]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
2817b000 SW |
2 | |
3 | #include <linux/spinlock.h> | |
4 | #include <linux/fs_struct.h> | |
5 | #include <linux/namei.h> | |
5a0e3ad6 | 6 | #include <linux/slab.h> |
2817b000 SW |
7 | #include <linux/sched.h> |
8 | ||
9 | #include "super.h" | |
3d14c5d2 | 10 | #include "mds_client.h" |
2817b000 SW |
11 | |
12 | /* | |
13 | * Directory operations: readdir, lookup, create, link, unlink, | |
14 | * rename, etc. | |
15 | */ | |
16 | ||
17 | /* | |
18 | * Ceph MDS operations are specified in terms of a base ino and | |
19 | * relative path. Thus, the client can specify an operation on a | |
20 | * specific inode (e.g., a getattr due to fstat(2)), or as a path | |
21 | * relative to, say, the root directory. | |
22 | * | |
23 | * Normally, we limit ourselves to strict inode ops (no path component) | |
24 | * or dentry operations (a single path component relative to an ino). The | |
25 | * exception to this is open_root_dentry(), which will open the mount | |
26 | * point by name. | |
27 | */ | |
28 | ||
29 | const struct inode_operations ceph_dir_iops; | |
30 | const struct file_operations ceph_dir_fops; | |
52dfb8ac | 31 | const struct dentry_operations ceph_dentry_ops; |
2817b000 SW |
32 | |
33 | /* | |
34 | * Initialize ceph dentry state. | |
35 | */ | |
36 | int ceph_init_dentry(struct dentry *dentry) | |
37 | { | |
38 | struct ceph_dentry_info *di; | |
39 | ||
40 | if (dentry->d_fsdata) | |
41 | return 0; | |
42 | ||
36e21687 | 43 | di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO); |
2817b000 SW |
44 | if (!di) |
45 | return -ENOMEM; /* oh well */ | |
46 | ||
47 | spin_lock(&dentry->d_lock); | |
8c6efb58 SW |
48 | if (dentry->d_fsdata) { |
49 | /* lost a race */ | |
50 | kmem_cache_free(ceph_dentry_cachep, di); | |
2817b000 | 51 | goto out_unlock; |
8c6efb58 | 52 | } |
48d0cbd1 SW |
53 | |
54 | if (dentry->d_parent == NULL || /* nfs fh_to_dentry */ | |
55 | ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) | |
56 | d_set_d_op(dentry, &ceph_dentry_ops); | |
57 | else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR) | |
58 | d_set_d_op(dentry, &ceph_snapdir_dentry_ops); | |
59 | else | |
60 | d_set_d_op(dentry, &ceph_snap_dentry_ops); | |
61 | ||
2817b000 SW |
62 | di->dentry = dentry; |
63 | di->lease_session = NULL; | |
2817b000 | 64 | dentry->d_time = jiffies; |
48d0cbd1 SW |
65 | /* avoid reordering d_fsdata setup so that the check above is safe */ |
66 | smp_mb(); | |
67 | dentry->d_fsdata = di; | |
2817b000 SW |
68 | ceph_dentry_lru_add(dentry); |
69 | out_unlock: | |
70 | spin_unlock(&dentry->d_lock); | |
71 | return 0; | |
72 | } | |
73 | ||
5f21c96d SW |
74 | struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry) |
75 | { | |
76 | struct inode *inode = NULL; | |
77 | ||
78 | if (!dentry) | |
79 | return NULL; | |
80 | ||
81 | spin_lock(&dentry->d_lock); | |
82 | if (dentry->d_parent) { | |
83 | inode = dentry->d_parent->d_inode; | |
84 | ihold(inode); | |
85 | } | |
86 | spin_unlock(&dentry->d_lock); | |
87 | return inode; | |
88 | } | |
2817b000 SW |
89 | |
90 | ||
91 | /* | |
92 | * for readdir, we encode the directory frag and offset within that | |
93 | * frag into f_pos. | |
94 | */ | |
95 | static unsigned fpos_frag(loff_t p) | |
96 | { | |
97 | return p >> 32; | |
98 | } | |
99 | static unsigned fpos_off(loff_t p) | |
100 | { | |
101 | return p & 0xffffffff; | |
102 | } | |
103 | ||
104 | /* | |
105 | * When possible, we try to satisfy a readdir by peeking at the | |
106 | * dcache. We make this work by carefully ordering dentries on | |
107 | * d_u.d_child when we initially get results back from the MDS, and | |
108 | * falling back to a "normal" sync readdir if any dentries in the dir | |
109 | * are dropped. | |
110 | * | |
111 | * I_COMPLETE tells indicates we have all dentries in the dir. It is | |
112 | * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by | |
113 | * the MDS if/when the directory is modified). | |
114 | */ | |
115 | static int __dcache_readdir(struct file *filp, | |
116 | void *dirent, filldir_t filldir) | |
117 | { | |
2817b000 SW |
118 | struct ceph_file_info *fi = filp->private_data; |
119 | struct dentry *parent = filp->f_dentry; | |
120 | struct inode *dir = parent->d_inode; | |
121 | struct list_head *p; | |
122 | struct dentry *dentry, *last; | |
123 | struct ceph_dentry_info *di; | |
124 | int err = 0; | |
125 | ||
126 | /* claim ref on last dentry we returned */ | |
127 | last = fi->dentry; | |
128 | fi->dentry = NULL; | |
129 | ||
130 | dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos, | |
131 | last); | |
132 | ||
2fd6b7f5 | 133 | spin_lock(&parent->d_lock); |
2817b000 SW |
134 | |
135 | /* start at beginning? */ | |
884ea892 SW |
136 | if (filp->f_pos == 2 || last == NULL || |
137 | filp->f_pos < ceph_dentry(last)->offset) { | |
2817b000 SW |
138 | if (list_empty(&parent->d_subdirs)) |
139 | goto out_unlock; | |
140 | p = parent->d_subdirs.prev; | |
141 | dout(" initial p %p/%p\n", p->prev, p->next); | |
142 | } else { | |
143 | p = last->d_u.d_child.prev; | |
144 | } | |
145 | ||
146 | more: | |
147 | dentry = list_entry(p, struct dentry, d_u.d_child); | |
148 | di = ceph_dentry(dentry); | |
149 | while (1) { | |
1cd3935b SW |
150 | dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next, |
151 | d_unhashed(dentry) ? "!hashed" : "hashed", | |
2817b000 SW |
152 | parent->d_subdirs.prev, parent->d_subdirs.next); |
153 | if (p == &parent->d_subdirs) { | |
9cfa1098 | 154 | fi->flags |= CEPH_F_ATEND; |
2817b000 SW |
155 | goto out_unlock; |
156 | } | |
2fd6b7f5 | 157 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2817b000 | 158 | if (!d_unhashed(dentry) && dentry->d_inode && |
09b8a7d2 | 159 | ceph_snap(dentry->d_inode) != CEPH_SNAPDIR && |
1d1de916 | 160 | ceph_ino(dentry->d_inode) != CEPH_INO_CEPH && |
2817b000 SW |
161 | filp->f_pos <= di->offset) |
162 | break; | |
163 | dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry, | |
164 | dentry->d_name.len, dentry->d_name.name, di->offset, | |
165 | filp->f_pos, d_unhashed(dentry) ? " unhashed" : "", | |
