IB/qib: Unnecessary delayed completions on RC connection
[linux/fpc-iii.git] / fs / ceph / dir.c
blobd902948a90d88622e9af8bbfc505a779fe735d99
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 dentry->d_op = &ceph_dentry_ops;
46 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
47 dentry->d_op = &ceph_snapdir_dentry_ops;
48 else
49 dentry->d_op = &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(&dcache_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 if (!d_unhashed(dentry) && dentry->d_inode &&
140 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
141 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
142 filp->f_pos <= di->offset)
143 break;
144 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
145 dentry->d_name.len, dentry->d_name.name, di->offset,
146 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
147 !dentry->d_inode ? " null" : "");
148 p = p->prev;
149 dentry = list_entry(p, struct dentry, d_u.d_child);
150 di = ceph_dentry(dentry);
153 atomic_inc(&dentry->d_count);
154 spin_unlock(&dcache_lock);
156 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
157 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
158 filp->f_pos = di->offset;
159 err = filldir(dirent, dentry->d_name.name,
160 dentry->d_name.len, di->offset,
161 dentry->d_inode->i_ino,
162 dentry->d_inode->i_mode >> 12);
164 if (last) {
165 if (err < 0) {
166 /* remember our position */
167 fi->dentry = last;
168 fi->next_offset = di->offset;
169 } else {
170 dput(last);
173 last = dentry;
175 if (err < 0)
176 goto out;
178 filp->f_pos++;
180 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
181 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
182 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
183 err = -EAGAIN;
184 goto out;
187 spin_lock(&dcache_lock);
188 p = p->prev; /* advance to next dentry */
189 goto more;
191 out_unlock:
192 spin_unlock(&dcache_lock);
193 out:
194 if (last)
195 dput(last);
196 return err;
200 * make note of the last dentry we read, so we can
201 * continue at the same lexicographical point,
202 * regardless of what dir changes take place on the
203 * server.
205 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
206 int len)
208 kfree(fi->last_name);
209 fi->last_name = kmalloc(len+1, GFP_NOFS);
210 if (!fi->last_name)
211 return -ENOMEM;
212 memcpy(fi->last_name, name, len);
213 fi->last_name[len] = 0;
214 dout("note_last_dentry '%s'\n", fi->last_name);
215 return 0;
218 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
220 struct ceph_file_info *fi = filp->private_data;
221 struct inode *inode = filp->f_dentry->d_inode;
222 struct ceph_inode_info *ci = ceph_inode(inode);
223 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
224 struct ceph_mds_client *mdsc = fsc->mdsc;
225 unsigned frag = fpos_frag(filp->f_pos);
226 int off = fpos_off(filp->f_pos);
227 int err;
228 u32 ftype;
229 struct ceph_mds_reply_info_parsed *rinfo;
230 const int max_entries = fsc->mount_options->max_readdir;
231 const int max_bytes = fsc->mount_options->max_readdir_bytes;
233 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
234 if (fi->at_end)
235 return 0;
237 /* always start with . and .. */
238 if (filp->f_pos == 0) {
239 /* note dir version at start of readdir so we can tell
240 * if any dentries get dropped */
241 fi->dir_release_count = ci->i_release_count;
243 dout("readdir off 0 -> '.'\n");
244 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
245 inode->i_ino, inode->i_mode >> 12) < 0)
246 return 0;
247 filp->f_pos = 1;
248 off = 1;
250 if (filp->f_pos == 1) {
251 dout("readdir off 1 -> '..'\n");
252 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
253 filp->f_dentry->d_parent->d_inode->i_ino,
254 inode->i_mode >> 12) < 0)
255 return 0;
256 filp->f_pos = 2;
257 off = 2;
260 /* can we use the dcache? */
261 spin_lock(&inode->i_lock);
262 if ((filp->f_pos == 2 || fi->dentry) &&
263 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
264 ceph_snap(inode) != CEPH_SNAPDIR &&
265 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
266 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
