeof(struct dentry *);
 122         pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
 123
 124         if (ptr_pos >= i_size_read(dir))
 125                 return NULL;
 126
 127         if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
 128                 ceph_readdir_cache_release(cache_ctl);
 129                 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
 130                 if (!cache_ctl->page) {
 131                         dout(" page %lu not found\n", ptr_pgoff);
 132                         return ERR_PTR(-EAGAIN);
 133                 }
 134                 /* reading/filling the cache are serialized by
 135                    i_mutex, no need to use page lock */
 136                 unlock_page(cache_ctl->page);
 137                 cache_ctl->dentries = kmap(cache_ctl->page);
 138         }
 139
 140         cache_ctl->index = idx & idx_mask;
 141
 142         rcu_read_lock();
 143         spin_lock(&parent->d_lock);
 144         /* check i_size again here, because empty directory can be
 145          * marked as complete while not holding the i_mutex. */
 146         if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
 147                 dentry = cache_ctl->dentries[cache_ctl->index];
 148         else
 149                 dentry = NULL;
 150         spin_unlock(&parent->d_lock);
 151         if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
 152                 dentry = NULL;
 153         rcu_read_unlock();
 154         return dentry ? : ERR_PTR(-EAGAIN);
 155 }
 156
 157 /*
 158  * When possible, we try to satisfy a readdir by peeking at the
 159  * dcache.  We make this work by carefully ordering dentries on
 160  * d_child when we initially get results back from the MDS, and
 161  * falling back to a "normal" sync readdir if any dentries in the dir
 162  * are dropped.
 163  *
 164  * Complete dir indicates that we have all dentries in the dir.  It is
 165  * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
 166  * the MDS if/when the directory is modified).
 167  */
 168 static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
 169                             u32 shared_gen)
 170 {
 171         struct ceph_file_info *fi = file->private_data;
 172         struct dentry *parent = file->f_path.dentry;
 173         struct inode *dir = d_inode(parent);
 174         struct dentry *dentry, *last = NULL;
 175         struct ceph_dentry_info *di;
 176         struct ceph_readdir_cache_control cache_ctl = {};
 177         u64 idx = 0;
 178         int err = 0;
 179
 180         dout("__dcache_readdir %p v%u at %llu\n", dir, shared_gen, ctx->pos);
 181
 182         /* search start position */
 183         if (ctx->pos > 2) {
 184                 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
 185                 while (count > 0) {
 186                         u64 step = count >> 1;
 187                         dentry = __dcache_find_get_entry(parent, idx + step,
 188                                                          &cache_ctl);
 189                         if (!dentry) {
 190                                 /* use linar search */
 191                                 idx = 0;
 192                                 break;
 193                         }
 194                         if (IS_ERR(dentry)) {
 195                                 err = PTR_ERR(dentry);
 196                                 goto out;
 197                         }
 198                         di = ceph_dentry(dentry);
 199                         spin_lock(&dentry->d_lock);
 200                         if (fpos_cmp(di->offset, ctx->pos) < 0) {
 201                                 idx += step + 1;
 202                                 count -= step + 1;
 203                         } else {
 204                                 count = step;
 205                         }
 206                         spin_unlock(&dentry->d_lock);
 207                         dput(dentry);
 208                 }
 209
 210                 dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
 211         }
 212
 213
 214         for (;;) {
 215                 bool emit_dentry = false;
 216                 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
 217                 if (!dentry) {
 218                         fi->flags |= CEPH_F_ATEND;
 219                         err = 0;
 220                         break;
 221                 }
 222                 if (IS_ERR(dentry)) {
 223                         err = PTR_ERR(dentry);
 224                         goto out;
 225                 }
 226
 227                 di = ceph_dentry(dentry);
 228                 spin_lock(&dentry->d_lock);
 229                 if (di->lease_shared_gen == shared_gen &&
 230                     d_really_is_positive(dentry) &&
 231                     ceph_snap(d_inode(dentry)) != CEPH_SNAPDIR &&
 232                     ceph_ino(d_inode(dentry)) != CEPH_INO_CEPH &&
 233                     fpos_cmp(ctx->pos, di->offset) <= 0) {
 234                         emit_dentry = true;
 235                 }
 236                 spin_unlock(&dentry->d_lock);
 237
 238                 if (emit_dentry) {
 239                         dout(" %llu (%llu) dentry %p %pd %p\n", di->offset, ctx->pos,
 240                              dentry, dentry, d_inode(dentry));
 241                         ctx->pos = di->offset;
 242                         if (!dir_emit(ctx, dentry->d_name.name,
 243                                       dentry->d_name.len,
 244                                       ceph_translate_ino(dentry->d_sb,
 245                                                          d_inode(dentry)->i_ino),
 246                                       d_inode(dentry)->i_mode >> 12)) {
 247                                 dput(dentry);
 248                                 err = 0;
 249                                 break;
 250                         }
 251                         ctx->pos++;
 252
 253                         if (last)
 254                                 dput(last);
 255                         last = dentry;
 256                 } else {
 257                         dput(dentry);
 258                 }
 259         }
 260 out:
 261         ceph_readdir_cache_release(&cache_ctl);
 262         if (last) {
 263                 int ret;
 264                 di = ceph_dentry(last);
 265                 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
 266                                        fpos_off(di->offset) + 1);
 267                 if (ret < 0)
 268                         err = ret;
 269                 dput(last);
 270         }
 271         return err;
 272 }
 273
 274 static int ceph_readdir(struct file *file, struct dir_context *ctx)
 275 {
 276         struct ceph_file_info *fi = file->private_data;
 277         struct inode *inode = file_inode(file);
 278         struct ceph_inode_info *ci = ceph_inode(inode);
 279         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 280         struct ceph_mds_client *mdsc = fsc->mdsc;
