fs/ceph/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/fs.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/xattr.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/random.h>
  15 #include <linux/sort.h>
  16 #include <linux/iversion.h>
  17
  18 #include "super.h"
  19 #include "mds_client.h"
  20 #include "cache.h"
  21 #include <linux/ceph/decode.h>
  22
  23 /*
  24  * Ceph inode operations
  25  *
  26  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  27  * setattr, etc.), xattr helpers, and helpers for assimilating
  28  * metadata returned by the MDS into our cache.
  29  *
  30  * Also define helpers for doing asynchronous writeback, invalidation,
  31  * and truncation for the benefit of those who can't afford to block
  32  * (typically because they are in the message handler path).
  33  */
  34
  35 static const struct inode_operations ceph_symlink_iops;
  36
  37 static void ceph_inode_work(struct work_struct *work);
  38
  39 /*
  40  * find or create an inode, given the ceph ino number
  41  */
  42 static int ceph_set_ino_cb(struct inode *inode, void *data)
  43 {
  44         struct ceph_inode_info *ci = ceph_inode(inode);
  45         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
  46
  47         ci->i_vino = *(struct ceph_vino *)data;
  48         inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
  49         inode_set_iversion_raw(inode, 0);
  50         percpu_counter_inc(&mdsc->metric.total_inodes);
  51
  52         return 0;
  53 }
  54
  55 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  56 {
  57         struct inode *inode;
  58
  59         if (ceph_vino_is_reserved(vino))
  60                 return ERR_PTR(-EREMOTEIO);
  61
  62         inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare,
  63                              ceph_set_ino_cb, &vino);
  64         if (!inode)
  65                 return ERR_PTR(-ENOMEM);
  66
  67         dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
  68              ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
  69         return inode;
  70 }
  71
  72 /*
  73  * get/constuct snapdir inode for a given directory
  74  */
  75 struct inode *ceph_get_snapdir(struct inode *parent)
  76 {
  77         struct ceph_vino vino = {
  78                 .ino = ceph_ino(parent),
  79                 .snap = CEPH_SNAPDIR,
  80         };
  81         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  82         struct ceph_inode_info *ci = ceph_inode(inode);
  83
  84         if (IS_ERR(inode))
  85                 return inode;
  86
  87         if (!S_ISDIR(parent->i_mode)) {
  88                 pr_warn_once("bad snapdir parent type (mode=0%o)\n",
  89                              parent->i_mode);
  90                 return ERR_PTR(-ENOTDIR);
  91         }
  92
  93         if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
  94                 pr_warn_once("bad snapdir inode type (mode=0%o)\n",
  95                              inode->i_mode);
  96                 return ERR_PTR(-ENOTDIR);
  97         }
  98
  99         inode->i_mode = parent->i_mode;
 100         inode->i_uid = parent->i_uid;
 101         inode->i_gid = parent->i_gid;
 102         inode->i_mtime = parent->i_mtime;
 103         inode->i_ctime = parent->i_ctime;
 104         inode->i_atime = parent->i_atime;
 105         ci->i_rbytes = 0;
 106         ci->i_btime = ceph_inode(parent)->i_btime;
 107
 108         if (inode->i_state & I_NEW) {
 109                 inode->i_op = &ceph_snapdir_iops;
 110                 inode->i_fop = &ceph_snapdir_fops;
 111                 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
 112                 unlock_new_inode(inode);
 113         }
 114
 115         return inode;
 116 }
 117
 118 const struct inode_operations ceph_file_iops = {
 119         .permission = ceph_permission,
 120         .setattr = ceph_setattr,
 121         .getattr = ceph_getattr,
 122         .listxattr = ceph_listxattr,
 123         .get_acl = ceph_get_acl,
 124         .set_acl = ceph_set_acl,
 125 };
 126
 127
 128 /*
 129  * We use a 'frag tree' to keep track of the MDS's directory fragments
 130  * for a given inode (usually there is just a single fragment).  We
 131  * need to know when a child frag is delegated to a new MDS, or when
 132  * it is flagged as replicated, so we can direct our requests
 133  * accordingly.
 134  */
 135
 136 /*
 137  * find/create a frag in the tree
 138  */
 139 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 140                                                     u32 f)
 141 {
 142         struct rb_node **p;
 143         struct rb_node *parent = NULL;
 144         struct ceph_inode_frag *frag;
 145         int c;
 146
 147         p = &ci->i_fragtree.rb_node;
 148         while (*p) {
 149                 parent = *p;
 150                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 151                 c = ceph_frag_compare(f, frag->frag);
 152                 if (c < 0)
 153                         p = &(*p)->rb_left;
 154                 else if (c > 0)
 155                         p = &(*p)->rb_right;
 156                 else
 157                         return frag;
 158         }
 159
 160         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 161         if (!frag)
 162                 return ERR_PTR(-ENOMEM);
 163
 164         frag->frag = f;
 165         frag->split_by = 0;
 166         frag->mds = -1;
 167         frag->ndist = 0;
 168
 169         rb_link_node(&frag->node, parent, p);
 170         rb_insert_color(&frag->node, &ci->i_fragtree);
 171
 172         dout("get_or_create_frag added %llx.%llx frag %x\n",
 173              ceph_vinop(&ci->vfs_inode), f);
 174         return frag;
 175 }
 176
 177 /*
 178  * find a specific frag @f
 179  */
 180 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 181 {
 182         struct rb_node *n = ci->i_fragtree.rb_node;
 183
 184         while (n) {
 185                 struct ceph_inode_frag *frag =
 186                         rb_entry(n, struct ceph_inode_frag, node);
 187                 int c = ceph_frag_compare(f, frag->frag);
 188                 if (c < 0)
 189                         n = n->rb_left;
 190                 else if (c > 0)
 191                         n = n->rb_right;
 192                 else
 193                         return frag;
 194         }
 195         return NULL;
 196 }
 197
 198 /*
 199  * Choose frag containing the given value @v.  If @pfrag is
 200  * specified, copy the frag delegation info to the caller if
 201  * it is present.
 202  */
 203 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 204                               struct ceph_inode_frag *pfrag, int *found)
 205 {
 206         u32 t = ceph_frag_make(0, 0);
 207         struct ceph_inode_frag *frag;
 208         unsigned nway, i;
 209         u32 n;
 210
 211         if (found)
 212                 *found = 0;
 213
 214         while (1) {
 215                 WARN_ON(!ceph_frag_contains_value(t, v));
 216                 frag = __ceph_find_frag(ci, t);
 217                 if (!frag)
 218                         break; /* t is a leaf */
 219                 if (frag->split_by == 0) {
 220                         if (pfrag)
 221                                 memcpy(pfrag, frag, sizeof(*pfrag));
 222                         if (found)
 223                                 *found = 1;
 224                         break;
 225                 }
 226
 227                 /* choose child */
 228                 nway = 1 << frag->split_by;
 229                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 230                      frag->split_by, nway);
 231                 for (i = 0; i < nway; i++) {
 232                         n = ceph_frag_make_child(t, frag->split_by, i);
 233                         if (ceph_frag_contains_value(n, v)) {
 234                                 t = n;
 235                                 break;
 236                         }
 237                 }
 238                 BUG_ON(i == nway);
 239         }
 240         dout("choose_frag(%x) = %x\n", v, t);
 241
 242         return t;
 243 }
 244
 245 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 246                      struct ceph_inode_frag *pfrag, int *found)
 247 {
 248         u32 ret;
 249         mutex_lock(&ci->i_fragtree_mutex);
 250         ret = __ceph_choose_frag(ci, v, pfrag, found);
 251         mutex_unlock(&ci->i_fragtree_mutex);
 252         return ret;
 253 }
 254
 255 /*
 256  * Process dirfrag (delegation) info from the mds.  Include leaf
 257  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 258  * branches/splits are included in i_fragtree)
 259  */
 260 static int ceph_fill_dirfrag(struct inode *inode,
 261                              struct ceph_mds_reply_dirfrag *dirinfo)
 262 {
 263         struct ceph_inode_info *ci = ceph_inode(inode);
 264         struct ceph_inode_frag *frag;
 265         u32 id = le32_to_cpu(dirinfo->frag);
 266         int mds = le32_to_cpu(dirinfo->auth);
 267         int ndist = le32_to_cpu(dirinfo->ndist);
 268         int diri_auth = -1;
 269         int i;
 270         int err = 0;
 271
 272         spin_lock(&ci->i_ceph_lock);
 273         if (ci->i_auth_cap)
 274                 diri_auth = ci->i_auth_cap->mds;
 275         spin_unlock(&ci->i_ceph_lock);
 276
 277         if (mds == -1) /* CDIR_AUTH_PARENT */
 278                 mds = diri_auth;
 279
 280         mutex_lock(&ci->i_fragtree_mutex);
 281         if (ndist == 0 && mds == diri_auth) {
 282                 /* no delegation info needed. */
 283                 frag = __ceph_find_frag(ci, id);
 284                 if (!frag)
 285                         goto out;
 286                 if (frag->split_by == 0) {
 287                         /* tree leaf, remove */
 288                         dout("fill_dirfrag removed %llx.%llx frag %x"
 289                              " (no ref)\n", ceph_vinop(inode), id);
 290                         rb_erase(&frag->node, &ci->i_fragtree);
 291                         kfree(frag);
 292                 } else {
 293                         /* tree branch, keep and clear */
 294                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 295                              " referral\n", ceph_vinop(inode), id);
 296                         frag->mds = -1;
 297                         frag->ndist = 0;
 298                 }
 299                 goto out;
 300         }
 301
 302
 303         /* find/add this frag to store mds delegation info */
 304         frag = __get_or_create_frag(ci, id);
 305         if (IS_ERR(frag)) {
 306                 /* this is not the end of the world; we can continue
 307                    with bad/inaccurate delegation info */
 308                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 309                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 310                 err = -ENOMEM;
 311                 goto out;
 312         }
 313
 314         frag->mds = mds;
 315         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 316         for (i = 0; i < frag->ndist; i++)
 317                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 318         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 319              ceph_vinop(inode), frag->frag, frag->ndist);
 320
 321 out:
 322         mutex_unlock(&ci->i_fragtree_mutex);
 323         return err;
 324 }
 325
 326 static int frag_tree_split_cmp(const void *l, const void *r)
 327 {
 328         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 329         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 330         return ceph_frag_compare(le32_to_cpu(ls->frag),
 331                                  le32_to_cpu(rs->frag));
 332 }
 333
 334 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 335 {
 336         if (!frag)
