fs/ceph/inode.c

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