Linux Audio

Check our new training course

Loading...
v3.15
   1#include <linux/ceph/ceph_debug.h>
   2
   3#include <linux/spinlock.h>
   4#include <linux/fs_struct.h>
   5#include <linux/namei.h>
   6#include <linux/slab.h>
   7#include <linux/sched.h>
   8
   9#include "super.h"
  10#include "mds_client.h"
  11
  12/*
  13 * Directory operations: readdir, lookup, create, link, unlink,
  14 * rename, etc.
  15 */
  16
  17/*
  18 * Ceph MDS operations are specified in terms of a base ino and
  19 * relative path.  Thus, the client can specify an operation on a
  20 * specific inode (e.g., a getattr due to fstat(2)), or as a path
  21 * relative to, say, the root directory.
  22 *
  23 * Normally, we limit ourselves to strict inode ops (no path component)
  24 * or dentry operations (a single path component relative to an ino).  The
  25 * exception to this is open_root_dentry(), which will open the mount
  26 * point by name.
  27 */
  28
  29const struct inode_operations ceph_dir_iops;
  30const struct file_operations ceph_dir_fops;
  31const struct dentry_operations ceph_dentry_ops;
  32
  33/*
  34 * Initialize ceph dentry state.
  35 */
  36int ceph_init_dentry(struct dentry *dentry)
  37{
  38	struct ceph_dentry_info *di;
  39
  40	if (dentry->d_fsdata)
  41		return 0;
  42
  43	di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
  44	if (!di)
  45		return -ENOMEM;          /* oh well */
  46
  47	spin_lock(&dentry->d_lock);
  48	if (dentry->d_fsdata) {
  49		/* lost a race */
  50		kmem_cache_free(ceph_dentry_cachep, di);
  51		goto out_unlock;
  52	}
  53
  54	if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
 
  55		d_set_d_op(dentry, &ceph_dentry_ops);
  56	else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
  57		d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
  58	else
  59		d_set_d_op(dentry, &ceph_snap_dentry_ops);
  60
  61	di->dentry = dentry;
  62	di->lease_session = NULL;
  63	dentry->d_time = jiffies;
  64	/* avoid reordering d_fsdata setup so that the check above is safe */
  65	smp_mb();
  66	dentry->d_fsdata = di;
  67	ceph_dentry_lru_add(dentry);
  68out_unlock:
  69	spin_unlock(&dentry->d_lock);
  70	return 0;
  71}
  72
  73struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
  74{
  75	struct inode *inode = NULL;
  76
  77	if (!dentry)
  78		return NULL;
  79
  80	spin_lock(&dentry->d_lock);
  81	if (!IS_ROOT(dentry)) {
  82		inode = dentry->d_parent->d_inode;
  83		ihold(inode);
  84	}
  85	spin_unlock(&dentry->d_lock);
  86	return inode;
  87}
  88
  89
  90/*
  91 * for readdir, we encode the directory frag and offset within that
  92 * frag into f_pos.
  93 */
  94static unsigned fpos_frag(loff_t p)
  95{
  96	return p >> 32;
  97}
  98static unsigned fpos_off(loff_t p)
  99{
 100	return p & 0xffffffff;
 101}
 102
 103static int fpos_cmp(loff_t l, loff_t r)
 104{
 105	int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
 106	if (v)
 107		return v;
 108	return (int)(fpos_off(l) - fpos_off(r));
 109}
 110
 111/*
 112 * When possible, we try to satisfy a readdir by peeking at the
 113 * dcache.  We make this work by carefully ordering dentries on
 114 * d_u.d_child when we initially get results back from the MDS, and
 115 * falling back to a "normal" sync readdir if any dentries in the dir
 116 * are dropped.
 117 *
 118 * Complete dir indicates that we have all dentries in the dir.  It is
 119 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
 120 * the MDS if/when the directory is modified).
 121 */
 122static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
 123			    u32 shared_gen)
 124{
 125	struct ceph_file_info *fi = file->private_data;
 126	struct dentry *parent = file->f_dentry;
 127	struct inode *dir = parent->d_inode;
 128	struct list_head *p;
 129	struct dentry *dentry, *last;
 130	struct ceph_dentry_info *di;
 131	int err = 0;
 132
 133	/* claim ref on last dentry we returned */
 134	last = fi->dentry;
 135	fi->dentry = NULL;
 136
 137	dout("__dcache_readdir %p v%u at %llu (last %p)\n",
 138	     dir, shared_gen, ctx->pos, last);
 139
 140	spin_lock(&parent->d_lock);
 141
 142	/* start at beginning? */
 143	if (ctx->pos == 2 || last == NULL ||
 144	    fpos_cmp(ctx->pos, ceph_dentry(last)->offset) < 0) {
 145		if (list_empty(&parent->d_subdirs))
 146			goto out_unlock;
 147		p = parent->d_subdirs.prev;
 148		dout(" initial p %p/%p\n", p->prev, p->next);
 149	} else {
 150		p = last->d_u.d_child.prev;
 151	}
 152
 153more:
 154	dentry = list_entry(p, struct dentry, d_u.d_child);
 155	di = ceph_dentry(dentry);
 156	while (1) {
 157		dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
 158		     d_unhashed(dentry) ? "!hashed" : "hashed",
 159		     parent->d_subdirs.prev, parent->d_subdirs.next);
 160		if (p == &parent->d_subdirs) {
 161			fi->flags |= CEPH_F_ATEND;
 162			goto out_unlock;
 163		}
 164		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 165		if (di->lease_shared_gen == shared_gen &&
 166		    !d_unhashed(dentry) && dentry->d_inode &&
 167		    ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
 168		    ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
 169		    fpos_cmp(ctx->pos, di->offset) <= 0)
 170			break;
 171		dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
 172		     dentry->d_name.len, dentry->d_name.name, di->offset,
 173		     ctx->pos, d_unhashed(dentry) ? " unhashed" : "",
 174		     !dentry->d_inode ? " null" : "");
 175		spin_unlock(&dentry->d_lock);
 176		p = p->prev;
 177		dentry = list_entry(p, struct dentry, d_u.d_child);
 178		di = ceph_dentry(dentry);
 179	}
 180
 181	dget_dlock(dentry);
 182	spin_unlock(&dentry->d_lock);
 183	spin_unlock(&parent->d_lock);
 184
 185	/* make sure a dentry wasn't dropped while we didn't have parent lock */
 186	if (!ceph_dir_is_complete(dir)) {
 187		dout(" lost dir complete on %p; falling back to mds\n", dir);
 188		dput(dentry);
 189		err = -EAGAIN;
 190		goto out;
 191	}
 192
 193	dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, ctx->pos,
 194	     dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
 195	if (!dir_emit(ctx, dentry->d_name.name,
 196		      dentry->d_name.len,
 
 197		      ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
 198		      dentry->d_inode->i_mode >> 12)) {
 199		if (last) {
 
 
 200			/* remember our position */
 201			fi->dentry = last;
 202			fi->next_offset = fpos_off(di->offset);
 
 
 203		}
 204		dput(dentry);
 205		return 0;
 206	}
 
 
 
 
 207
 208	ctx->pos = di->offset + 1;
 209
 210	if (last)
 211		dput(last);
 212	last = dentry;
 
 
 
 213
 214	spin_lock(&parent->d_lock);
 215	p = p->prev;	/* advance to next dentry */
 216	goto more;
 217
 218out_unlock:
 219	spin_unlock(&parent->d_lock);
 220out:
 221	if (last)
 222		dput(last);
 223	return err;
 224}
 225
 226/*
 227 * make note of the last dentry we read, so we can
 228 * continue at the same lexicographical point,
 229 * regardless of what dir changes take place on the
 230 * server.
 231 */
 232static int note_last_dentry(struct ceph_file_info *fi, const char *name,
 233			    int len)
 234{
 235	kfree(fi->last_name);
 236	fi->last_name = kmalloc(len+1, GFP_NOFS);
 237	if (!fi->last_name)
 238		return -ENOMEM;
 239	memcpy(fi->last_name, name, len);
 240	fi->last_name[len] = 0;
 241	dout("note_last_dentry '%s'\n", fi->last_name);
 242	return 0;
 243}
 244
 245static int ceph_readdir(struct file *file, struct dir_context *ctx)
 246{
 247	struct ceph_file_info *fi = file->private_data;
 248	struct inode *inode = file_inode(file);
 249	struct ceph_inode_info *ci = ceph_inode(inode);
 250	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 251	struct ceph_mds_client *mdsc = fsc->mdsc;
 252	unsigned frag = fpos_frag(ctx->pos);
 253	int off = fpos_off(ctx->pos);
 254	int err;
 255	u32 ftype;
 256	struct ceph_mds_reply_info_parsed *rinfo;
 
 
 257
 258	dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
 259	if (fi->flags & CEPH_F_ATEND)
 260		return 0;
 261
 262	/* always start with . and .. */
 263	if (ctx->pos == 0) {
 264		/* note dir version at start of readdir so we can tell
 265		 * if any dentries get dropped */
 266		fi->dir_release_count = atomic_read(&ci->i_release_count);
 267
 268		dout("readdir off 0 -> '.'\n");
 269		if (!dir_emit(ctx, ".", 1, 
 270			    ceph_translate_ino(inode->i_sb, inode->i_ino),
 271			    inode->i_mode >> 12))
 272			return 0;
 273		ctx->pos = 1;
 274		off = 1;
 275	}
 276	if (ctx->pos == 1) {
 277		ino_t ino = parent_ino(file->f_dentry);
 278		dout("readdir off 1 -> '..'\n");
 279		if (!dir_emit(ctx, "..", 2,
 280			    ceph_translate_ino(inode->i_sb, ino),
 281			    inode->i_mode >> 12))
 282			return 0;
 283		ctx->pos = 2;
 284		off = 2;
 285	}
 286
 287	/* can we use the dcache? */
 288	spin_lock(&ci->i_ceph_lock);
 289	if ((ctx->pos == 2 || fi->dentry) &&
 290	    !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
 291	    ceph_snap(inode) != CEPH_SNAPDIR &&
 292	    __ceph_dir_is_complete(ci) &&
 293	    __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
 294		u32 shared_gen = ci->i_shared_gen;
 295		spin_unlock(&ci->i_ceph_lock);
 296		err = __dcache_readdir(file, ctx, shared_gen);
 297		if (err != -EAGAIN)
 298			return err;
 299		frag = fpos_frag(ctx->pos);
 300		off = fpos_off(ctx->pos);
 301	} else {
 302		spin_unlock(&ci->i_ceph_lock);
 303	}
 304	if (fi->dentry) {
 305		err = note_last_dentry(fi, fi->dentry->d_name.name,
 306				       fi->dentry->d_name.len);
 307		if (err)
 308			return err;
 309		dput(fi->dentry);
 310		fi->dentry = NULL;
 311	}
 312
 313	/* proceed with a normal readdir */
 314
 315more:
 316	/* do we have the correct frag content buffered? */
 317	if (fi->frag != frag || fi->last_readdir == NULL) {
 318		struct ceph_mds_request *req;
 319		int op = ceph_snap(inode) == CEPH_SNAPDIR ?
 320			CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
 321
 322		/* discard old result, if any */
 323		if (fi->last_readdir) {
 324			ceph_mdsc_put_request(fi->last_readdir);
 325			fi->last_readdir = NULL;
 326		}
 327
 
 
 
 328		dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
 329		     ceph_vinop(inode), frag, fi->last_name);
 330		req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 331		if (IS_ERR(req))
 332			return PTR_ERR(req);
 333		err = ceph_alloc_readdir_reply_buffer(req, inode);
 334		if (err) {
 335			ceph_mdsc_put_request(req);
 336			return err;
 337		}
 338		req->r_inode = inode;
 339		ihold(inode);
 340		req->r_dentry = dget(file->f_dentry);
 341		/* hints to request -> mds selection code */
 342		req->r_direct_mode = USE_AUTH_MDS;
 343		req->r_direct_hash = ceph_frag_value(frag);
 344		req->r_direct_is_hash = true;
 345		req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
 346		req->r_readdir_offset = fi->next_offset;
 347		req->r_args.readdir.frag = cpu_to_le32(frag);
 