267 spin_unlock(&inode->i_lock);
268 err = __dcache_readdir(filp, dirent, filldir);
269 if (err != -EAGAIN)
270 return err;
271 } else {
272 spin_unlock(&inode->i_lock);
274 if (fi->dentry) {
275 err = note_last_dentry(fi, fi->dentry->d_name.name,
276 fi->dentry->d_name.len);
277 if (err)
278 return err;
279 dput(fi->dentry);
280 fi->dentry = NULL;
283 /* proceed with a normal readdir */
285 more:
286 /* do we have the correct frag content buffered? */
287 if (fi->frag != frag || fi->last_readdir == NULL) {
288 struct ceph_mds_request *req;
289 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
290 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
292 /* discard old result, if any */
293 if (fi->last_readdir) {
294 ceph_mdsc_put_request(fi->last_readdir);
295 fi->last_readdir = NULL;
298 /* requery frag tree, as the frag topology may have changed */
299 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
301 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
302 ceph_vinop(inode), frag, fi->last_name);
303 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
304 if (IS_ERR(req))
305 return PTR_ERR(req);
306 req->r_inode = igrab(inode);
307 req->r_dentry = dget(filp->f_dentry);
308 /* hints to request -> mds selection code */
309 req->r_direct_mode = USE_AUTH_MDS;
310 req->r_direct_hash = ceph_frag_value(frag);
311 req->r_direct_is_hash = true;
312 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
313 req->r_readdir_offset = fi->next_offset;
314 req->r_args.readdir.frag = cpu_to_le32(frag);
315 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
316 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
317 req->r_num_caps = max_entries + 1;
318 err = ceph_mdsc_do_request(mdsc, NULL, req);
319 if (err < 0) {
320 ceph_mdsc_put_request(req);
321 return err;
323 dout("readdir got and parsed readdir result=%d"
324 " on frag %x, end=%d, complete=%d\n", err, frag,
325 (int)req->r_reply_info.dir_end,
326 (int)req->r_reply_info.dir_complete);
328 if (!req->r_did_prepopulate) {
329 dout("readdir !did_prepopulate");
330 fi->dir_release_count--; /* preclude I_COMPLETE */
333 /* note next offset and last dentry name */
334 fi->offset = fi->next_offset;
335 fi->last_readdir = req;
337 if (req->r_reply_info.dir_end) {
338 kfree(fi->last_name);
339 fi->last_name = NULL;
340 if (ceph_frag_is_rightmost(frag))
341 fi->next_offset = 2;
342 else
343 fi->next_offset = 0;
344 } else {
345 rinfo = &req->r_reply_info;
346 err = note_last_dentry(fi,
347 rinfo->dir_dname[rinfo->dir_nr-1],
348 rinfo->dir_dname_len[rinfo->dir_nr-1]);
349 if (err)
350 return err;
351 fi->next_offset += rinfo->dir_nr;
355 rinfo = &fi->last_readdir->r_reply_info;
356 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
357 rinfo->dir_nr, off, fi->offset);
358 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
359 u64 pos = ceph_make_fpos(frag, off);
360 struct ceph_mds_reply_inode *in =
361 rinfo->dir_in[off - fi->offset].in;
362 struct ceph_vino vino;
363 ino_t ino;
365 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
366 off, off - fi->offset, rinfo->dir_nr, pos,
367 rinfo->dir_dname_len[off - fi->offset],
368 rinfo->dir_dname[off - fi->offset], in);
369 BUG_ON(!in);
370 ftype = le32_to_cpu(in->mode) >> 12;
371 vino.ino = le64_to_cpu(in->ino);
372 vino.snap = le64_to_cpu(in->snapid);
373 ino = ceph_vino_to_ino(vino);
374 if (filldir(dirent,
375 rinfo->dir_dname[off - fi->offset],
376 rinfo->dir_dname_len[off - fi->offset],
377 pos, ino, ftype) < 0) {
378 dout("filldir stopping us...\n");
379 return 0;
381 off++;
382 filp->f_pos = pos + 1;
385 if (fi->last_name) {
386 ceph_mdsc_put_request(fi->last_readdir);
387 fi->last_readdir = NULL;
388 goto more;
391 /* more frags? */
392 if (!ceph_frag_is_rightmost(frag)) {
393 frag = ceph_frag_next(frag);
394 off = 0;
395 filp->f_pos = ceph_make_fpos(frag, off);
396 dout("readdir next frag is %x\n", frag);
397 goto more;
399 fi->at_end = 1;
402 * if dir_release_count still matches the dir, no dentries
403 * were released during the whole readdir, and we should have
404 * the complete dir contents in our cache.