166 | !dentry->d_inode ? " null" : ""); | |
da502956 | 167 | spin_unlock(&dentry->d_lock); |
2817b000 SW |
168 | p = p->prev; |
169 | dentry = list_entry(p, struct dentry, d_u.d_child); | |
170 | di = ceph_dentry(dentry); | |
171 | } | |
172 | ||
da502956 | 173 | dget_dlock(dentry); |
b7ab39f6 | 174 | spin_unlock(&dentry->d_lock); |
2fd6b7f5 | 175 | spin_unlock(&parent->d_lock); |
2817b000 SW |
176 | |
177 | dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos, | |
178 | dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode); | |
179 | filp->f_pos = di->offset; | |
180 | err = filldir(dirent, dentry->d_name.name, | |
181 | dentry->d_name.len, di->offset, | |
ad1fee96 | 182 | ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino), |
2817b000 SW |
183 | dentry->d_inode->i_mode >> 12); |
184 | ||
185 | if (last) { | |
186 | if (err < 0) { | |
187 | /* remember our position */ | |
188 | fi->dentry = last; | |
189 | fi->next_offset = di->offset; | |
190 | } else { | |
191 | dput(last); | |
192 | } | |
2817b000 | 193 | } |
f5b06628 SW |
194 | last = dentry; |
195 | ||
2817b000 | 196 | if (err < 0) |
efa4c120 | 197 | goto out; |
2817b000 | 198 | |
2817b000 SW |
199 | filp->f_pos++; |
200 | ||
b5c84bf6 | 201 | /* make sure a dentry wasn't dropped while we didn't have parent lock */ |
efa4c120 SW |
202 | if (!ceph_i_test(dir, CEPH_I_COMPLETE)) { |
203 | dout(" lost I_COMPLETE on %p; falling back to mds\n", dir); | |
204 | err = -EAGAIN; | |
205 | goto out; | |
206 | } | |
207 | ||
2fd6b7f5 | 208 | spin_lock(&parent->d_lock); |
efa4c120 SW |
209 | p = p->prev; /* advance to next dentry */ |
210 | goto more; | |
2817b000 SW |
211 | |
212 | out_unlock: | |
2fd6b7f5 | 213 | spin_unlock(&parent->d_lock); |
efa4c120 SW |
214 | out: |
215 | if (last) | |
2817b000 | 216 | dput(last); |
2817b000 SW |
217 | return err; |
218 | } | |
219 | ||
220 | /* | |
221 | * make note of the last dentry we read, so we can | |
222 | * continue at the same lexicographical point, | |
223 | * regardless of what dir changes take place on the | |
224 | * server. | |
225 | */ | |
226 | static int note_last_dentry(struct ceph_file_info *fi, const char *name, | |
227 | int len) | |
228 | { | |
229 | kfree(fi->last_name); | |
230 | fi->last_name = kmalloc(len+1, GFP_NOFS); | |
231 | if (!fi->last_name) | |
232 | return -ENOMEM; | |
233 | memcpy(fi->last_name, name, len); | |
234 | fi->last_name[len] = 0; | |
235 | dout("note_last_dentry '%s'\n", fi->last_name); | |
236 | return 0; | |
237 | } | |
238 | ||
239 | static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
240 | { | |
241 | struct ceph_file_info *fi = filp->private_data; | |
242 | struct inode *inode = filp->f_dentry->d_inode; | |
243 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
244 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
245 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
246 | unsigned frag = fpos_frag(filp->f_pos); |
247 | int off = fpos_off(filp->f_pos); | |
248 | int err; | |
249 | u32 ftype; | |
250 | struct ceph_mds_reply_info_parsed *rinfo; | |
3d14c5d2 YS |
251 | const int max_entries = fsc->mount_options->max_readdir; |
252 | const int max_bytes = fsc->mount_options->max_readdir_bytes; | |
2817b000 SW |
253 | |
254 | dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off); | |
9cfa1098 | 255 | if (fi->flags & CEPH_F_ATEND) |
2817b000 SW |
256 | return 0; |
257 | ||
258 | /* always start with . and .. */ | |
259 | if (filp->f_pos == 0) { | |
260 | /* note dir version at start of readdir so we can tell | |
261 | * if any dentries get dropped */ | |
262 | fi->dir_release_count = ci->i_release_count; | |
263 | ||
264 | dout("readdir off 0 -> '.'\n"); | |
265 | if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0), | |
ad1fee96 YS |
266 | ceph_translate_ino(inode->i_sb, inode->i_ino), |
267 | inode->i_mode >> 12) < 0) | |
2817b000 SW |
268 | return 0; |
269 | filp->f_pos = 1; | |
270 | off = 1; | |
271 | } | |
272 | if (filp->f_pos == 1) { | |
ad1fee96 | 273 | ino_t ino = filp->f_dentry->d_parent->d_inode->i_ino; |
2817b000 SW |
274 | dout("readdir off 1 -> '..'\n"); |
275 | if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1), | |
ad1fee96 | 276 | ceph_translate_ino(inode->i_sb, ino), |
2817b000 SW |
277 | inode->i_mode >> 12) < 0) |
278 | return 0; | |
279 | filp->f_pos = 2; | |
280 | off = 2; | |
281 | } | |
282 | ||
283 | /* can we use the dcache? */ | |
284 | spin_lock(&inode->i_lock); | |
285 | if ((filp->f_pos == 2 || fi->dentry) && | |
3d14c5d2 | 286 | !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && |
a0dff78d | 287 | ceph_snap(inode) != CEPH_SNAPDIR && |
2817b000 SW |
288 | (ci->i_ceph_flags & CEPH_I_COMPLETE) && |
289 | __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) { | |
efa4c120 | 290 | spin_unlock(&inode->i_lock); |
2817b000 | 291 | err = __dcache_readdir(filp, dirent, filldir); |
efa4c120 | 292 | if (err != -EAGAIN) |
2817b000 | 293 | return err; |
efa4c120 SW |
294 | } else { |
295 | spin_unlock(&inode->i_lock); | |
2817b000 | 296 | } |
2817b000 SW |
297 | if (fi->dentry) { |
298 | err = note_last_dentry(fi, fi->dentry->d_name.name, | |
299 | fi->dentry->d_name.len); | |
300 | if (err) | |
301 | return err; | |
302 | dput(fi->dentry); | |
303 | fi->dentry = NULL; | |
304 | } | |
305 | ||
306 | /* proceed with a normal readdir */ | |
307 | ||
308 | more: | |
309 | /* do we have the correct frag content buffered? */ | |
310 | if (fi->frag != frag || fi->last_readdir == NULL) { | |
311 | struct ceph_mds_request *req; | |
312 | int op = ceph_snap(inode) == CEPH_SNAPDIR ? | |
313 | CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR; | |
314 | ||
315 | /* discard old result, if any */ | |
393f6620 | 316 | if (fi->last_readdir) { |
2817b000 | 317 | ceph_mdsc_put_request(fi->last_readdir); |
393f6620 SW |
318 | fi->last_readdir = NULL; |
319 | } | |
2817b000 SW |
320 | |
321 | /* requery frag tree, as the frag topology may have changed */ | |
322 | frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL); | |
323 | ||
324 | dout("readdir fetching %llx.%llx frag %x offset '%s'\n", | |
