blob: 6e0fedf6713b5130af5c79cdb14649e90ba8427c [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil2817b002009-10-06 11:31:08 -07002
3#include <linux/spinlock.h>
4#include <linux/fs_struct.h>
5#include <linux/namei.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09006#include <linux/slab.h>
Sage Weil2817b002009-10-06 11:31:08 -07007#include <linux/sched.h>
Andreas Gruenbacher2cdeb1e2016-04-14 00:30:17 +02008#include <linux/xattr.h>
Sage Weil2817b002009-10-06 11:31:08 -07009
10#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070011#include "mds_client.h"
Sage Weil2817b002009-10-06 11:31:08 -070012
13/*
14 * Directory operations: readdir, lookup, create, link, unlink,
15 * rename, etc.
16 */
17
18/*
19 * Ceph MDS operations are specified in terms of a base ino and
20 * relative path. Thus, the client can specify an operation on a
21 * specific inode (e.g., a getattr due to fstat(2)), or as a path
22 * relative to, say, the root directory.
23 *
24 * Normally, we limit ourselves to strict inode ops (no path component)
25 * or dentry operations (a single path component relative to an ino). The
26 * exception to this is open_root_dentry(), which will open the mount
27 * point by name.
28 */
29
Sage Weil52dfb8a2010-08-03 10:25:30 -070030const struct dentry_operations ceph_dentry_ops;
Sage Weil2817b002009-10-06 11:31:08 -070031
32/*
33 * Initialize ceph dentry state.
34 */
35int ceph_init_dentry(struct dentry *dentry)
36{
37 struct ceph_dentry_info *di;
38
39 if (dentry->d_fsdata)
40 return 0;
41
Geliang Tang99ec2692016-03-13 15:26:29 +080042 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
Sage Weil2817b002009-10-06 11:31:08 -070043 if (!di)
44 return -ENOMEM; /* oh well */
45
46 spin_lock(&dentry->d_lock);
Sage Weil8c6efb52010-04-23 11:36:54 -070047 if (dentry->d_fsdata) {
48 /* lost a race */
49 kmem_cache_free(ceph_dentry_cachep, di);
Sage Weil2817b002009-10-06 11:31:08 -070050 goto out_unlock;
Sage Weil8c6efb52010-04-23 11:36:54 -070051 }
Sage Weil48d0cbd2011-07-26 11:30:15 -070052
David Howells2b0143b2015-03-17 22:25:59 +000053 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP)
Sage Weil48d0cbd2011-07-26 11:30:15 -070054 d_set_d_op(dentry, &ceph_dentry_ops);
David Howells2b0143b2015-03-17 22:25:59 +000055 else if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
Sage Weil48d0cbd2011-07-26 11:30:15 -070056 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
57 else
58 d_set_d_op(dentry, &ceph_snap_dentry_ops);
59
Sage Weil2817b002009-10-06 11:31:08 -070060 di->dentry = dentry;
61 di->lease_session = NULL;
Sage Weil2817b002009-10-06 11:31:08 -070062 dentry->d_time = jiffies;
Sage Weil48d0cbd2011-07-26 11:30:15 -070063 /* avoid reordering d_fsdata setup so that the check above is safe */
64 smp_mb();
65 dentry->d_fsdata = di;
Sage Weil2817b002009-10-06 11:31:08 -070066 ceph_dentry_lru_add(dentry);
67out_unlock:
68 spin_unlock(&dentry->d_lock);
69 return 0;
70}
71
Sage Weil2817b002009-10-06 11:31:08 -070072/*
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +080073 * for f_pos for readdir:
74 * - hash order:
75 * (0xff << 52) | ((24 bits hash) << 28) |
76 * (the nth entry has hash collision);
77 * - frag+name order;
78 * ((frag value) << 28) | (the nth entry in frag);
Sage Weil2817b002009-10-06 11:31:08 -070079 */
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +080080#define OFFSET_BITS 28
81#define OFFSET_MASK ((1 << OFFSET_BITS) - 1)
82#define HASH_ORDER (0xffull << (OFFSET_BITS + 24))
83loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
84{
85 loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
86 if (hash_order)
87 fpos |= HASH_ORDER;
88 return fpos;
89}
90
91static bool is_hash_order(loff_t p)
92{
93 return (p & HASH_ORDER) == HASH_ORDER;
94}
95
Sage Weil2817b002009-10-06 11:31:08 -070096static unsigned fpos_frag(loff_t p)
97{
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +080098 return p >> OFFSET_BITS;
Sage Weil2817b002009-10-06 11:31:08 -070099}
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800100
101static unsigned fpos_hash(loff_t p)
102{
103 return ceph_frag_value(fpos_frag(p));
104}
105
Sage Weil2817b002009-10-06 11:31:08 -0700106static unsigned fpos_off(loff_t p)
107{
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800108 return p & OFFSET_MASK;
Sage Weil2817b002009-10-06 11:31:08 -0700109}
110
Yan, Zheng4d5f5df2014-02-13 19:40:26 +0800111static int fpos_cmp(loff_t l, loff_t r)
112{
113 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
114 if (v)
115 return v;
116 return (int)(fpos_off(l) - fpos_off(r));
117}
118
Sage Weil2817b002009-10-06 11:31:08 -0700119/*
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800120 * make note of the last dentry we read, so we can
121 * continue at the same lexicographical point,
122 * regardless of what dir changes take place on the
123 * server.
124 */
125static int note_last_dentry(struct ceph_file_info *fi, const char *name,
126 int len, unsigned next_offset)
127{
128 char *buf = kmalloc(len+1, GFP_KERNEL);
129 if (!buf)
130 return -ENOMEM;
131 kfree(fi->last_name);
132 fi->last_name = buf;
133 memcpy(fi->last_name, name, len);
134 fi->last_name[len] = 0;
135 fi->next_offset = next_offset;
136 dout("note_last_dentry '%s'\n", fi->last_name);
137 return 0;
138}
139
Yan, Zhengc530cd22016-04-28 17:43:35 +0800140
141static struct dentry *
142__dcache_find_get_entry(struct dentry *parent, u64 idx,
143 struct ceph_readdir_cache_control *cache_ctl)
144{
145 struct inode *dir = d_inode(parent);
146 struct dentry *dentry;
147 unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
148 loff_t ptr_pos = idx * sizeof(struct dentry *);
149 pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
150
151 if (ptr_pos >= i_size_read(dir))
152 return NULL;
153
154 if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
155 ceph_readdir_cache_release(cache_ctl);
156 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
157 if (!cache_ctl->page) {
158 dout(" page %lu not found\n", ptr_pgoff);
159 return ERR_PTR(-EAGAIN);
160 }
161 /* reading/filling the cache are serialized by
162 i_mutex, no need to use page lock */
163 unlock_page(cache_ctl->page);
164 cache_ctl->dentries = kmap(cache_ctl->page);
165 }
166
167 cache_ctl->index = idx & idx_mask;
168
169 rcu_read_lock();
170 spin_lock(&parent->d_lock);
171 /* check i_size again here, because empty directory can be
172 * marked as complete while not holding the i_mutex. */
173 if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
174 dentry = cache_ctl->dentries[cache_ctl->index];
175 else
176 dentry = NULL;
177 spin_unlock(&parent->d_lock);
178 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
179 dentry = NULL;
180 rcu_read_unlock();
181 return dentry ? : ERR_PTR(-EAGAIN);
182}
183
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800184/*
Sage Weil2817b002009-10-06 11:31:08 -0700185 * When possible, we try to satisfy a readdir by peeking at the
186 * dcache. We make this work by carefully ordering dentries on
Al Viro946e51f2014-10-26 19:19:16 -0400187 * d_child when we initially get results back from the MDS, and
Sage Weil2817b002009-10-06 11:31:08 -0700188 * falling back to a "normal" sync readdir if any dentries in the dir
189 * are dropped.