406 spin_lock(&inode->i_lock);
407 if (ci->i_release_count == fi->dir_release_count) {
408 dout(" marking %p complete\n", inode);
409 ci->i_ceph_flags |= CEPH_I_COMPLETE;
410 ci->i_max_offset = filp->f_pos;
412 spin_unlock(&inode->i_lock);
414 dout("readdir %p filp %p done.\n", inode, filp);
415 return 0;
418 static void reset_readdir(struct ceph_file_info *fi)
420 if (fi->last_readdir) {
421 ceph_mdsc_put_request(fi->last_readdir);
422 fi->last_readdir = NULL;
424 kfree(fi->last_name);
425 fi->last_name = NULL;
426 fi->next_offset = 2; /* compensate for . and .. */
427 if (fi->dentry) {
428 dput(fi->dentry);
429 fi->dentry = NULL;
431 fi->at_end = 0;
434 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
436 struct ceph_file_info *fi = file->private_data;
437 struct inode *inode = file->f_mapping->host;
438 loff_t old_offset = offset;
439 loff_t retval;
441 mutex_lock(&inode->i_mutex);
442 switch (origin) {
443 case SEEK_END:
444 offset += inode->i_size + 2; /* FIXME */
445 break;
446 case SEEK_CUR:
447 offset += file->f_pos;
449 retval = -EINVAL;
450 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
451 if (offset != file->f_pos) {
452 file->f_pos = offset;
453 file->f_version = 0;
454 fi->at_end = 0;
456 retval = offset;
459 * discard buffered readdir content on seekdir(0), or
460 * seek to new frag, or seek prior to current chunk.
462 if (offset == 0 ||
463 fpos_frag(offset) != fpos_frag(old_offset) ||
464 fpos_off(offset) < fi->offset) {
465 dout("dir_llseek dropping %p content\n", file);
466 reset_readdir(fi);
469 /* bump dir_release_count if we did a forward seek */
470 if (offset > old_offset)
471 fi->dir_release_count--;
473 mutex_unlock(&inode->i_mutex);
474 return retval;
478 * Process result of a lookup/open request.
480 * Mainly, make sure we return the final req->r_dentry (if it already
481 * existed) in place of the original VFS-provided dentry when they
482 * differ.
484 * Gracefully handle the case where the MDS replies with -ENOENT and
485 * no trace (which it may do, at its discretion, e.g., if it doesn't
486 * care to issue a lease on the negative dentry).
488 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
489 struct dentry *dentry, int err)
491 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
492 struct inode *parent = dentry->d_parent->d_inode;
494 /* .snap dir? */
495 if (err == -ENOENT &&
496 strcmp(dentry->d_name.name,
497 fsc->mount_options->snapdir_name) == 0) {
498 struct inode *inode = ceph_get_snapdir(parent);
499 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
500 dentry, dentry->d_name.len, dentry->d_name.name, inode);
501 BUG_ON(!d_unhashed(dentry));
502 d_add(dentry, inode);
503 err = 0;
506 if (err == -ENOENT) {
507 /* no trace? */
508 err = 0;
509 if (!req->r_reply_info.head->is_dentry) {
510 dout("ENOENT and no trace, dentry %p inode %p\n",
511 dentry, dentry->d_inode);
512 if (dentry->d_inode) {
513 d_drop(dentry);
514 err = -ENOENT;
515 } else {
516 d_add(dentry, NULL);
520 if (err)
521 dentry = ERR_PTR(err);
522 else if (dentry != req->r_dentry)
523 dentry = dget(req->r_dentry); /* we got spliced */
524 else
525 dentry = NULL;
526 return dentry;
529 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
531 return ceph_ino(inode) == CEPH_INO_ROOT &&
532 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
536 * Look up a single dir entry. If there is a lookup intent, inform
537 * the MDS so that it gets our 'caps wanted' value in a single op.
