Merge branch 'v6v7' into devel
[linux/fpc-iii.git] / fs / ceph / dir.c
blob099a58615b90f73e29368ffba4793e5a17cb66ae
1 #include <linux/ceph/ceph_debug.h>
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>
9 #include "super.h"
10 #include "mds_client.h"
13 * Directory operations: readdir, lookup, create, link, unlink,
14 * rename, etc.
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.
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.
29 const struct inode_operations ceph_dir_iops;
30 const struct file_operations ceph_dir_fops;
31 const struct dentry_operations ceph_dentry_ops;
34 * Initialize ceph dentry state.
36 int ceph_init_dentry(struct dentry *dentry)
38 struct ceph_dentry_info *di;
40 if (dentry->d_fsdata)
41 return 0;
43 if (dentry->d_parent == NULL || /* nfs fh_to_dentry */
44 ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
45 d_set_d_op(dentry, &ceph_dentry_ops);
46 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
47 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
48 else
49 d_set_d_op(dentry, &ceph_snap_dentry_ops);
51 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
52 if (!di)
53 return -ENOMEM; /* oh well */
55 spin_lock(&dentry->d_lock);
56 if (dentry->d_fsdata) {
57 /* lost a race */
58 kmem_cache_free(ceph_dentry_cachep, di);
59 goto out_unlock;
61 di->dentry = dentry;
62 di->lease_session = NULL;
63 dentry->d_fsdata = di;
64 dentry->d_time = jiffies;
65 ceph_dentry_lru_add(dentry);
66 out_unlock:
67 spin_unlock(&dentry->d_lock);
68 return 0;
74 * for readdir, we encode the directory frag and offset within that
75 * frag into f_pos.
77 static unsigned fpos_frag(loff_t p)
79 return p >> 32;
81 static unsigned fpos_off(loff_t p)
83 return p & 0xffffffff;
87 * When possible, we try to satisfy a readdir by peeking at the
88 * dcache. We make this work by carefully ordering dentries on
89 * d_u.d_child when we initially get results back from the MDS, and
90 * falling back to a "normal" sync readdir if any dentries in the dir
91 * are dropped.
93 * I_COMPLETE tells indicates we have all dentries in the dir. It is
94 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
95 * the MDS if/when the directory is modified).
97 static int __dcache_readdir(struct file *filp,
98 void *dirent, filldir_t filldir)
100 struct ceph_file_info *fi = filp->private_data;
101 struct dentry *parent = filp->f_dentry;
102 struct inode *dir = parent->d_inode;
103 struct list_head *p;
104 struct dentry *dentry, *last;
105 struct ceph_dentry_info *di;
106 int err = 0;
108 /* claim ref on last dentry we returned */
109 last = fi->dentry;
110 fi->dentry = NULL;
112 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
113 last);
115 spin_lock(&parent->d_lock);
117 /* start at beginning? */
118 if (filp->f_pos == 2 || last == NULL ||
119 filp->f_pos < ceph_dentry(last)->offset) {
120 if (list_empty(&parent->d_subdirs))
121 goto out_unlock;
122 p = parent->d_subdirs.prev;
123 dout(" initial p %p/%p\n", p->prev, p->next);
124 } else {
125 p = last->d_u.d_child.prev;
128 more:
129 dentry = list_entry(p, struct dentry, d_u.d_child);
130 di = ceph_dentry(dentry);
131 while (1) {
132 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
133 d_unhashed(dentry) ? "!hashed" : "hashed",
134 parent->d_subdirs.prev, parent->d_subdirs.next);
135 if (p == &parent->d_subdirs) {
136 fi->at_end = 1;
137 goto out_unlock;
139 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
140 if (!d_unhashed(dentry) && dentry->d_inode &&
141 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
142 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
143 filp->f_pos <= di->offset)
144 break;
145 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
146 dentry->d_name.len, dentry->d_name.name, di->offset,
147 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
148 !dentry->d_inode ? " null" : "");
149 spin_unlock(&dentry->d_lock);
150 p = p->prev;
151 dentry = list_entry(p, struct dentry, d_u.d_child);
152 di = ceph_dentry(dentry);
155 dget_dlock(dentry);
156 spin_unlock(&dentry->d_lock);
157 spin_unlock(&parent->d_lock);
159 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
160 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
161 filp->f_pos = di->offset;
162 err = filldir(dirent, dentry->d_name.name,
163 dentry->d_name.len, di->offset,
164 dentry->d_inode->i_ino,
165 dentry->d_inode->i_mode >> 12);
167 if (last) {
168 if (err < 0) {
169 /* remember our position */
170 fi->dentry = last;
171 fi->next_offset = di->offset;
172 } else {
173 dput(last);
176 last = dentry;
178 if (err < 0)
179 goto out;
181 filp->f_pos++;
183 /* make sure a dentry wasn't dropped while we didn't have parent lock */
184 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
185 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
186 err = -EAGAIN;
187 goto out;
190 spin_lock(&parent->d_lock);
191 p = p->prev; /* advance to next dentry */
192 goto more;
194 out_unlock:
195 spin_unlock(&parent->d_lock);
196 out:
197 if (last)
198 dput(last);
199 return err;
203 * make note of the last dentry we read, so we can
204 * continue at the same lexicographical point,
205 * regardless of what dir changes take place on the
206 * server.
