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
  37static const struct inode_operations ceph_symlink_iops;
  38static const struct inode_operations ceph_encrypted_symlink_iops;
  39
  40static void ceph_inode_work(struct work_struct *work);
  41
  42/*
  43 * find or create an inode, given the ceph ino number
  44 */
  45static 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 * Check if the parent inode matches the vino from directory reply info
  60 */
  61static inline bool ceph_vino_matches_parent(struct inode *parent,
  62					    struct ceph_vino vino)
  63{
  64	return ceph_ino(parent) == vino.ino && ceph_snap(parent) == vino.snap;
  65}
  66
  67/*
  68 * Validate that the directory inode referenced by @req->r_parent matches the
  69 * inode number and snapshot id contained in the reply's directory record.  If
  70 * they do not match – which can theoretically happen if the parent dentry was
  71 * moved between the time the request was issued and the reply arrived – fall
  72 * back to looking up the correct inode in the inode cache.
  73 *
  74 * A reference is *always* returned.  Callers that receive a different inode
  75 * than the original @parent are responsible for dropping the extra reference
  76 * once the reply has been processed.
  77 */
  78static struct inode *ceph_get_reply_dir(struct super_block *sb,
  79					struct inode *parent,
  80					struct ceph_mds_reply_info_parsed *rinfo)
  81{
  82	struct ceph_vino vino;
  83
  84	if (unlikely(!rinfo->diri.in))
  85		return parent; /* nothing to compare against */
  86
  87	/* If we didn't have a cached parent inode to begin with, just bail out. */
  88	if (!parent)
  89		return NULL;
  90
  91	vino.ino  = le64_to_cpu(rinfo->diri.in->ino);
  92	vino.snap = le64_to_cpu(rinfo->diri.in->snapid);
  93
  94	if (likely(ceph_vino_matches_parent(parent, vino)))
  95		return parent; /* matches – use the original reference */
  96
  97	/* Mismatch – this should be rare.  Emit a WARN and obtain the correct inode. */
  98	WARN_ONCE(1, "ceph: reply dir mismatch (parent valid %llx.%llx reply %llx.%llx)\n",
  99		  ceph_ino(parent), ceph_snap(parent), vino.ino, vino.snap);
 100
 101	return ceph_get_inode(sb, vino, NULL);
 102}
 103
 104/**
 105 * ceph_new_inode - allocate a new inode in advance of an expected create
 106 * @dir: parent directory for new inode
 107 * @dentry: dentry that may eventually point to new inode
 108 * @mode: mode of new inode
 109 * @as_ctx: pointer to inherited security context
 110 *
 111 * Allocate a new inode in advance of an operation to create a new inode.
 112 * This allocates the inode and sets up the acl_sec_ctx with appropriate
 113 * info for the new inode.
 114 *
 115 * Returns a pointer to the new inode or an ERR_PTR.
 116 */
 117struct inode *ceph_new_inode(struct inode *dir, struct dentry *dentry,
 118			     umode_t *mode, struct ceph_acl_sec_ctx *as_ctx)
 119{
 120	int err;
 121	struct inode *inode;
 122
 123	inode = new_inode(dir->i_sb);
 124	if (!inode)
 125		return ERR_PTR(-ENOMEM);
 126
 127	inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
 128
 129	if (!S_ISLNK(*mode)) {
 130		err = ceph_pre_init_acls(dir, mode, as_ctx);
 131		if (err < 0)
 132			goto out_err;
 133	}
 134
 135	inode_state_assign_raw(inode, 0);
 136	inode->i_mode = *mode;
 137
 138	err = ceph_security_init_secctx(dentry, *mode, as_ctx);
 139	if (err < 0)
 140		goto out_err;
 141
 142	/*
 143	 * We'll skip setting fscrypt context for snapshots, leaving that for
 144	 * the handle_reply().
 145	 */
 146	if (ceph_snap(dir) != CEPH_SNAPDIR) {
 147		err = ceph_fscrypt_prepare_context(dir, inode, as_ctx);
 148		if (err)
 149			goto out_err;
 150	}
 151
 152	return inode;
 153out_err:
 154	iput(inode);
 155	return ERR_PTR(err);
 156}
 157
 158void ceph_as_ctx_to_req(struct ceph_mds_request *req,
 159			struct ceph_acl_sec_ctx *as_ctx)
 160{
 161	if (as_ctx->pagelist) {
 162		req->r_pagelist = as_ctx->pagelist;
 163		as_ctx->pagelist = NULL;
 164	}
 165	ceph_fscrypt_as_ctx_to_req(req, as_ctx);
 166}
 167
 168/**
 169 * ceph_get_inode - find or create/hash a new inode
 170 * @sb: superblock to search and allocate in
 171 * @vino: vino to search for
 172 * @newino: optional new inode to insert if one isn't found (may be NULL)
 173 *
 174 * Search for or insert a new inode into the hash for the given vino, and
 175 * return a reference to it. If new is non-NULL, its reference is consumed.
 176 */
 177struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino,
 178			     struct inode *newino)
 179{
 180	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(sb);
 181	struct ceph_client *cl = mdsc->fsc->client;
 182	struct inode *inode;
 183
 184	if (ceph_vino_is_reserved(vino))
 185		return ERR_PTR(-EREMOTEIO);
 186
 187	if (newino) {
 188		inode = inode_insert5(newino, (unsigned long)vino.ino,
 189				      ceph_ino_compare, ceph_set_ino_cb, &vino);
 190		if (inode != newino)
 191			iput(newino);
 192	} else {
 193		inode = iget5_locked(sb, (unsigned long)vino.ino,
 194				     ceph_ino_compare, ceph_set_ino_cb, &vino);
 195	}
 196
 197	if (!inode) {
 198		doutc(cl, "no inode found for %llx.%llx\n", vino.ino, vino.snap);
 199		return ERR_PTR(-ENOMEM);
 200	}
 201
 202	doutc(cl, "on %llx=%llx.%llx got %p new %d\n",
 203	      ceph_present_inode(inode), ceph_vinop(inode), inode,
 204	      !!(inode_state_read_once(inode) & I_NEW));
 205	return inode;
 206}
 207
 208/*
 209 * get/construct snapdir inode for a given directory
 210 */
 211struct inode *ceph_get_snapdir(struct inode *parent)
 212{
 213	struct ceph_client *cl = ceph_inode_to_client(parent);
 214	struct ceph_vino vino = {
 215		.ino = ceph_ino(parent),
 216		.snap = CEPH_SNAPDIR,
 217	};
 218	struct inode *inode = ceph_get_inode(parent->i_sb, vino, NULL);
 219	struct ceph_inode_info *ci = ceph_inode(inode);
 220	int ret = -ENOTDIR;
 221
 222	if (IS_ERR(inode))
 223		return inode;
 224
 225	if (!S_ISDIR(parent->i_mode)) {
 226		pr_warn_once_client(cl, "bad snapdir parent type (mode=0%o)\n",
 227				    parent->i_mode);
 228		goto err;
 229	}
 230
 231	if (!(inode_state_read_once(inode) & I_NEW) && !S_ISDIR(inode->i_mode)) {
 232		pr_warn_once_client(cl, "bad snapdir inode type (mode=0%o)\n",
 233				    inode->i_mode);
 234		goto err;
 235	}
 236
 237	inode->i_mode = parent->i_mode;
 238	inode->i_uid = parent->i_uid;
 239	inode->i_gid = parent->i_gid;
 240	inode_set_mtime_to_ts(inode, inode_get_mtime(parent));
 241	inode_set_ctime_to_ts(inode, inode_get_ctime(parent));
 242	inode_set_atime_to_ts(inode, inode_get_atime(parent));
 243	ci->i_rbytes = 0;
 244	ci->i_btime = ceph_inode(parent)->i_btime;
 245
 246#ifdef CONFIG_FS_ENCRYPTION
 247	/* if encrypted, just borrow fscrypt_auth from parent */
 248	if (IS_ENCRYPTED(parent)) {
 249		struct ceph_inode_info *pci = ceph_inode(parent);
 250
 251		ci->fscrypt_auth = kmemdup(pci->fscrypt_auth,
 252					   pci->fscrypt_auth_len,
 253					   GFP_KERNEL);
 254		if (ci->fscrypt_auth) {
 255			inode->i_flags |= S_ENCRYPTED;
 256			ci->fscrypt_auth_len = pci->fscrypt_auth_len;
 257		} else {
 258			doutc(cl, "Failed to alloc snapdir fscrypt_auth\n");
 259			ret = -ENOMEM;
 260			goto err;
 261		}
 262	}
 263#endif
 264	if (inode_state_read_once(inode) & I_NEW) {
 265		inode->i_op = &ceph_snapdir_iops;
 266		inode->i_fop = &ceph_snapdir_fops;
 267		ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
 268		unlock_new_inode(inode);
 269	}
 270
 271	return inode;
 272err:
 273	if ((inode_state_read_once(inode) & I_NEW))
 274		discard_new_inode(inode);
 275	else
 276		iput(inode);
 277	return ERR_PTR(ret);
 278}
 279
 280const struct inode_operations ceph_file_iops = {
 281	.permission = ceph_permission,
 282	.setattr = ceph_setattr,
 283	.getattr = ceph_getattr,
 284	.listxattr = ceph_listxattr,
 285	.get_inode_acl = ceph_get_acl,
 286	.set_acl = ceph_set_acl,
 287};
 288
 289
 290/*
 291 * We use a 'frag tree' to keep track of the MDS's directory fragments
 292 * for a given inode (usually there is just a single fragment).  We
 293 * need to know when a child frag is delegated to a new MDS, or when
 294 * it is flagged as replicated, so we can direct our requests
 295 * accordingly.
 296 */
 297
 298/*
 299 * find/create a frag in the tree
 300 */
 301static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 302						    u32 f)
 303{
 304	struct inode *inode = &ci->netfs.inode;
 305	struct ceph_client *cl = ceph_inode_to_client(inode);
 306	struct rb_node **p;
 307	struct rb_node *parent = NULL;
 308	struct ceph_inode_frag *frag;
 309	int c;
 310
 311	p = &ci->i_fragtree.rb_node;
 312	while (*p) {
 313		parent = *p;
 314		frag = rb_entry(parent, struct ceph_inode_frag, node);
 315		c = ceph_frag_compare(f, frag->frag);
 316		if (c < 0)
 317			p = &(*p)->rb_left;
 318		else if (c > 0)
 319			p = &(*p)->rb_right;
 320		else
 321			return frag;
 322	}
 323
 324	frag = kmalloc(sizeof(*frag), GFP_NOFS);
 325	if (!frag)
 326		return ERR_PTR(-ENOMEM);
 327
 328	frag->frag = f;
 329	frag->split_by = 0;
 330	frag->mds = -1;
 331	frag->ndist = 0;
 332
 333	rb_link_node(&frag->node, parent, p);
 334	rb_insert_color(&frag->node, &ci->i_fragtree);
 335
 336	doutc(cl, "added %p %llx.%llx frag %x\n", inode, ceph_vinop(inode), f);
 337	return frag;
 338}
 339
 340/*
 341 * find a specific frag @f
 342 */
 343struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 344{
 345	struct rb_node *n = ci->i_fragtree.rb_node;
 346
 347	while (n) {
 348		struct ceph_inode_frag *frag =
 349			rb_entry(n, struct ceph_inode_frag, node);
 350		int c = ceph_frag_compare(f, frag->frag);
 351		if (c < 0)
 352			n = n->rb_left;
 353		else if (c > 0)
 354			n = n->rb_right;
 355		else
 356			return frag;
 357	}
 358	return NULL;
 359}
 360
 361/*
 362 * Choose frag containing the given value @v.  If @pfrag is
 363 * specified, copy the frag delegation info to the caller if
 364 * it is present.
