blob: 8a9116e15b701a809d6b6d78636c97d56bc81c97 [file] [log] [blame]
Sage Weil2817b002009-10-06 11:31:08 -07001#include "ceph_debug.h"
2
3#include <linux/spinlock.h>
4#include <linux/fs_struct.h>
5#include <linux/namei.h>
6#include <linux/sched.h>
7
8#include "super.h"
9
10/*
11 * Directory operations: readdir, lookup, create, link, unlink,
12 * rename, etc.
13 */
14
15/*
16 * Ceph MDS operations are specified in terms of a base ino and
17 * relative path. Thus, the client can specify an operation on a
18 * specific inode (e.g., a getattr due to fstat(2)), or as a path
19 * relative to, say, the root directory.
20 *
21 * Normally, we limit ourselves to strict inode ops (no path component)
22 * or dentry operations (a single path component relative to an ino). The
23 * exception to this is open_root_dentry(), which will open the mount
24 * point by name.
25 */
26
27const struct inode_operations ceph_dir_iops;
28const struct file_operations ceph_dir_fops;
29struct dentry_operations ceph_dentry_ops;
30
31/*
32 * Initialize ceph dentry state.
33 */
34int ceph_init_dentry(struct dentry *dentry)
35{
36 struct ceph_dentry_info *di;
37
38 if (dentry->d_fsdata)
39 return 0;
40
41 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
42 dentry->d_op = &ceph_dentry_ops;
43 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
44 dentry->d_op = &ceph_snapdir_dentry_ops;
45 else
46 dentry->d_op = &ceph_snap_dentry_ops;
47
48 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS);
49 if (!di)
50 return -ENOMEM; /* oh well */
51
52 spin_lock(&dentry->d_lock);
53 if (dentry->d_fsdata) /* lost a race */
54 goto out_unlock;
55 di->dentry = dentry;
56 di->lease_session = NULL;
57 dentry->d_fsdata = di;
58 dentry->d_time = jiffies;
59 ceph_dentry_lru_add(dentry);
60out_unlock:
61 spin_unlock(&dentry->d_lock);
62 return 0;
63}
64
65
66
67/*
68 * for readdir, we encode the directory frag and offset within that
69 * frag into f_pos.
70 */
71static unsigned fpos_frag(loff_t p)
72{
73 return p >> 32;
74}
75static unsigned fpos_off(loff_t p)
76{
77 return p & 0xffffffff;
78}
79
80/*
81 * When possible, we try to satisfy a readdir by peeking at the
82 * dcache. We make this work by carefully ordering dentries on
83 * d_u.d_child when we initially get results back from the MDS, and
84 * falling back to a "normal" sync readdir if any dentries in the dir
85 * are dropped.
86 *
87 * I_COMPLETE tells indicates we have all dentries in the dir. It is
88 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
89 * the MDS if/when the directory is modified).
90 */
91static int __dcache_readdir(struct file *filp,
92 void *dirent, filldir_t filldir)
93{
94 struct inode *inode = filp->f_dentry->d_inode;
95 struct ceph_file_info *fi = filp->private_data;
96 struct dentry *parent = filp->f_dentry;
97 struct inode *dir = parent->d_inode;
98 struct list_head *p;
99 struct dentry *dentry, *last;
100 struct ceph_dentry_info *di;
101 int err = 0;
102
103 /* claim ref on last dentry we returned */
104 last = fi->dentry;
105 fi->dentry = NULL;
106
107 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
108 last);
109
110 spin_lock(&dcache_lock);
111
112 /* start at beginning? */
113 if (filp->f_pos == 2 || (last &&
114 filp->f_pos < ceph_dentry(last)->offset)) {
115 if (list_empty(&parent->d_subdirs))
116 goto out_unlock;
117 p = parent->d_subdirs.prev;
118 dout(" initial p %p/%p\n", p->prev, p->next);
119 } else {
120 p = last->d_u.d_child.prev;
121 }
122
123more:
124 dentry = list_entry(p, struct dentry, d_u.d_child);
125 di = ceph_dentry(dentry);
126 while (1) {
127 dout(" p %p/%p d_subdirs %p/%p\n", p->prev, p->next,
128 parent->d_subdirs.prev, parent->d_subdirs.next);
129 if (p == &parent->d_subdirs) {
130 fi->at_end = 1;
131 goto out_unlock;
132 }
133 if (!d_unhashed(dentry) && dentry->d_inode &&
Sage Weil09b8a7d2009-11-11 15:21:27 -0800134 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
Sage Weil1d1de9162009-12-02 11:54:25 -0800135 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
Sage Weil2817b002009-10-06 11:31:08 -0700136 filp->f_pos <= di->offset)
137 break;
138 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
139 dentry->d_name.len, dentry->d_name.name, di->offset,
140 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
141 !dentry->d_inode ? " null" : "");
142 p = p->prev;
143 dentry = list_entry(p, struct dentry, d_u.d_child);
144 di = ceph_dentry(dentry);
145 }
146
147 atomic_inc(&dentry->d_count);
148 spin_unlock(&dcache_lock);
149 spin_unlock(&inode->i_lock);
150
151 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
152 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
153 filp->f_pos = di->offset;
154 err = filldir(dirent, dentry->d_name.name,
155 dentry->d_name.len, di->offset,
156 dentry->d_inode->i_ino,
157 dentry->d_inode->i_mode >> 12);
158
159 if (last) {
160 if (err < 0) {
161 /* remember our position */
162 fi->dentry = last;
163 fi->next_offset = di->offset;
164 } else {
165 dput(last);
166 }
167 last = NULL;
168 }
169
170 spin_lock(&inode->i_lock);
171 spin_lock(&dcache_lock);
172
173 if (err < 0)
174 goto out_unlock;
175
176 last = dentry;
177
178 p = p->prev;
179 filp->f_pos++;
180
181 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
182 if ((ceph_inode(dir)->i_ceph_flags & CEPH_I_COMPLETE))
183 goto more;
184 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
185 err = -EAGAIN;
186
187out_unlock:
188 spin_unlock(&dcache_lock);
189
190 if (last) {
191 spin_unlock(&inode->i_lock);
192 dput(last);
193 spin_lock(&inode->i_lock);
194 }
195
196 return err;
197}
198
199/*
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.