208 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
209 int len)
211 kfree(fi->last_name);
212 fi->last_name = kmalloc(len+1, GFP_NOFS);
213 if (!fi->last_name)
214 return -ENOMEM;
215 memcpy(fi->last_name, name, len);
216 fi->last_name[len] = 0;
217 dout("note_last_dentry '%s'\n", fi->last_name);
218 return 0;
221 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
223 struct ceph_file_info *fi = filp->private_data;
224 struct inode *inode = filp->f_dentry->d_inode;
225 struct ceph_inode_info *ci = ceph_inode(inode);
226 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
227 struct ceph_mds_client *mdsc = fsc->mdsc;
228 unsigned frag = fpos_frag(filp->f_pos);
229 int off = fpos_off(filp->f_pos);
230 int err;
231 u32 ftype;
232 struct ceph_mds_reply_info_parsed *rinfo;
233 const int max_entries = fsc->mount_options->max_readdir;
234 const int max_bytes = fsc->mount_options->max_readdir_bytes;
236 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
237 if (fi->at_end)
238 return 0;
240 /* always start with . and .. */
241 if (filp->f_pos == 0) {
242 /* note dir version at start of readdir so we can tell
243 * if any dentries get dropped */
244 fi->dir_release_count = ci->i_release_count;
246 dout("readdir off 0 -> '.'\n");
247 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
248 inode->i_ino, inode->i_mode >> 12) < 0)
249 return 0;
250 filp->f_pos = 1;
251 off = 1;
253 if (filp->f_pos == 1) {
254 dout("readdir off 1 -> '..'\n");
255 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
256 filp->f_dentry->d_parent->d_inode->i_ino,
257 inode->i_mode >> 12) < 0)
258 return 0;
259 filp->f_pos = 2;
260 off = 2;
263 /* can we use the dcache? */
264 spin_lock(&inode->i_lock);
265 if ((filp->f_pos == 2 || fi->dentry) &&
266 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
267 ceph_snap(inode) != CEPH_SNAPDIR &&
268 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
269 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
270 spin_unlock(&inode->i_lock);
271 err = __dcache_readdir(filp, dirent, filldir);
272 if (err != -EAGAIN)
273 return err;
274 } else {
275 spin_unlock(&inode->i_lock);
277 if (fi->dentry) {
278 err = note_last_dentry(fi, fi->dentry->d_name.name,
279 fi->dentry->d_name.len);
280 if (err)
281 return err;
282 dput(fi->dentry);
283 fi->dentry = NULL;
286 /* proceed with a normal readdir */
288 more:
289 /* do we have the correct frag content buffered? */
290 if (fi->frag != frag || fi->last_readdir == NULL) {
291 struct ceph_mds_request *req;
292 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
293 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
295 /* discard old result, if any */
296 if (fi->last_readdir) {
297 ceph_mdsc_put_request(fi->last_readdir);
298 fi->last_readdir = NULL;
301 /* requery frag tree, as the frag topology may have changed */
302 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
304 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
305 ceph_vinop(inode), frag, fi->last_name);
306 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
307 if (IS_ERR(req))
308 return PTR_ERR(req);
309 req->r_inode = igrab(inode);
310 req->r_dentry = dget(filp->f_dentry);
311 /* hints to request -> mds selection code */
312 req->r_direct_mode = USE_AUTH_MDS;
313 req->r_direct_hash = ceph_frag_value(frag);
314 req->r_direct_is_hash = true;
315 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
316 req->r_readdir_offset = fi->next_offset;
317 req->r_args.readdir.frag = cpu_to_le32(frag);
318 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
319 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
320 req->r_num_caps = max_entries + 1;
321 err = ceph_mdsc_do_request(mdsc, NULL, req);
322 if (err < 0) {
323 ceph_mdsc_put_request(req);
324 return err;
326 dout("readdir got and parsed readdir result=%d"
327 " on frag %x, end=%d, complete=%d\n", err, frag,
328 (int)req->r_reply_info.dir_end,
329 (int)req->r_reply_info.dir_complete);
331 if (!req->r_did_prepopulate) {
332 dout("readdir !did_prepopulate");
333 fi->dir_release_count--; /* preclude I_COMPLETE */