190 *
Yan, Zheng2f276c52013-03-13 19:44:32 +0800191 * Complete dir indicates that we have all dentries in the dir. It is
Sage Weil2817b002009-10-06 11:31:08 -0700192 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
193 * the MDS if/when the directory is modified).
194 */
Yan, Zhenga30be7c2014-04-06 14:10:04 +0800195static int __dcache_readdir(struct file *file, struct dir_context *ctx,
196 u32 shared_gen)
Sage Weil2817b002009-10-06 11:31:08 -0700197{
Al Viro77acfa22013-05-17 16:52:26 -0400198 struct ceph_file_info *fi = file->private_data;
Al Virob5830432014-10-31 01:22:04 -0400199 struct dentry *parent = file->f_path.dentry;
David Howells2b0143b2015-03-17 22:25:59 +0000200 struct inode *dir = d_inode(parent);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800201 struct dentry *dentry, *last = NULL;
Sage Weil2817b002009-10-06 11:31:08 -0700202 struct ceph_dentry_info *di;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800203 struct ceph_readdir_cache_control cache_ctl = {};
Yan, Zhengc530cd22016-04-28 17:43:35 +0800204 u64 idx = 0;
205 int err = 0;
Sage Weil2817b002009-10-06 11:31:08 -0700206
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800207 dout("__dcache_readdir %p v%u at %llx\n", dir, shared_gen, ctx->pos);
Sage Weil2817b002009-10-06 11:31:08 -0700208
Yan, Zhengc530cd22016-04-28 17:43:35 +0800209 /* search start position */
210 if (ctx->pos > 2) {
211 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
212 while (count > 0) {
213 u64 step = count >> 1;
214 dentry = __dcache_find_get_entry(parent, idx + step,
215 &cache_ctl);
216 if (!dentry) {
217 /* use linar search */
218 idx = 0;
219 break;
220 }
221 if (IS_ERR(dentry)) {
222 err = PTR_ERR(dentry);
223 goto out;
224 }
225 di = ceph_dentry(dentry);
226 spin_lock(&dentry->d_lock);
227 if (fpos_cmp(di->offset, ctx->pos) < 0) {
228 idx += step + 1;
229 count -= step + 1;
230 } else {
231 count = step;
232 }
233 spin_unlock(&dentry->d_lock);
234 dput(dentry);
235 }
236
237 dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
Sage Weil2817b002009-10-06 11:31:08 -0700238 }
239
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800240
Yan, Zhengc530cd22016-04-28 17:43:35 +0800241 for (;;) {
242 bool emit_dentry = false;
243 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
244 if (!dentry) {
Sage Weil9cfa1092011-07-26 11:26:18 -0700245 fi->flags |= CEPH_F_ATEND;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800246 err = 0;
247 break;
Sage Weil2817b002009-10-06 11:31:08 -0700248 }
Yan, Zhengc530cd22016-04-28 17:43:35 +0800249 if (IS_ERR(dentry)) {
250 err = PTR_ERR(dentry);
251 goto out;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800252 }
253
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800254 di = ceph_dentry(dentry);
255 spin_lock(&dentry->d_lock);
Yan, Zhenga30be7c2014-04-06 14:10:04 +0800256 if (di->lease_shared_gen == shared_gen &&
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800257 d_really_is_positive(dentry) &&
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800258 fpos_cmp(ctx->pos, di->offset) <= 0) {
259 emit_dentry = true;
Sage Weil2817b002009-10-06 11:31:08 -0700260 }
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800261 spin_unlock(&dentry->d_lock);
262
263 if (emit_dentry) {
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800264 dout(" %llx dentry %p %pd %p\n", di->offset,
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800265 dentry, dentry, d_inode(dentry));
266 ctx->pos = di->offset;
267 if (!dir_emit(ctx, dentry->d_name.name,
268 dentry->d_name.len,
269 ceph_translate_ino(dentry->d_sb,
270 d_inode(dentry)->i_ino),
271 d_inode(dentry)->i_mode >> 12)) {
272 dput(dentry);
273 err = 0;
274 break;
275 }
276 ctx->pos++;
277
278 if (last)
279 dput(last);
280 last = dentry;
281 } else {
282 dput(dentry);
283 }
Sage Weil2817b002009-10-06 11:31:08 -0700284 }
Yan, Zhengc530cd22016-04-28 17:43:35 +0800285out:
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800286 ceph_readdir_cache_release(&cache_ctl);
287 if (last) {
288 int ret;
289 di = ceph_dentry(last);
290 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
291 fpos_off(di->offset) + 1);
292 if (ret < 0)
293 err = ret;
Al Viro77acfa22013-05-17 16:52:26 -0400294 dput(last);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800295 }
Sage Weil2817b002009-10-06 11:31:08 -0700296 return err;
297}
298
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800299static bool need_send_readdir(struct ceph_file_info *fi, loff_t pos)
300{
301 if (!fi->last_readdir)
302 return true;
303 if (is_hash_order(pos))
304 return !ceph_frag_contains_value(fi->frag, fpos_hash(pos));
305 else
306 return fi->frag != fpos_frag(pos);
307}
308
Al Viro77acfa22013-05-17 16:52:26 -0400309static int ceph_readdir(struct file *file, struct dir_context *ctx)
Sage Weil2817b002009-10-06 11:31:08 -0700310{
Al Viro77acfa22013-05-17 16:52:26 -0400311 struct ceph_file_info *fi = file->private_data;
312 struct inode *inode = file_inode(file);
Sage Weil2817b002009-10-06 11:31:08 -0700313 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700314 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
315 struct ceph_mds_client *mdsc = fsc->mdsc;
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800316 int i;
Sage Weil2817b002009-10-06 11:31:08 -0700317 int err;
318 u32 ftype;
319 struct ceph_mds_reply_info_parsed *rinfo;
Sage Weil2817b002009-10-06 11:31:08 -0700320
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800321 dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
Sage Weil9cfa1092011-07-26 11:26:18 -0700322 if (fi->flags & CEPH_F_ATEND)
Sage Weil2817b002009-10-06 11:31:08 -0700323 return 0;
324
325 /* always start with . and .. */
Al Viro77acfa22013-05-17 16:52:26 -0400326 if (ctx->pos == 0) {
Sage Weil2817b002009-10-06 11:31:08 -0700327 dout("readdir off 0 -> '.'\n");
Al Viro77acfa22013-05-17 16:52:26 -0400328 if (!dir_emit(ctx, ".", 1,
Yehuda Sadehad1fee92011-01-21 16:44:03 -0800329 ceph_translate_ino(inode->i_sb, inode->i_ino),
Al Viro77acfa22013-05-17 16:52:26 -0400330 inode->i_mode >> 12))
Sage Weil2817b002009-10-06 11:31:08 -0700331 return 0;
Al Viro77acfa22013-05-17 16:52:26 -0400332 ctx->pos = 1;
Sage Weil2817b002009-10-06 11:31:08 -0700333 }
Al Viro77acfa22013-05-17 16:52:26 -0400334 if (ctx->pos == 1) {
Al Virob5830432014-10-31 01:22:04 -0400335 ino_t ino = parent_ino(file->f_path.dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700336 dout("readdir off 1 -> '..'\n");
Al Viro77acfa22013-05-17 16:52:26 -0400337 if (!dir_emit(ctx, "..", 2,