204 */
205static int note_last_dentry(struct ceph_file_info *fi, const char *name,
206 int len)
207{
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;
216}
217
218static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
219{
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_client *client = ceph_inode_to_client(inode);
224 struct ceph_mds_client *mdsc = &client->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;
Sage Weil6b805182009-10-27 11:50:50 -0700230 const int max_entries = client->mount_args->max_readdir;
Sage Weil2817b002009-10-06 11:31:08 -0700231
232 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
233 if (fi->at_end)
234 return 0;
235
236 /* always start with . and .. */
237 if (filp->f_pos == 0) {
238 /* note dir version at start of readdir so we can tell
239 * if any dentries get dropped */
240 fi->dir_release_count = ci->i_release_count;
241
242 dout("readdir off 0 -> '.'\n");
243 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
244 inode->i_ino, inode->i_mode >> 12) < 0)
245 return 0;
246 filp->f_pos = 1;
247 off = 1;
248 }
249 if (filp->f_pos == 1) {
250 dout("readdir off 1 -> '..'\n");
251 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
252 filp->f_dentry->d_parent->d_inode->i_ino,
253 inode->i_mode >> 12) < 0)
254 return 0;
255 filp->f_pos = 2;
256 off = 2;
257 }
258
259 /* can we use the dcache? */
260 spin_lock(&inode->i_lock);
261 if ((filp->f_pos == 2 || fi->dentry) &&
262 !ceph_test_opt(client, NOASYNCREADDIR) &&
263 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
264 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
265 err = __dcache_readdir(filp, dirent, filldir);
266 if (err != -EAGAIN) {
267 spin_unlock(&inode->i_lock);
268 return err;
269 }
270 }
271 spin_unlock(&inode->i_lock);
272 if (fi->dentry) {
273 err = note_last_dentry(fi, fi->dentry->d_name.name,
274 fi->dentry->d_name.len);
275 if (err)
276 return err;
277 dput(fi->dentry);
278 fi->dentry = NULL;
279 }
280
281 /* proceed with a normal readdir */
282
283more:
284 /* do we have the correct frag content buffered? */
285 if (fi->frag != frag || fi->last_readdir == NULL) {
286 struct ceph_mds_request *req;
287 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
288 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
289
290 /* discard old result, if any */
Sage Weil393f6622010-03-10 12:03:32 -0800291 if (fi->last_readdir) {
Sage Weil2817b002009-10-06 11:31:08 -0700292 ceph_mdsc_put_request(fi->last_readdir);
Sage Weil393f6622010-03-10 12:03:32 -0800293 fi->last_readdir = NULL;
294 }
Sage Weil2817b002009-10-06 11:31:08 -0700295
296 /* requery frag tree, as the frag topology may have changed */
297 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
298
299 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
300 ceph_vinop(inode), frag, fi->last_name);
301 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
302 if (IS_ERR(req))
303 return PTR_ERR(req);
304 req->r_inode = igrab(inode);
305 req->r_dentry = dget(filp->f_dentry);
306 /* hints to request -> mds selection code */
307 req->r_direct_mode = USE_AUTH_MDS;
308 req->r_direct_hash = ceph_frag_value(frag);
309 req->r_direct_is_hash = true;
310 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
311 req->r_readdir_offset = fi->next_offset;
312 req->r_args.readdir.frag = cpu_to_le32(frag);
313 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
314 req->r_num_caps = max_entries;
315 err = ceph_mdsc_do_request(mdsc, NULL, req);
316 if (err < 0) {
317 ceph_mdsc_put_request(req);
318 return err;
319 }
320 dout("readdir got and parsed readdir result=%d"
321 " on frag %x, end=%d, complete=%d\n", err, frag,
322 (int)req->r_reply_info.dir_end,
323 (int)req->r_reply_info.dir_complete);
324
325 if (!req->r_did_prepopulate) {
326 dout("readdir !did_prepopulate");
327 fi->dir_release_count--; /* preclude I_COMPLETE */
328 }
329
330 /* note next offset and last dentry name */
331 fi->offset = fi->next_offset;
332 fi->last_readdir = req;