325 | ceph_vinop(inode), frag, fi->last_name); | |
326 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
327 | if (IS_ERR(req)) | |
328 | return PTR_ERR(req); | |
70b666c3 SW |
329 | req->r_inode = inode; |
330 | ihold(inode); | |
2817b000 SW |
331 | req->r_dentry = dget(filp->f_dentry); |
332 | /* hints to request -> mds selection code */ | |
333 | req->r_direct_mode = USE_AUTH_MDS; | |
334 | req->r_direct_hash = ceph_frag_value(frag); | |
335 | req->r_direct_is_hash = true; | |
336 | req->r_path2 = kstrdup(fi->last_name, GFP_NOFS); | |
337 | req->r_readdir_offset = fi->next_offset; | |
338 | req->r_args.readdir.frag = cpu_to_le32(frag); | |
339 | req->r_args.readdir.max_entries = cpu_to_le32(max_entries); | |
23804d91 | 340 | req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes); |
e1e4dd0c | 341 | req->r_num_caps = max_entries + 1; |
2817b000 SW |
342 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
343 | if (err < 0) { | |
344 | ceph_mdsc_put_request(req); | |
345 | return err; | |
346 | } | |
347 | dout("readdir got and parsed readdir result=%d" | |
348 | " on frag %x, end=%d, complete=%d\n", err, frag, | |
349 | (int)req->r_reply_info.dir_end, | |
350 | (int)req->r_reply_info.dir_complete); | |
351 | ||
352 | if (!req->r_did_prepopulate) { | |
353 | dout("readdir !did_prepopulate"); | |
354 | fi->dir_release_count--; /* preclude I_COMPLETE */ | |
355 | } | |
356 | ||
357 | /* note next offset and last dentry name */ | |
358 | fi->offset = fi->next_offset; | |
359 | fi->last_readdir = req; | |
360 | ||
361 | if (req->r_reply_info.dir_end) { | |
362 | kfree(fi->last_name); | |
363 | fi->last_name = NULL; | |
7b88dadc SW |
364 | if (ceph_frag_is_rightmost(frag)) |
365 | fi->next_offset = 2; | |
366 | else | |
367 | fi->next_offset = 0; | |
2817b000 SW |
368 | } else { |
369 | rinfo = &req->r_reply_info; | |
370 | err = note_last_dentry(fi, | |
371 | rinfo->dir_dname[rinfo->dir_nr-1], | |
372 | rinfo->dir_dname_len[rinfo->dir_nr-1]); | |
373 | if (err) | |
374 | return err; | |
375 | fi->next_offset += rinfo->dir_nr; | |
376 | } | |
377 | } | |
378 | ||
379 | rinfo = &fi->last_readdir->r_reply_info; | |
380 | dout("readdir frag %x num %d off %d chunkoff %d\n", frag, | |
381 | rinfo->dir_nr, off, fi->offset); | |
da39822c | 382 | while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) { |
2817b000 SW |
383 | u64 pos = ceph_make_fpos(frag, off); |
384 | struct ceph_mds_reply_inode *in = | |
385 | rinfo->dir_in[off - fi->offset].in; | |
3105c19c SW |
386 | struct ceph_vino vino; |
387 | ino_t ino; | |
388 | ||
2817b000 SW |
389 | dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n", |
390 | off, off - fi->offset, rinfo->dir_nr, pos, | |
391 | rinfo->dir_dname_len[off - fi->offset], | |
392 | rinfo->dir_dname[off - fi->offset], in); | |
393 | BUG_ON(!in); | |
394 | ftype = le32_to_cpu(in->mode) >> 12; | |
3105c19c SW |
395 | vino.ino = le64_to_cpu(in->ino); |
396 | vino.snap = le64_to_cpu(in->snapid); | |
397 | ino = ceph_vino_to_ino(vino); | |
2817b000 SW |
398 | if (filldir(dirent, |
399 | rinfo->dir_dname[off - fi->offset], | |
400 | rinfo->dir_dname_len[off - fi->offset], | |
ad1fee96 YS |
401 | pos, |
402 | ceph_translate_ino(inode->i_sb, ino), ftype) < 0) { | |
2817b000 SW |
403 | dout("filldir stopping us...\n"); |
404 | return 0; | |
405 | } | |
406 | off++; | |
407 | filp->f_pos = pos + 1; | |
408 | } | |
409 | ||
410 | if (fi->last_name) { | |
411 | ceph_mdsc_put_request(fi->last_readdir); | |
412 | fi->last_readdir = NULL; | |
413 | goto more; | |
414 | } | |
415 | ||
416 | /* more frags? */ | |
417 | if (!ceph_frag_is_rightmost(frag)) { | |
418 | frag = ceph_frag_next(frag); | |
419 | off = 0; | |
420 | filp->f_pos = ceph_make_fpos(frag, off); | |
421 | dout("readdir next frag is %x\n", frag); | |
422 | goto more; | |
423 | } | |
9cfa1098 | 424 | fi->flags |= CEPH_F_ATEND; |
2817b000 SW |
425 | |
426 | /* | |
427 | * if dir_release_count still matches the dir, no dentries | |
428 | * were released during the whole readdir, and we should have | |
429 | * the complete dir contents in our cache. | |
430 | */ | |
431 | spin_lock(&inode->i_lock); | |
432 | if (ci->i_release_count == fi->dir_release_count) { | |
433 | dout(" marking %p complete\n", inode); | |
b545cc15 | 434 | /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */ |
2817b000 SW |
435 | ci->i_max_offset = filp->f_pos; |
436 | } | |
437 | spin_unlock(&inode->i_lock); | |
438 | ||
439 | dout("readdir %p filp %p done.\n", inode, filp); | |
440 | return 0; | |
441 | } | |
442 | ||
443 | static void reset_readdir(struct ceph_file_info *fi) | |
444 | { | |
445 | if (fi->last_readdir) { | |
446 | ceph_mdsc_put_request(fi->last_readdir); | |
447 | fi->last_readdir = NULL; | |
448 | } | |
449 | kfree(fi->last_name); | |
a1629c3b | 450 | fi->last_name = NULL; |
2817b000 SW |
451 | fi->next_offset = 2; /* compensate for . and .. */ |
452 | if (fi->dentry) { | |
453 | dput(fi->dentry); | |
454 | fi->dentry = NULL; | |
455 | } | |
9cfa1098 | 456 | fi->flags &= ~CEPH_F_ATEND; |
2817b000 SW |
457 | } |
458 | ||
459 | static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin) | |
460 | { | |
461 | struct ceph_file_info *fi = file->private_data; | |
462 | struct inode *inode = file->f_mapping->host; | |
463 | loff_t old_offset = offset; | |
464 | loff_t retval; | |
465 | ||
466 | mutex_lock(&inode->i_mutex); | |
467 | switch (origin) { | |
468 | case SEEK_END: | |
469 | offset += inode->i_size + 2; /* FIXME */ | |
470 | break; | |
471 | case SEEK_CUR: | |
472 | offset += file->f_pos; | |
473 | } | |
474 | retval = -EINVAL; | |
475 | if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) { | |
476 | if (offset != file->f_pos) { | |
477 | file->f_pos = offset; | |
478 | file->f_version = 0; | |
9cfa1098 | 479 | fi->flags &= ~CEPH_F_ATEND; |
2817b000 SW |
480 | } |
481 | retval = offset; | |
482 | ||
483 | /* | |
484 | * discard buffered readdir content on seekdir(0), or | |
485 | * seek to new frag, or seek prior to current chunk. | |
486 | */ | |
487 | if (offset == 0 || | |
488 | fpos_frag(offset) != fpos_frag(old_offset) || | |