Yehuda Sadehad1fee92011-01-21 16:44:03 -0800338 ceph_translate_ino(inode->i_sb, ino),
Al Viro77acfa22013-05-17 16:52:26 -0400339 inode->i_mode >> 12))
Sage Weil2817b002009-10-06 11:31:08 -0700340 return 0;
Al Viro77acfa22013-05-17 16:52:26 -0400341 ctx->pos = 2;
Sage Weil2817b002009-10-06 11:31:08 -0700342 }
343
344 /* can we use the dcache? */
Sage Weilbe655592011-11-30 09:47:09 -0800345 spin_lock(&ci->i_ceph_lock);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800346 if (ceph_test_mount_opt(fsc, DCACHE) &&
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700347 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
Sage Weila0dff782010-07-22 13:47:21 -0700348 ceph_snap(inode) != CEPH_SNAPDIR &&
Yan, Zheng70db4f32014-10-21 18:09:56 -0700349 __ceph_dir_is_complete_ordered(ci) &&
Sage Weil2817b002009-10-06 11:31:08 -0700350 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
Yan, Zhenga30be7c2014-04-06 14:10:04 +0800351 u32 shared_gen = ci->i_shared_gen;
Sage Weilbe655592011-11-30 09:47:09 -0800352 spin_unlock(&ci->i_ceph_lock);
Yan, Zhenga30be7c2014-04-06 14:10:04 +0800353 err = __dcache_readdir(file, ctx, shared_gen);
Sage Weilefa4c122010-10-18 14:04:31 -0700354 if (err != -EAGAIN)
Sage Weil2817b002009-10-06 11:31:08 -0700355 return err;
Sage Weilefa4c122010-10-18 14:04:31 -0700356 } else {
Sage Weilbe655592011-11-30 09:47:09 -0800357 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700358 }
Sage Weil2817b002009-10-06 11:31:08 -0700359
360 /* proceed with a normal readdir */
Sage Weil2817b002009-10-06 11:31:08 -0700361more:
362 /* do we have the correct frag content buffered? */
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800363 if (need_send_readdir(fi, ctx->pos)) {
Sage Weil2817b002009-10-06 11:31:08 -0700364 struct ceph_mds_request *req;
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800365 unsigned frag;
Sage Weil2817b002009-10-06 11:31:08 -0700366 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
367 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
368
369 /* discard old result, if any */
Sage Weil393f6622010-03-10 12:03:32 -0800370 if (fi->last_readdir) {
Sage Weil2817b002009-10-06 11:31:08 -0700371 ceph_mdsc_put_request(fi->last_readdir);
Sage Weil393f6622010-03-10 12:03:32 -0800372 fi->last_readdir = NULL;
373 }
Sage Weil2817b002009-10-06 11:31:08 -0700374
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800375 if (is_hash_order(ctx->pos)) {
376 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
377 NULL, NULL);
378 } else {
379 frag = fpos_frag(ctx->pos);
380 }
381
Sage Weil2817b002009-10-06 11:31:08 -0700382 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
383 ceph_vinop(inode), frag, fi->last_name);
384 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
385 if (IS_ERR(req))
386 return PTR_ERR(req);
Yan, Zheng54008392014-03-29 13:41:15 +0800387 err = ceph_alloc_readdir_reply_buffer(req, inode);
388 if (err) {
389 ceph_mdsc_put_request(req);
390 return err;
391 }
Sage Weil2817b002009-10-06 11:31:08 -0700392 /* hints to request -> mds selection code */
393 req->r_direct_mode = USE_AUTH_MDS;
394 req->r_direct_hash = ceph_frag_value(frag);
395 req->r_direct_is_hash = true;
Sanidhya Kashyapa149bb92015-03-21 12:54:58 -0400396 if (fi->last_name) {
Yan, Zheng687265e2015-06-13 17:27:05 +0800397 req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
Sanidhya Kashyapa149bb92015-03-21 12:54:58 -0400398 if (!req->r_path2) {
399 ceph_mdsc_put_request(req);
400 return -ENOMEM;
401 }
402 }
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800403 req->r_dir_release_cnt = fi->dir_release_count;
404 req->r_dir_ordered_cnt = fi->dir_ordered_count;
405 req->r_readdir_cache_idx = fi->readdir_cache_idx;
Sage Weil2817b002009-10-06 11:31:08 -0700406 req->r_readdir_offset = fi->next_offset;
407 req->r_args.readdir.frag = cpu_to_le32(frag);
Yan, Zheng956d39d2016-04-27 17:48:30 +0800408 req->r_args.readdir.flags =
409 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
Sanidhya Kashyapa149bb92015-03-21 12:54:58 -0400410
411 req->r_inode = inode;
412 ihold(inode);
413 req->r_dentry = dget(file->f_path.dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700414 err = ceph_mdsc_do_request(mdsc, NULL, req);
415 if (err < 0) {
416 ceph_mdsc_put_request(req);
417 return err;
418 }
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800419 dout("readdir got and parsed readdir result=%d on "
420 "frag %x, end=%d, complete=%d, hash_order=%d\n",
421 err, frag,
Sage Weil2817b002009-10-06 11:31:08 -0700422 (int)req->r_reply_info.dir_end,
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800423 (int)req->r_reply_info.dir_complete,
424 (int)req->r_reply_info.hash_order);
Sage Weil2817b002009-10-06 11:31:08 -0700425
Yan, Zheng81c6aea2013-09-18 09:44:13 +0800426 rinfo = &req->r_reply_info;
427 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
428 frag = le32_to_cpu(rinfo->dir_dir->frag);
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800429 if (!rinfo->hash_order) {
430 fi->next_offset = req->r_readdir_offset;
431 /* adjust ctx->pos to beginning of frag */
432 ctx->pos = ceph_make_fpos(frag,
433 fi->next_offset,
434 false);
435 }
Yan, Zheng81c6aea2013-09-18 09:44:13 +0800436 }
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800437
Yan, Zhengf0494202014-02-27 16:26:24 +0800438 fi->frag = frag;
Sage Weil2817b002009-10-06 11:31:08 -0700439 fi->last_readdir = req;
440
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800441 if (req->r_did_prepopulate) {
442 fi->readdir_cache_idx = req->r_readdir_cache_idx;
443 if (fi->readdir_cache_idx < 0) {
444 /* preclude from marking dir ordered */
445 fi->dir_ordered_count = 0;
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800446 } else if (ceph_frag_is_leftmost(frag) &&
447 fi->next_offset == 2) {
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800448 /* note dir version at start of readdir so
449 * we can tell if any dentries get dropped */
450 fi->dir_release_count = req->r_dir_release_cnt;
451 fi->dir_ordered_count = req->r_dir_ordered_cnt;
452 }
453 } else {
454 dout("readdir !did_prepopulate");
455 /* disable readdir cache */
456 fi->readdir_cache_idx = -1;
457 /* preclude from marking dir complete */
458 fi->dir_release_count = 0;
459 }
460
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800461 /* note next offset and last dentry name */
462 if (rinfo->dir_nr > 0) {
Yan, Zheng2a5beea2016-04-28 09:37:39 +0800463 struct ceph_mds_reply_dir_entry *rde =
464 rinfo->dir_entries + (rinfo->dir_nr-1);
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800465 unsigned next_offset = req->r_reply_info.dir_end ?