 281         unsigned frag = fpos_frag(ctx->pos);
 282         int off = fpos_off(ctx->pos);
 283         int err;
 284         u32 ftype;
 285         struct ceph_mds_reply_info_parsed *rinfo;
 286
 287         dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
 288         if (fi->flags & CEPH_F_ATEND)
 289                 return 0;
 290
 291         /* always start with . and .. */
 292         if (ctx->pos == 0) {
 293                 dout("readdir off 0 -> '.'\n");
 294                 if (!dir_emit(ctx, ".", 1,
 295                             ceph_translate_ino(inode->i_sb, inode->i_ino),
 296                             inode->i_mode >> 12))
 297                         return 0;
 298                 ctx->pos = 1;
 299                 off = 1;
 300         }
 301         if (ctx->pos == 1) {
 302                 ino_t ino = parent_ino(file->f_path.dentry);
 303                 dout("readdir off 1 -> '..'\n");
 304                 if (!dir_emit(ctx, "..", 2,
 305                             ceph_translate_ino(inode->i_sb, ino),
 306                             inode->i_mode >> 12))
 307                         return 0;
 308                 ctx->pos = 2;
 309                 off = 2;
 310         }
 311
 312         /* can we use the dcache? */
 313         spin_lock(&ci->i_ceph_lock);
 314         if (ceph_test_mount_opt(fsc, DCACHE) &&
 315             !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
 316             ceph_snap(inode) != CEPH_SNAPDIR &&
 317             __ceph_dir_is_complete_ordered(ci) &&
 318             __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
 319                 u32 shared_gen = ci->i_shared_gen;
 320                 spin_unlock(&ci->i_ceph_lock);
 321                 err = __dcache_readdir(file, ctx, shared_gen);
 322                 if (err != -EAGAIN)
 323                         return err;
 324                 frag = fpos_frag(ctx->pos);
 325                 off = fpos_off(ctx->pos);
 326         } else {
 327                 spin_unlock(&ci->i_ceph_lock);
 328         }
 329
 330         /* proceed with a normal readdir */
 331 more:
 332         /* do we have the correct frag content buffered? */
 333         if (fi->frag != frag || fi->last_readdir == NULL) {
 334                 struct ceph_mds_request *req;
 335                 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
 336                         CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
 337
 338                 /* discard old result, if any */
 339                 if (fi->last_readdir) {
 340                         ceph_mdsc_put_request(fi->last_readdir);
 341                         fi->last_readdir = NULL;
 342                 }
 343
 344                 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
 345                      ceph_vinop(inode), frag, fi->last_name);
 346                 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 347                 if (IS_ERR(req))
 348                         return PTR_ERR(req);
 349                 err = ceph_alloc_readdir_reply_buffer(req, inode);
 350                 if (err) {
 351                         ceph_mdsc_put_request(req);
 352                         return err;
 353                 }
 354                 /* hints to request -> mds selection code */
 355                 req->r_direct_mode = USE_AUTH_MDS;
 356                 req->r_direct_hash = ceph_frag_value(frag);
 357                 req->r_direct_is_hash = true;
 358                 if (fi->last_name) {
 359                         req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
 360                         if (!req->r_path2) {
 361                                 ceph_mdsc_put_request(req);
 362                                 return -ENOMEM;
 363                         }
 364                 }
 365                 req->r_dir_release_cnt = fi->dir_release_count;
 366                 req->r_dir_ordered_cnt = fi->dir_ordered_count;
 367                 req->r_readdir_cache_idx = fi->readdir_cache_idx;
 368                 req->r_readdir_offset = fi->next_offset;
 369                 req->r_args.readdir.frag = cpu_to_le32(frag);
 370
 371                 req->r_inode = inode;
 372                 ihold(inode);
 373                 req->r_dentry = dget(file->f_path.dentry);
 374                 err = ceph_mdsc_do_request(mdsc, NULL, req);
 375                 if (err < 0) {
 376                         ceph_mdsc_put_request(req);
 377                         return err;
 378                 }
 379                 dout("readdir got and parsed readdir result=%d"
 380                      " on frag %x, end=%d, complete=%d\n", err, frag,
 381                      (int)req->r_reply_info.dir_end,
 382                      (int)req->r_reply_info.dir_complete);
 383
 384
 385                 /* note next offset and last dentry name */
 386                 rinfo = &req->r_reply_info;
 387                 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
 388                         frag = le32_to_cpu(rinfo->dir_dir->frag);
 389                         off = req->r_readdir_offset;
 390                         fi->next_offset = off;
 391                 }
 392
 393                 fi->frag = frag;
 394                 fi->offset = fi->next_offset;
 395                 fi->last_readdir = req;
 396
 397                 if (req->r_did_prepopulate) {
 398                         fi->readdir_cache_idx = req->r_readdir_cache_idx;
 399                         if (fi->readdir_cache_idx < 0) {
 400                                 /* preclude from marking dir ordered */
 401                                 fi->dir_ordered_count = 0;
 402                         } else if (ceph_frag_is_leftmost(frag) && off == 2) {
 403                                 /* note dir version at start of readdir so
 404                                  * we can tell if any dentries get dropped */
 405                                 fi->dir_release_count = req->r_dir_release_cnt;
 406                                 fi->dir_ordered_count = req->r_dir_ordered_cnt;
 407                         }
 408                 } else {
 409                         dout("readdir !did_prepopulate");
 410                         /* disable readdir cache */
 411                         fi->readdir_cache_idx = -1;
 412                         /* preclude from marking dir complete */
 413                         fi->dir_release_count = 0;
 414                 }
 415
 416                 if (req->r_reply_info.dir_end) {
 417                         kfree(fi->last_name);
 418                         fi->last_name = NULL;
 419                         if (ceph_frag_is_rightmost(frag))
 420                                 fi->next_offset = 2;
 421                         else
 422                                 fi->next_offset = 0;
 423                 } else {
 424                         err = note_last_dentry(fi,
 425                                        rinfo->dir_dname[rinfo->dir_nr-1],