 365 */
 366static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 367			      struct ceph_inode_frag *pfrag, int *found)
 368{
 369	struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
 370	u32 t = ceph_frag_make(0, 0);
 371	struct ceph_inode_frag *frag;
 372	unsigned nway, i;
 373	u32 n;
 374
 375	if (found)
 376		*found = 0;
 377
 378	while (1) {
 379		WARN_ON(!ceph_frag_contains_value(t, v));
 380		frag = __ceph_find_frag(ci, t);
 381		if (!frag)
 382			break; /* t is a leaf */
 383		if (frag->split_by == 0) {
 384			if (pfrag)
 385				memcpy(pfrag, frag, sizeof(*pfrag));
 386			if (found)
 387				*found = 1;
 388			break;
 389		}
 390
 391		/* choose child */
 392		nway = 1 << frag->split_by;
 393		doutc(cl, "frag(%x) %x splits by %d (%d ways)\n", v, t,
 394		      frag->split_by, nway);
 395		for (i = 0; i < nway; i++) {
 396			n = ceph_frag_make_child(t, frag->split_by, i);
 397			if (ceph_frag_contains_value(n, v)) {
 398				t = n;
 399				break;
 400			}
 401		}
 402		BUG_ON(i == nway);
 403	}
 404	doutc(cl, "frag(%x) = %x\n", v, t);
 405
 406	return t;
 407}
 408
 409u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 410		     struct ceph_inode_frag *pfrag, int *found)
 411{
 412	u32 ret;
 413	mutex_lock(&ci->i_fragtree_mutex);
 414	ret = __ceph_choose_frag(ci, v, pfrag, found);
 415	mutex_unlock(&ci->i_fragtree_mutex);
 416	return ret;
 417}
 418
 419/*
 420 * Process dirfrag (delegation) info from the mds.  Include leaf
 421 * fragment in tree ONLY if ndist > 0.  Otherwise, only
 422 * branches/splits are included in i_fragtree)
 423 */
 424static int ceph_fill_dirfrag(struct inode *inode,
 425			     struct ceph_mds_reply_dirfrag *dirinfo)
 426{
 427	struct ceph_inode_info *ci = ceph_inode(inode);
 428	struct ceph_client *cl = ceph_inode_to_client(inode);
 429	struct ceph_inode_frag *frag;
 430	u32 id = le32_to_cpu(dirinfo->frag);
 431	int mds = le32_to_cpu(dirinfo->auth);
 432	int ndist = le32_to_cpu(dirinfo->ndist);
 433	int diri_auth = -1;
 434	int i;
 435	int err = 0;
 436
 437	spin_lock(&ci->i_ceph_lock);
 438	if (ci->i_auth_cap)
 439		diri_auth = ci->i_auth_cap->mds;
 440	spin_unlock(&ci->i_ceph_lock);
 441
 442	if (mds == -1) /* CDIR_AUTH_PARENT */
 443		mds = diri_auth;
 444
 445	mutex_lock(&ci->i_fragtree_mutex);
 446	if (ndist == 0 && mds == diri_auth) {
 447		/* no delegation info needed. */
 448		frag = __ceph_find_frag(ci, id);
 449		if (!frag)
 450			goto out;
 451		if (frag->split_by == 0) {
 452			/* tree leaf, remove */
 453			doutc(cl, "removed %p %llx.%llx frag %x (no ref)\n",
 454			      inode, ceph_vinop(inode), id);
 455			rb_erase(&frag->node, &ci->i_fragtree);
 456			kfree(frag);
 457		} else {
 458			/* tree branch, keep and clear */
 459			doutc(cl, "cleared %p %llx.%llx frag %x referral\n",
 460			      inode, ceph_vinop(inode), id);
 461			frag->mds = -1;
 462			frag->ndist = 0;
 463		}
 464		goto out;
 465	}
 466
 467
 468	/* find/add this frag to store mds delegation info */
 469	frag = __get_or_create_frag(ci, id);
 470	if (IS_ERR(frag)) {
 471		/* this is not the end of the world; we can continue
 472		   with bad/inaccurate delegation info */
 473		pr_err_client(cl, "ENOMEM on mds ref %p %llx.%llx fg %x\n",
 474			      inode, ceph_vinop(inode),
 475			      le32_to_cpu(dirinfo->frag));
 476		err = -ENOMEM;
 477		goto out;
 478	}
 479
 480	frag->mds = mds;
 481	frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 482	for (i = 0; i < frag->ndist; i++)
 483		frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 484	doutc(cl, "%p %llx.%llx frag %x ndist=%d\n", inode,
 485	      ceph_vinop(inode), frag->frag, frag->ndist);
 486
 487out:
 488	mutex_unlock(&ci->i_fragtree_mutex);
 489	return err;
 490}
 491
 492static int frag_tree_split_cmp(const void *l, const void *r)
 493{
 494	struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 495	struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 496	return ceph_frag_compare(le32_to_cpu(ls->frag),
 497				 le32_to_cpu(rs->frag));
 498}
 499
 500static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 501{
 502	if (!frag)
 503		return f == ceph_frag_make(0, 0);
 504	if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 505		return false;
 506	return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 507}
 508
 509static int ceph_fill_fragtree(struct inode *inode,
 510			      struct ceph_frag_tree_head *fragtree,
 511			      struct ceph_mds_reply_dirfrag *dirinfo)
 512{
 513	struct ceph_client *cl = ceph_inode_to_client(inode);
 514	struct ceph_inode_info *ci = ceph_inode(inode);
 515	struct ceph_inode_frag *frag, *prev_frag = NULL;
 516	struct rb_node *rb_node;
 517	unsigned i, split_by, nsplits;
 518	u32 id;
 519	bool update = false;
 520
 521	mutex_lock(&ci->i_fragtree_mutex);
 522	nsplits = le32_to_cpu(fragtree->nsplits);
 523	if (nsplits != ci->i_fragtree_nsplits) {
 524		update = true;
 525	} else if (nsplits) {
 526		i = get_random_u32_below(nsplits);
 527		id = le32_to_cpu(fragtree->splits[i].frag);
 528		if (!__ceph_find_frag(ci, id))
 529			update = true;
 530	} else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 531		rb_node = rb_first(&ci->i_fragtree);
 532		frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 533		if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 534			update = true;
 535	}
 536	if (!update && dirinfo) {
 537		id = le32_to_cpu(dirinfo->frag);
 538		if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 539			update = true;
 540	}
 541	if (!update)
 542		goto out_unlock;
 543
 544	if (nsplits > 1) {
 545		sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 546		     frag_tree_split_cmp, NULL);
 547	}
 548
 549	doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
 550	rb_node = rb_first(&ci->i_fragtree);
 551	for (i = 0; i < nsplits; i++) {
 552		id = le32_to_cpu(fragtree->splits[i].frag);
 553		split_by = le32_to_cpu(fragtree->splits[i].by);
 554		if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 555			pr_err_client(cl, "%p %llx.%llx invalid split %d/%u, "
 556			       "frag %x split by %d\n", inode,
 557			       ceph_vinop(inode), i, nsplits, id, split_by);
 558			continue;
 559		}
 560		frag = NULL;
 561		while (rb_node) {
 562			frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 563			if (ceph_frag_compare(frag->frag, id) >= 0) {
 564				if (frag->frag != id)
 565					frag = NULL;
 566				else
 567					rb_node = rb_next(rb_node);
 568				break;
 569			}
 570			rb_node = rb_next(rb_node);
 571			/* delete stale split/leaf node */
 572			if (frag->split_by > 0 ||
 573			    !is_frag_child(frag->frag, prev_frag)) {
 574				rb_erase(&frag->node, &ci->i_fragtree);
 575				if (frag->split_by > 0)
 576					ci->i_fragtree_nsplits--;
 577				kfree(frag);
 578			}
 579			frag = NULL;
 580		}
 581		if (!frag) {
 582			frag = __get_or_create_frag(ci, id);
 583			if (IS_ERR(frag))
 584				continue;
 585		}
 586		if (frag->split_by == 0)
 587			ci->i_fragtree_nsplits++;
 588		frag->split_by = split_by;
 589		doutc(cl, " frag %x split by %d\n", frag->frag, frag->split_by);
 590		prev_frag = frag;
 591	}
 592	while (rb_node) {
 593		frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 594		rb_node = rb_next(rb_node);
 595		/* delete stale split/leaf node */
 596		if (frag->split_by > 0 ||
 597		    !is_frag_child(frag->frag, prev_frag)) {
 598			rb_erase(&frag->node, &ci->i_fragtree);
 599			if (frag->split_by > 0)
 600				ci->i_fragtree_nsplits--;
 601			kfree(frag);
 602		}
 603	}
 604out_unlock:
 605	mutex_unlock(&ci->i_fragtree_mutex);
 606	return 0;
 607}
 608
 609/*
 610 * initialize a newly allocated inode.
 611 */
 612struct inode *ceph_alloc_inode(struct super_block *sb)
 613{
 614	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
 615	struct ceph_inode_info *ci;
 616	int i;
 617
 618	ci = alloc_inode_sb(sb, ceph_inode_cachep, GFP_NOFS);
 619	if (!ci)
 620		return NULL;
 621
 622	doutc(fsc->client, "%p\n", &ci->netfs.inode);
 623
 624	/* Set parameters for the netfs library */
 625	netfs_inode_init(&ci->netfs, &ceph_netfs_ops, false);
 626
 627	spin_lock_init(&ci->i_ceph_lock);
 628
 629	ci->i_version = 0;
 630	ci->i_inline_version = 0;
 631	ci->i_time_warp_seq = 0;
 632	ci->i_ceph_flags = 0;
 633	atomic64_set(&ci->i_ordered_count, 1);
 634	atomic64_set(&ci->i_release_count, 1);
 635	atomic64_set(&ci->i_complete_seq[0], 0);
 636	atomic64_set(&ci->i_complete_seq[1], 0);
 637	ci->i_symlink = NULL;
 638
 639	ci->i_max_bytes = 0;
 640	ci->i_max_files = 0;
 641
 642	memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 643	memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
 644	RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 645
 646	ci->i_fragtree = RB_ROOT;
 647	mutex_init(&ci->i_fragtree_mutex);
 648
 649	ci->i_xattrs.blob = NULL;
 650	ci->i_xattrs.prealloc_blob = NULL;
 651	ci->i_xattrs.dirty = false;
 652	ci->i_xattrs.index = RB_ROOT;
 653	ci->i_xattrs.count = 0;
 654	ci->i_xattrs.names_size = 0;
 655	ci->i_xattrs.vals_size = 0;
 656	ci->i_xattrs.version = 0;
 657	ci->i_xattrs.index_version = 0;
 658
 659	ci->i_caps = RB_ROOT;
 660	ci->i_auth_cap = NULL;
 661	ci->i_dirty_caps = 0;
 662	ci->i_flushing_caps = 0;
 663	INIT_LIST_HEAD(&ci->i_dirty_item);
 664	INIT_LIST_HEAD(&ci->i_flushing_item);
 665	ci->i_prealloc_cap_flush = NULL;
 666	INIT_LIST_HEAD(&ci->i_cap_flush_list);
 667	init_waitqueue_head(&ci->i_cap_wq);
 668	ci->i_hold_caps_max = 0;
 669	INIT_LIST_HEAD(&ci->i_cap_delay_list);
 670	INIT_LIST_HEAD(&ci->i_cap_snaps);
 671	ci->i_head_snapc = NULL;
 672	ci->i_snap_caps = 0;
 673
 674	ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
 675	for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 676		ci->i_nr_by_mode[i] = 0;
 677
 678	mutex_init(&ci->i_truncate_mutex);
 679	ci->i_truncate_seq = 0;
 680	ci->i_truncate_size = 0;
 681	ci->i_truncate_pending = 0;
 682	ci->i_truncate_pagecache_size = 0;
 683
 684	ci->i_max_size = 0;
 685	ci->i_reported_size = 0;
 686	ci->i_wanted_max_size = 0;
 687	ci->i_requested_max_size = 0;
 688
 689	ci->i_pin_ref = 0;
 690	ci->i_rd_ref = 0;
 691	ci->i_rdcache_ref = 0;
 692	ci->i_wr_ref = 0;
 693	ci->i_wb_ref = 0;
 694	ci->i_fx_ref = 0;
 695	ci->i_wrbuffer_ref = 0;
 696	ci->i_wrbuffer_ref_head = 0;
 697	atomic_set(&ci->i_filelock_ref, 0);
 698	atomic_set(&ci->i_shared_gen, 1);
 699	ci->i_rdcache_gen = 0;
 700	ci->i_rdcache_revoking = 0;
 701
 702	INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 703	INIT_LIST_HEAD(&ci->i_unsafe_iops);
 704	spin_lock_init(&ci->i_unsafe_lock);
 705
 706	ci->i_snap_realm = NULL;
 707	INIT_LIST_HEAD(&ci->i_snap_realm_item);
 708	INIT_LIST_HEAD(&ci->i_snap_flush_item);
 709
 710	INIT_WORK(&ci->i_work, ceph_inode_work);
 711	ci->i_work_mask = 0;
 712	memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 713#ifdef CONFIG_FS_ENCRYPTION
 714	ci->i_crypt_info = NULL;
 715	ci->fscrypt_auth = NULL;
 716	ci->fscrypt_auth_len = 0;
 717#endif
 718	return &ci->netfs.inode;
 719}
 720
 721void ceph_free_inode(struct inode *inode)
 722{
 723	struct ceph_inode_info *ci = ceph_inode(inode);
 724
 725	kfree(ci->i_symlink);
 726#ifdef CONFIG_FS_ENCRYPTION
 727	kfree(ci->fscrypt_auth);
 728#endif
 729	fscrypt_free_inode(inode);
 730	kmem_cache_free(ceph_inode_cachep, ci);
 731}
 732
 733void ceph_evict_inode(struct inode *inode)
 734{
 735	struct ceph_inode_info *ci = ceph_inode(inode);
 736	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 737	struct ceph_client *cl = ceph_inode_to_client(inode);
 738	struct ceph_inode_frag *frag;
 739	struct rb_node *n;
 740
 741	doutc(cl, "%p ino %llx.%llx\n", inode, ceph_vinop(inode));
 742
 743	percpu_counter_dec(&mdsc->metric.total_inodes);
 744
 745	netfs_wait_for_outstanding_io(inode);
 746	truncate_inode_pages_final(&inode->i_data);
 747	if (inode_state_read_once(inode) & I_PINNING_NETFS_WB)
 748		ceph_fscache_unuse_cookie(inode, true);
 749	clear_inode(inode);
 750
 751	ceph_fscache_unregister_inode_cookie(ci);
 752	fscrypt_put_encryption_info(inode);
 753
 754	__ceph_remove_caps(ci);
 755
 756	if (__ceph_has_quota(ci, QUOTA_GET_ANY))
 757		ceph_adjust_quota_realms_count(inode, false);
 758
 759	/*
 760	 * we may still have a snap_realm reference if there are stray
 761	 * caps in i_snap_caps.
 762	 */
 763	if (ci->i_snap_realm) {
 764		if (ceph_snap(inode) == CEPH_NOSNAP) {
 765			doutc(cl, " dropping residual ref to snap realm %p\n",
 766			      ci->i_snap_realm);
 767			ceph_change_snap_realm(inode, NULL);
 768		} else {
 769			ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 770			ci->i_snap_realm = NULL;
 771		}
 772	}
 773
 774	while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 775		frag = rb_entry(n, struct ceph_inode_frag, node);
 776		rb_erase(n, &ci->i_fragtree);
 777		kfree(frag);
 778	}
 779	ci->i_fragtree_nsplits = 0;
 780
 781	__ceph_destroy_xattrs(ci);
 782	if (ci->i_xattrs.blob)
 783		ceph_buffer_put(ci->i_xattrs.blob);
 784	if (ci->i_xattrs.prealloc_blob)
 785		ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 786
 787	ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 788	ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
 789}
 790
 791static inline blkcnt_t calc_inode_blocks(u64 size)
 792{
 793	return (size + (1<<9) - 1) >> 9;
 794}
 795
 796/*
 797 * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 798 * careful because either the client or MDS may have more up to date
 799 * info, depending on which capabilities are held, and whether
 800 * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 801 * and size are monotonically increasing, except when utimes() or
 802 * truncate() increments the corresponding _seq values.)
 803 */
 804int ceph_fill_file_size(struct inode *inode, int issued,
 805			u32 truncate_seq, u64 truncate_size, u64 size)
 806{
 807	struct ceph_client *cl = ceph_inode_to_client(inode);
 808	struct ceph_inode_info *ci = ceph_inode(inode);
 809	int queue_trunc = 0;
 810	loff_t isize = i_size_read(inode);
 811
 812	if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 813	    (truncate_seq == ci->i_truncate_seq && size > isize)) {
 814		doutc(cl, "size %lld -> %llu\n", isize, size);
 815		if (size > 0 && S_ISDIR(inode->i_mode)) {
 816			pr_err_client(cl, "non-zero size for directory\n");
 817			size = 0;
 818		}
 819		i_size_write(inode, size);
 820		inode->i_blocks = calc_inode_blocks(size);
 821		/*
 822		 * If we're expanding, then we should be able to just update
 823		 * the existing cookie.
 824		 */
 825		if (size > isize)
 826			ceph_fscache_update(inode);
 827		ci->i_reported_size = size;
 828		if (truncate_seq != ci->i_truncate_seq) {
 829			doutc(cl, "truncate_seq %u -> %u\n",
 830			      ci->i_truncate_seq, truncate_seq);
 831			ci->i_truncate_seq = truncate_seq;
 832
 833			/* the MDS should have revoked these caps */
 834			WARN_ON_ONCE(issued & (CEPH_CAP_FILE_RD |
 835					       CEPH_CAP_FILE_LAZYIO));
 836			/*
 837			 * If we hold relevant caps, or in the case where we're
 838			 * not the only client referencing this file and we
 839			 * don't hold those caps, then we need to check whether
 840			 * the file is either opened or mmaped
 841			 */
 842			if ((issued & (CEPH_CAP_FILE_CACHE|
 843				       CEPH_CAP_FILE_BUFFER)) ||
 844			    mapping_mapped(inode->i_mapping) ||
 845			    __ceph_is_file_opened(ci)) {
 846				ci->i_truncate_pending++;
 847				queue_trunc = 1;
 848			}
 849		}
 850	}
 851
 852	/*
 853	 * It's possible that the new sizes of the two consecutive
 854	 * size truncations will be in the same fscrypt last block,
 855	 * and we need to truncate the corresponding page caches
 856	 * anyway.