466 2 : (fpos_off(rde->offset) + 1);
Yan, Zheng2a5beea2016-04-28 09:37:39 +0800467 err = note_last_dentry(fi, rde->name, rde->name_len,
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800468 next_offset);
Sage Weil2817b002009-10-06 11:31:08 -0700469 if (err)
470 return err;
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800471 } else if (req->r_reply_info.dir_end) {
472 fi->next_offset = 2;
473 /* keep last name */
Sage Weil2817b002009-10-06 11:31:08 -0700474 }
475 }
476
477 rinfo = &fi->last_readdir->r_reply_info;
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800478 dout("readdir frag %x num %d pos %llx chunk first %llx\n",
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800479 fi->frag, rinfo->dir_nr, ctx->pos,
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800480 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
Al Viro77acfa22013-05-17 16:52:26 -0400481
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800482 i = 0;
483 /* search start position */
484 if (rinfo->dir_nr > 0) {
485 int step, nr = rinfo->dir_nr;
486 while (nr > 0) {
487 step = nr >> 1;
488 if (rinfo->dir_entries[i + step].offset < ctx->pos) {
489 i += step + 1;
490 nr -= step + 1;
491 } else {
492 nr = step;
493 }
494 }
495 }
496 for (; i < rinfo->dir_nr; i++) {
497 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
Sage Weil3105c192010-11-18 09:15:07 -0800498 struct ceph_vino vino;
499 ino_t ino;
500
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800501 BUG_ON(rde->offset < ctx->pos);
502
503 ctx->pos = rde->offset;
504 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
505 i, rinfo->dir_nr, ctx->pos,
Yan, Zheng2a5beea2016-04-28 09:37:39 +0800506 rde->name_len, rde->name, &rde->inode.in);
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800507
Yan, Zheng2a5beea2016-04-28 09:37:39 +0800508 BUG_ON(!rde->inode.in);
509 ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
510 vino.ino = le64_to_cpu(rde->inode.in->ino);
511 vino.snap = le64_to_cpu(rde->inode.in->snapid);
Sage Weil3105c192010-11-18 09:15:07 -0800512 ino = ceph_vino_to_ino(vino);
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800513
Yan, Zheng2a5beea2016-04-28 09:37:39 +0800514 if (!dir_emit(ctx, rde->name, rde->name_len,
515 ceph_translate_ino(inode->i_sb, ino), ftype)) {
Sage Weil2817b002009-10-06 11:31:08 -0700516 dout("filldir stopping us...\n");
517 return 0;
518 }
Al Viro77acfa22013-05-17 16:52:26 -0400519 ctx->pos++;
Sage Weil2817b002009-10-06 11:31:08 -0700520 }
521
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800522 if (fi->next_offset > 2) {
Sage Weil2817b002009-10-06 11:31:08 -0700523 ceph_mdsc_put_request(fi->last_readdir);
524 fi->last_readdir = NULL;
525 goto more;
526 }
527
528 /* more frags? */
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800529 if (!ceph_frag_is_rightmost(fi->frag)) {
530 unsigned frag = ceph_frag_next(fi->frag);
531 if (is_hash_order(ctx->pos)) {
532 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
533 fi->next_offset, true);
534 if (new_pos > ctx->pos)
535 ctx->pos = new_pos;
536 /* keep last_name */
537 } else {
538 ctx->pos = ceph_make_fpos(frag, fi->next_offset, false);
539 kfree(fi->last_name);
540 fi->last_name = NULL;
541 }
Sage Weil2817b002009-10-06 11:31:08 -0700542 dout("readdir next frag is %x\n", frag);
543 goto more;
544 }
Sage Weil9cfa1092011-07-26 11:26:18 -0700545 fi->flags |= CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700546
547 /*
548 * if dir_release_count still matches the dir, no dentries
549 * were released during the whole readdir, and we should have
550 * the complete dir contents in our cache.
551 */
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800552 if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
553 spin_lock(&ci->i_ceph_lock);
554 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
Yan, Zheng70db4f32014-10-21 18:09:56 -0700555 dout(" marking %p complete and ordered\n", inode);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800556 /* use i_size to track number of entries in
557 * readdir cache */
558 BUG_ON(fi->readdir_cache_idx < 0);
559 i_size_write(inode, fi->readdir_cache_idx *
560 sizeof(struct dentry*));
561 } else {
Yan, Zheng70db4f32014-10-21 18:09:56 -0700562 dout(" marking %p complete\n", inode);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800563 }
Yan, Zheng70db4f32014-10-21 18:09:56 -0700564 __ceph_dir_set_complete(ci, fi->dir_release_count,
565 fi->dir_ordered_count);
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800566 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700567 }
Sage Weil2817b002009-10-06 11:31:08 -0700568
Al Viro77acfa22013-05-17 16:52:26 -0400569 dout("readdir %p file %p done.\n", inode, file);
Sage Weil2817b002009-10-06 11:31:08 -0700570 return 0;
571}
572
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800573static void reset_readdir(struct ceph_file_info *fi)
Sage Weil2817b002009-10-06 11:31:08 -0700574{
575 if (fi->last_readdir) {
576 ceph_mdsc_put_request(fi->last_readdir);
577 fi->last_readdir = NULL;
578 }
579 kfree(fi->last_name);
Sage Weila1629c32010-11-11 15:24:06 -0800580 fi->last_name = NULL;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800581 fi->dir_release_count = 0;
582 fi->readdir_cache_idx = -1;
Yan, Zhenga78600e2016-04-27 17:32:34 +0800583 fi->next_offset = 2; /* compensate for . and .. */
Sage Weil9cfa1092011-07-26 11:26:18 -0700584 fi->flags &= ~CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700585}
586
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800587/*
588 * discard buffered readdir content on seekdir(0), or seek to new frag,
589 * or seek prior to current chunk
590 */
591static bool need_reset_readdir(struct ceph_file_info *fi, loff_t new_pos)
592{
593 struct ceph_mds_reply_info_parsed *rinfo;
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800594 loff_t chunk_offset;
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800595 if (new_pos == 0)
596 return true;
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800597 if (is_hash_order(new_pos)) {
598 /* no need to reset last_name for a forward seek when
599 * dentries are sotred in hash order */
600 } else if (fi->frag |= fpos_frag(new_pos)) {
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800601 return true;
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800602 }
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800603 rinfo = fi->last_readdir ? &fi->last_readdir->r_reply_info : NULL;
604 if (!rinfo || !rinfo->dir_nr)
605 return true;
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800606 chunk_offset = rinfo->dir_entries[0].offset;
607 return new_pos < chunk_offset ||
608 is_hash_order(new_pos) != is_hash_order(chunk_offset);
Yan, Zheng8974eeb2016-04-28 15:17:40 +0800609}
610
Andrew Morton965c8e52012-12-17 15:59:39 -0800611static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
Sage Weil2817b002009-10-06 11:31:08 -0700612{
613 struct ceph_file_info *fi = file->private_data;
614 struct inode *inode = file->f_mapping->host;
Sage Weil2817b002009-10-06 11:31:08 -0700615 loff_t retval;
616
Al Viro59551022016-01-22 15:40:57 -0500617 inode_lock(inode);
Josef Bacik06222e42011-07-18 13:21:38 -0400618 retval = -EINVAL;
Andrew Morton965c8e52012-12-17 15:59:39 -0800619 switch (whence) {
Sage Weil2817b002009-10-06 11:31:08 -0700620 case SEEK_CUR:
621 offset += file->f_pos;
Josef Bacik06222e42011-07-18 13:21:38 -0400622 case SEEK_SET:
623 break;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800624 case SEEK_END:
625 retval = -EOPNOTSUPP;
Josef Bacik06222e42011-07-18 13:21:38 -0400626 default:
627 goto out;
Sage Weil2817b002009-10-06 11:31:08 -0700628 }
Josef Bacik06222e42011-07-18 13:21:38 -0400629
Yan, Zhengf0494202014-02-27 16:26:24 +0800630 if (offset >= 0) {
Yan, Zhengf3c4ebe2016-04-29 11:27:30 +0800631 if (need_reset_readdir(fi, offset)) {
632 dout("dir_llseek dropping %p content\n", file);
633 reset_readdir(fi);
634 } else if (is_hash_order(offset) && offset > file->f_pos) {
635 /* for hash offset, we don't know if a forward seek
636 * is within same frag */
637 fi->dir_release_count = 0;
638 fi->readdir_cache_idx = -1;
639 }
640
Sage Weil2817b002009-10-06 11:31:08 -0700641 if (offset != file->f_pos) {
642 file->f_pos = offset;
643 file->f_version = 0;
Sage Weil9cfa1092011-07-26 11:26:18 -0700644 fi->flags &= ~CEPH_F_ATEND;
Sage Weil2817b002009-10-06 11:31:08 -0700645 }
646 retval = offset;
Sage Weil2817b002009-10-06 11:31:08 -0700647 }
Josef Bacik06222e42011-07-18 13:21:38 -0400648out:
Al Viro59551022016-01-22 15:40:57 -0500649 inode_unlock(inode);
Sage Weil2817b002009-10-06 11:31:08 -0700650 return retval;
651}
652
653/*
Sage Weil468640e2011-07-26 11:28:11 -0700654 * Handle lookups for the hidden .snap directory.
