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 #include <linux/xattr.h>
   9
  10 #include "super.h"
  11 #include "mds_client.h"
  12
  13 /*
  14  * Directory operations: readdir, lookup, create, link, unlink,
  15  * rename, etc.
  16  */
  17
  18 /*
  19  * Ceph MDS operations are specified in terms of a base ino and
  20  * relative path.  Thus, the client can specify an operation on a
  21  * specific inode (e.g., a getattr due to fstat(2)), or as a path
  22  * relative to, say, the root directory.
  23  *
  24  * Normally, we limit ourselves to strict inode ops (no path component)
  25  * or dentry operations (a single path component relative to an ino).  The
  26  * exception to this is open_root_dentry(), which will open the mount
  27  * point by name.
  28  */
  29
  30 const struct dentry_operations ceph_dentry_ops;
  31
  32 /*
  33  * Initialize ceph dentry state.
  34  */
  35 static int ceph_d_init(struct dentry *dentry)
  36 {
  37         struct ceph_dentry_info *di;
  38
  39         di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
  40         if (!di)
  41                 return -ENOMEM;          /* oh well */
  42
  43         di->dentry = dentry;
  44         di->lease_session = NULL;
  45         di->time = jiffies;
  46         dentry->d_fsdata = di;
  47         ceph_dentry_lru_add(dentry);
  48         return 0;
  49 }
  50
  51 /*
  52  * for f_pos for readdir:
  53  * - hash order:
  54  *      (0xff << 52) | ((24 bits hash) << 28) |
  55  *      (the nth entry has hash collision);
  56  * - frag+name order;
  57  *      ((frag value) << 28) | (the nth entry in frag);
  58  */
  59 #define OFFSET_BITS     28
  60 #define OFFSET_MASK     ((1 << OFFSET_BITS) - 1)
  61 #define HASH_ORDER      (0xffull << (OFFSET_BITS + 24))
  62 loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
  63 {
  64         loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
  65         if (hash_order)
  66                 fpos |= HASH_ORDER;
  67         return fpos;
  68 }
  69
  70 static bool is_hash_order(loff_t p)
  71 {
  72         return (p & HASH_ORDER) == HASH_ORDER;
  73 }
  74
  75 static unsigned fpos_frag(loff_t p)
  76 {
  77         return p >> OFFSET_BITS;
  78 }
  79
  80 static unsigned fpos_hash(loff_t p)
  81 {
  82         return ceph_frag_value(fpos_frag(p));
  83 }
  84
  85 static unsigned fpos_off(loff_t p)
  86 {
  87         return p & OFFSET_MASK;
  88 }
  89
  90 static int fpos_cmp(loff_t l, loff_t r)
  91 {
  92         int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
  93         if (v)
  94                 return v;
  95         return (int)(fpos_off(l) - fpos_off(r));
  96 }
  97
  98 /*
  99  * make note of the last dentry we read, so we can
 100  * continue at the same lexicographical point,
 101  * regardless of what dir changes take place on the
 102  * server.
 103  */
 104 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
 105                             int len, unsigned next_offset)
 106 {
 107         char *buf = kmalloc(len+1, GFP_KERNEL);
 108         if (!buf)
 109                 return -ENOMEM;
 110         kfree(fi->last_name);
 111         fi->last_name = buf;
 112         memcpy(fi->last_name, name, len);
 113         fi->last_name[len] = 0;
 114         fi->next_offset = next_offset;
 115         dout("note_last_dentry '%s'\n", fi->last_name);
 116         return 0;
 117 }
 118
 119
 120 static struct dentry *
 121 __dcache_find_get_entry(struct dentry *parent, u64 idx,
 122                         struct ceph_readdir_cache_control *cache_ctl)
 123 {
 124         struct inode *dir = d_inode(parent);
 125         struct dentry *dentry;
 126         unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
 127         loff_t ptr_pos = idx * sizeof(struct dentry *);
 128         pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
 129
 130         if (ptr_pos >= i_size_read(dir))
 131                 return NULL;
 132
 133         if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
 134                 ceph_readdir_cache_release(cache_ctl);
 135                 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
 136                 if (!cache_ctl->page) {
 137                         dout(" page %lu not found\n", ptr_pgoff);
 138                         return ERR_PTR(-EAGAIN);
 139                 }
 140                 /* reading/filling the cache are serialized by
 141                    i_mutex, no need to use page lock */
 142                 unlock_page(cache_ctl->page);
 143                 cache_ctl->dentries = kmap(cache_ctl->page);
 144         }
 145
 146         cache_ctl->index = idx & idx_mask;
 147
 148         rcu_read_lock();
 149         spin_lock(&parent->d_lock);
 150         /* check i_size again here, because empty directory can be
 151          * marked as complete while not holding the i_mutex. */
 152         if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
 153                 dentry = cache_ctl->dentries[cache_ctl->index];
 154         else
 155                 dentry = NULL;
 156         spin_unlock(&parent->d_lock);
 157         if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
 158                 dentry = NULL;
 159         rcu_read_unlock();
 160         return dentry ? : ERR_PTR(-EAGAIN);
 161 }
 162
 163 /*
 164  * When possible, we try to satisfy a readdir by peeking at the
 165  * dcache.  We make this work by carefully ordering dentries on
 166  * d_child when we initially get results back from the MDS, and
 167  * falling back to a "normal" sync readdir if any dentries in the dir
 168  * are dropped.
 169  *
 170  * Complete dir indicates that we have all dentries in the dir.  It is
 171  * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
 172  * the MDS if/when the directory is modified).