 857	 */
 858	if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0) {
 859		doutc(cl, "truncate_size %lld -> %llu, encrypted %d\n",
 860		      ci->i_truncate_size, truncate_size,
 861		      !!IS_ENCRYPTED(inode));
 862
 863		ci->i_truncate_size = truncate_size;
 864
 865		if (IS_ENCRYPTED(inode)) {
 866			doutc(cl, "truncate_pagecache_size %lld -> %llu\n",
 867			      ci->i_truncate_pagecache_size, size);
 868			ci->i_truncate_pagecache_size = size;
 869		} else {
 870			ci->i_truncate_pagecache_size = truncate_size;
 871		}
 872	}
 873	return queue_trunc;
 874}
 875
 876void ceph_fill_file_time(struct inode *inode, int issued,
 877			 u64 time_warp_seq, struct timespec64 *ctime,
 878			 struct timespec64 *mtime, struct timespec64 *atime)
 879{
 880	struct ceph_client *cl = ceph_inode_to_client(inode);
 881	struct ceph_inode_info *ci = ceph_inode(inode);
 882	struct timespec64 iatime = inode_get_atime(inode);
 883	struct timespec64 ictime = inode_get_ctime(inode);
 884	struct timespec64 imtime = inode_get_mtime(inode);
 885	int warn = 0;
 886
 887	if (issued & (CEPH_CAP_FILE_EXCL|
 888		      CEPH_CAP_FILE_WR|
 889		      CEPH_CAP_FILE_BUFFER|
 890		      CEPH_CAP_AUTH_EXCL|
 891		      CEPH_CAP_XATTR_EXCL)) {
 892		if (ci->i_version == 0 ||
 893		    timespec64_compare(ctime, &ictime) > 0) {
 894			doutc(cl, "ctime %ptSp -> %ptSp inc w/ cap\n", &ictime, ctime);
 895			inode_set_ctime_to_ts(inode, *ctime);
 896		}
 897		if (ci->i_version == 0 ||
 898		    ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 899			/* the MDS did a utimes() */
 900			doutc(cl, "mtime %ptSp -> %ptSp tw %d -> %d\n", &imtime, mtime,
 901			      ci->i_time_warp_seq, (int)time_warp_seq);
 902
 903			inode_set_mtime_to_ts(inode, *mtime);
 904			inode_set_atime_to_ts(inode, *atime);
 905			ci->i_time_warp_seq = time_warp_seq;
 906		} else if (time_warp_seq == ci->i_time_warp_seq) {
 907			/* nobody did utimes(); take the max */
 908			if (timespec64_compare(mtime, &imtime) > 0) {
 909				doutc(cl, "mtime %ptSp -> %ptSp inc\n", &imtime, mtime);
 910				inode_set_mtime_to_ts(inode, *mtime);
 911			}
 912			if (timespec64_compare(atime, &iatime) > 0) {
 913				doutc(cl, "atime %ptSp -> %ptSp inc\n", &iatime, atime);
 914				inode_set_atime_to_ts(inode, *atime);
 915			}
 916		} else if (issued & CEPH_CAP_FILE_EXCL) {
 917			/* we did a utimes(); ignore mds values */
 918		} else {
 919			warn = 1;
 920		}
 921	} else {
 922		/* we have no write|excl caps; whatever the MDS says is true */
 923		if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 924			inode_set_ctime_to_ts(inode, *ctime);
 925			inode_set_mtime_to_ts(inode, *mtime);
 926			inode_set_atime_to_ts(inode, *atime);
 927			ci->i_time_warp_seq = time_warp_seq;
 928		} else {
 929			warn = 1;
 930		}
 931	}
 932	if (warn) /* time_warp_seq shouldn't go backwards */
 933		doutc(cl, "%p mds time_warp_seq %llu < %u\n", inode,
 934		      time_warp_seq, ci->i_time_warp_seq);
 935}
 936
 937#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
 938static int decode_encrypted_symlink(struct ceph_mds_client *mdsc,
 939				    const char *encsym,
 940				    int enclen, u8 **decsym)
 941{
 942	struct ceph_client *cl = mdsc->fsc->client;
 943	int declen;
 944	u8 *sym;
 945
 946	sym = kmalloc(enclen + 1, GFP_NOFS);
 947	if (!sym)
 948		return -ENOMEM;
 949
 950	declen = base64_decode(encsym, enclen, sym, false, BASE64_IMAP);
 951	if (declen < 0) {
 952		pr_err_client(cl,
 953			"can't decode symlink (%d). Content: %.*s\n",
 954			declen, enclen, encsym);
 955		kfree(sym);
 956		return -EIO;
 957	}
 958	sym[declen + 1] = '\0';
 959	*decsym = sym;
 960	return declen;
 961}
 962#else
 963static int decode_encrypted_symlink(struct ceph_mds_client *mdsc,
 964				    const char *encsym,
 965				    int symlen, u8 **decsym)
 966{
 967	return -EOPNOTSUPP;
 968}
 969#endif
 970
 971/*
 972 * Populate an inode based on info from mds.  May be called on new or
 973 * existing inodes.
 974 */
 975int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 976		    struct ceph_mds_reply_info_in *iinfo,
 977		    struct ceph_mds_reply_dirfrag *dirinfo,
 978		    struct ceph_mds_session *session, int cap_fmode,
 979		    struct ceph_cap_reservation *caps_reservation)
 980{
 981	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 982	struct ceph_client *cl = mdsc->fsc->client;
 983	struct ceph_mds_reply_inode *info = iinfo->in;
 984	struct ceph_inode_info *ci = ceph_inode(inode);
 985	int issued, new_issued, info_caps;
 986	struct timespec64 mtime, atime, ctime;
 987	struct ceph_buffer *xattr_blob = NULL;
 988	struct ceph_buffer *old_blob = NULL;
 989	struct ceph_string *pool_ns = NULL;
 990	struct ceph_cap *new_cap = NULL;
 991	int err = 0;
 992	bool wake = false;
 993	bool queue_trunc = false;
 994	bool new_version = false;
 995	bool fill_inline = false;
 996	umode_t mode = le32_to_cpu(info->mode);
 997	dev_t rdev = le32_to_cpu(info->rdev);
 998
 999	lockdep_assert_held(&mdsc->snap_rwsem);
1000
1001	doutc(cl, "%p ino %llx.%llx v %llu had %llu\n", inode, ceph_vinop(inode),
1002	      le64_to_cpu(info->version), ci->i_version);
1003
1004	/* Once I_NEW is cleared, we can't change type or dev numbers */
1005	if (inode_state_read_once(inode) & I_NEW) {
1006		inode->i_mode = mode;
1007	} else {
1008		if (inode_wrong_type(inode, mode)) {
1009			pr_warn_once_client(cl,
1010				"inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
1011				ceph_vinop(inode), inode->i_mode, mode);
1012			return -ESTALE;
1013		}
1014
1015		if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
1016			pr_warn_once_client(cl,
1017				"dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
1018				ceph_vinop(inode), MAJOR(inode->i_rdev),
1019				MINOR(inode->i_rdev), MAJOR(rdev),
1020				MINOR(rdev));
1021			return -ESTALE;
1022		}
1023	}
1024
1025	info_caps = le32_to_cpu(info->cap.caps);
1026
1027	/* prealloc new cap struct */
1028	if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
1029		new_cap = ceph_get_cap(mdsc, caps_reservation);
1030		if (!new_cap)
1031			return -ENOMEM;
1032	}
1033
1034	/*
1035	 * prealloc xattr data, if it looks like we'll need it.  only
1036	 * if len > 4 (meaning there are actually xattrs; the first 4
1037	 * bytes are the xattr count).
1038	 */
1039	if (iinfo->xattr_len > 4) {
1040		xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
1041		if (!xattr_blob)
1042			pr_err_client(cl, "ENOMEM xattr blob %d bytes\n",
1043				      iinfo->xattr_len);
1044	}
1045
1046	if (iinfo->pool_ns_len > 0)
1047		pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
1048						     iinfo->pool_ns_len);
1049
1050	if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
1051		ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
1052
1053	spin_lock(&ci->i_ceph_lock);
1054
1055	/*
1056	 * provided version will be odd if inode value is projected,
1057	 * even if stable.  skip the update if we have newer stable
1058	 * info (ours>=theirs, e.g. due to racing mds replies), unless
1059	 * we are getting projected (unstable) info (in which case the
1060	 * version is odd, and we want ours>theirs).
1061	 *   us   them
1062	 *   2    2     skip
1063	 *   3    2     skip
1064	 *   3    3     update
1065	 */
1066	if (ci->i_version == 0 ||
1067	    ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1068	     le64_to_cpu(info->version) > (ci->i_version & ~1)))
1069		new_version = true;
1070
1071	/* Update change_attribute */
1072	inode_set_max_iversion_raw(inode, iinfo->change_attr);
1073
1074	__ceph_caps_issued(ci, &issued);
1075	issued |= __ceph_caps_dirty(ci);
1076	new_issued = ~issued & info_caps;
1077
1078	__ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
1079
1080#ifdef CONFIG_FS_ENCRYPTION
1081	if (iinfo->fscrypt_auth_len &&
1082	    ((inode_state_read_once(inode) & I_NEW) || (ci->fscrypt_auth_len == 0))) {
1083		kfree(ci->fscrypt_auth);
1084		ci->fscrypt_auth_len = iinfo->fscrypt_auth_len;
1085		ci->fscrypt_auth = iinfo->fscrypt_auth;
1086		iinfo->fscrypt_auth = NULL;
1087		iinfo->fscrypt_auth_len = 0;
1088		inode_set_flags(inode, S_ENCRYPTED, S_ENCRYPTED);
1089	}
1090#endif
1091
1092	if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
1093	    (issued & CEPH_CAP_AUTH_EXCL) == 0) {
1094		inode->i_mode = mode;
1095		inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
1096		inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
1097		doutc(cl, "%p %llx.%llx mode 0%o uid.gid %d.%d\n", inode,
1098		      ceph_vinop(inode), inode->i_mode,
1099		      from_kuid(&init_user_ns, inode->i_uid),
1100		      from_kgid(&init_user_ns, inode->i_gid));
1101		ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
1102		ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
1103	}
1104
1105	/* directories have fl_stripe_unit set to zero */
1106	if (IS_ENCRYPTED(inode))
1107		inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
1108	else if (le32_to_cpu(info->layout.fl_stripe_unit))
1109		inode->i_blkbits =
1110			fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
1111	else
1112		inode->i_blkbits = CEPH_BLOCK_SHIFT;
1113
1114	if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
1115	    (issued & CEPH_CAP_LINK_EXCL) == 0)
1116		set_nlink(inode, le32_to_cpu(info->nlink));
1117
1118	if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
1119		/* be careful with mtime, atime, size */
1120		ceph_decode_timespec64(&atime, &info->atime);
1121		ceph_decode_timespec64(&mtime, &info->mtime);
1122		ceph_decode_timespec64(&ctime, &info->ctime);
1123		ceph_fill_file_time(inode, issued,
1124				le32_to_cpu(info->time_warp_seq),
1125				&ctime, &mtime, &atime);
1126	}
1127
1128	if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
1129		ci->i_files = le64_to_cpu(info->files);
1130		ci->i_subdirs = le64_to_cpu(info->subdirs);
1131	}
1132
1133	if (new_version ||
1134	    (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
1135		u64 size = le64_to_cpu(info->size);
1136		s64 old_pool = ci->i_layout.pool_id;
1137		struct ceph_string *old_ns;
1138
1139		ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
1140		old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
1141					lockdep_is_held(&ci->i_ceph_lock));
1142		rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
1143
1144		if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
1145			ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
1146
1147		pool_ns = old_ns;
1148
1149		if (IS_ENCRYPTED(inode) && size &&
1150		    iinfo->fscrypt_file_len == sizeof(__le64)) {
1151			u64 fsize = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
1152
1153			if (size == round_up(fsize, CEPH_FSCRYPT_BLOCK_SIZE)) {
1154				size = fsize;
1155			} else {
1156				pr_warn_client(cl,
1157					"fscrypt size mismatch: size=%llu fscrypt_file=%llu, discarding fscrypt_file size.\n",
1158					info->size, size);
1159			}
1160		}
1161
1162		queue_trunc = ceph_fill_file_size(inode, issued,
1163					le32_to_cpu(info->truncate_seq),
1164					le64_to_cpu(info->truncate_size),
1165					size);
1166		/* only update max_size on auth cap */
1167		if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1168		    ci->i_max_size != le64_to_cpu(info->max_size)) {
1169			doutc(cl, "max_size %lld -> %llu\n",
1170			    ci->i_max_size, le64_to_cpu(info->max_size));
1171			ci->i_max_size = le64_to_cpu(info->max_size);
1172		}
1173	}
1174
1175	/* layout and rstat are not tracked by capability, update them if
1176	 * the inode info is from auth mds */
1177	if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
1178		if (S_ISDIR(inode->i_mode)) {
1179			ci->i_dir_layout = iinfo->dir_layout;
1180			ci->i_rbytes = le64_to_cpu(info->rbytes);
1181			ci->i_rfiles = le64_to_cpu(info->rfiles);
1182			ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
1183			ci->i_dir_pin = iinfo->dir_pin;
1184			ci->i_rsnaps = iinfo->rsnaps;
1185			ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
1186		}
1187	}
1188
1189	/* xattrs */
1190	/* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
1191	if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
1192	    le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
1193		if (ci->i_xattrs.blob)
1194			old_blob = ci->i_xattrs.blob;
1195		ci->i_xattrs.blob = xattr_blob;
1196		if (xattr_blob)
1197			memcpy(ci->i_xattrs.blob->vec.iov_base,
1198			       iinfo->xattr_data, iinfo->xattr_len);
1199		ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
1200		ceph_forget_all_cached_acls(inode);
1201		ceph_security_invalidate_secctx(inode);
1202		xattr_blob = NULL;
1203	}
1204
1205	/* finally update i_version */
1206	if (le64_to_cpu(info->version) > ci->i_version)
1207		ci->i_version = le64_to_cpu(info->version);
1208
1209	inode->i_mapping->a_ops = &ceph_aops;
1210
1211	switch (inode->i_mode & S_IFMT) {
1212	case S_IFIFO:
1213	case S_IFBLK:
1214	case S_IFCHR:
1215	case S_IFSOCK:
1216		inode->i_blkbits = PAGE_SHIFT;
1217		init_special_inode(inode, inode->i_mode, rdev);
1218		inode->i_op = &ceph_file_iops;
1219		break;
1220	case S_IFREG:
1221		inode->i_op = &ceph_file_iops;
1222		inode->i_fop = &ceph_file_fops;
1223		break;
1224	case S_IFLNK:
1225		if (!ci->i_symlink) {
1226			u32 symlen = iinfo->symlink_len;
1227			char *sym;
1228
1229			spin_unlock(&ci->i_ceph_lock);
1230
1231			if (IS_ENCRYPTED(inode)) {
1232				if (symlen != i_size_read(inode))
1233					pr_err_client(cl,
1234						"%p %llx.%llx BAD symlink size %lld\n",
1235						inode, ceph_vinop(inode),
1236						i_size_read(inode));
1237
1238				err = decode_encrypted_symlink(mdsc, iinfo->symlink,
1239							       symlen, (u8 **)&sym);
1240				if (err < 0) {
1241					pr_err_client(cl,
1242						"decoding encrypted symlink failed: %d\n",
1243						err);
1244					goto out;
1245				}
1246				symlen = err;
1247				i_size_write(inode, symlen);
1248				inode->i_blocks = calc_inode_blocks(symlen);
1249			} else {
1250				if (symlen != i_size_read(inode)) {
1251					pr_err_client(cl,
1252						"%p %llx.%llx BAD symlink size %lld\n",
1253						inode, ceph_vinop(inode),
1254						i_size_read(inode));
1255					i_size_write(inode, symlen);
1256					inode->i_blocks = calc_inode_blocks(symlen);
1257				}
1258
1259				err = -ENOMEM;
1260				sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
1261				if (!sym)
1262					goto out;
1263			}
1264
1265			spin_lock(&ci->i_ceph_lock);
1266			if (!ci->i_symlink)
1267				ci->i_symlink = sym;
1268			else
1269				kfree(sym); /* lost a race */
1270		}
1271
1272		if (IS_ENCRYPTED(inode)) {
1273			/*
1274			 * Encrypted symlinks need to be decrypted before we can
1275			 * cache their targets in i_link. Don't touch it here.