Sage Weil2817b002009-10-06 11:31:08 -0700655 */
Sage Weil468640e2011-07-26 11:28:11 -0700656int ceph_handle_snapdir(struct ceph_mds_request *req,
657 struct dentry *dentry, int err)
Sage Weil2817b002009-10-06 11:31:08 -0700658{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700659 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
David Howells2b0143b2015-03-17 22:25:59 +0000660 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
Sage Weil2817b002009-10-06 11:31:08 -0700661
662 /* .snap dir? */
663 if (err == -ENOENT &&
Sage Weil455cec02011-03-03 13:44:35 -0800664 ceph_snap(parent) == CEPH_NOSNAP &&
Sage Weil6b805182009-10-27 11:50:50 -0700665 strcmp(dentry->d_name.name,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700666 fsc->mount_options->snapdir_name) == 0) {
Sage Weil2817b002009-10-06 11:31:08 -0700667 struct inode *inode = ceph_get_snapdir(parent);
Al Viroa4555892014-10-21 20:11:25 -0400668 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
669 dentry, dentry, inode);
Sage Weil9358c6d2010-03-30 13:54:41 -0700670 BUG_ON(!d_unhashed(dentry));
Sage Weil2817b002009-10-06 11:31:08 -0700671 d_add(dentry, inode);
672 err = 0;
673 }
Sage Weil468640e2011-07-26 11:28:11 -0700674 return err;
675}
Sage Weil2817b002009-10-06 11:31:08 -0700676
Sage Weil468640e2011-07-26 11:28:11 -0700677/*
678 * Figure out final result of a lookup/open request.
679 *
680 * Mainly, make sure we return the final req->r_dentry (if it already
681 * existed) in place of the original VFS-provided dentry when they
682 * differ.
683 *
684 * Gracefully handle the case where the MDS replies with -ENOENT and
685 * no trace (which it may do, at its discretion, e.g., if it doesn't
686 * care to issue a lease on the negative dentry).
687 */
688struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
689 struct dentry *dentry, int err)
690{
Sage Weil2817b002009-10-06 11:31:08 -0700691 if (err == -ENOENT) {
692 /* no trace? */
693 err = 0;
694 if (!req->r_reply_info.head->is_dentry) {
695 dout("ENOENT and no trace, dentry %p inode %p\n",
David Howells2b0143b2015-03-17 22:25:59 +0000696 dentry, d_inode(dentry));
697 if (d_really_is_positive(dentry)) {
Sage Weil2817b002009-10-06 11:31:08 -0700698 d_drop(dentry);
699 err = -ENOENT;
700 } else {
701 d_add(dentry, NULL);
702 }
703 }
704 }
705 if (err)
706 dentry = ERR_PTR(err);
707 else if (dentry != req->r_dentry)
708 dentry = dget(req->r_dentry); /* we got spliced */
709 else
710 dentry = NULL;
711 return dentry;
712}
713
Zhang Zhuoyu3b33f692016-03-25 05:18:39 -0400714static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
Sage Weil1d1de9162009-12-02 11:54:25 -0800715{
716 return ceph_ino(inode) == CEPH_INO_ROOT &&
717 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
718}
719
Sage Weil2817b002009-10-06 11:31:08 -0700720/*
721 * Look up a single dir entry. If there is a lookup intent, inform
722 * the MDS so that it gets our 'caps wanted' value in a single op.
723 */
724static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
Al Viro00cd8dd2012-06-10 17:13:09 -0400725 unsigned int flags)
Sage Weil2817b002009-10-06 11:31:08 -0700726{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700727 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
728 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700729 struct ceph_mds_request *req;
730 int op;
Yan, Zheng315f2402016-03-07 10:34:50 +0800731 int mask;
Sage Weil2817b002009-10-06 11:31:08 -0700732 int err;
733
Al Viroa4555892014-10-21 20:11:25 -0400734 dout("lookup %p dentry %p '%pd'\n",
735 dir, dentry, dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700736
737 if (dentry->d_name.len > NAME_MAX)
738 return ERR_PTR(-ENAMETOOLONG);
739
740 err = ceph_init_dentry(dentry);
741 if (err < 0)
742 return ERR_PTR(err);
743
Sage Weil2817b002009-10-06 11:31:08 -0700744 /* can we conclude ENOENT locally? */
David Howells2b0143b2015-03-17 22:25:59 +0000745 if (d_really_is_negative(dentry)) {
Sage Weil2817b002009-10-06 11:31:08 -0700746 struct ceph_inode_info *ci = ceph_inode(dir);
747 struct ceph_dentry_info *di = ceph_dentry(dentry);
748
Sage Weilbe655592011-11-30 09:47:09 -0800749 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700750 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
751 if (strncmp(dentry->d_name.name,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700752 fsc->mount_options->snapdir_name,
Sage Weil2817b002009-10-06 11:31:08 -0700753 dentry->d_name.len) &&
Sage Weil1d1de9162009-12-02 11:54:25 -0800754 !is_root_ceph_dentry(dir, dentry) &&
Yan, Zhenge2c3de02015-03-04 16:05:04 +0800755 ceph_test_mount_opt(fsc, DCACHE) &&
Yan, Zheng2f276c52013-03-13 19:44:32 +0800756 __ceph_dir_is_complete(ci) &&
Sage Weil2817b002009-10-06 11:31:08 -0700757 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
Sage Weilbe655592011-11-30 09:47:09 -0800758 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700759 dout(" dir %p complete, -ENOENT\n", dir);
760 d_add(dentry, NULL);
761 di->lease_shared_gen = ci->i_shared_gen;
762 return NULL;
763 }
Sage Weilbe655592011-11-30 09:47:09 -0800764 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -0700765 }
766
767 op = ceph_snap(dir) == CEPH_SNAPDIR ?
768 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
769 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
770 if (IS_ERR(req))
Julia Lawall7e34bc52010-05-22 12:01:14 +0200771 return ERR_CAST(req);
Sage Weil2817b002009-10-06 11:31:08 -0700772 req->r_dentry = dget(dentry);
773 req->r_num_caps = 2;
Yan, Zheng315f2402016-03-07 10:34:50 +0800774
775 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
776 if (ceph_security_xattr_wanted(dir))
777 mask |= CEPH_CAP_XATTR_SHARED;
778 req->r_args.getattr.mask = cpu_to_le32(mask);
779
Sage Weil2817b002009-10-06 11:31:08 -0700780 req->r_locked_dir = dir;
781 err = ceph_mdsc_do_request(mdsc, NULL, req);
Sage Weil468640e2011-07-26 11:28:11 -0700782 err = ceph_handle_snapdir(req, dentry, err);
Sage Weil2817b002009-10-06 11:31:08 -0700783 dentry = ceph_finish_lookup(req, dentry, err);
784 ceph_mdsc_put_request(req); /* will dput(dentry) */
785 dout("lookup result=%p\n", dentry);
786 return dentry;
787}
788
789/*
790 * If we do a create but get no trace back from the MDS, follow up with
791 * a lookup (the VFS expects us to link up the provided dentry).
