fs/ceph/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/fs.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/xattr.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/random.h>
  15 #include <linux/sort.h>
  16 #include <linux/iversion.h>
  17 #include <linux/fscrypt.h>
  18
  19 #include "super.h"
  20 #include "mds_client.h"
  21 #include "cache.h"
  22 #include "crypto.h"
  23 #include <linux/ceph/decode.h>
  24
  25 /*
  26  * Ceph inode operations
  27  *
  28  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  29  * setattr, etc.), xattr helpers, and helpers for assimilating
  30  * metadata returned by the MDS into our cache.
  31  *
  32  * Also define helpers for doing asynchronous writeback, invalidation,
  33  * and truncation for the benefit of those who can't afford to block
  34  * (typically because they are in the message handler path).
  35  */
  36
  37 static const struct inode_operations ceph_symlink_iops;
  38 static const struct inode_operations ceph_encrypted_symlink_iops;
  39
  40 static void ceph_inode_work(struct work_struct *work);
  41
  42 /*
  43  * find or create an inode, given the ceph ino number
  44  */
  45 static int ceph_set_ino_cb(struct inode *inode, void *data)
  46 {
  47         struct ceph_inode_info *ci = ceph_inode(inode);
  48         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
  49
  50         ci->i_vino = *(struct ceph_vino *)data;
  51         inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
  52         inode_set_iversion_raw(inode, 0);
  53         percpu_counter_inc(&mdsc->metric.total_inodes);
  54
  55         return 0;
  56 }
  57
  58 /**
  59  * ceph_new_inode - allocate a new inode in advance of an expected create
  60  * @dir: parent directory for new inode
  61  * @dentry: dentry that may eventually point to new inode
  62  * @mode: mode of new inode
  63  * @as_ctx: pointer to inherited security context
  64  *
  65  * Allocate a new inode in advance of an operation to create a new inode.
  66  * This allocates the inode and sets up the acl_sec_ctx with appropriate
  67  * info for the new inode.
  68  *
  69  * Returns a pointer to the new inode or an ERR_PTR.
  70  */
  71 struct inode *ceph_new_inode(struct inode *dir, struct dentry *dentry,
  72                              umode_t *mode, struct ceph_acl_sec_ctx *as_ctx)
  73 {
  74         int err;
  75         struct inode *inode;
  76
  77         inode = new_inode(dir->i_sb);
  78         if (!inode)
  79                 return ERR_PTR(-ENOMEM);
  80
  81         inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
  82
  83         if (!S_ISLNK(*mode)) {
  84                 err = ceph_pre_init_acls(dir, mode, as_ctx);
  85                 if (err < 0)
  86                         goto out_err;
  87         }
  88
  89         inode->i_state = 0;
  90         inode->i_mode = *mode;
  91
  92         err = ceph_security_init_secctx(dentry, *mode, as_ctx);
  93         if (err < 0)
  94                 goto out_err;
  95
  96         /*
  97          * We'll skip setting fscrypt context for snapshots, leaving that for
  98          * the handle_reply().
  99          */
 100         if (ceph_snap(dir) != CEPH_SNAPDIR) {
 101                 err = ceph_fscrypt_prepare_context(dir, inode, as_ctx);
 102                 if (err)
 103                         goto out_err;
 104         }
 105
 106         return inode;
 107 out_err:
 108         iput(inode);
 109         return ERR_PTR(err);
 110 }
 111
 112 void ceph_as_ctx_to_req(struct ceph_mds_request *req,
 113                         struct ceph_acl_sec_ctx *as_ctx)
 114 {
 115         if (as_ctx->pagelist) {
 116                 req->r_pagelist = as_ctx->pagelist;
 117                 as_ctx->pagelist = NULL;
 118         }
 119         ceph_fscrypt_as_ctx_to_req(req, as_ctx);
 120 }
 121
 122 /**
 123  * ceph_get_inode - find or create/hash a new inode
 124  * @sb: superblock to search and allocate in
 125  * @vino: vino to search for
 126  * @newino: optional new inode to insert if one isn't found (may be NULL)
 127  *
 128  * Search for or insert a new inode into the hash for the given vino, and
 129  * return a reference to it. If new is non-NULL, its reference is consumed.
 130  */
 131 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino,
 132                              struct inode *newino)
 133 {
 134         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(sb);
 135         struct ceph_client *cl = mdsc->fsc->client;
 136         struct inode *inode;
 137
 138         if (ceph_vino_is_reserved(vino))
 139                 return ERR_PTR(-EREMOTEIO);
 140
 141         if (newino) {
 142                 inode = inode_insert5(newino, (unsigned long)vino.ino,
 143                                       ceph_ino_compare, ceph_set_ino_cb, &vino);
 144                 if (inode != newino)
 145                         iput(newino);
 146         } else {
 147                 inode = iget5_locked(sb, (unsigned long)vino.ino,
 148                                      ceph_ino_compare, ceph_set_ino_cb, &vino);
 149         }
 150
 151         if (!inode) {
 152                 doutc(cl, "no inode found for %llx.%llx\n", vino.ino, vino.snap);
 153                 return ERR_PTR(-ENOMEM);
 154         }
 155
 156         doutc(cl, "on %llx=%llx.%llx got %p new %d\n",
 157               ceph_present_inode(inode), ceph_vinop(inode), inode,
 158               !!(inode->i_state & I_NEW));
 159         return inode;
 160 }
 161
 162 /*
 163  * get/constuct snapdir inode for a given directory
 164  */
 165 struct inode *ceph_get_snapdir(struct inode *parent)
 166 {
 167         struct ceph_client *cl = ceph_inode_to_client(parent);
 168         struct ceph_vino vino = {
 169                 .ino = ceph_ino(parent),
 170                 .snap = CEPH_SNAPDIR,
 171         };
 172         struct inode *inode = ceph_get_inode(parent->i_sb, vino, NULL);
 173         struct ceph_inode_info *ci = ceph_inode(inode);
 174         int ret = -ENOTDIR;
 175
 176         if (IS_ERR(inode))
 177                 return inode;
 178
 179         if (!S_ISDIR(parent->i_mode)) {
 180                 pr_warn_once_client(cl, "bad snapdir parent type (mode=0%o)\n",
 181                                     parent->i_mode);
 182                 goto err;
 183         }
 184
 185         if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
 186                 pr_warn_once_client(cl, "bad snapdir inode type (mode=0%o)\n",
 187                                     inode->i_mode);
 188                 goto err;
 189         }
 190
 191         inode->i_mode = parent->i_mode;
 192         inode->i_uid = parent->i_uid;
 193         inode->i_gid = parent->i_gid;
 194         inode_set_mtime_to_ts(inode, inode_get_mtime(parent));
 195         inode_set_ctime_to_ts(inode, inode_get_ctime(parent));
 196         inode_set_atime_to_ts(inode, inode_get_atime(parent));
 197         ci->i_rbytes = 0;
 198         ci->i_btime = ceph_inode(parent)->i_btime;
 199
 200 #ifdef CONFIG_FS_ENCRYPTION
 201         /* if encrypted, just borrow fscrypt_auth from parent */
 202         if (IS_ENCRYPTED(parent)) {
 203                 struct ceph_inode_info *pci = ceph_inode(parent);
 204
 205                 ci->fscrypt_auth = kmemdup(pci->fscrypt_auth,
 206                                            pci->fscrypt_auth_len,
 207                                            GFP_KERNEL);
 208                 if (ci->fscrypt_auth) {
 209                         inode->i_flags |= S_ENCRYPTED;
 210                         ci->fscrypt_auth_len = pci->fscrypt_auth_len;
 211                 } else {
 212                         doutc(cl, "Failed to alloc snapdir fscrypt_auth\n");
 213                         ret = -ENOMEM;
 214                         goto err;
 215                 }
 216         }
 217 #endif
 218         if (inode->i_state & I_NEW) {
 219                 inode->i_op = &ceph_snapdir_iops;
 220                 inode->i_fop = &ceph_snapdir_fops;
 221                 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
 222                 unlock_new_inode(inode);
 223         }
 224
 225         return inode;
 226 err:
 227         if ((inode->i_state & I_NEW))
 228                 discard_new_inode(inode);
 229         else
 230                 iput(inode);
 231         return ERR_PTR(ret);
 232 }
 233
 234 const struct inode_operations ceph_file_iops = {
 235         .permission = ceph_permission,
 236         .setattr = ceph_setattr,
 237         .getattr = ceph_getattr,
 238         .listxattr = ceph_listxattr,
 239         .get_inode_acl = ceph_get_acl,
 240         .set_acl = ceph_set_acl,
 241 };
 242
 243
 244 /*
 245  * We use a 'frag tree' to keep track of the MDS's directory fragments
 246  * for a given inode (usually there is just a single fragment).  We
 247  * need to know when a child frag is delegated to a new MDS, or when
 248  * it is flagged as replicated, so we can direct our requests
 249  * accordingly.
 250  */
 251
 252 /*
 253  * find/create a frag in the tree
 254  */
 255 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 256                                                     u32 f)
 257 {
 258         struct inode *inode = &ci->netfs.inode;
 259         struct ceph_client *cl = ceph_inode_to_client(inode);
 260         struct rb_node **p;
 261         struct rb_node *parent = NULL;
 262         struct ceph_inode_frag *frag;
 263         int c;
 264
 265         p = &ci->i_fragtree.rb_node;
 266         while (*p) {
 267                 parent = *p;
 268                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 269                 c = ceph_frag_compare(f, frag->frag);
 270                 if (c < 0)
 271                         p = &(*p)->rb_left;
 272                 else if (c > 0)
 273                         p = &(*p)->rb_right;
 274                 else
 275                         return frag;
 276         }
 277
 278         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 279         if (!frag)
 280                 return ERR_PTR(-ENOMEM);
 281
 282         frag->frag = f;
 283         frag->split_by = 0;
 284         frag->mds = -1;
 285         frag->ndist = 0;
 286
 287         rb_link_node(&frag->node, parent, p);
 288         rb_insert_color(&frag->node, &ci->i_fragtree);
 289
 290         doutc(cl, "added %p %llx.%llx frag %x\n", inode, ceph_vinop(inode), f);
 291         return frag;
 292 }
 293
 294 /*
 295  * find a specific frag @f
 296  */
 297 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 298 {
 299         struct rb_node *n = ci->i_fragtree.rb_node;
 300
 301         while (n) {
 302                 struct ceph_inode_frag *frag =
 303                         rb_entry(n, struct ceph_inode_frag, node);
 304                 int c = ceph_frag_compare(f, frag->frag);
 305                 if (c < 0)
 306                         n = n->rb_left;
 307                 else if (c > 0)
 308                         n = n->rb_right;
 309                 else
 310                         return frag;
 311         }
 312         return NULL;
 313 }
 314
 315 /*
 316  * Choose frag containing the given value @v.  If @pfrag is
 317  * specified, copy the frag delegation info to the caller if
 318  * it is present.