 
 
 348		err = ceph_mdsc_do_request(mdsc, NULL, req);
 349		if (err < 0) {
 350			ceph_mdsc_put_request(req);
 351			return err;
 352		}
 353		dout("readdir got and parsed readdir result=%d"
 354		     " on frag %x, end=%d, complete=%d\n", err, frag,
 355		     (int)req->r_reply_info.dir_end,
 356		     (int)req->r_reply_info.dir_complete);
 357
 358		if (!req->r_did_prepopulate) {
 359			dout("readdir !did_prepopulate");
 360			/* preclude from marking dir complete */
 361			fi->dir_release_count--;
 362		}
 363
 364		/* note next offset and last dentry name */
 365		rinfo = &req->r_reply_info;
 366		if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
 367			frag = le32_to_cpu(rinfo->dir_dir->frag);
 368			if (ceph_frag_is_leftmost(frag))
 369				fi->next_offset = 2;
 370			else
 371				fi->next_offset = 0;
 372			off = fi->next_offset;
 373		}
 374		fi->frag = frag;
 375		fi->offset = fi->next_offset;
 376		fi->last_readdir = req;
 377
 378		if (req->r_reply_info.dir_end) {
 379			kfree(fi->last_name);
 380			fi->last_name = NULL;
 381			if (ceph_frag_is_rightmost(frag))
 382				fi->next_offset = 2;
 383			else
 384				fi->next_offset = 0;
 385		} else {
 
 386			err = note_last_dentry(fi,
 387				       rinfo->dir_dname[rinfo->dir_nr-1],
 388				       rinfo->dir_dname_len[rinfo->dir_nr-1]);
 389			if (err)
 390				return err;
 391			fi->next_offset += rinfo->dir_nr;
 392		}
 393	}
 394
 395	rinfo = &fi->last_readdir->r_reply_info;
 396	dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
 397	     rinfo->dir_nr, off, fi->offset);
 398
 399	ctx->pos = ceph_make_fpos(frag, off);
 400	while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
 
 401		struct ceph_mds_reply_inode *in =
 402			rinfo->dir_in[off - fi->offset].in;
 403		struct ceph_vino vino;
 404		ino_t ino;
 405
 406		dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
 407		     off, off - fi->offset, rinfo->dir_nr, ctx->pos,
 408		     rinfo->dir_dname_len[off - fi->offset],
 409		     rinfo->dir_dname[off - fi->offset], in);
 410		BUG_ON(!in);
 411		ftype = le32_to_cpu(in->mode) >> 12;
 412		vino.ino = le64_to_cpu(in->ino);
 413		vino.snap = le64_to_cpu(in->snapid);
 414		ino = ceph_vino_to_ino(vino);
 415		if (!dir_emit(ctx,
 416			    rinfo->dir_dname[off - fi->offset],
 417			    rinfo->dir_dname_len[off - fi->offset],
 418			    ceph_translate_ino(inode->i_sb, ino), ftype)) {
 
 419			dout("filldir stopping us...\n");
 420			return 0;
 421		}
 422		off++;
 423		ctx->pos++;
 424	}
 425
 426	if (fi->last_name) {
 427		ceph_mdsc_put_request(fi->last_readdir);
 428		fi->last_readdir = NULL;
 429		goto more;
 430	}
 431
 432	/* more frags? */
 433	if (!ceph_frag_is_rightmost(frag)) {
 434		frag = ceph_frag_next(frag);
 435		off = 0;
 436		ctx->pos = ceph_make_fpos(frag, off);
 437		dout("readdir next frag is %x\n", frag);
 438		goto more;
 439	}
 440	fi->flags |= CEPH_F_ATEND;
 441
 442	/*
 443	 * if dir_release_count still matches the dir, no dentries
 444	 * were released during the whole readdir, and we should have
 445	 * the complete dir contents in our cache.
 446	 */
 447	spin_lock(&ci->i_ceph_lock);
 448	if (atomic_read(&ci->i_release_count) == fi->dir_release_count) {
 449		dout(" marking %p complete\n", inode);
 450		__ceph_dir_set_complete(ci, fi->dir_release_count);
 451	}
 452	spin_unlock(&ci->i_ceph_lock);
 453
 454	dout("readdir %p file %p done.\n", inode, file);
 455	return 0;
 456}
 457
 458static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
 459{
 460	if (fi->last_readdir) {
 461		ceph_mdsc_put_request(fi->last_readdir);
 462		fi->last_readdir = NULL;
 463	}
 464	kfree(fi->last_name);
 465	fi->last_name = NULL;
 466	if (ceph_frag_is_leftmost(frag))
 467		fi->next_offset = 2;  /* compensate for . and .. */
 468	else
 469		fi->next_offset = 0;
 470	if (fi->dentry) {
 471		dput(fi->dentry);
 472		fi->dentry = NULL;
 473	}
 474	fi->flags &= ~CEPH_F_ATEND;
 475}
 476
 477static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
 478{
 479	struct ceph_file_info *fi = file->private_data;
 480	struct inode *inode = file->f_mapping->host;
 481	loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
 482	loff_t retval;
 483
 484	mutex_lock(&inode->i_mutex);
 485	retval = -EINVAL;
 486	switch (whence) {
 487	case SEEK_END:
 488		offset += inode->i_size + 2;   /* FIXME */
 489		break;
 490	case SEEK_CUR:
 491		offset += file->f_pos;
 492	case SEEK_SET:
 493		break;
 494	default:
 495		goto out;
 496	}
 497
 498	if (offset >= 0) {
 499		if (offset != file->f_pos) {
 500			file->f_pos = offset;
 501			file->f_version = 0;
 502			fi->flags &= ~CEPH_F_ATEND;
 503		}
 504		retval = offset;
 505
 506		/*
 507		 * discard buffered readdir content on seekdir(0), or
 508		 * seek to new frag, or seek prior to current chunk.
 509		 */
 510		if (offset == 0 ||
 511		    fpos_frag(offset) != fi->frag ||
 512		    fpos_off(offset) < fi->offset) {
 513			dout("dir_llseek dropping %p content\n", file);
 514			reset_readdir(fi, fpos_frag(offset));
 515		}
 516
 517		/* bump dir_release_count if we did a forward seek */
 518		if (fpos_cmp(offset, old_offset) > 0)
 519			fi->dir_release_count--;
 520	}
 521out:
 522	mutex_unlock(&inode->i_mutex);
 523	return retval;
 524}
 525
 526/*
 527 * Handle lookups for the hidden .snap directory.
 528 */
 529int ceph_handle_snapdir(struct ceph_mds_request *req,
 530			struct dentry *dentry, int err)
 531{
 532	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
 533	struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
 534
 535	/* .snap dir? */
 536	if (err == -ENOENT &&
 537	    ceph_snap(parent) == CEPH_NOSNAP &&
 538	    strcmp(dentry->d_name.name,
 539		   fsc->mount_options->snapdir_name) == 0) {
 540		struct inode *inode = ceph_get_snapdir(parent);
 541		dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
 542		     dentry, dentry->d_name.len, dentry->d_name.name, inode);
 543		BUG_ON(!d_unhashed(dentry));
 544		d_add(dentry, inode);
 545		err = 0;
 546	}
 547	return err;
 548}
 549
 550/*
 551 * Figure out final result of a lookup/open request.
 552 *
 553 * Mainly, make sure we return the final req->r_dentry (if it already
 554 * existed) in place of the original VFS-provided dentry when they
 555 * differ.
 556 *
 557 * Gracefully handle the case where the MDS replies with -ENOENT and
 558 * no trace (which it may do, at its discretion, e.g., if it doesn't
 559 * care to issue a lease on the negative dentry).
 560 */
 561struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
 562				  struct dentry *dentry, int err)
 563{
 564	if (err == -ENOENT) {
 565		/* no trace? */
 566		err = 0;
 567		if (!req->r_reply_info.head->is_dentry) {
 568			dout("ENOENT and no trace, dentry %p inode %p\n",
 569			     dentry, dentry->d_inode);
 570			if (dentry->d_inode) {
 571				d_drop(dentry);
 572				err = -ENOENT;
 573			} else {
 574				d_add(dentry, NULL);
 575			}
 576		}
 577	}
 578	if (err)
 579		dentry = ERR_PTR(err);
 580	else if (dentry != req->r_dentry)
 581		dentry = dget(req->r_dentry);   /* we got spliced */
 582	else
 583		dentry = NULL;
 584	return dentry;
 585}
 586
 587static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
 588{
 589	return ceph_ino(inode) == CEPH_INO_ROOT &&
 590		strncmp(dentry->d_name.name, ".ceph", 5) == 0;
 591}
 592
 593/*
 594 * Look up a single dir entry.  If there is a lookup intent, inform
 595 * the MDS so that it gets our 'caps wanted' value in a single op.
 596 */
 597static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
 598				  unsigned int flags)
 599{
 600	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 601	struct ceph_mds_client *mdsc = fsc->mdsc;
 602	struct ceph_mds_request *req;
 603	int op;
 604	int err;
 605
 606	dout("lookup %p dentry %p '%.*s'\n",
 607	     dir, dentry, dentry->d_name.len, dentry->d_name.name);
 608
 609	if (dentry->d_name.len > NAME_MAX)
 610		return ERR_PTR(-ENAMETOOLONG);
 611
 612	err = ceph_init_dentry(dentry);
 613	if (err < 0)
 614		return ERR_PTR(err);
 615
 
 
 
 
 
 
 