792 */
793int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
794{
Al Viro00cd8dd2012-06-10 17:13:09 -0400795 struct dentry *result = ceph_lookup(dir, dentry, 0);
Sage Weil2817b002009-10-06 11:31:08 -0700796
797 if (result && !IS_ERR(result)) {
798 /*
799 * We created the item, then did a lookup, and found
800 * it was already linked to another inode we already
Yan, Zheng4d41cef2015-02-04 15:10:48 +0800801 * had in our cache (and thus got spliced). To not
802 * confuse VFS (especially when inode is a directory),
803 * we don't link our dentry to that inode, return an
804 * error instead.
805 *
806 * This event should be rare and it happens only when
807 * we talk to old MDS. Recent MDS does not send traceless
808 * reply for request that creates new inode.
Sage Weil2817b002009-10-06 11:31:08 -0700809 */
Yan, Zheng5cba3722015-02-02 11:27:56 +0800810 d_drop(result);
Yan, Zheng4d41cef2015-02-04 15:10:48 +0800811 return -ESTALE;
Sage Weil2817b002009-10-06 11:31:08 -0700812 }
813 return PTR_ERR(result);
814}
815
816static int ceph_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -0400817 umode_t mode, dev_t rdev)
Sage Weil2817b002009-10-06 11:31:08 -0700818{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700819 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
820 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700821 struct ceph_mds_request *req;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800822 struct ceph_acls_info acls = {};
Sage Weil2817b002009-10-06 11:31:08 -0700823 int err;
824
825 if (ceph_snap(dir) != CEPH_NOSNAP)
826 return -EROFS;
827
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800828 err = ceph_pre_init_acls(dir, &mode, &acls);
829 if (err < 0)
830 return err;
831
Al Viro1a67aaf2011-07-26 01:52:52 -0400832 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
Sage Weil2817b002009-10-06 11:31:08 -0700833 dir, dentry, mode, rdev);
834 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
835 if (IS_ERR(req)) {
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800836 err = PTR_ERR(req);
837 goto out;
Sage Weil2817b002009-10-06 11:31:08 -0700838 }
839 req->r_dentry = dget(dentry);
840 req->r_num_caps = 2;
841 req->r_locked_dir = dir;
842 req->r_args.mknod.mode = cpu_to_le32(mode);
843 req->r_args.mknod.rdev = cpu_to_le32(rdev);
844 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
845 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800846 if (acls.pagelist) {
847 req->r_pagelist = acls.pagelist;
848 acls.pagelist = NULL;
849 }
Sage Weil2817b002009-10-06 11:31:08 -0700850 err = ceph_mdsc_do_request(mdsc, dir, req);
851 if (!err && !req->r_reply_info.head->is_dentry)
852 err = ceph_handle_notrace_create(dir, dentry);
853 ceph_mdsc_put_request(req);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800854out:
Guangliang Zhao7221fe42013-11-11 15:18:03 +0800855 if (!err)
David Howells2b0143b2015-03-17 22:25:59 +0000856 ceph_init_inode_acls(d_inode(dentry), &acls);
Yan, Zhengb20a95a2014-02-11 12:55:05 +0800857 else
Sage Weil2817b002009-10-06 11:31:08 -0700858 d_drop(dentry);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800859 ceph_release_acls_info(&acls);
Sage Weil2817b002009-10-06 11:31:08 -0700860 return err;
861}
862
Al Viro4acdaf22011-07-26 01:42:34 -0400863static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -0400864 bool excl)
Sage Weil2817b002009-10-06 11:31:08 -0700865{
Miklos Szeredi2d83bde2012-06-05 15:10:25 +0200866 return ceph_mknod(dir, dentry, mode, 0);
Sage Weil2817b002009-10-06 11:31:08 -0700867}
868
869static int ceph_symlink(struct inode *dir, struct dentry *dentry,
870 const char *dest)
871{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700872 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
873 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700874 struct ceph_mds_request *req;
875 int err;
876
877 if (ceph_snap(dir) != CEPH_NOSNAP)
878 return -EROFS;
879
880 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
881 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
882 if (IS_ERR(req)) {
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800883 err = PTR_ERR(req);
884 goto out;
Sage Weil2817b002009-10-06 11:31:08 -0700885 }
Yan, Zheng687265e2015-06-13 17:27:05 +0800886 req->r_path2 = kstrdup(dest, GFP_KERNEL);
Sanidhya Kashyapa149bb92015-03-21 12:54:58 -0400887 if (!req->r_path2) {
888 err = -ENOMEM;
889 ceph_mdsc_put_request(req);
890 goto out;
891 }
892 req->r_locked_dir = dir;
Sage Weil2817b002009-10-06 11:31:08 -0700893 req->r_dentry = dget(dentry);
894 req->r_num_caps = 2;
Sage Weil2817b002009-10-06 11:31:08 -0700895 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
896 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
897 err = ceph_mdsc_do_request(mdsc, dir, req);
898 if (!err && !req->r_reply_info.head->is_dentry)
899 err = ceph_handle_notrace_create(dir, dentry);
900 ceph_mdsc_put_request(req);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800901out:
902 if (err)
Sage Weil2817b002009-10-06 11:31:08 -0700903 d_drop(dentry);
904 return err;
905}
906
Al Viro18bb1db2011-07-26 01:41:39 -0400907static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Sage Weil2817b002009-10-06 11:31:08 -0700908{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700909 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
910 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700911 struct ceph_mds_request *req;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800912 struct ceph_acls_info acls = {};
Sage Weil2817b002009-10-06 11:31:08 -0700913 int err = -EROFS;
914 int op;
915
916 if (ceph_snap(dir) == CEPH_SNAPDIR) {
917 /* mkdir .snap/foo is a MKSNAP */
918 op = CEPH_MDS_OP_MKSNAP;
Al Viroa4555892014-10-21 20:11:25 -0400919 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
920 dentry, dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700921 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
Al Viro18bb1db2011-07-26 01:41:39 -0400922 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
Sage Weil2817b002009-10-06 11:31:08 -0700923 op = CEPH_MDS_OP_MKDIR;
924 } else {
925 goto out;
926 }
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800927
928 mode |= S_IFDIR;
929 err = ceph_pre_init_acls(dir, &mode, &acls);
930 if (err < 0)
931 goto out;
932
Sage Weil2817b002009-10-06 11:31:08 -0700933 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
934 if (IS_ERR(req)) {
935 err = PTR_ERR(req);
936 goto out;
937 }
938
939 req->r_dentry = dget(dentry);
940 req->r_num_caps = 2;
941 req->r_locked_dir = dir;
942 req->r_args.mkdir.mode = cpu_to_le32(mode);
943 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
944 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800945 if (acls.pagelist) {
946 req->r_pagelist = acls.pagelist;
947 acls.pagelist = NULL;
948 }
Sage Weil2817b002009-10-06 11:31:08 -0700949 err = ceph_mdsc_do_request(mdsc, dir, req);
Yan, Zheng275dd192014-12-10 16:17:31 +0800950 if (!err &&
951 !req->r_reply_info.head->is_target &&
952 !req->r_reply_info.head->is_dentry)
Sage Weil2817b002009-10-06 11:31:08 -0700953 err = ceph_handle_notrace_create(dir, dentry);
954 ceph_mdsc_put_request(req);
955out:
Yan, Zhengb20a95a2014-02-11 12:55:05 +0800956 if (!err)
David Howells2b0143b2015-03-17 22:25:59 +0000957 ceph_init_inode_acls(d_inode(dentry), &acls);
Yan, Zhengb20a95a2014-02-11 12:55:05 +0800958 else
Sage Weil2817b002009-10-06 11:31:08 -0700959 d_drop(dentry);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800960 ceph_release_acls_info(&acls);
Sage Weil2817b002009-10-06 11:31:08 -0700961 return err;
962}
963
964static int ceph_link(struct dentry *old_dentry, struct inode *dir,
965 struct dentry *dentry)
966{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700967 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
968 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -0700969 struct ceph_mds_request *req;
970 int err;
971
972 if (ceph_snap(dir) != CEPH_NOSNAP)
973 return -EROFS;
974
975 dout("link in dir %p old_dentry %p dentry %p\n", dir,
976 old_dentry, dentry);
977 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
978 if (IS_ERR(req)) {
979 d_drop(dentry);
980 return PTR_ERR(req);
981 }
982 req->r_dentry = dget(dentry);
983 req->r_num_caps = 2;
Sage Weil4b58c9b2013-02-05 13:41:23 -0800984 req->r_old_dentry = dget(old_dentry);
Sage Weil2817b002009-10-06 11:31:08 -0700985 req->r_locked_dir = dir;
986 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
987 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengad88f232013-07-21 20:25:25 +0800988 /* release LINK_SHARED on source inode (mds will lock it) */
989 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
Sage Weil2817b002009-10-06 11:31:08 -0700990 err = ceph_mdsc_do_request(mdsc, dir, req);
Sage Weil70b666c2011-05-27 09:24:26 -0700991 if (err) {
Sage Weil2817b002009-10-06 11:31:08 -0700992 d_drop(dentry);
Sage Weil70b666c2011-05-27 09:24:26 -0700993 } else if (!req->r_reply_info.head->is_dentry) {
David Howells2b0143b2015-03-17 22:25:59 +0000994 ihold(d_inode(old_dentry));
995 d_instantiate(dentry, d_inode(old_dentry));
Sage Weil70b666c2011-05-27 09:24:26 -0700996 }
Sage Weil2817b002009-10-06 11:31:08 -0700997 ceph_mdsc_put_request(req);
998 return err;
999}
1000
1001/*
1002 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
1003 * looks like the link count will hit 0, drop any other caps (other
1004 * than PIN) we don't specifically want (due to the file still being
1005 * open).