336 /* note next offset and last dentry name */
337 fi->offset = fi->next_offset;
338 fi->last_readdir = req;
340 if (req->r_reply_info.dir_end) {
341 kfree(fi->last_name);
342 fi->last_name = NULL;
343 if (ceph_frag_is_rightmost(frag))
344 fi->next_offset = 2;
345 else
346 fi->next_offset = 0;
347 } else {
348 rinfo = &req->r_reply_info;
349 err = note_last_dentry(fi,
350 rinfo->dir_dname[rinfo->dir_nr-1],
351 rinfo->dir_dname_len[rinfo->dir_nr-1]);
352 if (err)
353 return err;
354 fi->next_offset += rinfo->dir_nr;
358 rinfo = &fi->last_readdir->r_reply_info;
359 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
360 rinfo->dir_nr, off, fi->offset);
361 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
362 u64 pos = ceph_make_fpos(frag, off);
363 struct ceph_mds_reply_inode *in =
364 rinfo->dir_in[off - fi->offset].in;
365 struct ceph_vino vino;
366 ino_t ino;
368 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
369 off, off - fi->offset, rinfo->dir_nr, pos,
370 rinfo->dir_dname_len[off - fi->offset],
371 rinfo->dir_dname[off - fi->offset], in);
372 BUG_ON(!in);
373 ftype = le32_to_cpu(in->mode) >> 12;
374 vino.ino = le64_to_cpu(in->ino);
375 vino.snap = le64_to_cpu(in->snapid);
376 ino = ceph_vino_to_ino(vino);
377 if (filldir(dirent,
378 rinfo->dir_dname[off - fi->offset],
379 rinfo->dir_dname_len[off - fi->offset],
380 pos, ino, ftype) < 0) {
381 dout("filldir stopping us...\n");
382 return 0;
384 off++;
385 filp->f_pos = pos + 1;
388 if (fi->last_name) {
389 ceph_mdsc_put_request(fi->last_readdir);
390 fi->last_readdir = NULL;
391 goto more;
394 /* more frags? */
395 if (!ceph_frag_is_rightmost(frag)) {
396 frag = ceph_frag_next(frag);
397 off = 0;
398 filp->f_pos = ceph_make_fpos(frag, off);
399 dout("readdir next frag is %x\n", frag);
400 goto more;
402 fi->at_end = 1;
405 * if dir_release_count still matches the dir, no dentries
406 * were released during the whole readdir, and we should have
407 * the complete dir contents in our cache.
409 spin_lock(&inode->i_lock);
410 if (ci->i_release_count == fi->dir_release_count) {
411 dout(" marking %p complete\n", inode);
412 /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */
413 ci->i_max_offset = filp->f_pos;
415 spin_unlock(&inode->i_lock);
417 dout("readdir %p filp %p done.\n", inode, filp);
418 return 0;
421 static void reset_readdir(struct ceph_file_info *fi)
423 if (fi->last_readdir) {
424 ceph_mdsc_put_request(fi->last_readdir);
425 fi->last_readdir = NULL;
427 kfree(fi->last_name);
428 fi->last_name = NULL;
429 fi->next_offset = 2; /* compensate for . and .. */
430 if (fi->dentry) {
431 dput(fi->dentry);
432 fi->dentry = NULL;
434 fi->at_end = 0;
437 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
439 struct ceph_file_info *fi = file->private_data;
440 struct inode *inode = file->f_mapping->host;
441 loff_t old_offset = offset;
442 loff_t retval;
444 mutex_lock(&inode->i_mutex);
445 switch (origin) {
446 case SEEK_END:
447 offset += inode->i_size + 2; /* FIXME */
448 break;
449 case SEEK_CUR:
450 offset += file->f_pos;
452 retval = -EINVAL;
453 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
454 if (offset != file->f_pos) {
455 file->f_pos = offset;
456 file->f_version = 0;
457 fi->at_end = 0;
459 retval = offset;
462 * discard buffered readdir content on seekdir(0), or
463 * seek to new frag, or seek prior to current chunk.
465 if (offset == 0 ||
466 fpos_frag(offset) != fpos_frag(old_offset) ||
467 fpos_off(offset) < fi->offset) {
468 dout("dir_llseek dropping %p content\n", file);
469 reset_readdir(fi);
472 /* bump dir_release_count if we did a forward seek */
473 if (offset > old_offset)
474 fi->dir_release_count--;
476 mutex_unlock(&inode->i_mutex);
477 return retval;
481 * Process result of a lookup/open request.
483 * Mainly, make sure we return the final req->r_dentry (if it already
484 * existed) in place of the original VFS-provided dentry when they
485 * differ.
487 * Gracefully handle the case where the MDS replies with -ENOENT and
488 * no trace (which it may do, at its discretion, e.g., if it doesn't
489 * care to issue a lease on the negative dentry).