 
 616	/* can we conclude ENOENT locally? */
 617	if (dentry->d_inode == NULL) {
 618		struct ceph_inode_info *ci = ceph_inode(dir);
 619		struct ceph_dentry_info *di = ceph_dentry(dentry);
 620
 621		spin_lock(&ci->i_ceph_lock);
 622		dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
 623		if (strncmp(dentry->d_name.name,
 624			    fsc->mount_options->snapdir_name,
 625			    dentry->d_name.len) &&
 626		    !is_root_ceph_dentry(dir, dentry) &&
 627		    __ceph_dir_is_complete(ci) &&
 628		    (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
 629			spin_unlock(&ci->i_ceph_lock);
 630			dout(" dir %p complete, -ENOENT\n", dir);
 631			d_add(dentry, NULL);
 632			di->lease_shared_gen = ci->i_shared_gen;
 633			return NULL;
 634		}
 635		spin_unlock(&ci->i_ceph_lock);
 636	}
 637
 638	op = ceph_snap(dir) == CEPH_SNAPDIR ?
 639		CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
 640	req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
 641	if (IS_ERR(req))
 642		return ERR_CAST(req);
 643	req->r_dentry = dget(dentry);
 644	req->r_num_caps = 2;
 645	/* we only need inode linkage */
 646	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 647	req->r_locked_dir = dir;
 648	err = ceph_mdsc_do_request(mdsc, NULL, req);
 649	err = ceph_handle_snapdir(req, dentry, err);
 650	dentry = ceph_finish_lookup(req, dentry, err);
 651	ceph_mdsc_put_request(req);  /* will dput(dentry) */
 652	dout("lookup result=%p\n", dentry);
 653	return dentry;
 654}
 655
 656/*
 657 * If we do a create but get no trace back from the MDS, follow up with
 658 * a lookup (the VFS expects us to link up the provided dentry).
 659 */
 660int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
 661{
 662	struct dentry *result = ceph_lookup(dir, dentry, 0);
 663
 664	if (result && !IS_ERR(result)) {
 665		/*
 666		 * We created the item, then did a lookup, and found
 667		 * it was already linked to another inode we already
 668		 * had in our cache (and thus got spliced).  Link our
 669		 * dentry to that inode, but don't hash it, just in
 670		 * case the VFS wants to dereference it.
 671		 */
 672		BUG_ON(!result->d_inode);
 673		d_instantiate(dentry, result->d_inode);
 674		return 0;
 675	}
 676	return PTR_ERR(result);
 677}
 678
 679static int ceph_mknod(struct inode *dir, struct dentry *dentry,
 680		      umode_t mode, dev_t rdev)
 681{
 682	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 683	struct ceph_mds_client *mdsc = fsc->mdsc;
 684	struct ceph_mds_request *req;
 685	int err;
 686
 687	if (ceph_snap(dir) != CEPH_NOSNAP)
 688		return -EROFS;
 689
 690	dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
 691	     dir, dentry, mode, rdev);
 692	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
 693	if (IS_ERR(req)) {
 694		d_drop(dentry);
 695		return PTR_ERR(req);
 696	}
 697	req->r_dentry = dget(dentry);
 698	req->r_num_caps = 2;
 699	req->r_locked_dir = dir;
 700	req->r_args.mknod.mode = cpu_to_le32(mode);
 701	req->r_args.mknod.rdev = cpu_to_le32(rdev);
 702	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 703	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 704	err = ceph_mdsc_do_request(mdsc, dir, req);
 705	if (!err && !req->r_reply_info.head->is_dentry)
 706		err = ceph_handle_notrace_create(dir, dentry);
 707	ceph_mdsc_put_request(req);
 708
 709	if (!err)
 710		ceph_init_acl(dentry, dentry->d_inode, dir);
 711	else
 712		d_drop(dentry);
 713	return err;
 714}
 715
 716static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 717		       bool excl)
 718{
 719	return ceph_mknod(dir, dentry, mode, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 720}
 721
 722static int ceph_symlink(struct inode *dir, struct dentry *dentry,
 723			    const char *dest)
 724{
 725	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 726	struct ceph_mds_client *mdsc = fsc->mdsc;
 727	struct ceph_mds_request *req;
 728	int err;
 729
 730	if (ceph_snap(dir) != CEPH_NOSNAP)
 731		return -EROFS;
 732
 733	dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
 734	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
 735	if (IS_ERR(req)) {
 736		d_drop(dentry);
 737		return PTR_ERR(req);
 738	}
 739	req->r_dentry = dget(dentry);
 740	req->r_num_caps = 2;
 741	req->r_path2 = kstrdup(dest, GFP_NOFS);
 742	req->r_locked_dir = dir;
 743	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 744	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 745	err = ceph_mdsc_do_request(mdsc, dir, req);
 746	if (!err && !req->r_reply_info.head->is_dentry)
 747		err = ceph_handle_notrace_create(dir, dentry);
 748	ceph_mdsc_put_request(req);
 749	if (!err)
 750		ceph_init_acl(dentry, dentry->d_inode, dir);
 751	else
 752		d_drop(dentry);
 753	return err;
 754}
 755
 756static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 757{
 758	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 759	struct ceph_mds_client *mdsc = fsc->mdsc;
 760	struct ceph_mds_request *req;
 761	int err = -EROFS;
 762	int op;
 763
 764	if (ceph_snap(dir) == CEPH_SNAPDIR) {
 765		/* mkdir .snap/foo is a MKSNAP */
 766		op = CEPH_MDS_OP_MKSNAP;
 767		dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
 768		     dentry->d_name.len, dentry->d_name.name, dentry);
 769	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
 770		dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
 771		op = CEPH_MDS_OP_MKDIR;
 772	} else {
 773		goto out;
 774	}
 775	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 776	if (IS_ERR(req)) {
 777		err = PTR_ERR(req);
 778		goto out;
 779	}
 780
 781	req->r_dentry = dget(dentry);
 782	req->r_num_caps = 2;
 783	req->r_locked_dir = dir;
 784	req->r_args.mkdir.mode = cpu_to_le32(mode);
 785	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 786	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 787	err = ceph_mdsc_do_request(mdsc, dir, req);
 788	if (!err && !req->r_reply_info.head->is_dentry)
 789		err = ceph_handle_notrace_create(dir, dentry);
 790	ceph_mdsc_put_request(req);
 791out:
 792	if (!err)
 793		ceph_init_acl(dentry, dentry->d_inode, dir);
 794	else
 795		d_drop(dentry);
 796	return err;
 797}
 798
 799static int ceph_link(struct dentry *old_dentry, struct inode *dir,
 800		     struct dentry *dentry)
 801{
 802	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 803	struct ceph_mds_client *mdsc = fsc->mdsc;
 804	struct ceph_mds_request *req;
 805	int err;
 806
 807	if (ceph_snap(dir) != CEPH_NOSNAP)
 808		return -EROFS;
 809
 810	dout("link in dir %p old_dentry %p dentry %p\n", dir,
 811	     old_dentry, dentry);
 812	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
 813	if (IS_ERR(req)) {
 814		d_drop(dentry);
 815		return PTR_ERR(req);
 816	}
 817	req->r_dentry = dget(dentry);
 818	req->r_num_caps = 2;
 819	req->r_old_dentry = dget(old_dentry);
 
 820	req->r_locked_dir = dir;
 821	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 822	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 823	/* release LINK_SHARED on source inode (mds will lock it) */
 824	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
 825	err = ceph_mdsc_do_request(mdsc, dir, req);
 826	if (err) {
 827		d_drop(dentry);
 828	} else if (!req->r_reply_info.head->is_dentry) {
 829		ihold(old_dentry->d_inode);
 830		d_instantiate(dentry, old_dentry->d_inode);
 831	}
 832	ceph_mdsc_put_request(req);
 833	return err;
 834}
 835
 836/*
 837 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps.  If it
 838 * looks like the link count will hit 0, drop any other caps (other
 839 * than PIN) we don't specifically want (due to the file still being
 840 * open).
 841 */
 842static int drop_caps_for_unlink(struct inode *inode)
 843{
 844	struct ceph_inode_info *ci = ceph_inode(inode);
 845	int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
 846
 847	spin_lock(&ci->i_ceph_lock);
 848	if (inode->i_nlink == 1) {
 849		drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
 850		ci->i_ceph_flags |= CEPH_I_NODELAY;
 851	}
 852	spin_unlock(&ci->i_ceph_lock);
 853	return drop;
 854}
 855
 856/*
 857 * rmdir and unlink are differ only by the metadata op code
 858 */
 859static int ceph_unlink(struct inode *dir, struct dentry *dentry)
 860{
 861	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 862	struct ceph_mds_client *mdsc = fsc->mdsc;
 863	struct inode *inode = dentry->d_inode;
 864	struct ceph_mds_request *req;
 865	int err = -EROFS;
 866	int op;
 867
 868	if (ceph_snap(dir) == CEPH_SNAPDIR) {
 869		/* rmdir .snap/foo is RMSNAP */
 870		dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
 871		     dentry->d_name.name, dentry);
 872		op = CEPH_MDS_OP_RMSNAP;
 873	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
 874		dout("unlink/rmdir dir %p dn %p inode %p\n",
 875		     dir, dentry, inode);
 876		op = S_ISDIR(dentry->d_inode->i_mode) ?
 877			CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
 878	} else
 879		goto out;
 880	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 881	if (IS_ERR(req)) {
 882		err = PTR_ERR(req);
 883		goto out;
 884	}
 885	req->r_dentry = dget(dentry);
 886	req->r_num_caps = 2;
 887	req->r_locked_dir = dir;
 888	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 889	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 890	req->r_inode_drop = drop_caps_for_unlink(inode);
 891	err = ceph_mdsc_do_request(mdsc, dir, req);
 892	if (!err && !req->r_reply_info.head->is_dentry)
 893		d_delete(dentry);
 894	ceph_mdsc_put_request(req);
 895out:
 896	return err;
 897}
 898
 899static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
 900		       struct inode *new_dir, struct dentry *new_dentry)
 901{
 902	struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
 903	struct ceph_mds_client *mdsc = fsc->mdsc;
 904	struct ceph_mds_request *req;
 905	int err;
 906
 907	if (ceph_snap(old_dir) != ceph_snap(new_dir))
 908		return -EXDEV;
 909	if (ceph_snap(old_dir) != CEPH_NOSNAP ||
 910	    ceph_snap(new_dir) != CEPH_NOSNAP)
 911		return -EROFS;
 912	dout("rename dir %p dentry %p to dir %p dentry %p\n",
 913	     old_dir, old_dentry, new_dir, new_dentry);
 914	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
 915	if (IS_ERR(req))
 916		return PTR_ERR(req);
 917	ihold(old_dir);
 918	req->r_dentry = dget(new_dentry);
 919	req->r_num_caps = 2;
 920	req->r_old_dentry = dget(old_dentry);
 921	req->r_old_dentry_dir = old_dir;
 922	req->r_locked_dir = new_dir;
 923	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
 924	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
 925	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 926	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 927	/* release LINK_RDCACHE on source inode (mds will lock it) */
 928	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
 929	if (new_dentry->d_inode)
 930		req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
 931	err = ceph_mdsc_do_request(mdsc, old_dir, req);
 932	if (!err && !req->r_reply_info.head->is_dentry) {
 933		/*
 934		 * Normally d_move() is done by fill_trace (called by
 935		 * do_request, above).  If there is no trace, we need
 936		 * to do it here.
 937		 */
 938
 
 
 