 426                                        rinfo->dir_dname_len[rinfo->dir_nr-1],
 427                                        fi->next_offset + rinfo->dir_nr);
 428                         if (err)
 429                                 return err;
 430                 }
 431         }
 432
 433         rinfo = &fi->last_readdir->r_reply_info;
 434         dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
 435              rinfo->dir_nr, off, fi->offset);
 436
 437         ctx->pos = ceph_make_fpos(frag, off);
 438         while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
 439                 struct ceph_mds_reply_inode *in =
 440                         rinfo->dir_in[off - fi->offset].in;
 441                 struct ceph_vino vino;
 442                 ino_t ino;
 443
 444                 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
 445                      off, off - fi->offset, rinfo->dir_nr, ctx->pos,
 446                      rinfo->dir_dname_len[off - fi->offset],
 447                      rinfo->dir_dname[off - fi->offset], in);
 448                 BUG_ON(!in);
 449                 ftype = le32_to_cpu(in->mode) >> 12;
 450                 vino.ino = le64_to_cpu(in->ino);
 451                 vino.snap = le64_to_cpu(in->snapid);
 452                 ino = ceph_vino_to_ino(vino);
 453                 if (!dir_emit(ctx,
 454                             rinfo->dir_dname[off - fi->offset],
 455                             rinfo->dir_dname_len[off - fi->offset],
 456                             ceph_translate_ino(inode->i_sb, ino), ftype)) {
 457                         dout("filldir stopping us...\n");
 458                         return 0;
 459                 }
 460                 off++;
 461                 ctx->pos++;
 462         }
 463
 464         if (fi->last_name) {
 465                 ceph_mdsc_put_request(fi->last_readdir);
 466                 fi->last_readdir = NULL;
 467                 goto more;
 468         }
 469
 470         /* more frags? */
 471         if (!ceph_frag_is_rightmost(frag)) {
 472                 frag = ceph_frag_next(frag);
 473                 off = 0;
 474                 ctx->pos = ceph_make_fpos(frag, off);
 475                 dout("readdir next frag is %x\n", frag);
 476                 goto more;
 477         }
 478         fi->flags |= CEPH_F_ATEND;
 479
 480         /*
 481          * if dir_release_count still matches the dir, no dentries
 482          * were released during the whole readdir, and we should have
 483          * the complete dir contents in our cache.
 484          */
 485         if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
 486                 spin_lock(&ci->i_ceph_lock);
 487                 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
 488                         dout(" marking %p complete and ordered\n", inode);
 489                         /* use i_size to track number of entries in
 490                          * readdir cache */
 491                         BUG_ON(fi->readdir_cache_idx < 0);
 492                         i_size_write(inode, fi->readdir_cache_idx *
 493                                      sizeof(struct dentry*));
 494                 } else {
 495                         dout(" marking %p complete\n", inode);
 496                 }
 497                 __ceph_dir_set_complete(ci, fi->dir_release_count,
 498                                         fi->dir_ordered_count);
 499                 spin_unlock(&ci->i_ceph_lock);
 500         }
 501
 502         dout("readdir %p file %p done.\n", inode, file);
 503         return 0;
 504 }
 505
 506 static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
 507 {
 508         if (fi->last_readdir) {
 509                 ceph_mdsc_put_request(fi->last_readdir);
 510                 fi->last_readdir = NULL;
 511         }
 512         kfree(fi->last_name);
 513         fi->last_name = NULL;
 514         fi->dir_release_count = 0;
 515         fi->readdir_cache_idx = -1;
 516         if (ceph_frag_is_leftmost(frag))
 517                 fi->next_offset = 2;  /* compensate for . and .. */
 518         else
 519                 fi->next_offset = 0;
 520         fi->flags &= ~CEPH_F_ATEND;
 521 }
 522
 523 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
 524 {
 525         struct ceph_file_info *fi = file->private_data;
 526         struct inode *inode = file->f_mapping->host;
 527         loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
 528         loff_t retval;
 529
 530         inode_lock(inode);
 531         retval = -EINVAL;
 532         switch (whence) {
 533         case SEEK_CUR:
 534                 offset += file->f_pos;
 535         case SEEK_SET:
 536                 break;
 537         case SEEK_END:
 538                 retval = -EOPNOTSUPP;
 539         default:
 540                 goto out;
 541         }
 542
 543         if (offset >= 0) {
 544                 if (offset != file->f_pos) {
 545                         file->f_pos = offset;
 546                         file->f_version = 0;
 547                         fi->flags &= ~CEPH_F_ATEND;
 548                 }
 549                 retval = offset;
 550
 551                 if (offset == 0 ||
 552                     fpos_frag(offset) != fi->frag ||
 553                     fpos_off(offset) < fi->offset) {
 554                         /* discard buffered readdir content on seekdir(0), or
 555                          * seek to new frag, or seek prior to current chunk */
 556                         dout("dir_llseek dropping %p content\n", file);
 557                         reset_readdir(fi, fpos_frag(offset));
 558                 } else if (fpos_cmp(offset, old_offset) > 0) {
 559                         /* reset dir_release_count if we did a forward seek */
 560                         fi->dir_release_count = 0;
 561                         fi->readdir_cache_idx = -1;
 562                 }
 563         }
 564 out:
 565         inode_unlock(inode);
 566         return retval;
 567 }
 568
 569 /*
 570  * Handle lookups for the hidden .snap directory.
 571  */
 572 int ceph_handle_snapdir(struct ceph_mds_request *req,
 573                         struct dentry *dentry, int err)
 574 {
 575         struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
 576         struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
 577
 578         /* .snap dir? */
 579         if (err == -ENOENT &&
 580             ceph_snap(parent) == CEPH_NOSNAP &&
 581             strcmp(dentry->d_name.name,
 582                    fsc->mount_options->snapdir_name) == 0) {
 583                 struct inode *inode = ceph_get_snapdir(parent);
 584                 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
 585                      dentry, dentry, inode);
 586                 BUG_ON(!d_unhashed(dentry));
 587                 d_add(dentry, inode);
 588                 err = 0;
 589         }
 590         return err;
 591 }
 592
 593 /*
 594  * Figure out final result of a lookup/open request.