 173  */
 174 static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
 175                             u32 shared_gen)
 176 {
 177         struct ceph_file_info *fi = file->private_data;
 178         struct dentry *parent = file->f_path.dentry;
 179         struct inode *dir = d_inode(parent);
 180         struct dentry *dentry, *last = NULL;
 181         struct ceph_dentry_info *di;
 182         struct ceph_readdir_cache_control cache_ctl = {};
 183         u64 idx = 0;
 184         int err = 0;
 185
 186         dout("__dcache_readdir %p v%u at %llx\n", dir, shared_gen, ctx->pos);
 187
 188         /* search start position */
 189         if (ctx->pos > 2) {
 190                 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
 191                 while (count > 0) {
 192                         u64 step = count >> 1;
 193                         dentry = __dcache_find_get_entry(parent, idx + step,
 194                                                          &cache_ctl);
 195                         if (!dentry) {
 196                                 /* use linar search */
 197                                 idx = 0;
 198                                 break;
 199                         }
 200                         if (IS_ERR(dentry)) {
 201                                 err = PTR_ERR(dentry);
 202                                 goto out;
 203                         }
 204                         di = ceph_dentry(dentry);
 205                         spin_lock(&dentry->d_lock);
 206                         if (fpos_cmp(di->offset, ctx->pos) < 0) {
 207                                 idx += step + 1;
 208                                 count -= step + 1;
 209                         } else {
 210                                 count = step;
 211                         }
 212                         spin_unlock(&dentry->d_lock);
 213                         dput(dentry);
 214                 }
 215
 216                 dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
 217         }
 218
 219
 220         for (;;) {
 221                 bool emit_dentry = false;
 222                 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
 223                 if (!dentry) {
 224                         fi->flags |= CEPH_F_ATEND;
 225                         err = 0;
 226                         break;
 227                 }
 228                 if (IS_ERR(dentry)) {
 229                         err = PTR_ERR(dentry);
 230                         goto out;
 231                 }
 232
 233                 di = ceph_dentry(dentry);
 234                 spin_lock(&dentry->d_lock);
 235                 if (di->lease_shared_gen == shared_gen &&
 236                     d_really_is_positive(dentry) &&
 237                     fpos_cmp(ctx->pos, di->offset) <= 0) {
 238                         emit_dentry = true;
 239                 }
 240                 spin_unlock(&dentry->d_lock);
 241
 242                 if (emit_dentry) {
 243                         dout(" %llx dentry %p %pd %p\n", di->offset,
 244                              dentry, dentry, d_inode(dentry));
 245                         ctx->pos = di->offset;
 246                         if (!dir_emit(ctx, dentry->d_name.name,
 247                                       dentry->d_name.len,
 248                                       ceph_translate_ino(dentry->d_sb,
 249                                                          d_inode(dentry)->i_ino),
 250                                       d_inode(dentry)->i_mode >> 12)) {
 251                                 dput(dentry);
 252                                 err = 0;
 253                                 break;
 254                         }
 255                         ctx->pos++;
 256
 257                         if (last)
 258                                 dput(last);
 259                         last = dentry;
 260                 } else {
 261                         dput(dentry);
 262                 }
 263         }
 264 out:
 265         ceph_readdir_cache_release(&cache_ctl);
 266         if (last) {
 267                 int ret;
 268                 di = ceph_dentry(last);
 269                 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
 270                                        fpos_off(di->offset) + 1);
 271                 if (ret < 0)
 272                         err = ret;
 273                 dput(last);
 274         }
 275         return err;
 276 }
 277
 278 static bool need_send_readdir(struct ceph_file_info *fi, loff_t pos)
 279 {
 280         if (!fi->last_readdir)
 281                 return true;
 282         if (is_hash_order(pos))
 283                 return !ceph_frag_contains_value(fi->frag, fpos_hash(pos));
 284         else
 285                 return fi->frag != fpos_frag(pos);
 286 }
 287
 288 static int ceph_readdir(struct file *file, struct dir_context *ctx)
 289 {
 290         struct ceph_file_info *fi = file->private_data;
 291         struct inode *inode = file_inode(file);
 292         struct ceph_inode_info *ci = ceph_inode(inode);
 293         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 294         struct ceph_mds_client *mdsc = fsc->mdsc;
 295         int i;
 296         int err;
 297         u32 ftype;
 298         struct ceph_mds_reply_info_parsed *rinfo;
 299
 300         dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
 301         if (fi->flags & CEPH_F_ATEND)
 302                 return 0;
 303
 304         /* always start with . and .. */
 305         if (ctx->pos == 0) {
 306                 dout("readdir off 0 -> '.'\n");
 307                 if (!dir_emit(ctx, ".", 1,
 308                             ceph_translate_ino(inode->i_sb, inode->i_ino),
 309                             inode->i_mode >> 12))
 310                         return 0;
 311                 ctx->pos = 1;
 312         }
 313         if (ctx->pos == 1) {
 314                 ino_t ino = parent_ino(file->f_path.dentry);
 315                 dout("readdir off 1 -> '..'\n");
 316                 if (!dir_emit(ctx, "..", 2,
 317                             ceph_translate_ino(inode->i_sb, ino),
 318                             inode->i_mode >> 12))
 319                         return 0;
 320                 ctx->pos = 2;
 321         }
 322
 323         /* can we use the dcache? */
 324         spin_lock(&ci->i_ceph_lock);
 325         if (ceph_test_mount_opt(fsc, DCACHE) &&
 326             !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
 327             ceph_snap(inode) != CEPH_SNAPDIR &&
 328             __ceph_dir_is_complete_ordered(ci) &&
 329             __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
 330                 u32 shared_gen = ci->i_shared_gen;
 331                 spin_unlock(&ci->i_ceph_lock);
 332                 err = __dcache_readdir(file, ctx, shared_gen);
 333                 if (err != -EAGAIN)
 334                         return err;
 335         } else {
 336                 spin_unlock(&ci->i_ceph_lock);
 337         }
 338
 339         /* proceed with a normal readdir */
 340 more:
 341         /* do we have the correct frag content buffered? */
 342         if (need_send_readdir(fi, ctx->pos)) {
 343                 struct ceph_mds_request *req;
 344                 unsigned frag;
 345                 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
 346                         CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
 347
 348                 /* discard old result, if any */
 349                 if (fi->last_readdir) {
 350                         ceph_mdsc_put_request(fi->last_readdir);
 351                         fi->last_readdir = NULL;
 352                 }
 353
 354                 if (is_hash_order(ctx->pos)) {
 355                         frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
 356                                                 NULL, NULL);
 357                 } else {
 358                         frag = fpos_frag(ctx->pos);
 359                 }
 360
 361                 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
 362                      ceph_vinop(inode), frag, fi->last_name);
 363                 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 364                 if (IS_ERR(req))