 337                 return f == ceph_frag_make(0, 0);
 338         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 339                 return false;
 340         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 341 }
 342
 343 static int ceph_fill_fragtree(struct inode *inode,
 344                               struct ceph_frag_tree_head *fragtree,
 345                               struct ceph_mds_reply_dirfrag *dirinfo)
 346 {
 347         struct ceph_inode_info *ci = ceph_inode(inode);
 348         struct ceph_inode_frag *frag, *prev_frag = NULL;
 349         struct rb_node *rb_node;
 350         unsigned i, split_by, nsplits;
 351         u32 id;
 352         bool update = false;
 353
 354         mutex_lock(&ci->i_fragtree_mutex);
 355         nsplits = le32_to_cpu(fragtree->nsplits);
 356         if (nsplits != ci->i_fragtree_nsplits) {
 357                 update = true;
 358         } else if (nsplits) {
 359                 i = prandom_u32() % nsplits;
 360                 id = le32_to_cpu(fragtree->splits[i].frag);
 361                 if (!__ceph_find_frag(ci, id))
 362                         update = true;
 363         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 364                 rb_node = rb_first(&ci->i_fragtree);
 365                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 366                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 367                         update = true;
 368         }
 369         if (!update && dirinfo) {
 370                 id = le32_to_cpu(dirinfo->frag);
 371                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 372                         update = true;
 373         }
 374         if (!update)
 375                 goto out_unlock;
 376
 377         if (nsplits > 1) {
 378                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 379                      frag_tree_split_cmp, NULL);
 380         }
 381
 382         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 383         rb_node = rb_first(&ci->i_fragtree);
 384         for (i = 0; i < nsplits; i++) {
 385                 id = le32_to_cpu(fragtree->splits[i].frag);
 386                 split_by = le32_to_cpu(fragtree->splits[i].by);
 387                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 388                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 389                                "frag %x split by %d\n", ceph_vinop(inode),
 390                                i, nsplits, id, split_by);
 391                         continue;
 392                 }
 393                 frag = NULL;
 394                 while (rb_node) {
 395                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 396                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 397                                 if (frag->frag != id)
 398                                         frag = NULL;
 399                                 else
 400                                         rb_node = rb_next(rb_node);
 401                                 break;
 402                         }
 403                         rb_node = rb_next(rb_node);
 404                         /* delete stale split/leaf node */
 405                         if (frag->split_by > 0 ||
 406                             !is_frag_child(frag->frag, prev_frag)) {
 407                                 rb_erase(&frag->node, &ci->i_fragtree);
 408                                 if (frag->split_by > 0)
 409                                         ci->i_fragtree_nsplits--;
 410                                 kfree(frag);
 411                         }
 412                         frag = NULL;
 413                 }
 414                 if (!frag) {
 415                         frag = __get_or_create_frag(ci, id);
 416                         if (IS_ERR(frag))
 417                                 continue;
 418                 }
 419                 if (frag->split_by == 0)
 420                         ci->i_fragtree_nsplits++;
 421                 frag->split_by = split_by;
 422                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 423                 prev_frag = frag;
 424         }
 425         while (rb_node) {
 426                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 427                 rb_node = rb_next(rb_node);
 428                 /* delete stale split/leaf node */
 429                 if (frag->split_by > 0 ||
 430                     !is_frag_child(frag->frag, prev_frag)) {
 431                         rb_erase(&frag->node, &ci->i_fragtree);
 432                         if (frag->split_by > 0)
 433                                 ci->i_fragtree_nsplits--;
 434                         kfree(frag);
 435                 }
 436         }
 437 out_unlock:
 438         mutex_unlock(&ci->i_fragtree_mutex);
 439         return 0;
 440 }
 441
 442 /*
 443  * initialize a newly allocated inode.
 444  */
 445 struct inode *ceph_alloc_inode(struct super_block *sb)
 446 {
 447         struct ceph_inode_info *ci;
 448         int i;
 449
 450         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
 451         if (!ci)
 452                 return NULL;
 453
 454         dout("alloc_inode %p\n", &ci->vfs_inode);
 455
 456         spin_lock_init(&ci->i_ceph_lock);
 457
 458         ci->i_version = 0;
 459         ci->i_inline_version = 0;
 460         ci->i_time_warp_seq = 0;
 461         ci->i_ceph_flags = 0;
 462         atomic64_set(&ci->i_ordered_count, 1);
 463         atomic64_set(&ci->i_release_count, 1);
 464         atomic64_set(&ci->i_complete_seq[0], 0);
 465         atomic64_set(&ci->i_complete_seq[1], 0);
 466         ci->i_symlink = NULL;
 467
 468         ci->i_max_bytes = 0;
 469         ci->i_max_files = 0;
 470
 471         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 472         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
 473         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 474
 475         ci->i_fragtree = RB_ROOT;
 476         mutex_init(&ci->i_fragtree_mutex);
 477
 478         ci->i_xattrs.blob = NULL;
 479         ci->i_xattrs.prealloc_blob = NULL;
 480         ci->i_xattrs.dirty = false;
 481         ci->i_xattrs.index = RB_ROOT;
 482         ci->i_xattrs.count = 0;
 483         ci->i_xattrs.names_size = 0;
 484         ci->i_xattrs.vals_size = 0;
 485         ci->i_xattrs.version = 0;
 486         ci->i_xattrs.index_version = 0;
 487
 488         ci->i_caps = RB_ROOT;
 489         ci->i_auth_cap = NULL;
 490         ci->i_dirty_caps = 0;
 491         ci->i_flushing_caps = 0;
 492         INIT_LIST_HEAD(&ci->i_dirty_item);
 493         INIT_LIST_HEAD(&ci->i_flushing_item);
 494         ci->i_prealloc_cap_flush = NULL;
 495         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 496         init_waitqueue_head(&ci->i_cap_wq);
 497         ci->i_hold_caps_max = 0;
 498         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 499         INIT_LIST_HEAD(&ci->i_cap_snaps);
 500         ci->i_head_snapc = NULL;
 501         ci->i_snap_caps = 0;
 502
 503         ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
 504         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 505                 ci->i_nr_by_mode[i] = 0;
 506
 507         mutex_init(&ci->i_truncate_mutex);
 508         ci->i_truncate_seq = 0;
 509         ci->i_truncate_size = 0;
 510         ci->i_truncate_pending = 0;
 511
 512         ci->i_max_size = 0;
 513         ci->i_reported_size = 0;
 514         ci->i_wanted_max_size = 0;
 515         ci->i_requested_max_size = 0;
 516
 517         ci->i_pin_ref = 0;
 518         ci->i_rd_ref = 0;
 519         ci->i_rdcache_ref = 0;
 520         ci->i_wr_ref = 0;
 521         ci->i_wb_ref = 0;
 522         ci->i_fx_ref = 0;
 523         ci->i_wrbuffer_ref = 0;
 524         ci->i_wrbuffer_ref_head = 0;
 525         atomic_set(&ci->i_filelock_ref, 0);
 526         atomic_set(&ci->i_shared_gen, 1);
 527         ci->i_rdcache_gen = 0;
 528         ci->i_rdcache_revoking = 0;
 529
 530         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 531         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 532         spin_lock_init(&ci->i_unsafe_lock);
 533
 534         ci->i_snap_realm = NULL;
 535         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 536         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 537
 538         INIT_WORK(&ci->i_work, ceph_inode_work);
 539         ci->i_work_mask = 0;
 540         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 541
 542         ceph_fscache_inode_init(ci);
 543
 544         ci->i_meta_err = 0;
 545
 546         return &ci->vfs_inode;
 547 }
 548
 549 void ceph_free_inode(struct inode *inode)
 550 {
 551         struct ceph_inode_info *ci = ceph_inode(inode);
 552
 553         kfree(ci->i_symlink);
 554         kmem_cache_free(ceph_inode_cachep, ci);
 555 }
 556
 557 void ceph_evict_inode(struct inode *inode)
 558 {
 559         struct ceph_inode_info *ci = ceph_inode(inode);
 560         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 561         struct ceph_inode_frag *frag;
 562         struct rb_node *n;
 563
 564         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 565
 566         percpu_counter_dec(&mdsc->metric.total_inodes);
 567
 568         truncate_inode_pages_final(&inode->i_data);
 569         clear_inode(inode);
 570
 571         ceph_fscache_unregister_inode_cookie(ci);
 572
 573         __ceph_remove_caps(ci);
 574
 575         if (__ceph_has_any_quota(ci))
 576                 ceph_adjust_quota_realms_count(inode, false);
 577
 578         /*
 579          * we may still have a snap_realm reference if there are stray
 580          * caps in i_snap_caps.
 581          */
 582         if (ci->i_snap_realm) {
 583                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 584                         struct ceph_snap_realm *realm = ci->i_snap_realm;
 585                         dout(" dropping residual ref to snap realm %p\n",
 586                              realm);
 587                         spin_lock(&realm->inodes_with_caps_lock);
 588                         list_del_init(&ci->i_snap_realm_item);
 589                         ci->i_snap_realm = NULL;
 590                         if (realm->ino == ci->i_vino.ino)
 591                                 realm->inode = NULL;
 592                         spin_unlock(&realm->inodes_with_caps_lock);
 593                         ceph_put_snap_realm(mdsc, realm);
 594                 } else {
 595                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 596                         ci->i_snap_realm = NULL;
 597                 }
 598         }
 599
 600         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 601                 frag = rb_entry(n, struct ceph_inode_frag, node);
 602                 rb_erase(n, &ci->i_fragtree);
 603                 kfree(frag);
 604         }
 605         ci->i_fragtree_nsplits = 0;
 606
 607         __ceph_destroy_xattrs(ci);
 608         if (ci->i_xattrs.blob)
 609                 ceph_buffer_put(ci->i_xattrs.blob);
 610         if (ci->i_xattrs.prealloc_blob)
 611                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 612
 613         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 614         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
 615 }
 616
 617 static inline blkcnt_t calc_inode_blocks(u64 size)
 618 {
 619         return (size + (1<<9) - 1) >> 9;
 620 }
 621
 622 /*
 623  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 624  * careful because either the client or MDS may have more up to date
 625  * info, depending on which capabilities are held, and whether
 626  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 627  * and size are monotonically increasing, except when utimes() or
 628  * truncate() increments the corresponding _seq values.)