 595  *
 596  * Mainly, make sure we return the final req->r_dentry (if it already
 597  * existed) in place of the original VFS-provided dentry when they
 598  * differ.
 599  *
 600  * Gracefully handle the case where the MDS replies with -ENOENT and
 601  * no trace (which it may do, at its discretion, e.g., if it doesn't
 602  * care to issue a lease on the negative dentry).
 603  */
 604 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
 605                                   struct dentry *dentry, int err)
 606 {
 607         if (err == -ENOENT) {
 608                 /* no trace? */
 609                 err = 0;
 610                 if (!req->r_reply_info.head->is_dentry) {
 611                         dout("ENOENT and no trace, dentry %p inode %p\n",
 612                              dentry, d_inode(dentry));
 613                         if (d_really_is_positive(dentry)) {
 614                                 d_drop(dentry);
 615                                 err = -ENOENT;
 616                         } else {
 617                                 d_add(dentry, NULL);
 618                         }
 619                 }
 620         }
 621         if (err)
 622                 dentry = ERR_PTR(err);
 623         else if (dentry != req->r_dentry)
 624                 dentry = dget(req->r_dentry);   /* we got spliced */
 625         else
 626                 dentry = NULL;
 627         return dentry;
 628 }
 629
 630 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
 631 {
 632         return ceph_ino(inode) == CEPH_INO_ROOT &&
 633                 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
 634 }
 635
 636 /*
 637  * Look up a single dir entry.  If there is a lookup intent, inform
 638  * the MDS so that it gets our 'caps wanted' value in a single op.
 639  */
 640 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
 641                                   unsigned int flags)
 642 {
 643         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 644         struct ceph_mds_client *mdsc = fsc->mdsc;
 645         struct ceph_mds_request *req;
 646         int op;
 647         int mask;
 648         int err;
 649
 650         dout("lookup %p dentry %p '%pd'\n",
 651              dir, dentry, dentry);
 652
 653         if (dentry->d_name.len > NAME_MAX)
 654                 return ERR_PTR(-ENAMETOOLONG);
 655
 656         err = ceph_init_dentry(dentry);
 657         if (err < 0)
 658                 return ERR_PTR(err);
 659
 660         /* can we conclude ENOENT locally? */
 661         if (d_really_is_negative(dentry)) {
 662                 struct ceph_inode_info *ci = ceph_inode(dir);
 663                 struct ceph_dentry_info *di = ceph_dentry(dentry);
 664
 665                 spin_lock(&ci->i_ceph_lock);
 666                 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
 667                 if (strncmp(dentry->d_name.name,
 668                             fsc->mount_options->snapdir_name,
 669                             dentry->d_name.len) &&
 670                     !is_root_ceph_dentry(dir, dentry) &&
 671                     ceph_test_mount_opt(fsc, DCACHE) &&
 672                     __ceph_dir_is_complete(ci) &&
 673                     (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
 674                         spin_unlock(&ci->i_ceph_lock);
 675                         dout(" dir %p complete, -ENOENT\n", dir);
 676                         d_add(dentry, NULL);
 677                         di->lease_shared_gen = ci->i_shared_gen;
 678                         return NULL;
 679                 }
 680                 spin_unlock(&ci->i_ceph_lock);
 681         }
 682
 683         op = ceph_snap(dir) == CEPH_SNAPDIR ?
 684                 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
 685         req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
 686         if (IS_ERR(req))
 687                 return ERR_CAST(req);
 688         req->r_dentry = dget(dentry);
 689         req->r_num_caps = 2;
 690
 691         mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
 692         if (ceph_security_xattr_wanted(dir))
 693                 mask |= CEPH_CAP_XATTR_SHARED;
 694         req->r_args.getattr.mask = cpu_to_le32(mask);
 695
 696         req->r_locked_dir = dir;
 697         err = ceph_mdsc_do_request(mdsc, NULL, req);
 698         err = ceph_handle_snapdir(req, dentry, err);
 699         dentry = ceph_finish_lookup(req, dentry, err);
 700         ceph_mdsc_put_request(req);  /* will dput(dentry) */
 701         dout("lookup result=%p\n", dentry);
 702         return dentry;
 703 }
 704
 705 /*
 706  * If we do a create but get no trace back from the MDS, follow up with
 707  * a lookup (the VFS expects us to link up the provided dentry).
 708  */
 709 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
 710 {
 711         struct dentry *result = ceph_lookup(dir, dentry, 0);
 712
 713         if (result && !IS_ERR(result)) {
 714                 /*
 715                  * We created the item, then did a lookup, and found
 716                  * it was already linked to another inode we already
 717                  * had in our cache (and thus got spliced). To not
 718                  * confuse VFS (especially when inode is a directory),
 719                  * we don't link our dentry to that inode, return an
 720                  * error instead.
 721                  *
 722                  * This event should be rare and it happens only when
 723                  * we talk to old MDS. Recent MDS does not send traceless
 724                  * reply for request that creates new inode.