 365                         return PTR_ERR(req);
 366                 err = ceph_alloc_readdir_reply_buffer(req, inode);
 367                 if (err) {
 368                         ceph_mdsc_put_request(req);
 369                         return err;
 370                 }
 371                 /* hints to request -> mds selection code */
 372                 req->r_direct_mode = USE_AUTH_MDS;
 373                 req->r_direct_hash = ceph_frag_value(frag);
 374                 req->r_direct_is_hash = true;
 375                 if (fi->last_name) {
 376                         req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
 377                         if (!req->r_path2) {
 378                                 ceph_mdsc_put_request(req);
 379                                 return -ENOMEM;
 380                         }
 381                 }
 382                 req->r_dir_release_cnt = fi->dir_release_count;
 383                 req->r_dir_ordered_cnt = fi->dir_ordered_count;
 384                 req->r_readdir_cache_idx = fi->readdir_cache_idx;
 385                 req->r_readdir_offset = fi->next_offset;
 386                 req->r_args.readdir.frag = cpu_to_le32(frag);
 387                 req->r_args.readdir.flags =
 388                                 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
 389
 390                 req->r_inode = inode;
 391                 ihold(inode);
 392                 req->r_dentry = dget(file->f_path.dentry);
 393                 err = ceph_mdsc_do_request(mdsc, NULL, req);
 394                 if (err < 0) {
 395                         ceph_mdsc_put_request(req);
 396                         return err;
 397                 }
 398                 dout("readdir got and parsed readdir result=%d on "
 399                      "frag %x, end=%d, complete=%d, hash_order=%d\n",
 400                      err, frag,
 401                      (int)req->r_reply_info.dir_end,
 402                      (int)req->r_reply_info.dir_complete,
 403                      (int)req->r_reply_info.hash_order);
 404
 405                 rinfo = &req->r_reply_info;
 406                 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
 407                         frag = le32_to_cpu(rinfo->dir_dir->frag);
 408                         if (!rinfo->hash_order) {
 409                                 fi->next_offset = req->r_readdir_offset;
 410                                 /* adjust ctx->pos to beginning of frag */
 411                                 ctx->pos = ceph_make_fpos(frag,
 412                                                           fi->next_offset,
 413                                                           false);
 414                         }
 415                 }
 416
 417                 fi->frag = frag;
 418                 fi->last_readdir = req;
 419
 420                 if (req->r_did_prepopulate) {
 421                         fi->readdir_cache_idx = req->r_readdir_cache_idx;
 422                         if (fi->readdir_cache_idx < 0) {
 423                                 /* preclude from marking dir ordered */
 424                                 fi->dir_ordered_count = 0;
 425                         } else if (ceph_frag_is_leftmost(frag) &&
 426                                    fi->next_offset == 2) {
 427                                 /* note dir version at start of readdir so
 428                                  * we can tell if any dentries get dropped */
 429                                 fi->dir_release_count = req->r_dir_release_cnt;
 430                                 fi->dir_ordered_count = req->r_dir_ordered_cnt;
 431                         }
 432                 } else {
 433                         dout("readdir !did_prepopulate");
 434                         /* disable readdir cache */
 435                         fi->readdir_cache_idx = -1;
 436                         /* preclude from marking dir complete */
 437                         fi->dir_release_count = 0;
 438                 }
 439
 440                 /* note next offset and last dentry name */
 441                 if (rinfo->dir_nr > 0) {
 442                         struct ceph_mds_reply_dir_entry *rde =
 443                                         rinfo->dir_entries + (rinfo->dir_nr-1);
 444                         unsigned next_offset = req->r_reply_info.dir_end ?
 445                                         2 : (fpos_off(rde->offset) + 1);
 446                         err = note_last_dentry(fi, rde->name, rde->name_len,
 447                                                next_offset);
 448                         if (err)
 449                                 return err;
 450                 } else if (req->r_reply_info.dir_end) {
 451                         fi->next_offset = 2;
 452                         /* keep last name */
 453                 }
 454         }
 455
 456         rinfo = &fi->last_readdir->r_reply_info;
 457         dout("readdir frag %x num %d pos %llx chunk first %llx\n",
 458              fi->frag, rinfo->dir_nr, ctx->pos,
 459              rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
 460
 461         i = 0;
 462         /* search start position */
 463         if (rinfo->dir_nr > 0) {
 464                 int step, nr = rinfo->dir_nr;
 465                 while (nr > 0) {
 466                         step = nr >> 1;
 467                         if (rinfo->dir_entries[i + step].offset < ctx->pos) {
 468                                 i +=  step + 1;
 469                                 nr -= step + 1;
 470                         } else {
 471                                 nr = step;
 472                         }
 473                 }
 474         }
 475         for (; i < rinfo->dir_nr; i++) {
 476                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
 477                 struct ceph_vino vino;
 478                 ino_t ino;
 479
 480                 BUG_ON(rde->offset < ctx->pos);
 481
 482                 ctx->pos = rde->offset;
 483                 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
 484                      i, rinfo->dir_nr, ctx->pos,
 485                      rde->name_len, rde->name, &rde->inode.in);
 486
 487                 BUG_ON(!rde->inode.in);
 488                 ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
 489                 vino.ino = le64_to_cpu(rde->inode.in->ino);
 490                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
 491                 ino = ceph_vino_to_ino(vino);
 492
 493                 if (!dir_emit(ctx, rde->name, rde->name_len,
 494                               ceph_translate_ino(inode->i_sb, ino), ftype)) {
 495                         dout("filldir stopping us...\n");
 496                         return 0;
 497                 }
 498                 ctx->pos++;
 499         }
 500
 501         if (fi->next_offset > 2) {
 502                 ceph_mdsc_put_request(fi->last_readdir);
 503                 fi->last_readdir = NULL;
 504                 goto more;
 505         }
 506
 507         /* more frags? */
 508         if (!ceph_frag_is_rightmost(fi->frag)) {
 509                 unsigned frag = ceph_frag_next(fi->frag);
 510                 if (is_hash_order(ctx->pos)) {
 511                         loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
 512                                                         fi->next_offset, true);
 513                         if (new_pos > ctx->pos)
 514                                 ctx->pos = new_pos;
 515                         /* keep last_name */
 516                 } else {
 517                         ctx->pos = ceph_make_fpos(frag, fi->next_offset, false);
 518                         kfree(fi->last_name);
 519                         fi->last_name = NULL;
 520                 }
 521                 dout("readdir next frag is %x\n", frag);
 522                 goto more;
 523         }
 524         fi->flags |= CEPH_F_ATEND;
 525
 526         /*
 527          * if dir_release_count still matches the dir, no dentries
 528          * were released during the whole readdir, and we should have
 529          * the complete dir contents in our cache.