1276			 */
1277			inode->i_op = &ceph_encrypted_symlink_iops;
1278		} else {
1279			inode->i_link = ci->i_symlink;
1280			inode->i_op = &ceph_symlink_iops;
1281		}
1282		break;
1283	case S_IFDIR:
1284		inode->i_op = &ceph_dir_iops;
1285		inode->i_fop = &ceph_dir_fops;
1286		break;
1287	default:
1288		pr_err_client(cl, "%p %llx.%llx BAD mode 0%o\n", inode,
1289			      ceph_vinop(inode), inode->i_mode);
1290	}
1291
1292	/* were we issued a capability? */
1293	if (info_caps) {
1294		if (ceph_snap(inode) == CEPH_NOSNAP) {
1295			ceph_add_cap(inode, session,
1296				     le64_to_cpu(info->cap.cap_id),
1297				     info_caps,
1298				     le32_to_cpu(info->cap.wanted),
1299				     le32_to_cpu(info->cap.seq),
1300				     le32_to_cpu(info->cap.mseq),
1301				     le64_to_cpu(info->cap.realm),
1302				     info->cap.flags, &new_cap);
1303
1304			/* set dir completion flag? */
1305			if (S_ISDIR(inode->i_mode) &&
1306			    ci->i_files == 0 && ci->i_subdirs == 0 &&
1307			    (info_caps & CEPH_CAP_FILE_SHARED) &&
1308			    (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1309			    !__ceph_dir_is_complete(ci)) {
1310				doutc(cl, " marking %p complete (empty)\n",
1311				      inode);
1312				i_size_write(inode, 0);
1313				__ceph_dir_set_complete(ci,
1314					atomic64_read(&ci->i_release_count),
1315					atomic64_read(&ci->i_ordered_count));
1316			}
1317
1318			wake = true;
1319		} else {
1320			doutc(cl, " %p got snap_caps %s\n", inode,
1321			      ceph_cap_string(info_caps));
1322			ci->i_snap_caps |= info_caps;
1323		}
1324	}
1325
1326	if (iinfo->inline_version > 0 &&
1327	    iinfo->inline_version >= ci->i_inline_version) {
1328		int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1329		ci->i_inline_version = iinfo->inline_version;
1330		if (ceph_has_inline_data(ci) &&
1331		    (locked_page || (info_caps & cache_caps)))
1332			fill_inline = true;
1333	}
1334
1335	if (cap_fmode >= 0) {
1336		if (!info_caps)
1337			pr_warn_client(cl, "mds issued no caps on %llx.%llx\n",
1338				       ceph_vinop(inode));
1339		__ceph_touch_fmode(ci, mdsc, cap_fmode);
1340	}
1341
1342	spin_unlock(&ci->i_ceph_lock);
1343
1344	ceph_fscache_register_inode_cookie(inode);
1345
1346	if (fill_inline)
1347		ceph_fill_inline_data(inode, locked_page,
1348				      iinfo->inline_data, iinfo->inline_len);
1349
1350	if (wake)
1351		wake_up_all(&ci->i_cap_wq);
1352
1353	/* queue truncate if we saw i_size decrease */
1354	if (queue_trunc)
1355		ceph_queue_vmtruncate(inode);
1356
1357	/* populate frag tree */
1358	if (S_ISDIR(inode->i_mode))
1359		ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1360
1361	/* update delegation info? */
1362	if (dirinfo)
1363		ceph_fill_dirfrag(inode, dirinfo);
1364
1365	err = 0;
1366out:
1367	if (new_cap)
1368		ceph_put_cap(mdsc, new_cap);
1369	ceph_buffer_put(old_blob);
1370	ceph_buffer_put(xattr_blob);
1371	ceph_put_string(pool_ns);
1372	return err;
1373}
1374
1375/*
1376 * caller should hold session s_mutex and dentry->d_lock.
1377 */
1378static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1379				  struct ceph_mds_reply_lease *lease,
1380				  struct ceph_mds_session *session,
1381				  unsigned long from_time,
1382				  struct ceph_mds_session **old_lease_session)
1383{
1384	struct ceph_client *cl = ceph_inode_to_client(dir);
1385	struct ceph_dentry_info *di = ceph_dentry(dentry);
1386	unsigned mask = le16_to_cpu(lease->mask);
1387	long unsigned duration = le32_to_cpu(lease->duration_ms);
1388	long unsigned ttl = from_time + (duration * HZ) / 1000;
1389	long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1390
1391	doutc(cl, "%p duration %lu ms ttl %lu\n", dentry, duration, ttl);
1392
1393	/* only track leases on regular dentries */
1394	if (ceph_snap(dir) != CEPH_NOSNAP)
1395		return;
1396
1397	if (mask & CEPH_LEASE_PRIMARY_LINK)
1398		di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1399	else
1400		di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1401
1402	di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1403	if (!(mask & CEPH_LEASE_VALID)) {
1404		__ceph_dentry_dir_lease_touch(di);
1405		return;
1406	}
1407
1408	if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1409	    time_before(ttl, di->time))
1410		return;  /* we already have a newer lease. */
1411
1412	if (di->lease_session && di->lease_session != session) {
1413		*old_lease_session = di->lease_session;
1414		di->lease_session = NULL;
1415	}
1416
1417	if (!di->lease_session)
1418		di->lease_session = ceph_get_mds_session(session);
1419	di->lease_gen = atomic_read(&session->s_cap_gen);
1420	di->lease_seq = le32_to_cpu(lease->seq);
1421	di->lease_renew_after = half_ttl;
1422	di->lease_renew_from = 0;
1423	di->time = ttl;
1424
1425	__ceph_dentry_lease_touch(di);
1426}
1427
1428static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1429					struct ceph_mds_reply_lease *lease,
1430					struct ceph_mds_session *session,
1431					unsigned long from_time)
1432{
1433	struct ceph_mds_session *old_lease_session = NULL;
1434	spin_lock(&dentry->d_lock);
1435	__update_dentry_lease(dir, dentry, lease, session, from_time,
1436			      &old_lease_session);
1437	spin_unlock(&dentry->d_lock);
1438	ceph_put_mds_session(old_lease_session);
1439}
1440
1441/*
1442 * update dentry lease without having parent inode locked
1443 */
1444static void update_dentry_lease_careful(struct dentry *dentry,
1445					struct ceph_mds_reply_lease *lease,
1446					struct ceph_mds_session *session,
1447					unsigned long from_time,
1448					char *dname, u32 dname_len,
1449					struct ceph_vino *pdvino,
1450					struct ceph_vino *ptvino)
1451
1452{
1453	struct inode *dir;
1454	struct ceph_mds_session *old_lease_session = NULL;
1455
1456	spin_lock(&dentry->d_lock);
1457	/* make sure dentry's name matches target */
1458	if (dentry->d_name.len != dname_len ||
1459	    memcmp(dentry->d_name.name, dname, dname_len))
1460		goto out_unlock;
1461
1462	dir = d_inode(dentry->d_parent);
1463	/* make sure parent matches dvino */
1464	if (!ceph_ino_compare(dir, pdvino))
1465		goto out_unlock;
1466
1467	/* make sure dentry's inode matches target. NULL ptvino means that
1468	 * we expect a negative dentry */
1469	if (ptvino) {
1470		if (d_really_is_negative(dentry))
1471			goto out_unlock;
1472		if (!ceph_ino_compare(d_inode(dentry), ptvino))
1473			goto out_unlock;
1474	} else {
1475		if (d_really_is_positive(dentry))
1476			goto out_unlock;
1477	}
1478
1479	__update_dentry_lease(dir, dentry, lease, session,
1480			      from_time, &old_lease_session);
1481out_unlock:
1482	spin_unlock(&dentry->d_lock);
1483	ceph_put_mds_session(old_lease_session);
1484}
1485
1486/*
1487 * splice a dentry to an inode.
1488 * caller must hold directory i_rwsem for this to be safe.
1489 */
1490static int splice_dentry(struct dentry **pdn, struct inode *in)
1491{
1492	struct ceph_client *cl = ceph_inode_to_client(in);
1493	struct dentry *dn = *pdn;
1494	struct dentry *realdn;
1495
1496	BUG_ON(d_inode(dn));
1497
1498	if (S_ISDIR(in->i_mode)) {
1499		/* If inode is directory, d_splice_alias() below will remove
1500		 * 'realdn' from its origin parent. We need to ensure that
1501		 * origin parent's readdir cache will not reference 'realdn'
1502		 */
1503		realdn = d_find_any_alias(in);
1504		if (realdn) {
1505			struct ceph_dentry_info *di = ceph_dentry(realdn);
1506			spin_lock(&realdn->d_lock);
1507
1508			realdn->d_op->d_prune(realdn);
1509
1510			di->time = jiffies;
1511			di->lease_shared_gen = 0;
1512			di->offset = 0;
1513
1514			spin_unlock(&realdn->d_lock);
1515			dput(realdn);
1516		}
1517	}
1518
1519	/* dn must be unhashed */
1520	if (!d_unhashed(dn))
1521		d_drop(dn);
1522	realdn = d_splice_alias(in, dn);
1523	if (IS_ERR(realdn)) {
1524		pr_err_client(cl, "error %ld %p inode %p ino %llx.%llx\n",
1525			      PTR_ERR(realdn), dn, in, ceph_vinop(in));
1526		return PTR_ERR(realdn);
1527	}
1528
1529	if (realdn) {
1530		doutc(cl, "dn %p (%d) spliced with %p (%d) inode %p ino %llx.%llx\n",
1531		      dn, d_count(dn), realdn, d_count(realdn),
1532		      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1533		dput(dn);
1534		*pdn = realdn;
1535	} else {
1536		BUG_ON(!ceph_dentry(dn));
1537		doutc(cl, "dn %p attached to %p ino %llx.%llx\n", dn,
1538		      d_inode(dn), ceph_vinop(d_inode(dn)));
1539	}
1540	return 0;
1541}
1542
1543/*
1544 * Incorporate results into the local cache.  This is either just
1545 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1546 * after a lookup).
1547 *
1548 * A reply may contain
1549 *         a directory inode along with a dentry.
1550 *  and/or a target inode
1551 *
1552 * Called with snap_rwsem (read).