 629  */
 630 int ceph_fill_file_size(struct inode *inode, int issued,
 631                         u32 truncate_seq, u64 truncate_size, u64 size)
 632 {
 633         struct ceph_inode_info *ci = ceph_inode(inode);
 634         int queue_trunc = 0;
 635         loff_t isize = i_size_read(inode);
 636
 637         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 638             (truncate_seq == ci->i_truncate_seq && size > isize)) {
 639                 dout("size %lld -> %llu\n", isize, size);
 640                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 641                         pr_err("fill_file_size non-zero size for directory\n");
 642                         size = 0;
 643                 }
 644                 i_size_write(inode, size);
 645                 inode->i_blocks = calc_inode_blocks(size);
 646                 ci->i_reported_size = size;
 647                 if (truncate_seq != ci->i_truncate_seq) {
 648                         dout("truncate_seq %u -> %u\n",
 649                              ci->i_truncate_seq, truncate_seq);
 650                         ci->i_truncate_seq = truncate_seq;
 651
 652                         /* the MDS should have revoked these caps */
 653                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
 654                                                CEPH_CAP_FILE_RD |
 655                                                CEPH_CAP_FILE_WR |
 656                                                CEPH_CAP_FILE_LAZYIO));
 657                         /*
 658                          * If we hold relevant caps, or in the case where we're
 659                          * not the only client referencing this file and we
 660                          * don't hold those caps, then we need to check whether
 661                          * the file is either opened or mmaped
 662                          */
 663                         if ((issued & (CEPH_CAP_FILE_CACHE|
 664                                        CEPH_CAP_FILE_BUFFER)) ||
 665                             mapping_mapped(inode->i_mapping) ||
 666                             __ceph_is_file_opened(ci)) {
 667                                 ci->i_truncate_pending++;
 668                                 queue_trunc = 1;
 669                         }
 670                 }
 671         }
 672         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 673             ci->i_truncate_size != truncate_size) {
 674                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 675                      truncate_size);
 676                 ci->i_truncate_size = truncate_size;
 677         }
 678
 679         if (queue_trunc)
 680                 ceph_fscache_invalidate(inode);
 681
 682         return queue_trunc;
 683 }
 684
 685 void ceph_fill_file_time(struct inode *inode, int issued,
 686                          u64 time_warp_seq, struct timespec64 *ctime,
 687                          struct timespec64 *mtime, struct timespec64 *atime)
 688 {
 689         struct ceph_inode_info *ci = ceph_inode(inode);
 690         int warn = 0;
 691
 692         if (issued & (CEPH_CAP_FILE_EXCL|
 693                       CEPH_CAP_FILE_WR|
 694                       CEPH_CAP_FILE_BUFFER|
 695                       CEPH_CAP_AUTH_EXCL|
 696                       CEPH_CAP_XATTR_EXCL)) {
 697                 if (ci->i_version == 0 ||
 698                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 699                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 700                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 701                              ctime->tv_sec, ctime->tv_nsec);
 702                         inode->i_ctime = *ctime;
 703                 }
 704                 if (ci->i_version == 0 ||
 705                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 706                         /* the MDS did a utimes() */
 707                         dout("mtime %lld.%09ld -> %lld.%09ld "
 708                              "tw %d -> %d\n",
 709                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 710                              mtime->tv_sec, mtime->tv_nsec,
 711                              ci->i_time_warp_seq, (int)time_warp_seq);
 712
 713                         inode->i_mtime = *mtime;
 714                         inode->i_atime = *atime;
 715                         ci->i_time_warp_seq = time_warp_seq;
 716                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 717                         /* nobody did utimes(); take the max */
 718                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 719                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 720                                      inode->i_mtime.tv_sec,
 721                                      inode->i_mtime.tv_nsec,
 722                                      mtime->tv_sec, mtime->tv_nsec);
 723                                 inode->i_mtime = *mtime;
 724                         }
 725                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 726                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 727                                      inode->i_atime.tv_sec,
 728                                      inode->i_atime.tv_nsec,
 729                                      atime->tv_sec, atime->tv_nsec);
 730                                 inode->i_atime = *atime;
 731                         }
 732                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 733                         /* we did a utimes(); ignore mds values */
 734                 } else {
 735                         warn = 1;
 736                 }
 737         } else {
 738                 /* we have no write|excl caps; whatever the MDS says is true */
 739                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 740                         inode->i_ctime = *ctime;
 741                         inode->i_mtime = *mtime;
 742                         inode->i_atime = *atime;
 743                         ci->i_time_warp_seq = time_warp_seq;
 744                 } else {
 745                         warn = 1;
 746                 }
 747         }
 748         if (warn) /* time_warp_seq shouldn't go backwards */
 749                 dout("%p mds time_warp_seq %llu < %u\n",
 750                      inode, time_warp_seq, ci->i_time_warp_seq);
 751 }
 752
 753 /*
 754  * Populate an inode based on info from mds.  May be called on new or
 755  * existing inodes.
 756  */
 757 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 758                     struct ceph_mds_reply_info_in *iinfo,
 759                     struct ceph_mds_reply_dirfrag *dirinfo,
 760                     struct ceph_mds_session *session, int cap_fmode,
 761                     struct ceph_cap_reservation *caps_reservation)
 762 {
 763         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 764         struct ceph_mds_reply_inode *info = iinfo->in;
 765         struct ceph_inode_info *ci = ceph_inode(inode);
 766         int issued, new_issued, info_caps;
 767         struct timespec64 mtime, atime, ctime;
 768         struct ceph_buffer *xattr_blob = NULL;
 769         struct ceph_buffer *old_blob = NULL;
 770         struct ceph_string *pool_ns = NULL;
 771         struct ceph_cap *new_cap = NULL;
 772         int err = 0;
 773         bool wake = false;
 774         bool queue_trunc = false;
 775         bool new_version = false;
 776         bool fill_inline = false;
 777         umode_t mode = le32_to_cpu(info->mode);
 778         dev_t rdev = le32_to_cpu(info->rdev);
 779
 780         lockdep_assert_held(&mdsc->snap_rwsem);
 781
 782         dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
 783              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 784              ci->i_version);
 785
 786         /* Once I_NEW is cleared, we can't change type or dev numbers */
 787         if (inode->i_state & I_NEW) {
 788                 inode->i_mode = mode;
 789         } else {
 790                 if (inode_wrong_type(inode, mode)) {
 791                         pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
 792                                      ceph_vinop(inode), inode->i_mode, mode);
 793                         return -ESTALE;
 794                 }
 795
 796                 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
 797                         pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
 798                                      ceph_vinop(inode), MAJOR(inode->i_rdev),
 799                                      MINOR(inode->i_rdev), MAJOR(rdev),
 800                                      MINOR(rdev));
 801                         return -ESTALE;
 802                 }
 803         }
 804
 805         info_caps = le32_to_cpu(info->cap.caps);
 806
 807         /* prealloc new cap struct */
 808         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 809                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 810                 if (!new_cap)
 811                         return -ENOMEM;
 812         }
 813
 814         /*
 815          * prealloc xattr data, if it looks like we'll need it.  only
 816          * if len > 4 (meaning there are actually xattrs; the first 4
 817          * bytes are the xattr count).
 818          */
 819         if (iinfo->xattr_len > 4) {
 820                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 821                 if (!xattr_blob)
 822                         pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
 823                                iinfo->xattr_len);
 824         }
 825
 826         if (iinfo->pool_ns_len > 0)
 827                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 828                                                      iinfo->pool_ns_len);
 829
 830         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 831                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 832
 833         spin_lock(&ci->i_ceph_lock);
 834
 835         /*
 836          * provided version will be odd if inode value is projected,
 837          * even if stable.  skip the update if we have newer stable
 838          * info (ours>=theirs, e.g. due to racing mds replies), unless
 839          * we are getting projected (unstable) info (in which case the
 840          * version is odd, and we want ours>theirs).