491 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
492 struct dentry *dentry, int err)
494 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
495 struct inode *parent = dentry->d_parent->d_inode;
497 /* .snap dir? */
498 if (err == -ENOENT &&
499 ceph_snap(parent) == CEPH_NOSNAP &&
500 strcmp(dentry->d_name.name,
501 fsc->mount_options->snapdir_name) == 0) {
502 struct inode *inode = ceph_get_snapdir(parent);
503 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
504 dentry, dentry->d_name.len, dentry->d_name.name, inode);
505 BUG_ON(!d_unhashed(dentry));
506 d_add(dentry, inode);
507 err = 0;
510 if (err == -ENOENT) {
511 /* no trace? */
512 err = 0;
513 if (!req->r_reply_info.head->is_dentry) {
514 dout("ENOENT and no trace, dentry %p inode %p\n",
515 dentry, dentry->d_inode);
516 if (dentry->d_inode) {
517 d_drop(dentry);
518 err = -ENOENT;
519 } else {
520 d_add(dentry, NULL);
524 if (err)
525 dentry = ERR_PTR(err);
526 else if (dentry != req->r_dentry)
527 dentry = dget(req->r_dentry); /* we got spliced */
528 else
529 dentry = NULL;
530 return dentry;
533 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
535 return ceph_ino(inode) == CEPH_INO_ROOT &&
536 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
540 * Look up a single dir entry. If there is a lookup intent, inform
541 * the MDS so that it gets our 'caps wanted' value in a single op.
543 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
544 struct nameidata *nd)
546 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
547 struct ceph_mds_client *mdsc = fsc->mdsc;
548 struct ceph_mds_request *req;
549 int op;
550 int err;
552 dout("lookup %p dentry %p '%.*s'\n",
553 dir, dentry, dentry->d_name.len, dentry->d_name.name);
555 if (dentry->d_name.len > NAME_MAX)
556 return ERR_PTR(-ENAMETOOLONG);
558 err = ceph_init_dentry(dentry);
559 if (err < 0)
560 return ERR_PTR(err);
562 /* open (but not create!) intent? */
563 if (nd &&
564 (nd->flags & LOOKUP_OPEN) &&
565 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
566 !(nd->intent.open.flags & O_CREAT)) {
567 int mode = nd->intent.open.create_mode & ~current->fs->umask;
568 return ceph_lookup_open(dir, dentry, nd, mode, 1);
571 /* can we conclude ENOENT locally? */
572 if (dentry->d_inode == NULL) {
573 struct ceph_inode_info *ci = ceph_inode(dir);
574 struct ceph_dentry_info *di = ceph_dentry(dentry);
576 spin_lock(&dir->i_lock);
577 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
578 if (strncmp(dentry->d_name.name,
579 fsc->mount_options->snapdir_name,
580 dentry->d_name.len) &&
581 !is_root_ceph_dentry(dir, dentry) &&
582 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
583 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
584 spin_unlock(&dir->i_lock);
585 dout(" dir %p complete, -ENOENT\n", dir);
586 d_add(dentry, NULL);
587 di->lease_shared_gen = ci->i_shared_gen;
588 return NULL;
590 spin_unlock(&dir->i_lock);
593 op = ceph_snap(dir) == CEPH_SNAPDIR ?
594 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
595 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
596 if (IS_ERR(req))
597 return ERR_CAST(req);
598 req->r_dentry = dget(dentry);
599 req->r_num_caps = 2;
600 /* we only need inode linkage */
601 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
602 req->r_locked_dir = dir;
603 err = ceph_mdsc_do_request(mdsc, NULL, req);
604 dentry = ceph_finish_lookup(req, dentry, err);
605 ceph_mdsc_put_request(req); /* will dput(dentry) */
606 dout("lookup result=%p\n", dentry);
607 return dentry;
611 * If we do a create but get no trace back from the MDS, follow up with
612 * a lookup (the VFS expects us to link up the provided dentry).
614 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
616 struct dentry *result = ceph_lookup(dir, dentry, NULL);
618 if (result && !IS_ERR(result)) {
620 * We created the item, then did a lookup, and found
621 * it was already linked to another inode we already
622 * had in our cache (and thus got spliced). Link our
623 * dentry to that inode, but don't hash it, just in
624 * case the VFS wants to dereference it.