1553 */
1554int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1555{
1556	struct ceph_mds_session *session = req->r_session;
1557	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1558	struct inode *in = NULL;
1559	struct ceph_vino tvino, dvino;
1560	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1561	struct ceph_client *cl = fsc->client;
1562	struct inode *parent_dir = NULL;
1563	int err = 0;
1564
1565	doutc(cl, "%p is_dentry %d is_target %d\n", req,
1566	      rinfo->head->is_dentry, rinfo->head->is_target);
1567
1568	if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1569		doutc(cl, "reply is empty!\n");
1570		if (rinfo->head->result == 0 && req->r_parent)
1571			ceph_invalidate_dir_request(req);
1572		return 0;
1573	}
1574
1575	if (rinfo->head->is_dentry) {
1576		/*
1577		 * r_parent may be stale, in cases when R_PARENT_LOCKED is not set,
1578		 * so we need to get the correct inode
1579		 */
1580		parent_dir = ceph_get_reply_dir(sb, req->r_parent, rinfo);
1581		if (unlikely(IS_ERR(parent_dir))) {
1582			err = PTR_ERR(parent_dir);
1583			goto done;
1584		}
1585		if (parent_dir) {
1586			err = ceph_fill_inode(parent_dir, NULL, &rinfo->diri,
1587					      rinfo->dirfrag, session, -1,
1588					      &req->r_caps_reservation);
1589			if (err < 0)
1590				goto done;
1591		} else {
1592			WARN_ON_ONCE(1);
1593		}
1594
1595		if (parent_dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1596		    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1597		    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1598			bool is_nokey = false;
1599			struct qstr dname;
1600			struct dentry *dn, *parent;
1601			struct fscrypt_str oname = FSTR_INIT(NULL, 0);
1602			struct ceph_fname fname = { .dir	= parent_dir,
1603						    .name	= rinfo->dname,
1604						    .ctext	= rinfo->altname,
1605						    .name_len	= rinfo->dname_len,
1606						    .ctext_len	= rinfo->altname_len };
1607
1608			BUG_ON(!rinfo->head->is_target);
1609			BUG_ON(req->r_dentry);
1610
1611			parent = d_find_any_alias(parent_dir);
1612			BUG_ON(!parent);
1613
1614			err = ceph_fname_alloc_buffer(parent_dir, &oname);
1615			if (err < 0) {
1616				dput(parent);
1617				goto done;
1618			}
1619
1620			err = ceph_fname_to_usr(&fname, NULL, &oname, &is_nokey);
1621			if (err < 0) {
1622				dput(parent);
1623				ceph_fname_free_buffer(parent_dir, &oname);
1624				goto done;
1625			}
1626			dname.name = oname.name;
1627			dname.len = oname.len;
1628			dname.hash = full_name_hash(parent, dname.name, dname.len);
1629			tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1630			tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1631retry_lookup:
1632			dn = d_lookup(parent, &dname);
1633			doutc(cl, "d_lookup on parent=%p name=%.*s got %p\n",
1634			      parent, dname.len, dname.name, dn);
1635
1636			if (!dn) {
1637				dn = d_alloc(parent, &dname);
1638				doutc(cl, "d_alloc %p '%.*s' = %p\n", parent,
1639				      dname.len, dname.name, dn);
1640				if (!dn) {
1641					dput(parent);
1642					ceph_fname_free_buffer(parent_dir, &oname);
1643					err = -ENOMEM;
1644					goto done;
1645				}
1646				if (is_nokey) {
1647					spin_lock(&dn->d_lock);
1648					dn->d_flags |= DCACHE_NOKEY_NAME;
1649					spin_unlock(&dn->d_lock);
1650				}
1651				err = 0;
1652			} else if (d_really_is_positive(dn) &&
1653				   (ceph_ino(d_inode(dn)) != tvino.ino ||
1654				    ceph_snap(d_inode(dn)) != tvino.snap)) {
1655				doutc(cl, " dn %p points to wrong inode %p\n",
1656				      dn, d_inode(dn));
1657				ceph_dir_clear_ordered(parent_dir);
1658				d_delete(dn);
1659				dput(dn);
1660				goto retry_lookup;
1661			}
1662			ceph_fname_free_buffer(parent_dir, &oname);
1663
1664			req->r_dentry = dn;
1665			dput(parent);
1666		}
1667	}
1668
1669	if (rinfo->head->is_target) {
1670		/* Should be filled in by handle_reply */
1671		BUG_ON(!req->r_target_inode);
1672
1673		in = req->r_target_inode;
1674		err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1675				NULL, session,
1676				(!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1677				 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1678				 rinfo->head->result == 0) ?  req->r_fmode : -1,
1679				&req->r_caps_reservation);
1680		if (err < 0) {
1681			pr_err_client(cl, "badness %p %llx.%llx\n", in,
1682				      ceph_vinop(in));
1683			req->r_target_inode = NULL;
1684			if (inode_state_read_once(in) & I_NEW)
1685				discard_new_inode(in);
1686			else
1687				iput(in);
1688			goto done;
1689		}
1690		if (inode_state_read_once(in) & I_NEW)
1691			unlock_new_inode(in);
1692	}
1693
1694	/*
1695	 * ignore null lease/binding on snapdir ENOENT, or else we
1696	 * will have trouble splicing in the virtual snapdir later
1697	 */
1698	if (rinfo->head->is_dentry &&
1699            !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1700	    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1701	    (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1702					       fsc->mount_options->snapdir_name,
1703					       req->r_dentry->d_name.len))) {
1704		/*
1705		 * lookup link rename   : null -> possibly existing inode
1706		 * mknod symlink mkdir  : null -> new inode
1707		 * unlink               : linked -> null
1708		 */
1709		struct inode *dir = req->r_parent;
1710		struct dentry *dn = req->r_dentry;
1711		bool have_dir_cap, have_lease;
1712
1713		BUG_ON(!dn);
1714		BUG_ON(!dir);
1715		BUG_ON(d_inode(dn->d_parent) != dir);
1716
1717		dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1718		dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1719
1720		BUG_ON(ceph_ino(dir) != dvino.ino);
1721		BUG_ON(ceph_snap(dir) != dvino.snap);
1722
1723		/* do we have a lease on the whole dir? */
1724		have_dir_cap =
1725			(le32_to_cpu(rinfo->diri.in->cap.caps) &
1726			 CEPH_CAP_FILE_SHARED);
1727
1728		/* do we have a dn lease? */
1729		have_lease = have_dir_cap ||
1730			le32_to_cpu(rinfo->dlease->duration_ms);
1731		if (!have_lease)
1732			doutc(cl, "no dentry lease or dir cap\n");
1733
1734		/* rename? */
1735		if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1736			struct inode *olddir = req->r_old_dentry_dir;
1737			BUG_ON(!olddir);
1738
1739			doutc(cl, " src %p '%pd' dst %p '%pd'\n",
1740			      req->r_old_dentry, req->r_old_dentry, dn, dn);
1741			doutc(cl, "doing d_move %p -> %p\n", req->r_old_dentry, dn);
1742
1743			/* d_move screws up sibling dentries' offsets */
1744			ceph_dir_clear_ordered(dir);
1745			ceph_dir_clear_ordered(olddir);
1746
1747			d_move(req->r_old_dentry, dn);
1748			doutc(cl, " src %p '%pd' dst %p '%pd'\n",
1749			      req->r_old_dentry, req->r_old_dentry, dn, dn);
1750
1751			/* ensure target dentry is invalidated, despite
1752			   rehashing bug in vfs_rename_dir */
1753			ceph_invalidate_dentry_lease(dn);
1754
1755			doutc(cl, "dn %p gets new offset %lld\n",
1756			      req->r_old_dentry,
1757			      ceph_dentry(req->r_old_dentry)->offset);
1758
1759			/* swap r_dentry and r_old_dentry in case that
1760			 * splice_dentry() gets called later. This is safe
1761			 * because no other place will use them */
1762			req->r_dentry = req->r_old_dentry;
1763			req->r_old_dentry = dn;
1764			dn = req->r_dentry;
1765		}
1766
1767		/* null dentry? */
1768		if (!rinfo->head->is_target) {
1769			doutc(cl, "null dentry\n");
1770			if (d_really_is_positive(dn)) {
1771				doutc(cl, "d_delete %p\n", dn);
1772				ceph_dir_clear_ordered(dir);
1773				d_delete(dn);
1774			} else if (have_lease) {
1775				if (d_unhashed(dn))
1776					d_add(dn, NULL);
1777			}
1778
1779			if (!d_unhashed(dn) && have_lease)
1780				update_dentry_lease(dir, dn,
1781						    rinfo->dlease, session,
1782						    req->r_request_started);
1783			goto done;
1784		}
1785
1786		if (unlikely(!in)) {
1787			err = -EINVAL;
1788			goto done;
1789		}
1790
1791		/* attach proper inode */
1792		if (d_really_is_negative(dn)) {
1793			ceph_dir_clear_ordered(dir);
1794			ihold(in);
1795			err = splice_dentry(&req->r_dentry, in);
1796			if (err < 0)
1797				goto done;
1798			dn = req->r_dentry;  /* may have spliced */
1799		} else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1800			doutc(cl, " %p links to %p %llx.%llx, not %llx.%llx\n",
1801			      dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1802			      ceph_vinop(in));
1803			d_invalidate(dn);
1804			have_lease = false;
1805		}
1806
1807		if (have_lease) {
1808			update_dentry_lease(dir, dn,
1809					    rinfo->dlease, session,
1810					    req->r_request_started);
1811		}
1812		doutc(cl, " final dn %p\n", dn);
1813	} else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1814		    req->r_op == CEPH_MDS_OP_MKSNAP) &&
1815	           test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1816		   !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1817		struct inode *dir = req->r_parent;
1818
1819		/* fill out a snapdir LOOKUPSNAP dentry */
1820		BUG_ON(!dir);
1821		BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1822		BUG_ON(!req->r_dentry);
1823		doutc(cl, " linking snapped dir %p to dn %p\n", in,
1824		      req->r_dentry);
1825		ceph_dir_clear_ordered(dir);
1826
1827		if (unlikely(!in)) {
1828			err = -EINVAL;
1829			goto done;
1830		}
1831
1832		ihold(in);
1833		err = splice_dentry(&req->r_dentry, in);
1834		if (err < 0)
1835			goto done;
1836	} else if (rinfo->head->is_dentry && req->r_dentry) {
1837		/* parent inode is not locked, be careful */
1838		struct ceph_vino *ptvino = NULL;
1839		dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1840		dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1841		if (rinfo->head->is_target) {
1842			tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1843			tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1844			ptvino = &tvino;
1845		}
1846		update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1847					    session, req->r_request_started,
1848					    rinfo->dname, rinfo->dname_len,
1849					    &dvino, ptvino);
1850	}
1851done:
1852	/* Drop extra ref from ceph_get_reply_dir() if it returned a new inode */
1853	if (unlikely(!IS_ERR_OR_NULL(parent_dir) && parent_dir != req->r_parent))
1854		iput(parent_dir);
1855	doutc(cl, "done err=%d\n", err);
1856	return err;
1857}
1858
1859/*
1860 * Prepopulate our cache with readdir results, leases, etc.
1861 */
1862static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1863					   struct ceph_mds_session *session)
1864{
1865	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1866	struct ceph_client *cl = session->s_mdsc->fsc->client;
1867	int i, err = 0;
1868
1869	for (i = 0; i < rinfo->dir_nr; i++) {
1870		struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1871		struct ceph_vino vino;
1872		struct inode *in;
1873		int rc;
1874
1875		vino.ino = le64_to_cpu(rde->inode.in->ino);
1876		vino.snap = le64_to_cpu(rde->inode.in->snapid);
1877
1878		in = ceph_get_inode(req->r_dentry->d_sb, vino, NULL);
1879		if (IS_ERR(in)) {
1880			err = PTR_ERR(in);
1881			doutc(cl, "badness got %d\n", err);
1882			continue;
1883		}
1884		rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1885				     -1, &req->r_caps_reservation);
1886		if (rc < 0) {
1887			pr_err_client(cl, "inode badness on %p got %d\n", in,
1888				      rc);
1889			err = rc;
1890			if (inode_state_read_once(in) & I_NEW) {
1891				ihold(in);
1892				discard_new_inode(in);
1893			}
1894		} else if (inode_state_read_once(in) & I_NEW) {
1895			unlock_new_inode(in);
1896		}
1897
1898		iput(in);
1899	}
1900
1901	return err;
1902}
1903
1904void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1905{
1906	if (ctl->folio) {
1907		folio_release_kmap(ctl->folio, ctl->dentries);
1908		ctl->folio = NULL;
1909	}
1910}
1911
1912static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1913			      struct ceph_readdir_cache_control *ctl,
1914			      struct ceph_mds_request *req)
1915{
1916	struct ceph_client *cl = ceph_inode_to_client(dir);
1917	struct ceph_inode_info *ci = ceph_inode(dir);
1918	unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1919	unsigned idx = ctl->index % nsize;
1920	pgoff_t pgoff = ctl->index / nsize;
1921
1922	if (!ctl->folio || pgoff != ctl->folio->index) {
1923		ceph_readdir_cache_release(ctl);
1924		fgf_t fgf = FGP_LOCK;
1925
1926		if (idx == 0)
1927			fgf |= FGP_ACCESSED | FGP_CREAT;
1928
1929		ctl->folio = __filemap_get_folio(&dir->i_data, pgoff,
1930				fgf, mapping_gfp_mask(&dir->i_data));
1931		if (IS_ERR(ctl->folio)) {
1932			int err = PTR_ERR(ctl->folio);
1933
1934			ctl->folio = NULL;
1935			ctl->index = -1;
1936			return idx == 0 ? err : 0;
1937		}
1938		/* reading/filling the cache are serialized by
1939		 * i_rwsem, no need to use folio lock */
1940		folio_unlock(ctl->folio);
1941		ctl->dentries = kmap_local_folio(ctl->folio, 0);
1942		if (idx == 0)
1943			memset(ctl->dentries, 0, PAGE_SIZE);
1944	}
1945
1946	if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1947	    req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1948		doutc(cl, "dn %p idx %d\n", dn, ctl->index);
1949		ctl->dentries[idx] = dn;
1950		ctl->index++;
1951	} else {
1952		doutc(cl, "disable readdir cache\n");
1953		ctl->index = -1;
1954	}
1955	return 0;
1956}
1957
1958int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1959			     struct ceph_mds_session *session)
1960{
1961	struct dentry *parent = req->r_dentry;
1962	struct inode *inode = d_inode(parent);
1963	struct ceph_inode_info *ci = ceph_inode(inode);
1964	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1965	struct ceph_client *cl = session->s_mdsc->fsc->client;
1966	struct qstr dname;
1967	struct dentry *dn;
1968	struct inode *in;
1969	int err = 0, skipped = 0, ret, i;
1970	u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1971	u32 last_hash = 0;
1972	u32 fpos_offset;
1973	struct ceph_readdir_cache_control cache_ctl = {};
1974
1975	if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1976		return readdir_prepopulate_inodes_only(req, session);
1977
1978	if (rinfo->hash_order) {
1979		if (req->r_path2) {
1980			last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1981						  req->r_path2,
1982						  strlen(req->r_path2));
1983			last_hash = ceph_frag_value(last_hash);
1984		} else if (rinfo->offset_hash) {
1985			/* mds understands offset_hash */
1986			WARN_ON_ONCE(req->r_readdir_offset != 2);
1987			last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1988		}
1989	}
1990
1991	if (rinfo->dir_dir &&
1992	    le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1993		doutc(cl, "got new frag %x -> %x\n", frag,