333
334 if (req->r_reply_info.dir_end) {
335 kfree(fi->last_name);
336 fi->last_name = NULL;
337 fi->next_offset = 0;
338 } else {
339 rinfo = &req->r_reply_info;
340 err = note_last_dentry(fi,
341 rinfo->dir_dname[rinfo->dir_nr-1],
342 rinfo->dir_dname_len[rinfo->dir_nr-1]);
343 if (err)
344 return err;
345 fi->next_offset += rinfo->dir_nr;
346 }
347 }
348
349 rinfo = &fi->last_readdir->r_reply_info;
350 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
351 rinfo->dir_nr, off, fi->offset);
352 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
353 u64 pos = ceph_make_fpos(frag, off);
354 struct ceph_mds_reply_inode *in =
355 rinfo->dir_in[off - fi->offset].in;
356 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
357 off, off - fi->offset, rinfo->dir_nr, pos,
358 rinfo->dir_dname_len[off - fi->offset],
359 rinfo->dir_dname[off - fi->offset], in);
360 BUG_ON(!in);
361 ftype = le32_to_cpu(in->mode) >> 12;
362 if (filldir(dirent,
363 rinfo->dir_dname[off - fi->offset],
364 rinfo->dir_dname_len[off - fi->offset],
365 pos,
366 le64_to_cpu(in->ino),
367 ftype) < 0) {
368 dout("filldir stopping us...\n");
369 return 0;
370 }
371 off++;
372 filp->f_pos = pos + 1;
373 }
374
375 if (fi->last_name) {
376 ceph_mdsc_put_request(fi->last_readdir);
377 fi->last_readdir = NULL;
378 goto more;
379 }
380
381 /* more frags? */
382 if (!ceph_frag_is_rightmost(frag)) {
383 frag = ceph_frag_next(frag);
384 off = 0;
385 filp->f_pos = ceph_make_fpos(frag, off);
386 dout("readdir next frag is %x\n", frag);
387 goto more;
388 }
389 fi->at_end = 1;
390
391 /*
392 * if dir_release_count still matches the dir, no dentries
393 * were released during the whole readdir, and we should have
394 * the complete dir contents in our cache.
395 */
396 spin_lock(&inode->i_lock);
397 if (ci->i_release_count == fi->dir_release_count) {
398 dout(" marking %p complete\n", inode);
399 ci->i_ceph_flags |= CEPH_I_COMPLETE;
400 ci->i_max_offset = filp->f_pos;
401 }
402 spin_unlock(&inode->i_lock);
403
404 dout("readdir %p filp %p done.\n", inode, filp);
405 return 0;
406}
407
408static void reset_readdir(struct ceph_file_info *fi)
409{
410 if (fi->last_readdir) {
411 ceph_mdsc_put_request(fi->last_readdir);
412 fi->last_readdir = NULL;
413 }
414 kfree(fi->last_name);
415 fi->next_offset = 2; /* compensate for . and .. */
416 if (fi->dentry) {
417 dput(fi->dentry);
418 fi->dentry = NULL;
419 }
420 fi->at_end = 0;
421}
422
423static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
424{
425 struct ceph_file_info *fi = file->private_data;
426 struct inode *inode = file->f_mapping->host;
427 loff_t old_offset = offset;
428 loff_t retval;
429
430 mutex_lock(&inode->i_mutex);
431 switch (origin) {
432 case SEEK_END:
433 offset += inode->i_size + 2; /* FIXME */
434 break;
435 case SEEK_CUR:
436 offset += file->f_pos;
437 }
438 retval = -EINVAL;
439 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
440 if (offset != file->f_pos) {
441 file->f_pos = offset;
442 file->f_version = 0;
443 fi->at_end = 0;
444 }
445 retval = offset;
446
447 /*
448 * discard buffered readdir content on seekdir(0), or
449 * seek to new frag, or seek prior to current chunk.
450 */
451 if (offset == 0 ||
452 fpos_frag(offset) != fpos_frag(old_offset) ||
453 fpos_off(offset) < fi->offset) {
454 dout("dir_llseek dropping %p content\n", file);
455 reset_readdir(fi);
456 }
457
458 /* bump dir_release_count if we did a forward seek */
459 if (offset > old_offset)
460 fi->dir_release_count--;
461 }
462 mutex_unlock(&inode->i_mutex);
463 return retval;
464}
465
466/*
467 * Process result of a lookup/open request.
468 *
469 * Mainly, make sure we return the final req->r_dentry (if it already
470 * existed) in place of the original VFS-provided dentry when they
471 * differ.
472 *
473 * Gracefully handle the case where the MDS replies with -ENOENT and
474 * no trace (which it may do, at its discretion, e.g., if it doesn't
475 * care to issue a lease on the negative dentry).