539 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
540 struct nameidata *nd)
542 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
543 struct ceph_mds_client *mdsc = fsc->mdsc;
544 struct ceph_mds_request *req;
545 int op;
546 int err;
548 dout("lookup %p dentry %p '%.*s'\n",
549 dir, dentry, dentry->d_name.len, dentry->d_name.name);
551 if (dentry->d_name.len > NAME_MAX)
552 return ERR_PTR(-ENAMETOOLONG);
554 err = ceph_init_dentry(dentry);
555 if (err < 0)
556 return ERR_PTR(err);
558 /* open (but not create!) intent? */
559 if (nd &&
560 (nd->flags & LOOKUP_OPEN) &&
561 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
562 !(nd->intent.open.flags & O_CREAT)) {
563 int mode = nd->intent.open.create_mode & ~current->fs->umask;
564 return ceph_lookup_open(dir, dentry, nd, mode, 1);
567 /* can we conclude ENOENT locally? */
568 if (dentry->d_inode == NULL) {
569 struct ceph_inode_info *ci = ceph_inode(dir);
570 struct ceph_dentry_info *di = ceph_dentry(dentry);
572 spin_lock(&dir->i_lock);
573 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
574 if (strncmp(dentry->d_name.name,
575 fsc->mount_options->snapdir_name,
576 dentry->d_name.len) &&
577 !is_root_ceph_dentry(dir, dentry) &&
578 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
579 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
580 spin_unlock(&dir->i_lock);
581 dout(" dir %p complete, -ENOENT\n", dir);
582 d_add(dentry, NULL);
583 di->lease_shared_gen = ci->i_shared_gen;
584 return NULL;
586 spin_unlock(&dir->i_lock);
589 op = ceph_snap(dir) == CEPH_SNAPDIR ?
590 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
591 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
592 if (IS_ERR(req))
593 return ERR_CAST(req);
594 req->r_dentry = dget(dentry);
595 req->r_num_caps = 2;
596 /* we only need inode linkage */
597 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
598 req->r_locked_dir = dir;
599 err = ceph_mdsc_do_request(mdsc, NULL, req);
600 dentry = ceph_finish_lookup(req, dentry, err);
601 ceph_mdsc_put_request(req); /* will dput(dentry) */
602 dout("lookup result=%p\n", dentry);
603 return dentry;
607 * If we do a create but get no trace back from the MDS, follow up with
608 * a lookup (the VFS expects us to link up the provided dentry).
610 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
612 struct dentry *result = ceph_lookup(dir, dentry, NULL);
614 if (result && !IS_ERR(result)) {
616 * We created the item, then did a lookup, and found
617 * it was already linked to another inode we already
618 * had in our cache (and thus got spliced). Link our
619 * dentry to that inode, but don't hash it, just in
620 * case the VFS wants to dereference it.
622 BUG_ON(!result->d_inode);
623 d_instantiate(dentry, result->d_inode);
624 return 0;
626 return PTR_ERR(result);
629 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
630 int mode, dev_t rdev)
632 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
633 struct ceph_mds_client *mdsc = fsc->mdsc;
634 struct ceph_mds_request *req;
635 int err;
637 if (ceph_snap(dir) != CEPH_NOSNAP)
638 return -EROFS;
640 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
641 dir, dentry, mode, rdev);
642 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
643 if (IS_ERR(req)) {
644 d_drop(dentry);
645 return PTR_ERR(req);
647 req->r_dentry = dget(dentry);
648 req->r_num_caps = 2;
649 req->r_locked_dir = dir;
650 req->r_args.mknod.mode = cpu_to_le32(mode);
651 req->r_args.mknod.rdev = cpu_to_le32(rdev);
652 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
653 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
654 err = ceph_mdsc_do_request(mdsc, dir, req);
655 if (!err && !req->r_reply_info.head->is_dentry)
656 err = ceph_handle_notrace_create(dir, dentry);
657 ceph_mdsc_put_request(req);
658 if (err)
659 d_drop(dentry);
660 return err;
663 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
664 struct nameidata *nd)
666 dout("create in dir %p dentry %p name '%.*s'\n",
667 dir, dentry, dentry->d_name.len, dentry->d_name.name);
669 if (ceph_snap(dir) != CEPH_NOSNAP)
670 return -EROFS;
672 if (nd) {
673 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
674 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
675 /* hrm, what should i do here if we get aliased? */
676 if (IS_ERR(dentry))
677 return PTR_ERR(dentry);
678 return 0;
681 /* fall back to mknod */
682 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
685 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
686 const char *dest)
688 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
689 struct ceph_mds_client *mdsc = fsc->mdsc;
690 struct ceph_mds_request *req;
691 int err;
693 if (ceph_snap(dir) != CEPH_NOSNAP)
694 return -EROFS;
696 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
697 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
698 if (IS_ERR(req)) {
699 d_drop(dentry);
700 return PTR_ERR(req);
702 req->r_dentry = dget(dentry);
703 req->r_num_caps = 2;
704 req->r_path2 = kstrdup(dest, GFP_NOFS);
705 req->r_locked_dir = dir;
706 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
707 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
708 err = ceph_mdsc_do_request(mdsc, dir, req);
709 if (!err && !req->r_reply_info.head->is_dentry)
710 err = ceph_handle_notrace_create(dir, dentry);
711 ceph_mdsc_put_request(req);
712 if (err)
713 d_drop(dentry);
714 return err;
717 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
719 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