 319  */
 320 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 321                               struct ceph_inode_frag *pfrag, int *found)
 322 {
 323         struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
 324         u32 t = ceph_frag_make(0, 0);
 325         struct ceph_inode_frag *frag;
 326         unsigned nway, i;
 327         u32 n;
 328
 329         if (found)
 330                 *found = 0;
 331
 332         while (1) {
 333                 WARN_ON(!ceph_frag_contains_value(t, v));
 334                 frag = __ceph_find_frag(ci, t);
 335                 if (!frag)
 336                         break; /* t is a leaf */
 337                 if (frag->split_by == 0) {
 338                         if (pfrag)
 339                                 memcpy(pfrag, frag, sizeof(*pfrag));
 340                         if (found)
 341                                 *found = 1;
 342                         break;
 343                 }
 344
 345                 /* choose child */
 346                 nway = 1 << frag->split_by;
 347                 doutc(cl, "frag(%x) %x splits by %d (%d ways)\n", v, t,
 348                       frag->split_by, nway);
 349                 for (i = 0; i < nway; i++) {
 350                         n = ceph_frag_make_child(t, frag->split_by, i);
 351                         if (ceph_frag_contains_value(n, v)) {
 352                                 t = n;
 353                                 break;
 354                         }
 355                 }
 356                 BUG_ON(i == nway);
 357         }
 358         doutc(cl, "frag(%x) = %x\n", v, t);
 359
 360         return t;
 361 }
 362
 363 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 364                      struct ceph_inode_frag *pfrag, int *found)
 365 {
 366         u32 ret;
 367         mutex_lock(&ci->i_fragtree_mutex);
 368         ret = __ceph_choose_frag(ci, v, pfrag, found);
 369         mutex_unlock(&ci->i_fragtree_mutex);
 370         return ret;
 371 }
 372
 373 /*
 374  * Process dirfrag (delegation) info from the mds.  Include leaf
 375  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 376  * branches/splits are included in i_fragtree)
 377  */
 378 static int ceph_fill_dirfrag(struct inode *inode,
 379                              struct ceph_mds_reply_dirfrag *dirinfo)
 380 {
 381         struct ceph_inode_info *ci = ceph_inode(inode);
 382         struct ceph_client *cl = ceph_inode_to_client(inode);
 383         struct ceph_inode_frag *frag;
 384         u32 id = le32_to_cpu(dirinfo->frag);
 385         int mds = le32_to_cpu(dirinfo->auth);
 386         int ndist = le32_to_cpu(dirinfo->ndist);
 387         int diri_auth = -1;
 388         int i;
 389         int err = 0;
 390
 391         spin_lock(&ci->i_ceph_lock);
 392         if (ci->i_auth_cap)
 393                 diri_auth = ci->i_auth_cap->mds;
 394         spin_unlock(&ci->i_ceph_lock);
 395
 396         if (mds == -1) /* CDIR_AUTH_PARENT */
 397                 mds = diri_auth;
 398
 399         mutex_lock(&ci->i_fragtree_mutex);
 400         if (ndist == 0 && mds == diri_auth) {
 401                 /* no delegation info needed. */
 402                 frag = __ceph_find_frag(ci, id);
 403                 if (!frag)
 404                         goto out;
 405                 if (frag->split_by == 0) {
 406                         /* tree leaf, remove */
 407                         doutc(cl, "removed %p %llx.%llx frag %x (no ref)\n",
 408                               inode, ceph_vinop(inode), id);
 409                         rb_erase(&frag->node, &ci->i_fragtree);
 410                         kfree(frag);
 411                 } else {
 412                         /* tree branch, keep and clear */
 413                         doutc(cl, "cleared %p %llx.%llx frag %x referral\n",
 414                               inode, ceph_vinop(inode), id);
 415                         frag->mds = -1;
 416                         frag->ndist = 0;
 417                 }
 418                 goto out;
 419         }
 420
 421
 422         /* find/add this frag to store mds delegation info */
 423         frag = __get_or_create_frag(ci, id);
 424         if (IS_ERR(frag)) {
 425                 /* this is not the end of the world; we can continue
 426                    with bad/inaccurate delegation info */
 427                 pr_err_client(cl, "ENOMEM on mds ref %p %llx.%llx fg %x\n",
 428                               inode, ceph_vinop(inode),
 429                               le32_to_cpu(dirinfo->frag));
 430                 err = -ENOMEM;
 431                 goto out;
 432         }
 433
 434         frag->mds = mds;
 435         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 436         for (i = 0; i < frag->ndist; i++)
 437                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 438         doutc(cl, "%p %llx.%llx frag %x ndist=%d\n", inode,
 439               ceph_vinop(inode), frag->frag, frag->ndist);
 440
 441 out:
 442         mutex_unlock(&ci->i_fragtree_mutex);
 443         return err;
 444 }
 445
 446 static int frag_tree_split_cmp(const void *l, const void *r)
 447 {
 448         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 449         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 450         return ceph_frag_compare(le32_to_cpu(ls->frag),
 451                                  le32_to_cpu(rs->frag));
 452 }
 453
 454 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 455 {
 456         if (!frag)
 457                 return f == ceph_frag_make(0, 0);
 458         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 459                 return false;
 460         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 461 }
 462
 463 static int ceph_fill_fragtree(struct inode *inode,
 464                               struct ceph_frag_tree_head *fragtree,
 465                               struct ceph_mds_reply_dirfrag *dirinfo)
 466 {
 467         struct ceph_client *cl = ceph_inode_to_client(inode);
 468         struct ceph_inode_info *ci = ceph_inode(inode);
 469         struct ceph_inode_frag *frag, *prev_frag = NULL;
 470         struct rb_node *rb_node;
 471         unsigned i, split_by, nsplits;
 472         u32 id;
 473         bool update = false;
 474
 475         mutex_lock(&ci->i_fragtree_mutex);
 476         nsplits = le32_to_cpu(fragtree->nsplits);
 477         if (nsplits != ci->i_fragtree_nsplits) {
 478                 update = true;
 479         } else if (nsplits) {
 480                 i = get_random_u32_below(nsplits);
 481                 id = le32_to_cpu(fragtree->splits[i].frag);
 482                 if (!__ceph_find_frag(ci, id))
 483                         update = true;
 484         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 485                 rb_node = rb_first(&ci->i_fragtree);
 486                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 487                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 488                         update = true;
 489         }
 490         if (!update && dirinfo) {
 491                 id = le32_to_cpu(dirinfo->frag);
 492                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 493                         update = true;
 494         }
 495         if (!update)
 496                 goto out_unlock;
 497
 498         if (nsplits > 1) {
 499                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 500                      frag_tree_split_cmp, NULL);
 501         }
 502
 503         doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
 504         rb_node = rb_first(&ci->i_fragtree);
 505         for (i = 0; i < nsplits; i++) {
 506                 id = le32_to_cpu(fragtree->splits[i].frag);
 507                 split_by = le32_to_cpu(fragtree->splits[i].by);
 508                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 509                         pr_err_client(cl, "%p %llx.%llx invalid split %d/%u, "
 510                                "frag %x split by %d\n", inode,
 511                                ceph_vinop(inode), i, nsplits, id, split_by);
 512                         continue;
 513                 }
 514                 frag = NULL;
 515                 while (rb_node) {
 516                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 517                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 518                                 if (frag->frag != id)
 519                                         frag = NULL;
 520                                 else
 521                                         rb_node = rb_next(rb_node);
 522                                 break;
 523                         }
 524                         rb_node = rb_next(rb_node);
 525                         /* delete stale split/leaf node */
 526                         if (frag->split_by > 0 ||
 527                             !is_frag_child(frag->frag, prev_frag)) {
 528                                 rb_erase(&frag->node, &ci->i_fragtree);
 529                                 if (frag->split_by > 0)
 530                                         ci->i_fragtree_nsplits--;
 531                                 kfree(frag);
 532                         }
 533                         frag = NULL;
 534                 }
 535                 if (!frag) {
 536                         frag = __get_or_create_frag(ci, id);
 537                         if (IS_ERR(frag))
 538                                 continue;
 539                 }
 540                 if (frag->split_by == 0)
 541                         ci->i_fragtree_nsplits++;
 542                 frag->split_by = split_by;
 543                 doutc(cl, " frag %x split by %d\n", frag->frag, frag->split_by);
 544                 prev_frag = frag;
 545         }
 546         while (rb_node) {
 547                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 548                 rb_node = rb_next(rb_node);
 549                 /* delete stale split/leaf node */
 550                 if (frag->split_by > 0 ||
 551                     !is_frag_child(frag->frag, prev_frag)) {
 552                         rb_erase(&frag->node, &ci->i_fragtree);
 553                         if (frag->split_by > 0)
 554                                 ci->i_fragtree_nsplits--;
 555                         kfree(frag);
 556                 }
 557         }
 558 out_unlock:
 559         mutex_unlock(&ci->i_fragtree_mutex);
 560         return 0;
 561 }
 562
 563 /*
 564  * initialize a newly allocated inode.
 565  */
 566 struct inode *ceph_alloc_inode(struct super_block *sb)
 567 {
 568         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
 569         struct ceph_inode_info *ci;
 570         int i;
 571
 572         ci = alloc_inode_sb(sb, ceph_inode_cachep, GFP_NOFS);
 573         if (!ci)
 574                 return NULL;
 575
 576         doutc(fsc->client, "%p\n", &ci->netfs.inode);
 577
 578         /* Set parameters for the netfs library */
 579         netfs_inode_init(&ci->netfs, &ceph_netfs_ops, false);
 580         /* [DEPRECATED] Use PG_private_2 to mark folio being written to the cache. */
 581         __set_bit(NETFS_ICTX_USE_PGPRIV2, &ci->netfs.flags);
 582
 583         spin_lock_init(&ci->i_ceph_lock);
 584
 585         ci->i_version = 0;
 586         ci->i_inline_version = 0;
 587         ci->i_time_warp_seq = 0;
 588         ci->i_ceph_flags = 0;
 589         atomic64_set(&ci->i_ordered_count, 1);
 590         atomic64_set(&ci->i_release_count, 1);
 591         atomic64_set(&ci->i_complete_seq[0], 0);
 592         atomic64_set(&ci->i_complete_seq[1], 0);
 593         ci->i_symlink = NULL;
 594
 595         ci->i_max_bytes = 0;
 596         ci->i_max_files = 0;
 597
 598         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 599         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
 600         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 601
 602         ci->i_fragtree = RB_ROOT;
 603         mutex_init(&ci->i_fragtree_mutex);
 604
 605         ci->i_xattrs.blob = NULL;
 606         ci->i_xattrs.prealloc_blob = NULL;
 607         ci->i_xattrs.dirty = false;
 608         ci->i_xattrs.index = RB_ROOT;
 609         ci->i_xattrs.count = 0;
 610         ci->i_xattrs.names_size = 0;
 611         ci->i_xattrs.vals_size = 0;
 612         ci->i_xattrs.version = 0;
 613         ci->i_xattrs.index_version = 0;
 614
 615         ci->i_caps = RB_ROOT;
 616         ci->i_auth_cap = NULL;
 617         ci->i_dirty_caps = 0;
 618         ci->i_flushing_caps = 0;
 619         INIT_LIST_HEAD(&ci->i_dirty_item);
 620         INIT_LIST_HEAD(&ci->i_flushing_item);
 621         ci->i_prealloc_cap_flush = NULL;
 622         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 623         init_waitqueue_head(&ci->i_cap_wq);
 624         ci->i_hold_caps_max = 0;
 625         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 626         INIT_LIST_HEAD(&ci->i_cap_snaps);
 627         ci->i_head_snapc = NULL;
 628         ci->i_snap_caps = 0;
 629
 630         ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
 631         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 632                 ci->i_nr_by_mode[i] = 0;
 633
 634         mutex_init(&ci->i_truncate_mutex);
 635         ci->i_truncate_seq = 0;
 636         ci->i_truncate_size = 0;
 637         ci->i_truncate_pending = 0;
 638         ci->i_truncate_pagecache_size = 0;
 639
 640         ci->i_max_size = 0;
 641         ci->i_reported_size = 0;
 642         ci->i_wanted_max_size = 0;
 643         ci->i_requested_max_size = 0;
 644
 645         ci->i_pin_ref = 0;
 646         ci->i_rd_ref = 0;
 647         ci->i_rdcache_ref = 0;
 648         ci->i_wr_ref = 0;
 649         ci->i_wb_ref = 0;
 650         ci->i_fx_ref = 0;
 651         ci->i_wrbuffer_ref = 0;
 652         ci->i_wrbuffer_ref_head = 0;
 653         atomic_set(&ci->i_filelock_ref, 0);
 654         atomic_set(&ci->i_shared_gen, 1);
 655         ci->i_rdcache_gen = 0;
 656         ci->i_rdcache_revoking = 0;
 657
 658         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 659         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 660         spin_lock_init(&ci->i_unsafe_lock);
 661
 662         ci->i_snap_realm = NULL;
 663         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 664         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 665
 666         INIT_WORK(&ci->i_work, ceph_inode_work);
 667         ci->i_work_mask = 0;
 668         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 669 #ifdef CONFIG_FS_ENCRYPTION
 670         ci->fscrypt_auth = NULL;
 671         ci->fscrypt_auth_len = 0;
 672 #endif
 673         return &ci->netfs.inode;
 674 }
 675
 676 void ceph_free_inode(struct inode *inode)
 677 {
 678         struct ceph_inode_info *ci = ceph_inode(inode);
 679
 680         kfree(ci->i_symlink);
 681 #ifdef CONFIG_FS_ENCRYPTION
 682         kfree(ci->fscrypt_auth);
 683 #endif
 684         fscrypt_free_inode(inode);
 685         kmem_cache_free(ceph_inode_cachep, ci);
 686 }
 687
 688 void ceph_evict_inode(struct inode *inode)
 689 {
 690         struct ceph_inode_info *ci = ceph_inode(inode);
 691         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 692         struct ceph_client *cl = ceph_inode_to_client(inode);
 693         struct ceph_inode_frag *frag;
 694         struct rb_node *n;
 695
 696         doutc(cl, "%p ino %llx.%llx\n", inode, ceph_vinop(inode));
 697
 698         percpu_counter_dec(&mdsc->metric.total_inodes);
 699
 700         truncate_inode_pages_final(&inode->i_data);
 701         if (inode->i_state & I_PINNING_NETFS_WB)
 702                 ceph_fscache_unuse_cookie(inode, true);
 703         clear_inode(inode);
 704
 705         ceph_fscache_unregister_inode_cookie(ci);
 706         fscrypt_put_encryption_info(inode);
 707
 708         __ceph_remove_caps(ci);
 709
 710         if (__ceph_has_quota(ci, QUOTA_GET_ANY))
 711                 ceph_adjust_quota_realms_count(inode, false);
 712
 713         /*
 714          * we may still have a snap_realm reference if there are stray
 715          * caps in i_snap_caps.
 716          */
 717         if (ci->i_snap_realm) {
 718                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 719                         doutc(cl, " dropping residual ref to snap realm %p\n",
 720                               ci->i_snap_realm);
 721                         ceph_change_snap_realm(inode, NULL);
 722                 } else {
 723                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 724                         ci->i_snap_realm = NULL;
 725                 }
 726         }
 727
 728         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 729                 frag = rb_entry(n, struct ceph_inode_frag, node);
 730                 rb_erase(n, &ci->i_fragtree);
 731                 kfree(frag);
 732         }
 733         ci->i_fragtree_nsplits = 0;
 734
 735         __ceph_destroy_xattrs(ci);
 736         if (ci->i_xattrs.blob)
 737                 ceph_buffer_put(ci->i_xattrs.blob);
 738         if (ci->i_xattrs.prealloc_blob)
 739                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 740
 741         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 742         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
 743 }
 744
 745 static inline blkcnt_t calc_inode_blocks(u64 size)
 746 {
 747         return (size + (1<<9) - 1) >> 9;
 748 }
 749
 750 /*
 751  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 752  * careful because either the client or MDS may have more up to date
 753  * info, depending on which capabilities are held, and whether
 754  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 755  * and size are monotonically increasing, except when utimes() or
 756  * truncate() increments the corresponding _seq values.)
 757  */
 758 int ceph_fill_file_size(struct inode *inode, int issued,
 759                         u32 truncate_seq, u64 truncate_size, u64 size)
 760 {
 761         struct ceph_client *cl = ceph_inode_to_client(inode);
 762         struct ceph_inode_info *ci = ceph_inode(inode);
 763         int queue_trunc = 0;
 764         loff_t isize = i_size_read(inode);
 765
 766         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 767             (truncate_seq == ci->i_truncate_seq && size > isize)) {
 768                 doutc(cl, "size %lld -> %llu\n", isize, size);
 769                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 770                         pr_err_client(cl, "non-zero size for directory\n");
 771                         size = 0;
 772                 }
 773                 i_size_write(inode, size);
 774                 inode->i_blocks = calc_inode_blocks(size);
 775                 /*
 776                  * If we're expanding, then we should be able to just update
 777                  * the existing cookie.
 778                  */
 779                 if (size > isize)
 780                         ceph_fscache_update(inode);
 781                 ci->i_reported_size = size;
 782                 if (truncate_seq != ci->i_truncate_seq) {
 783                         doutc(cl, "truncate_seq %u -> %u\n",
 784                               ci->i_truncate_seq, truncate_seq);
 785                         ci->i_truncate_seq = truncate_seq;
 786
 787                         /* the MDS should have revoked these caps */
 788                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_RD |
 789                                                CEPH_CAP_FILE_LAZYIO));
 790                         /*
 791                          * If we hold relevant caps, or in the case where we're
 792                          * not the only client referencing this file and we
 793                          * don't hold those caps, then we need to check whether
 794                          * the file is either opened or mmaped
 795                          */
 796                         if ((issued & (CEPH_CAP_FILE_CACHE|
 797                                        CEPH_CAP_FILE_BUFFER)) ||
 798                             mapping_mapped(inode->i_mapping) ||
 799                             __ceph_is_file_opened(ci)) {
 800                                 ci->i_truncate_pending++;
 801                                 queue_trunc = 1;
 802                         }
 803                 }
 804         }
 805
 806         /*
 807          * It's possible that the new sizes of the two consecutive
 808          * size truncations will be in the same fscrypt last block,
 809          * and we need to truncate the corresponding page caches
 810          * anyway.