 841          *   us   them
 842          *   2    2     skip
 843          *   3    2     skip
 844          *   3    3     update
 845          */
 846         if (ci->i_version == 0 ||
 847             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 848              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 849                 new_version = true;
 850
 851         /* Update change_attribute */
 852         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 853
 854         __ceph_caps_issued(ci, &issued);
 855         issued |= __ceph_caps_dirty(ci);
 856         new_issued = ~issued & info_caps;
 857
 858         /* directories have fl_stripe_unit set to zero */
 859         if (le32_to_cpu(info->layout.fl_stripe_unit))
 860                 inode->i_blkbits =
 861                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 862         else
 863                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 864
 865         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 866
 867         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 868             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 869                 inode->i_mode = mode;
 870                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 871                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 872                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 873                      from_kuid(&init_user_ns, inode->i_uid),
 874                      from_kgid(&init_user_ns, inode->i_gid));
 875                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 876                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 877         }
 878
 879         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 880             (issued & CEPH_CAP_LINK_EXCL) == 0)
 881                 set_nlink(inode, le32_to_cpu(info->nlink));
 882
 883         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 884                 /* be careful with mtime, atime, size */
 885                 ceph_decode_timespec64(&atime, &info->atime);
 886                 ceph_decode_timespec64(&mtime, &info->mtime);
 887                 ceph_decode_timespec64(&ctime, &info->ctime);
 888                 ceph_fill_file_time(inode, issued,
 889                                 le32_to_cpu(info->time_warp_seq),
 890                                 &ctime, &mtime, &atime);
 891         }
 892
 893         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 894                 ci->i_files = le64_to_cpu(info->files);
 895                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 896         }
 897
 898         if (new_version ||
 899             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 900                 s64 old_pool = ci->i_layout.pool_id;
 901                 struct ceph_string *old_ns;
 902
 903                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 904                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 905                                         lockdep_is_held(&ci->i_ceph_lock));
 906                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 907
 908                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 909                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 910
 911                 pool_ns = old_ns;
 912
 913                 queue_trunc = ceph_fill_file_size(inode, issued,
 914                                         le32_to_cpu(info->truncate_seq),
 915                                         le64_to_cpu(info->truncate_size),
 916                                         le64_to_cpu(info->size));
 917                 /* only update max_size on auth cap */
 918                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 919                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 920                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 921                                         le64_to_cpu(info->max_size));
 922                         ci->i_max_size = le64_to_cpu(info->max_size);
 923                 }
 924         }
 925
 926         /* layout and rstat are not tracked by capability, update them if
 927          * the inode info is from auth mds */
 928         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 929                 if (S_ISDIR(inode->i_mode)) {
 930                         ci->i_dir_layout = iinfo->dir_layout;
 931                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 932                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 933                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 934                         ci->i_dir_pin = iinfo->dir_pin;
 935                         ci->i_rsnaps = iinfo->rsnaps;
 936                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 937                 }
 938         }
 939
 940         /* xattrs */
 941         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 942         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 943             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 944                 if (ci->i_xattrs.blob)
 945                         old_blob = ci->i_xattrs.blob;
 946                 ci->i_xattrs.blob = xattr_blob;
 947                 if (xattr_blob)
 948                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 949                                iinfo->xattr_data, iinfo->xattr_len);
 950                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 951                 ceph_forget_all_cached_acls(inode);
 952                 ceph_security_invalidate_secctx(inode);
 953                 xattr_blob = NULL;
 954         }
 955
 956         /* finally update i_version */
 957         if (le64_to_cpu(info->version) > ci->i_version)
 958                 ci->i_version = le64_to_cpu(info->version);
 959
 960         inode->i_mapping->a_ops = &ceph_aops;
 961
 962         switch (inode->i_mode & S_IFMT) {
 963         case S_IFIFO:
 964         case S_IFBLK:
 965         case S_IFCHR:
 966         case S_IFSOCK:
 967                 inode->i_blkbits = PAGE_SHIFT;
 968                 init_special_inode(inode, inode->i_mode, rdev);
 969                 inode->i_op = &ceph_file_iops;
 970                 break;
 971         case S_IFREG:
 972                 inode->i_op = &ceph_file_iops;
 973                 inode->i_fop = &ceph_file_fops;
 974                 break;
 975         case S_IFLNK:
 976                 inode->i_op = &ceph_symlink_iops;
 977                 if (!ci->i_symlink) {
 978                         u32 symlen = iinfo->symlink_len;
 979                         char *sym;
 980
 981                         spin_unlock(&ci->i_ceph_lock);
 982
 983                         if (symlen != i_size_read(inode)) {
 984                                 pr_err("%s %llx.%llx BAD symlink "
 985                                         "size %lld\n", __func__,
 986                                         ceph_vinop(inode),
 987                                         i_size_read(inode));
 988                                 i_size_write(inode, symlen);
 989                                 inode->i_blocks = calc_inode_blocks(symlen);
 990                         }
 991
 992                         err = -ENOMEM;
 993                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 994                         if (!sym)
 995                                 goto out;
 996
 997                         spin_lock(&ci->i_ceph_lock);
 998                         if (!ci->i_symlink)
 999                                 ci->i_symlink = sym;
1000                         else
1001                                 kfree(sym); /* lost a race */
1002                 }
1003                 inode->i_link = ci->i_symlink;
1004                 break;
1005         case S_IFDIR:
1006                 inode->i_op = &ceph_dir_iops;
1007                 inode->i_fop = &ceph_dir_fops;
1008                 break;
1009         default:
1010                 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
1011                        ceph_vinop(inode), inode->i_mode);
1012         }
1013
1014         /* were we issued a capability? */
1015         if (info_caps) {
1016                 if (ceph_snap(inode) == CEPH_NOSNAP) {
1017                         ceph_add_cap(inode, session,
1018                                      le64_to_cpu(info->cap.cap_id),
1019                                      info_caps,
1020                                      le32_to_cpu(info->cap.wanted),
1021                                      le32_to_cpu(info->cap.seq),
1022                                      le32_to_cpu(info->cap.mseq),
1023                                      le64_to_cpu(info->cap.realm),
1024                                      info->cap.flags, &new_cap);
1025
1026                         /* set dir completion flag? */
1027                         if (S_ISDIR(inode->i_mode) &&
1028                             ci->i_files == 0 && ci->i_subdirs == 0 &&
1029                             (info_caps & CEPH_CAP_FILE_SHARED) &&
1030                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1031                             !__ceph_dir_is_complete(ci)) {
1032                                 dout(" marking %p complete (empty)\n", inode);
1033                                 i_size_write(inode, 0);
1034                                 __ceph_dir_set_complete(ci,
1035                                         atomic64_read(&ci->i_release_count),
1036                                         atomic64_read(&ci->i_ordered_count));
1037                         }
1038
1039                         wake = true;
1040                 } else {
1041                         dout(" %p got snap_caps %s\n", inode,
1042                              ceph_cap_string(info_caps));
1043                         ci->i_snap_caps |= info_caps;
1044                 }
1045         }
1046
1047         if (iinfo->inline_version > 0 &&
1048             iinfo->inline_version >= ci->i_inline_version) {
1049                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1050                 ci->i_inline_version = iinfo->inline_version;
1051                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1052                     (locked_page || (info_caps & cache_caps)))
1053                         fill_inline = true;
1054         }
1055
1056         if (cap_fmode >= 0) {
1057                 if (!info_caps)
1058                         pr_warn("mds issued no caps on %llx.%llx\n",
1059                                 ceph_vinop(inode));
1060                 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1061         }
1062
1063         spin_unlock(&ci->i_ceph_lock);
1064
1065         if (fill_inline)
1066                 ceph_fill_inline_data(inode, locked_page,
1067                                       iinfo->inline_data, iinfo->inline_len);
1068
1069         if (wake)
1070                 wake_up_all(&ci->i_cap_wq);
1071
1072         /* queue truncate if we saw i_size decrease */
1073         if (queue_trunc)
1074                 ceph_queue_vmtruncate(inode);
1075
1076         /* populate frag tree */
1077         if (S_ISDIR(inode->i_mode))
1078                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1079
1080         /* update delegation info? */
1081         if (dirinfo)
1082                 ceph_fill_dirfrag(inode, dirinfo);
1083
1084         err = 0;
1085 out:
1086         if (new_cap)
1087                 ceph_put_cap(mdsc, new_cap);
1088         ceph_buffer_put(old_blob);
1089         ceph_buffer_put(xattr_blob);
1090         ceph_put_string(pool_ns);
1091         return err;
1092 }
1093
1094 /*
1095  * caller should hold session s_mutex and dentry->d_lock.
1096  */
1097 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1098                                   struct ceph_mds_reply_lease *lease,
1099                                   struct ceph_mds_session *session,
1100                                   unsigned long from_time,
1101                                   struct ceph_mds_session **old_lease_session)
1102 {
1103         struct ceph_dentry_info *di = ceph_dentry(dentry);
1104         unsigned mask = le16_to_cpu(lease->mask);
1105         long unsigned duration = le32_to_cpu(lease->duration_ms);
1106         long unsigned ttl = from_time + (duration * HZ) / 1000;
1107         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1108
1109         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1110              dentry, duration, ttl);
1111
1112         /* only track leases on regular dentries */
1113         if (ceph_snap(dir) != CEPH_NOSNAP)
1114                 return;
1115
1116         if (mask & CEPH_LEASE_PRIMARY_LINK)
1117                 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1118         else
1119                 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1120
1121         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1122         if (!(mask & CEPH_LEASE_VALID)) {
1123                 __ceph_dentry_dir_lease_touch(di);
1124                 return;
1125         }
1126
1127         if (di->lease_gen == session->s_cap_gen &&
1128             time_before(ttl, di->time))
1129                 return;  /* we already have a newer lease. */
1130
1131         if (di->lease_session && di->lease_session != session) {
1132                 *old_lease_session = di->lease_session;
1133                 di->lease_session = NULL;
1134         }
1135
1136         if (!di->lease_session)
1137                 di->lease_session = ceph_get_mds_session(session);
1138         di->lease_gen = session->s_cap_gen;
1139         di->lease_seq = le32_to_cpu(lease->seq);
1140         di->lease_renew_after = half_ttl;
1141         di->lease_renew_from = 0;
1142         di->time = ttl;
1143
1144         __ceph_dentry_lease_touch(di);
1145 }
1146
1147 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1148                                         struct ceph_mds_reply_lease *lease,
1149                                         struct ceph_mds_session *session,
1150                                         unsigned long from_time)
1151 {
1152         struct ceph_mds_session *old_lease_session = NULL;
1153         spin_lock(&dentry->d_lock);
1154         __update_dentry_lease(dir, dentry, lease, session, from_time,
1155                               &old_lease_session);
1156         spin_unlock(&dentry->d_lock);
1157         if (old_lease_session)
1158                 ceph_put_mds_session(old_lease_session);
1159 }
1160
1161 /*
1162  * update dentry lease without having parent inode locked
1163  */
1164 static void update_dentry_lease_careful(struct dentry *dentry,
1165                                         struct ceph_mds_reply_lease *lease,
1166                                         struct ceph_mds_session *session,
1167                                         unsigned long from_time,
1168                                         char *dname, u32 dname_len,
1169                                         struct ceph_vino *pdvino,
1170                                         struct ceph_vino *ptvino)
1171
1172 {
1173         struct inode *dir;
1174         struct ceph_mds_session *old_lease_session = NULL;
1175
1176         spin_lock(&dentry->d_lock);
1177         /* make sure dentry's name matches target */
1178         if (dentry->d_name.len != dname_len ||
1179             memcmp(dentry->d_name.name, dname, dname_len))
1180                 goto out_unlock;
1181
1182         dir = d_inode(dentry->d_parent);
1183         /* make sure parent matches dvino */
1184         if (!ceph_ino_compare(dir, pdvino))
1185                 goto out_unlock;
1186
1187         /* make sure dentry's inode matches target. NULL ptvino means that
1188          * we expect a negative dentry */
1189         if (ptvino) {
1190                 if (d_really_is_negative(dentry))
1191                         goto out_unlock;
1192                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1193                         goto out_unlock;
1194         } else {
1195                 if (d_really_is_positive(dentry))
1196                         goto out_unlock;
1197         }
1198
1199         __update_dentry_lease(dir, dentry, lease, session,
1200                               from_time, &old_lease_session);
1201 out_unlock:
1202         spin_unlock(&dentry->d_lock);
1203         if (old_lease_session)
1204                 ceph_put_mds_session(old_lease_session);
1205 }
1206
1207 /*
1208  * splice a dentry to an inode.