720 struct ceph_mds_client *mdsc = fsc->mdsc;
721 struct ceph_mds_request *req;
722 int err = -EROFS;
723 int op;
725 if (ceph_snap(dir) == CEPH_SNAPDIR) {
726 /* mkdir .snap/foo is a MKSNAP */
727 op = CEPH_MDS_OP_MKSNAP;
728 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
729 dentry->d_name.len, dentry->d_name.name, dentry);
730 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
731 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
732 op = CEPH_MDS_OP_MKDIR;
733 } else {
734 goto out;
736 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
737 if (IS_ERR(req)) {
738 err = PTR_ERR(req);
739 goto out;
742 req->r_dentry = dget(dentry);
743 req->r_num_caps = 2;
744 req->r_locked_dir = dir;
745 req->r_args.mkdir.mode = cpu_to_le32(mode);
746 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
747 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
748 err = ceph_mdsc_do_request(mdsc, dir, req);
749 if (!err && !req->r_reply_info.head->is_dentry)
750 err = ceph_handle_notrace_create(dir, dentry);
751 ceph_mdsc_put_request(req);
752 out:
753 if (err < 0)
754 d_drop(dentry);
755 return err;
758 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
759 struct dentry *dentry)
761 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
762 struct ceph_mds_client *mdsc = fsc->mdsc;
763 struct ceph_mds_request *req;
764 int err;
766 if (ceph_snap(dir) != CEPH_NOSNAP)
767 return -EROFS;
769 dout("link in dir %p old_dentry %p dentry %p\n", dir,
770 old_dentry, dentry);
771 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
772 if (IS_ERR(req)) {
773 d_drop(dentry);
774 return PTR_ERR(req);
776 req->r_dentry = dget(dentry);
777 req->r_num_caps = 2;
778 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
779 req->r_locked_dir = dir;
780 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
781 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
782 err = ceph_mdsc_do_request(mdsc, dir, req);
783 if (err)
784 d_drop(dentry);
785 else if (!req->r_reply_info.head->is_dentry)
786 d_instantiate(dentry, igrab(old_dentry->d_inode));
787 ceph_mdsc_put_request(req);
788 return err;
792 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
793 * looks like the link count will hit 0, drop any other caps (other
794 * than PIN) we don't specifically want (due to the file still being
795 * open).
797 static int drop_caps_for_unlink(struct inode *inode)
799 struct ceph_inode_info *ci = ceph_inode(inode);
800 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
802 spin_lock(&inode->i_lock);
803 if (inode->i_nlink == 1) {
804 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
805 ci->i_ceph_flags |= CEPH_I_NODELAY;
807 spin_unlock(&inode->i_lock);
808 return drop;
812 * rmdir and unlink are differ only by the metadata op code
814 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
816 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
817 struct ceph_mds_client *mdsc = fsc->mdsc;
818 struct inode *inode = dentry->d_inode;
819 struct ceph_mds_request *req;
820 int err = -EROFS;
821 int op;
823 if (ceph_snap(dir) == CEPH_SNAPDIR) {
824 /* rmdir .snap/foo is RMSNAP */
825 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
826 dentry->d_name.name, dentry);
827 op = CEPH_MDS_OP_RMSNAP;
828 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
829 dout("unlink/rmdir dir %p dn %p inode %p\n",
830 dir, dentry, inode);
831 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
832 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
833 } else
834 goto out;
835 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
836 if (IS_ERR(req)) {
837 err = PTR_ERR(req);
838 goto out;
840 req->r_dentry = dget(dentry);
841 req->r_num_caps = 2;
842 req->r_locked_dir = dir;
843 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
844 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
845 req->r_inode_drop = drop_caps_for_unlink(inode);
846 err = ceph_mdsc_do_request(mdsc, dir, req);
847 if (!err && !req->r_reply_info.head->is_dentry)
848 d_delete(dentry);
849 ceph_mdsc_put_request(req);
850 out:
851 return err;
854 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
855 struct inode *new_dir, struct dentry *new_dentry)
857 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
858 struct ceph_mds_client *mdsc = fsc->mdsc;
859 struct ceph_mds_request *req;
860 int err;
862 if (ceph_snap(old_dir) != ceph_snap(new_dir))
863 return -EXDEV;
864 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
865 ceph_snap(new_dir) != CEPH_NOSNAP)
866 return -EROFS;
867 dout("rename dir %p dentry %p to dir %p dentry %p\n",
868 old_dir, old_dentry, new_dir, new_dentry);
869 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
870 if (IS_ERR(req))
871 return PTR_ERR(req);
872 req->r_dentry = dget(new_dentry);
873 req->r_num_caps = 2;
874 req->r_old_dentry = dget(old_dentry);
875 req->r_locked_dir = new_dir;
876 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
877 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
878 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
879 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
880 /* release LINK_RDCACHE on source inode (mds will lock it) */
881 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
882 if (new_dentry->d_inode)
883 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
884 err = ceph_mdsc_do_request(mdsc, old_dir, req);
885 if (!err && !req->r_reply_info.head->is_dentry) {
887 * Normally d_move() is done by fill_trace (called by
888 * do_request, above). If there is no trace, we need
889 * to do it here.