489 | fpos_off(offset) < fi->offset) { | |
490 | dout("dir_llseek dropping %p content\n", file); | |
491 | reset_readdir(fi); | |
492 | } | |
493 | ||
494 | /* bump dir_release_count if we did a forward seek */ | |
495 | if (offset > old_offset) | |
496 | fi->dir_release_count--; | |
497 | } | |
498 | mutex_unlock(&inode->i_mutex); | |
499 | return retval; | |
500 | } | |
501 | ||
502 | /* | |
468640e3 | 503 | * Handle lookups for the hidden .snap directory. |
2817b000 | 504 | */ |
468640e3 SW |
505 | int ceph_handle_snapdir(struct ceph_mds_request *req, |
506 | struct dentry *dentry, int err) | |
2817b000 | 507 | { |
3d14c5d2 | 508 | struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); |
2817b000 SW |
509 | struct inode *parent = dentry->d_parent->d_inode; |
510 | ||
511 | /* .snap dir? */ | |
512 | if (err == -ENOENT && | |
455cec0a | 513 | ceph_snap(parent) == CEPH_NOSNAP && |
6b805185 | 514 | strcmp(dentry->d_name.name, |
3d14c5d2 | 515 | fsc->mount_options->snapdir_name) == 0) { |
2817b000 SW |
516 | struct inode *inode = ceph_get_snapdir(parent); |
517 | dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n", | |
518 | dentry, dentry->d_name.len, dentry->d_name.name, inode); | |
9358c6d4 | 519 | BUG_ON(!d_unhashed(dentry)); |
2817b000 SW |
520 | d_add(dentry, inode); |
521 | err = 0; | |
522 | } | |
468640e3 SW |
523 | return err; |
524 | } | |
2817b000 | 525 | |
468640e3 SW |
526 | /* |
527 | * Figure out final result of a lookup/open request. | |
528 | * | |
529 | * Mainly, make sure we return the final req->r_dentry (if it already | |
530 | * existed) in place of the original VFS-provided dentry when they | |
531 | * differ. | |
532 | * | |
533 | * Gracefully handle the case where the MDS replies with -ENOENT and | |
534 | * no trace (which it may do, at its discretion, e.g., if it doesn't | |
535 | * care to issue a lease on the negative dentry). | |
536 | */ | |
537 | struct dentry *ceph_finish_lookup(struct ceph_mds_request *req, | |
538 | struct dentry *dentry, int err) | |
539 | { | |
2817b000 SW |
540 | if (err == -ENOENT) { |
541 | /* no trace? */ | |
542 | err = 0; | |
543 | if (!req->r_reply_info.head->is_dentry) { | |
544 | dout("ENOENT and no trace, dentry %p inode %p\n", | |
545 | dentry, dentry->d_inode); | |
546 | if (dentry->d_inode) { | |
547 | d_drop(dentry); | |
548 | err = -ENOENT; | |
549 | } else { | |
550 | d_add(dentry, NULL); | |
551 | } | |
552 | } | |
553 | } | |
554 | if (err) | |
555 | dentry = ERR_PTR(err); | |
556 | else if (dentry != req->r_dentry) | |
557 | dentry = dget(req->r_dentry); /* we got spliced */ | |
558 | else | |
559 | dentry = NULL; | |
560 | return dentry; | |
561 | } | |
562 | ||
1d1de916 SW |
563 | static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry) |
564 | { | |
565 | return ceph_ino(inode) == CEPH_INO_ROOT && | |
566 | strncmp(dentry->d_name.name, ".ceph", 5) == 0; | |
567 | } | |
568 | ||
2817b000 SW |
569 | /* |
570 | * Look up a single dir entry. If there is a lookup intent, inform | |
571 | * the MDS so that it gets our 'caps wanted' value in a single op. | |
572 | */ | |
573 | static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry, | |
574 | struct nameidata *nd) | |
575 | { | |
3d14c5d2 YS |
576 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
577 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
578 | struct ceph_mds_request *req; |
579 | int op; | |
580 | int err; | |
581 | ||
582 | dout("lookup %p dentry %p '%.*s'\n", | |
583 | dir, dentry, dentry->d_name.len, dentry->d_name.name); | |
584 | ||
585 | if (dentry->d_name.len > NAME_MAX) | |
586 | return ERR_PTR(-ENAMETOOLONG); | |
587 | ||
588 | err = ceph_init_dentry(dentry); | |
589 | if (err < 0) | |
590 | return ERR_PTR(err); | |
591 | ||
592 | /* open (but not create!) intent? */ | |
593 | if (nd && | |
594 | (nd->flags & LOOKUP_OPEN) && | |
595 | (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */ | |
596 | !(nd->intent.open.flags & O_CREAT)) { | |
597 | int mode = nd->intent.open.create_mode & ~current->fs->umask; | |
598 | return ceph_lookup_open(dir, dentry, nd, mode, 1); | |
599 | } | |
600 | ||
601 | /* can we conclude ENOENT locally? */ | |
602 | if (dentry->d_inode == NULL) { | |
603 | struct ceph_inode_info *ci = ceph_inode(dir); | |
604 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
605 | ||
606 | spin_lock(&dir->i_lock); | |
607 | dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags); | |
608 | if (strncmp(dentry->d_name.name, | |
3d14c5d2 | 609 | fsc->mount_options->snapdir_name, |
2817b000 | 610 | dentry->d_name.len) && |
1d1de916 | 611 | !is_root_ceph_dentry(dir, dentry) && |
2817b000 SW |
612 | (ci->i_ceph_flags & CEPH_I_COMPLETE) && |
613 | (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) { | |
2817b000 SW |
614 | spin_unlock(&dir->i_lock); |
615 | dout(" dir %p complete, -ENOENT\n", dir); | |
616 | d_add(dentry, NULL); | |
617 | di->lease_shared_gen = ci->i_shared_gen; | |
618 | return NULL; | |
619 | } | |
620 | spin_unlock(&dir->i_lock); | |
621 | } | |
622 | ||
623 | op = ceph_snap(dir) == CEPH_SNAPDIR ? | |
624 | CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
625 | req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS); | |
626 | if (IS_ERR(req)) | |
7e34bc52 | 627 | return ERR_CAST(req); |
2817b000 SW |
628 | req->r_dentry = dget(dentry); |
629 | req->r_num_caps = 2; | |
630 | /* we only need inode linkage */ | |
631 | req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE); | |
632 | req->r_locked_dir = dir; | |
633 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
468640e3 | 634 | err = ceph_handle_snapdir(req, dentry, err); |
2817b000 SW |
635 | dentry = ceph_finish_lookup(req, dentry, err); |
636 | ceph_mdsc_put_request(req); /* will dput(dentry) */ | |
637 | dout("lookup result=%p\n", dentry); | |
638 | return dentry; | |
639 | } | |
640 | ||
641 | /* | |
642 | * If we do a create but get no trace back from the MDS, follow up with | |
643 | * a lookup (the VFS expects us to link up the provided dentry). | |
644 | */ | |
645 | int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry) | |