 939		d_move(old_dentry, new_dentry);
 940
 941		/* ensure target dentry is invalidated, despite
 942		   rehashing bug in vfs_rename_dir */
 943		ceph_invalidate_dentry_lease(new_dentry);
 944
 945		/* d_move screws up sibling dentries' offsets */
 946		ceph_dir_clear_complete(old_dir);
 947		ceph_dir_clear_complete(new_dir);
 948
 949	}
 950	ceph_mdsc_put_request(req);
 951	return err;
 952}
 953
 954/*
 955 * Ensure a dentry lease will no longer revalidate.
 956 */
 957void ceph_invalidate_dentry_lease(struct dentry *dentry)
 958{
 959	spin_lock(&dentry->d_lock);
 960	dentry->d_time = jiffies;
 961	ceph_dentry(dentry)->lease_shared_gen = 0;
 962	spin_unlock(&dentry->d_lock);
 963}
 964
 965/*
 966 * Check if dentry lease is valid.  If not, delete the lease.  Try to
 967 * renew if the least is more than half up.
 968 */
 969static int dentry_lease_is_valid(struct dentry *dentry)
 970{
 971	struct ceph_dentry_info *di;
 972	struct ceph_mds_session *s;
 973	int valid = 0;
 974	u32 gen;
 975	unsigned long ttl;
 976	struct ceph_mds_session *session = NULL;
 977	struct inode *dir = NULL;
 978	u32 seq = 0;
 979
 980	spin_lock(&dentry->d_lock);
 981	di = ceph_dentry(dentry);
 982	if (di->lease_session) {
 983		s = di->lease_session;
 984		spin_lock(&s->s_gen_ttl_lock);
 985		gen = s->s_cap_gen;
 986		ttl = s->s_cap_ttl;
 987		spin_unlock(&s->s_gen_ttl_lock);
 988
 989		if (di->lease_gen == gen &&
 990		    time_before(jiffies, dentry->d_time) &&
 991		    time_before(jiffies, ttl)) {
 992			valid = 1;
 993			if (di->lease_renew_after &&
 994			    time_after(jiffies, di->lease_renew_after)) {
 995				/* we should renew */
 996				dir = dentry->d_parent->d_inode;
 997				session = ceph_get_mds_session(s);
 998				seq = di->lease_seq;
 999				di->lease_renew_after = 0;
1000				di->lease_renew_from = jiffies;
1001			}
1002		}
1003	}
1004	spin_unlock(&dentry->d_lock);
1005
1006	if (session) {
1007		ceph_mdsc_lease_send_msg(session, dir, dentry,
1008					 CEPH_MDS_LEASE_RENEW, seq);
1009		ceph_put_mds_session(session);
1010	}
1011	dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1012	return valid;
1013}
1014
1015/*
1016 * Check if directory-wide content lease/cap is valid.
1017 */
1018static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1019{
1020	struct ceph_inode_info *ci = ceph_inode(dir);
1021	struct ceph_dentry_info *di = ceph_dentry(dentry);
1022	int valid = 0;
1023
1024	spin_lock(&ci->i_ceph_lock);
1025	if (ci->i_shared_gen == di->lease_shared_gen)
1026		valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1027	spin_unlock(&ci->i_ceph_lock);
1028	dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1029	     dir, (unsigned)ci->i_shared_gen, dentry,
1030	     (unsigned)di->lease_shared_gen, valid);
1031	return valid;
1032}
1033
1034/*
1035 * Check if cached dentry can be trusted.
1036 */
1037static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1038{
1039	int valid = 0;
1040	struct inode *dir;
1041
1042	if (flags & LOOKUP_RCU)
1043		return -ECHILD;
1044
1045	dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1046	     dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1047	     ceph_dentry(dentry)->offset);
1048
1049	dir = ceph_get_dentry_parent_inode(dentry);
1050
1051	/* always trust cached snapped dentries, snapdir dentry */
1052	if (ceph_snap(dir) != CEPH_NOSNAP) {
1053		dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1054		     dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1055		valid = 1;
1056	} else if (dentry->d_inode &&
1057		   ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1058		valid = 1;
1059	} else if (dentry_lease_is_valid(dentry) ||
1060		   dir_lease_is_valid(dir, dentry)) {
1061		if (dentry->d_inode)
1062			valid = ceph_is_any_caps(dentry->d_inode);
1063		else
1064			valid = 1;
1065	}
1066
1067	dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1068	if (valid) {
1069		ceph_dentry_lru_touch(dentry);
1070	} else {
1071		ceph_dir_clear_complete(dir);
1072		d_drop(dentry);
1073	}
1074	iput(dir);
1075	return valid;
1076}
1077
1078/*
1079 * Release our ceph_dentry_info.
1080 */
1081static void ceph_d_release(struct dentry *dentry)
1082{
1083	struct ceph_dentry_info *di = ceph_dentry(dentry);
1084
1085	dout("d_release %p\n", dentry);
1086	ceph_dentry_lru_del(dentry);
1087	if (di->lease_session)
1088		ceph_put_mds_session(di->lease_session);
1089	kmem_cache_free(ceph_dentry_cachep, di);
1090	dentry->d_fsdata = NULL;
1091}
1092
1093static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1094					  unsigned int flags)
1095{
1096	/*
1097	 * Eventually, we'll want to revalidate snapped metadata
1098	 * too... probably...
1099	 */
1100	return 1;
1101}
1102
1103/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1104 * When the VFS prunes a dentry from the cache, we need to clear the
1105 * complete flag on the parent directory.
1106 *
1107 * Called under dentry->d_lock.
1108 */
1109static void ceph_d_prune(struct dentry *dentry)
1110{
 
 
1111	dout("ceph_d_prune %p\n", dentry);
1112
1113	/* do we have a valid parent? */
1114	if (IS_ROOT(dentry))
1115		return;
1116
1117	/* if we are not hashed, we don't affect dir's completeness */
1118	if (d_unhashed(dentry))
1119		return;
1120
1121	/*
1122	 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1123	 * cleared until d_release
1124	 */
1125	ceph_dir_clear_complete(dentry->d_parent->d_inode);
 
1126}
1127
1128/*
1129 * read() on a dir.  This weird interface hack only works if mounted
1130 * with '-o dirstat'.
1131 */
1132static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1133			     loff_t *ppos)
1134{
1135	struct ceph_file_info *cf = file->private_data;
1136	struct inode *inode = file_inode(file);
1137	struct ceph_inode_info *ci = ceph_inode(inode);
1138	int left;
1139	const int bufsize = 1024;
1140
1141	if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1142		return -EISDIR;
1143
1144	if (!cf->dir_info) {
1145		cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1146		if (!cf->dir_info)
1147			return -ENOMEM;
1148		cf->dir_info_len =
1149			snprintf(cf->dir_info, bufsize,
1150				"entries:   %20lld\n"
1151				" files:    %20lld\n"
1152				" subdirs:  %20lld\n"
1153				"rentries:  %20lld\n"
1154				" rfiles:   %20lld\n"
1155				" rsubdirs: %20lld\n"
1156				"rbytes:    %20lld\n"
1157				"rctime:    %10ld.%09ld\n",
1158				ci->i_files + ci->i_subdirs,
1159				ci->i_files,
1160				ci->i_subdirs,
1161				ci->i_rfiles + ci->i_rsubdirs,
1162				ci->i_rfiles,
1163				ci->i_rsubdirs,
1164				ci->i_rbytes,
1165				(long)ci->i_rctime.tv_sec,
1166				(long)ci->i_rctime.tv_nsec);
1167	}
1168
1169	if (*ppos >= cf->dir_info_len)
1170		return 0;
1171	size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1172	left = copy_to_user(buf, cf->dir_info + *ppos, size);
1173	if (left == size)
1174		return -EFAULT;
1175	*ppos += (size - left);
1176	return size - left;
1177}
1178
1179/*
1180 * an fsync() on a dir will wait for any uncommitted directory
1181 * operations to commit.
1182 */
1183static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1184			  int datasync)
1185{
1186	struct inode *inode = file_inode(file);
1187	struct ceph_inode_info *ci = ceph_inode(inode);
1188	struct list_head *head = &ci->i_unsafe_dirops;
1189	struct ceph_mds_request *req;
1190	u64 last_tid;
1191	int ret = 0;
1192
1193	dout("dir_fsync %p\n", inode);
1194	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1195	if (ret)
1196		return ret;
1197	mutex_lock(&inode->i_mutex);
1198
1199	spin_lock(&ci->i_unsafe_lock);
1200	if (list_empty(head))
1201		goto out;
1202
1203	req = list_entry(head->prev,
1204			 struct ceph_mds_request, r_unsafe_dir_item);
1205	last_tid = req->r_tid;
1206
1207	do {
1208		ceph_mdsc_get_request(req);
1209		spin_unlock(&ci->i_unsafe_lock);
1210
1211		dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1212		     inode, req->r_tid, last_tid);
1213		if (req->r_timeout) {
1214			ret = wait_for_completion_timeout(
1215				&req->r_safe_completion, req->r_timeout);
1216			if (ret > 0)
1217				ret = 0;
1218			else if (ret == 0)
1219				ret = -EIO;  /* timed out */
1220		} else {
1221			wait_for_completion(&req->r_safe_completion);
1222		}
1223		ceph_mdsc_put_request(req);
1224
1225		spin_lock(&ci->i_unsafe_lock);
1226		if (ret || list_empty(head))
1227			break;
1228		req = list_entry(head->next,
1229				 struct ceph_mds_request, r_unsafe_dir_item);
1230	} while (req->r_tid < last_tid);
1231out:
1232	spin_unlock(&ci->i_unsafe_lock);
1233	mutex_unlock(&inode->i_mutex);
1234
1235	return ret;
1236}
1237
1238/*
1239 * We maintain a private dentry LRU.
1240 *
1241 * FIXME: this needs to be changed to a per-mds lru to be useful.
1242 */
1243void ceph_dentry_lru_add(struct dentry *dn)
1244{
1245	struct ceph_dentry_info *di = ceph_dentry(dn);
1246	struct ceph_mds_client *mdsc;
1247
1248	dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1249	     dn->d_name.len, dn->d_name.name);
1250	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1251	spin_lock(&mdsc->dentry_lru_lock);
1252	list_add_tail(&di->lru, &mdsc->dentry_lru);
1253	mdsc->num_dentry++;
1254	spin_unlock(&mdsc->dentry_lru_lock);
1255}
1256
1257void ceph_dentry_lru_touch(struct dentry *dn)
1258{
1259	struct ceph_dentry_info *di = ceph_dentry(dn);
1260	struct ceph_mds_client *mdsc;
1261
1262	dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1263	     dn->d_name.len, dn->d_name.name, di->offset);
1264	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1265	spin_lock(&mdsc->dentry_lru_lock);
1266	list_move_tail(&di->lru, &mdsc->dentry_lru);
1267	spin_unlock(&mdsc->dentry_lru_lock);
1268}
1269
1270void ceph_dentry_lru_del(struct dentry *dn)
1271{
1272	struct ceph_dentry_info *di = ceph_dentry(dn);
1273	struct ceph_mds_client *mdsc;
1274
1275	dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1276	     dn->d_name.len, dn->d_name.name);
1277	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1278	spin_lock(&mdsc->dentry_lru_lock);
1279	list_del_init(&di->lru);
1280	mdsc->num_dentry--;
1281	spin_unlock(&mdsc->dentry_lru_lock);
1282}
1283
1284/*
1285 * Return name hash for a given dentry.  This is dependent on
1286 * the parent directory's hash function.
1287 */
1288unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1289{
1290	struct ceph_inode_info *dci = ceph_inode(dir);
1291
1292	switch (dci->i_dir_layout.dl_dir_hash) {
1293	case 0:	/* for backward compat */
1294	case CEPH_STR_HASH_LINUX:
1295		return dn->d_name.hash;
1296
1297	default:
1298		return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1299				     dn->d_name.name, dn->d_name.len);
1300	}
1301}
1302
1303const struct file_operations ceph_dir_fops = {
1304	.read = ceph_read_dir,
1305	.iterate = ceph_readdir,
1306	.llseek = ceph_dir_llseek,
1307	.open = ceph_open,
1308	.release = ceph_release,
1309	.unlocked_ioctl = ceph_ioctl,
1310	.fsync = ceph_dir_fsync,
1311};
1312
1313const struct inode_operations ceph_dir_iops = {
1314	.lookup = ceph_lookup,
1315	.permission = ceph_permission,
1316	.getattr = ceph_getattr,
1317	.setattr = ceph_setattr,
1318	.setxattr = ceph_setxattr,
1319	.getxattr = ceph_getxattr,
1320	.listxattr = ceph_listxattr,
1321	.removexattr = ceph_removexattr,
1322	.get_acl = ceph_get_acl,
1323	.set_acl = ceph_set_acl,
1324	.mknod = ceph_mknod,
1325	.symlink = ceph_symlink,
1326	.mkdir = ceph_mkdir,
1327	.link = ceph_link,
1328	.unlink = ceph_unlink,
1329	.rmdir = ceph_unlink,
1330	.rename = ceph_rename,
1331	.create = ceph_create,
1332	.atomic_open = ceph_atomic_open,
1333};
1334
1335const struct dentry_operations ceph_dentry_ops = {
1336	.d_revalidate = ceph_d_revalidate,
1337	.d_release = ceph_d_release,
1338	.d_prune = ceph_d_prune,
1339};
1340
1341const struct dentry_operations ceph_snapdir_dentry_ops = {
1342	.d_revalidate = ceph_snapdir_d_revalidate,
1343	.d_release = ceph_d_release,
1344};
1345
1346const struct dentry_operations ceph_snap_dentry_ops = {
1347	.d_release = ceph_d_release,
1348	.d_prune = ceph_d_prune,
1349};
v3.5.6
   1#include <linux/ceph/ceph_debug.h>
   2
   3#include <linux/spinlock.h>
   4#include <linux/fs_struct.h>
   5#include <linux/namei.h>
   6#include <linux/slab.h>
   7#include <linux/sched.h>
   8
   9#include "super.h"
  10#include "mds_client.h"
  11
  12/*
  13 * Directory operations: readdir, lookup, create, link, unlink,
  14 * rename, etc.
  15 */
  16
  17/*
  18 * Ceph MDS operations are specified in terms of a base ino and
  19 * relative path.  Thus, the client can specify an operation on a
  20 * specific inode (e.g., a getattr due to fstat(2)), or as a path
  21 * relative to, say, the root directory.
  22 *
  23 * Normally, we limit ourselves to strict inode ops (no path component)
  24 * or dentry operations (a single path component relative to an ino).  The
  25 * exception to this is open_root_dentry(), which will open the mount
  26 * point by name.
  27 */
  28
  29const struct inode_operations ceph_dir_iops;
  30const struct file_operations ceph_dir_fops;
  31const struct dentry_operations ceph_dentry_ops;
  32
  33/*
  34 * Initialize ceph dentry state.
  35 */
  36int ceph_init_dentry(struct dentry *dentry)
  37{
  38	struct ceph_dentry_info *di;
  39
  40	if (dentry->d_fsdata)
  41		return 0;
  42
  43	di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
  44	if (!di)
  45		return -ENOMEM;          /* oh well */
  46
  47	spin_lock(&dentry->d_lock);
  48	if (dentry->d_fsdata) {
  49		/* lost a race */
  50		kmem_cache_free(ceph_dentry_cachep, di);
  51		goto out_unlock;
  52	}
  53
  54	if (dentry->d_parent == NULL ||   /* nfs fh_to_dentry */
  55	    ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
  56		d_set_d_op(dentry, &ceph_dentry_ops);
  57	else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
  58		d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
  59	else
  60		d_set_d_op(dentry, &ceph_snap_dentry_ops);
  61
  62	di->dentry = dentry;
  63	di->lease_session = NULL;
  64	dentry->d_time = jiffies;
  65	/* avoid reordering d_fsdata setup so that the check above is safe */
  66	smp_mb();
  67	dentry->d_fsdata = di;
  68	ceph_dentry_lru_add(dentry);
  69out_unlock:
  70	spin_unlock(&dentry->d_lock);
  71	return 0;
  72}
  73
  74struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
  75{
  76	struct inode *inode = NULL;
  77
  78	if (!dentry)
  79		return NULL;
  80
  81	spin_lock(&dentry->d_lock);
  82	if (dentry->d_parent) {
  83		inode = dentry->d_parent->d_inode;
  84		ihold(inode);
  85	}
  86	spin_unlock(&dentry->d_lock);
  87	return inode;
  88}
  89
  90
  91/*
  92 * for readdir, we encode the directory frag and offset within that
  93 * frag into f_pos.
  94 */
  95static unsigned fpos_frag(loff_t p)
  96{
  97	return p >> 32;
  98}
  99static unsigned fpos_off(loff_t p)
 100{
 101	return p & 0xffffffff;
 102}
 103
 