892 /* d_move screws up d_subdirs order */
893 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
895 d_move(old_dentry, new_dentry);
897 /* ensure target dentry is invalidated, despite
898 rehashing bug in vfs_rename_dir */
899 ceph_invalidate_dentry_lease(new_dentry);
901 ceph_mdsc_put_request(req);
902 return err;
906 * Ensure a dentry lease will no longer revalidate.
908 void ceph_invalidate_dentry_lease(struct dentry *dentry)
910 spin_lock(&dentry->d_lock);
911 dentry->d_time = jiffies;
912 ceph_dentry(dentry)->lease_shared_gen = 0;
913 spin_unlock(&dentry->d_lock);
917 * Check if dentry lease is valid. If not, delete the lease. Try to
918 * renew if the least is more than half up.
920 static int dentry_lease_is_valid(struct dentry *dentry)
922 struct ceph_dentry_info *di;
923 struct ceph_mds_session *s;
924 int valid = 0;
925 u32 gen;
926 unsigned long ttl;
927 struct ceph_mds_session *session = NULL;
928 struct inode *dir = NULL;
929 u32 seq = 0;
931 spin_lock(&dentry->d_lock);
932 di = ceph_dentry(dentry);
933 if (di && di->lease_session) {
934 s = di->lease_session;
935 spin_lock(&s->s_cap_lock);
936 gen = s->s_cap_gen;
937 ttl = s->s_cap_ttl;
938 spin_unlock(&s->s_cap_lock);
940 if (di->lease_gen == gen &&
941 time_before(jiffies, dentry->d_time) &&
942 time_before(jiffies, ttl)) {
943 valid = 1;
944 if (di->lease_renew_after &&
945 time_after(jiffies, di->lease_renew_after)) {
946 /* we should renew */
947 dir = dentry->d_parent->d_inode;
948 session = ceph_get_mds_session(s);
949 seq = di->lease_seq;
950 di->lease_renew_after = 0;
951 di->lease_renew_from = jiffies;
955 spin_unlock(&dentry->d_lock);
957 if (session) {
958 ceph_mdsc_lease_send_msg(session, dir, dentry,
959 CEPH_MDS_LEASE_RENEW, seq);
960 ceph_put_mds_session(session);
962 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
963 return valid;
967 * Check if directory-wide content lease/cap is valid.
969 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
971 struct ceph_inode_info *ci = ceph_inode(dir);
972 struct ceph_dentry_info *di = ceph_dentry(dentry);
973 int valid = 0;
975 spin_lock(&dir->i_lock);
976 if (ci->i_shared_gen == di->lease_shared_gen)
977 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
978 spin_unlock(&dir->i_lock);
979 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
980 dir, (unsigned)ci->i_shared_gen, dentry,
981 (unsigned)di->lease_shared_gen, valid);
982 return valid;
986 * Check if cached dentry can be trusted.