 811          */
 812         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0) {
 813                 doutc(cl, "truncate_size %lld -> %llu, encrypted %d\n",
 814                       ci->i_truncate_size, truncate_size,
 815                       !!IS_ENCRYPTED(inode));
 816
 817                 ci->i_truncate_size = truncate_size;
 818
 819                 if (IS_ENCRYPTED(inode)) {
 820                         doutc(cl, "truncate_pagecache_size %lld -> %llu\n",
 821                               ci->i_truncate_pagecache_size, size);
 822                         ci->i_truncate_pagecache_size = size;
 823                 } else {
 824                         ci->i_truncate_pagecache_size = truncate_size;
 825                 }
 826         }
 827         return queue_trunc;
 828 }
 829
 830 void ceph_fill_file_time(struct inode *inode, int issued,
 831                          u64 time_warp_seq, struct timespec64 *ctime,
 832                          struct timespec64 *mtime, struct timespec64 *atime)
 833 {
 834         struct ceph_client *cl = ceph_inode_to_client(inode);
 835         struct ceph_inode_info *ci = ceph_inode(inode);
 836         struct timespec64 ictime = inode_get_ctime(inode);
 837         int warn = 0;
 838
 839         if (issued & (CEPH_CAP_FILE_EXCL|
 840                       CEPH_CAP_FILE_WR|
 841                       CEPH_CAP_FILE_BUFFER|
 842                       CEPH_CAP_AUTH_EXCL|
 843                       CEPH_CAP_XATTR_EXCL)) {
 844                 if (ci->i_version == 0 ||
 845                     timespec64_compare(ctime, &ictime) > 0) {
 846                         doutc(cl, "ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 847                              ictime.tv_sec, ictime.tv_nsec,
 848                              ctime->tv_sec, ctime->tv_nsec);
 849                         inode_set_ctime_to_ts(inode, *ctime);
 850                 }
 851                 if (ci->i_version == 0 ||
 852                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 853                         /* the MDS did a utimes() */
 854                         doutc(cl, "mtime %lld.%09ld -> %lld.%09ld tw %d -> %d\n",
 855                              inode_get_mtime_sec(inode),
 856                              inode_get_mtime_nsec(inode),
 857                              mtime->tv_sec, mtime->tv_nsec,
 858                              ci->i_time_warp_seq, (int)time_warp_seq);
 859
 860                         inode_set_mtime_to_ts(inode, *mtime);
 861                         inode_set_atime_to_ts(inode, *atime);
 862                         ci->i_time_warp_seq = time_warp_seq;
 863                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 864                         struct timespec64       ts;
 865
 866                         /* nobody did utimes(); take the max */
 867                         ts = inode_get_mtime(inode);
 868                         if (timespec64_compare(mtime, &ts) > 0) {
 869                                 doutc(cl, "mtime %lld.%09ld -> %lld.%09ld inc\n",
 870                                      ts.tv_sec, ts.tv_nsec,
 871                                      mtime->tv_sec, mtime->tv_nsec);
 872                                 inode_set_mtime_to_ts(inode, *mtime);
 873                         }
 874                         ts = inode_get_atime(inode);
 875                         if (timespec64_compare(atime, &ts) > 0) {
 876                                 doutc(cl, "atime %lld.%09ld -> %lld.%09ld inc\n",
 877                                      ts.tv_sec, ts.tv_nsec,
 878                                      atime->tv_sec, atime->tv_nsec);
 879                                 inode_set_atime_to_ts(inode, *atime);
 880                         }
 881                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 882                         /* we did a utimes(); ignore mds values */
 883                 } else {
 884                         warn = 1;
 885                 }
 886         } else {
 887                 /* we have no write|excl caps; whatever the MDS says is true */
 888                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 889                         inode_set_ctime_to_ts(inode, *ctime);
 890                         inode_set_mtime_to_ts(inode, *mtime);
 891                         inode_set_atime_to_ts(inode, *atime);
 892                         ci->i_time_warp_seq = time_warp_seq;
 893                 } else {
 894                         warn = 1;
 895                 }
 896         }
 897         if (warn) /* time_warp_seq shouldn't go backwards */
 898                 doutc(cl, "%p mds time_warp_seq %llu < %u\n", inode,
 899                       time_warp_seq, ci->i_time_warp_seq);
 900 }
 901
 902 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
 903 static int decode_encrypted_symlink(struct ceph_mds_client *mdsc,
 904                                     const char *encsym,
 905                                     int enclen, u8 **decsym)
 906 {
 907         struct ceph_client *cl = mdsc->fsc->client;
 908         int declen;
 909         u8 *sym;
 910
 911         sym = kmalloc(enclen + 1, GFP_NOFS);
 912         if (!sym)
 913                 return -ENOMEM;
 914
 915         declen = ceph_base64_decode(encsym, enclen, sym);
 916         if (declen < 0) {
 917                 pr_err_client(cl,
 918                         "can't decode symlink (%d). Content: %.*s\n",
 919                         declen, enclen, encsym);
 920                 kfree(sym);
 921                 return -EIO;
 922         }
 923         sym[declen + 1] = '\0';
 924         *decsym = sym;
 925         return declen;
 926 }
 927 #else
 928 static int decode_encrypted_symlink(struct ceph_mds_client *mdsc,
 929                                     const char *encsym,
 930                                     int symlen, u8 **decsym)
 931 {
 932         return -EOPNOTSUPP;
 933 }
 934 #endif
 935
 936 /*
 937  * Populate an inode based on info from mds.  May be called on new or
 938  * existing inodes.
 939  */
 940 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 941                     struct ceph_mds_reply_info_in *iinfo,
 942                     struct ceph_mds_reply_dirfrag *dirinfo,
 943                     struct ceph_mds_session *session, int cap_fmode,
 944                     struct ceph_cap_reservation *caps_reservation)
 945 {
 946         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 947         struct ceph_client *cl = mdsc->fsc->client;
 948         struct ceph_mds_reply_inode *info = iinfo->in;
 949         struct ceph_inode_info *ci = ceph_inode(inode);
 950         int issued, new_issued, info_caps;
 951         struct timespec64 mtime, atime, ctime;
 952         struct ceph_buffer *xattr_blob = NULL;
 953         struct ceph_buffer *old_blob = NULL;
 954         struct ceph_string *pool_ns = NULL;
 955         struct ceph_cap *new_cap = NULL;
 956         int err = 0;
 957         bool wake = false;
 958         bool queue_trunc = false;
 959         bool new_version = false;
 960         bool fill_inline = false;
 961         umode_t mode = le32_to_cpu(info->mode);
 962         dev_t rdev = le32_to_cpu(info->rdev);
 963
 964         lockdep_assert_held(&mdsc->snap_rwsem);
 965
 966         doutc(cl, "%p ino %llx.%llx v %llu had %llu\n", inode, ceph_vinop(inode),
 967               le64_to_cpu(info->version), ci->i_version);
 968
 969         /* Once I_NEW is cleared, we can't change type or dev numbers */
 970         if (inode->i_state & I_NEW) {
 971                 inode->i_mode = mode;
 972         } else {
 973                 if (inode_wrong_type(inode, mode)) {
 974                         pr_warn_once_client(cl,
 975                                 "inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
 976                                 ceph_vinop(inode), inode->i_mode, mode);
 977                         return -ESTALE;
 978                 }
 979
 980                 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
 981                         pr_warn_once_client(cl,
 982                                 "dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
 983                                 ceph_vinop(inode), MAJOR(inode->i_rdev),
 984                                 MINOR(inode->i_rdev), MAJOR(rdev),
 985                                 MINOR(rdev));
 986                         return -ESTALE;
 987                 }
 988         }
 989
 990         info_caps = le32_to_cpu(info->cap.caps);
 991
 992         /* prealloc new cap struct */
 993         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 994                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 995                 if (!new_cap)
 996                         return -ENOMEM;
 997         }
 998
 999         /*
1000          * prealloc xattr data, if it looks like we'll need it.  only
1001          * if len > 4 (meaning there are actually xattrs; the first 4
1002          * bytes are the xattr count).
1003          */
1004         if (iinfo->xattr_len > 4) {
1005                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
1006                 if (!xattr_blob)
1007                         pr_err_client(cl, "ENOMEM xattr blob %d bytes\n",
1008                                       iinfo->xattr_len);
1009         }
1010
1011         if (iinfo->pool_ns_len > 0)
1012                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
1013                                                      iinfo->pool_ns_len);
1014
1015         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
1016                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
1017
1018         spin_lock(&ci->i_ceph_lock);
1019
1020         /*
1021          * provided version will be odd if inode value is projected,
1022          * even if stable.  skip the update if we have newer stable
1023          * info (ours>=theirs, e.g. due to racing mds replies), unless
1024          * we are getting projected (unstable) info (in which case the
1025          * version is odd, and we want ours>theirs).
1026          *   us   them
1027          *   2    2     skip
1028          *   3    2     skip
1029          *   3    3     update
1030          */
1031         if (ci->i_version == 0 ||
1032             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1033              le64_to_cpu(info->version) > (ci->i_version & ~1)))
1034                 new_version = true;
1035
1036         /* Update change_attribute */
1037         inode_set_max_iversion_raw(inode, iinfo->change_attr);
1038
1039         __ceph_caps_issued(ci, &issued);
1040         issued |= __ceph_caps_dirty(ci);
1041         new_issued = ~issued & info_caps;
1042
1043         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
1044
1045 #ifdef CONFIG_FS_ENCRYPTION
1046         if (iinfo->fscrypt_auth_len &&
1047             ((inode->i_state & I_NEW) || (ci->fscrypt_auth_len == 0))) {
1048                 kfree(ci->fscrypt_auth);
1049                 ci->fscrypt_auth_len = iinfo->fscrypt_auth_len;
1050                 ci->fscrypt_auth = iinfo->fscrypt_auth;
1051                 iinfo->fscrypt_auth = NULL;
1052                 iinfo->fscrypt_auth_len = 0;
1053                 inode_set_flags(inode, S_ENCRYPTED, S_ENCRYPTED);
1054         }
1055 #endif
1056
1057         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
1058             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
1059                 inode->i_mode = mode;
1060                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
1061                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
1062                 doutc(cl, "%p %llx.%llx mode 0%o uid.gid %d.%d\n", inode,
1063                       ceph_vinop(inode), inode->i_mode,
1064                       from_kuid(&init_user_ns, inode->i_uid),
1065                       from_kgid(&init_user_ns, inode->i_gid));
1066                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
1067                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
1068         }
1069
1070         /* directories have fl_stripe_unit set to zero */
1071         if (IS_ENCRYPTED(inode))
1072                 inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
1073         else if (le32_to_cpu(info->layout.fl_stripe_unit))
1074                 inode->i_blkbits =
1075                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
1076         else
1077                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
1078
1079         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
1080             (issued & CEPH_CAP_LINK_EXCL) == 0)
1081                 set_nlink(inode, le32_to_cpu(info->nlink));
1082
1083         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
1084                 /* be careful with mtime, atime, size */
1085                 ceph_decode_timespec64(&atime, &info->atime);
1086                 ceph_decode_timespec64(&mtime, &info->mtime);
1087                 ceph_decode_timespec64(&ctime, &info->ctime);
1088                 ceph_fill_file_time(inode, issued,
1089                                 le32_to_cpu(info->time_warp_seq),
1090                                 &ctime, &mtime, &atime);
1091         }
1092
1093         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
1094                 ci->i_files = le64_to_cpu(info->files);
1095                 ci->i_subdirs = le64_to_cpu(info->subdirs);
1096         }
1097
1098         if (new_version ||
1099             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
1100                 u64 size = le64_to_cpu(info->size);
1101                 s64 old_pool = ci->i_layout.pool_id;
1102                 struct ceph_string *old_ns;
1103
1104                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
1105                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
1106                                         lockdep_is_held(&ci->i_ceph_lock));
1107                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
1108
1109                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
1110                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
1111
1112                 pool_ns = old_ns;
1113
1114                 if (IS_ENCRYPTED(inode) && size &&
1115                     iinfo->fscrypt_file_len == sizeof(__le64)) {
1116                         u64 fsize = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
1117
1118                         if (size == round_up(fsize, CEPH_FSCRYPT_BLOCK_SIZE)) {
1119                                 size = fsize;
1120                         } else {
1121                                 pr_warn_client(cl,
1122                                         "fscrypt size mismatch: size=%llu fscrypt_file=%llu, discarding fscrypt_file size.\n",
1123                                         info->size, size);
1124                         }
1125                 }
1126
1127                 queue_trunc = ceph_fill_file_size(inode, issued,
1128                                         le32_to_cpu(info->truncate_seq),
1129                                         le64_to_cpu(info->truncate_size),
1130                                         size);
1131                 /* only update max_size on auth cap */
1132                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1133                     ci->i_max_size != le64_to_cpu(info->max_size)) {
1134                         doutc(cl, "max_size %lld -> %llu\n",
1135                             ci->i_max_size, le64_to_cpu(info->max_size));
1136                         ci->i_max_size = le64_to_cpu(info->max_size);
1137                 }
1138         }
1139
1140         /* layout and rstat are not tracked by capability, update them if
1141          * the inode info is from auth mds */
1142         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
1143                 if (S_ISDIR(inode->i_mode)) {
1144                         ci->i_dir_layout = iinfo->dir_layout;
1145                         ci->i_rbytes = le64_to_cpu(info->rbytes);
1146                         ci->i_rfiles = le64_to_cpu(info->rfiles);
1147                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
1148                         ci->i_dir_pin = iinfo->dir_pin;
1149                         ci->i_rsnaps = iinfo->rsnaps;
1150                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
1151                 }
1152         }
1153
1154         /* xattrs */
1155         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
1156         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
1157             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
1158                 if (ci->i_xattrs.blob)
1159                         old_blob = ci->i_xattrs.blob;
1160                 ci->i_xattrs.blob = xattr_blob;
1161                 if (xattr_blob)
1162                         memcpy(ci->i_xattrs.blob->vec.iov_base,
1163                                iinfo->xattr_data, iinfo->xattr_len);
1164                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
1165                 ceph_forget_all_cached_acls(inode);
1166                 ceph_security_invalidate_secctx(inode);
1167                 xattr_blob = NULL;
1168         }
1169
1170         /* finally update i_version */
1171         if (le64_to_cpu(info->version) > ci->i_version)
1172                 ci->i_version = le64_to_cpu(info->version);
1173
1174         inode->i_mapping->a_ops = &ceph_aops;
1175
1176         switch (inode->i_mode & S_IFMT) {
1177         case S_IFIFO:
1178         case S_IFBLK:
1179         case S_IFCHR:
1180         case S_IFSOCK:
1181                 inode->i_blkbits = PAGE_SHIFT;
1182                 init_special_inode(inode, inode->i_mode, rdev);
1183                 inode->i_op = &ceph_file_iops;
1184                 break;
1185         case S_IFREG:
1186                 inode->i_op = &ceph_file_iops;
1187                 inode->i_fop = &ceph_file_fops;
1188                 break;
1189         case S_IFLNK:
1190                 if (!ci->i_symlink) {
1191                         u32 symlen = iinfo->symlink_len;
1192                         char *sym;
1193
1194                         spin_unlock(&ci->i_ceph_lock);
1195
1196                         if (IS_ENCRYPTED(inode)) {
1197                                 if (symlen != i_size_read(inode))
1198                                         pr_err_client(cl,
1199                                                 "%p %llx.%llx BAD symlink size %lld\n",
1200                                                 inode, ceph_vinop(inode),
1201                                                 i_size_read(inode));
1202
1203                                 err = decode_encrypted_symlink(mdsc, iinfo->symlink,
1204                                                                symlen, (u8 **)&sym);
1205                                 if (err < 0) {
1206                                         pr_err_client(cl,
1207                                                 "decoding encrypted symlink failed: %d\n",
1208                                                 err);
1209                                         goto out;
1210                                 }
1211                                 symlen = err;
1212                                 i_size_write(inode, symlen);
1213                                 inode->i_blocks = calc_inode_blocks(symlen);
1214                         } else {
1215                                 if (symlen != i_size_read(inode)) {
1216                                         pr_err_client(cl,
1217                                                 "%p %llx.%llx BAD symlink size %lld\n",
1218                                                 inode, ceph_vinop(inode),
1219                                                 i_size_read(inode));
1220                                         i_size_write(inode, symlen);
1221                                         inode->i_blocks = calc_inode_blocks(symlen);
1222                                 }
1223
1224                                 err = -ENOMEM;
1225                                 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
1226                                 if (!sym)
1227                                         goto out;
1228                         }
1229
1230                         spin_lock(&ci->i_ceph_lock);
1231                         if (!ci->i_symlink)
1232                                 ci->i_symlink = sym;
1233                         else
1234                                 kfree(sym); /* lost a race */
1235                 }
1236
1237                 if (IS_ENCRYPTED(inode)) {
1238                         /*
1239                          * Encrypted symlinks need to be decrypted before we can
1240                          * cache their targets in i_link. Don't touch it here.