 
 
 
 
 
 
 
 104/*
 105 * When possible, we try to satisfy a readdir by peeking at the
 106 * dcache.  We make this work by carefully ordering dentries on
 107 * d_u.d_child when we initially get results back from the MDS, and
 108 * falling back to a "normal" sync readdir if any dentries in the dir
 109 * are dropped.
 110 *
 111 * D_COMPLETE tells indicates we have all dentries in the dir.  It is
 112 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
 113 * the MDS if/when the directory is modified).
 114 */
 115static int __dcache_readdir(struct file *filp,
 116			    void *dirent, filldir_t filldir)
 117{
 118	struct ceph_file_info *fi = filp->private_data;
 119	struct dentry *parent = filp->f_dentry;
 120	struct inode *dir = parent->d_inode;
 121	struct list_head *p;
 122	struct dentry *dentry, *last;
 123	struct ceph_dentry_info *di;
 124	int err = 0;
 125
 126	/* claim ref on last dentry we returned */
 127	last = fi->dentry;
 128	fi->dentry = NULL;
 129
 130	dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
 131	     last);
 132
 133	spin_lock(&parent->d_lock);
 134
 135	/* start at beginning? */
 136	if (filp->f_pos == 2 || last == NULL ||
 137	    filp->f_pos < ceph_dentry(last)->offset) {
 138		if (list_empty(&parent->d_subdirs))
 139			goto out_unlock;
 140		p = parent->d_subdirs.prev;
 141		dout(" initial p %p/%p\n", p->prev, p->next);
 142	} else {
 143		p = last->d_u.d_child.prev;
 144	}
 145
 146more:
 147	dentry = list_entry(p, struct dentry, d_u.d_child);
 148	di = ceph_dentry(dentry);
 149	while (1) {
 150		dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
 151		     d_unhashed(dentry) ? "!hashed" : "hashed",
 152		     parent->d_subdirs.prev, parent->d_subdirs.next);
 153		if (p == &parent->d_subdirs) {
 154			fi->flags |= CEPH_F_ATEND;
 155			goto out_unlock;
 156		}
 157		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 158		if (!d_unhashed(dentry) && dentry->d_inode &&
 
 159		    ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
 160		    ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
 161		    filp->f_pos <= di->offset)
 162			break;
 163		dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
 164		     dentry->d_name.len, dentry->d_name.name, di->offset,
 165		     filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
 166		     !dentry->d_inode ? " null" : "");
 167		spin_unlock(&dentry->d_lock);
 168		p = p->prev;
 169		dentry = list_entry(p, struct dentry, d_u.d_child);
 170		di = ceph_dentry(dentry);
 171	}
 172
 173	dget_dlock(dentry);
 174	spin_unlock(&dentry->d_lock);
 175	spin_unlock(&parent->d_lock);
 176
 177	dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
 
 
 
 
 
 
 
 
 178	     dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
 179	filp->f_pos = di->offset;
 180	err = filldir(dirent, dentry->d_name.name,
 181		      dentry->d_name.len, di->offset,
 182		      ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
 183		      dentry->d_inode->i_mode >> 12);
 184
 185	if (last) {
 186		if (err < 0) {
 187			/* remember our position */
 188			fi->dentry = last;
 189			fi->next_offset = di->offset;
 190		} else {
 191			dput(last);
 192		}
 
 
 193	}
 194	last = dentry;
 195
 196	if (err < 0)
 197		goto out;
 198
 199	filp->f_pos++;
 200
 201	/* make sure a dentry wasn't dropped while we didn't have parent lock */
 202	if (!ceph_dir_test_complete(dir)) {
 203		dout(" lost D_COMPLETE on %p; falling back to mds\n", dir);
 204		err = -EAGAIN;
 205		goto out;
 206	}
 207
 208	spin_lock(&parent->d_lock);
 209	p = p->prev;	/* advance to next dentry */
 210	goto more;
 211
 212out_unlock:
 213	spin_unlock(&parent->d_lock);
 214out:
 215	if (last)
 216		dput(last);
 217	return err;
 218}
 219
 220/*
 221 * make note of the last dentry we read, so we can
 222 * continue at the same lexicographical point,
 223 * regardless of what dir changes take place on the
 224 * server.
 225 */
 226static int note_last_dentry(struct ceph_file_info *fi, const char *name,
 227			    int len)
 228{
 229	kfree(fi->last_name);
 230	fi->last_name = kmalloc(len+1, GFP_NOFS);
 231	if (!fi->last_name)
 232		return -ENOMEM;
 233	memcpy(fi->last_name, name, len);
 234	fi->last_name[len] = 0;
 235	dout("note_last_dentry '%s'\n", fi->last_name);
 236	return 0;
 237}
 238
 239static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
 240{
 241	struct ceph_file_info *fi = filp->private_data;
 242	struct inode *inode = filp->f_dentry->d_inode;
 243	struct ceph_inode_info *ci = ceph_inode(inode);
 244	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 245	struct ceph_mds_client *mdsc = fsc->mdsc;
 246	unsigned frag = fpos_frag(filp->f_pos);
 247	int off = fpos_off(filp->f_pos);
 248	int err;
 249	u32 ftype;
 250	struct ceph_mds_reply_info_parsed *rinfo;
 251	const int max_entries = fsc->mount_options->max_readdir;
 252	const int max_bytes = fsc->mount_options->max_readdir_bytes;
 253
 254	dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
 255	if (fi->flags & CEPH_F_ATEND)
 256		return 0;
 257
 258	/* always start with . and .. */
 259	if (filp->f_pos == 0) {
 260		/* note dir version at start of readdir so we can tell
 261		 * if any dentries get dropped */
 262		fi->dir_release_count = ci->i_release_count;
 263
 264		dout("readdir off 0 -> '.'\n");
 265		if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
 266			    ceph_translate_ino(inode->i_sb, inode->i_ino),
 267			    inode->i_mode >> 12) < 0)
 268			return 0;
 269		filp->f_pos = 1;
 270		off = 1;
 271	}
 272	if (filp->f_pos == 1) {
 273		ino_t ino = parent_ino(filp->f_dentry);
 274		dout("readdir off 1 -> '..'\n");
 275		if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
 276			    ceph_translate_ino(inode->i_sb, ino),
 277			    inode->i_mode >> 12) < 0)
 278			return 0;
 279		filp->f_pos = 2;
 280		off = 2;
 281	}
 282
 283	/* can we use the dcache? */
 284	spin_lock(&ci->i_ceph_lock);
 285	if ((filp->f_pos == 2 || fi->dentry) &&
 286	    !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
 287	    ceph_snap(inode) != CEPH_SNAPDIR &&
 288	    ceph_dir_test_complete(inode) &&
 289	    __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
 
 290		spin_unlock(&ci->i_ceph_lock);
 291		err = __dcache_readdir(filp, dirent, filldir);
 292		if (err != -EAGAIN)
 293			return err;
 
 
 294	} else {
 295		spin_unlock(&ci->i_ceph_lock);
 296	}
 297	if (fi->dentry) {
 298		err = note_last_dentry(fi, fi->dentry->d_name.name,
 299				       fi->dentry->d_name.len);
 300		if (err)
 301			return err;
 302		dput(fi->dentry);
 303		fi->dentry = NULL;
 304	}
 305
 306	/* proceed with a normal readdir */
 307
 308more:
 309	/* do we have the correct frag content buffered? */
 310	if (fi->frag != frag || fi->last_readdir == NULL) {
 311		struct ceph_mds_request *req;
 312		int op = ceph_snap(inode) == CEPH_SNAPDIR ?
 313			CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
 314
 315		/* discard old result, if any */
 316		if (fi->last_readdir) {
 317			ceph_mdsc_put_request(fi->last_readdir);
 318			fi->last_readdir = NULL;
 319		}
 320
 321		/* requery frag tree, as the frag topology may have changed */
 322		frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
 323
 324		dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
 325		     ceph_vinop(inode), frag, fi->last_name);
 326		req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 327		if (IS_ERR(req))
 328			return PTR_ERR(req);
 
 
 
 
 
 329		req->r_inode = inode;
 330		ihold(inode);
 331		req->r_dentry = dget(filp->f_dentry);
 332		/* hints to request -> mds selection code */
 333		req->r_direct_mode = USE_AUTH_MDS;
 334		req->r_direct_hash = ceph_frag_value(frag);
 335		req->r_direct_is_hash = true;
 336		req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
 337		req->r_readdir_offset = fi->next_offset;
 338		req->r_args.readdir.frag = cpu_to_le32(frag);
 339		req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
 340		req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
 341		req->r_num_caps = max_entries + 1;
 342		err = ceph_mdsc_do_request(mdsc, NULL, req);
 343		if (err < 0) {
 344			ceph_mdsc_put_request(req);
 345			return err;
 346		}
 347		dout("readdir got and parsed readdir result=%d"
 348		     " on frag %x, end=%d, complete=%d\n", err, frag,
 349		     (int)req->r_reply_info.dir_end,
 350		     (int)req->r_reply_info.dir_complete);
 351
 352		if (!req->r_did_prepopulate) {
 353			dout("readdir !did_prepopulate");
 354			fi->dir_release_count--;    /* preclude D_COMPLETE */
 