988 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
990 struct inode *dir = dentry->d_parent->d_inode;
992 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
993 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
994 ceph_dentry(dentry)->offset);
996 /* always trust cached snapped dentries, snapdir dentry */
997 if (ceph_snap(dir) != CEPH_NOSNAP) {
998 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
999 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1000 goto out_touch;
1002 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
1003 goto out_touch;
1005 if (dentry_lease_is_valid(dentry) ||
1006 dir_lease_is_valid(dir, dentry))
1007 goto out_touch;
1009 dout("d_revalidate %p invalid\n", dentry);
1010 d_drop(dentry);
1011 return 0;
1012 out_touch:
1013 ceph_dentry_lru_touch(dentry);
1014 return 1;
1018 * When a dentry is released, clear the dir I_COMPLETE if it was part
1019 * of the current dir gen or if this is in the snapshot namespace.
1021 static void ceph_dentry_release(struct dentry *dentry)
1023 struct ceph_dentry_info *di = ceph_dentry(dentry);
1024 struct inode *parent_inode = NULL;
1025 u64 snapid = CEPH_NOSNAP;
1027 if (!IS_ROOT(dentry)) {
1028 parent_inode = dentry->d_parent->d_inode;
1029 if (parent_inode)
1030 snapid = ceph_snap(parent_inode);
1032 dout("dentry_release %p parent %p\n", dentry, parent_inode);
1033 if (parent_inode && snapid != CEPH_SNAPDIR) {
1034 struct ceph_inode_info *ci = ceph_inode(parent_inode);
1036 spin_lock(&parent_inode->i_lock);
1037 if (ci->i_shared_gen == di->lease_shared_gen ||
1038 snapid <= CEPH_MAXSNAP) {
1039 dout(" clearing %p complete (d_release)\n",
1040 parent_inode);
1041 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1042 ci->i_release_count++;
1044 spin_unlock(&parent_inode->i_lock);
1046 if (di) {
1047 ceph_dentry_lru_del(dentry);
1048 if (di->lease_session)
1049 ceph_put_mds_session(di->lease_session);
1050 kmem_cache_free(ceph_dentry_cachep, di);
1051 dentry->d_fsdata = NULL;
1055 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1056 struct nameidata *nd)
1059 * Eventually, we'll want to revalidate snapped metadata
1060 * too... probably...
1062 return 1;
1068 * read() on a dir. This weird interface hack only works if mounted
1069 * with '-o dirstat'.
1071 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1072 loff_t *ppos)
1074 struct ceph_file_info *cf = file->private_data;
1075 struct inode *inode = file->f_dentry->d_inode;
1076 struct ceph_inode_info *ci = ceph_inode(inode);
1077 int left;
1079 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1080 return -EISDIR;
1082 if (!cf->dir_info) {
1083 cf->dir_info = kmalloc(1024, GFP_NOFS);
1084 if (!cf->dir_info)
1085 return -ENOMEM;
1086 cf->dir_info_len =
1087 sprintf(cf->dir_info,
1088 "entries: %20lld\n"
1089 " files: %20lld\n"
1090 " subdirs: %20lld\n"
1091 "rentries: %20lld\n"
1092 " rfiles: %20lld\n"
1093 " rsubdirs: %20lld\n"
1094 "rbytes: %20lld\n"
1095 "rctime: %10ld.%09ld\n",
1096 ci->i_files + ci->i_subdirs,
1097 ci->i_files,
1098 ci->i_subdirs,
1099 ci->i_rfiles + ci->i_rsubdirs,
1100 ci->i_rfiles,
1101 ci->i_rsubdirs,
1102 ci->i_rbytes,
1103 (long)ci->i_rctime.tv_sec,
1104 (long)ci->i_rctime.tv_nsec);
1107 if (*ppos >= cf->dir_info_len)
1108 return 0;
1109 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1110 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1111 if (left == size)
1112 return -EFAULT;
1113 *ppos += (size - left);
1114 return size - left;
1118 * an fsync() on a dir will wait for any uncommitted directory
1119 * operations to commit.