1994			    le32_to_cpu(rinfo->dir_dir->frag));
1995		frag = le32_to_cpu(rinfo->dir_dir->frag);
1996		if (!rinfo->hash_order)
1997			req->r_readdir_offset = 2;
1998	}
1999
2000	if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
2001		doutc(cl, "%d items under SNAPDIR dn %p\n",
2002		      rinfo->dir_nr, parent);
2003	} else {
2004		doutc(cl, "%d items under dn %p\n", rinfo->dir_nr, parent);
2005		if (rinfo->dir_dir)
2006			ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
2007
2008		if (ceph_frag_is_leftmost(frag) &&
2009		    req->r_readdir_offset == 2 &&
2010		    !(rinfo->hash_order && last_hash)) {
2011			/* note dir version at start of readdir so we can
2012			 * tell if any dentries get dropped */
2013			req->r_dir_release_cnt =
2014				atomic64_read(&ci->i_release_count);
2015			req->r_dir_ordered_cnt =
2016				atomic64_read(&ci->i_ordered_count);
2017			req->r_readdir_cache_idx = 0;
2018		}
2019	}
2020
2021	cache_ctl.index = req->r_readdir_cache_idx;
2022	fpos_offset = req->r_readdir_offset;
2023
2024	/* FIXME: release caps/leases if error occurs */
2025	for (i = 0; i < rinfo->dir_nr; i++) {
2026		struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
2027		struct ceph_vino tvino;
2028
2029		dname.name = rde->name;
2030		dname.len = rde->name_len;
2031		dname.hash = full_name_hash(parent, dname.name, dname.len);
2032
2033		tvino.ino = le64_to_cpu(rde->inode.in->ino);
2034		tvino.snap = le64_to_cpu(rde->inode.in->snapid);
2035
2036		if (rinfo->hash_order) {
2037			u32 hash = ceph_frag_value(rde->raw_hash);
2038			if (hash != last_hash)
2039				fpos_offset = 2;
2040			last_hash = hash;
2041			rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
2042		} else {
2043			rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
2044		}
2045
2046retry_lookup:
2047		dn = d_lookup(parent, &dname);
2048		doutc(cl, "d_lookup on parent=%p name=%.*s got %p\n",
2049		      parent, dname.len, dname.name, dn);
2050
2051		if (!dn) {
2052			dn = d_alloc(parent, &dname);
2053			doutc(cl, "d_alloc %p '%.*s' = %p\n", parent,
2054			      dname.len, dname.name, dn);
2055			if (!dn) {
2056				doutc(cl, "d_alloc badness\n");
2057				err = -ENOMEM;
2058				goto out;
2059			}
2060			if (rde->is_nokey) {
2061				spin_lock(&dn->d_lock);
2062				dn->d_flags |= DCACHE_NOKEY_NAME;
2063				spin_unlock(&dn->d_lock);
2064			}
2065		} else if (d_really_is_positive(dn) &&
2066			   (ceph_ino(d_inode(dn)) != tvino.ino ||
2067			    ceph_snap(d_inode(dn)) != tvino.snap)) {
2068			struct ceph_dentry_info *di = ceph_dentry(dn);
2069			doutc(cl, " dn %p points to wrong inode %p\n",
2070			      dn, d_inode(dn));
2071
2072			spin_lock(&dn->d_lock);
2073			if (di->offset > 0 &&
2074			    di->lease_shared_gen ==
2075			    atomic_read(&ci->i_shared_gen)) {
2076				__ceph_dir_clear_ordered(ci);
2077				di->offset = 0;
2078			}
2079			spin_unlock(&dn->d_lock);
2080
2081			d_delete(dn);
2082			dput(dn);
2083			goto retry_lookup;
2084		}
2085
2086		/* inode */
2087		if (d_really_is_positive(dn)) {
2088			in = d_inode(dn);
2089		} else {
2090			in = ceph_get_inode(parent->d_sb, tvino, NULL);
2091			if (IS_ERR(in)) {
2092				doutc(cl, "new_inode badness\n");
2093				d_drop(dn);
2094				dput(dn);
2095				err = PTR_ERR(in);
2096				goto out;
2097			}
2098		}
2099
2100		ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
2101				      -1, &req->r_caps_reservation);
2102		if (ret < 0) {
2103			pr_err_client(cl, "badness on %p %llx.%llx\n", in,
2104				      ceph_vinop(in));
2105			if (d_really_is_negative(dn)) {
2106				if (inode_state_read_once(in) & I_NEW) {
2107					ihold(in);
2108					discard_new_inode(in);
2109				}
2110				iput(in);
2111			}
2112			d_drop(dn);
2113			err = ret;
2114			goto next_item;
2115		}
2116		if (inode_state_read_once(in) & I_NEW)
2117			unlock_new_inode(in);
2118
2119		if (d_really_is_negative(dn)) {
2120			if (ceph_security_xattr_deadlock(in)) {
2121				doutc(cl, " skip splicing dn %p to inode %p"
2122				      " (security xattr deadlock)\n", dn, in);
2123				iput(in);
2124				skipped++;
2125				goto next_item;
2126			}
2127
2128			err = splice_dentry(&dn, in);
2129			if (err < 0)
2130				goto next_item;
2131		}
2132
2133		ceph_dentry(dn)->offset = rde->offset;
2134
2135		update_dentry_lease(d_inode(parent), dn,
2136				    rde->lease, req->r_session,
2137				    req->r_request_started);
2138
2139		if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
2140			ret = fill_readdir_cache(d_inode(parent), dn,
2141						 &cache_ctl, req);
2142			if (ret < 0)
2143				err = ret;
2144		}
2145next_item:
2146		dput(dn);
2147	}
2148out:
2149	if (err == 0 && skipped == 0) {
2150		set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
2151		req->r_readdir_cache_idx = cache_ctl.index;
2152	}
2153	ceph_readdir_cache_release(&cache_ctl);
2154	doutc(cl, "done\n");
2155	return err;
2156}
2157
2158bool ceph_inode_set_size(struct inode *inode, loff_t size)
2159{
2160	struct ceph_client *cl = ceph_inode_to_client(inode);
2161	struct ceph_inode_info *ci = ceph_inode(inode);
2162	bool ret;
2163
2164	spin_lock(&ci->i_ceph_lock);
2165	doutc(cl, "set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
2166	i_size_write(inode, size);
2167	ceph_fscache_update(inode);
2168	inode->i_blocks = calc_inode_blocks(size);
2169
2170	ret = __ceph_should_report_size(ci);
2171
2172	spin_unlock(&ci->i_ceph_lock);
2173
2174	return ret;
2175}
2176
2177void ceph_queue_inode_work(struct inode *inode, int work_bit)
2178{
2179	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2180	struct ceph_client *cl = fsc->client;
2181	struct ceph_inode_info *ci = ceph_inode(inode);
2182	set_bit(work_bit, &ci->i_work_mask);
2183
2184	ihold(inode);
2185	if (queue_work(fsc->inode_wq, &ci->i_work)) {
2186		doutc(cl, "%p %llx.%llx mask=%lx\n", inode,
2187		      ceph_vinop(inode), ci->i_work_mask);
2188	} else {
2189		doutc(cl, "%p %llx.%llx already queued, mask=%lx\n",
2190		      inode, ceph_vinop(inode), ci->i_work_mask);
2191		iput(inode);
2192	}
2193}
2194
2195static void ceph_do_invalidate_pages(struct inode *inode)
2196{
2197	struct ceph_client *cl = ceph_inode_to_client(inode);
2198	struct ceph_inode_info *ci = ceph_inode(inode);
2199	u32 orig_gen;
2200	int check = 0;
2201
2202	ceph_fscache_invalidate(inode, false);
2203
2204	mutex_lock(&ci->i_truncate_mutex);
2205
2206	if (ceph_inode_is_shutdown(inode)) {
2207		pr_warn_ratelimited_client(cl,
2208			"%p %llx.%llx is shut down\n", inode,
2209			ceph_vinop(inode));
2210		mapping_set_error(inode->i_mapping, -EIO);
2211		truncate_pagecache(inode, 0);
2212		mutex_unlock(&ci->i_truncate_mutex);
2213		goto out;
2214	}
2215
2216	spin_lock(&ci->i_ceph_lock);
2217	doutc(cl, "%p %llx.%llx gen %d revoking %d\n", inode,
2218	      ceph_vinop(inode), ci->i_rdcache_gen, ci->i_rdcache_revoking);
2219	if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
2220		if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2221			check = 1;
2222		spin_unlock(&ci->i_ceph_lock);
2223		mutex_unlock(&ci->i_truncate_mutex);
2224		goto out;
2225	}
2226	orig_gen = ci->i_rdcache_gen;
2227	spin_unlock(&ci->i_ceph_lock);
2228
2229	if (invalidate_inode_pages2(inode->i_mapping) < 0) {
2230		pr_err_client(cl, "invalidate_inode_pages2 %llx.%llx failed\n",
2231			      ceph_vinop(inode));
2232	}
2233
2234	spin_lock(&ci->i_ceph_lock);
2235	if (orig_gen == ci->i_rdcache_gen &&
2236	    orig_gen == ci->i_rdcache_revoking) {
2237		doutc(cl, "%p %llx.%llx gen %d successful\n", inode,
2238		      ceph_vinop(inode), ci->i_rdcache_gen);
2239		ci->i_rdcache_revoking--;
2240		check = 1;
2241	} else {
2242		doutc(cl, "%p %llx.%llx gen %d raced, now %d revoking %d\n",
2243		      inode, ceph_vinop(inode), orig_gen, ci->i_rdcache_gen,
2244		      ci->i_rdcache_revoking);
2245		if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2246			check = 1;
2247	}
2248	spin_unlock(&ci->i_ceph_lock);
2249	mutex_unlock(&ci->i_truncate_mutex);
2250out:
2251	if (check)
2252		ceph_check_caps(ci, 0);
2253}
2254
2255/*
2256 * Make sure any pending truncation is applied before doing anything
2257 * that may depend on it.
2258 */
2259void __ceph_do_pending_vmtruncate(struct inode *inode)
2260{
2261	struct ceph_client *cl = ceph_inode_to_client(inode);
2262	struct ceph_inode_info *ci = ceph_inode(inode);
2263	u64 to;
2264	int wrbuffer_refs, finish = 0;
2265
2266	mutex_lock(&ci->i_truncate_mutex);
2267retry:
2268	spin_lock(&ci->i_ceph_lock);
2269	if (ci->i_truncate_pending == 0) {
2270		doutc(cl, "%p %llx.%llx none pending\n", inode,
2271		      ceph_vinop(inode));
2272		spin_unlock(&ci->i_ceph_lock);
2273		mutex_unlock(&ci->i_truncate_mutex);
2274		return;
2275	}
2276
2277	/*
2278	 * make sure any dirty snapped pages are flushed before we
2279	 * possibly truncate them.. so write AND block!
2280	 */
2281	if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
2282		spin_unlock(&ci->i_ceph_lock);
2283		doutc(cl, "%p %llx.%llx flushing snaps first\n", inode,
2284		      ceph_vinop(inode));
2285		filemap_write_and_wait_range(&inode->i_data, 0,
2286					     inode->i_sb->s_maxbytes);
2287		goto retry;
2288	}
2289
2290	/* there should be no reader or writer */
2291	WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
2292
2293	to = ci->i_truncate_pagecache_size;
2294	wrbuffer_refs = ci->i_wrbuffer_ref;
2295	doutc(cl, "%p %llx.%llx (%d) to %lld\n", inode, ceph_vinop(inode),
2296	      ci->i_truncate_pending, to);
2297	spin_unlock(&ci->i_ceph_lock);
2298
2299	ceph_fscache_resize(inode, to);
2300	truncate_pagecache(inode, to);
2301
2302	spin_lock(&ci->i_ceph_lock);
2303	if (to == ci->i_truncate_pagecache_size) {
2304		ci->i_truncate_pending = 0;
2305		finish = 1;
2306	}
2307	spin_unlock(&ci->i_ceph_lock);
2308	if (!finish)
2309		goto retry;
2310
2311	mutex_unlock(&ci->i_truncate_mutex);
2312
2313	if (wrbuffer_refs == 0)
2314		ceph_check_caps(ci, 0);
2315
2316	wake_up_all(&ci->i_cap_wq);
2317}
2318
2319static void ceph_inode_work(struct work_struct *work)
2320{
2321	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
2322						 i_work);
2323	struct inode *inode = &ci->netfs.inode;
2324	struct ceph_client *cl = ceph_inode_to_client(inode);
2325
2326	if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
2327		doutc(cl, "writeback %p %llx.%llx\n", inode, ceph_vinop(inode));
2328		filemap_fdatawrite(&inode->i_data);
2329	}
2330	if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2331		ceph_do_invalidate_pages(inode);
2332
2333	if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2334		__ceph_do_pending_vmtruncate(inode);
2335
2336	if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
2337		ceph_check_caps(ci, 0);
2338
2339	if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
2340		ceph_flush_snaps(ci, NULL);
2341
2342	iput(inode);
2343}
2344
2345static const char *ceph_encrypted_get_link(struct dentry *dentry,
2346					   struct inode *inode,
2347					   struct delayed_call *done)
2348{
2349	struct ceph_inode_info *ci = ceph_inode(inode);
2350
2351	if (!dentry)
2352		return ERR_PTR(-ECHILD);
2353
2354	return fscrypt_get_symlink(inode, ci->i_symlink, i_size_read(inode),
2355				   done);
2356}
2357
2358static int ceph_encrypted_symlink_getattr(struct mnt_idmap *idmap,
2359					  const struct path *path,
2360					  struct kstat *stat, u32 request_mask,
2361					  unsigned int query_flags)
2362{
2363	int ret;
2364
2365	ret = ceph_getattr(idmap, path, stat, request_mask, query_flags);
2366	if (ret)
2367		return ret;
2368	return fscrypt_symlink_getattr(path, stat);
2369}
2370
2371/*
2372 * symlinks
2373 */
2374static const struct inode_operations ceph_symlink_iops = {
2375	.get_link = simple_get_link,
2376	.setattr = ceph_setattr,
2377	.getattr = ceph_getattr,
2378	.listxattr = ceph_listxattr,
2379};
2380
2381static const struct inode_operations ceph_encrypted_symlink_iops = {
2382	.get_link = ceph_encrypted_get_link,
2383	.setattr = ceph_setattr,
2384	.getattr = ceph_encrypted_symlink_getattr,
2385	.listxattr = ceph_listxattr,
2386};
2387
2388/*
2389 * Transfer the encrypted last block to the MDS and the MDS
2390 * will help update it when truncating a smaller size.
2391 *
2392 * We don't support a PAGE_SIZE that is smaller than the
2393 * CEPH_FSCRYPT_BLOCK_SIZE.
2394 */
2395static int fill_fscrypt_truncate(struct inode *inode,
2396				 struct ceph_mds_request *req,
2397				 struct iattr *attr)
2398{
2399	struct ceph_client *cl = ceph_inode_to_client(inode);
2400	struct ceph_inode_info *ci = ceph_inode(inode);
2401	int boff = attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE;
2402	loff_t pos, orig_pos = round_down(attr->ia_size,
2403					  CEPH_FSCRYPT_BLOCK_SIZE);
2404	u64 block = orig_pos >> CEPH_FSCRYPT_BLOCK_SHIFT;
2405	struct ceph_pagelist *pagelist = NULL;
2406	struct kvec iov = {0};
2407	struct iov_iter iter;
2408	struct page *page = NULL;
2409	struct ceph_fscrypt_truncate_size_header header;
2410	int retry_op = 0;
2411	int len = CEPH_FSCRYPT_BLOCK_SIZE;
2412	loff_t i_size = i_size_read(inode);
2413	int got, ret, issued;
2414	u64 objver;
2415
2416	ret = __ceph_get_caps(inode, NULL, CEPH_CAP_FILE_RD, 0, -1, &got);
2417	if (ret < 0)
2418		return ret;
2419
2420	issued = __ceph_caps_issued(ci, NULL);
2421
2422	doutc(cl, "size %lld -> %lld got cap refs on %s, issued %s\n",
2423	      i_size, attr->ia_size, ceph_cap_string(got),
2424	      ceph_cap_string(issued));
2425
2426	/* Try to writeback the dirty pagecaches */
2427	if (issued & (CEPH_CAP_FILE_BUFFER)) {
2428		loff_t lend = orig_pos + CEPH_FSCRYPT_BLOCK_SIZE - 1;
2429
2430		ret = filemap_write_and_wait_range(inode->i_mapping,
2431						   orig_pos, lend);
2432		if (ret < 0)
2433			goto out;
2434	}
2435
2436	page = __page_cache_alloc(GFP_KERNEL);
2437	if (page == NULL) {
2438		ret = -ENOMEM;
2439		goto out;
2440	}
2441
2442	pagelist = ceph_pagelist_alloc(GFP_KERNEL);
2443	if (!pagelist) {
2444		ret = -ENOMEM;
2445		goto out;
2446	}
2447
2448	iov.iov_base = kmap_local_page(page);
2449	iov.iov_len = len;
2450	iov_iter_kvec(&iter, READ, &iov, 1, len);
2451
2452	pos = orig_pos;
2453	ret = __ceph_sync_read(inode, &pos, &iter, &retry_op, &objver);
2454	if (ret < 0)
2455		goto out;
2456
2457	/* Insert the header first */
2458	header.ver = 1;
2459	header.compat = 1;
2460	header.change_attr = cpu_to_le64(inode_peek_iversion_raw(inode));
2461
2462	/*
2463	 * Always set the block_size to CEPH_FSCRYPT_BLOCK_SIZE,
2464	 * because in MDS it may need this to do the truncate.