646 | { | |
647 | struct dentry *result = ceph_lookup(dir, dentry, NULL); | |
648 | ||
649 | if (result && !IS_ERR(result)) { | |
650 | /* | |
651 | * We created the item, then did a lookup, and found | |
652 | * it was already linked to another inode we already | |
653 | * had in our cache (and thus got spliced). Link our | |
654 | * dentry to that inode, but don't hash it, just in | |
655 | * case the VFS wants to dereference it. | |
656 | */ | |
657 | BUG_ON(!result->d_inode); | |
658 | d_instantiate(dentry, result->d_inode); | |
659 | return 0; | |
660 | } | |
661 | return PTR_ERR(result); | |
662 | } | |
663 | ||
664 | static int ceph_mknod(struct inode *dir, struct dentry *dentry, | |
665 | int mode, dev_t rdev) | |
666 | { | |
3d14c5d2 YS |
667 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
668 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
669 | struct ceph_mds_request *req; |
670 | int err; | |
671 | ||
672 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
673 | return -EROFS; | |
674 | ||
675 | dout("mknod in dir %p dentry %p mode 0%o rdev %d\n", | |
676 | dir, dentry, mode, rdev); | |
677 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS); | |
678 | if (IS_ERR(req)) { | |
679 | d_drop(dentry); | |
680 | return PTR_ERR(req); | |
681 | } | |
682 | req->r_dentry = dget(dentry); | |
683 | req->r_num_caps = 2; | |
684 | req->r_locked_dir = dir; | |
685 | req->r_args.mknod.mode = cpu_to_le32(mode); | |
686 | req->r_args.mknod.rdev = cpu_to_le32(rdev); | |
687 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
688 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
689 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
690 | if (!err && !req->r_reply_info.head->is_dentry) | |
691 | err = ceph_handle_notrace_create(dir, dentry); | |
692 | ceph_mdsc_put_request(req); | |
693 | if (err) | |
694 | d_drop(dentry); | |
695 | return err; | |
696 | } | |
697 | ||
698 | static int ceph_create(struct inode *dir, struct dentry *dentry, int mode, | |
699 | struct nameidata *nd) | |
700 | { | |
701 | dout("create in dir %p dentry %p name '%.*s'\n", | |
702 | dir, dentry, dentry->d_name.len, dentry->d_name.name); | |
703 | ||
704 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
705 | return -EROFS; | |
706 | ||
707 | if (nd) { | |
708 | BUG_ON((nd->flags & LOOKUP_OPEN) == 0); | |
709 | dentry = ceph_lookup_open(dir, dentry, nd, mode, 0); | |
710 | /* hrm, what should i do here if we get aliased? */ | |
711 | if (IS_ERR(dentry)) | |
712 | return PTR_ERR(dentry); | |
713 | return 0; | |
714 | } | |
715 | ||
716 | /* fall back to mknod */ | |
717 | return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0); | |
718 | } | |
719 | ||
720 | static int ceph_symlink(struct inode *dir, struct dentry *dentry, | |
721 | const char *dest) | |
722 | { | |
3d14c5d2 YS |
723 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
724 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
725 | struct ceph_mds_request *req; |
726 | int err; | |
727 | ||
728 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
729 | return -EROFS; | |
730 | ||
731 | dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest); | |
732 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS); | |
733 | if (IS_ERR(req)) { | |
734 | d_drop(dentry); | |
735 | return PTR_ERR(req); | |
736 | } | |
737 | req->r_dentry = dget(dentry); | |
738 | req->r_num_caps = 2; | |
739 | req->r_path2 = kstrdup(dest, GFP_NOFS); | |
740 | req->r_locked_dir = dir; | |
741 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
742 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
743 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
744 | if (!err && !req->r_reply_info.head->is_dentry) | |
745 | err = ceph_handle_notrace_create(dir, dentry); | |
746 | ceph_mdsc_put_request(req); | |
747 | if (err) | |
748 | d_drop(dentry); | |
749 | return err; | |
750 | } | |
751 | ||
752 | static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
753 | { | |
3d14c5d2 YS |
754 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
755 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
756 | struct ceph_mds_request *req; |
757 | int err = -EROFS; | |
758 | int op; | |
759 | ||
760 | if (ceph_snap(dir) == CEPH_SNAPDIR) { | |
761 | /* mkdir .snap/foo is a MKSNAP */ | |
762 | op = CEPH_MDS_OP_MKSNAP; | |
763 | dout("mksnap dir %p snap '%.*s' dn %p\n", dir, | |
764 | dentry->d_name.len, dentry->d_name.name, dentry); | |
765 | } else if (ceph_snap(dir) == CEPH_NOSNAP) { | |
766 | dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode); | |
767 | op = CEPH_MDS_OP_MKDIR; | |
768 | } else { | |
769 | goto out; | |
770 | } | |
771 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
772 | if (IS_ERR(req)) { | |
773 | err = PTR_ERR(req); | |
774 | goto out; | |
775 | } | |
776 | ||
777 | req->r_dentry = dget(dentry); | |
778 | req->r_num_caps = 2; | |
779 | req->r_locked_dir = dir; | |
780 | req->r_args.mkdir.mode = cpu_to_le32(mode); | |
781 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
782 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
783 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
784 | if (!err && !req->r_reply_info.head->is_dentry) | |
785 | err = ceph_handle_notrace_create(dir, dentry); | |
786 | ceph_mdsc_put_request(req); | |
787 | out: | |
788 | if (err < 0) | |
789 | d_drop(dentry); | |
790 | return err; | |
791 | } | |
792 | ||
793 | static int ceph_link(struct dentry *old_dentry, struct inode *dir, | |
794 | struct dentry *dentry) | |
795 | { | |
3d14c5d2 YS |
796 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
797 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
798 | struct ceph_mds_request *req; |
799 | int err; | |
800 | ||
801 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
802 | return -EROFS; | |
803 | ||
804 | dout("link in dir %p old_dentry %p dentry %p\n", dir, | |
805 | old_dentry, dentry); | |
806 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS); | |
807 | if (IS_ERR(req)) { | |
808 | d_drop(dentry); | |
809 | return PTR_ERR(req); | |
810 | } | |
811 | req->r_dentry = dget(dentry); | |