626 BUG_ON(!result->d_inode);
627 d_instantiate(dentry, result->d_inode);
628 return 0;
630 return PTR_ERR(result);
633 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
634 int mode, dev_t rdev)
636 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
637 struct ceph_mds_client *mdsc = fsc->mdsc;
638 struct ceph_mds_request *req;
639 int err;
641 if (ceph_snap(dir) != CEPH_NOSNAP)
642 return -EROFS;
644 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
645 dir, dentry, mode, rdev);
646 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
647 if (IS_ERR(req)) {
648 d_drop(dentry);
649 return PTR_ERR(req);
651 req->r_dentry = dget(dentry);
652 req->r_num_caps = 2;
653 req->r_locked_dir = dir;
654 req->r_args.mknod.mode = cpu_to_le32(mode);
655 req->r_args.mknod.rdev = cpu_to_le32(rdev);
656 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
657 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
658 err = ceph_mdsc_do_request(mdsc, dir, req);
659 if (!err && !req->r_reply_info.head->is_dentry)
660 err = ceph_handle_notrace_create(dir, dentry);
661 ceph_mdsc_put_request(req);
662 if (err)
663 d_drop(dentry);
664 return err;
667 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
668 struct nameidata *nd)
670 dout("create in dir %p dentry %p name '%.*s'\n",
671 dir, dentry, dentry->d_name.len, dentry->d_name.name);
673 if (ceph_snap(dir) != CEPH_NOSNAP)
674 return -EROFS;
676 if (nd) {
677 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
678 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
679 /* hrm, what should i do here if we get aliased? */
680 if (IS_ERR(dentry))
681 return PTR_ERR(dentry);
682 return 0;
685 /* fall back to mknod */
686 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
689 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
690 const char *dest)
692 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
693 struct ceph_mds_client *mdsc = fsc->mdsc;
694 struct ceph_mds_request *req;
695 int err;
697 if (ceph_snap(dir) != CEPH_NOSNAP)
698 return -EROFS;
700 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
701 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
702 if (IS_ERR(req)) {
703 d_drop(dentry);
704 return PTR_ERR(req);
706 req->r_dentry = dget(dentry);
707 req->r_num_caps = 2;
708 req->r_path2 = kstrdup(dest, GFP_NOFS);
709 req->r_locked_dir = dir;
710 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
711 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
712 err = ceph_mdsc_do_request(mdsc, dir, req);
713 if (!err && !req->r_reply_info.head->is_dentry)
714 err = ceph_handle_notrace_create(dir, dentry);
715 ceph_mdsc_put_request(req);
716 if (err)
717 d_drop(dentry);
718 return err;
721 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
723 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
724 struct ceph_mds_client *mdsc = fsc->mdsc;
725 struct ceph_mds_request *req;
726 int err = -EROFS;
727 int op;
729 if (ceph_snap(dir) == CEPH_SNAPDIR) {
730 /* mkdir .snap/foo is a MKSNAP */
731 op = CEPH_MDS_OP_MKSNAP;
732 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
733 dentry->d_name.len, dentry->d_name.name, dentry);
734 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
735 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
736 op = CEPH_MDS_OP_MKDIR;
737 } else {
738 goto out;
740 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
741 if (IS_ERR(req)) {
742 err = PTR_ERR(req);
743 goto out;
746 req->r_dentry = dget(dentry);
747 req->r_num_caps = 2;
748 req->r_locked_dir = dir;
749 req->r_args.mkdir.mode = cpu_to_le32(mode);
750 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
751 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
752 err = ceph_mdsc_do_request(mdsc, dir, req);
753 if (!err && !req->r_reply_info.head->is_dentry)
754 err = ceph_handle_notrace_create(dir, dentry);
755 ceph_mdsc_put_request(req);
756 out:
757 if (err < 0)
758 d_drop(dentry);
759 return err;
762 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
763 struct dentry *dentry)
765 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
766 struct ceph_mds_client *mdsc = fsc->mdsc;
767 struct ceph_mds_request *req;
768 int err;
770 if (ceph_snap(dir) != CEPH_NOSNAP)
771 return -EROFS;
773 dout("link in dir %p old_dentry %p dentry %p\n", dir,
774 old_dentry, dentry);
775 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
776 if (IS_ERR(req)) {
777 d_drop(dentry);
778 return PTR_ERR(req);
780 req->r_dentry = dget(dentry);
781 req->r_num_caps = 2;
782 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
783 req->r_locked_dir = dir;
784 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
785 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
786 err = ceph_mdsc_do_request(mdsc, dir, req);
787 if (err)
788 d_drop(dentry);
789 else if (!req->r_reply_info.head->is_dentry)
790 d_instantiate(dentry, igrab(old_dentry->d_inode));
791 ceph_mdsc_put_request(req);
792 return err;
796 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
797 * looks like the link count will hit 0, drop any other caps (other
798 * than PIN) we don't specifically want (due to the file still being
799 * open).