 725                  */
 726                 d_drop(result);
 727                 return -ESTALE;
 728         }
 729         return PTR_ERR(result);
 730 }
 731
 732 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
 733                       umode_t mode, dev_t rdev)
 734 {
 735         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 736         struct ceph_mds_client *mdsc = fsc->mdsc;
 737         struct ceph_mds_request *req;
 738         struct ceph_acls_info acls = {};
 739         int err;
 740
 741         if (ceph_snap(dir) != CEPH_NOSNAP)
 742                 return -EROFS;
 743
 744         err = ceph_pre_init_acls(dir, &mode, &acls);
 745         if (err < 0)
 746                 return err;
 747
 748         dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
 749              dir, dentry, mode, rdev);
 750         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
 751         if (IS_ERR(req)) {
 752                 err = PTR_ERR(req);
 753                 goto out;
 754         }
 755         req->r_dentry = dget(dentry);
 756         req->r_num_caps = 2;
 757         req->r_locked_dir = dir;
 758         req->r_args.mknod.mode = cpu_to_le32(mode);
 759         req->r_args.mknod.rdev = cpu_to_le32(rdev);
 760         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 761         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 762         if (acls.pagelist) {
 763                 req->r_pagelist = acls.pagelist;
 764                 acls.pagelist = NULL;
 765         }
 766         err = ceph_mdsc_do_request(mdsc, dir, req);
 767         if (!err && !req->r_reply_info.head->is_dentry)
 768                 err = ceph_handle_notrace_create(dir, dentry);
 769         ceph_mdsc_put_request(req);
 770 out:
 771         if (!err)
 772                 ceph_init_inode_acls(d_inode(dentry), &acls);
 773         else
 774                 d_drop(dentry);
 775         ceph_release_acls_info(&acls);
 776         return err;
 777 }
 778
 779 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 780                        bool excl)
 781 {
 782         return ceph_mknod(dir, dentry, mode, 0);
 783 }
 784
 785 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
 786                             const char *dest)
 787 {
 788         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 789         struct ceph_mds_client *mdsc = fsc->mdsc;
 790         struct ceph_mds_request *req;
 791         int err;
 792
 793         if (ceph_snap(dir) != CEPH_NOSNAP)
 794                 return -EROFS;
 795
 796         dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
 797         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
 798         if (IS_ERR(req)) {
 799                 err = PTR_ERR(req);
 800                 goto out;
 801         }
 802         req->r_path2 = kstrdup(dest, GFP_KERNEL);
 803         if (!req->r_path2) {
 804                 err = -ENOMEM;
 805                 ceph_mdsc_put_request(req);
 806                 goto out;
 807         }
 808         req->r_locked_dir = dir;
 809         req->r_dentry = dget(dentry);
 810         req->r_num_caps = 2;
 811         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 812         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 813         err = ceph_mdsc_do_request(mdsc, dir, req);
 814         if (!err && !req->r_reply_info.head->is_dentry)
 815                 err = ceph_handle_notrace_create(dir, dentry);
 816         ceph_mdsc_put_request(req);
 817 out:
 818         if (err)
 819                 d_drop(dentry);
 820         return err;
 821 }
 822
 823 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 824 {
 825         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 826         struct ceph_mds_client *mdsc = fsc->mdsc;
 827         struct ceph_mds_request *req;
 828         struct ceph_acls_info acls = {};
 829         int err = -EROFS;
 830         int op;
 831
 832         if (ceph_snap(dir) == CEPH_SNAPDIR) {
 833                 /* mkdir .snap/foo is a MKSNAP */
 834                 op = CEPH_MDS_OP_MKSNAP;
 835                 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
 836                      dentry, dentry);
 837         } else if (ceph_snap(dir) == CEPH_NOSNAP) {
 838                 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
 839                 op = CEPH_MDS_OP_MKDIR;
 840         } else {
 841                 goto out;
 842         }
 843
 844         mode |= S_IFDIR;
 845         err = ceph_pre_init_acls(dir, &mode, &acls);
 846         if (err < 0)
 847                 goto out;
 848
 849         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 850         if (IS_ERR(req)) {
 851                 err = PTR_ERR(req);
 852                 goto out;
 853         }
 854
 855         req->r_dentry = dget(dentry);
 856         req->r_num_caps = 2;
 857         req->r_locked_dir = dir;
 858         req->r_args.mkdir.mode = cpu_to_le32(mode);
 859         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 860         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 861         if (acls.pagelist) {
 862                 req->r_pagelist = acls.pagelist;
 863                 acls.pagelist = NULL;
 864         }
 865         err = ceph_mdsc_do_request(mdsc, dir, req);
 866         if (!err &&
 867             !req->r_reply_info.head->is_target &&
 868             !req->r_reply_info.head->is_dentry)
 869                 err = ceph_handle_notrace_create(dir, dentry);
 870         ceph_mdsc_put_request(req);
 871 out:
 872         if (!err)
 873                 ceph_init_inode_acls(d_inode(dentry), &acls);
 874         else
 875                 d_drop(dentry);
 876         ceph_release_acls_info(&acls);
 877         return err;
 878 }
 879
 880 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
 881                      struct dentry *dentry)
 882 {
 883         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 884         struct ceph_mds_client *mdsc = fsc->mdsc;
 885         struct ceph_mds_request *req;
 886         int err;
 887
 888         if (ceph_snap(dir) != CEPH_NOSNAP)
 889                 return -EROFS;
 890
 891         dout("link in dir %p old_dentry %p dentry %p\n", dir,
 892              old_dentry, dentry);
 893         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
 894         if (IS_ERR(req)) {
 895                 d_drop(dentry);
 896                 return PTR_ERR(req);
 897         }
 898         req->r_dentry = dget(dentry);
 899         req->r_num_caps = 2;
 900         req->r_old_dentry = dget(old_dentry);
 901         req->r_locked_dir = dir;
 902         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 903         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 904         /* release LINK_SHARED on source inode (mds will lock it) */
 905         req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
 906         err = ceph_mdsc_do_request(mdsc, dir, req);
 907         if (err) {
 908                 d_drop(dentry);
 909         } else if (!req->r_reply_info.head->is_dentry) {
 910                 ihold(d_inode(old_dentry));
 911                 d_instantiate(dentry, d_inode(old_dentry));
 912         }
 913         ceph_mdsc_put_request(req);
 914         return err;
 915 }
 916
 917 /*
 918  * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps.  If it
 919  * looks like the link count will hit 0, drop any other caps (other
 920  * than PIN) we don't specifically want (due to the file still being
 921  * open).