1241                          */
1242                         inode->i_op = &ceph_encrypted_symlink_iops;
1243                 } else {
1244                         inode->i_link = ci->i_symlink;
1245                         inode->i_op = &ceph_symlink_iops;
1246                 }
1247                 break;
1248         case S_IFDIR:
1249                 inode->i_op = &ceph_dir_iops;
1250                 inode->i_fop = &ceph_dir_fops;
1251                 break;
1252         default:
1253                 pr_err_client(cl, "%p %llx.%llx BAD mode 0%o\n", inode,
1254                               ceph_vinop(inode), inode->i_mode);
1255         }
1256
1257         /* were we issued a capability? */
1258         if (info_caps) {
1259                 if (ceph_snap(inode) == CEPH_NOSNAP) {
1260                         ceph_add_cap(inode, session,
1261                                      le64_to_cpu(info->cap.cap_id),
1262                                      info_caps,
1263                                      le32_to_cpu(info->cap.wanted),
1264                                      le32_to_cpu(info->cap.seq),
1265                                      le32_to_cpu(info->cap.mseq),
1266                                      le64_to_cpu(info->cap.realm),
1267                                      info->cap.flags, &new_cap);
1268
1269                         /* set dir completion flag? */
1270                         if (S_ISDIR(inode->i_mode) &&
1271                             ci->i_files == 0 && ci->i_subdirs == 0 &&
1272                             (info_caps & CEPH_CAP_FILE_SHARED) &&
1273                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1274                             !__ceph_dir_is_complete(ci)) {
1275                                 doutc(cl, " marking %p complete (empty)\n",
1276                                       inode);
1277                                 i_size_write(inode, 0);
1278                                 __ceph_dir_set_complete(ci,
1279                                         atomic64_read(&ci->i_release_count),
1280                                         atomic64_read(&ci->i_ordered_count));
1281                         }
1282
1283                         wake = true;
1284                 } else {
1285                         doutc(cl, " %p got snap_caps %s\n", inode,
1286                               ceph_cap_string(info_caps));
1287                         ci->i_snap_caps |= info_caps;
1288                 }
1289         }
1290
1291         if (iinfo->inline_version > 0 &&
1292             iinfo->inline_version >= ci->i_inline_version) {
1293                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1294                 ci->i_inline_version = iinfo->inline_version;
1295                 if (ceph_has_inline_data(ci) &&
1296                     (locked_page || (info_caps & cache_caps)))
1297                         fill_inline = true;
1298         }
1299
1300         if (cap_fmode >= 0) {
1301                 if (!info_caps)
1302                         pr_warn_client(cl, "mds issued no caps on %llx.%llx\n",
1303                                        ceph_vinop(inode));
1304                 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1305         }
1306
1307         spin_unlock(&ci->i_ceph_lock);
1308
1309         ceph_fscache_register_inode_cookie(inode);
1310
1311         if (fill_inline)
1312                 ceph_fill_inline_data(inode, locked_page,
1313                                       iinfo->inline_data, iinfo->inline_len);
1314
1315         if (wake)
1316                 wake_up_all(&ci->i_cap_wq);
1317
1318         /* queue truncate if we saw i_size decrease */
1319         if (queue_trunc)
1320                 ceph_queue_vmtruncate(inode);
1321
1322         /* populate frag tree */
1323         if (S_ISDIR(inode->i_mode))
1324                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1325
1326         /* update delegation info? */
1327         if (dirinfo)
1328                 ceph_fill_dirfrag(inode, dirinfo);
1329
1330         err = 0;
1331 out:
1332         if (new_cap)
1333                 ceph_put_cap(mdsc, new_cap);
1334         ceph_buffer_put(old_blob);
1335         ceph_buffer_put(xattr_blob);
1336         ceph_put_string(pool_ns);
1337         return err;
1338 }
1339
1340 /*
1341  * caller should hold session s_mutex and dentry->d_lock.
1342  */
1343 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1344                                   struct ceph_mds_reply_lease *lease,
1345                                   struct ceph_mds_session *session,
1346                                   unsigned long from_time,
1347                                   struct ceph_mds_session **old_lease_session)
1348 {
1349         struct ceph_client *cl = ceph_inode_to_client(dir);
1350         struct ceph_dentry_info *di = ceph_dentry(dentry);
1351         unsigned mask = le16_to_cpu(lease->mask);
1352         long unsigned duration = le32_to_cpu(lease->duration_ms);
1353         long unsigned ttl = from_time + (duration * HZ) / 1000;
1354         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1355
1356         doutc(cl, "%p duration %lu ms ttl %lu\n", dentry, duration, ttl);
1357
1358         /* only track leases on regular dentries */
1359         if (ceph_snap(dir) != CEPH_NOSNAP)
1360                 return;
1361
1362         if (mask & CEPH_LEASE_PRIMARY_LINK)
1363                 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1364         else
1365                 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1366
1367         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1368         if (!(mask & CEPH_LEASE_VALID)) {
1369                 __ceph_dentry_dir_lease_touch(di);
1370                 return;
1371         }
1372
1373         if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1374             time_before(ttl, di->time))
1375                 return;  /* we already have a newer lease. */
1376
1377         if (di->lease_session && di->lease_session != session) {
1378                 *old_lease_session = di->lease_session;
1379                 di->lease_session = NULL;
1380         }
1381
1382         if (!di->lease_session)
1383                 di->lease_session = ceph_get_mds_session(session);
1384         di->lease_gen = atomic_read(&session->s_cap_gen);
1385         di->lease_seq = le32_to_cpu(lease->seq);
1386         di->lease_renew_after = half_ttl;
1387         di->lease_renew_from = 0;
1388         di->time = ttl;
1389
1390         __ceph_dentry_lease_touch(di);
1391 }
1392
1393 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1394                                         struct ceph_mds_reply_lease *lease,
1395                                         struct ceph_mds_session *session,
1396                                         unsigned long from_time)
1397 {
1398         struct ceph_mds_session *old_lease_session = NULL;
1399         spin_lock(&dentry->d_lock);
1400         __update_dentry_lease(dir, dentry, lease, session, from_time,
1401                               &old_lease_session);
1402         spin_unlock(&dentry->d_lock);
1403         ceph_put_mds_session(old_lease_session);
1404 }
1405
1406 /*
1407  * update dentry lease without having parent inode locked
1408  */
1409 static void update_dentry_lease_careful(struct dentry *dentry,
1410                                         struct ceph_mds_reply_lease *lease,
1411                                         struct ceph_mds_session *session,
1412                                         unsigned long from_time,
1413                                         char *dname, u32 dname_len,
1414                                         struct ceph_vino *pdvino,
1415                                         struct ceph_vino *ptvino)
1416
1417 {
1418         struct inode *dir;
1419         struct ceph_mds_session *old_lease_session = NULL;
1420
1421         spin_lock(&dentry->d_lock);
1422         /* make sure dentry's name matches target */
1423         if (dentry->d_name.len != dname_len ||
1424             memcmp(dentry->d_name.name, dname, dname_len))
1425                 goto out_unlock;
1426
1427         dir = d_inode(dentry->d_parent);
1428         /* make sure parent matches dvino */
1429         if (!ceph_ino_compare(dir, pdvino))
1430                 goto out_unlock;
1431
1432         /* make sure dentry's inode matches target. NULL ptvino means that
1433          * we expect a negative dentry */
1434         if (ptvino) {
1435                 if (d_really_is_negative(dentry))
1436                         goto out_unlock;
1437                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1438                         goto out_unlock;
1439         } else {
1440                 if (d_really_is_positive(dentry))
1441                         goto out_unlock;
1442         }
1443
1444         __update_dentry_lease(dir, dentry, lease, session,
1445                               from_time, &old_lease_session);
1446 out_unlock:
1447         spin_unlock(&dentry->d_lock);
1448         ceph_put_mds_session(old_lease_session);
1449 }
1450
1451 /*
1452  * splice a dentry to an inode.
1453  * caller must hold directory i_rwsem for this to be safe.
1454  */
1455 static int splice_dentry(struct dentry **pdn, struct inode *in)
1456 {
1457         struct ceph_client *cl = ceph_inode_to_client(in);
1458         struct dentry *dn = *pdn;
1459         struct dentry *realdn;
1460
1461         BUG_ON(d_inode(dn));
1462
1463         if (S_ISDIR(in->i_mode)) {
1464                 /* If inode is directory, d_splice_alias() below will remove
1465                  * 'realdn' from its origin parent. We need to ensure that
1466                  * origin parent's readdir cache will not reference 'realdn'
1467                  */
1468                 realdn = d_find_any_alias(in);
1469                 if (realdn) {
1470                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1471                         spin_lock(&realdn->d_lock);
1472
1473                         realdn->d_op->d_prune(realdn);
1474
1475                         di->time = jiffies;
1476                         di->lease_shared_gen = 0;
1477                         di->offset = 0;
1478
1479                         spin_unlock(&realdn->d_lock);
1480                         dput(realdn);
1481                 }
1482         }
1483
1484         /* dn must be unhashed */
1485         if (!d_unhashed(dn))
1486                 d_drop(dn);
1487         realdn = d_splice_alias(in, dn);
1488         if (IS_ERR(realdn)) {
1489                 pr_err_client(cl, "error %ld %p inode %p ino %llx.%llx\n",
1490                               PTR_ERR(realdn), dn, in, ceph_vinop(in));
1491                 return PTR_ERR(realdn);
1492         }
1493
1494         if (realdn) {
1495                 doutc(cl, "dn %p (%d) spliced with %p (%d) inode %p ino %llx.%llx\n",
1496                       dn, d_count(dn), realdn, d_count(realdn),
1497                       d_inode(realdn), ceph_vinop(d_inode(realdn)));
1498                 dput(dn);
1499                 *pdn = realdn;
1500         } else {
1501                 BUG_ON(!ceph_dentry(dn));
1502                 doutc(cl, "dn %p attached to %p ino %llx.%llx\n", dn,
1503                       d_inode(dn), ceph_vinop(d_inode(dn)));
1504         }
1505         return 0;
1506 }
1507
1508 /*
1509  * Incorporate results into the local cache.  This is either just
1510  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1511  * after a lookup).
1512  *
1513  * A reply may contain
1514  *         a directory inode along with a dentry.
1515  *  and/or a target inode
1516  *
1517  * Called with snap_rwsem (read).