801 static int drop_caps_for_unlink(struct inode *inode)
803 struct ceph_inode_info *ci = ceph_inode(inode);
804 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
806 spin_lock(&inode->i_lock);
807 if (inode->i_nlink == 1) {
808 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
809 ci->i_ceph_flags |= CEPH_I_NODELAY;
811 spin_unlock(&inode->i_lock);
812 return drop;
816 * rmdir and unlink are differ only by the metadata op code
818 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
820 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
821 struct ceph_mds_client *mdsc = fsc->mdsc;
822 struct inode *inode = dentry->d_inode;
823 struct ceph_mds_request *req;
824 int err = -EROFS;
825 int op;
827 if (ceph_snap(dir) == CEPH_SNAPDIR) {
828 /* rmdir .snap/foo is RMSNAP */
829 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
830 dentry->d_name.name, dentry);
831 op = CEPH_MDS_OP_RMSNAP;
832 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
833 dout("unlink/rmdir dir %p dn %p inode %p\n",
834 dir, dentry, inode);
835 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
836 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
837 } else
838 goto out;
839 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
840 if (IS_ERR(req)) {
841 err = PTR_ERR(req);
842 goto out;
844 req->r_dentry = dget(dentry);
845 req->r_num_caps = 2;
846 req->r_locked_dir = dir;
847 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
848 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
849 req->r_inode_drop = drop_caps_for_unlink(inode);
850 err = ceph_mdsc_do_request(mdsc, dir, req);
851 if (!err && !req->r_reply_info.head->is_dentry)
852 d_delete(dentry);
853 ceph_mdsc_put_request(req);
854 out:
855 return err;
858 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
859 struct inode *new_dir, struct dentry *new_dentry)
861 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
862 struct ceph_mds_client *mdsc = fsc->mdsc;
863 struct ceph_mds_request *req;
864 int err;
866 if (ceph_snap(old_dir) != ceph_snap(new_dir))
867 return -EXDEV;
868 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
869 ceph_snap(new_dir) != CEPH_NOSNAP)
870 return -EROFS;
871 dout("rename dir %p dentry %p to dir %p dentry %p\n",
872 old_dir, old_dentry, new_dir, new_dentry);
873 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
874 if (IS_ERR(req))
875 return PTR_ERR(req);
876 req->r_dentry = dget(new_dentry);
877 req->r_num_caps = 2;
878 req->r_old_dentry = dget(old_dentry);
879 req->r_locked_dir = new_dir;
880 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
881 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
882 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
883 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
884 /* release LINK_RDCACHE on source inode (mds will lock it) */
885 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
886 if (new_dentry->d_inode)
887 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
888 err = ceph_mdsc_do_request(mdsc, old_dir, req);
889 if (!err && !req->r_reply_info.head->is_dentry) {
891 * Normally d_move() is done by fill_trace (called by
892 * do_request, above). If there is no trace, we need
893 * to do it here.
896 /* d_move screws up d_subdirs order */
897 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
899 d_move(old_dentry, new_dentry);
901 /* ensure target dentry is invalidated, despite
902 rehashing bug in vfs_rename_dir */
903 ceph_invalidate_dentry_lease(new_dentry);
905 ceph_mdsc_put_request(req);
906 return err;
910 * Ensure a dentry lease will no longer revalidate.
912 void ceph_invalidate_dentry_lease(struct dentry *dentry)
914 spin_lock(&dentry->d_lock);
915 dentry->d_time = jiffies;
916 ceph_dentry(dentry)->lease_shared_gen = 0;
917 spin_unlock(&dentry->d_lock);
921 * Check if dentry lease is valid. If not, delete the lease. Try to
922 * renew if the least is more than half up.
924 static int dentry_lease_is_valid(struct dentry *dentry)
926 struct ceph_dentry_info *di;
927 struct ceph_mds_session *s;
928 int valid = 0;
929 u32 gen;
930 unsigned long ttl;
931 struct ceph_mds_session *session = NULL;
932 struct inode *dir = NULL;
933 u32 seq = 0;
935 spin_lock(&dentry->d_lock);
936 di = ceph_dentry(dentry);
937 if (di && di->lease_session) {
938 s = di->lease_session;
939 spin_lock(&s->s_cap_lock);
940 gen = s->s_cap_gen;
941 ttl = s->s_cap_ttl;
942 spin_unlock(&s->s_cap_lock);
944 if (di->lease_gen == gen &&
945 time_before(jiffies, dentry->d_time) &&
946 time_before(jiffies, ttl)) {
947 valid = 1;
948 if (di->lease_renew_after &&
949 time_after(jiffies, di->lease_renew_after)) {
950 /* we should renew */
951 dir = dentry->d_parent->d_inode;
952 session = ceph_get_mds_session(s);
953 seq = di->lease_seq;
954 di->lease_renew_after = 0;
955 di->lease_renew_from = jiffies;
959 spin_unlock(&dentry->d_lock);
961 if (session) {
962 ceph_mdsc_lease_send_msg(session, dir, dentry,
963 CEPH_MDS_LEASE_RENEW, seq);
964 ceph_put_mds_session(session);
966 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
967 return valid;
971 * Check if directory-wide content lease/cap is valid.