1006 */
1007static int drop_caps_for_unlink(struct inode *inode)
1008{
1009 struct ceph_inode_info *ci = ceph_inode(inode);
1010 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1011
Sage Weilbe655592011-11-30 09:47:09 -08001012 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001013 if (inode->i_nlink == 1) {
1014 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
1015 ci->i_ceph_flags |= CEPH_I_NODELAY;
1016 }
Sage Weilbe655592011-11-30 09:47:09 -08001017 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001018 return drop;
1019}
1020
1021/*
1022 * rmdir and unlink are differ only by the metadata op code
1023 */
1024static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1025{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001026 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1027 struct ceph_mds_client *mdsc = fsc->mdsc;
David Howells2b0143b2015-03-17 22:25:59 +00001028 struct inode *inode = d_inode(dentry);
Sage Weil2817b002009-10-06 11:31:08 -07001029 struct ceph_mds_request *req;
1030 int err = -EROFS;
1031 int op;
1032
1033 if (ceph_snap(dir) == CEPH_SNAPDIR) {
1034 /* rmdir .snap/foo is RMSNAP */
Al Viroa4555892014-10-21 20:11:25 -04001035 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
Sage Weil2817b002009-10-06 11:31:08 -07001036 op = CEPH_MDS_OP_RMSNAP;
1037 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1038 dout("unlink/rmdir dir %p dn %p inode %p\n",
1039 dir, dentry, inode);
David Howellse36cb0b2015-01-29 12:02:35 +00001040 op = d_is_dir(dentry) ?
Sage Weil2817b002009-10-06 11:31:08 -07001041 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1042 } else
1043 goto out;
1044 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1045 if (IS_ERR(req)) {
1046 err = PTR_ERR(req);
1047 goto out;
1048 }
1049 req->r_dentry = dget(dentry);
1050 req->r_num_caps = 2;
1051 req->r_locked_dir = dir;
1052 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1053 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1054 req->r_inode_drop = drop_caps_for_unlink(inode);
1055 err = ceph_mdsc_do_request(mdsc, dir, req);
1056 if (!err && !req->r_reply_info.head->is_dentry)
1057 d_delete(dentry);
1058 ceph_mdsc_put_request(req);
1059out:
1060 return err;
1061}
1062
1063static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
1064 struct inode *new_dir, struct dentry *new_dentry)
1065{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001066 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
1067 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil2817b002009-10-06 11:31:08 -07001068 struct ceph_mds_request *req;
Yan, Zheng0ea611a2015-04-07 15:36:32 +08001069 int op = CEPH_MDS_OP_RENAME;
Sage Weil2817b002009-10-06 11:31:08 -07001070 int err;
1071
1072 if (ceph_snap(old_dir) != ceph_snap(new_dir))
1073 return -EXDEV;
Yan, Zheng0ea611a2015-04-07 15:36:32 +08001074 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1075 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1076 op = CEPH_MDS_OP_RENAMESNAP;
1077 else
1078 return -EROFS;
1079 }
Sage Weil2817b002009-10-06 11:31:08 -07001080 dout("rename dir %p dentry %p to dir %p dentry %p\n",
1081 old_dir, old_dentry, new_dir, new_dentry);
Yan, Zheng0ea611a2015-04-07 15:36:32 +08001082 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
Sage Weil2817b002009-10-06 11:31:08 -07001083 if (IS_ERR(req))
1084 return PTR_ERR(req);
Sage Weil180061a2013-02-05 13:36:05 -08001085 ihold(old_dir);
Sage Weil2817b002009-10-06 11:31:08 -07001086 req->r_dentry = dget(new_dentry);
1087 req->r_num_caps = 2;
1088 req->r_old_dentry = dget(old_dentry);
Sage Weil180061a2013-02-05 13:36:05 -08001089 req->r_old_dentry_dir = old_dir;
Sage Weil2817b002009-10-06 11:31:08 -07001090 req->r_locked_dir = new_dir;
1091 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1092 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1093 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1094 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1095 /* release LINK_RDCACHE on source inode (mds will lock it) */
1096 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
David Howells2b0143b2015-03-17 22:25:59 +00001097 if (d_really_is_positive(new_dentry))
1098 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
Sage Weil2817b002009-10-06 11:31:08 -07001099 err = ceph_mdsc_do_request(mdsc, old_dir, req);
1100 if (!err && !req->r_reply_info.head->is_dentry) {
1101 /*
1102 * Normally d_move() is done by fill_trace (called by
1103 * do_request, above). If there is no trace, we need
1104 * to do it here.
1105 */
Sage Weilea1409f2010-04-28 16:12:06 -07001106
Yan, Zhengfdd4e152015-06-16 20:48:56 +08001107 /* d_move screws up sibling dentries' offsets */
1108 ceph_dir_clear_complete(old_dir);
1109 ceph_dir_clear_complete(new_dir);
1110
Sage Weil2817b002009-10-06 11:31:08 -07001111 d_move(old_dentry, new_dentry);
Sage Weilea1409f2010-04-28 16:12:06 -07001112
1113 /* ensure target dentry is invalidated, despite
1114 rehashing bug in vfs_rename_dir */
Sage Weil81a6cf22010-05-14 09:35:38 -07001115 ceph_invalidate_dentry_lease(new_dentry);
Sage Weil2817b002009-10-06 11:31:08 -07001116 }
1117 ceph_mdsc_put_request(req);
1118 return err;
1119}
1120
Sage Weil81a6cf22010-05-14 09:35:38 -07001121/*
1122 * Ensure a dentry lease will no longer revalidate.