 922  */
 923 static int drop_caps_for_unlink(struct inode *inode)
 924 {
 925         struct ceph_inode_info *ci = ceph_inode(inode);
 926         int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
 927
 928         spin_lock(&ci->i_ceph_lock);
 929         if (inode->i_nlink == 1) {
 930                 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
 931                 ci->i_ceph_flags |= CEPH_I_NODELAY;
 932         }
 933         spin_unlock(&ci->i_ceph_lock);
 934         return drop;
 935 }
 936
 937 /*
 938  * rmdir and unlink are differ only by the metadata op code
 939  */
 940 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
 941 {
 942         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 943         struct ceph_mds_client *mdsc = fsc->mdsc;
 944         struct inode *inode = d_inode(dentry);
 945         struct ceph_mds_request *req;
 946         int err = -EROFS;
 947         int op;
 948
 949         if (ceph_snap(dir) == CEPH_SNAPDIR) {
 950                 /* rmdir .snap/foo is RMSNAP */
 951                 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
 952                 op = CEPH_MDS_OP_RMSNAP;
 953         } else if (ceph_snap(dir) == CEPH_NOSNAP) {
 954                 dout("unlink/rmdir dir %p dn %p inode %p\n",
 955                      dir, dentry, inode);
 956                 op = d_is_dir(dentry) ?
 957                         CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
 958         } else
 959                 goto out;
 960         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 961         if (IS_ERR(req)) {
 962                 err = PTR_ERR(req);
 963                 goto out;
 964         }
 965         req->r_dentry = dget(dentry);
 966         req->r_num_caps = 2;
 967         req->r_locked_dir = dir;
 968         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 969         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 970         req->r_inode_drop = drop_caps_for_unlink(inode);
 971         err = ceph_mdsc_do_request(mdsc, dir, req);
 972         if (!err && !req->r_reply_info.head->is_dentry)
 973                 d_delete(dentry);
 974         ceph_mdsc_put_request(req);
 975 out:
 976         return err;
 977 }
 978
 979 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
 980                        struct inode *new_dir, struct dentry *new_dentry)
 981 {
 982         struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
 983         struct ceph_mds_client *mdsc = fsc->mdsc;
 984         struct ceph_mds_request *req;
 985         int op = CEPH_MDS_OP_RENAME;
 986         int err;
 987
 988         if (ceph_snap(old_dir) != ceph_snap(new_dir))
 989                 return -EXDEV;
 990         if (ceph_snap(old_dir) != CEPH_NOSNAP) {
 991                 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
 992                         op = CEPH_MDS_OP_RENAMESNAP;
 993                 else
 994                         return -EROFS;
 995         }
 996         dout("rename dir %p dentry %p to dir %p dentry %p\n",
 997              old_dir, old_dentry, new_dir, new_dentry);
 998         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 999         if (IS_ERR(req))
1000                 return PTR_ERR(req);
1001         ihold(old_dir);
1002         req->r_dentry = dget(new_dentry);
1003         req->r_num_caps = 2;
1004         req->r_old_dentry = dget(old_dentry);
1005         req->r_old_dentry_dir = old_dir;
1006         req->r_locked_dir = new_dir;
1007         req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1008         req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1009         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1010         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1011         /* release LINK_RDCACHE on source inode (mds will lock it) */
1012         req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
1013         if (d_really_is_positive(new_dentry))
1014                 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
1015         err = ceph_mdsc_do_request(mdsc, old_dir, req);
1016         if (!err && !req->r_reply_info.head->is_dentry) {
1017                 /*
1018                  * Normally d_move() is done by fill_trace (called by
1019                  * do_request, above).  If there is no trace, we need
1020                  * to do it here.
1021                  */
1022
1023                 /* d_move screws up sibling dentries' offsets */
1024                 ceph_dir_clear_complete(old_dir);
1025                 ceph_dir_clear_complete(new_dir);
1026
1027                 d_move(old_dentry, new_dentry);
1028
1029                 /* ensure target dentry is invalidated, despite
1030                    rehashing bug in vfs_rename_dir */
1031                 ceph_invalidate_dentry_lease(new_dentry);
1032         }
1033         ceph_mdsc_put_request(req);
1034         return err;
1035 }
1036
1037 /*
1038  * Ensure a dentry lease will no longer revalidate.
1039  */
1040 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1041 {
1042         spin_lock(&dentry->d_lock);
1043         dentry->d_time = jiffies;
1044         ceph_dentry(dentry)->lease_shared_gen = 0;
1045         spin_unlock(&dentry->d_lock);
1046 }
1047
1048 /*
1049  * Check if dentry lease is valid.  If not, delete the lease.  Try to
1050  * renew if the least is more than half up.
1051  */
1052 static int dentry_lease_is_valid(struct dentry *dentry)
1053 {
1054         struct ceph_dentry_info *di;
1055         struct ceph_mds_session *s;
1056         int valid = 0;
1057         u32 gen;
1058         unsigned long ttl;
1059         struct ceph_mds_session *session = NULL;
1060         struct inode *dir = NULL;
1061         u32 seq = 0;
1062
1063         spin_lock(&dentry->d_lock);
1064         di = ceph_dentry(dentry);
1065         if (di->lease_session) {
1066                 s = di->lease_session;
1067                 spin_lock(&s->s_gen_ttl_lock);
1068                 gen = s->s_cap_gen;
1069                 ttl = s->s_cap_ttl;
1070                 spin_unlock(&s->s_gen_ttl_lock);
1071
1072                 if (di->lease_gen == gen &&
1073                     time_before(jiffies, dentry->d_time) &&
1074                     time_before(jiffies, ttl)) {
1075                         valid = 1;
1076                         if (di->lease_renew_after &&
1077                             time_after(jiffies, di->lease_renew_after)) {
1078                                 /* we should renew */
1079                                 dir = d_inode(dentry->d_parent);
1080                                 session = ceph_get_mds_session(s);
1081                                 seq = di->lease_seq;
1082                                 di->lease_renew_after = 0;
1083                                 di->lease_renew_from = jiffies;
1084                         }
1085                 }
1086         }
1087         spin_unlock(&dentry->d_lock);
1088
1089         if (session) {
1090                 ceph_mdsc_lease_send_msg(session, dir, dentry,
1091                                          CEPH_MDS_LEASE_RENEW, seq);
1092                 ceph_put_mds_session(session);
1093         }
1094         dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1095         return valid;
1096 }
1097
1098 /*
1099  * Check if directory-wide content lease/cap is valid.