973 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
975 struct ceph_inode_info *ci = ceph_inode(dir);
976 struct ceph_dentry_info *di = ceph_dentry(dentry);
977 int valid = 0;
979 spin_lock(&dir->i_lock);
980 if (ci->i_shared_gen == di->lease_shared_gen)
981 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
982 spin_unlock(&dir->i_lock);
983 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
984 dir, (unsigned)ci->i_shared_gen, dentry,
985 (unsigned)di->lease_shared_gen, valid);
986 return valid;
990 * Check if cached dentry can be trusted.
992 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
994 struct inode *dir;
996 if (nd->flags & LOOKUP_RCU)
997 return -ECHILD;
999 dir = dentry->d_parent->d_inode;
1001 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1002 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1003 ceph_dentry(dentry)->offset);
1005 /* always trust cached snapped dentries, snapdir dentry */
1006 if (ceph_snap(dir) != CEPH_NOSNAP) {
1007 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1008 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1009 goto out_touch;
1011 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
1012 goto out_touch;
1014 if (dentry_lease_is_valid(dentry) ||
1015 dir_lease_is_valid(dir, dentry))
1016 goto out_touch;
1018 dout("d_revalidate %p invalid\n", dentry);
1019 d_drop(dentry);
1020 return 0;
1021 out_touch:
1022 ceph_dentry_lru_touch(dentry);
1023 return 1;
1027 * When a dentry is released, clear the dir I_COMPLETE if it was part
1028 * of the current dir gen or if this is in the snapshot namespace.
1030 static void ceph_dentry_release(struct dentry *dentry)
1032 struct ceph_dentry_info *di = ceph_dentry(dentry);
1034 dout("dentry_release %p\n", dentry);
1035 if (di) {
1036 ceph_dentry_lru_del(dentry);
1037 if (di->lease_session)
1038 ceph_put_mds_session(di->lease_session);
1039 kmem_cache_free(ceph_dentry_cachep, di);
1040 dentry->d_fsdata = NULL;
1044 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1045 struct nameidata *nd)
1048 * Eventually, we'll want to revalidate snapped metadata
1049 * too... probably...
1051 return 1;
1057 * read() on a dir. This weird interface hack only works if mounted
1058 * with '-o dirstat'.
1060 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1061 loff_t *ppos)
1063 struct ceph_file_info *cf = file->private_data;
1064 struct inode *inode = file->f_dentry->d_inode;
1065 struct ceph_inode_info *ci = ceph_inode(inode);
1066 int left;
1068 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1069 return -EISDIR;
1071 if (!cf->dir_info) {
1072 cf->dir_info = kmalloc(1024, GFP_NOFS);
1073 if (!cf->dir_info)
1074 return -ENOMEM;
1075 cf->dir_info_len =
1076 sprintf(cf->dir_info,
1077 "entries: %20lld\n"
1078 " files: %20lld\n"
1079 " subdirs: %20lld\n"
1080 "rentries: %20lld\n"
1081 " rfiles: %20lld\n"
1082 " rsubdirs: %20lld\n"
1083 "rbytes: %20lld\n"
1084 "rctime: %10ld.%09ld\n",
1085 ci->i_files + ci->i_subdirs,
1086 ci->i_files,
1087 ci->i_subdirs,
1088 ci->i_rfiles + ci->i_rsubdirs,
1089 ci->i_rfiles,
1090 ci->i_rsubdirs,
1091 ci->i_rbytes,
1092 (long)ci->i_rctime.tv_sec,
1093 (long)ci->i_rctime.tv_nsec);
1096 if (*ppos >= cf->dir_info_len)
1097 return 0;
1098 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1099 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1100 if (left == size)
1101 return -EFAULT;
1102 *ppos += (size - left);
1103 return size - left;
1107 * an fsync() on a dir will wait for any uncommitted directory
1108 * operations to commit.