476 */
477struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
478 struct dentry *dentry, int err)
479{
480 struct ceph_client *client = ceph_client(dentry->d_sb);
481 struct inode *parent = dentry->d_parent->d_inode;
482
483 /* .snap dir? */
484 if (err == -ENOENT &&
485 ceph_vino(parent).ino != CEPH_INO_ROOT && /* no .snap in root dir */
Sage Weil6b805182009-10-27 11:50:50 -0700486 strcmp(dentry->d_name.name,
487 client->mount_args->snapdir_name) == 0) {
Sage Weil2817b002009-10-06 11:31:08 -0700488 struct inode *inode = ceph_get_snapdir(parent);
489 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
490 dentry, dentry->d_name.len, dentry->d_name.name, inode);
491 d_add(dentry, inode);
492 err = 0;
493 }
494
495 if (err == -ENOENT) {
496 /* no trace? */
497 err = 0;
498 if (!req->r_reply_info.head->is_dentry) {
499 dout("ENOENT and no trace, dentry %p inode %p\n",
500 dentry, dentry->d_inode);
501 if (dentry->d_inode) {
502 d_drop(dentry);
503 err = -ENOENT;
504 } else {
505 d_add(dentry, NULL);
506 }
507 }
508 }
509 if (err)
510 dentry = ERR_PTR(err);
511 else if (dentry != req->r_dentry)
512 dentry = dget(req->r_dentry); /* we got spliced */
513 else
514 dentry = NULL;
515 return dentry;
516}
517
Sage Weil1d1de9162009-12-02 11:54:25 -0800518static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
519{
520 return ceph_ino(inode) == CEPH_INO_ROOT &&
521 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
522}
523
Sage Weil2817b002009-10-06 11:31:08 -0700524/*
525 * Look up a single dir entry. If there is a lookup intent, inform
526 * the MDS so that it gets our 'caps wanted' value in a single op.
527 */
528static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
529 struct nameidata *nd)
530{
531 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
532 struct ceph_mds_client *mdsc = &client->mdsc;
533 struct ceph_mds_request *req;
534 int op;
535 int err;
536
537 dout("lookup %p dentry %p '%.*s'\n",
538 dir, dentry, dentry->d_name.len, dentry->d_name.name);
539
540 if (dentry->d_name.len > NAME_MAX)
541 return ERR_PTR(-ENAMETOOLONG);
542
543 err = ceph_init_dentry(dentry);
544 if (err < 0)
545 return ERR_PTR(err);
546
547 /* open (but not create!) intent? */
548 if (nd &&
549 (nd->flags & LOOKUP_OPEN) &&
550 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
551 !(nd->intent.open.flags & O_CREAT)) {
552 int mode = nd->intent.open.create_mode & ~current->fs->umask;
553 return ceph_lookup_open(dir, dentry, nd, mode, 1);
554 }
555
556 /* can we conclude ENOENT locally? */
557 if (dentry->d_inode == NULL) {
558 struct ceph_inode_info *ci = ceph_inode(dir);
559 struct ceph_dentry_info *di = ceph_dentry(dentry);
560
561 spin_lock(&dir->i_lock);
562 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
563 if (strncmp(dentry->d_name.name,
Sage Weil6b805182009-10-27 11:50:50 -0700564 client->mount_args->snapdir_name,
Sage Weil2817b002009-10-06 11:31:08 -0700565 dentry->d_name.len) &&
Sage Weil1d1de9162009-12-02 11:54:25 -0800566 !is_root_ceph_dentry(dir, dentry) &&
Sage Weil2817b002009-10-06 11:31:08 -0700567 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
568 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
569 di->offset = ci->i_max_offset++;
570 spin_unlock(&dir->i_lock);
571 dout(" dir %p complete, -ENOENT\n", dir);
572 d_add(dentry, NULL);
573 di->lease_shared_gen = ci->i_shared_gen;
574 return NULL;
575 }
576 spin_unlock(&dir->i_lock);
577 }
578
579 op = ceph_snap(dir) == CEPH_SNAPDIR ?
580 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
581 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
582 if (IS_ERR(req))
583 return ERR_PTR(PTR_ERR(req));
584 req->r_dentry = dget(dentry);
585 req->r_num_caps = 2;
586 /* we only need inode linkage */
587 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
588 req->r_locked_dir = dir;
589 err = ceph_mdsc_do_request(mdsc, NULL, req);
590 dentry = ceph_finish_lookup(req, dentry, err);
591 ceph_mdsc_put_request(req); /* will dput(dentry) */
592 dout("lookup result=%p\n", dentry);
593 return dentry;
594}
595
596/*
597 * If we do a create but get no trace back from the MDS, follow up with
598 * a lookup (the VFS expects us to link up the provided dentry).
599 */
600int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
601{
602 struct dentry *result = ceph_lookup(dir, dentry, NULL);
603
604 if (result && !IS_ERR(result)) {
605 /*
606 * We created the item, then did a lookup, and found
607 * it was already linked to another inode we already
608 * had in our cache (and thus got spliced). Link our
609 * dentry to that inode, but don't hash it, just in
610 * case the VFS wants to dereference it.
