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