2465	 */
2466	header.block_size = cpu_to_le32(CEPH_FSCRYPT_BLOCK_SIZE);
2467
2468	/*
2469	 * If we hit a hole here, we should just skip filling
2470	 * the fscrypt for the request, because once the fscrypt
2471	 * is enabled, the file will be split into many blocks
2472	 * with the size of CEPH_FSCRYPT_BLOCK_SIZE, if there
2473	 * has a hole, the hole size should be multiple of block
2474	 * size.
2475	 *
2476	 * If the Rados object doesn't exist, it will be set to 0.
2477	 */
2478	if (!objver) {
2479		doutc(cl, "hit hole, ppos %lld < size %lld\n", pos, i_size);
2480
2481		header.data_len = cpu_to_le32(8 + 8 + 4);
2482		header.file_offset = 0;
2483		ret = 0;
2484	} else {
2485		header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
2486		header.file_offset = cpu_to_le64(orig_pos);
2487
2488		doutc(cl, "encrypt block boff/bsize %d/%lu\n", boff,
2489		      CEPH_FSCRYPT_BLOCK_SIZE);
2490
2491		/* truncate and zero out the extra contents for the last block */
2492		memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
2493
2494		/* encrypt the last block */
2495		ret = ceph_fscrypt_encrypt_block_inplace(inode, page,
2496						    CEPH_FSCRYPT_BLOCK_SIZE,
2497						    0, block);
2498		if (ret)
2499			goto out;
2500	}
2501
2502	/* Insert the header */
2503	ret = ceph_pagelist_append(pagelist, &header, sizeof(header));
2504	if (ret)
2505		goto out;
2506
2507	if (header.block_size) {
2508		/* Append the last block contents to pagelist */
2509		ret = ceph_pagelist_append(pagelist, iov.iov_base,
2510					   CEPH_FSCRYPT_BLOCK_SIZE);
2511		if (ret)
2512			goto out;
2513	}
2514	req->r_pagelist = pagelist;
2515out:
2516	doutc(cl, "%p %llx.%llx size dropping cap refs on %s\n", inode,
2517	      ceph_vinop(inode), ceph_cap_string(got));
2518	ceph_put_cap_refs(ci, got);
2519	if (iov.iov_base)
2520		kunmap_local(iov.iov_base);
2521	if (page)
2522		__free_pages(page, 0);
2523	if (ret && pagelist)
2524		ceph_pagelist_release(pagelist);
2525	return ret;
2526}
2527
2528int __ceph_setattr(struct mnt_idmap *idmap, struct inode *inode,
2529		   struct iattr *attr, struct ceph_iattr *cia)
2530{
2531	struct ceph_inode_info *ci = ceph_inode(inode);
2532	unsigned int ia_valid = attr->ia_valid;
2533	struct ceph_mds_request *req;
2534	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
2535	struct ceph_client *cl = ceph_inode_to_client(inode);
2536	struct ceph_cap_flush *prealloc_cf;
2537	loff_t isize = i_size_read(inode);
2538	int issued;
2539	int release = 0, dirtied = 0;
2540	int mask = 0;
2541	int err = 0;
2542	int inode_dirty_flags = 0;
2543	bool lock_snap_rwsem = false;
2544	bool fill_fscrypt;
2545	int truncate_retry = 20; /* The RMW will take around 50ms */
2546	struct dentry *dentry;
2547	char *path;
2548	bool do_sync = false;
2549
2550	dentry = d_find_alias(inode);
2551	if (!dentry) {
2552		do_sync = true;
2553	} else {
2554		struct ceph_path_info path_info = {0};
2555		path = ceph_mdsc_build_path(mdsc, dentry, &path_info, 0);
2556		if (IS_ERR(path)) {
2557			do_sync = true;
2558			err = 0;
2559		} else {
2560			err = ceph_mds_check_access(mdsc, path, MAY_WRITE);
2561		}
2562		ceph_mdsc_free_path_info(&path_info);
2563		dput(dentry);
2564
2565		/* For none EACCES cases will let the MDS do the mds auth check */
2566		if (err == -EACCES) {
2567			return err;
2568		} else if (err < 0) {
2569			do_sync = true;
2570			err = 0;
2571		}
2572	}
2573
2574retry:
2575	prealloc_cf = ceph_alloc_cap_flush();
2576	if (!prealloc_cf)
2577		return -ENOMEM;
2578
2579	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2580				       USE_AUTH_MDS);
2581	if (IS_ERR(req)) {
2582		ceph_free_cap_flush(prealloc_cf);
2583		return PTR_ERR(req);
2584	}
2585
2586	fill_fscrypt = false;
2587	spin_lock(&ci->i_ceph_lock);
2588	issued = __ceph_caps_issued(ci, NULL);
2589
2590	if (!ci->i_head_snapc &&
2591	    (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2592		lock_snap_rwsem = true;
2593		if (!down_read_trylock(&mdsc->snap_rwsem)) {
2594			spin_unlock(&ci->i_ceph_lock);
2595			down_read(&mdsc->snap_rwsem);
2596			spin_lock(&ci->i_ceph_lock);
2597			issued = __ceph_caps_issued(ci, NULL);
2598		}
2599	}
2600
2601	doutc(cl, "%p %llx.%llx issued %s\n", inode, ceph_vinop(inode),
2602	      ceph_cap_string(issued));
2603#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
2604	if (cia && cia->fscrypt_auth) {
2605		u32 len = ceph_fscrypt_auth_len(cia->fscrypt_auth);
2606
2607		if (len > sizeof(*cia->fscrypt_auth)) {
2608			err = -EINVAL;
2609			spin_unlock(&ci->i_ceph_lock);
2610			goto out;
2611		}
2612
2613		doutc(cl, "%p %llx.%llx fscrypt_auth len %u to %u)\n", inode,
2614		      ceph_vinop(inode), ci->fscrypt_auth_len, len);
2615
2616		/* It should never be re-set once set */
2617		WARN_ON_ONCE(ci->fscrypt_auth);
2618
2619		if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2620			dirtied |= CEPH_CAP_AUTH_EXCL;
2621			kfree(ci->fscrypt_auth);
2622			ci->fscrypt_auth = (u8 *)cia->fscrypt_auth;
2623			ci->fscrypt_auth_len = len;
2624		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2625			   ci->fscrypt_auth_len != len ||
2626			   memcmp(ci->fscrypt_auth, cia->fscrypt_auth, len)) {
2627			req->r_fscrypt_auth = cia->fscrypt_auth;
2628			mask |= CEPH_SETATTR_FSCRYPT_AUTH;
2629			release |= CEPH_CAP_AUTH_SHARED;
2630		}
2631		cia->fscrypt_auth = NULL;
2632	}
2633#else
2634	if (cia && cia->fscrypt_auth) {
2635		err = -EINVAL;
2636		spin_unlock(&ci->i_ceph_lock);
2637		goto out;
2638	}
2639#endif /* CONFIG_FS_ENCRYPTION */
2640
2641	if (ia_valid & ATTR_UID) {
2642		kuid_t fsuid = from_vfsuid(idmap, i_user_ns(inode), attr->ia_vfsuid);
2643
2644		doutc(cl, "%p %llx.%llx uid %d -> %d\n", inode,
2645		      ceph_vinop(inode),
2646		      from_kuid(&init_user_ns, inode->i_uid),
2647		      from_kuid(&init_user_ns, attr->ia_uid));
2648		if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2649			inode->i_uid = fsuid;
2650			dirtied |= CEPH_CAP_AUTH_EXCL;
2651		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2652			   !uid_eq(fsuid, inode->i_uid)) {
2653			req->r_args.setattr.uid = cpu_to_le32(
2654				from_kuid(&init_user_ns, fsuid));
2655			mask |= CEPH_SETATTR_UID;
2656			release |= CEPH_CAP_AUTH_SHARED;
2657		}
2658	}
2659	if (ia_valid & ATTR_GID) {
2660		kgid_t fsgid = from_vfsgid(idmap, i_user_ns(inode), attr->ia_vfsgid);
2661
2662		doutc(cl, "%p %llx.%llx gid %d -> %d\n", inode,
2663		      ceph_vinop(inode),
2664		      from_kgid(&init_user_ns, inode->i_gid),
2665		      from_kgid(&init_user_ns, attr->ia_gid));
2666		if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2667			inode->i_gid = fsgid;
2668			dirtied |= CEPH_CAP_AUTH_EXCL;
2669		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2670			   !gid_eq(fsgid, inode->i_gid)) {
2671			req->r_args.setattr.gid = cpu_to_le32(
2672				from_kgid(&init_user_ns, fsgid));
2673			mask |= CEPH_SETATTR_GID;
2674			release |= CEPH_CAP_AUTH_SHARED;
2675		}
2676	}
2677	if (ia_valid & ATTR_MODE) {
2678		doutc(cl, "%p %llx.%llx mode 0%o -> 0%o\n", inode,
2679		      ceph_vinop(inode), inode->i_mode, attr->ia_mode);
2680		if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2681			inode->i_mode = attr->ia_mode;
2682			dirtied |= CEPH_CAP_AUTH_EXCL;
2683		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2684			   attr->ia_mode != inode->i_mode) {
2685			inode->i_mode = attr->ia_mode;
2686			req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2687			mask |= CEPH_SETATTR_MODE;
2688			release |= CEPH_CAP_AUTH_SHARED;
2689		}
2690	}
2691
2692	if (ia_valid & ATTR_ATIME) {
2693		struct timespec64 atime = inode_get_atime(inode);
2694
2695		doutc(cl, "%p %llx.%llx atime %ptSp -> %ptSp\n",
2696		      inode, ceph_vinop(inode), &atime, &attr->ia_atime);
2697		if (!do_sync && (issued & CEPH_CAP_FILE_EXCL)) {
2698			ci->i_time_warp_seq++;
2699			inode_set_atime_to_ts(inode, attr->ia_atime);
2700			dirtied |= CEPH_CAP_FILE_EXCL;
2701		} else if (!do_sync && (issued & CEPH_CAP_FILE_WR) &&
2702			   timespec64_compare(&atime,
2703					      &attr->ia_atime) < 0) {
2704			inode_set_atime_to_ts(inode, attr->ia_atime);
2705			dirtied |= CEPH_CAP_FILE_WR;
2706		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2707			   !timespec64_equal(&atime, &attr->ia_atime)) {
2708			ceph_encode_timespec64(&req->r_args.setattr.atime,
2709					       &attr->ia_atime);
2710			mask |= CEPH_SETATTR_ATIME;
2711			release |= CEPH_CAP_FILE_SHARED |
2712				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2713		}
2714	}
2715	if (ia_valid & ATTR_SIZE) {
2716		doutc(cl, "%p %llx.%llx size %lld -> %lld\n", inode,
2717		      ceph_vinop(inode), isize, attr->ia_size);
2718		/*
2719		 * Only when the new size is smaller and not aligned to
2720		 * CEPH_FSCRYPT_BLOCK_SIZE will the RMW is needed.
2721		 */
2722		if (IS_ENCRYPTED(inode) && attr->ia_size < isize &&
2723		    (attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE)) {
2724			mask |= CEPH_SETATTR_SIZE;
2725			release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2726				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2727			set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2728			mask |= CEPH_SETATTR_FSCRYPT_FILE;
2729			req->r_args.setattr.size =
2730				cpu_to_le64(round_up(attr->ia_size,
2731						     CEPH_FSCRYPT_BLOCK_SIZE));
2732			req->r_args.setattr.old_size =
2733				cpu_to_le64(round_up(isize,
2734						     CEPH_FSCRYPT_BLOCK_SIZE));
2735			req->r_fscrypt_file = attr->ia_size;
2736			fill_fscrypt = true;
2737		} else if (!do_sync && (issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
2738			if (attr->ia_size > isize) {
2739				i_size_write(inode, attr->ia_size);
2740				inode->i_blocks = calc_inode_blocks(attr->ia_size);
2741				ci->i_reported_size = attr->ia_size;
2742				dirtied |= CEPH_CAP_FILE_EXCL;
2743				ia_valid |= ATTR_MTIME;
2744			}
2745		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2746			   attr->ia_size != isize) {
2747			mask |= CEPH_SETATTR_SIZE;
2748			release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2749				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2750			if (IS_ENCRYPTED(inode) && attr->ia_size) {
2751				set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2752				mask |= CEPH_SETATTR_FSCRYPT_FILE;
2753				req->r_args.setattr.size =
2754					cpu_to_le64(round_up(attr->ia_size,
2755							     CEPH_FSCRYPT_BLOCK_SIZE));
2756				req->r_args.setattr.old_size =
2757					cpu_to_le64(round_up(isize,
2758							     CEPH_FSCRYPT_BLOCK_SIZE));
2759				req->r_fscrypt_file = attr->ia_size;
2760			} else {
2761				req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2762				req->r_args.setattr.old_size = cpu_to_le64(isize);
2763				req->r_fscrypt_file = 0;
2764			}
2765		}
2766	}
2767	if (ia_valid & ATTR_MTIME) {
2768		struct timespec64 mtime = inode_get_mtime(inode);
2769
2770		doutc(cl, "%p %llx.%llx mtime %ptSp -> %ptSp\n",
2771		      inode, ceph_vinop(inode), &mtime, &attr->ia_mtime);
2772		if (!do_sync && (issued & CEPH_CAP_FILE_EXCL)) {
2773			ci->i_time_warp_seq++;
2774			inode_set_mtime_to_ts(inode, attr->ia_mtime);
2775			dirtied |= CEPH_CAP_FILE_EXCL;
2776		} else if (!do_sync && (issued & CEPH_CAP_FILE_WR) &&
2777			   timespec64_compare(&mtime, &attr->ia_mtime) < 0) {
2778			inode_set_mtime_to_ts(inode, attr->ia_mtime);
2779			dirtied |= CEPH_CAP_FILE_WR;
2780		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2781			   !timespec64_equal(&mtime, &attr->ia_mtime)) {
2782			ceph_encode_timespec64(&req->r_args.setattr.mtime,
2783					       &attr->ia_mtime);
2784			mask |= CEPH_SETATTR_MTIME;
2785			release |= CEPH_CAP_FILE_SHARED |
2786				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2787		}
2788	}
2789
2790	/* these do nothing */
2791	if (ia_valid & ATTR_CTIME) {
2792		struct timespec64 ictime = inode_get_ctime(inode);
2793		bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2794					 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2795		doutc(cl, "%p %llx.%llx ctime %ptSp -> %ptSp (%s)\n",
2796		      inode, ceph_vinop(inode), &ictime, &attr->ia_ctime,
2797		      only ? "ctime only" : "ignored");
2798		if (only) {
2799			/*
2800			 * if kernel wants to dirty ctime but nothing else,
2801			 * we need to choose a cap to dirty under, or do
2802			 * a almost-no-op setattr
2803			 */
2804			if (issued & CEPH_CAP_AUTH_EXCL)
2805				dirtied |= CEPH_CAP_AUTH_EXCL;
2806			else if (issued & CEPH_CAP_FILE_EXCL)
2807				dirtied |= CEPH_CAP_FILE_EXCL;
2808			else if (issued & CEPH_CAP_XATTR_EXCL)
2809				dirtied |= CEPH_CAP_XATTR_EXCL;
2810			else
2811				mask |= CEPH_SETATTR_CTIME;
2812		}
2813	}
2814	if (ia_valid & ATTR_FILE)
2815		doutc(cl, "%p %llx.%llx ATTR_FILE ... hrm!\n", inode,
2816		      ceph_vinop(inode));
2817
2818	if (dirtied) {
2819		inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2820							   &prealloc_cf);
2821		inode_set_ctime_to_ts(inode, attr->ia_ctime);
2822		inode_inc_iversion_raw(inode);
2823	}
2824
2825	release &= issued;
2826	spin_unlock(&ci->i_ceph_lock);
2827	if (lock_snap_rwsem) {
2828		up_read(&mdsc->snap_rwsem);
2829		lock_snap_rwsem = false;
2830	}
2831
2832	if (inode_dirty_flags)
2833		__mark_inode_dirty(inode, inode_dirty_flags);
2834
2835	if (mask) {
2836		req->r_inode = inode;
2837		ihold(inode);
2838		req->r_inode_drop = release;
2839		req->r_args.setattr.mask = cpu_to_le32(mask);
2840		req->r_num_caps = 1;
2841		req->r_stamp = attr->ia_ctime;
2842		if (fill_fscrypt) {
2843			err = fill_fscrypt_truncate(inode, req, attr);
2844			if (err)
2845				goto out;
2846		}
2847
2848		/*
2849		 * The truncate request will return -EAGAIN when the
2850		 * last block has been updated just before the MDS
2851		 * successfully gets the xlock for the FILE lock. To
2852		 * avoid corrupting the file contents we need to retry
2853		 * it.