1518  */
1519 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1520 {
1521         struct ceph_mds_session *session = req->r_session;
1522         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1523         struct inode *in = NULL;
1524         struct ceph_vino tvino, dvino;
1525         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1526         struct ceph_client *cl = fsc->client;
1527         int err = 0;
1528
1529         doutc(cl, "%p is_dentry %d is_target %d\n", req,
1530               rinfo->head->is_dentry, rinfo->head->is_target);
1531
1532         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1533                 doutc(cl, "reply is empty!\n");
1534                 if (rinfo->head->result == 0 && req->r_parent)
1535                         ceph_invalidate_dir_request(req);
1536                 return 0;
1537         }
1538
1539         if (rinfo->head->is_dentry) {
1540                 struct inode *dir = req->r_parent;
1541
1542                 if (dir) {
1543                         err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1544                                               rinfo->dirfrag, session, -1,
1545                                               &req->r_caps_reservation);
1546                         if (err < 0)
1547                                 goto done;
1548                 } else {
1549                         WARN_ON_ONCE(1);
1550                 }
1551
1552                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1553                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1554                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1555                         bool is_nokey = false;
1556                         struct qstr dname;
1557                         struct dentry *dn, *parent;
1558                         struct fscrypt_str oname = FSTR_INIT(NULL, 0);
1559                         struct ceph_fname fname = { .dir        = dir,
1560                                                     .name       = rinfo->dname,
1561                                                     .ctext      = rinfo->altname,
1562                                                     .name_len   = rinfo->dname_len,
1563                                                     .ctext_len  = rinfo->altname_len };
1564
1565                         BUG_ON(!rinfo->head->is_target);
1566                         BUG_ON(req->r_dentry);
1567
1568                         parent = d_find_any_alias(dir);
1569                         BUG_ON(!parent);
1570
1571                         err = ceph_fname_alloc_buffer(dir, &oname);
1572                         if (err < 0) {
1573                                 dput(parent);
1574                                 goto done;
1575                         }
1576
1577                         err = ceph_fname_to_usr(&fname, NULL, &oname, &is_nokey);
1578                         if (err < 0) {
1579                                 dput(parent);
1580                                 ceph_fname_free_buffer(dir, &oname);
1581                                 goto done;
1582                         }
1583                         dname.name = oname.name;
1584                         dname.len = oname.len;
1585                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1586                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1587                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1588 retry_lookup:
1589                         dn = d_lookup(parent, &dname);
1590                         doutc(cl, "d_lookup on parent=%p name=%.*s got %p\n",
1591                               parent, dname.len, dname.name, dn);
1592
1593                         if (!dn) {
1594                                 dn = d_alloc(parent, &dname);
1595                                 doutc(cl, "d_alloc %p '%.*s' = %p\n", parent,
1596                                       dname.len, dname.name, dn);
1597                                 if (!dn) {
1598                                         dput(parent);
1599                                         ceph_fname_free_buffer(dir, &oname);
1600                                         err = -ENOMEM;
1601                                         goto done;
1602                                 }
1603                                 if (is_nokey) {
1604                                         spin_lock(&dn->d_lock);
1605                                         dn->d_flags |= DCACHE_NOKEY_NAME;
1606                                         spin_unlock(&dn->d_lock);
1607                                 }
1608                                 err = 0;
1609                         } else if (d_really_is_positive(dn) &&
1610                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1611                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1612                                 doutc(cl, " dn %p points to wrong inode %p\n",
1613                                       dn, d_inode(dn));
1614                                 ceph_dir_clear_ordered(dir);
1615                                 d_delete(dn);
1616                                 dput(dn);
1617                                 goto retry_lookup;
1618                         }
1619                         ceph_fname_free_buffer(dir, &oname);
1620
1621                         req->r_dentry = dn;
1622                         dput(parent);
1623                 }
1624         }
1625
1626         if (rinfo->head->is_target) {
1627                 /* Should be filled in by handle_reply */
1628                 BUG_ON(!req->r_target_inode);
1629
1630                 in = req->r_target_inode;
1631                 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1632                                 NULL, session,
1633                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1634                                  !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1635                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1636                                 &req->r_caps_reservation);
1637                 if (err < 0) {
1638                         pr_err_client(cl, "badness %p %llx.%llx\n", in,
1639                                       ceph_vinop(in));
1640                         req->r_target_inode = NULL;
1641                         if (in->i_state & I_NEW)
1642                                 discard_new_inode(in);
1643                         else
1644                                 iput(in);
1645                         goto done;
1646                 }
1647                 if (in->i_state & I_NEW)
1648                         unlock_new_inode(in);
1649         }
1650
1651         /*
1652          * ignore null lease/binding on snapdir ENOENT, or else we
1653          * will have trouble splicing in the virtual snapdir later
1654          */
1655         if (rinfo->head->is_dentry &&
1656             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1657             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1658             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1659                                                fsc->mount_options->snapdir_name,
1660                                                req->r_dentry->d_name.len))) {
1661                 /*
1662                  * lookup link rename   : null -> possibly existing inode
1663                  * mknod symlink mkdir  : null -> new inode
1664                  * unlink               : linked -> null
1665                  */
1666                 struct inode *dir = req->r_parent;
1667                 struct dentry *dn = req->r_dentry;
1668                 bool have_dir_cap, have_lease;
1669
1670                 BUG_ON(!dn);
1671                 BUG_ON(!dir);
1672                 BUG_ON(d_inode(dn->d_parent) != dir);
1673
1674                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1675                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1676
1677                 BUG_ON(ceph_ino(dir) != dvino.ino);
1678                 BUG_ON(ceph_snap(dir) != dvino.snap);
1679
1680                 /* do we have a lease on the whole dir? */
1681                 have_dir_cap =
1682                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1683                          CEPH_CAP_FILE_SHARED);
1684
1685                 /* do we have a dn lease? */
1686                 have_lease = have_dir_cap ||
1687                         le32_to_cpu(rinfo->dlease->duration_ms);
1688                 if (!have_lease)
1689                         doutc(cl, "no dentry lease or dir cap\n");
1690
1691                 /* rename? */
1692                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1693                         struct inode *olddir = req->r_old_dentry_dir;
1694                         BUG_ON(!olddir);
1695
1696                         doutc(cl, " src %p '%pd' dst %p '%pd'\n",
1697                               req->r_old_dentry, req->r_old_dentry, dn, dn);
1698                         doutc(cl, "doing d_move %p -> %p\n", req->r_old_dentry, dn);
1699
1700                         /* d_move screws up sibling dentries' offsets */
1701                         ceph_dir_clear_ordered(dir);
1702                         ceph_dir_clear_ordered(olddir);
1703
1704                         d_move(req->r_old_dentry, dn);
1705                         doutc(cl, " src %p '%pd' dst %p '%pd'\n",
1706                               req->r_old_dentry, req->r_old_dentry, dn, dn);
1707
1708                         /* ensure target dentry is invalidated, despite
1709                            rehashing bug in vfs_rename_dir */
1710                         ceph_invalidate_dentry_lease(dn);
1711
1712                         doutc(cl, "dn %p gets new offset %lld\n",
1713                               req->r_old_dentry,
1714                               ceph_dentry(req->r_old_dentry)->offset);
1715
1716                         /* swap r_dentry and r_old_dentry in case that
1717                          * splice_dentry() gets called later. This is safe
1718                          * because no other place will use them */
1719                         req->r_dentry = req->r_old_dentry;
1720                         req->r_old_dentry = dn;
1721                         dn = req->r_dentry;
1722                 }
1723
1724                 /* null dentry? */
1725                 if (!rinfo->head->is_target) {
1726                         doutc(cl, "null dentry\n");
1727                         if (d_really_is_positive(dn)) {
1728                                 doutc(cl, "d_delete %p\n", dn);
1729                                 ceph_dir_clear_ordered(dir);
1730                                 d_delete(dn);
1731                         } else if (have_lease) {
1732                                 if (d_unhashed(dn))
1733                                         d_add(dn, NULL);
1734                         }
1735
1736                         if (!d_unhashed(dn) && have_lease)
1737                                 update_dentry_lease(dir, dn,
1738                                                     rinfo->dlease, session,
1739                                                     req->r_request_started);
1740                         goto done;
1741                 }
1742
1743                 /* attach proper inode */
1744                 if (d_really_is_negative(dn)) {
1745                         ceph_dir_clear_ordered(dir);
1746                         ihold(in);
1747                         err = splice_dentry(&req->r_dentry, in);
1748                         if (err < 0)
1749                                 goto done;
1750                         dn = req->r_dentry;  /* may have spliced */
1751                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1752                         doutc(cl, " %p links to %p %llx.%llx, not %llx.%llx\n",
1753                               dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1754                               ceph_vinop(in));
1755                         d_invalidate(dn);
1756                         have_lease = false;
1757                 }
1758
1759                 if (have_lease) {
1760                         update_dentry_lease(dir, dn,
1761                                             rinfo->dlease, session,
1762                                             req->r_request_started);
1763                 }
1764                 doutc(cl, " final dn %p\n", dn);
1765         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1766                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1767                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1768                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1769                 struct inode *dir = req->r_parent;
1770
1771                 /* fill out a snapdir LOOKUPSNAP dentry */
1772                 BUG_ON(!dir);
1773                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1774                 BUG_ON(!req->r_dentry);
1775                 doutc(cl, " linking snapped dir %p to dn %p\n", in,
1776                       req->r_dentry);
1777                 ceph_dir_clear_ordered(dir);
1778                 ihold(in);
1779                 err = splice_dentry(&req->r_dentry, in);
1780                 if (err < 0)
1781                         goto done;
1782         } else if (rinfo->head->is_dentry && req->r_dentry) {
1783                 /* parent inode is not locked, be carefull */
1784                 struct ceph_vino *ptvino = NULL;
1785                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1786                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1787                 if (rinfo->head->is_target) {
1788                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1789                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1790                         ptvino = &tvino;
1791                 }
1792                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1793                                             session, req->r_request_started,
1794                                             rinfo->dname, rinfo->dname_len,
1795                                             &dvino, ptvino);
1796         }
1797 done:
1798         doutc(cl, "done err=%d\n", err);
1799         return err;
1800 }
1801
1802 /*
1803  * Prepopulate our cache with readdir results, leases, etc.
1804  */
1805 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1806                                            struct ceph_mds_session *session)
1807 {
1808         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1809         struct ceph_client *cl = session->s_mdsc->fsc->client;
1810         int i, err = 0;
1811
1812         for (i = 0; i < rinfo->dir_nr; i++) {
1813                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1814                 struct ceph_vino vino;
1815                 struct inode *in;
1816                 int rc;
1817
1818                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1819                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1820
1821                 in = ceph_get_inode(req->r_dentry->d_sb, vino, NULL);
1822                 if (IS_ERR(in)) {
1823                         err = PTR_ERR(in);
1824                         doutc(cl, "badness got %d\n", err);
1825                         continue;
1826                 }
1827                 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1828                                      -1, &req->r_caps_reservation);
1829                 if (rc < 0) {
1830                         pr_err_client(cl, "inode badness on %p got %d\n", in,
1831                                       rc);
1832                         err = rc;
1833                         if (in->i_state & I_NEW) {
1834                                 ihold(in);
1835                                 discard_new_inode(in);
1836                         }
1837                 } else if (in->i_state & I_NEW) {
1838                         unlock_new_inode(in);
1839                 }
1840
1841                 iput(in);
1842         }
1843
1844         return err;
1845 }
1846
1847 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1848 {
1849         if (ctl->page) {
1850                 kunmap(ctl->page);
1851                 put_page(ctl->page);
1852                 ctl->page = NULL;
1853         }
1854 }
1855
1856 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1857                               struct ceph_readdir_cache_control *ctl,
1858                               struct ceph_mds_request *req)
1859 {
1860         struct ceph_client *cl = ceph_inode_to_client(dir);
1861         struct ceph_inode_info *ci = ceph_inode(dir);
1862         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1863         unsigned idx = ctl->index % nsize;
1864         pgoff_t pgoff = ctl->index / nsize;
1865
1866         if (!ctl->page || pgoff != ctl->page->index) {
1867                 ceph_readdir_cache_release(ctl);
1868                 if (idx == 0)
1869                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1870                 else
1871                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1872                 if (!ctl->page) {
1873                         ctl->index = -1;
1874                         return idx == 0 ? -ENOMEM : 0;
1875                 }
1876                 /* reading/filling the cache are serialized by
1877                  * i_rwsem, no need to use page lock */
1878                 unlock_page(ctl->page);
1879                 ctl->dentries = kmap(ctl->page);
1880                 if (idx == 0)
1881                         memset(ctl->dentries, 0, PAGE_SIZE);
1882         }
1883
1884         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1885             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1886                 doutc(cl, "dn %p idx %d\n", dn, ctl->index);
1887                 ctl->dentries[idx] = dn;
1888                 ctl->index++;
1889         } else {
1890                 doutc(cl, "disable readdir cache\n");
1891                 ctl->index = -1;
1892         }
1893         return 0;
1894 }
1895
1896 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1897                              struct ceph_mds_session *session)
1898 {
1899         struct dentry *parent = req->r_dentry;
1900         struct inode *inode = d_inode(parent);
1901         struct ceph_inode_info *ci = ceph_inode(inode);
1902         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1903         struct ceph_client *cl = session->s_mdsc->fsc->client;
1904         struct qstr dname;
1905         struct dentry *dn;
1906         struct inode *in;
1907         int err = 0, skipped = 0, ret, i;
1908         u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1909         u32 last_hash = 0;
1910         u32 fpos_offset;
1911         struct ceph_readdir_cache_control cache_ctl = {};
1912
1913         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1914                 return readdir_prepopulate_inodes_only(req, session);
1915
1916         if (rinfo->hash_order) {
1917                 if (req->r_path2) {
1918                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1919                                                   req->r_path2,
1920                                                   strlen(req->r_path2));
1921                         last_hash = ceph_frag_value(last_hash);
1922                 } else if (rinfo->offset_hash) {
1923                         /* mds understands offset_hash */
1924                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1925                         last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1926                 }
1927         }
1928
1929         if (rinfo->dir_dir &&
1930             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1931                 doutc(cl, "got new frag %x -> %x\n", frag,
1932                             le32_to_cpu(rinfo->dir_dir->frag));