611 */
612 BUG_ON(!result->d_inode);
613 d_instantiate(dentry, result->d_inode);
614 return 0;
615 }
616 return PTR_ERR(result);
617}
618
619static int ceph_mknod(struct inode *dir, struct dentry *dentry,
620 int mode, dev_t rdev)
621{
622 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
623 struct ceph_mds_client *mdsc = &client->mdsc;
624 struct ceph_mds_request *req;
625 int err;
626
627 if (ceph_snap(dir) != CEPH_NOSNAP)
628 return -EROFS;
629
630 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
631 dir, dentry, mode, rdev);
632 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
633 if (IS_ERR(req)) {
634 d_drop(dentry);
635 return PTR_ERR(req);
636 }
637 req->r_dentry = dget(dentry);
638 req->r_num_caps = 2;
639 req->r_locked_dir = dir;
640 req->r_args.mknod.mode = cpu_to_le32(mode);
641 req->r_args.mknod.rdev = cpu_to_le32(rdev);
642 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
643 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
644 err = ceph_mdsc_do_request(mdsc, dir, req);
645 if (!err && !req->r_reply_info.head->is_dentry)
646 err = ceph_handle_notrace_create(dir, dentry);
647 ceph_mdsc_put_request(req);
648 if (err)
649 d_drop(dentry);
650 return err;
651}
652
653static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
654 struct nameidata *nd)
655{
656 dout("create in dir %p dentry %p name '%.*s'\n",
657 dir, dentry, dentry->d_name.len, dentry->d_name.name);
658
659 if (ceph_snap(dir) != CEPH_NOSNAP)
660 return -EROFS;
661
662 if (nd) {
663 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
664 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
665 /* hrm, what should i do here if we get aliased? */
666 if (IS_ERR(dentry))
667 return PTR_ERR(dentry);
668 return 0;
669 }
670
671 /* fall back to mknod */
672 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
673}
674
675static int ceph_symlink(struct inode *dir, struct dentry *dentry,
676 const char *dest)
677{
678 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
679 struct ceph_mds_client *mdsc = &client->mdsc;
680 struct ceph_mds_request *req;
681 int err;
682
683 if (ceph_snap(dir) != CEPH_NOSNAP)
684 return -EROFS;
685
686 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
687 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
688 if (IS_ERR(req)) {
689 d_drop(dentry);
690 return PTR_ERR(req);
691 }
692 req->r_dentry = dget(dentry);
693 req->r_num_caps = 2;
694 req->r_path2 = kstrdup(dest, GFP_NOFS);
695 req->r_locked_dir = dir;
696 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
697 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
698 err = ceph_mdsc_do_request(mdsc, dir, req);
699 if (!err && !req->r_reply_info.head->is_dentry)
700 err = ceph_handle_notrace_create(dir, dentry);
701 ceph_mdsc_put_request(req);
702 if (err)
703 d_drop(dentry);
704 return err;
705}
706
707static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
708{
709 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
710 struct ceph_mds_client *mdsc = &client->mdsc;
711 struct ceph_mds_request *req;
712 int err = -EROFS;
713 int op;
714
715 if (ceph_snap(dir) == CEPH_SNAPDIR) {
716 /* mkdir .snap/foo is a MKSNAP */
717 op = CEPH_MDS_OP_MKSNAP;
718 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
719 dentry->d_name.len, dentry->d_name.name, dentry);
720 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
721 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
722 op = CEPH_MDS_OP_MKDIR;
723 } else {
724 goto out;
725 }
726 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
727 if (IS_ERR(req)) {
728 err = PTR_ERR(req);
729 goto out;
730 }
731
732 req->r_dentry = dget(dentry);
733 req->r_num_caps = 2;
734 req->r_locked_dir = dir;
735 req->r_args.mkdir.mode = cpu_to_le32(mode);
736 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
737 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
738 err = ceph_mdsc_do_request(mdsc, dir, req);
739 if (!err && !req->r_reply_info.head->is_dentry)
740 err = ceph_handle_notrace_create(dir, dentry);
741 ceph_mdsc_put_request(req);
742out:
743 if (err < 0)
744 d_drop(dentry);
745 return err;
746}
747
748static int ceph_link(struct dentry *old_dentry, struct inode *dir,
749 struct dentry *dentry)
750{
751 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
752 struct ceph_mds_client *mdsc = &client->mdsc;
753 struct ceph_mds_request *req;
754 int err;
755
756 if (ceph_snap(dir) != CEPH_NOSNAP)
757 return -EROFS;
758
759 dout("link in dir %p old_dentry %p dentry %p\n", dir,
760 old_dentry, dentry);
761 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
762 if (IS_ERR(req)) {
763 d_drop(dentry);
764 return PTR_ERR(req);
765 }
766 req->r_dentry = dget(dentry);
767 req->r_num_caps = 2;
768 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
769 req->r_locked_dir = dir;
770 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
771 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
772 err = ceph_mdsc_do_request(mdsc, dir, req);
773 if (err)
774 d_drop(dentry);
775 else if (!req->r_reply_info.head->is_dentry)
776 d_instantiate(dentry, igrab(old_dentry->d_inode));
777 ceph_mdsc_put_request(req);
778 return err;
779}
780
781/*
782 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
783 * looks like the link count will hit 0, drop any other caps (other
784 * than PIN) we don't specifically want (due to the file still being
785 * open).