812 | req->r_num_caps = 2; | |
813 | req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */ | |
814 | req->r_locked_dir = dir; | |
815 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
816 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
817 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
70b666c3 | 818 | if (err) { |
2817b000 | 819 | d_drop(dentry); |
70b666c3 SW |
820 | } else if (!req->r_reply_info.head->is_dentry) { |
821 | ihold(old_dentry->d_inode); | |
822 | d_instantiate(dentry, old_dentry->d_inode); | |
823 | } | |
2817b000 SW |
824 | ceph_mdsc_put_request(req); |
825 | return err; | |
826 | } | |
827 | ||
828 | /* | |
829 | * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it | |
830 | * looks like the link count will hit 0, drop any other caps (other | |
831 | * than PIN) we don't specifically want (due to the file still being | |
832 | * open). | |
833 | */ | |
834 | static int drop_caps_for_unlink(struct inode *inode) | |
835 | { | |
836 | struct ceph_inode_info *ci = ceph_inode(inode); | |
837 | int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
838 | ||
839 | spin_lock(&inode->i_lock); | |
840 | if (inode->i_nlink == 1) { | |
841 | drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN); | |
842 | ci->i_ceph_flags |= CEPH_I_NODELAY; | |
843 | } | |
844 | spin_unlock(&inode->i_lock); | |
845 | return drop; | |
846 | } | |
847 | ||
848 | /* | |
849 | * rmdir and unlink are differ only by the metadata op code | |
850 | */ | |
851 | static int ceph_unlink(struct inode *dir, struct dentry *dentry) | |
852 | { | |
3d14c5d2 YS |
853 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
854 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
855 | struct inode *inode = dentry->d_inode; |
856 | struct ceph_mds_request *req; | |
857 | int err = -EROFS; | |
858 | int op; | |
859 | ||
860 | if (ceph_snap(dir) == CEPH_SNAPDIR) { | |
861 | /* rmdir .snap/foo is RMSNAP */ | |
862 | dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len, | |
863 | dentry->d_name.name, dentry); | |
864 | op = CEPH_MDS_OP_RMSNAP; | |
865 | } else if (ceph_snap(dir) == CEPH_NOSNAP) { | |
866 | dout("unlink/rmdir dir %p dn %p inode %p\n", | |
867 | dir, dentry, inode); | |
868 | op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ? | |
869 | CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK; | |
870 | } else | |
871 | goto out; | |
872 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
873 | if (IS_ERR(req)) { | |
874 | err = PTR_ERR(req); | |
875 | goto out; | |
876 | } | |
877 | req->r_dentry = dget(dentry); | |
878 | req->r_num_caps = 2; | |
879 | req->r_locked_dir = dir; | |
880 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
881 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
882 | req->r_inode_drop = drop_caps_for_unlink(inode); | |
883 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
884 | if (!err && !req->r_reply_info.head->is_dentry) | |
885 | d_delete(dentry); | |
886 | ceph_mdsc_put_request(req); | |
887 | out: | |
888 | return err; | |
889 | } | |
890 | ||
891 | static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry, | |
892 | struct inode *new_dir, struct dentry *new_dentry) | |
893 | { | |
3d14c5d2 YS |
894 | struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb); |
895 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
896 | struct ceph_mds_request *req; |
897 | int err; | |
898 | ||
899 | if (ceph_snap(old_dir) != ceph_snap(new_dir)) | |
900 | return -EXDEV; | |
901 | if (ceph_snap(old_dir) != CEPH_NOSNAP || | |
902 | ceph_snap(new_dir) != CEPH_NOSNAP) | |
903 | return -EROFS; | |
904 | dout("rename dir %p dentry %p to dir %p dentry %p\n", | |
905 | old_dir, old_dentry, new_dir, new_dentry); | |
906 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS); | |
907 | if (IS_ERR(req)) | |
908 | return PTR_ERR(req); | |
909 | req->r_dentry = dget(new_dentry); | |
910 | req->r_num_caps = 2; | |
911 | req->r_old_dentry = dget(old_dentry); | |
912 | req->r_locked_dir = new_dir; | |
913 | req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
914 | req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
915 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
916 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
917 | /* release LINK_RDCACHE on source inode (mds will lock it) */ | |
918 | req->r_old_inode_drop = CEPH_CAP_LINK_SHARED; | |
919 | if (new_dentry->d_inode) | |
920 | req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode); | |
921 | err = ceph_mdsc_do_request(mdsc, old_dir, req); | |
922 | if (!err && !req->r_reply_info.head->is_dentry) { | |
923 | /* | |
924 | * Normally d_move() is done by fill_trace (called by | |
925 | * do_request, above). If there is no trace, we need | |
926 | * to do it here. | |
927 | */ | |
ea1409f9 SW |
928 | |
929 | /* d_move screws up d_subdirs order */ | |
930 | ceph_i_clear(new_dir, CEPH_I_COMPLETE); | |
931 | ||
2817b000 | 932 | d_move(old_dentry, new_dentry); |
ea1409f9 SW |
933 | |
934 | /* ensure target dentry is invalidated, despite | |
935 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d | 936 | ceph_invalidate_dentry_lease(new_dentry); |
2817b000 SW |
937 | } |
938 | ceph_mdsc_put_request(req); | |
939 | return err; | |
940 | } | |
941 | ||
81a6cf2d SW |
942 | /* |
943 | * Ensure a dentry lease will no longer revalidate. | |
944 | */ | |
945 | void ceph_invalidate_dentry_lease(struct dentry *dentry) | |
946 | { | |
947 | spin_lock(&dentry->d_lock); | |
948 | dentry->d_time = jiffies; | |
949 | ceph_dentry(dentry)->lease_shared_gen = 0; | |
950 | spin_unlock(&dentry->d_lock); | |
951 | } | |
2817b000 SW |
952 | |
953 | /* | |
954 | * Check if dentry lease is valid. If not, delete the lease. Try to | |
955 | * renew if the least is more than half up. | |
956 | */ | |
957 | static int dentry_lease_is_valid(struct dentry *dentry) | |
958 | { | |
959 | struct ceph_dentry_info *di; | |
960 | struct ceph_mds_session *s; | |
961 | int valid = 0; | |
962 | u32 gen; | |
963 | unsigned long ttl; | |
964 | struct ceph_mds_session *session = NULL; | |
965 | struct inode *dir = NULL; | |