1933                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1934                 if (!rinfo->hash_order)
1935                         req->r_readdir_offset = 2;
1936         }
1937
1938         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1939                 doutc(cl, "%d items under SNAPDIR dn %p\n",
1940                       rinfo->dir_nr, parent);
1941         } else {
1942                 doutc(cl, "%d items under dn %p\n", rinfo->dir_nr, parent);
1943                 if (rinfo->dir_dir)
1944                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1945
1946                 if (ceph_frag_is_leftmost(frag) &&
1947                     req->r_readdir_offset == 2 &&
1948                     !(rinfo->hash_order && last_hash)) {
1949                         /* note dir version at start of readdir so we can
1950                          * tell if any dentries get dropped */
1951                         req->r_dir_release_cnt =
1952                                 atomic64_read(&ci->i_release_count);
1953                         req->r_dir_ordered_cnt =
1954                                 atomic64_read(&ci->i_ordered_count);
1955                         req->r_readdir_cache_idx = 0;
1956                 }
1957         }
1958
1959         cache_ctl.index = req->r_readdir_cache_idx;
1960         fpos_offset = req->r_readdir_offset;
1961
1962         /* FIXME: release caps/leases if error occurs */
1963         for (i = 0; i < rinfo->dir_nr; i++) {
1964                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1965                 struct ceph_vino tvino;
1966
1967                 dname.name = rde->name;
1968                 dname.len = rde->name_len;
1969                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1970
1971                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1972                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1973
1974                 if (rinfo->hash_order) {
1975                         u32 hash = ceph_frag_value(rde->raw_hash);
1976                         if (hash != last_hash)
1977                                 fpos_offset = 2;
1978                         last_hash = hash;
1979                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1980                 } else {
1981                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1982                 }
1983
1984 retry_lookup:
1985                 dn = d_lookup(parent, &dname);
1986                 doutc(cl, "d_lookup on parent=%p name=%.*s got %p\n",
1987                       parent, dname.len, dname.name, dn);
1988
1989                 if (!dn) {
1990                         dn = d_alloc(parent, &dname);
1991                         doutc(cl, "d_alloc %p '%.*s' = %p\n", parent,
1992                               dname.len, dname.name, dn);
1993                         if (!dn) {
1994                                 doutc(cl, "d_alloc badness\n");
1995                                 err = -ENOMEM;
1996                                 goto out;
1997                         }
1998                         if (rde->is_nokey) {
1999                                 spin_lock(&dn->d_lock);
2000                                 dn->d_flags |= DCACHE_NOKEY_NAME;
2001                                 spin_unlock(&dn->d_lock);
2002                         }
2003                 } else if (d_really_is_positive(dn) &&
2004                            (ceph_ino(d_inode(dn)) != tvino.ino ||
2005                             ceph_snap(d_inode(dn)) != tvino.snap)) {
2006                         struct ceph_dentry_info *di = ceph_dentry(dn);
2007                         doutc(cl, " dn %p points to wrong inode %p\n",
2008                               dn, d_inode(dn));
2009
2010                         spin_lock(&dn->d_lock);
2011                         if (di->offset > 0 &&
2012                             di->lease_shared_gen ==
2013                             atomic_read(&ci->i_shared_gen)) {
2014                                 __ceph_dir_clear_ordered(ci);
2015                                 di->offset = 0;
2016                         }
2017                         spin_unlock(&dn->d_lock);
2018
2019                         d_delete(dn);
2020                         dput(dn);
2021                         goto retry_lookup;
2022                 }
2023
2024                 /* inode */
2025                 if (d_really_is_positive(dn)) {
2026                         in = d_inode(dn);
2027                 } else {
2028                         in = ceph_get_inode(parent->d_sb, tvino, NULL);
2029                         if (IS_ERR(in)) {
2030                                 doutc(cl, "new_inode badness\n");
2031                                 d_drop(dn);
2032                                 dput(dn);
2033                                 err = PTR_ERR(in);
2034                                 goto out;
2035                         }
2036                 }
2037
2038                 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
2039                                       -1, &req->r_caps_reservation);
2040                 if (ret < 0) {
2041                         pr_err_client(cl, "badness on %p %llx.%llx\n", in,
2042                                       ceph_vinop(in));
2043                         if (d_really_is_negative(dn)) {
2044                                 if (in->i_state & I_NEW) {
2045                                         ihold(in);
2046                                         discard_new_inode(in);
2047                                 }
2048                                 iput(in);
2049                         }
2050                         d_drop(dn);
2051                         err = ret;
2052                         goto next_item;
2053                 }
2054                 if (in->i_state & I_NEW)
2055                         unlock_new_inode(in);
2056
2057                 if (d_really_is_negative(dn)) {
2058                         if (ceph_security_xattr_deadlock(in)) {
2059                                 doutc(cl, " skip splicing dn %p to inode %p"
2060                                       " (security xattr deadlock)\n", dn, in);
2061                                 iput(in);
2062                                 skipped++;
2063                                 goto next_item;
2064                         }
2065
2066                         err = splice_dentry(&dn, in);
2067                         if (err < 0)
2068                                 goto next_item;
2069                 }
2070
2071                 ceph_dentry(dn)->offset = rde->offset;
2072
2073                 update_dentry_lease(d_inode(parent), dn,
2074                                     rde->lease, req->r_session,
2075                                     req->r_request_started);
2076
2077                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
2078                         ret = fill_readdir_cache(d_inode(parent), dn,
2079                                                  &cache_ctl, req);
2080                         if (ret < 0)
2081                                 err = ret;
2082                 }
2083 next_item:
2084                 dput(dn);
2085         }
2086 out:
2087         if (err == 0 && skipped == 0) {
2088                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
2089                 req->r_readdir_cache_idx = cache_ctl.index;
2090         }
2091         ceph_readdir_cache_release(&cache_ctl);
2092         doutc(cl, "done\n");
2093         return err;
2094 }
2095
2096 bool ceph_inode_set_size(struct inode *inode, loff_t size)
2097 {
2098         struct ceph_client *cl = ceph_inode_to_client(inode);
2099         struct ceph_inode_info *ci = ceph_inode(inode);
2100         bool ret;
2101
2102         spin_lock(&ci->i_ceph_lock);
2103         doutc(cl, "set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
2104         i_size_write(inode, size);
2105         ceph_fscache_update(inode);
2106         inode->i_blocks = calc_inode_blocks(size);
2107
2108         ret = __ceph_should_report_size(ci);
2109
2110         spin_unlock(&ci->i_ceph_lock);
2111
2112         return ret;
2113 }
2114
2115 void ceph_queue_inode_work(struct inode *inode, int work_bit)
2116 {
2117         struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2118         struct ceph_client *cl = fsc->client;
2119         struct ceph_inode_info *ci = ceph_inode(inode);
2120         set_bit(work_bit, &ci->i_work_mask);
2121
2122         ihold(inode);
2123         if (queue_work(fsc->inode_wq, &ci->i_work)) {
2124                 doutc(cl, "%p %llx.%llx mask=%lx\n", inode,
2125                       ceph_vinop(inode), ci->i_work_mask);
2126         } else {
2127                 doutc(cl, "%p %llx.%llx already queued, mask=%lx\n",
2128                       inode, ceph_vinop(inode), ci->i_work_mask);
2129                 iput(inode);
2130         }
2131 }
2132
2133 static void ceph_do_invalidate_pages(struct inode *inode)
2134 {
2135         struct ceph_client *cl = ceph_inode_to_client(inode);
2136         struct ceph_inode_info *ci = ceph_inode(inode);
2137         u32 orig_gen;
2138         int check = 0;
2139
2140         ceph_fscache_invalidate(inode, false);
2141
2142         mutex_lock(&ci->i_truncate_mutex);
2143
2144         if (ceph_inode_is_shutdown(inode)) {
2145                 pr_warn_ratelimited_client(cl,
2146                         "%p %llx.%llx is shut down\n", inode,
2147                         ceph_vinop(inode));
2148                 mapping_set_error(inode->i_mapping, -EIO);
2149                 truncate_pagecache(inode, 0);
2150                 mutex_unlock(&ci->i_truncate_mutex);
2151                 goto out;
2152         }
2153
2154         spin_lock(&ci->i_ceph_lock);
2155         doutc(cl, "%p %llx.%llx gen %d revoking %d\n", inode,
2156               ceph_vinop(inode), ci->i_rdcache_gen, ci->i_rdcache_revoking);
2157         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
2158                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2159                         check = 1;
2160                 spin_unlock(&ci->i_ceph_lock);
2161                 mutex_unlock(&ci->i_truncate_mutex);
2162                 goto out;
2163         }
2164         orig_gen = ci->i_rdcache_gen;
2165         spin_unlock(&ci->i_ceph_lock);
2166
2167         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
2168                 pr_err_client(cl, "invalidate_inode_pages2 %llx.%llx failed\n",
2169                               ceph_vinop(inode));
2170         }
2171
2172         spin_lock(&ci->i_ceph_lock);
2173         if (orig_gen == ci->i_rdcache_gen &&
2174             orig_gen == ci->i_rdcache_revoking) {
2175                 doutc(cl, "%p %llx.%llx gen %d successful\n", inode,
2176                       ceph_vinop(inode), ci->i_rdcache_gen);
2177                 ci->i_rdcache_revoking--;
2178                 check = 1;
2179         } else {
2180                 doutc(cl, "%p %llx.%llx gen %d raced, now %d revoking %d\n",
2181                       inode, ceph_vinop(inode), orig_gen, ci->i_rdcache_gen,
2182                       ci->i_rdcache_revoking);
2183                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2184                         check = 1;
2185         }
2186         spin_unlock(&ci->i_ceph_lock);
2187         mutex_unlock(&ci->i_truncate_mutex);
2188 out:
2189         if (check)
2190                 ceph_check_caps(ci, 0);
2191 }
2192
2193 /*
2194  * Make sure any pending truncation is applied before doing anything
2195  * that may depend on it.
2196  */
2197 void __ceph_do_pending_vmtruncate(struct inode *inode)
2198 {
2199         struct ceph_client *cl = ceph_inode_to_client(inode);
2200         struct ceph_inode_info *ci = ceph_inode(inode);
2201         u64 to;
2202         int wrbuffer_refs, finish = 0;
2203
2204         mutex_lock(&ci->i_truncate_mutex);
2205 retry:
2206         spin_lock(&ci->i_ceph_lock);
2207         if (ci->i_truncate_pending == 0) {
2208                 doutc(cl, "%p %llx.%llx none pending\n", inode,
2209                       ceph_vinop(inode));
2210                 spin_unlock(&ci->i_ceph_lock);
2211                 mutex_unlock(&ci->i_truncate_mutex);
2212                 return;
2213         }
2214
2215         /*
2216          * make sure any dirty snapped pages are flushed before we
2217          * possibly truncate them.. so write AND block!
2218          */
2219         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
2220                 spin_unlock(&ci->i_ceph_lock);
2221                 doutc(cl, "%p %llx.%llx flushing snaps first\n", inode,
2222                       ceph_vinop(inode));
2223                 filemap_write_and_wait_range(&inode->i_data, 0,
2224                                              inode->i_sb->s_maxbytes);
2225                 goto retry;
2226         }
2227
2228         /* there should be no reader or writer */
2229         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
2230
2231         to = ci->i_truncate_pagecache_size;
2232         wrbuffer_refs = ci->i_wrbuffer_ref;
2233         doutc(cl, "%p %llx.%llx (%d) to %lld\n", inode, ceph_vinop(inode),
2234               ci->i_truncate_pending, to);
2235         spin_unlock(&ci->i_ceph_lock);
2236
2237         ceph_fscache_resize(inode, to);
2238         truncate_pagecache(inode, to);
2239
2240         spin_lock(&ci->i_ceph_lock);
2241         if (to == ci->i_truncate_pagecache_size) {
2242                 ci->i_truncate_pending = 0;
2243                 finish = 1;
2244         }
2245         spin_unlock(&ci->i_ceph_lock);
2246         if (!finish)
2247                 goto retry;
2248
2249         mutex_unlock(&ci->i_truncate_mutex);
2250
2251         if (wrbuffer_refs == 0)
2252                 ceph_check_caps(ci, 0);
2253
2254         wake_up_all(&ci->i_cap_wq);
2255 }
2256
2257 static void ceph_inode_work(struct work_struct *work)
2258 {
2259         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
2260                                                  i_work);
2261         struct inode *inode = &ci->netfs.inode;
2262         struct ceph_client *cl = ceph_inode_to_client(inode);
2263
2264         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
2265                 doutc(cl, "writeback %p %llx.%llx\n", inode, ceph_vinop(inode));
2266                 filemap_fdatawrite(&inode->i_data);
2267         }
2268         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2269                 ceph_do_invalidate_pages(inode);
2270
2271         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2272                 __ceph_do_pending_vmtruncate(inode);
2273
2274         if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
2275                 ceph_check_caps(ci, 0);
2276
2277         if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
2278                 ceph_flush_snaps(ci, NULL);
2279
2280         iput(inode);
2281 }
2282
2283 static const char *ceph_encrypted_get_link(struct dentry *dentry,
2284                                            struct inode *inode,
2285                                            struct delayed_call *done)
2286 {
2287         struct ceph_inode_info *ci = ceph_inode(inode);
2288
2289         if (!dentry)
2290                 return ERR_PTR(-ECHILD);
2291
2292         return fscrypt_get_symlink(inode, ci->i_symlink, i_size_read(inode),
2293                                    done);
2294 }
2295
2296 static int ceph_encrypted_symlink_getattr(struct mnt_idmap *idmap,
2297                                           const struct path *path,
2298                                           struct kstat *stat, u32 request_mask,
2299                                           unsigned int query_flags)
2300 {
2301         int ret;
2302
2303         ret = ceph_getattr(idmap, path, stat, request_mask, query_flags);
2304         if (ret)
2305                 return ret;
2306         return fscrypt_symlink_getattr(path, stat);
2307 }
2308
2309 /*
2310  * symlinks
2311  */
2312 static const struct inode_operations ceph_symlink_iops = {
2313         .get_link = simple_get_link,
2314         .setattr = ceph_setattr,
2315         .getattr = ceph_getattr,
2316         .listxattr = ceph_listxattr,
2317 };
2318
2319 static const struct inode_operations ceph_encrypted_symlink_iops = {
2320         .get_link = ceph_encrypted_get_link,
2321         .setattr = ceph_setattr,
2322         .getattr = ceph_encrypted_symlink_getattr,
2323         .listxattr = ceph_listxattr,
2324 };
2325
2326 /*
2327  * Transfer the encrypted last block to the MDS and the MDS
2328  * will help update it when truncating a smaller size.
2329  *
2330  * We don't support a PAGE_SIZE that is smaller than the
2331  * CEPH_FSCRYPT_BLOCK_SIZE.
2332  */
2333 static int fill_fscrypt_truncate(struct inode *inode,
2334                                  struct ceph_mds_request *req,
2335                                  struct iattr *attr)
2336 {
2337         struct ceph_client *cl = ceph_inode_to_client(inode);
2338         struct ceph_inode_info *ci = ceph_inode(inode);
2339         int boff = attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE;
2340         loff_t pos, orig_pos = round_down(attr->ia_size,
2341                                           CEPH_FSCRYPT_BLOCK_SIZE);
2342         u64 block = orig_pos >> CEPH_FSCRYPT_BLOCK_SHIFT;
2343         struct ceph_pagelist *pagelist = NULL;
2344         struct kvec iov = {0};
2345         struct iov_iter iter;
2346         struct page *page = NULL;
2347         struct ceph_fscrypt_truncate_size_header header;
2348         int retry_op = 0;
2349         int len = CEPH_FSCRYPT_BLOCK_SIZE;
2350         loff_t i_size = i_size_read(inode);
2351         int got, ret, issued;
2352         u64 objver;
2353
2354         ret = __ceph_get_caps(inode, NULL, CEPH_CAP_FILE_RD, 0, -1, &got);
2355         if (ret < 0)
2356                 return ret;
2357
2358         issued = __ceph_caps_issued(ci, NULL);
2359
2360         doutc(cl, "size %lld -> %lld got cap refs on %s, issued %s\n",
2361               i_size, attr->ia_size, ceph_cap_string(got),
2362               ceph_cap_string(issued));
2363
2364         /* Try to writeback the dirty pagecaches */
2365         if (issued & (CEPH_CAP_FILE_BUFFER)) {
2366                 loff_t lend = orig_pos + CEPH_FSCRYPT_BLOCK_SHIFT - 1;
2367
2368                 ret = filemap_write_and_wait_range(inode->i_mapping,
2369                                                    orig_pos, lend);
2370                 if (ret < 0)
2371                         goto out;
2372         }
2373
2374         page = __page_cache_alloc(GFP_KERNEL);
2375         if (page == NULL) {
2376                 ret = -ENOMEM;
2377                 goto out;
2378         }
2379
2380         pagelist = ceph_pagelist_alloc(GFP_KERNEL);
2381         if (!pagelist) {
2382                 ret = -ENOMEM;
2383                 goto out;
2384         }
2385
2386         iov.iov_base = kmap_local_page(page);
2387         iov.iov_len = len;
2388         iov_iter_kvec(&iter, READ, &iov, 1, len);
2389
2390         pos = orig_pos;
2391         ret = __ceph_sync_read(inode, &pos, &iter, &retry_op, &objver);
2392         if (ret < 0)
2393                 goto out;
2394
2395         /* Insert the header first */
2396         header.ver = 1;
2397         header.compat = 1;
2398         header.change_attr = cpu_to_le64(inode_peek_iversion_raw(inode));
2399
2400         /*
2401          * Always set the block_size to CEPH_FSCRYPT_BLOCK_SIZE,
2402          * because in MDS it may need this to do the truncate.
2403          */
2404         header.block_size = cpu_to_le32(CEPH_FSCRYPT_BLOCK_SIZE);
2405
2406         /*
2407          * If we hit a hole here, we should just skip filling
2408          * the fscrypt for the request, because once the fscrypt
2409          * is enabled, the file will be split into many blocks
2410          * with the size of CEPH_FSCRYPT_BLOCK_SIZE, if there
2411          * has a hole, the hole size should be multiple of block
2412          * size.
2413          *
2414          * If the Rados object doesn't exist, it will be set to 0.