786 */
787static int drop_caps_for_unlink(struct inode *inode)
788{
789 struct ceph_inode_info *ci = ceph_inode(inode);
790 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
791
792 spin_lock(&inode->i_lock);
793 if (inode->i_nlink == 1) {
794 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
795 ci->i_ceph_flags |= CEPH_I_NODELAY;
796 }
797 spin_unlock(&inode->i_lock);
798 return drop;
799}
800
801/*
802 * rmdir and unlink are differ only by the metadata op code
803 */
804static int ceph_unlink(struct inode *dir, struct dentry *dentry)
805{
806 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
807 struct ceph_mds_client *mdsc = &client->mdsc;
808 struct inode *inode = dentry->d_inode;
809 struct ceph_mds_request *req;
810 int err = -EROFS;
811 int op;
812
813 if (ceph_snap(dir) == CEPH_SNAPDIR) {
814 /* rmdir .snap/foo is RMSNAP */
815 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
816 dentry->d_name.name, dentry);
817 op = CEPH_MDS_OP_RMSNAP;
818 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
819 dout("unlink/rmdir dir %p dn %p inode %p\n",
820 dir, dentry, inode);
821 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
822 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
823 } else
824 goto out;
825 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
826 if (IS_ERR(req)) {
827 err = PTR_ERR(req);
828 goto out;
829 }
830 req->r_dentry = dget(dentry);
831 req->r_num_caps = 2;
832 req->r_locked_dir = dir;
833 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
834 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
835 req->r_inode_drop = drop_caps_for_unlink(inode);
836 err = ceph_mdsc_do_request(mdsc, dir, req);
837 if (!err && !req->r_reply_info.head->is_dentry)
838 d_delete(dentry);
839 ceph_mdsc_put_request(req);
840out:
841 return err;
842}
843
844static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
845 struct inode *new_dir, struct dentry *new_dentry)
846{
847 struct ceph_client *client = ceph_sb_to_client(old_dir->i_sb);
848 struct ceph_mds_client *mdsc = &client->mdsc;
849 struct ceph_mds_request *req;
850 int err;
851
852 if (ceph_snap(old_dir) != ceph_snap(new_dir))
853 return -EXDEV;
854 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
855 ceph_snap(new_dir) != CEPH_NOSNAP)
856 return -EROFS;
857 dout("rename dir %p dentry %p to dir %p dentry %p\n",
858 old_dir, old_dentry, new_dir, new_dentry);
859 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
860 if (IS_ERR(req))
861 return PTR_ERR(req);
862 req->r_dentry = dget(new_dentry);
863 req->r_num_caps = 2;
864 req->r_old_dentry = dget(old_dentry);
865 req->r_locked_dir = new_dir;
866 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
867 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
868 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
869 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
870 /* release LINK_RDCACHE on source inode (mds will lock it) */
871 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
872 if (new_dentry->d_inode)
873 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
874 err = ceph_mdsc_do_request(mdsc, old_dir, req);
875 if (!err && !req->r_reply_info.head->is_dentry) {
876 /*
877 * Normally d_move() is done by fill_trace (called by
878 * do_request, above). If there is no trace, we need
879 * to do it here.
880 */
881 d_move(old_dentry, new_dentry);
882 }
883 ceph_mdsc_put_request(req);
884 return err;
885}
886
887
888/*
889 * Check if dentry lease is valid. If not, delete the lease. Try to
890 * renew if the least is more than half up.
891 */
892static int dentry_lease_is_valid(struct dentry *dentry)
893{
894 struct ceph_dentry_info *di;
895 struct ceph_mds_session *s;
896 int valid = 0;
897 u32 gen;
898 unsigned long ttl;
899 struct ceph_mds_session *session = NULL;
900 struct inode *dir = NULL;
901 u32 seq = 0;
902
903 spin_lock(&dentry->d_lock);
904 di = ceph_dentry(dentry);
905 if (di && di->lease_session) {
906 s = di->lease_session;
907 spin_lock(&s->s_cap_lock);
908 gen = s->s_cap_gen;
909 ttl = s->s_cap_ttl;
910 spin_unlock(&s->s_cap_lock);
911
912 if (di->lease_gen == gen &&
913 time_before(jiffies, dentry->d_time) &&
914 time_before(jiffies, ttl)) {
915 valid = 1;
916 if (di->lease_renew_after &&
917 time_after(jiffies, di->lease_renew_after)) {
918 /* we should renew */
919 dir = dentry->d_parent->d_inode;
920 session = ceph_get_mds_session(s);
921 seq = di->lease_seq;
922 di->lease_renew_after = 0;
923 di->lease_renew_from = jiffies;
924 }
Sage Weil2817b002009-10-06 11:31:08 -0700925 }
926 }
927 spin_unlock(&dentry->d_lock);
928
929 if (session) {
930 ceph_mdsc_lease_send_msg(session, dir, dentry,
931 CEPH_MDS_LEASE_RENEW, seq);
932 ceph_put_mds_session(session);
933 }
934 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
935 return valid;
936}
937
938/*
939 * Check if directory-wide content lease/cap is valid.