 355		}
 356
 357		/* note next offset and last dentry name */
 
 
 
 
 
 
 
 
 
 
 358		fi->offset = fi->next_offset;
 359		fi->last_readdir = req;
 360
 361		if (req->r_reply_info.dir_end) {
 362			kfree(fi->last_name);
 363			fi->last_name = NULL;
 364			if (ceph_frag_is_rightmost(frag))
 365				fi->next_offset = 2;
 366			else
 367				fi->next_offset = 0;
 368		} else {
 369			rinfo = &req->r_reply_info;
 370			err = note_last_dentry(fi,
 371				       rinfo->dir_dname[rinfo->dir_nr-1],
 372				       rinfo->dir_dname_len[rinfo->dir_nr-1]);
 373			if (err)
 374				return err;
 375			fi->next_offset += rinfo->dir_nr;
 376		}
 377	}
 378
 379	rinfo = &fi->last_readdir->r_reply_info;
 380	dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
 381	     rinfo->dir_nr, off, fi->offset);
 
 
 382	while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
 383		u64 pos = ceph_make_fpos(frag, off);
 384		struct ceph_mds_reply_inode *in =
 385			rinfo->dir_in[off - fi->offset].in;
 386		struct ceph_vino vino;
 387		ino_t ino;
 388
 389		dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
 390		     off, off - fi->offset, rinfo->dir_nr, pos,
 391		     rinfo->dir_dname_len[off - fi->offset],
 392		     rinfo->dir_dname[off - fi->offset], in);
 393		BUG_ON(!in);
 394		ftype = le32_to_cpu(in->mode) >> 12;
 395		vino.ino = le64_to_cpu(in->ino);
 396		vino.snap = le64_to_cpu(in->snapid);
 397		ino = ceph_vino_to_ino(vino);
 398		if (filldir(dirent,
 399			    rinfo->dir_dname[off - fi->offset],
 400			    rinfo->dir_dname_len[off - fi->offset],
 401			    pos,
 402			    ceph_translate_ino(inode->i_sb, ino), ftype) < 0) {
 403			dout("filldir stopping us...\n");
 404			return 0;
 405		}
 406		off++;
 407		filp->f_pos = pos + 1;
 408	}
 409
 410	if (fi->last_name) {
 411		ceph_mdsc_put_request(fi->last_readdir);
 412		fi->last_readdir = NULL;
 413		goto more;
 414	}
 415
 416	/* more frags? */
 417	if (!ceph_frag_is_rightmost(frag)) {
 418		frag = ceph_frag_next(frag);
 419		off = 0;
 420		filp->f_pos = ceph_make_fpos(frag, off);
 421		dout("readdir next frag is %x\n", frag);
 422		goto more;
 423	}
 424	fi->flags |= CEPH_F_ATEND;
 425
 426	/*
 427	 * if dir_release_count still matches the dir, no dentries
 428	 * were released during the whole readdir, and we should have
 429	 * the complete dir contents in our cache.
 430	 */
 431	spin_lock(&ci->i_ceph_lock);
 432	if (ci->i_release_count == fi->dir_release_count) {
 433		ceph_dir_set_complete(inode);
 434		ci->i_max_offset = filp->f_pos;
 435	}
 436	spin_unlock(&ci->i_ceph_lock);
 437
 438	dout("readdir %p filp %p done.\n", inode, filp);
 439	return 0;
 440}
 441
 442static void reset_readdir(struct ceph_file_info *fi)
 443{
 444	if (fi->last_readdir) {
 445		ceph_mdsc_put_request(fi->last_readdir);
 446		fi->last_readdir = NULL;
 447	}
 448	kfree(fi->last_name);
 449	fi->last_name = NULL;
 450	fi->next_offset = 2;  /* compensate for . and .. */
 
 
 
 451	if (fi->dentry) {
 452		dput(fi->dentry);
 453		fi->dentry = NULL;
 454	}
 455	fi->flags &= ~CEPH_F_ATEND;
 456}
 457
 458static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
 459{
 460	struct ceph_file_info *fi = file->private_data;
 461	struct inode *inode = file->f_mapping->host;
 462	loff_t old_offset = offset;
 463	loff_t retval;
 464
 465	mutex_lock(&inode->i_mutex);
 466	retval = -EINVAL;
 467	switch (origin) {
 468	case SEEK_END:
 469		offset += inode->i_size + 2;   /* FIXME */
 470		break;
 471	case SEEK_CUR:
 472		offset += file->f_pos;
 473	case SEEK_SET:
 474		break;
 475	default:
 476		goto out;
 477	}
 478
 479	if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
 480		if (offset != file->f_pos) {
 481			file->f_pos = offset;
 482			file->f_version = 0;
 483			fi->flags &= ~CEPH_F_ATEND;
 484		}
 485		retval = offset;
 486
 487		/*
 488		 * discard buffered readdir content on seekdir(0), or
 489		 * seek to new frag, or seek prior to current chunk.
 490		 */
 491		if (offset == 0 ||
 492		    fpos_frag(offset) != fpos_frag(old_offset) ||
 493		    fpos_off(offset) < fi->offset) {
 494			dout("dir_llseek dropping %p content\n", file);
 495			reset_readdir(fi);
 496		}
 497
 498		/* bump dir_release_count if we did a forward seek */
 499		if (offset > old_offset)
 500			fi->dir_release_count--;
 501	}
 502out:
 503	mutex_unlock(&inode->i_mutex);
 504	return retval;
 505}
 506
 507/*
 508 * Handle lookups for the hidden .snap directory.
 509 */
 510int ceph_handle_snapdir(struct ceph_mds_request *req,
 511			struct dentry *dentry, int err)
 512{
 513	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
 514	struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
 515
 516	/* .snap dir? */
 517	if (err == -ENOENT &&
 518	    ceph_snap(parent) == CEPH_NOSNAP &&
 519	    strcmp(dentry->d_name.name,
 520		   fsc->mount_options->snapdir_name) == 0) {
 521		struct inode *inode = ceph_get_snapdir(parent);
 522		dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
 523		     dentry, dentry->d_name.len, dentry->d_name.name, inode);
 524		BUG_ON(!d_unhashed(dentry));
 525		d_add(dentry, inode);
 526		err = 0;
 527	}
 528	return err;
 529}
 530
 531/*
 532 * Figure out final result of a lookup/open request.
 533 *
 534 * Mainly, make sure we return the final req->r_dentry (if it already
 535 * existed) in place of the original VFS-provided dentry when they
 536 * differ.
 537 *
 538 * Gracefully handle the case where the MDS replies with -ENOENT and
 539 * no trace (which it may do, at its discretion, e.g., if it doesn't
 540 * care to issue a lease on the negative dentry).
 541 */
 542struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
 543				  struct dentry *dentry, int err)
 544{
 545	if (err == -ENOENT) {
 546		/* no trace? */
 547		err = 0;
 548		if (!req->r_reply_info.head->is_dentry) {
 549			dout("ENOENT and no trace, dentry %p inode %p\n",
 550			     dentry, dentry->d_inode);
 551			if (dentry->d_inode) {
 552				d_drop(dentry);
 553				err = -ENOENT;
 554			} else {
 555				d_add(dentry, NULL);
 556			}
 557		}
 558	}
 559	if (err)
 560		dentry = ERR_PTR(err);
 561	else if (dentry != req->r_dentry)
 562		dentry = dget(req->r_dentry);   /* we got spliced */
 563	else
 564		dentry = NULL;
 565	return dentry;
 566}
 567
 568static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
 569{
 570	return ceph_ino(inode) == CEPH_INO_ROOT &&
 571		strncmp(dentry->d_name.name, ".ceph", 5) == 0;
 572}
 573
 574/*
 575 * Look up a single dir entry.  If there is a lookup intent, inform
 576 * the MDS so that it gets our 'caps wanted' value in a single op.
 577 */
 578static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
 579				  struct nameidata *nd)
 580{
 581	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 582	struct ceph_mds_client *mdsc = fsc->mdsc;
 583	struct ceph_mds_request *req;
 584	int op;
 585	int err;
 586
 587	dout("lookup %p dentry %p '%.*s'\n",
 588	     dir, dentry, dentry->d_name.len, dentry->d_name.name);
 589
 590	if (dentry->d_name.len > NAME_MAX)
 591		return ERR_PTR(-ENAMETOOLONG);
 592
 593	err = ceph_init_dentry(dentry);
 594	if (err < 0)
 595		return ERR_PTR(err);
 596
 597	/* open (but not create!) intent? */
 598	if (nd &&
 599	    (nd->flags & LOOKUP_OPEN) &&
 600	    !(nd->intent.open.flags & O_CREAT)) {
 601		int mode = nd->intent.open.create_mode & ~current->fs->umask;
 602		return ceph_lookup_open(dir, dentry, nd, mode, 1);
 603	}
 604
 605	/* can we conclude ENOENT locally? */
 606	if (dentry->d_inode == NULL) {
 607		struct ceph_inode_info *ci = ceph_inode(dir);
 608		struct ceph_dentry_info *di = ceph_dentry(dentry);
 609
 610		spin_lock(&ci->i_ceph_lock);
 611		dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
 612		if (strncmp(dentry->d_name.name,
 613			    fsc->mount_options->snapdir_name,
 614			    dentry->d_name.len) &&
 615		    !is_root_ceph_dentry(dir, dentry) &&
 616		    ceph_dir_test_complete(dir) &&
 617		    (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
 618			spin_unlock(&ci->i_ceph_lock);
 619			dout(" dir %p complete, -ENOENT\n", dir);
 620			d_add(dentry, NULL);
 621			di->lease_shared_gen = ci->i_shared_gen;
 622			return NULL;
 623		}
 624		spin_unlock(&ci->i_ceph_lock);
 625	}
 626
 627	op = ceph_snap(dir) == CEPH_SNAPDIR ?
 628		CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
 629	req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
 630	if (IS_ERR(req))
 631		return ERR_CAST(req);
 632	req->r_dentry = dget(dentry);
 633	req->r_num_caps = 2;
 634	/* we only need inode linkage */
 635	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
 636	req->r_locked_dir = dir;
 637	err = ceph_mdsc_do_request(mdsc, NULL, req);
 638	err = ceph_handle_snapdir(req, dentry, err);
 639	dentry = ceph_finish_lookup(req, dentry, err);
 640	ceph_mdsc_put_request(req);  /* will dput(dentry) */
 641	dout("lookup result=%p\n", dentry);
 642	return dentry;
 643}
 644
 645/*
 646 * If we do a create but get no trace back from the MDS, follow up with
 647 * a lookup (the VFS expects us to link up the provided dentry).
 648 */
 649int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
 650{
 651	struct dentry *result = ceph_lookup(dir, dentry, NULL);
 652
 653	if (result && !IS_ERR(result)) {
 654		/*
 655		 * We created the item, then did a lookup, and found
 656		 * it was already linked to another inode we already
 657		 * had in our cache (and thus got spliced).  Link our
 658		 * dentry to that inode, but don't hash it, just in
 659		 * case the VFS wants to dereference it.
 660		 */
 661		BUG_ON(!result->d_inode);
 662		d_instantiate(dentry, result->d_inode);
 663		return 0;
 664	}
 665	return PTR_ERR(result);
 666}
 667
 668static int ceph_mknod(struct inode *dir, struct dentry *dentry,
 669		      umode_t mode, dev_t rdev)
 670{
 671	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 672	struct ceph_mds_client *mdsc = fsc->mdsc;
 673	struct ceph_mds_request *req;
 674	int err;
 675
 676	if (ceph_snap(dir) != CEPH_NOSNAP)
 677		return -EROFS;
 678
 679	dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
 680	     dir, dentry, mode, rdev);
 681	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
 682	if (IS_ERR(req)) {
 683		d_drop(dentry);
 684		return PTR_ERR(req);
 685	}
 686	req->r_dentry = dget(dentry);
 687	req->r_num_caps = 2;
 688	req->r_locked_dir = dir;
 689	req->r_args.mknod.mode = cpu_to_le32(mode);
 690	req->r_args.mknod.rdev = cpu_to_le32(rdev);
 691	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 692	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 693	err = ceph_mdsc_do_request(mdsc, dir, req);
 694	if (!err && !req->r_reply_info.head->is_dentry)
 695		err = ceph_handle_notrace_create(dir, dentry);
 696	ceph_mdsc_put_request(req);
 697	if (err)
 