2415          */
2416         if (!objver) {
2417                 doutc(cl, "hit hole, ppos %lld < size %lld\n", pos, i_size);
2418
2419                 header.data_len = cpu_to_le32(8 + 8 + 4);
2420                 header.file_offset = 0;
2421                 ret = 0;
2422         } else {
2423                 header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
2424                 header.file_offset = cpu_to_le64(orig_pos);
2425
2426                 doutc(cl, "encrypt block boff/bsize %d/%lu\n", boff,
2427                       CEPH_FSCRYPT_BLOCK_SIZE);
2428
2429                 /* truncate and zero out the extra contents for the last block */
2430                 memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
2431
2432                 /* encrypt the last block */
2433                 ret = ceph_fscrypt_encrypt_block_inplace(inode, page,
2434                                                     CEPH_FSCRYPT_BLOCK_SIZE,
2435                                                     0, block,
2436                                                     GFP_KERNEL);
2437                 if (ret)
2438                         goto out;
2439         }
2440
2441         /* Insert the header */
2442         ret = ceph_pagelist_append(pagelist, &header, sizeof(header));
2443         if (ret)
2444                 goto out;
2445
2446         if (header.block_size) {
2447                 /* Append the last block contents to pagelist */
2448                 ret = ceph_pagelist_append(pagelist, iov.iov_base,
2449                                            CEPH_FSCRYPT_BLOCK_SIZE);
2450                 if (ret)
2451                         goto out;
2452         }
2453         req->r_pagelist = pagelist;
2454 out:
2455         doutc(cl, "%p %llx.%llx size dropping cap refs on %s\n", inode,
2456               ceph_vinop(inode), ceph_cap_string(got));
2457         ceph_put_cap_refs(ci, got);
2458         if (iov.iov_base)
2459                 kunmap_local(iov.iov_base);
2460         if (page)
2461                 __free_pages(page, 0);
2462         if (ret && pagelist)
2463                 ceph_pagelist_release(pagelist);
2464         return ret;
2465 }
2466
2467 int __ceph_setattr(struct mnt_idmap *idmap, struct inode *inode,
2468                    struct iattr *attr, struct ceph_iattr *cia)
2469 {
2470         struct ceph_inode_info *ci = ceph_inode(inode);
2471         unsigned int ia_valid = attr->ia_valid;
2472         struct ceph_mds_request *req;
2473         struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
2474         struct ceph_client *cl = ceph_inode_to_client(inode);
2475         struct ceph_cap_flush *prealloc_cf;
2476         loff_t isize = i_size_read(inode);
2477         int issued;
2478         int release = 0, dirtied = 0;
2479         int mask = 0;
2480         int err = 0;
2481         int inode_dirty_flags = 0;
2482         bool lock_snap_rwsem = false;
2483         bool fill_fscrypt;
2484         int truncate_retry = 20; /* The RMW will take around 50ms */
2485         struct dentry *dentry;
2486         char *path;
2487         int pathlen;
2488         u64 pathbase;
2489         bool do_sync = false;
2490
2491         dentry = d_find_alias(inode);
2492         if (!dentry) {
2493                 do_sync = true;
2494         } else {
2495                 path = ceph_mdsc_build_path(mdsc, dentry, &pathlen, &pathbase, 0);
2496                 if (IS_ERR(path)) {
2497                         do_sync = true;
2498                         err = 0;
2499                 } else {
2500                         err = ceph_mds_check_access(mdsc, path, MAY_WRITE);
2501                 }
2502                 ceph_mdsc_free_path(path, pathlen);
2503                 dput(dentry);
2504
2505                 /* For none EACCES cases will let the MDS do the mds auth check */
2506                 if (err == -EACCES) {
2507                         return err;
2508                 } else if (err < 0) {
2509                         do_sync = true;
2510                         err = 0;
2511                 }
2512         }
2513
2514 retry:
2515         prealloc_cf = ceph_alloc_cap_flush();
2516         if (!prealloc_cf)
2517                 return -ENOMEM;
2518
2519         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2520                                        USE_AUTH_MDS);
2521         if (IS_ERR(req)) {
2522                 ceph_free_cap_flush(prealloc_cf);
2523                 return PTR_ERR(req);
2524         }
2525
2526         fill_fscrypt = false;
2527         spin_lock(&ci->i_ceph_lock);
2528         issued = __ceph_caps_issued(ci, NULL);
2529
2530         if (!ci->i_head_snapc &&
2531             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2532                 lock_snap_rwsem = true;
2533                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2534                         spin_unlock(&ci->i_ceph_lock);
2535                         down_read(&mdsc->snap_rwsem);
2536                         spin_lock(&ci->i_ceph_lock);
2537                         issued = __ceph_caps_issued(ci, NULL);
2538                 }
2539         }
2540
2541         doutc(cl, "%p %llx.%llx issued %s\n", inode, ceph_vinop(inode),
2542               ceph_cap_string(issued));
2543 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
2544         if (cia && cia->fscrypt_auth) {
2545                 u32 len = ceph_fscrypt_auth_len(cia->fscrypt_auth);
2546
2547                 if (len > sizeof(*cia->fscrypt_auth)) {
2548                         err = -EINVAL;
2549                         spin_unlock(&ci->i_ceph_lock);
2550                         goto out;
2551                 }
2552
2553                 doutc(cl, "%p %llx.%llx fscrypt_auth len %u to %u)\n", inode,
2554                       ceph_vinop(inode), ci->fscrypt_auth_len, len);
2555
2556                 /* It should never be re-set once set */
2557                 WARN_ON_ONCE(ci->fscrypt_auth);
2558
2559                 if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2560                         dirtied |= CEPH_CAP_AUTH_EXCL;
2561                         kfree(ci->fscrypt_auth);
2562                         ci->fscrypt_auth = (u8 *)cia->fscrypt_auth;
2563                         ci->fscrypt_auth_len = len;
2564                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2565                            ci->fscrypt_auth_len != len ||
2566                            memcmp(ci->fscrypt_auth, cia->fscrypt_auth, len)) {
2567                         req->r_fscrypt_auth = cia->fscrypt_auth;
2568                         mask |= CEPH_SETATTR_FSCRYPT_AUTH;
2569                         release |= CEPH_CAP_AUTH_SHARED;
2570                 }
2571                 cia->fscrypt_auth = NULL;
2572         }
2573 #else
2574         if (cia && cia->fscrypt_auth) {
2575                 err = -EINVAL;
2576                 spin_unlock(&ci->i_ceph_lock);
2577                 goto out;
2578         }
2579 #endif /* CONFIG_FS_ENCRYPTION */
2580
2581         if (ia_valid & ATTR_UID) {
2582                 kuid_t fsuid = from_vfsuid(idmap, i_user_ns(inode), attr->ia_vfsuid);
2583
2584                 doutc(cl, "%p %llx.%llx uid %d -> %d\n", inode,
2585                       ceph_vinop(inode),
2586                       from_kuid(&init_user_ns, inode->i_uid),
2587                       from_kuid(&init_user_ns, attr->ia_uid));
2588                 if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2589                         inode->i_uid = fsuid;
2590                         dirtied |= CEPH_CAP_AUTH_EXCL;
2591                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2592                            !uid_eq(fsuid, inode->i_uid)) {
2593                         req->r_args.setattr.uid = cpu_to_le32(
2594                                 from_kuid(&init_user_ns, fsuid));
2595                         mask |= CEPH_SETATTR_UID;
2596                         release |= CEPH_CAP_AUTH_SHARED;
2597                 }
2598         }
2599         if (ia_valid & ATTR_GID) {
2600                 kgid_t fsgid = from_vfsgid(idmap, i_user_ns(inode), attr->ia_vfsgid);
2601
2602                 doutc(cl, "%p %llx.%llx gid %d -> %d\n", inode,
2603                       ceph_vinop(inode),
2604                       from_kgid(&init_user_ns, inode->i_gid),
2605                       from_kgid(&init_user_ns, attr->ia_gid));
2606                 if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2607                         inode->i_gid = fsgid;
2608                         dirtied |= CEPH_CAP_AUTH_EXCL;
2609                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2610                            !gid_eq(fsgid, inode->i_gid)) {
2611                         req->r_args.setattr.gid = cpu_to_le32(
2612                                 from_kgid(&init_user_ns, fsgid));
2613                         mask |= CEPH_SETATTR_GID;
2614                         release |= CEPH_CAP_AUTH_SHARED;
2615                 }
2616         }
2617         if (ia_valid & ATTR_MODE) {
2618                 doutc(cl, "%p %llx.%llx mode 0%o -> 0%o\n", inode,
2619                       ceph_vinop(inode), inode->i_mode, attr->ia_mode);
2620                 if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2621                         inode->i_mode = attr->ia_mode;
2622                         dirtied |= CEPH_CAP_AUTH_EXCL;
2623                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2624                            attr->ia_mode != inode->i_mode) {
2625                         inode->i_mode = attr->ia_mode;
2626                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2627                         mask |= CEPH_SETATTR_MODE;
2628                         release |= CEPH_CAP_AUTH_SHARED;
2629                 }
2630         }
2631
2632         if (ia_valid & ATTR_ATIME) {
2633                 struct timespec64 atime = inode_get_atime(inode);
2634
2635                 doutc(cl, "%p %llx.%llx atime %lld.%09ld -> %lld.%09ld\n",
2636                       inode, ceph_vinop(inode),
2637                       atime.tv_sec, atime.tv_nsec,
2638                       attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2639                 if (!do_sync && (issued & CEPH_CAP_FILE_EXCL)) {
2640                         ci->i_time_warp_seq++;
2641                         inode_set_atime_to_ts(inode, attr->ia_atime);
2642                         dirtied |= CEPH_CAP_FILE_EXCL;
2643                 } else if (!do_sync && (issued & CEPH_CAP_FILE_WR) &&
2644                            timespec64_compare(&atime,
2645                                               &attr->ia_atime) < 0) {
2646                         inode_set_atime_to_ts(inode, attr->ia_atime);
2647                         dirtied |= CEPH_CAP_FILE_WR;
2648                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2649                            !timespec64_equal(&atime, &attr->ia_atime)) {
2650                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2651                                                &attr->ia_atime);
2652                         mask |= CEPH_SETATTR_ATIME;
2653                         release |= CEPH_CAP_FILE_SHARED |
2654                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2655                 }
2656         }
2657         if (ia_valid & ATTR_SIZE) {
2658                 doutc(cl, "%p %llx.%llx size %lld -> %lld\n", inode,
2659                       ceph_vinop(inode), isize, attr->ia_size);
2660                 /*
2661                  * Only when the new size is smaller and not aligned to
2662                  * CEPH_FSCRYPT_BLOCK_SIZE will the RMW is needed.
2663                  */
2664                 if (IS_ENCRYPTED(inode) && attr->ia_size < isize &&
2665                     (attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE)) {
2666                         mask |= CEPH_SETATTR_SIZE;
2667                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2668                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2669                         set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2670                         mask |= CEPH_SETATTR_FSCRYPT_FILE;
2671                         req->r_args.setattr.size =
2672                                 cpu_to_le64(round_up(attr->ia_size,
2673                                                      CEPH_FSCRYPT_BLOCK_SIZE));
2674                         req->r_args.setattr.old_size =
2675                                 cpu_to_le64(round_up(isize,
2676                                                      CEPH_FSCRYPT_BLOCK_SIZE));
2677                         req->r_fscrypt_file = attr->ia_size;
2678                         fill_fscrypt = true;
2679                 } else if (!do_sync && (issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
2680                         if (attr->ia_size > isize) {
2681                                 i_size_write(inode, attr->ia_size);
2682                                 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2683                                 ci->i_reported_size = attr->ia_size;
2684                                 dirtied |= CEPH_CAP_FILE_EXCL;
2685                                 ia_valid |= ATTR_MTIME;
2686                         }
2687                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2688                            attr->ia_size != isize) {
2689                         mask |= CEPH_SETATTR_SIZE;
2690                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2691                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2692                         if (IS_ENCRYPTED(inode) && attr->ia_size) {
2693                                 set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2694                                 mask |= CEPH_SETATTR_FSCRYPT_FILE;
2695                                 req->r_args.setattr.size =
2696                                         cpu_to_le64(round_up(attr->ia_size,
2697                                                              CEPH_FSCRYPT_BLOCK_SIZE));
2698                                 req->r_args.setattr.old_size =
2699                                         cpu_to_le64(round_up(isize,
2700                                                              CEPH_FSCRYPT_BLOCK_SIZE));
2701                                 req->r_fscrypt_file = attr->ia_size;
2702                         } else {
2703                                 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2704                                 req->r_args.setattr.old_size = cpu_to_le64(isize);
2705                                 req->r_fscrypt_file = 0;
2706                         }
2707                 }
2708         }
2709         if (ia_valid & ATTR_MTIME) {
2710                 struct timespec64 mtime = inode_get_mtime(inode);
2711
2712                 doutc(cl, "%p %llx.%llx mtime %lld.%09ld -> %lld.%09ld\n",
2713                       inode, ceph_vinop(inode),
2714                       mtime.tv_sec, mtime.tv_nsec,
2715                       attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2716                 if (!do_sync && (issued & CEPH_CAP_FILE_EXCL)) {
2717                         ci->i_time_warp_seq++;
2718                         inode_set_mtime_to_ts(inode, attr->ia_mtime);
2719                         dirtied |= CEPH_CAP_FILE_EXCL;
2720                 } else if (!do_sync && (issued & CEPH_CAP_FILE_WR) &&
2721                            timespec64_compare(&mtime, &attr->ia_mtime) < 0) {
2722                         inode_set_mtime_to_ts(inode, attr->ia_mtime);
2723                         dirtied |= CEPH_CAP_FILE_WR;
2724                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2725                            !timespec64_equal(&mtime, &attr->ia_mtime)) {
2726                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2727                                                &attr->ia_mtime);
2728                         mask |= CEPH_SETATTR_MTIME;
2729                         release |= CEPH_CAP_FILE_SHARED |
2730                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2731                 }
2732         }
2733
2734         /* these do nothing */
2735         if (ia_valid & ATTR_CTIME) {
2736                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2737                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2738                 doutc(cl, "%p %llx.%llx ctime %lld.%09ld -> %lld.%09ld (%s)\n",
2739                       inode, ceph_vinop(inode),
2740                       inode_get_ctime_sec(inode),
2741                       inode_get_ctime_nsec(inode),
2742                       attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2743                       only ? "ctime only" : "ignored");
2744                 if (only) {
2745                         /*
2746                          * if kernel wants to dirty ctime but nothing else,
2747                          * we need to choose a cap to dirty under, or do
2748                          * a almost-no-op setattr
2749                          */
2750                         if (issued & CEPH_CAP_AUTH_EXCL)
2751                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2752                         else if (issued & CEPH_CAP_FILE_EXCL)
2753                                 dirtied |= CEPH_CAP_FILE_EXCL;
2754                         else if (issued & CEPH_CAP_XATTR_EXCL)
2755                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2756                         else
2757                                 mask |= CEPH_SETATTR_CTIME;
2758                 }
2759         }
2760         if (ia_valid & ATTR_FILE)
2761                 doutc(cl, "%p %llx.%llx ATTR_FILE ... hrm!\n", inode,
2762                       ceph_vinop(inode));
2763
2764         if (dirtied) {
2765                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2766                                                            &prealloc_cf);
2767                 inode_set_ctime_to_ts(inode, attr->ia_ctime);
2768                 inode_inc_iversion_raw(inode);
2769         }
2770
2771         release &= issued;
2772         spin_unlock(&ci->i_ceph_lock);
2773         if (lock_snap_rwsem) {
2774                 up_read(&mdsc->snap_rwsem);
2775                 lock_snap_rwsem = false;
2776         }
2777
2778         if (inode_dirty_flags)
2779                 __mark_inode_dirty(inode, inode_dirty_flags);
2780
2781         if (mask) {
2782                 req->r_inode = inode;
2783                 ihold(inode);
2784                 req->r_inode_drop = release;
2785                 req->r_args.setattr.mask = cpu_to_le32(mask);
2786                 req->r_num_caps = 1;
2787                 req->r_stamp = attr->ia_ctime;
2788                 if (fill_fscrypt) {
2789                         err = fill_fscrypt_truncate(inode, req, attr);
2790                         if (err)
2791                                 goto out;
2792                 }
2793
2794                 /*
2795                  * The truncate request will return -EAGAIN when the
2796                  * last block has been updated just before the MDS
2797                  * successfully gets the xlock for the FILE lock. To
2798                  * avoid corrupting the file contents we need to retry
2799                  * it.