1209  * caller must hold directory i_mutex for this to be safe.
1210  */
1211 static int splice_dentry(struct dentry **pdn, struct inode *in)
1212 {
1213         struct dentry *dn = *pdn;
1214         struct dentry *realdn;
1215
1216         BUG_ON(d_inode(dn));
1217
1218         if (S_ISDIR(in->i_mode)) {
1219                 /* If inode is directory, d_splice_alias() below will remove
1220                  * 'realdn' from its origin parent. We need to ensure that
1221                  * origin parent's readdir cache will not reference 'realdn'
1222                  */
1223                 realdn = d_find_any_alias(in);
1224                 if (realdn) {
1225                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1226                         spin_lock(&realdn->d_lock);
1227
1228                         realdn->d_op->d_prune(realdn);
1229
1230                         di->time = jiffies;
1231                         di->lease_shared_gen = 0;
1232                         di->offset = 0;
1233
1234                         spin_unlock(&realdn->d_lock);
1235                         dput(realdn);
1236                 }
1237         }
1238
1239         /* dn must be unhashed */
1240         if (!d_unhashed(dn))
1241                 d_drop(dn);
1242         realdn = d_splice_alias(in, dn);
1243         if (IS_ERR(realdn)) {
1244                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1245                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1246                 return PTR_ERR(realdn);
1247         }
1248
1249         if (realdn) {
1250                 dout("dn %p (%d) spliced with %p (%d) "
1251                      "inode %p ino %llx.%llx\n",
1252                      dn, d_count(dn),
1253                      realdn, d_count(realdn),
1254                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1255                 dput(dn);
1256                 *pdn = realdn;
1257         } else {
1258                 BUG_ON(!ceph_dentry(dn));
1259                 dout("dn %p attached to %p ino %llx.%llx\n",
1260                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1261         }
1262         return 0;
1263 }
1264
1265 /*
1266  * Incorporate results into the local cache.  This is either just
1267  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1268  * after a lookup).
1269  *
1270  * A reply may contain
1271  *         a directory inode along with a dentry.
1272  *  and/or a target inode
1273  *
1274  * Called with snap_rwsem (read).
1275  */
1276 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1277 {
1278         struct ceph_mds_session *session = req->r_session;
1279         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1280         struct inode *in = NULL;
1281         struct ceph_vino tvino, dvino;
1282         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1283         int err = 0;
1284
1285         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1286              rinfo->head->is_dentry, rinfo->head->is_target);
1287
1288         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1289                 dout("fill_trace reply is empty!\n");
1290                 if (rinfo->head->result == 0 && req->r_parent)
1291                         ceph_invalidate_dir_request(req);
1292                 return 0;
1293         }
1294
1295         if (rinfo->head->is_dentry) {
1296                 struct inode *dir = req->r_parent;
1297
1298                 if (dir) {
1299                         err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1300                                               rinfo->dirfrag, session, -1,
1301                                               &req->r_caps_reservation);
1302                         if (err < 0)
1303                                 goto done;
1304                 } else {
1305                         WARN_ON_ONCE(1);
1306                 }
1307
1308                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1309                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1310                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1311                         struct qstr dname;
1312                         struct dentry *dn, *parent;
1313
1314                         BUG_ON(!rinfo->head->is_target);
1315                         BUG_ON(req->r_dentry);
1316
1317                         parent = d_find_any_alias(dir);
1318                         BUG_ON(!parent);
1319
1320                         dname.name = rinfo->dname;
1321                         dname.len = rinfo->dname_len;
1322                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1323                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1324                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1325 retry_lookup:
1326                         dn = d_lookup(parent, &dname);
1327                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1328                              parent, dname.len, dname.name, dn);
1329
1330                         if (!dn) {
1331                                 dn = d_alloc(parent, &dname);
1332                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1333                                      dname.len, dname.name, dn);
1334                                 if (!dn) {
1335                                         dput(parent);
1336                                         err = -ENOMEM;
1337                                         goto done;
1338                                 }
1339                                 err = 0;
1340                         } else if (d_really_is_positive(dn) &&
1341                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1342                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1343                                 dout(" dn %p points to wrong inode %p\n",
1344                                      dn, d_inode(dn));
1345                                 ceph_dir_clear_ordered(dir);
1346                                 d_delete(dn);
1347                                 dput(dn);
1348                                 goto retry_lookup;
1349                         }
1350
1351                         req->r_dentry = dn;
1352                         dput(parent);
1353                 }
1354         }
1355
1356         if (rinfo->head->is_target) {
1357                 /* Should be filled in by handle_reply */
1358                 BUG_ON(!req->r_target_inode);
1359
1360                 in = req->r_target_inode;
1361                 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1362                                 NULL, session,
1363                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1364                                  !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1365                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1366                                 &req->r_caps_reservation);
1367                 if (err < 0) {
1368                         pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1369                                 in, ceph_vinop(in));
1370                         req->r_target_inode = NULL;
1371                         if (in->i_state & I_NEW)
1372                                 discard_new_inode(in);
1373                         else
1374                                 iput(in);
1375                         goto done;
1376                 }
1377                 if (in->i_state & I_NEW)
1378                         unlock_new_inode(in);
1379         }
1380
1381         /*
1382          * ignore null lease/binding on snapdir ENOENT, or else we
1383          * will have trouble splicing in the virtual snapdir later
1384          */
1385         if (rinfo->head->is_dentry &&
1386             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1387             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1388             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1389                                                fsc->mount_options->snapdir_name,
1390                                                req->r_dentry->d_name.len))) {
1391                 /*
1392                  * lookup link rename   : null -> possibly existing inode
1393                  * mknod symlink mkdir  : null -> new inode
1394                  * unlink               : linked -> null
1395                  */
1396                 struct inode *dir = req->r_parent;
1397                 struct dentry *dn = req->r_dentry;
1398                 bool have_dir_cap, have_lease;
1399
1400                 BUG_ON(!dn);
1401                 BUG_ON(!dir);
1402                 BUG_ON(d_inode(dn->d_parent) != dir);
1403
1404                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1405                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1406
1407                 BUG_ON(ceph_ino(dir) != dvino.ino);
1408                 BUG_ON(ceph_snap(dir) != dvino.snap);
1409
1410                 /* do we have a lease on the whole dir? */
1411                 have_dir_cap =
1412                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1413                          CEPH_CAP_FILE_SHARED);
1414
1415                 /* do we have a dn lease? */
1416                 have_lease = have_dir_cap ||
1417                         le32_to_cpu(rinfo->dlease->duration_ms);
1418                 if (!have_lease)
1419                         dout("fill_trace  no dentry lease or dir cap\n");
1420
1421                 /* rename? */
1422                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1423                         struct inode *olddir = req->r_old_dentry_dir;
1424                         BUG_ON(!olddir);
1425
1426                         dout(" src %p '%pd' dst %p '%pd'\n",
1427                              req->r_old_dentry,
1428                              req->r_old_dentry,
1429                              dn, dn);
1430                         dout("fill_trace doing d_move %p -> %p\n",
1431                              req->r_old_dentry, dn);
1432
1433                         /* d_move screws up sibling dentries' offsets */
1434                         ceph_dir_clear_ordered(dir);
1435                         ceph_dir_clear_ordered(olddir);
1436
1437                         d_move(req->r_old_dentry, dn);
1438                         dout(" src %p '%pd' dst %p '%pd'\n",
1439                              req->r_old_dentry,
1440                              req->r_old_dentry,
1441                              dn, dn);
1442
1443                         /* ensure target dentry is invalidated, despite
1444                            rehashing bug in vfs_rename_dir */
1445                         ceph_invalidate_dentry_lease(dn);
1446
1447                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1448                              ceph_dentry(req->r_old_dentry)->offset);
1449
1450                         /* swap r_dentry and r_old_dentry in case that
1451                          * splice_dentry() gets called later. This is safe
1452                          * because no other place will use them */
1453                         req->r_dentry = req->r_old_dentry;
1454                         req->r_old_dentry = dn;
1455                         dn = req->r_dentry;
1456                 }
1457
1458                 /* null dentry? */
1459                 if (!rinfo->head->is_target) {
1460                         dout("fill_trace null dentry\n");
1461                         if (d_really_is_positive(dn)) {
1462                                 dout("d_delete %p\n", dn);
1463                                 ceph_dir_clear_ordered(dir);
1464                                 d_delete(dn);
1465                         } else if (have_lease) {
1466                                 if (d_unhashed(dn))
1467                                         d_add(dn, NULL);
1468                                 update_dentry_lease(dir, dn,
1469                                                     rinfo->dlease, session,
1470                                                     req->r_request_started);
1471                         }
1472                         goto done;
1473                 }
1474
1475                 /* attach proper inode */
1476                 if (d_really_is_negative(dn)) {
1477                         ceph_dir_clear_ordered(dir);
1478                         ihold(in);
1479                         err = splice_dentry(&req->r_dentry, in);
1480                         if (err < 0)
1481                                 goto done;
1482                         dn = req->r_dentry;  /* may have spliced */
1483                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1484                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1485                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1486                              ceph_vinop(in));
1487                         d_invalidate(dn);
1488                         have_lease = false;
1489                 }
1490
1491                 if (have_lease) {
1492                         update_dentry_lease(dir, dn,
1493                                             rinfo->dlease, session,
1494                                             req->r_request_started);
1495                 }
1496                 dout(" final dn %p\n", dn);
1497         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1498                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1499                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1500                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1501                 struct inode *dir = req->r_parent;
1502
1503                 /* fill out a snapdir LOOKUPSNAP dentry */
1504                 BUG_ON(!dir);
1505                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1506                 BUG_ON(!req->r_dentry);
1507                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1508                 ceph_dir_clear_ordered(dir);
1509                 ihold(in);
1510                 err = splice_dentry(&req->r_dentry, in);
1511                 if (err < 0)
1512                         goto done;
1513         } else if (rinfo->head->is_dentry && req->r_dentry) {
1514                 /* parent inode is not locked, be carefull */
1515                 struct ceph_vino *ptvino = NULL;
1516                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1517                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1518                 if (rinfo->head->is_target) {
1519                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1520                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1521                         ptvino = &tvino;
1522                 }
1523                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1524                                             session, req->r_request_started,
1525                                             rinfo->dname, rinfo->dname_len,
1526                                             &dvino, ptvino);
1527         }
1528 done:
1529         dout("fill_trace done err=%d\n", err);
1530         return err;
1531 }
1532
1533 /*
1534  * Prepopulate our cache with readdir results, leases, etc.