 
 
 698		d_drop(dentry);
 699	return err;
 700}
 701
 702static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 703		       struct nameidata *nd)
 704{
 705	dout("create in dir %p dentry %p name '%.*s'\n",
 706	     dir, dentry, dentry->d_name.len, dentry->d_name.name);
 707
 708	if (ceph_snap(dir) != CEPH_NOSNAP)
 709		return -EROFS;
 710
 711	if (nd) {
 712		BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
 713		dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
 714		/* hrm, what should i do here if we get aliased? */
 715		if (IS_ERR(dentry))
 716			return PTR_ERR(dentry);
 717		return 0;
 718	}
 719
 720	/* fall back to mknod */
 721	return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
 722}
 723
 724static int ceph_symlink(struct inode *dir, struct dentry *dentry,
 725			    const char *dest)
 726{
 727	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 728	struct ceph_mds_client *mdsc = fsc->mdsc;
 729	struct ceph_mds_request *req;
 730	int err;
 731
 732	if (ceph_snap(dir) != CEPH_NOSNAP)
 733		return -EROFS;
 734
 735	dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
 736	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
 737	if (IS_ERR(req)) {
 738		d_drop(dentry);
 739		return PTR_ERR(req);
 740	}
 741	req->r_dentry = dget(dentry);
 742	req->r_num_caps = 2;
 743	req->r_path2 = kstrdup(dest, GFP_NOFS);
 744	req->r_locked_dir = dir;
 745	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 746	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 747	err = ceph_mdsc_do_request(mdsc, dir, req);
 748	if (!err && !req->r_reply_info.head->is_dentry)
 749		err = ceph_handle_notrace_create(dir, dentry);
 750	ceph_mdsc_put_request(req);
 751	if (err)
 
 
 752		d_drop(dentry);
 753	return err;
 754}
 755
 756static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 757{
 758	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 759	struct ceph_mds_client *mdsc = fsc->mdsc;
 760	struct ceph_mds_request *req;
 761	int err = -EROFS;
 762	int op;
 763
 764	if (ceph_snap(dir) == CEPH_SNAPDIR) {
 765		/* mkdir .snap/foo is a MKSNAP */
 766		op = CEPH_MDS_OP_MKSNAP;
 767		dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
 768		     dentry->d_name.len, dentry->d_name.name, dentry);
 769	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
 770		dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
 771		op = CEPH_MDS_OP_MKDIR;
 772	} else {
 773		goto out;
 774	}
 775	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 776	if (IS_ERR(req)) {
 777		err = PTR_ERR(req);
 778		goto out;
 779	}
 780
 781	req->r_dentry = dget(dentry);
 782	req->r_num_caps = 2;
 783	req->r_locked_dir = dir;
 784	req->r_args.mkdir.mode = cpu_to_le32(mode);
 785	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 786	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 787	err = ceph_mdsc_do_request(mdsc, dir, req);
 788	if (!err && !req->r_reply_info.head->is_dentry)
 789		err = ceph_handle_notrace_create(dir, dentry);
 790	ceph_mdsc_put_request(req);
 791out:
 792	if (err < 0)
 
 
 793		d_drop(dentry);
 794	return err;
 795}
 796
 797static int ceph_link(struct dentry *old_dentry, struct inode *dir,
 798		     struct dentry *dentry)
 799{
 800	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 801	struct ceph_mds_client *mdsc = fsc->mdsc;
 802	struct ceph_mds_request *req;
 803	int err;
 804
 805	if (ceph_snap(dir) != CEPH_NOSNAP)
 806		return -EROFS;
 807
 808	dout("link in dir %p old_dentry %p dentry %p\n", dir,
 809	     old_dentry, dentry);
 810	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
 811	if (IS_ERR(req)) {
 812		d_drop(dentry);
 813		return PTR_ERR(req);
 814	}
 815	req->r_dentry = dget(dentry);
 816	req->r_num_caps = 2;
 817	req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
 818	req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
 819	req->r_locked_dir = dir;
 820	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 821	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 
 
 822	err = ceph_mdsc_do_request(mdsc, dir, req);
 823	if (err) {
 824		d_drop(dentry);
 825	} else if (!req->r_reply_info.head->is_dentry) {
 826		ihold(old_dentry->d_inode);
 827		d_instantiate(dentry, old_dentry->d_inode);
 828	}
 829	ceph_mdsc_put_request(req);
 830	return err;
 831}
 832
 833/*
 834 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps.  If it
 835 * looks like the link count will hit 0, drop any other caps (other
 836 * than PIN) we don't specifically want (due to the file still being
 837 * open).
 838 */
 839static int drop_caps_for_unlink(struct inode *inode)
 840{
 841	struct ceph_inode_info *ci = ceph_inode(inode);
 842	int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
 843
 844	spin_lock(&ci->i_ceph_lock);
 845	if (inode->i_nlink == 1) {
 846		drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
 847		ci->i_ceph_flags |= CEPH_I_NODELAY;
 848	}
 849	spin_unlock(&ci->i_ceph_lock);
 850	return drop;
 851}
 852
 853/*
 854 * rmdir and unlink are differ only by the metadata op code
 855 */
 856static int ceph_unlink(struct inode *dir, struct dentry *dentry)
 857{
 858	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 859	struct ceph_mds_client *mdsc = fsc->mdsc;
 860	struct inode *inode = dentry->d_inode;
 861	struct ceph_mds_request *req;
 862	int err = -EROFS;
 863	int op;
 864
 865	if (ceph_snap(dir) == CEPH_SNAPDIR) {
 866		/* rmdir .snap/foo is RMSNAP */
 867		dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
 868		     dentry->d_name.name, dentry);
 869		op = CEPH_MDS_OP_RMSNAP;
 870	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
 871		dout("unlink/rmdir dir %p dn %p inode %p\n",
 872		     dir, dentry, inode);
 873		op = S_ISDIR(dentry->d_inode->i_mode) ?
 874			CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
 875	} else
 876		goto out;
 877	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 878	if (IS_ERR(req)) {
 879		err = PTR_ERR(req);
 880		goto out;
 881	}
 882	req->r_dentry = dget(dentry);
 883	req->r_num_caps = 2;
 884	req->r_locked_dir = dir;
 885	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 886	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 887	req->r_inode_drop = drop_caps_for_unlink(inode);
 888	err = ceph_mdsc_do_request(mdsc, dir, req);
 889	if (!err && !req->r_reply_info.head->is_dentry)
 890		d_delete(dentry);
 891	ceph_mdsc_put_request(req);
 892out:
 893	return err;
 894}
 895
 896static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
 897		       struct inode *new_dir, struct dentry *new_dentry)
 898{
 899	struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
 900	struct ceph_mds_client *mdsc = fsc->mdsc;
 901	struct ceph_mds_request *req;
 902	int err;
 903
 904	if (ceph_snap(old_dir) != ceph_snap(new_dir))
 905		return -EXDEV;
 906	if (ceph_snap(old_dir) != CEPH_NOSNAP ||
 907	    ceph_snap(new_dir) != CEPH_NOSNAP)
 908		return -EROFS;
 909	dout("rename dir %p dentry %p to dir %p dentry %p\n",
 910	     old_dir, old_dentry, new_dir, new_dentry);
 911	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
 912	if (IS_ERR(req))
 913		return PTR_ERR(req);
 
 914	req->r_dentry = dget(new_dentry);
 915	req->r_num_caps = 2;
 916	req->r_old_dentry = dget(old_dentry);
 917	req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
 918	req->r_locked_dir = new_dir;
 919	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
 920	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
 921	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 922	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 923	/* release LINK_RDCACHE on source inode (mds will lock it) */
 924	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
 925	if (new_dentry->d_inode)
 926		req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
 927	err = ceph_mdsc_do_request(mdsc, old_dir, req);
 928	if (!err && !req->r_reply_info.head->is_dentry) {
 929		/*
 930		 * Normally d_move() is done by fill_trace (called by
 931		 * do_request, above).  If there is no trace, we need
 932		 * to do it here.
 933		 */
 934
 935		/* d_move screws up d_subdirs order */
 936		ceph_dir_clear_complete(new_dir);
 937
 938		d_move(old_dentry, new_dentry);
 939
 940		/* ensure target dentry is invalidated, despite
 941		   rehashing bug in vfs_rename_dir */
 942		ceph_invalidate_dentry_lease(new_dentry);
 
 
 
 
 
 943	}
 944	ceph_mdsc_put_request(req);
 945	return err;
 946}
 947
 948/*
 949 * Ensure a dentry lease will no longer revalidate.
 950 */
 951void ceph_invalidate_dentry_lease(struct dentry *dentry)
 952{
 953	spin_lock(&dentry->d_lock);
 954	dentry->d_time = jiffies;
 955	ceph_dentry(dentry)->lease_shared_gen = 0;
 956	spin_unlock(&dentry->d_lock);
 957}
 958
 959/*
 960 * Check if dentry lease is valid.  If not, delete the lease.  Try to
 961 * renew if the least is more than half up.
 962 */
 963static int dentry_lease_is_valid(struct dentry *dentry)
 964{
 965	struct ceph_dentry_info *di;
 966	struct ceph_mds_session *s;
 967	int valid = 0;
 968	u32 gen;
 969	unsigned long ttl;
 970	struct ceph_mds_session *session = NULL;
 971	struct inode *dir = NULL;
 972	u32 seq = 0;
 973
 974	spin_lock(&dentry->d_lock);
 975	di = ceph_dentry(dentry);
 976	if (di->lease_session) {
 977		s = di->lease_session;
 978		spin_lock(&s->s_gen_ttl_lock);
 979		gen = s->s_cap_gen;
 980		ttl = s->s_cap_ttl;
 981		spin_unlock(&s->s_gen_ttl_lock);
 982
 983		if (di->lease_gen == gen &&
 984		    time_before(jiffies, dentry->d_time) &&
 985		    time_before(jiffies, ttl)) {
 986			valid = 1;
 987			if (di->lease_renew_after &&
 988			    time_after(jiffies, di->lease_renew_after)) {
 989				/* we should renew */
 990				dir = dentry->d_parent->d_inode;
 991				session = ceph_get_mds_session(s);
 992				seq = di->lease_seq;
 993				di->lease_renew_after = 0;
 994				di->lease_renew_from = jiffies;
 995			}
 996		}
 997	}
 998	spin_unlock(&dentry->d_lock);
 999
1000	if (session) {
1001		ceph_mdsc_lease_send_msg(session, dir, dentry,
1002					 CEPH_MDS_LEASE_RENEW, seq);
1003		ceph_put_mds_session(session);
1004	}
1005	dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1006	return valid;
1007}
1008
1009/*
1010 * Check if directory-wide content lease/cap is valid.
1011 */
1012static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1013{
1014	struct ceph_inode_info *ci = ceph_inode(dir);
1015	struct ceph_dentry_info *di = ceph_dentry(dentry);
1016	int valid = 0;
1017
1018	spin_lock(&ci->i_ceph_lock);
1019	if (ci->i_shared_gen == di->lease_shared_gen)
1020		valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1021	spin_unlock(&ci->i_ceph_lock);
1022	dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1023	     dir, (unsigned)ci->i_shared_gen, dentry,
1024	     (unsigned)di->lease_shared_gen, valid);
1025	return valid;
1026}
1027
1028/*
1029 * Check if cached dentry can be trusted.
1030 */
1031static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
1032{
1033	int valid = 0;
1034	struct inode *dir;
1035
1036	if (nd && nd->flags & LOOKUP_RCU)
1037		return -ECHILD;
1038
1039	dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1040	     dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1041	     ceph_dentry(dentry)->offset);
1042
1043	dir = ceph_get_dentry_parent_inode(dentry);
1044
1045	/* always trust cached snapped dentries, snapdir dentry */
1046	if (ceph_snap(dir) != CEPH_NOSNAP) {
1047		dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1048		     dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1049		valid = 1;
1050	} else if (dentry->d_inode &&
1051		   ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1052		valid = 1;
1053	} else if (dentry_lease_is_valid(dentry) ||
1054		   dir_lease_is_valid(dir, dentry)) {
1055		valid = 1;
 