2854		 */
2855		err = ceph_mdsc_do_request(mdsc, NULL, req);
2856		if (err == -EAGAIN && truncate_retry--) {
2857			doutc(cl, "%p %llx.%llx result=%d (%s locally, %d remote), retry it!\n",
2858			      inode, ceph_vinop(inode), err,
2859			      ceph_cap_string(dirtied), mask);
2860			ceph_mdsc_put_request(req);
2861			ceph_free_cap_flush(prealloc_cf);
2862			goto retry;
2863		}
2864	}
2865out:
2866	doutc(cl, "%p %llx.%llx result=%d (%s locally, %d remote)\n", inode,
2867	      ceph_vinop(inode), err, ceph_cap_string(dirtied), mask);
2868
2869	ceph_mdsc_put_request(req);
2870	ceph_free_cap_flush(prealloc_cf);
2871
2872	if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2873		__ceph_do_pending_vmtruncate(inode);
2874
2875	return err;
2876}
2877
2878/*
2879 * setattr
2880 */
2881int ceph_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
2882		 struct iattr *attr)
2883{
2884	struct inode *inode = d_inode(dentry);
2885	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2886	int err;
2887
2888	if (ceph_snap(inode) != CEPH_NOSNAP)
2889		return -EROFS;
2890
2891	if (ceph_inode_is_shutdown(inode))
2892		return -ESTALE;
2893
2894	err = fscrypt_prepare_setattr(dentry, attr);
2895	if (err)
2896		return err;
2897
2898	err = setattr_prepare(idmap, dentry, attr);
2899	if (err != 0)
2900		return err;
2901
2902	if ((attr->ia_valid & ATTR_SIZE) &&
2903	    attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2904		return -EFBIG;
2905
2906	if ((attr->ia_valid & ATTR_SIZE) &&
2907	    ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2908		return -EDQUOT;
2909
2910	err = __ceph_setattr(idmap, inode, attr, NULL);
2911
2912	if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2913		err = posix_acl_chmod(idmap, dentry, attr->ia_mode);
2914
2915	return err;
2916}
2917
2918int ceph_try_to_choose_auth_mds(struct inode *inode, int mask)
2919{
2920	int issued = ceph_caps_issued(ceph_inode(inode));
2921
2922	/*
2923	 * If any 'x' caps is issued we can just choose the auth MDS
2924	 * instead of the random replica MDSes. Because only when the
2925	 * Locker is in LOCK_EXEC state will the loner client could
2926	 * get the 'x' caps. And if we send the getattr requests to
2927	 * any replica MDS it must auth pin and tries to rdlock from
2928	 * the auth MDS, and then the auth MDS need to do the Locker
2929	 * state transition to LOCK_SYNC. And after that the lock state
2930	 * will change back.
2931	 *
2932	 * This cost much when doing the Locker state transition and
2933	 * usually will need to revoke caps from clients.
2934	 *
2935	 * And for the 'Xs' caps for getxattr we will also choose the
2936	 * auth MDS, because the MDS side code is buggy due to setxattr
2937	 * won't notify the replica MDSes when the values changed and
2938	 * the replica MDS will return the old values. Though we will
2939	 * fix it in MDS code, but this still makes sense for old ceph.
2940	 */
2941	if (((mask & CEPH_CAP_ANY_SHARED) && (issued & CEPH_CAP_ANY_EXCL))
2942	    || (mask & (CEPH_STAT_RSTAT | CEPH_STAT_CAP_XATTR)))
2943		return USE_AUTH_MDS;
2944	else
2945		return USE_ANY_MDS;
2946}
2947
2948/*
2949 * Verify that we have a lease on the given mask.  If not,
2950 * do a getattr against an mds.
2951 */
2952int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2953		      int mask, bool force)
2954{
2955	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
2956	struct ceph_client *cl = fsc->client;
2957	struct ceph_mds_client *mdsc = fsc->mdsc;
2958	struct ceph_mds_request *req;
2959	int mode;
2960	int err;
2961
2962	if (ceph_snap(inode) == CEPH_SNAPDIR) {
2963		doutc(cl, "inode %p %llx.%llx SNAPDIR\n", inode,
2964		      ceph_vinop(inode));
2965		return 0;
2966	}
2967
2968	doutc(cl, "inode %p %llx.%llx mask %s mode 0%o\n", inode,
2969	      ceph_vinop(inode), ceph_cap_string(mask), inode->i_mode);
2970	if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2971			return 0;
2972
2973	mode = ceph_try_to_choose_auth_mds(inode, mask);
2974	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2975	if (IS_ERR(req))
2976		return PTR_ERR(req);
2977	req->r_inode = inode;
2978	ihold(inode);
2979	req->r_num_caps = 1;
2980	req->r_args.getattr.mask = cpu_to_le32(mask);
2981	req->r_locked_page = locked_page;
2982	err = ceph_mdsc_do_request(mdsc, NULL, req);
2983	if (locked_page && err == 0) {
2984		u64 inline_version = req->r_reply_info.targeti.inline_version;
2985		if (inline_version == 0) {
2986			/* the reply is supposed to contain inline data */
2987			err = -EINVAL;
2988		} else if (inline_version == CEPH_INLINE_NONE ||
2989			   inline_version == 1) {
2990			err = -ENODATA;
2991		} else {
2992			err = req->r_reply_info.targeti.inline_len;
2993		}
2994	}
2995	ceph_mdsc_put_request(req);
2996	doutc(cl, "result=%d\n", err);
2997	return err;
2998}
2999
3000int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
3001		      size_t size)
3002{
3003	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
3004	struct ceph_client *cl = fsc->client;
3005	struct ceph_mds_client *mdsc = fsc->mdsc;
3006	struct ceph_mds_request *req;
3007	int mode = USE_AUTH_MDS;
3008	int err;
3009	char *xattr_value;
3010	size_t xattr_value_len;
3011
3012	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode);
3013	if (IS_ERR(req)) {
3014		err = -ENOMEM;
3015		goto out;
3016	}
3017
3018	req->r_feature_needed = CEPHFS_FEATURE_OP_GETVXATTR;
3019	req->r_path2 = kstrdup(name, GFP_NOFS);
3020	if (!req->r_path2) {
3021		err = -ENOMEM;
3022		goto put;
3023	}
3024
3025	ihold(inode);
3026	req->r_inode = inode;
3027	err = ceph_mdsc_do_request(mdsc, NULL, req);
3028	if (err < 0)
3029		goto put;
3030
3031	xattr_value = req->r_reply_info.xattr_info.xattr_value;
3032	xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len;
3033
3034	doutc(cl, "xattr_value_len:%zu, size:%zu\n", xattr_value_len, size);
3035
3036	err = (int)xattr_value_len;
3037	if (size == 0)
3038		goto put;
3039
3040	if (xattr_value_len > size) {
3041		err = -ERANGE;
3042		goto put;
3043	}
3044
3045	memcpy(value, xattr_value, xattr_value_len);
3046put:
3047	ceph_mdsc_put_request(req);
3048out:
3049	doutc(cl, "result=%d\n", err);
3050	return err;
3051}
3052
3053
3054/*
3055 * Check inode permissions.  We verify we have a valid value for
3056 * the AUTH cap, then call the generic handler.
3057 */
3058int ceph_permission(struct mnt_idmap *idmap, struct inode *inode,
3059		    int mask)
3060{
3061	int err;
3062
3063	if (mask & MAY_NOT_BLOCK)
3064		return -ECHILD;
3065
3066	err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
3067
3068	if (!err)
3069		err = generic_permission(idmap, inode, mask);
3070	return err;
3071}
3072
3073/* Craft a mask of needed caps given a set of requested statx attrs. */
3074static int statx_to_caps(u32 want, umode_t mode)
3075{
3076	int mask = 0;
3077
3078	if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME|STATX_CHANGE_COOKIE))
3079		mask |= CEPH_CAP_AUTH_SHARED;
3080
3081	if (want & (STATX_NLINK|STATX_CTIME|STATX_CHANGE_COOKIE)) {
3082		/*
3083		 * The link count for directories depends on inode->i_subdirs,
3084		 * and that is only updated when Fs caps are held.
3085		 */
3086		if (S_ISDIR(mode))
3087			mask |= CEPH_CAP_FILE_SHARED;
3088		else
3089			mask |= CEPH_CAP_LINK_SHARED;
3090	}
3091
3092	if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|STATX_BLOCKS|STATX_CHANGE_COOKIE))
3093		mask |= CEPH_CAP_FILE_SHARED;
3094
3095	if (want & (STATX_CTIME|STATX_CHANGE_COOKIE))
3096		mask |= CEPH_CAP_XATTR_SHARED;
3097
3098	return mask;
3099}
3100
3101/*
3102 * Get all the attributes. If we have sufficient caps for the requested attrs,
3103 * then we can avoid talking to the MDS at all.
3104 */
3105int ceph_getattr(struct mnt_idmap *idmap, const struct path *path,
3106		 struct kstat *stat, u32 request_mask, unsigned int flags)
3107{
3108	struct inode *inode = d_inode(path->dentry);
3109	struct super_block *sb = inode->i_sb;
3110	struct ceph_inode_info *ci = ceph_inode(inode);
3111	u32 valid_mask = STATX_BASIC_STATS;
3112	int err = 0;
3113
3114	if (ceph_inode_is_shutdown(inode))
3115		return -ESTALE;
3116
3117	/* Skip the getattr altogether if we're asked not to sync */
3118	if ((flags & AT_STATX_SYNC_TYPE) != AT_STATX_DONT_SYNC) {
3119		err = ceph_do_getattr(inode,
3120				statx_to_caps(request_mask, inode->i_mode),
3121				flags & AT_STATX_FORCE_SYNC);
3122		if (err)
3123			return err;
3124	}
3125
3126	generic_fillattr(idmap, request_mask, inode, stat);
3127	stat->ino = ceph_present_inode(inode);
3128
3129	/*
3130	 * btime on newly-allocated inodes is 0, so if this is still set to
3131	 * that, then assume that it's not valid.
3132	 */
3133	if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
3134		stat->btime = ci->i_btime;
3135		valid_mask |= STATX_BTIME;
3136	}
3137
3138	if (request_mask & STATX_CHANGE_COOKIE) {
3139		stat->change_cookie = inode_peek_iversion_raw(inode);
3140		valid_mask |= STATX_CHANGE_COOKIE;
3141	}
3142
3143	if (ceph_snap(inode) == CEPH_NOSNAP)
3144		stat->dev = sb->s_dev;
3145	else
3146		stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
3147
3148	if (S_ISDIR(inode->i_mode)) {
3149		if (ceph_test_mount_opt(ceph_sb_to_fs_client(sb), RBYTES)) {
3150			stat->size = ci->i_rbytes;
3151		} else if (ceph_snap(inode) == CEPH_SNAPDIR) {
3152			struct ceph_inode_info *pci;
3153			struct ceph_snap_realm *realm;
3154			struct inode *parent;
3155
3156			parent = ceph_lookup_inode(sb, ceph_ino(inode));
3157			if (IS_ERR(parent))
3158				return PTR_ERR(parent);
3159
3160			pci = ceph_inode(parent);
3161			spin_lock(&pci->i_ceph_lock);
3162			realm = pci->i_snap_realm;
3163			if (realm)
3164				stat->size = realm->num_snaps;
3165			else
3166				stat->size = 0;
3167			spin_unlock(&pci->i_ceph_lock);
3168			iput(parent);
3169		} else {
3170			stat->size = ci->i_files + ci->i_subdirs;
3171		}
3172		stat->blocks = 0;
3173		stat->blksize = 65536;
3174		/*
3175		 * Some applications rely on the number of st_nlink
3176		 * value on directories to be either 0 (if unlinked)
3177		 * or 2 + number of subdirectories.
3178		 */
3179		if (stat->nlink == 1)
3180			/* '.' + '..' + subdirs */
3181			stat->nlink = 1 + 1 + ci->i_subdirs;
3182	}
3183
3184	stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
3185	if (IS_ENCRYPTED(inode))
3186		stat->attributes |= STATX_ATTR_ENCRYPTED;
3187	stat->attributes_mask |= (STATX_ATTR_CHANGE_MONOTONIC |
3188				  STATX_ATTR_ENCRYPTED);
3189
3190	stat->result_mask = request_mask & valid_mask;
3191	return err;
3192}
3193
3194void ceph_inode_shutdown(struct inode *inode)
3195{
3196	struct ceph_inode_info *ci = ceph_inode(inode);
3197	struct rb_node *p;
3198	int iputs = 0;
3199	bool invalidate = false;
3200
3201	spin_lock(&ci->i_ceph_lock);
3202	ci->i_ceph_flags |= CEPH_I_SHUTDOWN;
3203	p = rb_first(&ci->i_caps);
3204	while (p) {
3205		struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
3206
3207		p = rb_next(p);
3208		iputs += ceph_purge_inode_cap(inode, cap, &invalidate);
3209	}
3210	spin_unlock(&ci->i_ceph_lock);
3211
3212	if (invalidate)
3213		ceph_queue_invalidate(inode);
3214	while (iputs--)
3215		iput(inode);
3216}