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