 530          */
 531         if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
 532                 spin_lock(&ci->i_ceph_lock);
 533                 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
 534                         dout(" marking %p complete and ordered\n", inode);
 535                         /* use i_size to track number of entries in
 536                          * readdir cache */
 537                         BUG_ON(fi->readdir_cache_idx < 0);
 538                         i_size_write(inode, fi->readdir_cache_idx *
 539                                      sizeof(struct dentry*));
 540                 } else {
 541                         dout(" marking %p complete\n", inode);
 542                 }
 543                 __ceph_dir_set_complete(ci, fi->dir_release_count,
 544                                         fi->dir_ordered_count);
 545                 spin_unlock(&ci->i_ceph_lock);
 546         }
 547
 548         dout("readdir %p file %p done.\n", inode, file);
 549         return 0;
 550 }
 551
 552 static void reset_readdir(struct ceph_file_info *fi)
 553 {
 554         if (fi->last_readdir) {
 555                 ceph_mdsc_put_request(fi->last_readdir);
 556                 fi->last_readdir = NULL;
 557         }
 558         kfree(fi->last_name);
 559         fi->last_name = NULL;
 560         fi->dir_release_count = 0;
 561         fi->readdir_cache_idx = -1;
 562         fi->next_offset = 2;  /* compensate for . and .. */
 563         fi->flags &= ~CEPH_F_ATEND;
 564 }
 565
 566 /*
 567  * discard buffered readdir content on seekdir(0), or seek to new frag,
 568  * or seek prior to current chunk
 569  */
 570 static bool need_reset_readdir(struct ceph_file_info *fi, loff_t new_pos)
 571 {
 572         struct ceph_mds_reply_info_parsed *rinfo;
 573         loff_t chunk_offset;
 574         if (new_pos == 0)
 575                 return true;
 576         if (is_hash_order(new_pos)) {
 577                 /* no need to reset last_name for a forward seek when
 578                  * dentries are sotred in hash order */
 579         } else if (fi->frag != fpos_frag(new_pos)) {
 580                 return true;
 581         }
 582         rinfo = fi->last_readdir ? &fi->last_readdir->r_reply_info : NULL;
 583         if (!rinfo || !rinfo->dir_nr)
 584                 return true;
 585         chunk_offset = rinfo->dir_entries[0].offset;
 586         return new_pos < chunk_offset ||
 587                is_hash_order(new_pos) != is_hash_order(chunk_offset);
 588 }
 589
 590 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
 591 {
 592         struct ceph_file_info *fi = file->private_data;
 593         struct inode *inode = file->f_mapping->host;
 594         loff_t retval;
 595
 596         inode_lock(inode);
 597         retval = -EINVAL;
 598         switch (whence) {
 599         case SEEK_CUR:
 600                 offset += file->f_pos;
 601         case SEEK_SET:
 602                 break;
 603         case SEEK_END:
 604                 retval = -EOPNOTSUPP;
 605         default:
 606                 goto out;
 607         }
 608
 609         if (offset >= 0) {
 610                 if (need_reset_readdir(fi, offset)) {
 611                         dout("dir_llseek dropping %p content\n", file);
 612                         reset_readdir(fi);
 613                 } else if (is_hash_order(offset) && offset > file->f_pos) {
 614                         /* for hash offset, we don't know if a forward seek
 615                          * is within same frag */
 616                         fi->dir_release_count = 0;
 617                         fi->readdir_cache_idx = -1;
 618                 }
 619
 620                 if (offset != file->f_pos) {
 621                         file->f_pos = offset;
 622                         file->f_version = 0;
 623                         fi->flags &= ~CEPH_F_ATEND;
 624                 }
 625                 retval = offset;
 626         }
 627 out:
 628         inode_unlock(inode);
 629         return retval;
 630 }
 631
 632 /*
 633  * Handle lookups for the hidden .snap directory.
 634  */
 635 int ceph_handle_snapdir(struct ceph_mds_request *req,
 636                         struct dentry *dentry, int err)
 637 {
 638         struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
 639         struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
 640
 641         /* .snap dir? */
 642         if (err == -ENOENT &&
 643             ceph_snap(parent) == CEPH_NOSNAP &&
 644             strcmp(dentry->d_name.name,
 645                    fsc->mount_options->snapdir_name) == 0) {
 646                 struct inode *inode = ceph_get_snapdir(parent);
 647                 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
 648                      dentry, dentry, inode);
 649                 BUG_ON(!d_unhashed(dentry));
 650                 d_add(dentry, inode);
 651                 err = 0;
 652         }
 653         return err;
 654 }
 655
 656 /*
 657  * Figure out final result of a lookup/open request.
 658  *
 659  * Mainly, make sure we return the final req->r_dentry (if it already
 660  * existed) in place of the original VFS-provided dentry when they
 661  * differ.
 662  *
 663  * Gracefully handle the case where the MDS replies with -ENOENT and
 664  * no trace (which it may do, at its discretion, e.g., if it doesn't
 665  * care to issue a lease on the negative dentry).
 666  */
 667 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
 668                                   struct dentry *dentry, int err)
 669 {
 670         if (err == -ENOENT) {
 671                 /* no trace? */
 672                 err = 0;
 673                 if (!req->r_reply_info.head->is_dentry) {
 674                         dout("ENOENT and no trace, dentry %p inode %p\n",
 675                              dentry, d_inode(dentry));
 676                         if (d_really_is_positive(dentry)) {
 677                                 d_drop(dentry);
 678                                 err = -ENOENT;
 679                         } else {
 680                                 d_add(dentry, NULL);
 681                         }
 682                 }
 683         }
 684         if (err)
 685                 dentry = ERR_PTR(err);
 686         else if (dentry != req->r_dentry)
 687                 dentry = dget(req->r_dentry);   /* we got spliced */
 688         else
 689                 dentry = NULL;
 690         return dentry;
 691 }
 692
 693 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
 694 {
 695         return ceph_ino(inode) == CEPH_INO_ROOT &&
 696                 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
 697 }
 698
 699 /*
 700  * Look up a single dir entry.  If there is a lookup intent, inform
 701  * the MDS so that it gets our 'caps wanted' value in a single op.
 702  */
 703 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
 704                                   unsigned int flags)
 705 {
 706         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 707         struct ceph_mds_client *mdsc = fsc->mdsc;
 708         struct ceph_mds_request *req;
 709         int op;
 710         int mask;
 711         int err;
 712
 713         dout("lookup %p dentry %p '%pd'\n",
 714              dir, dentry, dentry);
 715
 716         if (dentry->d_name.len > NAME_MAX)
 717                 return ERR_PTR(-ENAMETOOLONG);
 718
 719         /* can we conclude ENOENT locally? */
 720         if (d_really_is_negative(dentry)) {
 721                 struct ceph_inode_info *ci = ceph_inode(dir);
 722                 struct ceph_dentry_info *di = ceph_dentry(dentry);
 723
 724                 spin_lock(&ci->i_ceph_lock);
 725                 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
 726                 if (strncmp(dentry->d_name.name,
 727                             fsc->mount_options->snapdir_name,
 728                             dentry->d_name.len) &&
 729                     !is_root_ceph_dentry(dir, dentry) &&
 730                     ceph_test_mount_opt(fsc, DCACHE) &&
 731                     __ceph_dir_is_complete(ci) &&
 732                     (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
 733                         spin_unlock(&ci->i_ceph_lock);
 734                         dout(" dir %p complete, -ENOENT\n", dir);
 735                         d_add(dentry, NULL);
 736                         di->lease_shared_gen = ci->i_shared_gen;
 737                         return NULL;
 738                 }
 739                 spin_unlock(&ci->i_ceph_lock);
 740         }
 741
 742         op = ceph_snap(dir) == CEPH_SNAPDIR ?