1100  */
1101 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1102 {
1103         struct ceph_inode_info *ci = ceph_inode(dir);
1104         struct ceph_dentry_info *di = ceph_dentry(dentry);
1105         int valid = 0;
1106
1107         spin_lock(&ci->i_ceph_lock);
1108         if (ci->i_shared_gen == di->lease_shared_gen)
1109                 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1110         spin_unlock(&ci->i_ceph_lock);
1111         dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1112              dir, (unsigned)ci->i_shared_gen, dentry,
1113              (unsigned)di->lease_shared_gen, valid);
1114         return valid;
1115 }
1116
1117 /*
1118  * Check if cached dentry can be trusted.
1119  */
1120 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1121 {
1122         int valid = 0;
1123         struct dentry *parent;
1124         struct inode *dir;
1125
1126         if (flags & LOOKUP_RCU)
1127                 return -ECHILD;
1128
1129         dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1130              dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1131
1132         parent = dget_parent(dentry);
1133         dir = d_inode(parent);
1134
1135         /* always trust cached snapped dentries, snapdir dentry */
1136         if (ceph_snap(dir) != CEPH_NOSNAP) {
1137                 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1138                      dentry, d_inode(dentry));
1139                 valid = 1;
1140         } else if (d_really_is_positive(dentry) &&
1141                    ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1142                 valid = 1;
1143         } else if (dentry_lease_is_valid(dentry) ||
1144                    dir_lease_is_valid(dir, dentry)) {
1145                 if (d_really_is_positive(dentry))
1146                         valid = ceph_is_any_caps(d_inode(dentry));
1147                 else
1148                         valid = 1;
1149         }
1150
1151         if (!valid) {
1152                 struct ceph_mds_client *mdsc =
1153                         ceph_sb_to_client(dir->i_sb)->mdsc;
1154                 struct ceph_mds_request *req;
1155                 int op, mask, err;
1156
1157                 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1158                         CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1159                 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1160                 if (!IS_ERR(req)) {
1161                         req->r_dentry = dget(dentry);
1162                         req->r_num_caps = 2;
1163
1164                         mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1165                         if (ceph_security_xattr_wanted(dir))
1166                                 mask |= CEPH_CAP_XATTR_SHARED;
1167                         req->r_args.getattr.mask = mask;
1168
1169                         req->r_locked_dir = dir;
1170                         err = ceph_mdsc_do_request(mdsc, NULL, req);
1171                         if (err == 0 || err == -ENOENT) {
1172                                 if (dentry == req->r_dentry) {
1173                                         valid = !d_unhashed(dentry);
1174                                 } else {
1175                                         d_invalidate(req->r_dentry);
1176                                         err = -EAGAIN;
1177                                 }
1178                         }
1179                         ceph_mdsc_put_request(req);
1180                         dout("d_revalidate %p lookup result=%d\n",
1181                              dentry, err);
1182                 }
1183         }
1184
1185         dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1186         if (valid) {
1187                 ceph_dentry_lru_touch(dentry);
1188         } else {
1189                 ceph_dir_clear_complete(dir);
1190         }
1191
1192         dput(parent);
1193         return valid;
1194 }
1195
1196 /*
1197  * Release our ceph_dentry_info.
1198  */
1199 static void ceph_d_release(struct dentry *dentry)
1200 {
1201         struct ceph_dentry_info *di = ceph_dentry(dentry);
1202
1203         dout("d_release %p\n", dentry);
1204         ceph_dentry_lru_del(dentry);
1205         if (di->lease_session)
1206                 ceph_put_mds_session(di->lease_session);
1207         kmem_cache_free(ceph_dentry_cachep, di);
1208         dentry->d_fsdata = NULL;
1209 }
1210
1211 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1212                                           unsigned int flags)
1213 {
1214         /*
1215          * Eventually, we'll want to revalidate snapped metadata
1216          * too... probably...
1217          */
1218         return 1;
1219 }
1220
1221 /*
1222  * When the VFS prunes a dentry from the cache, we need to clear the
1223  * complete flag on the parent directory.
1224  *
1225  * Called under dentry->d_lock.
1226  */
1227 static void ceph_d_prune(struct dentry *dentry)
1228 {
1229         dout("ceph_d_prune %p\n", dentry);
1230
1231         /* do we have a valid parent? */
1232         if (IS_ROOT(dentry))
1233                 return;
1234
1235         /* if we are not hashed, we don't affect dir's completeness */
1236         if (d_unhashed(dentry))
1237                 return;
1238
1239         /*
1240          * we hold d_lock, so d_parent is stable, and d_fsdata is never
1241          * cleared until d_release
1242          */
1243         ceph_dir_clear_complete(d_inode(dentry->d_parent));
1244 }
1245
1246 /*
1247  * read() on a dir.  This weird interface hack only works if mounted
1248  * with '-o dirstat'.