1121 static int ceph_dir_fsync(struct file *file, int datasync)
1123 struct inode *inode = file->f_path.dentry->d_inode;
1124 struct ceph_inode_info *ci = ceph_inode(inode);
1125 struct list_head *head = &ci->i_unsafe_dirops;
1126 struct ceph_mds_request *req;
1127 u64 last_tid;
1128 int ret = 0;
1130 dout("dir_fsync %p\n", inode);
1131 spin_lock(&ci->i_unsafe_lock);
1132 if (list_empty(head))
1133 goto out;
1135 req = list_entry(head->prev,
1136 struct ceph_mds_request, r_unsafe_dir_item);
1137 last_tid = req->r_tid;
1139 do {
1140 ceph_mdsc_get_request(req);
1141 spin_unlock(&ci->i_unsafe_lock);
1142 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1143 inode, req->r_tid, last_tid);
1144 if (req->r_timeout) {
1145 ret = wait_for_completion_timeout(
1146 &req->r_safe_completion, req->r_timeout);
1147 if (ret > 0)
1148 ret = 0;
1149 else if (ret == 0)
1150 ret = -EIO; /* timed out */
1151 } else {
1152 wait_for_completion(&req->r_safe_completion);
1154 spin_lock(&ci->i_unsafe_lock);
1155 ceph_mdsc_put_request(req);
1157 if (ret || list_empty(head))
1158 break;
1159 req = list_entry(head->next,
1160 struct ceph_mds_request, r_unsafe_dir_item);
1161 } while (req->r_tid < last_tid);
1162 out:
1163 spin_unlock(&ci->i_unsafe_lock);
1164 return ret;
1168 * We maintain a private dentry LRU.
1170 * FIXME: this needs to be changed to a per-mds lru to be useful.
1172 void ceph_dentry_lru_add(struct dentry *dn)
1174 struct ceph_dentry_info *di = ceph_dentry(dn);
1175 struct ceph_mds_client *mdsc;
1177 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1178 dn->d_name.len, dn->d_name.name);
1179 if (di) {
1180 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1181 spin_lock(&mdsc->dentry_lru_lock);
1182 list_add_tail(&di->lru, &mdsc->dentry_lru);
1183 mdsc->num_dentry++;
1184 spin_unlock(&mdsc->dentry_lru_lock);
1188 void ceph_dentry_lru_touch(struct dentry *dn)
1190 struct ceph_dentry_info *di = ceph_dentry(dn);
1191 struct ceph_mds_client *mdsc;
1193 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1194 dn->d_name.len, dn->d_name.name, di->offset);
1195 if (di) {
1196 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1197 spin_lock(&mdsc->dentry_lru_lock);
1198 list_move_tail(&di->lru, &mdsc->dentry_lru);
1199 spin_unlock(&mdsc->dentry_lru_lock);
1203 void ceph_dentry_lru_del(struct dentry *dn)
1205 struct ceph_dentry_info *di = ceph_dentry(dn);
1206 struct ceph_mds_client *mdsc;
1208 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1209 dn->d_name.len, dn->d_name.name);
1210 if (di) {
1211 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1212 spin_lock(&mdsc->dentry_lru_lock);
1213 list_del_init(&di->lru);
1214 mdsc->num_dentry--;
1215 spin_unlock(&mdsc->dentry_lru_lock);
1219 const struct file_operations ceph_dir_fops = {
1220 .read = ceph_read_dir,
1221 .readdir = ceph_readdir,
1222 .llseek = ceph_dir_llseek,
1223 .open = ceph_open,
1224 .release = ceph_release,
1225 .unlocked_ioctl = ceph_ioctl,
1226 .fsync = ceph_dir_fsync,
1229 const struct inode_operations ceph_dir_iops = {
1230 .lookup = ceph_lookup,
1231 .permission = ceph_permission,
1232 .getattr = ceph_getattr,
1233 .setattr = ceph_setattr,
1234 .setxattr = ceph_setxattr,
1235 .getxattr = ceph_getxattr,
1236 .listxattr = ceph_listxattr,
1237 .removexattr = ceph_removexattr,
1238 .mknod = ceph_mknod,
1239 .symlink = ceph_symlink,
1240 .mkdir = ceph_mkdir,
1241 .link = ceph_link,
1242 .unlink = ceph_unlink,
1243 .rmdir = ceph_unlink,
1244 .rename = ceph_rename,
1245 .create = ceph_create,
1248 const struct dentry_operations ceph_dentry_ops = {
1249 .d_revalidate = ceph_d_revalidate,
1250 .d_release = ceph_dentry_release,
1253 const struct dentry_operations ceph_snapdir_dentry_ops = {
1254 .d_revalidate = ceph_snapdir_d_revalidate,
1255 .d_release = ceph_dentry_release,
1258 const struct dentry_operations ceph_snap_dentry_ops = {
1259 .d_release = ceph_dentry_release,