1535  */
1536 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1537                                            struct ceph_mds_session *session)
1538 {
1539         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1540         int i, err = 0;
1541
1542         for (i = 0; i < rinfo->dir_nr; i++) {
1543                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1544                 struct ceph_vino vino;
1545                 struct inode *in;
1546                 int rc;
1547
1548                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1549                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1550
1551                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1552                 if (IS_ERR(in)) {
1553                         err = PTR_ERR(in);
1554                         dout("new_inode badness got %d\n", err);
1555                         continue;
1556                 }
1557                 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1558                                      -1, &req->r_caps_reservation);
1559                 if (rc < 0) {
1560                         pr_err("ceph_fill_inode badness on %p got %d\n",
1561                                in, rc);
1562                         err = rc;
1563                         if (in->i_state & I_NEW) {
1564                                 ihold(in);
1565                                 discard_new_inode(in);
1566                         }
1567                 } else if (in->i_state & I_NEW) {
1568                         unlock_new_inode(in);
1569                 }
1570
1571                 /* avoid calling iput_final() in mds dispatch threads */
1572                 ceph_async_iput(in);
1573         }
1574
1575         return err;
1576 }
1577
1578 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1579 {
1580         if (ctl->page) {
1581                 kunmap(ctl->page);
1582                 put_page(ctl->page);
1583                 ctl->page = NULL;
1584         }
1585 }
1586
1587 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1588                               struct ceph_readdir_cache_control *ctl,
1589                               struct ceph_mds_request *req)
1590 {
1591         struct ceph_inode_info *ci = ceph_inode(dir);
1592         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1593         unsigned idx = ctl->index % nsize;
1594         pgoff_t pgoff = ctl->index / nsize;
1595
1596         if (!ctl->page || pgoff != page_index(ctl->page)) {
1597                 ceph_readdir_cache_release(ctl);
1598                 if (idx == 0)
1599                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1600                 else
1601                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1602                 if (!ctl->page) {
1603                         ctl->index = -1;
1604                         return idx == 0 ? -ENOMEM : 0;
1605                 }
1606                 /* reading/filling the cache are serialized by
1607                  * i_mutex, no need to use page lock */
1608                 unlock_page(ctl->page);
1609                 ctl->dentries = kmap(ctl->page);
1610                 if (idx == 0)
1611                         memset(ctl->dentries, 0, PAGE_SIZE);
1612         }
1613
1614         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1615             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1616                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1617                 ctl->dentries[idx] = dn;
1618                 ctl->index++;
1619         } else {
1620                 dout("disable readdir cache\n");
1621                 ctl->index = -1;
1622         }
1623         return 0;
1624 }
1625
1626 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1627                              struct ceph_mds_session *session)
1628 {
1629         struct dentry *parent = req->r_dentry;
1630         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1631         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1632         struct qstr dname;
1633         struct dentry *dn;
1634         struct inode *in;
1635         int err = 0, skipped = 0, ret, i;
1636         u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1637         u32 last_hash = 0;
1638         u32 fpos_offset;
1639         struct ceph_readdir_cache_control cache_ctl = {};
1640
1641         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1642                 return readdir_prepopulate_inodes_only(req, session);
1643
1644         if (rinfo->hash_order) {
1645                 if (req->r_path2) {
1646                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1647                                                   req->r_path2,
1648                                                   strlen(req->r_path2));
1649                         last_hash = ceph_frag_value(last_hash);
1650                 } else if (rinfo->offset_hash) {
1651                         /* mds understands offset_hash */
1652                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1653                         last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1654                 }
1655         }
1656
1657         if (rinfo->dir_dir &&
1658             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1659                 dout("readdir_prepopulate got new frag %x -> %x\n",
1660                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1661                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1662                 if (!rinfo->hash_order)
1663                         req->r_readdir_offset = 2;
1664         }
1665
1666         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1667                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1668                      rinfo->dir_nr, parent);
1669         } else {
1670                 dout("readdir_prepopulate %d items under dn %p\n",
1671                      rinfo->dir_nr, parent);
1672                 if (rinfo->dir_dir)
1673                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1674
1675                 if (ceph_frag_is_leftmost(frag) &&
1676                     req->r_readdir_offset == 2 &&
1677                     !(rinfo->hash_order && last_hash)) {
1678                         /* note dir version at start of readdir so we can
1679                          * tell if any dentries get dropped */
1680                         req->r_dir_release_cnt =
1681                                 atomic64_read(&ci->i_release_count);
1682                         req->r_dir_ordered_cnt =
1683                                 atomic64_read(&ci->i_ordered_count);
1684                         req->r_readdir_cache_idx = 0;
1685                 }
1686         }
1687
1688         cache_ctl.index = req->r_readdir_cache_idx;
1689         fpos_offset = req->r_readdir_offset;
1690
1691         /* FIXME: release caps/leases if error occurs */
1692         for (i = 0; i < rinfo->dir_nr; i++) {
1693                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1694                 struct ceph_vino tvino;
1695
1696                 dname.name = rde->name;
1697                 dname.len = rde->name_len;
1698                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1699
1700                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1701                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1702
1703                 if (rinfo->hash_order) {
1704                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1705                                                  rde->name, rde->name_len);
1706                         hash = ceph_frag_value(hash);
1707                         if (hash != last_hash)
1708                                 fpos_offset = 2;
1709                         last_hash = hash;
1710                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1711                 } else {
1712                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1713                 }
1714
1715 retry_lookup:
1716                 dn = d_lookup(parent, &dname);
1717                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1718                      parent, dname.len, dname.name, dn);
1719
1720                 if (!dn) {
1721                         dn = d_alloc(parent, &dname);
1722                         dout("d_alloc %p '%.*s' = %p\n", parent,
1723                              dname.len, dname.name, dn);
1724                         if (!dn) {
1725                                 dout("d_alloc badness\n");
1726                                 err = -ENOMEM;
1727                                 goto out;
1728                         }
1729                 } else if (d_really_is_positive(dn) &&
1730                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1731                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1732                         struct ceph_dentry_info *di = ceph_dentry(dn);
1733                         dout(" dn %p points to wrong inode %p\n",
1734                              dn, d_inode(dn));
1735
1736                         spin_lock(&dn->d_lock);
1737                         if (di->offset > 0 &&
1738                             di->lease_shared_gen ==
1739                             atomic_read(&ci->i_shared_gen)) {
1740                                 __ceph_dir_clear_ordered(ci);
1741                                 di->offset = 0;
1742                         }
1743                         spin_unlock(&dn->d_lock);
1744
1745                         d_delete(dn);
1746                         dput(dn);
1747                         goto retry_lookup;
1748                 }
1749
1750                 /* inode */
1751                 if (d_really_is_positive(dn)) {
1752                         in = d_inode(dn);
1753                 } else {
1754                         in = ceph_get_inode(parent->d_sb, tvino);
1755                         if (IS_ERR(in)) {
1756                                 dout("new_inode badness\n");
1757                                 d_drop(dn);
1758                                 dput(dn);
1759                                 err = PTR_ERR(in);
1760                                 goto out;
1761                         }
1762                 }
1763
1764                 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1765                                       -1, &req->r_caps_reservation);
1766                 if (ret < 0) {
1767                         pr_err("ceph_fill_inode badness on %p\n", in);
1768                         if (d_really_is_negative(dn)) {
1769                                 /* avoid calling iput_final() in mds
1770                                  * dispatch threads */
1771                                 if (in->i_state & I_NEW) {
1772                                         ihold(in);
1773                                         discard_new_inode(in);
1774                                 }
1775                                 ceph_async_iput(in);
1776                         }
1777                         d_drop(dn);
1778                         err = ret;
1779                         goto next_item;
1780                 }
1781                 if (in->i_state & I_NEW)
1782                         unlock_new_inode(in);
1783
1784                 if (d_really_is_negative(dn)) {
1785                         if (ceph_security_xattr_deadlock(in)) {
1786                                 dout(" skip splicing dn %p to inode %p"
1787                                      " (security xattr deadlock)\n", dn, in);
1788                                 ceph_async_iput(in);
1789                                 skipped++;
1790                                 goto next_item;
1791                         }
1792
1793                         err = splice_dentry(&dn, in);
1794                         if (err < 0)
1795                                 goto next_item;
1796                 }
1797
1798                 ceph_dentry(dn)->offset = rde->offset;
1799
1800                 update_dentry_lease(d_inode(parent), dn,
1801                                     rde->lease, req->r_session,
1802                                     req->r_request_started);
1803
1804                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1805                         ret = fill_readdir_cache(d_inode(parent), dn,
1806                                                  &cache_ctl, req);
1807                         if (ret < 0)
1808                                 err = ret;
1809                 }
1810 next_item:
1811                 dput(dn);
1812         }
1813 out:
1814         if (err == 0 && skipped == 0) {
1815                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1816                 req->r_readdir_cache_idx = cache_ctl.index;
1817         }
1818         ceph_readdir_cache_release(&cache_ctl);
1819         dout("readdir_prepopulate done\n");
1820         return err;
1821 }
1822
1823 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1824 {
1825         struct ceph_inode_info *ci = ceph_inode(inode);
1826         bool ret;
1827
1828         spin_lock(&ci->i_ceph_lock);
1829         dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
1830         i_size_write(inode, size);
1831         inode->i_blocks = calc_inode_blocks(size);
1832
1833         ret = __ceph_should_report_size(ci);
1834
1835         spin_unlock(&ci->i_ceph_lock);
1836         return ret;
1837 }
1838
1839 /*
1840  * Put reference to inode, but avoid calling iput_final() in current thread.