 743                 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
 744         req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
 745         if (IS_ERR(req))
 746                 return ERR_CAST(req);
 747         req->r_dentry = dget(dentry);
 748         req->r_num_caps = 2;
 749
 750         mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
 751         if (ceph_security_xattr_wanted(dir))
 752                 mask |= CEPH_CAP_XATTR_SHARED;
 753         req->r_args.getattr.mask = cpu_to_le32(mask);
 754
 755         req->r_locked_dir = dir;
 756         err = ceph_mdsc_do_request(mdsc, NULL, req);
 757         err = ceph_handle_snapdir(req, dentry, err);
 758         dentry = ceph_finish_lookup(req, dentry, err);
 759         ceph_mdsc_put_request(req);  /* will dput(dentry) */
 760         dout("lookup result=%p\n", dentry);
 761         return dentry;
 762 }
 763
 764 /*
 765  * If we do a create but get no trace back from the MDS, follow up with
 766  * a lookup (the VFS expects us to link up the provided dentry).
 767  */
 768 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
 769 {
 770         struct dentry *result = ceph_lookup(dir, dentry, 0);
 771
 772         if (result && !IS_ERR(result)) {
 773                 /*
 774                  * We created the item, then did a lookup, and found
 775                  * it was already linked to another inode we already
 776                  * had in our cache (and thus got spliced). To not
 777                  * confuse VFS (especially when inode is a directory),
 778                  * we don't link our dentry to that inode, return an
 779                  * error instead.
 780                  *
 781                  * This event should be rare and it happens only when
 782                  * we talk to old MDS. Recent MDS does not send traceless
 783                  * reply for request that creates new inode.
 784                  */
 785                 d_drop(result);
 786                 return -ESTALE;
 787         }
 788         return PTR_ERR(result);
 789 }
 790
 791 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
 792                       umode_t mode, dev_t rdev)
 793 {
 794         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 795         struct ceph_mds_client *mdsc = fsc->mdsc;
 796         struct ceph_mds_request *req;
 797         struct ceph_acls_info acls = {};
 798         int err;
 799
 800         if (ceph_snap(dir) != CEPH_NOSNAP)
 801                 return -EROFS;
 802
 803         err = ceph_pre_init_acls(dir, &mode, &acls);
 804         if (err < 0)
 805                 return err;
 806
 807         dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
 808              dir, dentry, mode, rdev);
 809         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
 810         if (IS_ERR(req)) {
 811                 err = PTR_ERR(req);
 812                 goto out;
 813         }
 814         req->r_dentry = dget(dentry);
 815         req->r_num_caps = 2;
 816         req->r_locked_dir = dir;
 817         req->r_args.mknod.mode = cpu_to_le32(mode);
 818         req->r_args.mknod.rdev = cpu_to_le32(rdev);
 819         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 820         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 821         if (acls.pagelist) {
 822                 req->r_pagelist = acls.pagelist;
 823                 acls.pagelist = NULL;
 824         }
 825         err = ceph_mdsc_do_request(mdsc, dir, req);
 826         if (!err && !req->r_reply_info.head->is_dentry)
 827                 err = ceph_handle_notrace_create(dir, dentry);
 828         ceph_mdsc_put_request(req);
 829 out:
 830         if (!err)
 831                 ceph_init_inode_acls(d_inode(dentry), &acls);
 832         else
 833                 d_drop(dentry);
 834         ceph_release_acls_info(&acls);
 835         return err;
 836 }
 837
 838 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 839                        bool excl)
 840 {
 841         return ceph_mknod(dir, dentry, mode, 0);
 842 }
 843
 844 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
 845                             const char *dest)
 846 {
 847         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 848         struct ceph_mds_client *mdsc = fsc->mdsc;
 849         struct ceph_mds_request *req;
 850         int err;
 851
 852         if (ceph_snap(dir) != CEPH_NOSNAP)
 853                 return -EROFS;
 854
 855         dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
 856         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
 857         if (IS_ERR(req)) {
 858                 err = PTR_ERR(req);
 859                 goto out;
 860         }
 861         req->r_path2 = kstrdup(dest, GFP_KERNEL);
 862         if (!req->r_path2) {
 863                 err = -ENOMEM;
 864                 ceph_mdsc_put_request(req);
 865                 goto out;
 866         }
 867         req->r_locked_dir = dir;
 868         req->r_dentry = dget(dentry);
 869         req->r_num_caps = 2;
 870         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 871         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 872         err = ceph_mdsc_do_request(mdsc, dir, req);
 873         if (!err && !req->r_reply_info.head->is_dentry)
 874                 err = ceph_handle_notrace_create(dir, dentry);
 875         ceph_mdsc_put_request(req);
 876 out:
 877         if (err)
 878                 d_drop(dentry);
 879         return err;
 880 }
 881
 882 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 883 {
 884         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 885         struct ceph_mds_client *mdsc = fsc->mdsc;
 886         struct ceph_mds_request *req;
 887         struct ceph_acls_info acls = {};
 888         int err = -EROFS;
 889         int op;
 890
 891         if (ceph_snap(dir) == CEPH_SNAPDIR) {
 892                 /* mkdir .snap/foo is a MKSNAP */
 893                 op = CEPH_MDS_OP_MKSNAP;
 894                 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
 895                      dentry, dentry);
 896         } else if (ceph_snap(dir) == CEPH_NOSNAP) {
 897                 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
 898                 op = CEPH_MDS_OP_MKDIR;
 899         } else {
 900                 goto out;
 901         }
 902
 903         mode |= S_IFDIR;
 904         err = ceph_pre_init_acls(dir, &mode, &acls);
 905         if (err < 0)
 906                 goto out;
 907
 908         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 909         if (IS_ERR(req)) {
 910                 err = PTR_ERR(req);
 911                 goto out;
 912         }
 913
 914         req->r_dentry = dget(dentry);
 915         req->r_num_caps = 2;
 916         req->r_locked_dir = dir;
 917         req->r_args.mkdir.mode = cpu_to_le32(mode);
 918         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 919         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 920         if (acls.pagelist) {
 921                 req->r_pagelist = acls.pagelist;
 922                 acls.pagelist = NULL;
 923         }
 924         err = ceph_mdsc_do_request(mdsc, dir, req);
 925         if (!err &&
 926             !req->r_reply_info.head->is_target &&
 927             !req->r_reply_info.head->is_dentry)
 928                 err = ceph_handle_notrace_create(dir, dentry);
 929         ceph_mdsc_put_request(req);
 930 out:
 931         if (!err)
 932                 ceph_init_inode_acls(d_inode(dentry), &acls);
 933         else
 934                 d_drop(dentry);
 935         ceph_release_acls_info(&acls);
 936         return err;
 937 }
 938
 939 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
 940                      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 ceph_mds_request *req;
 945         int err;
 946
 947         if (ceph_snap(dir) != CEPH_NOSNAP)
 948                 return -EROFS;
 949
 950         dout("link in dir %p old_dentry %p dentry %p\n", dir,
 951              old_dentry, dentry);
 952         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
 953         if (IS_ERR(req)) {
 954                 d_drop(dentry);
 955                 return PTR_ERR(req);
 956         }
 957         req->r_dentry = dget(dentry);
 958         req->r_num_caps = 2;
 959         req->r_old_dentry = dget(old_dentry);
 960         req->r_locked_dir = dir;
 961         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 962         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 963         /* release LINK_SHARED on source inode (mds will lock it) */
 964         req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
 965         err = ceph_mdsc_do_request(mdsc, dir, req);
 966         if (err) {
 967                 d_drop(dentry);
 968         } else if (!req->r_reply_info.head->is_dentry) {
 969                 ihold(d_inode(old_dentry));
 970                 d_instantiate(dentry, d_inode(old_dentry));
 971         }
 972         ceph_mdsc_put_request(req);
 973         return err;
 974 }
 975
 976 /*
 977  * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps.  If it
 978  * looks like the link count will hit 0, drop any other caps (other
 979  * than PIN) we don't specifically want (due to the file still being
 980  * open).