940 */
941static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
942{
943 struct ceph_inode_info *ci = ceph_inode(dir);
944 struct ceph_dentry_info *di = ceph_dentry(dentry);
945 int valid = 0;
946
947 spin_lock(&dir->i_lock);
948 if (ci->i_shared_gen == di->lease_shared_gen)
949 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
950 spin_unlock(&dir->i_lock);
951 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
952 dir, (unsigned)ci->i_shared_gen, dentry,
953 (unsigned)di->lease_shared_gen, valid);
954 return valid;
955}
956
957/*
958 * Check if cached dentry can be trusted.
959 */
960static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
961{
962 struct inode *dir = dentry->d_parent->d_inode;
963
964 dout("d_revalidate %p '%.*s' inode %p\n", dentry,
965 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
966
967 /* always trust cached snapped dentries, snapdir dentry */
968 if (ceph_snap(dir) != CEPH_NOSNAP) {
969 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
970 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
971 goto out_touch;
972 }
973 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
974 goto out_touch;
975
976 if (dentry_lease_is_valid(dentry) ||
977 dir_lease_is_valid(dir, dentry))
978 goto out_touch;
979
980 dout("d_revalidate %p invalid\n", dentry);
981 d_drop(dentry);
982 return 0;
983out_touch:
984 ceph_dentry_lru_touch(dentry);
985 return 1;
986}
987
988/*
989 * When a dentry is released, clear the dir I_COMPLETE if it was part
990 * of the current dir gen.
991 */
992static void ceph_dentry_release(struct dentry *dentry)
993{
994 struct ceph_dentry_info *di = ceph_dentry(dentry);
995 struct inode *parent_inode = dentry->d_parent->d_inode;
996
997 if (parent_inode) {
998 struct ceph_inode_info *ci = ceph_inode(parent_inode);
999
1000 spin_lock(&parent_inode->i_lock);
1001 if (ci->i_shared_gen == di->lease_shared_gen) {
1002 dout(" clearing %p complete (d_release)\n",
1003 parent_inode);
1004 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1005 ci->i_release_count++;
1006 }
1007 spin_unlock(&parent_inode->i_lock);
1008 }
1009 if (di) {
1010 ceph_dentry_lru_del(dentry);
1011 if (di->lease_session)
1012 ceph_put_mds_session(di->lease_session);
1013 kmem_cache_free(ceph_dentry_cachep, di);
1014 dentry->d_fsdata = NULL;
1015 }
1016}
1017
1018static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1019 struct nameidata *nd)
1020{
1021 /*
1022 * Eventually, we'll want to revalidate snapped metadata
1023 * too... probably...
1024 */
1025 return 1;
1026}
1027
1028
1029
1030/*
1031 * read() on a dir. This weird interface hack only works if mounted
1032 * with '-o dirstat'.
1033 */
1034static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1035 loff_t *ppos)
1036{
1037 struct ceph_file_info *cf = file->private_data;
1038 struct inode *inode = file->f_dentry->d_inode;
1039 struct ceph_inode_info *ci = ceph_inode(inode);
1040 int left;
1041
1042 if (!ceph_test_opt(ceph_client(inode->i_sb), DIRSTAT))
1043 return -EISDIR;
1044
1045 if (!cf->dir_info) {
1046 cf->dir_info = kmalloc(1024, GFP_NOFS);
1047 if (!cf->dir_info)
1048 return -ENOMEM;
1049 cf->dir_info_len =
1050 sprintf(cf->dir_info,
1051 "entries: %20lld\n"
1052 " files: %20lld\n"
1053 " subdirs: %20lld\n"
1054 "rentries: %20lld\n"
1055 " rfiles: %20lld\n"
1056 " rsubdirs: %20lld\n"
1057 "rbytes: %20lld\n"
1058 "rctime: %10ld.%09ld\n",
1059 ci->i_files + ci->i_subdirs,
1060 ci->i_files,
1061 ci->i_subdirs,
1062 ci->i_rfiles + ci->i_rsubdirs,
1063 ci->i_rfiles,
1064 ci->i_rsubdirs,
1065 ci->i_rbytes,
1066 (long)ci->i_rctime.tv_sec,
1067 (long)ci->i_rctime.tv_nsec);
1068 }
1069
1070 if (*ppos >= cf->dir_info_len)
1071 return 0;
1072 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1073 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1074 if (left == size)
1075 return -EFAULT;
1076 *ppos += (size - left);
1077 return size - left;
1078}
1079
1080/*
1081 * an fsync() on a dir will wait for any uncommitted directory
1082 * operations to commit.