1841  * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1842  * some contexts.
1843  */
1844 void ceph_async_iput(struct inode *inode)
1845 {
1846         if (!inode)
1847                 return;
1848         for (;;) {
1849                 if (atomic_add_unless(&inode->i_count, -1, 1))
1850                         break;
1851                 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1852                                &ceph_inode(inode)->i_work))
1853                         break;
1854                 /* queue work failed, i_count must be at least 2 */
1855         }
1856 }
1857
1858 void ceph_queue_inode_work(struct inode *inode, int work_bit)
1859 {
1860         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1861         struct ceph_inode_info *ci = ceph_inode(inode);
1862         set_bit(work_bit, &ci->i_work_mask);
1863
1864         ihold(inode);
1865         if (queue_work(fsc->inode_wq, &ci->i_work)) {
1866                 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask);
1867         } else {
1868                 dout("queue_inode_work %p already queued, mask=%lx\n",
1869                      inode, ci->i_work_mask);
1870                 iput(inode);
1871         }
1872 }
1873
1874 static void ceph_do_invalidate_pages(struct inode *inode)
1875 {
1876         struct ceph_inode_info *ci = ceph_inode(inode);
1877         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1878         u32 orig_gen;
1879         int check = 0;
1880
1881         mutex_lock(&ci->i_truncate_mutex);
1882
1883         if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1884                 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1885                                     inode, ceph_ino(inode));
1886                 mapping_set_error(inode->i_mapping, -EIO);
1887                 truncate_pagecache(inode, 0);
1888                 mutex_unlock(&ci->i_truncate_mutex);
1889                 goto out;
1890         }
1891
1892         spin_lock(&ci->i_ceph_lock);
1893         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1894              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1895         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1896                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1897                         check = 1;
1898                 spin_unlock(&ci->i_ceph_lock);
1899                 mutex_unlock(&ci->i_truncate_mutex);
1900                 goto out;
1901         }
1902         orig_gen = ci->i_rdcache_gen;
1903         spin_unlock(&ci->i_ceph_lock);
1904
1905         ceph_fscache_invalidate(inode);
1906         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1907                 pr_err("invalidate_pages %p fails\n", inode);
1908         }
1909
1910         spin_lock(&ci->i_ceph_lock);
1911         if (orig_gen == ci->i_rdcache_gen &&
1912             orig_gen == ci->i_rdcache_revoking) {
1913                 dout("invalidate_pages %p gen %d successful\n", inode,
1914                      ci->i_rdcache_gen);
1915                 ci->i_rdcache_revoking--;
1916                 check = 1;
1917         } else {
1918                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1919                      inode, orig_gen, ci->i_rdcache_gen,
1920                      ci->i_rdcache_revoking);
1921                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1922                         check = 1;
1923         }
1924         spin_unlock(&ci->i_ceph_lock);
1925         mutex_unlock(&ci->i_truncate_mutex);
1926 out:
1927         if (check)
1928                 ceph_check_caps(ci, 0, NULL);
1929 }
1930
1931 /*
1932  * Make sure any pending truncation is applied before doing anything
1933  * that may depend on it.
1934  */
1935 void __ceph_do_pending_vmtruncate(struct inode *inode)
1936 {
1937         struct ceph_inode_info *ci = ceph_inode(inode);
1938         u64 to;
1939         int wrbuffer_refs, finish = 0;
1940
1941         mutex_lock(&ci->i_truncate_mutex);
1942 retry:
1943         spin_lock(&ci->i_ceph_lock);
1944         if (ci->i_truncate_pending == 0) {
1945                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1946                 spin_unlock(&ci->i_ceph_lock);
1947                 mutex_unlock(&ci->i_truncate_mutex);
1948                 return;
1949         }
1950
1951         /*
1952          * make sure any dirty snapped pages are flushed before we
1953          * possibly truncate them.. so write AND block!
1954          */
1955         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1956                 spin_unlock(&ci->i_ceph_lock);
1957                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1958                      inode);
1959                 filemap_write_and_wait_range(&inode->i_data, 0,
1960                                              inode->i_sb->s_maxbytes);
1961                 goto retry;
1962         }
1963
1964         /* there should be no reader or writer */
1965         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1966
1967         to = ci->i_truncate_size;
1968         wrbuffer_refs = ci->i_wrbuffer_ref;
1969         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1970              ci->i_truncate_pending, to);
1971         spin_unlock(&ci->i_ceph_lock);
1972
1973         truncate_pagecache(inode, to);
1974
1975         spin_lock(&ci->i_ceph_lock);
1976         if (to == ci->i_truncate_size) {
1977                 ci->i_truncate_pending = 0;
1978                 finish = 1;
1979         }
1980         spin_unlock(&ci->i_ceph_lock);
1981         if (!finish)
1982                 goto retry;
1983
1984         mutex_unlock(&ci->i_truncate_mutex);
1985
1986         if (wrbuffer_refs == 0)
1987                 ceph_check_caps(ci, 0, NULL);
1988
1989         wake_up_all(&ci->i_cap_wq);
1990 }
1991
1992 static void ceph_inode_work(struct work_struct *work)
1993 {
1994         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1995                                                  i_work);
1996         struct inode *inode = &ci->vfs_inode;
1997
1998         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1999                 dout("writeback %p\n", inode);
2000                 filemap_fdatawrite(&inode->i_data);
2001         }
2002         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2003                 ceph_do_invalidate_pages(inode);
2004
2005         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2006                 __ceph_do_pending_vmtruncate(inode);
2007
2008         if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
2009                 ceph_check_caps(ci, 0, NULL);
2010
2011         if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
2012                 ceph_flush_snaps(ci, NULL);
2013
2014         iput(inode);
2015 }
2016
2017 /*
2018  * symlinks
2019  */
2020 static const struct inode_operations ceph_symlink_iops = {
2021         .get_link = simple_get_link,
2022         .setattr = ceph_setattr,
2023         .getattr = ceph_getattr,
2024         .listxattr = ceph_listxattr,
2025 };
2026
2027 int __ceph_setattr(struct inode *inode, struct iattr *attr)
2028 {
2029         struct ceph_inode_info *ci = ceph_inode(inode);
2030         unsigned int ia_valid = attr->ia_valid;
2031         struct ceph_mds_request *req;
2032         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2033         struct ceph_cap_flush *prealloc_cf;
2034         int issued;
2035         int release = 0, dirtied = 0;
2036         int mask = 0;
2037         int err = 0;
2038         int inode_dirty_flags = 0;
2039         bool lock_snap_rwsem = false;
2040
2041         prealloc_cf = ceph_alloc_cap_flush();
2042         if (!prealloc_cf)
2043                 return -ENOMEM;
2044
2045         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2046                                        USE_AUTH_MDS);
2047         if (IS_ERR(req)) {
2048                 ceph_free_cap_flush(prealloc_cf);
2049                 return PTR_ERR(req);
2050         }
2051
2052         spin_lock(&ci->i_ceph_lock);
2053         issued = __ceph_caps_issued(ci, NULL);
2054
2055         if (!ci->i_head_snapc &&
2056             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2057                 lock_snap_rwsem = true;
2058                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2059                         spin_unlock(&ci->i_ceph_lock);
2060                         down_read(&mdsc->snap_rwsem);
2061                         spin_lock(&ci->i_ceph_lock);
2062                         issued = __ceph_caps_issued(ci, NULL);
2063                 }
2064         }
2065
2066         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2067
2068         if (ia_valid & ATTR_UID) {
2069                 dout("setattr %p uid %d -> %d\n", inode,
2070                      from_kuid(&init_user_ns, inode->i_uid),
2071                      from_kuid(&init_user_ns, attr->ia_uid));
2072                 if (issued & CEPH_CAP_AUTH_EXCL) {
2073                         inode->i_uid = attr->ia_uid;
2074                         dirtied |= CEPH_CAP_AUTH_EXCL;
2075                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2076                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2077                         req->r_args.setattr.uid = cpu_to_le32(
2078                                 from_kuid(&init_user_ns, attr->ia_uid));
2079                         mask |= CEPH_SETATTR_UID;
2080                         release |= CEPH_CAP_AUTH_SHARED;
2081                 }
2082         }
2083         if (ia_valid & ATTR_GID) {
2084                 dout("setattr %p gid %d -> %d\n", inode,
2085                      from_kgid(&init_user_ns, inode->i_gid),
2086                      from_kgid(&init_user_ns, attr->ia_gid));
2087                 if (issued & CEPH_CAP_AUTH_EXCL) {
2088                         inode->i_gid = attr->ia_gid;
2089                         dirtied |= CEPH_CAP_AUTH_EXCL;
2090                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2091                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2092                         req->r_args.setattr.gid = cpu_to_le32(
2093                                 from_kgid(&init_user_ns, attr->ia_gid));
2094                         mask |= CEPH_SETATTR_GID;
2095                         release |= CEPH_CAP_AUTH_SHARED;
2096                 }
2097         }
2098         if (ia_valid & ATTR_MODE) {
2099                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2100                      attr->ia_mode);
2101                 if (issued & CEPH_CAP_AUTH_EXCL) {
2102                         inode->i_mode = attr->ia_mode;
2103                         dirtied |= CEPH_CAP_AUTH_EXCL;
2104                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2105                            attr->ia_mode != inode->i_mode) {
2106                         inode->i_mode = attr->ia_mode;
2107                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2108                         mask |= CEPH_SETATTR_MODE;
2109                         release |= CEPH_CAP_AUTH_SHARED;
2110                 }
2111         }
2112
2113         if (ia_valid & ATTR_ATIME) {
2114                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2115                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2116                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2117                 if (issued & CEPH_CAP_FILE_EXCL) {
2118                         ci->i_time_warp_seq++;
2119                         inode->i_atime = attr->ia_atime;
2120                         dirtied |= CEPH_CAP_FILE_EXCL;
2121                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2122                            timespec64_compare(&inode->i_atime,
2123                                             &attr->ia_atime) < 0) {
2124                         inode->i_atime = attr->ia_atime;
2125                         dirtied |= CEPH_CAP_FILE_WR;
2126                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2127                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2128                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2129                                                &attr->ia_atime);
2130                         mask |= CEPH_SETATTR_ATIME;
2131                         release |= CEPH_CAP_FILE_SHARED |
2132                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2133                 }