 981  */
 982 static int drop_caps_for_unlink(struct inode *inode)
 983 {
 984         struct ceph_inode_info *ci = ceph_inode(inode);
 985         int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
 986
 987         spin_lock(&ci->i_ceph_lock);
 988         if (inode->i_nlink == 1) {
 989                 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
 990                 ci->i_ceph_flags |= CEPH_I_NODELAY;
 991         }
 992         spin_unlock(&ci->i_ceph_lock);
 993         return drop;
 994 }
 995
 996 /*
 997  * rmdir and unlink are differ only by the metadata op code
 998  */
 999 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1000 {
1001         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1002         struct ceph_mds_client *mdsc = fsc->mdsc;
1003         struct inode *inode = d_inode(dentry);
1004         struct ceph_mds_request *req;
1005         int err = -EROFS;
1006         int op;
1007
1008         if (ceph_snap(dir) == CEPH_SNAPDIR) {
1009                 /* rmdir .snap/foo is RMSNAP */
1010                 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
1011                 op = CEPH_MDS_OP_RMSNAP;
1012         } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1013                 dout("unlink/rmdir dir %p dn %p inode %p\n",
1014                      dir, dentry, inode);
1015                 op = d_is_dir(dentry) ?
1016                         CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1017         } else
1018                 goto out;
1019         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1020         if (IS_ERR(req)) {
1021                 err = PTR_ERR(req);
1022                 goto out;
1023         }
1024         req->r_dentry = dget(dentry);
1025         req->r_num_caps = 2;
1026         req->r_locked_dir = dir;
1027         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1028         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1029         req->r_inode_drop = drop_caps_for_unlink(inode);
1030         err = ceph_mdsc_do_request(mdsc, dir, req);
1031         if (!err && !req->r_reply_info.head->is_dentry)
1032                 d_delete(dentry);
1033         ceph_mdsc_put_request(req);
1034 out:
1035         return err;
1036 }
1037
1038 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
1039                        struct inode *new_dir, struct dentry *new_dentry,
1040                        unsigned int flags)
1041 {
1042         struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
1043         struct ceph_mds_client *mdsc = fsc->mdsc;
1044         struct ceph_mds_request *req;
1045         int op = CEPH_MDS_OP_RENAME;
1046         int err;
1047
1048         if (flags)
1049                 return -EINVAL;
1050
1051         if (ceph_snap(old_dir) != ceph_snap(new_dir))
1052                 return -EXDEV;
1053         if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1054                 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1055                         op = CEPH_MDS_OP_RENAMESNAP;
1056                 else
1057                         return -EROFS;
1058         }
1059         dout("rename dir %p dentry %p to dir %p dentry %p\n",
1060              old_dir, old_dentry, new_dir, new_dentry);
1061         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1062         if (IS_ERR(req))
1063                 return PTR_ERR(req);
1064         ihold(old_dir);
1065         req->r_dentry = dget(new_dentry);
1066         req->r_num_caps = 2;
1067         req->r_old_dentry = dget(old_dentry);
1068         req->r_old_dentry_dir = old_dir;
1069         req->r_locked_dir = new_dir;
1070         req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1071         req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1072         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1073         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1074         /* release LINK_RDCACHE on source inode (mds will lock it) */
1075         req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
1076         if (d_really_is_positive(new_dentry))
1077                 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
1078         err = ceph_mdsc_do_request(mdsc, old_dir, req);
1079         if (!err && !req->r_reply_info.head->is_dentry) {
1080                 /*
1081                  * Normally d_move() is done by fill_trace (called by
1082                  * do_request, above).  If there is no trace, we need
1083                  * to do it here.
1084                  */
1085
1086                 /* d_move screws up sibling dentries' offsets */
1087                 ceph_dir_clear_complete(old_dir);
1088                 ceph_dir_clear_complete(new_dir);
1089
1090                 d_move(old_dentry, new_dentry);
1091
1092                 /* ensure target dentry is invalidated, despite
1093                    rehashing bug in vfs_rename_dir */
1094                 ceph_invalidate_dentry_lease(new_dentry);
1095         }
1096         ceph_mdsc_put_request(req);
1097         return err;
1098 }
1099
1100 /*
1101  * Ensure a dentry lease will no longer revalidate.
1102  */
1103 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1104 {
1105         spin_lock(&dentry->d_lock);
1106         ceph_dentry(dentry)->time = jiffies;
1107         ceph_dentry(dentry)->lease_shared_gen = 0;
1108         spin_unlock(&dentry->d_lock);
1109 }
1110
1111 /*
1112  * Check if dentry lease is valid.  If not, delete the lease.  Try to
1113  * renew if the least is more than half up.