966 | u32 seq = 0; | |
967 | ||
968 | spin_lock(&dentry->d_lock); | |
969 | di = ceph_dentry(dentry); | |
970 | if (di && di->lease_session) { | |
971 | s = di->lease_session; | |
972 | spin_lock(&s->s_cap_lock); | |
973 | gen = s->s_cap_gen; | |
974 | ttl = s->s_cap_ttl; | |
975 | spin_unlock(&s->s_cap_lock); | |
976 | ||
977 | if (di->lease_gen == gen && | |
978 | time_before(jiffies, dentry->d_time) && | |
979 | time_before(jiffies, ttl)) { | |
980 | valid = 1; | |
981 | if (di->lease_renew_after && | |
982 | time_after(jiffies, di->lease_renew_after)) { | |
983 | /* we should renew */ | |
984 | dir = dentry->d_parent->d_inode; | |
985 | session = ceph_get_mds_session(s); | |
986 | seq = di->lease_seq; | |
987 | di->lease_renew_after = 0; | |
988 | di->lease_renew_from = jiffies; | |
989 | } | |
2817b000 SW |
990 | } |
991 | } | |
992 | spin_unlock(&dentry->d_lock); | |
993 | ||
994 | if (session) { | |
995 | ceph_mdsc_lease_send_msg(session, dir, dentry, | |
996 | CEPH_MDS_LEASE_RENEW, seq); | |
997 | ceph_put_mds_session(session); | |
998 | } | |
999 | dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid); | |
1000 | return valid; | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Check if directory-wide content lease/cap is valid. | |
1005 | */ | |
1006 | static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry) | |
1007 | { | |
1008 | struct ceph_inode_info *ci = ceph_inode(dir); | |
1009 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
1010 | int valid = 0; | |
1011 | ||
1012 | spin_lock(&dir->i_lock); | |
1013 | if (ci->i_shared_gen == di->lease_shared_gen) | |
1014 | valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1); | |
1015 | spin_unlock(&dir->i_lock); | |
1016 | dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n", | |
1017 | dir, (unsigned)ci->i_shared_gen, dentry, | |
1018 | (unsigned)di->lease_shared_gen, valid); | |
1019 | return valid; | |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * Check if cached dentry can be trusted. | |
1024 | */ | |
1025 | static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd) | |
1026 | { | |
bf1c6aca | 1027 | int valid = 0; |
34286d66 NP |
1028 | struct inode *dir; |
1029 | ||
0eb980e3 | 1030 | if (nd && nd->flags & LOOKUP_RCU) |
34286d66 NP |
1031 | return -ECHILD; |
1032 | ||
1cd3935b SW |
1033 | dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry, |
1034 | dentry->d_name.len, dentry->d_name.name, dentry->d_inode, | |
1035 | ceph_dentry(dentry)->offset); | |
2817b000 | 1036 | |
bf1c6aca SW |
1037 | dir = ceph_get_dentry_parent_inode(dentry); |
1038 | ||
2817b000 SW |
1039 | /* always trust cached snapped dentries, snapdir dentry */ |
1040 | if (ceph_snap(dir) != CEPH_NOSNAP) { | |
1041 | dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry, | |
1042 | dentry->d_name.len, dentry->d_name.name, dentry->d_inode); | |
bf1c6aca SW |
1043 | valid = 1; |
1044 | } else if (dentry->d_inode && | |
1045 | ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) { | |
1046 | valid = 1; | |
1047 | } else if (dentry_lease_is_valid(dentry) || | |
1048 | dir_lease_is_valid(dir, dentry)) { | |
1049 | valid = 1; | |
2817b000 | 1050 | } |
2817b000 | 1051 | |
bf1c6aca SW |
1052 | dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid"); |
1053 | if (valid) | |
1054 | ceph_dentry_lru_touch(dentry); | |
1055 | else | |
1056 | d_drop(dentry); | |
1057 | iput(dir); | |
1058 | return valid; | |
2817b000 SW |
1059 | } |
1060 | ||
1061 | /* | |
147851d2 | 1062 | * Release our ceph_dentry_info. |
2817b000 | 1063 | */ |
147851d2 | 1064 | static void ceph_d_release(struct dentry *dentry) |
2817b000 SW |
1065 | { |
1066 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
2817b000 | 1067 | |
147851d2 | 1068 | dout("d_release %p\n", dentry); |
2817b000 SW |
1069 | if (di) { |
1070 | ceph_dentry_lru_del(dentry); | |
1071 | if (di->lease_session) | |
1072 | ceph_put_mds_session(di->lease_session); | |
1073 | kmem_cache_free(ceph_dentry_cachep, di); | |
1074 | dentry->d_fsdata = NULL; | |
1075 | } | |
1076 | } | |
1077 | ||
1078 | static int ceph_snapdir_d_revalidate(struct dentry *dentry, | |
1079 | struct nameidata *nd) | |
1080 | { | |
1081 | /* | |
1082 | * Eventually, we'll want to revalidate snapped metadata | |
1083 | * too... probably... | |
1084 | */ | |
1085 | return 1; | |
1086 | } | |
1087 | ||
1088 | ||
1089 | ||
1090 | /* | |
1091 | * read() on a dir. This weird interface hack only works if mounted | |
1092 | * with '-o dirstat'. | |
1093 | */ | |
1094 | static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size, | |
1095 | loff_t *ppos) | |
1096 | { | |
1097 | struct ceph_file_info *cf = file->private_data; | |
1098 | struct inode *inode = file->f_dentry->d_inode; | |
1099 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1100 | int left; | |
ae598083 | 1101 | const int bufsize = 1024; |
2817b000 | 1102 | |
3d14c5d2 | 1103 | if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT)) |
2817b000 SW |
1104 | return -EISDIR; |
1105 | ||
1106 | if (!cf->dir_info) { | |
ae598083 | 1107 | cf->dir_info = kmalloc(bufsize, GFP_NOFS); |
2817b000 SW |
1108 | if (!cf->dir_info) |
1109 | return -ENOMEM; | |
1110 | cf->dir_info_len = | |
ae598083 | 1111 | snprintf(cf->dir_info, bufsize, |
2817b000 SW |
1112 | "entries: %20lld\n" |
1113 | " files: %20lld\n" | |
1114 | " subdirs: %20lld\n" | |
1115 | "rentries: %20lld\n" | |
1116 | " rfiles: %20lld\n" | |
1117 | " rsubdirs: %20lld\n" | |
1118 | "rbytes: %20lld\n" | |
1119 | "rctime: %10ld.%09ld\n", | |
1120 | ci->i_files + ci->i_subdirs, | |
1121 | ci->i_files, | |
1122 | ci->i_subdirs, | |
1123 | ci->i_rfiles + ci->i_rsubdirs, | |
1124 | ci->i_rfiles, | |
1125 | ci->i_rsubdirs, | |
1126 | ci->i_rbytes, | |
1127 | (long)ci->i_rctime.tv_sec, | |
1128 | (long)ci->i_rctime.tv_nsec); | |
1129 | } | |
1130 | ||
1131 | if (*ppos >= cf->dir_info_len) | |
1132 | return 0; | |
1133 | size = min_t(unsigned, size, cf->dir_info_len-*ppos); | |
1134 | left = copy_to_user(buf, cf->dir_info + *ppos, size); | |
1135 | if (left == size) | |
1136 | return -EFAULT; | |