 
 
1056	}
1057
1058	dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1059	if (valid)
1060		ceph_dentry_lru_touch(dentry);
1061	else
 
1062		d_drop(dentry);
 
1063	iput(dir);
1064	return valid;
1065}
1066
1067/*
1068 * Release our ceph_dentry_info.
1069 */
1070static void ceph_d_release(struct dentry *dentry)
1071{
1072	struct ceph_dentry_info *di = ceph_dentry(dentry);
1073
1074	dout("d_release %p\n", dentry);
1075	ceph_dentry_lru_del(dentry);
1076	if (di->lease_session)
1077		ceph_put_mds_session(di->lease_session);
1078	kmem_cache_free(ceph_dentry_cachep, di);
1079	dentry->d_fsdata = NULL;
1080}
1081
1082static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1083					  struct nameidata *nd)
1084{
1085	/*
1086	 * Eventually, we'll want to revalidate snapped metadata
1087	 * too... probably...
1088	 */
1089	return 1;
1090}
1091
1092/*
1093 * Set/clear/test dir complete flag on the dir's dentry.
1094 */
1095void ceph_dir_set_complete(struct inode *inode)
1096{
1097	struct dentry *dentry = d_find_any_alias(inode);
1098	
1099	if (dentry && ceph_dentry(dentry) &&
1100	    ceph_test_mount_opt(ceph_sb_to_client(dentry->d_sb), DCACHE)) {
1101		dout(" marking %p (%p) complete\n", inode, dentry);
1102		set_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1103	}
1104	dput(dentry);
1105}
1106
1107void ceph_dir_clear_complete(struct inode *inode)
1108{
1109	struct dentry *dentry = d_find_any_alias(inode);
1110
1111	if (dentry && ceph_dentry(dentry)) {
1112		dout(" marking %p (%p) complete\n", inode, dentry);
1113		set_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1114	}
1115	dput(dentry);
1116}
1117
1118bool ceph_dir_test_complete(struct inode *inode)
1119{
1120	struct dentry *dentry = d_find_any_alias(inode);
1121
1122	if (dentry && ceph_dentry(dentry)) {
1123		dout(" marking %p (%p) NOT complete\n", inode, dentry);
1124		clear_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1125	}
1126	dput(dentry);
1127	return false;
1128}
1129
1130/*
1131 * When the VFS prunes a dentry from the cache, we need to clear the
1132 * complete flag on the parent directory.
1133 *
1134 * Called under dentry->d_lock.
1135 */
1136static void ceph_d_prune(struct dentry *dentry)
1137{
1138	struct ceph_dentry_info *di;
1139
1140	dout("ceph_d_prune %p\n", dentry);
1141
1142	/* do we have a valid parent? */
1143	if (!dentry->d_parent || IS_ROOT(dentry))
1144		return;
1145
1146	/* if we are not hashed, we don't affect D_COMPLETE */
1147	if (d_unhashed(dentry))
1148		return;
1149
1150	/*
1151	 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1152	 * cleared until d_release
1153	 */
1154	di = ceph_dentry(dentry->d_parent);
1155	clear_bit(CEPH_D_COMPLETE, &di->flags);
1156}
1157
1158/*
1159 * read() on a dir.  This weird interface hack only works if mounted
1160 * with '-o dirstat'.
1161 */
1162static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1163			     loff_t *ppos)
1164{
1165	struct ceph_file_info *cf = file->private_data;
1166	struct inode *inode = file->f_dentry->d_inode;
1167	struct ceph_inode_info *ci = ceph_inode(inode);
1168	int left;
1169	const int bufsize = 1024;
1170
1171	if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1172		return -EISDIR;
1173
1174	if (!cf->dir_info) {
1175		cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1176		if (!cf->dir_info)
1177			return -ENOMEM;
1178		cf->dir_info_len =
1179			snprintf(cf->dir_info, bufsize,
1180				"entries:   %20lld\n"
1181				" files:    %20lld\n"
1182				" subdirs:  %20lld\n"
1183				"rentries:  %20lld\n"
1184				" rfiles:   %20lld\n"
1185				" rsubdirs: %20lld\n"
1186				"rbytes:    %20lld\n"
1187				"rctime:    %10ld.%09ld\n",
1188				ci->i_files + ci->i_subdirs,
1189				ci->i_files,
1190				ci->i_subdirs,
1191				ci->i_rfiles + ci->i_rsubdirs,
1192				ci->i_rfiles,
1193				ci->i_rsubdirs,
1194				ci->i_rbytes,
1195				(long)ci->i_rctime.tv_sec,
1196				(long)ci->i_rctime.tv_nsec);
1197	}
1198
1199	if (*ppos >= cf->dir_info_len)
1200		return 0;
1201	size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1202	left = copy_to_user(buf, cf->dir_info + *ppos, size);
1203	if (left == size)
1204		return -EFAULT;
1205	*ppos += (size - left);
1206	return size - left;
1207}
1208
1209/*
1210 * an fsync() on a dir will wait for any uncommitted directory
1211 * operations to commit.
1212 */
1213static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1214			  int datasync)
1215{
1216	struct inode *inode = file->f_path.dentry->d_inode;
1217	struct ceph_inode_info *ci = ceph_inode(inode);
1218	struct list_head *head = &ci->i_unsafe_dirops;
1219	struct ceph_mds_request *req;
1220	u64 last_tid;
1221	int ret = 0;
1222
1223	dout("dir_fsync %p\n", inode);
1224	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1225	if (ret)
1226		return ret;
1227	mutex_lock(&inode->i_mutex);
1228
1229	spin_lock(&ci->i_unsafe_lock);
1230	if (list_empty(head))
1231		goto out;
1232
1233	req = list_entry(head->prev,
1234			 struct ceph_mds_request, r_unsafe_dir_item);
1235	last_tid = req->r_tid;
1236
1237	do {
1238		ceph_mdsc_get_request(req);
1239		spin_unlock(&ci->i_unsafe_lock);
1240
1241		dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1242		     inode, req->r_tid, last_tid);
1243		if (req->r_timeout) {
1244			ret = wait_for_completion_timeout(
1245				&req->r_safe_completion, req->r_timeout);
1246			if (ret > 0)
1247				ret = 0;
1248			else if (ret == 0)
1249				ret = -EIO;  /* timed out */
1250		} else {
1251			wait_for_completion(&req->r_safe_completion);
1252		}
1253		ceph_mdsc_put_request(req);
1254
1255		spin_lock(&ci->i_unsafe_lock);
1256		if (ret || list_empty(head))
1257			break;
1258		req = list_entry(head->next,
1259				 struct ceph_mds_request, r_unsafe_dir_item);
1260	} while (req->r_tid < last_tid);
1261out:
1262	spin_unlock(&ci->i_unsafe_lock);
1263	mutex_unlock(&inode->i_mutex);
1264
1265	return ret;
1266}
1267
1268/*
1269 * We maintain a private dentry LRU.
1270 *
1271 * FIXME: this needs to be changed to a per-mds lru to be useful.
1272 */
1273void ceph_dentry_lru_add(struct dentry *dn)
1274{
1275	struct ceph_dentry_info *di = ceph_dentry(dn);
1276	struct ceph_mds_client *mdsc;
1277
1278	dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1279	     dn->d_name.len, dn->d_name.name);
1280	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1281	spin_lock(&mdsc->dentry_lru_lock);
1282	list_add_tail(&di->lru, &mdsc->dentry_lru);
1283	mdsc->num_dentry++;
1284	spin_unlock(&mdsc->dentry_lru_lock);
1285}
1286
1287void ceph_dentry_lru_touch(struct dentry *dn)
1288{
1289	struct ceph_dentry_info *di = ceph_dentry(dn);
1290	struct ceph_mds_client *mdsc;
1291
1292	dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1293	     dn->d_name.len, dn->d_name.name, di->offset);
1294	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1295	spin_lock(&mdsc->dentry_lru_lock);
1296	list_move_tail(&di->lru, &mdsc->dentry_lru);
1297	spin_unlock(&mdsc->dentry_lru_lock);
1298}
1299
1300void ceph_dentry_lru_del(struct dentry *dn)
1301{
1302	struct ceph_dentry_info *di = ceph_dentry(dn);
1303	struct ceph_mds_client *mdsc;
1304
1305	dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1306	     dn->d_name.len, dn->d_name.name);
1307	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1308	spin_lock(&mdsc->dentry_lru_lock);
1309	list_del_init(&di->lru);
1310	mdsc->num_dentry--;
1311	spin_unlock(&mdsc->dentry_lru_lock);
1312}
1313
1314/*
1315 * Return name hash for a given dentry.  This is dependent on
1316 * the parent directory's hash function.
1317 */
1318unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1319{
1320	struct ceph_inode_info *dci = ceph_inode(dir);
1321
1322	switch (dci->i_dir_layout.dl_dir_hash) {
1323	case 0:	/* for backward compat */
1324	case CEPH_STR_HASH_LINUX:
1325		return dn->d_name.hash;
1326
1327	default:
1328		return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1329				     dn->d_name.name, dn->d_name.len);
1330	}
1331}
1332
1333const struct file_operations ceph_dir_fops = {
1334	.read = ceph_read_dir,
1335	.readdir = ceph_readdir,
1336	.llseek = ceph_dir_llseek,
1337	.open = ceph_open,
1338	.release = ceph_release,
1339	.unlocked_ioctl = ceph_ioctl,
1340	.fsync = ceph_dir_fsync,
1341};
1342
1343const struct inode_operations ceph_dir_iops = {
1344	.lookup = ceph_lookup,
1345	.permission = ceph_permission,
1346	.getattr = ceph_getattr,
1347	.setattr = ceph_setattr,
1348	.setxattr = ceph_setxattr,
1349	.getxattr = ceph_getxattr,
1350	.listxattr = ceph_listxattr,
1351	.removexattr = ceph_removexattr,
 
 
1352	.mknod = ceph_mknod,
1353	.symlink = ceph_symlink,
1354	.mkdir = ceph_mkdir,
1355	.link = ceph_link,
1356	.unlink = ceph_unlink,
1357	.rmdir = ceph_unlink,
1358	.rename = ceph_rename,
1359	.create = ceph_create,
 
1360};
1361
1362const struct dentry_operations ceph_dentry_ops = {
1363	.d_revalidate = ceph_d_revalidate,
1364	.d_release = ceph_d_release,
1365	.d_prune = ceph_d_prune,
1366};
1367
1368const struct dentry_operations ceph_snapdir_dentry_ops = {
1369	.d_revalidate = ceph_snapdir_d_revalidate,
1370	.d_release = ceph_d_release,
1371};
1372
1373const struct dentry_operations ceph_snap_dentry_ops = {
1374	.d_release = ceph_d_release,
1375	.d_prune = ceph_d_prune,
1376};