1114  */
1115 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags,
1116                                  struct inode *dir)
1117 {
1118         struct ceph_dentry_info *di;
1119         struct ceph_mds_session *s;
1120         int valid = 0;
1121         u32 gen;
1122         unsigned long ttl;
1123         struct ceph_mds_session *session = NULL;
1124         u32 seq = 0;
1125
1126         spin_lock(&dentry->d_lock);
1127         di = ceph_dentry(dentry);
1128         if (di && di->lease_session) {
1129                 s = di->lease_session;
1130                 spin_lock(&s->s_gen_ttl_lock);
1131                 gen = s->s_cap_gen;
1132                 ttl = s->s_cap_ttl;
1133                 spin_unlock(&s->s_gen_ttl_lock);
1134
1135                 if (di->lease_gen == gen &&
1136                     time_before(jiffies, di->time) &&
1137                     time_before(jiffies, ttl)) {
1138                         valid = 1;
1139                         if (di->lease_renew_after &&
1140                             time_after(jiffies, di->lease_renew_after)) {
1141                                 /*
1142                                  * We should renew. If we're in RCU walk mode
1143                                  * though, we can't do that so just return
1144                                  * -ECHILD.
1145                                  */
1146                                 if (flags & LOOKUP_RCU) {
1147                                         valid = -ECHILD;
1148                                 } else {
1149                                         session = ceph_get_mds_session(s);
1150                                         seq = di->lease_seq;
1151                                         di->lease_renew_after = 0;
1152                                         di->lease_renew_from = jiffies;
1153                                 }
1154                         }
1155                 }
1156         }
1157         spin_unlock(&dentry->d_lock);
1158
1159         if (session) {
1160                 ceph_mdsc_lease_send_msg(session, dir, dentry,
1161                                          CEPH_MDS_LEASE_RENEW, seq);
1162                 ceph_put_mds_session(session);
1163         }
1164         dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1165         return valid;
1166 }
1167
1168 /*
1169  * Check if directory-wide content lease/cap is valid.
1170  */
1171 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1172 {
1173         struct ceph_inode_info *ci = ceph_inode(dir);
1174         struct ceph_dentry_info *di = ceph_dentry(dentry);
1175         int valid = 0;
1176
1177         spin_lock(&ci->i_ceph_lock);
1178         if (ci->i_shared_gen == di->lease_shared_gen)
1179                 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1180         spin_unlock(&ci->i_ceph_lock);
1181         dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1182              dir, (unsigned)ci->i_shared_gen, dentry,
1183              (unsigned)di->lease_shared_gen, valid);
1184         return valid;
1185 }
1186
1187 /*
1188  * Check if cached dentry can be trusted.
1189  */
1190 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1191 {
1192         int valid = 0;
1193         struct dentry *parent;
1194         struct inode *dir;
1195
1196         if (flags & LOOKUP_RCU) {
1197                 parent = ACCESS_ONCE(dentry->d_parent);
1198                 dir = d_inode_rcu(parent);
1199                 if (!dir)
1200                         return -ECHILD;
1201         } else {
1202                 parent = dget_parent(dentry);
1203                 dir = d_inode(parent);
1204         }
1205
1206         dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1207              dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1208
1209         /* always trust cached snapped dentries, snapdir dentry */
1210         if (ceph_snap(dir) != CEPH_NOSNAP) {
1211                 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1212                      dentry, d_inode(dentry));
1213                 valid = 1;
1214         } else if (d_really_is_positive(dentry) &&
1215                    ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1216                 valid = 1;
1217         } else {
1218                 valid = dentry_lease_is_valid(dentry, flags, dir);
1219                 if (valid == -ECHILD)
1220                         return valid;
1221                 if (valid || dir_lease_is_valid(dir, dentry)) {
1222                         if (d_really_is_positive(dentry))
1223                                 valid = ceph_is_any_caps(d_inode(dentry));
1224                         else
1225                                 valid = 1;
1226                 }
1227         }
1228
1229         if (!valid) {
1230                 struct ceph_mds_client *mdsc =
1231                         ceph_sb_to_client(dir->i_sb)->mdsc;
1232                 struct ceph_mds_request *req;
1233                 int op, err;
1234                 u32 mask;
1235
1236                 if (flags & LOOKUP_RCU)
1237                         return -ECHILD;
1238
1239                 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1240                         CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_GETATTR;
1241                 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1242                 if (!IS_ERR(req)) {
1243                         req->r_dentry = dget(dentry);
1244                         req->r_num_caps = op == CEPH_MDS_OP_GETATTR ? 1 : 2;
1245
1246                         mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1247                         if (ceph_security_xattr_wanted(dir))
1248                                 mask |= CEPH_CAP_XATTR_SHARED;
1249                         req->r_args.getattr.mask = cpu_to_le32(mask);
1250
1251                         err = ceph_mdsc_do_request(mdsc, NULL, req);
1252                         switch (err) {
1253                         case 0:
1254                                 if (d_really_is_positive(dentry) &&
1255                                     d_inode(dentry) == req->r_target_inode)
1256                                         valid = 1;
1257                                 break;
1258                         case -ENOENT:
1259                                 if (d_really_is_negative(dentry))
1260                                         valid = 1;
1261                                 /* Fallthrough */
1262                         default:
1263                                 break;
1264                         }
1265                         ceph_mdsc_put_request(req);
1266                         dout("d_revalidate %p lookup result=%d\n",
1267                              dentry, err);
1268                 }
1269         }
1270
1271         dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1272         if (valid) {
1273                 ceph_dentry_lru_touch(dentry);
1274         } else {
1275                 ceph_dir_clear_complete(dir);
1276         }
1277
1278         if (!(flags & LOOKUP_RCU))
1279                 dput(parent);
1280         return valid;
1281 }
1282
1283 /*
1284  * Release our ceph_dentry_info.
1285  */
1286 static void ceph_d_release(struct dentry *dentry)
1287 {
1288         struct ceph_dentry_info *di = ceph_dentry(dentry);
1289
1290         dout("d_release %p\n", dentry);
1291         ceph_dentry_lru_del(dentry);
1292
1293         spin_lock(&dentry->d_lock);
1294         dentry->d_fsdata = NULL;
1295         spin_unlock(&dentry->d_lock);
1296
1297         if (di->lease_session)
1298                 ceph_put_mds_session(di->lease_session);
1299         kmem_cache_free(ceph_dentry_cachep, di);
1300 }
1301
1302 /*
1303  * When the VFS prunes a dentry from the cache, we need to clear the
1304  * complete flag on the parent directory.
1305  *
1306  * Called under dentry->d_lock.