1137 | *ppos += (size - left); | |
1138 | return size - left; | |
1139 | } | |
1140 | ||
1141 | /* | |
1142 | * an fsync() on a dir will wait for any uncommitted directory | |
1143 | * operations to commit. | |
1144 | */ | |
7ea80859 | 1145 | static int ceph_dir_fsync(struct file *file, int datasync) |
2817b000 | 1146 | { |
7ea80859 | 1147 | struct inode *inode = file->f_path.dentry->d_inode; |
2817b000 SW |
1148 | struct ceph_inode_info *ci = ceph_inode(inode); |
1149 | struct list_head *head = &ci->i_unsafe_dirops; | |
1150 | struct ceph_mds_request *req; | |
1151 | u64 last_tid; | |
1152 | int ret = 0; | |
1153 | ||
1154 | dout("dir_fsync %p\n", inode); | |
1155 | spin_lock(&ci->i_unsafe_lock); | |
1156 | if (list_empty(head)) | |
1157 | goto out; | |
1158 | ||
1159 | req = list_entry(head->prev, | |
1160 | struct ceph_mds_request, r_unsafe_dir_item); | |
1161 | last_tid = req->r_tid; | |
1162 | ||
1163 | do { | |
1164 | ceph_mdsc_get_request(req); | |
1165 | spin_unlock(&ci->i_unsafe_lock); | |
1166 | dout("dir_fsync %p wait on tid %llu (until %llu)\n", | |
1167 | inode, req->r_tid, last_tid); | |
1168 | if (req->r_timeout) { | |
1169 | ret = wait_for_completion_timeout( | |
1170 | &req->r_safe_completion, req->r_timeout); | |
1171 | if (ret > 0) | |
1172 | ret = 0; | |
1173 | else if (ret == 0) | |
1174 | ret = -EIO; /* timed out */ | |
1175 | } else { | |
1176 | wait_for_completion(&req->r_safe_completion); | |
1177 | } | |
1178 | spin_lock(&ci->i_unsafe_lock); | |
1179 | ceph_mdsc_put_request(req); | |
1180 | ||
1181 | if (ret || list_empty(head)) | |
1182 | break; | |
1183 | req = list_entry(head->next, | |
1184 | struct ceph_mds_request, r_unsafe_dir_item); | |
1185 | } while (req->r_tid < last_tid); | |
1186 | out: | |
1187 | spin_unlock(&ci->i_unsafe_lock); | |
1188 | return ret; | |
1189 | } | |
1190 | ||
1191 | /* | |
1192 | * We maintain a private dentry LRU. | |
1193 | * | |
1194 | * FIXME: this needs to be changed to a per-mds lru to be useful. | |
1195 | */ | |
1196 | void ceph_dentry_lru_add(struct dentry *dn) | |
1197 | { | |
1198 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1199 | struct ceph_mds_client *mdsc; | |
2817b000 | 1200 | |
04a419f9 SW |
1201 | dout("dentry_lru_add %p %p '%.*s'\n", di, dn, |
1202 | dn->d_name.len, dn->d_name.name); | |
2817b000 | 1203 | if (di) { |
3d14c5d2 | 1204 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
2817b000 SW |
1205 | spin_lock(&mdsc->dentry_lru_lock); |
1206 | list_add_tail(&di->lru, &mdsc->dentry_lru); | |
1207 | mdsc->num_dentry++; | |
1208 | spin_unlock(&mdsc->dentry_lru_lock); | |
1209 | } | |
1210 | } | |
1211 | ||
1212 | void ceph_dentry_lru_touch(struct dentry *dn) | |
1213 | { | |
1214 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1215 | struct ceph_mds_client *mdsc; | |
2817b000 | 1216 | |
1cd3935b SW |
1217 | dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn, |
1218 | dn->d_name.len, dn->d_name.name, di->offset); | |
2817b000 | 1219 | if (di) { |
3d14c5d2 | 1220 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
2817b000 SW |
1221 | spin_lock(&mdsc->dentry_lru_lock); |
1222 | list_move_tail(&di->lru, &mdsc->dentry_lru); | |
1223 | spin_unlock(&mdsc->dentry_lru_lock); | |
1224 | } | |
1225 | } | |
1226 | ||
1227 | void ceph_dentry_lru_del(struct dentry *dn) | |
1228 | { | |
1229 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1230 | struct ceph_mds_client *mdsc; | |
1231 | ||
04a419f9 SW |
1232 | dout("dentry_lru_del %p %p '%.*s'\n", di, dn, |
1233 | dn->d_name.len, dn->d_name.name); | |
2817b000 | 1234 | if (di) { |
3d14c5d2 | 1235 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
2817b000 SW |
1236 | spin_lock(&mdsc->dentry_lru_lock); |
1237 | list_del_init(&di->lru); | |
1238 | mdsc->num_dentry--; | |
1239 | spin_unlock(&mdsc->dentry_lru_lock); | |
1240 | } | |
1241 | } | |
1242 | ||
6c0f3af7 SW |
1243 | /* |
1244 | * Return name hash for a given dentry. This is dependent on | |
1245 | * the parent directory's hash function. | |
1246 | */ | |
e5f86dc3 | 1247 | unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn) |
6c0f3af7 | 1248 | { |
6c0f3af7 SW |
1249 | struct ceph_inode_info *dci = ceph_inode(dir); |
1250 | ||
1251 | switch (dci->i_dir_layout.dl_dir_hash) { | |
1252 | case 0: /* for backward compat */ | |
1253 | case CEPH_STR_HASH_LINUX: | |
1254 | return dn->d_name.hash; | |
1255 | ||
1256 | default: | |
1257 | return ceph_str_hash(dci->i_dir_layout.dl_dir_hash, | |
1258 | dn->d_name.name, dn->d_name.len); | |
1259 | } | |
1260 | } | |
1261 | ||
2817b000 SW |
1262 | const struct file_operations ceph_dir_fops = { |
1263 | .read = ceph_read_dir, | |
1264 | .readdir = ceph_readdir, | |
1265 | .llseek = ceph_dir_llseek, | |
1266 | .open = ceph_open, | |
1267 | .release = ceph_release, | |
1268 | .unlocked_ioctl = ceph_ioctl, | |
1269 | .fsync = ceph_dir_fsync, | |
1270 | }; | |
1271 | ||
1272 | const struct inode_operations ceph_dir_iops = { | |
1273 | .lookup = ceph_lookup, | |
1274 | .permission = ceph_permission, | |
1275 | .getattr = ceph_getattr, | |
1276 | .setattr = ceph_setattr, | |
1277 | .setxattr = ceph_setxattr, | |
1278 | .getxattr = ceph_getxattr, | |
1279 | .listxattr = ceph_listxattr, | |
1280 | .removexattr = ceph_removexattr, | |
1281 | .mknod = ceph_mknod, | |
1282 | .symlink = ceph_symlink, | |
1283 | .mkdir = ceph_mkdir, | |
1284 | .link = ceph_link, | |
1285 | .unlink = ceph_unlink, | |
1286 | .rmdir = ceph_unlink, | |
1287 | .rename = ceph_rename, | |
1288 | .create = ceph_create, | |
1289 | }; | |
1290 | ||
52dfb8ac | 1291 | const struct dentry_operations ceph_dentry_ops = { |
2817b000 | 1292 | .d_revalidate = ceph_d_revalidate, |
147851d2 | 1293 | .d_release = ceph_d_release, |
2817b000 SW |
1294 | }; |
1295 | ||
52dfb8ac | 1296 | const struct dentry_operations ceph_snapdir_dentry_ops = { |
2817b000 | 1297 | .d_revalidate = ceph_snapdir_d_revalidate, |
147851d2 | 1298 | .d_release = ceph_d_release, |
2817b000 SW |
1299 | }; |
1300 | ||
52dfb8ac | 1301 | const struct dentry_operations ceph_snap_dentry_ops = { |
147851d2 | 1302 | .d_release = ceph_d_release, |
2817b000 | 1303 | }; |