2134         }
2135         if (ia_valid & ATTR_SIZE) {
2136                 loff_t isize = i_size_read(inode);
2137
2138                 dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
2139                 if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size > isize) {
2140                         i_size_write(inode, attr->ia_size);
2141                         inode->i_blocks = calc_inode_blocks(attr->ia_size);
2142                         ci->i_reported_size = attr->ia_size;
2143                         dirtied |= CEPH_CAP_FILE_EXCL;
2144                         ia_valid |= ATTR_MTIME;
2145                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2146                            attr->ia_size != isize) {
2147                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2148                         req->r_args.setattr.old_size = cpu_to_le64(isize);
2149                         mask |= CEPH_SETATTR_SIZE;
2150                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2151                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2152                 }
2153         }
2154         if (ia_valid & ATTR_MTIME) {
2155                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2156                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2157                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2158                 if (issued & CEPH_CAP_FILE_EXCL) {
2159                         ci->i_time_warp_seq++;
2160                         inode->i_mtime = attr->ia_mtime;
2161                         dirtied |= CEPH_CAP_FILE_EXCL;
2162                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2163                            timespec64_compare(&inode->i_mtime,
2164                                             &attr->ia_mtime) < 0) {
2165                         inode->i_mtime = attr->ia_mtime;
2166                         dirtied |= CEPH_CAP_FILE_WR;
2167                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2168                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2169                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2170                                                &attr->ia_mtime);
2171                         mask |= CEPH_SETATTR_MTIME;
2172                         release |= CEPH_CAP_FILE_SHARED |
2173                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2174                 }
2175         }
2176
2177         /* these do nothing */
2178         if (ia_valid & ATTR_CTIME) {
2179                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2180                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2181                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2182                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2183                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2184                      only ? "ctime only" : "ignored");
2185                 if (only) {
2186                         /*
2187                          * if kernel wants to dirty ctime but nothing else,
2188                          * we need to choose a cap to dirty under, or do
2189                          * a almost-no-op setattr
2190                          */
2191                         if (issued & CEPH_CAP_AUTH_EXCL)
2192                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2193                         else if (issued & CEPH_CAP_FILE_EXCL)
2194                                 dirtied |= CEPH_CAP_FILE_EXCL;
2195                         else if (issued & CEPH_CAP_XATTR_EXCL)
2196                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2197                         else
2198                                 mask |= CEPH_SETATTR_CTIME;
2199                 }
2200         }
2201         if (ia_valid & ATTR_FILE)
2202                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2203
2204         if (dirtied) {
2205                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2206                                                            &prealloc_cf);
2207                 inode->i_ctime = attr->ia_ctime;
2208         }
2209
2210         release &= issued;
2211         spin_unlock(&ci->i_ceph_lock);
2212         if (lock_snap_rwsem)
2213                 up_read(&mdsc->snap_rwsem);
2214
2215         if (inode_dirty_flags)
2216                 __mark_inode_dirty(inode, inode_dirty_flags);
2217
2218
2219         if (mask) {
2220                 req->r_inode = inode;
2221                 ihold(inode);
2222                 req->r_inode_drop = release;
2223                 req->r_args.setattr.mask = cpu_to_le32(mask);
2224                 req->r_num_caps = 1;
2225                 req->r_stamp = attr->ia_ctime;
2226                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2227         }
2228         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2229              ceph_cap_string(dirtied), mask);
2230
2231         ceph_mdsc_put_request(req);
2232         ceph_free_cap_flush(prealloc_cf);
2233
2234         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2235                 __ceph_do_pending_vmtruncate(inode);
2236
2237         return err;
2238 }
2239
2240 /*
2241  * setattr
2242  */
2243 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
2244                  struct iattr *attr)
2245 {
2246         struct inode *inode = d_inode(dentry);
2247         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2248         int err;
2249
2250         if (ceph_snap(inode) != CEPH_NOSNAP)
2251                 return -EROFS;
2252
2253         err = setattr_prepare(&init_user_ns, dentry, attr);
2254         if (err != 0)
2255                 return err;
2256
2257         if ((attr->ia_valid & ATTR_SIZE) &&
2258             attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2259                 return -EFBIG;
2260
2261         if ((attr->ia_valid & ATTR_SIZE) &&
2262             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2263                 return -EDQUOT;
2264
2265         err = __ceph_setattr(inode, attr);
2266
2267         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2268                 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode);
2269
2270         return err;
2271 }
2272
2273 /*
2274  * Verify that we have a lease on the given mask.  If not,
2275  * do a getattr against an mds.
2276  */
2277 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2278                       int mask, bool force)
2279 {
2280         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2281         struct ceph_mds_client *mdsc = fsc->mdsc;
2282         struct ceph_mds_request *req;
2283         int mode;
2284         int err;
2285
2286         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2287                 dout("do_getattr inode %p SNAPDIR\n", inode);
2288                 return 0;
2289         }
2290
2291         dout("do_getattr inode %p mask %s mode 0%o\n",
2292              inode, ceph_cap_string(mask), inode->i_mode);
2293         if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2294                         return 0;
2295
2296         mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2297         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2298         if (IS_ERR(req))
2299                 return PTR_ERR(req);
2300         req->r_inode = inode;
2301         ihold(inode);
2302         req->r_num_caps = 1;
2303         req->r_args.getattr.mask = cpu_to_le32(mask);
2304         req->r_locked_page = locked_page;
2305         err = ceph_mdsc_do_request(mdsc, NULL, req);
2306         if (locked_page && err == 0) {
2307                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2308                 if (inline_version == 0) {
2309                         /* the reply is supposed to contain inline data */
2310                         err = -EINVAL;
2311                 } else if (inline_version == CEPH_INLINE_NONE) {
2312                         err = -ENODATA;
2313                 } else {
2314                         err = req->r_reply_info.targeti.inline_len;
2315                 }
2316         }
2317         ceph_mdsc_put_request(req);
2318         dout("do_getattr result=%d\n", err);
2319         return err;
2320 }
2321
2322
2323 /*
2324  * Check inode permissions.  We verify we have a valid value for
2325  * the AUTH cap, then call the generic handler.
2326  */
2327 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode,
2328                     int mask)
2329 {
2330         int err;
2331
2332         if (mask & MAY_NOT_BLOCK)
2333                 return -ECHILD;
2334
2335         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2336
2337         if (!err)
2338                 err = generic_permission(&init_user_ns, inode, mask);
2339         return err;
2340 }
2341
2342 /* Craft a mask of needed caps given a set of requested statx attrs. */
2343 static int statx_to_caps(u32 want, umode_t mode)
2344 {
2345         int mask = 0;
2346
2347         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2348                 mask |= CEPH_CAP_AUTH_SHARED;
2349
2350         if (want & (STATX_NLINK|STATX_CTIME)) {
2351                 /*
2352                  * The link count for directories depends on inode->i_subdirs,
2353                  * and that is only updated when Fs caps are held.
2354                  */
2355                 if (S_ISDIR(mode))
2356                         mask |= CEPH_CAP_FILE_SHARED;
2357                 else
2358                         mask |= CEPH_CAP_LINK_SHARED;
2359         }
2360
2361         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2362                     STATX_BLOCKS))
2363                 mask |= CEPH_CAP_FILE_SHARED;
2364
2365         if (want & (STATX_CTIME))
2366                 mask |= CEPH_CAP_XATTR_SHARED;
2367
2368         return mask;
2369 }
2370
2371 /*
2372  * Get all the attributes. If we have sufficient caps for the requested attrs,
2373  * then we can avoid talking to the MDS at all.
2374  */
2375 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path,
2376                  struct kstat *stat, u32 request_mask, unsigned int flags)
2377 {
2378         struct inode *inode = d_inode(path->dentry);
2379         struct ceph_inode_info *ci = ceph_inode(inode);
2380         u32 valid_mask = STATX_BASIC_STATS;
2381         int err = 0;
2382
2383         /* Skip the getattr altogether if we're asked not to sync */
2384         if (!(flags & AT_STATX_DONT_SYNC)) {
2385                 err = ceph_do_getattr(inode,
2386                                 statx_to_caps(request_mask, inode->i_mode),
2387                                 flags & AT_STATX_FORCE_SYNC);
2388                 if (err)
2389                         return err;
2390         }
2391
2392         generic_fillattr(&init_user_ns, inode, stat);
2393         stat->ino = ceph_present_inode(inode);
2394
2395         /*
2396          * btime on newly-allocated inodes is 0, so if this is still set to
2397          * that, then assume that it's not valid.
2398          */
2399         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2400                 stat->btime = ci->i_btime;
2401                 valid_mask |= STATX_BTIME;
2402         }
2403
2404         if (ceph_snap(inode) == CEPH_NOSNAP)
2405                 stat->dev = inode->i_sb->s_dev;
2406         else
2407                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2408
2409         if (S_ISDIR(inode->i_mode)) {
2410                 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2411                                         RBYTES))
2412                         stat->size = ci->i_rbytes;
2413                 else
2414                         stat->size = ci->i_files + ci->i_subdirs;
2415                 stat->blocks = 0;
2416                 stat->blksize = 65536;
2417                 /*
2418                  * Some applications rely on the number of st_nlink
2419                  * value on directories to be either 0 (if unlinked)
2420                  * or 2 + number of subdirectories.
2421                  */
2422                 if (stat->nlink == 1)
2423                         /* '.' + '..' + subdirs */
2424                         stat->nlink = 1 + 1 + ci->i_subdirs;
2425         }
2426
2427         stat->result_mask = request_mask & valid_mask;
2428         return err;
2429 }