1307  */
1308 static void ceph_d_prune(struct dentry *dentry)
1309 {
1310         dout("ceph_d_prune %p\n", dentry);
1311
1312         /* do we have a valid parent? */
1313         if (IS_ROOT(dentry))
1314                 return;
1315
1316         /* if we are not hashed, we don't affect dir's completeness */
1317         if (d_unhashed(dentry))
1318                 return;
1319
1320         if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
1321                 return;
1322
1323         /*
1324          * we hold d_lock, so d_parent is stable, and d_fsdata is never
1325          * cleared until d_release
1326          */
1327         ceph_dir_clear_complete(d_inode(dentry->d_parent));
1328 }
1329
1330 /*
1331  * read() on a dir.  This weird interface hack only works if mounted
1332  * with '-o dirstat'.
1333  */
1334 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1335                              loff_t *ppos)
1336 {
1337         struct ceph_file_info *cf = file->private_data;
1338         struct inode *inode = file_inode(file);
1339         struct ceph_inode_info *ci = ceph_inode(inode);
1340         int left;
1341         const int bufsize = 1024;
1342
1343         if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1344                 return -EISDIR;
1345
1346         if (!cf->dir_info) {
1347                 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1348                 if (!cf->dir_info)
1349                         return -ENOMEM;
1350                 cf->dir_info_len =
1351                         snprintf(cf->dir_info, bufsize,
1352                                 "entries:   %20lld\n"
1353                                 " files:    %20lld\n"
1354                                 " subdirs:  %20lld\n"
1355                                 "rentries:  %20lld\n"
1356                                 " rfiles:   %20lld\n"
1357                                 " rsubdirs: %20lld\n"
1358                                 "rbytes:    %20lld\n"
1359                                 "rctime:    %10ld.%09ld\n",
1360                                 ci->i_files + ci->i_subdirs,
1361                                 ci->i_files,
1362                                 ci->i_subdirs,
1363                                 ci->i_rfiles + ci->i_rsubdirs,
1364                                 ci->i_rfiles,
1365                                 ci->i_rsubdirs,
1366                                 ci->i_rbytes,
1367                                 (long)ci->i_rctime.tv_sec,
1368                                 (long)ci->i_rctime.tv_nsec);
1369         }
1370
1371         if (*ppos >= cf->dir_info_len)
1372                 return 0;
1373         size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1374         left = copy_to_user(buf, cf->dir_info + *ppos, size);
1375         if (left == size)
1376                 return -EFAULT;
1377         *ppos += (size - left);
1378         return size - left;
1379 }
1380
1381 /*
1382  * We maintain a private dentry LRU.
1383  *
1384  * FIXME: this needs to be changed to a per-mds lru to be useful.
1385  */
1386 void ceph_dentry_lru_add(struct dentry *dn)
1387 {
1388         struct ceph_dentry_info *di = ceph_dentry(dn);
1389         struct ceph_mds_client *mdsc;
1390
1391         dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1392         mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1393         spin_lock(&mdsc->dentry_lru_lock);
1394         list_add_tail(&di->lru, &mdsc->dentry_lru);
1395         mdsc->num_dentry++;
1396         spin_unlock(&mdsc->dentry_lru_lock);
1397 }
1398
1399 void ceph_dentry_lru_touch(struct dentry *dn)
1400 {
1401         struct ceph_dentry_info *di = ceph_dentry(dn);
1402         struct ceph_mds_client *mdsc;
1403
1404         dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1405              di->offset);
1406         mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1407         spin_lock(&mdsc->dentry_lru_lock);
1408         list_move_tail(&di->lru, &mdsc->dentry_lru);
1409         spin_unlock(&mdsc->dentry_lru_lock);
1410 }
1411
1412 void ceph_dentry_lru_del(struct dentry *dn)
1413 {
1414         struct ceph_dentry_info *di = ceph_dentry(dn);
1415         struct ceph_mds_client *mdsc;
1416
1417         dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1418         mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1419         spin_lock(&mdsc->dentry_lru_lock);
1420         list_del_init(&di->lru);
1421         mdsc->num_dentry--;
1422         spin_unlock(&mdsc->dentry_lru_lock);
1423 }
1424
1425 /*
1426  * Return name hash for a given dentry.  This is dependent on
1427  * the parent directory's hash function.
1428  */
1429 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1430 {
1431         struct ceph_inode_info *dci = ceph_inode(dir);
1432
1433         switch (dci->i_dir_layout.dl_dir_hash) {
1434         case 0: /* for backward compat */
1435         case CEPH_STR_HASH_LINUX:
1436                 return dn->d_name.hash;
1437
1438         default:
1439                 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1440                                      dn->d_name.name, dn->d_name.len);
1441         }
1442 }
1443
1444 const struct file_operations ceph_dir_fops = {
1445         .read = ceph_read_dir,
1446         .iterate = ceph_readdir,
1447         .llseek = ceph_dir_llseek,
1448         .open = ceph_open,
1449         .release = ceph_release,
1450         .unlocked_ioctl = ceph_ioctl,
1451         .fsync = ceph_fsync,
1452 };
1453
1454 const struct file_operations ceph_snapdir_fops = {
1455         .iterate = ceph_readdir,
1456         .llseek = ceph_dir_llseek,
1457         .open = ceph_open,
1458         .release = ceph_release,
1459 };
1460
1461 const struct inode_operations ceph_dir_iops = {
1462         .lookup = ceph_lookup,
1463         .permission = ceph_permission,
1464         .getattr = ceph_getattr,
1465         .setattr = ceph_setattr,
1466         .listxattr = ceph_listxattr,
1467         .get_acl = ceph_get_acl,
1468         .set_acl = ceph_set_acl,
1469         .mknod = ceph_mknod,
1470         .symlink = ceph_symlink,
1471         .mkdir = ceph_mkdir,
1472         .link = ceph_link,
1473         .unlink = ceph_unlink,
1474         .rmdir = ceph_unlink,
1475         .rename = ceph_rename,
1476         .create = ceph_create,
1477         .atomic_open = ceph_atomic_open,
1478 };
1479
1480 const struct inode_operations ceph_snapdir_iops = {
1481         .lookup = ceph_lookup,
1482         .permission = ceph_permission,
1483         .getattr = ceph_getattr,
1484         .mkdir = ceph_mkdir,
1485         .rmdir = ceph_unlink,
1486         .rename = ceph_rename,
1487 };
1488
1489 const struct dentry_operations ceph_dentry_ops = {
1490         .d_revalidate = ceph_d_revalidate,
1491         .d_release = ceph_d_release,
1492         .d_prune = ceph_d_prune,
1493         .d_init = ceph_d_init,
1494 };