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