1110 static int ceph_dir_fsync(struct file *file, int datasync)
1112 struct inode *inode = file->f_path.dentry->d_inode;
1113 struct ceph_inode_info *ci = ceph_inode(inode);
1114 struct list_head *head = &ci->i_unsafe_dirops;
1115 struct ceph_mds_request *req;
1116 u64 last_tid;
1117 int ret = 0;
1119 dout("dir_fsync %p\n", inode);
1120 spin_lock(&ci->i_unsafe_lock);
1121 if (list_empty(head))
1122 goto out;
1124 req = list_entry(head->prev,
1125 struct ceph_mds_request, r_unsafe_dir_item);
1126 last_tid = req->r_tid;
1128 do {
1129 ceph_mdsc_get_request(req);
1130 spin_unlock(&ci->i_unsafe_lock);
1131 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1132 inode, req->r_tid, last_tid);
1133 if (req->r_timeout) {
1134 ret = wait_for_completion_timeout(
1135 &req->r_safe_completion, req->r_timeout);
1136 if (ret > 0)
1137 ret = 0;
1138 else if (ret == 0)
1139 ret = -EIO; /* timed out */
1140 } else {
1141 wait_for_completion(&req->r_safe_completion);
1143 spin_lock(&ci->i_unsafe_lock);
1144 ceph_mdsc_put_request(req);
1146 if (ret || list_empty(head))
1147 break;
1148 req = list_entry(head->next,
1149 struct ceph_mds_request, r_unsafe_dir_item);
1150 } while (req->r_tid < last_tid);
1151 out:
1152 spin_unlock(&ci->i_unsafe_lock);
1153 return ret;
1157 * We maintain a private dentry LRU.
1159 * FIXME: this needs to be changed to a per-mds lru to be useful.
1161 void ceph_dentry_lru_add(struct dentry *dn)
1163 struct ceph_dentry_info *di = ceph_dentry(dn);
1164 struct ceph_mds_client *mdsc;
1166 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1167 dn->d_name.len, dn->d_name.name);
1168 if (di) {
1169 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1170 spin_lock(&mdsc->dentry_lru_lock);
1171 list_add_tail(&di->lru, &mdsc->dentry_lru);
1172 mdsc->num_dentry++;
1173 spin_unlock(&mdsc->dentry_lru_lock);
1177 void ceph_dentry_lru_touch(struct dentry *dn)
1179 struct ceph_dentry_info *di = ceph_dentry(dn);
1180 struct ceph_mds_client *mdsc;
1182 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1183 dn->d_name.len, dn->d_name.name, di->offset);
1184 if (di) {
1185 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1186 spin_lock(&mdsc->dentry_lru_lock);
1187 list_move_tail(&di->lru, &mdsc->dentry_lru);
1188 spin_unlock(&mdsc->dentry_lru_lock);
1192 void ceph_dentry_lru_del(struct dentry *dn)
1194 struct ceph_dentry_info *di = ceph_dentry(dn);
1195 struct ceph_mds_client *mdsc;
1197 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1198 dn->d_name.len, dn->d_name.name);
1199 if (di) {
1200 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1201 spin_lock(&mdsc->dentry_lru_lock);
1202 list_del_init(&di->lru);
1203 mdsc->num_dentry--;
1204 spin_unlock(&mdsc->dentry_lru_lock);
1209 * Return name hash for a given dentry. This is dependent on
1210 * the parent directory's hash function.
1212 unsigned ceph_dentry_hash(struct dentry *dn)
1214 struct inode *dir = dn->d_parent->d_inode;
1215 struct ceph_inode_info *dci = ceph_inode(dir);
1217 switch (dci->i_dir_layout.dl_dir_hash) {
1218 case 0: /* for backward compat */
1219 case CEPH_STR_HASH_LINUX:
1220 return dn->d_name.hash;
1222 default:
1223 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1224 dn->d_name.name, dn->d_name.len);
1228 const struct file_operations ceph_dir_fops = {
1229 .read = ceph_read_dir,
1230 .readdir = ceph_readdir,
1231 .llseek = ceph_dir_llseek,
1232 .open = ceph_open,
1233 .release = ceph_release,
1234 .unlocked_ioctl = ceph_ioctl,
1235 .fsync = ceph_dir_fsync,
1238 const struct inode_operations ceph_dir_iops = {
1239 .lookup = ceph_lookup,
1240 .permission = ceph_permission,
1241 .getattr = ceph_getattr,
1242 .setattr = ceph_setattr,
1243 .setxattr = ceph_setxattr,
1244 .getxattr = ceph_getxattr,
1245 .listxattr = ceph_listxattr,
1246 .removexattr = ceph_removexattr,
1247 .mknod = ceph_mknod,
1248 .symlink = ceph_symlink,
1249 .mkdir = ceph_mkdir,
1250 .link = ceph_link,
1251 .unlink = ceph_unlink,
1252 .rmdir = ceph_unlink,
1253 .rename = ceph_rename,
1254 .create = ceph_create,
1257 const struct dentry_operations ceph_dentry_ops = {
1258 .d_revalidate = ceph_d_revalidate,
1259 .d_release = ceph_dentry_release,
1262 const struct dentry_operations ceph_snapdir_dentry_ops = {
1263 .d_revalidate = ceph_snapdir_d_revalidate,
1264 .d_release = ceph_dentry_release,
1267 const struct dentry_operations ceph_snap_dentry_ops = {
1268 .d_release = ceph_dentry_release,