2800                  */
2801                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2802                 if (err == -EAGAIN && truncate_retry--) {
2803                         doutc(cl, "%p %llx.%llx result=%d (%s locally, %d remote), retry it!\n",
2804                               inode, ceph_vinop(inode), err,
2805                               ceph_cap_string(dirtied), mask);
2806                         ceph_mdsc_put_request(req);
2807                         ceph_free_cap_flush(prealloc_cf);
2808                         goto retry;
2809                 }
2810         }
2811 out:
2812         doutc(cl, "%p %llx.%llx result=%d (%s locally, %d remote)\n", inode,
2813               ceph_vinop(inode), err, ceph_cap_string(dirtied), mask);
2814
2815         ceph_mdsc_put_request(req);
2816         ceph_free_cap_flush(prealloc_cf);
2817
2818         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2819                 __ceph_do_pending_vmtruncate(inode);
2820
2821         return err;
2822 }
2823
2824 /*
2825  * setattr
2826  */
2827 int ceph_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
2828                  struct iattr *attr)
2829 {
2830         struct inode *inode = d_inode(dentry);
2831         struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2832         int err;
2833
2834         if (ceph_snap(inode) != CEPH_NOSNAP)
2835                 return -EROFS;
2836
2837         if (ceph_inode_is_shutdown(inode))
2838                 return -ESTALE;
2839
2840         err = fscrypt_prepare_setattr(dentry, attr);
2841         if (err)
2842                 return err;
2843
2844         err = setattr_prepare(idmap, dentry, attr);
2845         if (err != 0)
2846                 return err;
2847
2848         if ((attr->ia_valid & ATTR_SIZE) &&
2849             attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2850                 return -EFBIG;
2851
2852         if ((attr->ia_valid & ATTR_SIZE) &&
2853             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2854                 return -EDQUOT;
2855
2856         err = __ceph_setattr(idmap, inode, attr, NULL);
2857
2858         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2859                 err = posix_acl_chmod(idmap, dentry, attr->ia_mode);
2860
2861         return err;
2862 }
2863
2864 int ceph_try_to_choose_auth_mds(struct inode *inode, int mask)
2865 {
2866         int issued = ceph_caps_issued(ceph_inode(inode));
2867
2868         /*
2869          * If any 'x' caps is issued we can just choose the auth MDS
2870          * instead of the random replica MDSes. Because only when the
2871          * Locker is in LOCK_EXEC state will the loner client could
2872          * get the 'x' caps. And if we send the getattr requests to
2873          * any replica MDS it must auth pin and tries to rdlock from
2874          * the auth MDS, and then the auth MDS need to do the Locker
2875          * state transition to LOCK_SYNC. And after that the lock state
2876          * will change back.
2877          *
2878          * This cost much when doing the Locker state transition and
2879          * usually will need to revoke caps from clients.
2880          *
2881          * And for the 'Xs' caps for getxattr we will also choose the
2882          * auth MDS, because the MDS side code is buggy due to setxattr
2883          * won't notify the replica MDSes when the values changed and
2884          * the replica MDS will return the old values. Though we will
2885          * fix it in MDS code, but this still makes sense for old ceph.
2886          */
2887         if (((mask & CEPH_CAP_ANY_SHARED) && (issued & CEPH_CAP_ANY_EXCL))
2888             || (mask & (CEPH_STAT_RSTAT | CEPH_STAT_CAP_XATTR)))
2889                 return USE_AUTH_MDS;
2890         else
2891                 return USE_ANY_MDS;
2892 }
2893
2894 /*
2895  * Verify that we have a lease on the given mask.  If not,
2896  * do a getattr against an mds.
2897  */
2898 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2899                       int mask, bool force)
2900 {
2901         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
2902         struct ceph_client *cl = fsc->client;
2903         struct ceph_mds_client *mdsc = fsc->mdsc;
2904         struct ceph_mds_request *req;
2905         int mode;
2906         int err;
2907
2908         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2909                 doutc(cl, "inode %p %llx.%llx SNAPDIR\n", inode,
2910                       ceph_vinop(inode));
2911                 return 0;
2912         }
2913
2914         doutc(cl, "inode %p %llx.%llx mask %s mode 0%o\n", inode,
2915               ceph_vinop(inode), ceph_cap_string(mask), inode->i_mode);
2916         if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2917                         return 0;
2918
2919         mode = ceph_try_to_choose_auth_mds(inode, mask);
2920         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2921         if (IS_ERR(req))
2922                 return PTR_ERR(req);
2923         req->r_inode = inode;
2924         ihold(inode);
2925         req->r_num_caps = 1;
2926         req->r_args.getattr.mask = cpu_to_le32(mask);
2927         req->r_locked_page = locked_page;
2928         err = ceph_mdsc_do_request(mdsc, NULL, req);
2929         if (locked_page && err == 0) {
2930                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2931                 if (inline_version == 0) {
2932                         /* the reply is supposed to contain inline data */
2933                         err = -EINVAL;
2934                 } else if (inline_version == CEPH_INLINE_NONE ||
2935                            inline_version == 1) {
2936                         err = -ENODATA;
2937                 } else {
2938                         err = req->r_reply_info.targeti.inline_len;
2939                 }
2940         }
2941         ceph_mdsc_put_request(req);
2942         doutc(cl, "result=%d\n", err);
2943         return err;
2944 }
2945
2946 int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
2947                       size_t size)
2948 {
2949         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
2950         struct ceph_client *cl = fsc->client;
2951         struct ceph_mds_client *mdsc = fsc->mdsc;
2952         struct ceph_mds_request *req;
2953         int mode = USE_AUTH_MDS;
2954         int err;
2955         char *xattr_value;
2956         size_t xattr_value_len;
2957
2958         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode);
2959         if (IS_ERR(req)) {
2960                 err = -ENOMEM;
2961                 goto out;
2962         }
2963
2964         req->r_feature_needed = CEPHFS_FEATURE_OP_GETVXATTR;
2965         req->r_path2 = kstrdup(name, GFP_NOFS);
2966         if (!req->r_path2) {
2967                 err = -ENOMEM;
2968                 goto put;
2969         }
2970
2971         ihold(inode);
2972         req->r_inode = inode;
2973         err = ceph_mdsc_do_request(mdsc, NULL, req);
2974         if (err < 0)
2975                 goto put;
2976
2977         xattr_value = req->r_reply_info.xattr_info.xattr_value;
2978         xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len;
2979
2980         doutc(cl, "xattr_value_len:%zu, size:%zu\n", xattr_value_len, size);
2981
2982         err = (int)xattr_value_len;
2983         if (size == 0)
2984                 goto put;
2985
2986         if (xattr_value_len > size) {
2987                 err = -ERANGE;
2988                 goto put;
2989         }
2990
2991         memcpy(value, xattr_value, xattr_value_len);
2992 put:
2993         ceph_mdsc_put_request(req);
2994 out:
2995         doutc(cl, "result=%d\n", err);
2996         return err;
2997 }
2998
2999
3000 /*
3001  * Check inode permissions.  We verify we have a valid value for
3002  * the AUTH cap, then call the generic handler.
3003  */
3004 int ceph_permission(struct mnt_idmap *idmap, struct inode *inode,
3005                     int mask)
3006 {
3007         int err;
3008
3009         if (mask & MAY_NOT_BLOCK)
3010                 return -ECHILD;
3011
3012         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
3013
3014         if (!err)
3015                 err = generic_permission(idmap, inode, mask);
3016         return err;
3017 }
3018
3019 /* Craft a mask of needed caps given a set of requested statx attrs. */
3020 static int statx_to_caps(u32 want, umode_t mode)
3021 {
3022         int mask = 0;
3023
3024         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME|STATX_CHANGE_COOKIE))
3025                 mask |= CEPH_CAP_AUTH_SHARED;
3026
3027         if (want & (STATX_NLINK|STATX_CTIME|STATX_CHANGE_COOKIE)) {
3028                 /*
3029                  * The link count for directories depends on inode->i_subdirs,
3030                  * and that is only updated when Fs caps are held.
3031                  */
3032                 if (S_ISDIR(mode))
3033                         mask |= CEPH_CAP_FILE_SHARED;
3034                 else
3035                         mask |= CEPH_CAP_LINK_SHARED;
3036         }
3037
3038         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|STATX_BLOCKS|STATX_CHANGE_COOKIE))
3039                 mask |= CEPH_CAP_FILE_SHARED;
3040
3041         if (want & (STATX_CTIME|STATX_CHANGE_COOKIE))
3042                 mask |= CEPH_CAP_XATTR_SHARED;
3043
3044         return mask;
3045 }
3046
3047 /*
3048  * Get all the attributes. If we have sufficient caps for the requested attrs,
3049  * then we can avoid talking to the MDS at all.
3050  */
3051 int ceph_getattr(struct mnt_idmap *idmap, const struct path *path,
3052                  struct kstat *stat, u32 request_mask, unsigned int flags)
3053 {
3054         struct inode *inode = d_inode(path->dentry);
3055         struct super_block *sb = inode->i_sb;
3056         struct ceph_inode_info *ci = ceph_inode(inode);
3057         u32 valid_mask = STATX_BASIC_STATS;
3058         int err = 0;
3059
3060         if (ceph_inode_is_shutdown(inode))
3061                 return -ESTALE;
3062
3063         /* Skip the getattr altogether if we're asked not to sync */
3064         if ((flags & AT_STATX_SYNC_TYPE) != AT_STATX_DONT_SYNC) {
3065                 err = ceph_do_getattr(inode,
3066                                 statx_to_caps(request_mask, inode->i_mode),
3067                                 flags & AT_STATX_FORCE_SYNC);
3068                 if (err)
3069                         return err;
3070         }
3071
3072         generic_fillattr(idmap, request_mask, inode, stat);
3073         stat->ino = ceph_present_inode(inode);
3074
3075         /*
3076          * btime on newly-allocated inodes is 0, so if this is still set to
3077          * that, then assume that it's not valid.
3078          */
3079         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
3080                 stat->btime = ci->i_btime;
3081                 valid_mask |= STATX_BTIME;
3082         }
3083
3084         if (request_mask & STATX_CHANGE_COOKIE) {
3085                 stat->change_cookie = inode_peek_iversion_raw(inode);
3086                 valid_mask |= STATX_CHANGE_COOKIE;
3087         }
3088
3089         if (ceph_snap(inode) == CEPH_NOSNAP)
3090                 stat->dev = sb->s_dev;
3091         else
3092                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
3093
3094         if (S_ISDIR(inode->i_mode)) {
3095                 if (ceph_test_mount_opt(ceph_sb_to_fs_client(sb), RBYTES)) {
3096                         stat->size = ci->i_rbytes;
3097                 } else if (ceph_snap(inode) == CEPH_SNAPDIR) {
3098                         struct ceph_inode_info *pci;
3099                         struct ceph_snap_realm *realm;
3100                         struct inode *parent;
3101
3102                         parent = ceph_lookup_inode(sb, ceph_ino(inode));
3103                         if (IS_ERR(parent))
3104                                 return PTR_ERR(parent);
3105
3106                         pci = ceph_inode(parent);
3107                         spin_lock(&pci->i_ceph_lock);
3108                         realm = pci->i_snap_realm;
3109                         if (realm)
3110                                 stat->size = realm->num_snaps;
3111                         else
3112                                 stat->size = 0;
3113                         spin_unlock(&pci->i_ceph_lock);
3114                         iput(parent);
3115                 } else {
3116                         stat->size = ci->i_files + ci->i_subdirs;
3117                 }
3118                 stat->blocks = 0;
3119                 stat->blksize = 65536;
3120                 /*
3121                  * Some applications rely on the number of st_nlink
3122                  * value on directories to be either 0 (if unlinked)
3123                  * or 2 + number of subdirectories.
3124                  */
3125                 if (stat->nlink == 1)
3126                         /* '.' + '..' + subdirs */
3127                         stat->nlink = 1 + 1 + ci->i_subdirs;
3128         }
3129
3130         stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
3131         if (IS_ENCRYPTED(inode))
3132                 stat->attributes |= STATX_ATTR_ENCRYPTED;
3133         stat->attributes_mask |= (STATX_ATTR_CHANGE_MONOTONIC |
3134                                   STATX_ATTR_ENCRYPTED);
3135
3136         stat->result_mask = request_mask & valid_mask;
3137         return err;
3138 }
3139
3140 void ceph_inode_shutdown(struct inode *inode)
3141 {
3142         struct ceph_inode_info *ci = ceph_inode(inode);
3143         struct rb_node *p;
3144         int iputs = 0;
3145         bool invalidate = false;
3146
3147         spin_lock(&ci->i_ceph_lock);
3148         ci->i_ceph_flags |= CEPH_I_SHUTDOWN;
3149         p = rb_first(&ci->i_caps);
3150         while (p) {
3151                 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
3152
3153                 p = rb_next(p);
3154                 iputs += ceph_purge_inode_cap(inode, cap, &invalidate);
3155         }
3156         spin_unlock(&ci->i_ceph_lock);
3157
3158         if (invalidate)
3159                 ceph_queue_invalidate(inode);
3160         while (iputs--)
3161                 iput(inode);
3162 }