1083 */
1084static int ceph_dir_fsync(struct file *file, struct dentry *dentry,
1085 int datasync)
1086{
1087 struct inode *inode = dentry->d_inode;
1088 struct ceph_inode_info *ci = ceph_inode(inode);
1089 struct list_head *head = &ci->i_unsafe_dirops;
1090 struct ceph_mds_request *req;
1091 u64 last_tid;
1092 int ret = 0;
1093
1094 dout("dir_fsync %p\n", inode);
1095 spin_lock(&ci->i_unsafe_lock);
1096 if (list_empty(head))
1097 goto out;
1098
1099 req = list_entry(head->prev,
1100 struct ceph_mds_request, r_unsafe_dir_item);
1101 last_tid = req->r_tid;
1102
1103 do {
1104 ceph_mdsc_get_request(req);
1105 spin_unlock(&ci->i_unsafe_lock);
1106 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1107 inode, req->r_tid, last_tid);
1108 if (req->r_timeout) {
1109 ret = wait_for_completion_timeout(
1110 &req->r_safe_completion, req->r_timeout);
1111 if (ret > 0)
1112 ret = 0;
1113 else if (ret == 0)
1114 ret = -EIO; /* timed out */
1115 } else {
1116 wait_for_completion(&req->r_safe_completion);
1117 }
1118 spin_lock(&ci->i_unsafe_lock);
1119 ceph_mdsc_put_request(req);
1120
1121 if (ret || list_empty(head))
1122 break;
1123 req = list_entry(head->next,
1124 struct ceph_mds_request, r_unsafe_dir_item);
1125 } while (req->r_tid < last_tid);
1126out:
1127 spin_unlock(&ci->i_unsafe_lock);
1128 return ret;
1129}
1130
1131/*
1132 * We maintain a private dentry LRU.
1133 *
1134 * FIXME: this needs to be changed to a per-mds lru to be useful.
1135 */
1136void ceph_dentry_lru_add(struct dentry *dn)
1137{
1138 struct ceph_dentry_info *di = ceph_dentry(dn);
1139 struct ceph_mds_client *mdsc;
Sage Weil2817b002009-10-06 11:31:08 -07001140
Sage Weil04a419f2009-12-23 09:30:21 -08001141 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1142 dn->d_name.len, dn->d_name.name);
Sage Weil2817b002009-10-06 11:31:08 -07001143 if (di) {
1144 mdsc = &ceph_client(dn->d_sb)->mdsc;
1145 spin_lock(&mdsc->dentry_lru_lock);
1146 list_add_tail(&di->lru, &mdsc->dentry_lru);
1147 mdsc->num_dentry++;
1148 spin_unlock(&mdsc->dentry_lru_lock);
1149 }
1150}
1151
1152void ceph_dentry_lru_touch(struct dentry *dn)
1153{
1154 struct ceph_dentry_info *di = ceph_dentry(dn);
1155 struct ceph_mds_client *mdsc;
Sage Weil2817b002009-10-06 11:31:08 -07001156
Sage Weil04a419f2009-12-23 09:30:21 -08001157 dout("dentry_lru_touch %p %p '%.*s'\n", di, dn,
1158 dn->d_name.len, dn->d_name.name);
Sage Weil2817b002009-10-06 11:31:08 -07001159 if (di) {
1160 mdsc = &ceph_client(dn->d_sb)->mdsc;
1161 spin_lock(&mdsc->dentry_lru_lock);
1162 list_move_tail(&di->lru, &mdsc->dentry_lru);
1163 spin_unlock(&mdsc->dentry_lru_lock);
1164 }
1165}
1166
1167void ceph_dentry_lru_del(struct dentry *dn)
1168{
1169 struct ceph_dentry_info *di = ceph_dentry(dn);
1170 struct ceph_mds_client *mdsc;
1171
Sage Weil04a419f2009-12-23 09:30:21 -08001172 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1173 dn->d_name.len, dn->d_name.name);
Sage Weil2817b002009-10-06 11:31:08 -07001174 if (di) {
1175 mdsc = &ceph_client(dn->d_sb)->mdsc;
1176 spin_lock(&mdsc->dentry_lru_lock);
1177 list_del_init(&di->lru);
1178 mdsc->num_dentry--;
1179 spin_unlock(&mdsc->dentry_lru_lock);
1180 }
1181}
1182
1183const struct file_operations ceph_dir_fops = {
1184 .read = ceph_read_dir,
1185 .readdir = ceph_readdir,
1186 .llseek = ceph_dir_llseek,
1187 .open = ceph_open,
1188 .release = ceph_release,
1189 .unlocked_ioctl = ceph_ioctl,
1190 .fsync = ceph_dir_fsync,
1191};
1192
1193const struct inode_operations ceph_dir_iops = {
1194 .lookup = ceph_lookup,
1195 .permission = ceph_permission,
1196 .getattr = ceph_getattr,
1197 .setattr = ceph_setattr,
1198 .setxattr = ceph_setxattr,
1199 .getxattr = ceph_getxattr,
1200 .listxattr = ceph_listxattr,
1201 .removexattr = ceph_removexattr,
1202 .mknod = ceph_mknod,
1203 .symlink = ceph_symlink,
1204 .mkdir = ceph_mkdir,
1205 .link = ceph_link,
1206 .unlink = ceph_unlink,
1207 .rmdir = ceph_unlink,
1208 .rename = ceph_rename,
1209 .create = ceph_create,
1210};
1211
1212struct dentry_operations ceph_dentry_ops = {
1213 .d_revalidate = ceph_d_revalidate,
1214 .d_release = ceph_dentry_release,
1215};
1216
1217struct dentry_operations ceph_snapdir_dentry_ops = {
1218 .d_revalidate = ceph_snapdir_d_revalidate,
1219};
1220
1221struct dentry_operations ceph_snap_dentry_ops = {
1222};