1249  */
1250 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1251                              loff_t *ppos)
1252 {
1253         struct ceph_file_info *cf = file->private_data;
1254         struct inode *inode = file_inode(file);
1255         struct ceph_inode_info *ci = ceph_inode(inode);
1256         int left;
1257         const int bufsize = 1024;
1258
1259         if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1260                 return -EISDIR;
1261
1262         if (!cf->dir_info) {
1263                 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1264                 if (!cf->dir_info)
1265                         return -ENOMEM;
1266                 cf->dir_info_len =
1267                         snprintf(cf->dir_info, bufsize,
1268                                 "entries:   %20lld\n"
1269                                 " files:    %20lld\n"
1270                                 " subdirs:  %20lld\n"
1271                                 "rentries:  %20lld\n"
1272                                 " rfiles:   %20lld\n"
1273                                 " rsubdirs: %20lld\n"
1274                                 "rbytes:    %20lld\n"
1275                                 "rctime:    %10ld.%09ld\n",
1276                                 ci->i_files + ci->i_subdirs,
1277                                 ci->i_files,
1278                                 ci->i_subdirs,
1279                                 ci->i_rfiles + ci->i_rsubdirs,
1280                                 ci->i_rfiles,
1281                                 ci->i_rsubdirs,
1282                                 ci->i_rbytes,
1283                                 (long)ci->i_rctime.tv_sec,
1284                                 (long)ci->i_rctime.tv_nsec);
1285         }
1286
1287         if (*ppos >= cf->dir_info_len)
1288                 return 0;
1289         size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1290         left = copy_to_user(buf, cf->dir_info + *ppos, size);
1291         if (left == size)
1292                 return -EFAULT;
1293         *ppos += (size - left);
1294         return size - left;
1295 }
1296
1297 /*
1298  * We maintain a private dentry LRU.
1299  *
1300  * FIXME: this needs to be changed to a per-mds lru to be useful.
1301  */
1302 void ceph_dentry_lru_add(struct dentry *dn)
1303 {
1304         struct ceph_dentry_info *di = ceph_dentry(dn);
1305         struct ceph_mds_client *mdsc;
1306
1307         dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1308         mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1309         spin_lock(&mdsc->dentry_lru_lock);
1310         list_add_tail(&di->lru, &mdsc->dentry_lru);
1311         mdsc->num_dentry++;
1312         spin_unlock(&mdsc->dentry_lru_lock);
1313 }
1314
1315 void ceph_dentry_lru_touch(struct dentry *dn)
1316 {
1317         struct ceph_dentry_info *di = ceph_dentry(dn);
1318         struct ceph_mds_client *mdsc;
1319
1320         dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1321              di->offset);
1322         mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1323         spin_lock(&mdsc->dentry_lru_lock);
1324         list_move_tail(&di->lru, &mdsc->dentry_lru);
1325         spin_unlock(&mdsc->dentry_lru_lock);
1326 }
1327
1328 void ceph_dentry_lru_del(struct dentry *dn)
1329 {
1330         struct ceph_dentry_info *di = ceph_dentry(dn);
1331         struct ceph_mds_client *mdsc;
1332
1333         dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1334         mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1335         spin_lock(&mdsc->dentry_lru_lock);
1336         list_del_init(&di->lru);
1337         mdsc->num_dentry--;
1338         spin_unlock(&mdsc->dentry_lru_lock);
1339 }
1340
1341 /*
1342  * Return name hash for a given dentry.  This is dependent on
1343  * the parent directory's hash function.
1344  */
1345 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1346 {
1347         struct ceph_inode_info *dci = ceph_inode(dir);
1348
1349         switch (dci->i_dir_layout.dl_dir_hash) {
1350         case 0: /* for backward compat */
1351         case CEPH_STR_HASH_LINUX:
1352                 return dn->d_name.hash;
1353
1354         default:
1355                 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1356                                      dn->d_name.name, dn->d_name.len);
1357         }
1358 }
1359
1360 const struct file_operations ceph_dir_fops = {
1361         .read = ceph_read_dir,
1362         .iterate = ceph_readdir,
1363         .llseek = ceph_dir_llseek,
1364         .open = ceph_open,
1365         .release = ceph_release,
1366         .unlocked_ioctl = ceph_ioctl,
1367         .fsync = ceph_fsync,
1368 };
1369
1370 const struct file_operations ceph_snapdir_fops = {
1371         .iterate = ceph_readdir,
1372         .llseek = ceph_dir_llseek,
1373         .open = ceph_open,
1374         .release = ceph_release,
1375 };
1376
1377 const struct inode_operations ceph_dir_iops = {
1378         .lookup = ceph_lookup,
1379         .permission = ceph_permission,
1380         .getattr = ceph_getattr,
1381         .setattr = ceph_setattr,
1382         .setxattr = ceph_setxattr,
1383         .getxattr = ceph_getxattr,
1384         .listxattr = ceph_listxattr,
1385         .removexattr = ceph_removexattr,
1386         .get_acl = ceph_get_acl,
1387         .set_acl = ceph_set_acl,
1388         .mknod = ceph_mknod,
1389         .symlink = ceph_symlink,
1390         .mkdir = ceph_mkdir,
1391         .link = ceph_link,
1392         .unlink = ceph_unlink,
1393         .rmdir = ceph_unlink,
1394         .rename = ceph_rename,
1395         .create = ceph_create,
1396         .atomic_open = ceph_atomic_open,
1397 };
1398
1399 const struct inode_operations ceph_snapdir_iops = {
1400         .lookup = ceph_lookup,
1401         .permission = ceph_permission,
1402         .getattr = ceph_getattr,
1403         .mkdir = ceph_mkdir,
1404         .rmdir = ceph_unlink,
1405         .rename = ceph_rename,
1406 };
1407
1408 const struct dentry_operations ceph_dentry_ops = {
1409         .d_revalidate = ceph_d_revalidate,
1410         .d_release = ceph_d_release,
1411         .d_prune = ceph_d_prune,
1412 };
1413
1414 const struct dentry_operations ceph_snapdir_dentry_ops = {
1415         .d_revalidate = ceph_snapdir_d_revalidate,
1416         .d_release = ceph_d_release,
1417 };
1418
1419 const struct dentry_operations ceph_snap_dentry_ops = {
1420         .d_release = ceph_d_release,
1421         .d_prune = ceph_d_prune,
1422 };