1123 */
1124void ceph_invalidate_dentry_lease(struct dentry *dentry)
1125{
1126 spin_lock(&dentry->d_lock);
1127 dentry->d_time = jiffies;
1128 ceph_dentry(dentry)->lease_shared_gen = 0;
1129 spin_unlock(&dentry->d_lock);
1130}
Sage Weil2817b002009-10-06 11:31:08 -07001131
1132/*
1133 * Check if dentry lease is valid. If not, delete the lease. Try to
1134 * renew if the least is more than half up.
1135 */
1136static int dentry_lease_is_valid(struct dentry *dentry)
1137{
1138 struct ceph_dentry_info *di;
1139 struct ceph_mds_session *s;
1140 int valid = 0;
1141 u32 gen;
1142 unsigned long ttl;
1143 struct ceph_mds_session *session = NULL;
1144 struct inode *dir = NULL;
1145 u32 seq = 0;
1146
1147 spin_lock(&dentry->d_lock);
1148 di = ceph_dentry(dentry);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001149 if (di->lease_session) {
Sage Weil2817b002009-10-06 11:31:08 -07001150 s = di->lease_session;
Alex Elderd8fb02a2012-01-12 17:48:10 -08001151 spin_lock(&s->s_gen_ttl_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001152 gen = s->s_cap_gen;
1153 ttl = s->s_cap_ttl;
Alex Elderd8fb02a2012-01-12 17:48:10 -08001154 spin_unlock(&s->s_gen_ttl_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001155
1156 if (di->lease_gen == gen &&
1157 time_before(jiffies, dentry->d_time) &&
1158 time_before(jiffies, ttl)) {
1159 valid = 1;
1160 if (di->lease_renew_after &&
1161 time_after(jiffies, di->lease_renew_after)) {
1162 /* we should renew */
David Howells2b0143b2015-03-17 22:25:59 +00001163 dir = d_inode(dentry->d_parent);
Sage Weil2817b002009-10-06 11:31:08 -07001164 session = ceph_get_mds_session(s);
1165 seq = di->lease_seq;
1166 di->lease_renew_after = 0;
1167 di->lease_renew_from = jiffies;
1168 }
Sage Weil2817b002009-10-06 11:31:08 -07001169 }
1170 }
1171 spin_unlock(&dentry->d_lock);
1172
1173 if (session) {
1174 ceph_mdsc_lease_send_msg(session, dir, dentry,
1175 CEPH_MDS_LEASE_RENEW, seq);
1176 ceph_put_mds_session(session);
1177 }
1178 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1179 return valid;
1180}
1181
1182/*
1183 * Check if directory-wide content lease/cap is valid.
1184 */
1185static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1186{
1187 struct ceph_inode_info *ci = ceph_inode(dir);
1188 struct ceph_dentry_info *di = ceph_dentry(dentry);
1189 int valid = 0;
1190
Sage Weilbe655592011-11-30 09:47:09 -08001191 spin_lock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001192 if (ci->i_shared_gen == di->lease_shared_gen)
1193 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
Sage Weilbe655592011-11-30 09:47:09 -08001194 spin_unlock(&ci->i_ceph_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001195 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1196 dir, (unsigned)ci->i_shared_gen, dentry,
1197 (unsigned)di->lease_shared_gen, valid);
1198 return valid;
1199}
1200
1201/*
1202 * Check if cached dentry can be trusted.
1203 */
Al Viro0b728e12012-06-10 16:03:43 -04001204static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
Sage Weil2817b002009-10-06 11:31:08 -07001205{
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001206 int valid = 0;
Yan, Zheng641235d2016-03-16 16:40:23 +08001207 struct dentry *parent;
Nick Piggin34286d62011-01-07 17:49:57 +11001208 struct inode *dir;
1209
Al Viro0b728e12012-06-10 16:03:43 -04001210 if (flags & LOOKUP_RCU)
Nick Piggin34286d62011-01-07 17:49:57 +11001211 return -ECHILD;
1212
Al Viroa4555892014-10-21 20:11:25 -04001213 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
David Howells2b0143b2015-03-17 22:25:59 +00001214 dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
Sage Weil2817b002009-10-06 11:31:08 -07001215
Yan, Zheng641235d2016-03-16 16:40:23 +08001216 parent = dget_parent(dentry);
1217 dir = d_inode(parent);
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001218
Sage Weil2817b002009-10-06 11:31:08 -07001219 /* always trust cached snapped dentries, snapdir dentry */
1220 if (ceph_snap(dir) != CEPH_NOSNAP) {
Al Viroa4555892014-10-21 20:11:25 -04001221 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
David Howells2b0143b2015-03-17 22:25:59 +00001222 dentry, d_inode(dentry));
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001223 valid = 1;
David Howells2b0143b2015-03-17 22:25:59 +00001224 } else if (d_really_is_positive(dentry) &&
1225 ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001226 valid = 1;
1227 } else if (dentry_lease_is_valid(dentry) ||
1228 dir_lease_is_valid(dir, dentry)) {
David Howells2b0143b2015-03-17 22:25:59 +00001229 if (d_really_is_positive(dentry))
1230 valid = ceph_is_any_caps(d_inode(dentry));
Yan, Zheng9215aee2013-11-30 12:47:41 +08001231 else
1232 valid = 1;
Sage Weil2817b002009-10-06 11:31:08 -07001233 }
Sage Weil2817b002009-10-06 11:31:08 -07001234
Yan, Zheng200fd272016-03-17 14:41:59 +08001235 if (!valid) {
1236 struct ceph_mds_client *mdsc =
1237 ceph_sb_to_client(dir->i_sb)->mdsc;
1238 struct ceph_mds_request *req;
1239 int op, mask, err;
1240
1241 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1242 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1243 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1244 if (!IS_ERR(req)) {
1245 req->r_dentry = dget(dentry);
1246 req->r_num_caps = 2;
1247
1248 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1249 if (ceph_security_xattr_wanted(dir))
1250 mask |= CEPH_CAP_XATTR_SHARED;
1251 req->r_args.getattr.mask = mask;
1252
1253 req->r_locked_dir = dir;
1254 err = ceph_mdsc_do_request(mdsc, NULL, req);
1255 if (err == 0 || err == -ENOENT) {
1256 if (dentry == req->r_dentry) {
1257 valid = !d_unhashed(dentry);
1258 } else {
1259 d_invalidate(req->r_dentry);
1260 err = -EAGAIN;
1261 }
1262 }
1263 ceph_mdsc_put_request(req);
1264 dout("d_revalidate %p lookup result=%d\n",
1265 dentry, err);
1266 }
1267 }
1268
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001269 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
Yan, Zheng9215aee2013-11-30 12:47:41 +08001270 if (valid) {
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001271 ceph_dentry_lru_touch(dentry);
Yan, Zheng9215aee2013-11-30 12:47:41 +08001272 } else {
1273 ceph_dir_clear_complete(dir);
Yan, Zheng9215aee2013-11-30 12:47:41 +08001274 }
Yan, Zheng641235d2016-03-16 16:40:23 +08001275
1276 dput(parent);
Sage Weilbf1c6ac2011-07-26 11:30:43 -07001277 return valid;
Sage Weil2817b002009-10-06 11:31:08 -07001278}
1279
1280/*
Sage Weil147851d2011-03-15 14:57:41 -07001281 * Release our ceph_dentry_info.
Sage Weil2817b002009-10-06 11:31:08 -07001282 */
Sage Weil147851d2011-03-15 14:57:41 -07001283static void ceph_d_release(struct dentry *dentry)
Sage Weil2817b002009-10-06 11:31:08 -07001284{
1285 struct ceph_dentry_info *di = ceph_dentry(dentry);
Sage Weil2817b002009-10-06 11:31:08 -07001286
Sage Weil147851d2011-03-15 14:57:41 -07001287 dout("d_release %p\n", dentry);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001288 ceph_dentry_lru_del(dentry);
1289 if (di->lease_session)
1290 ceph_put_mds_session(di->lease_session);
1291 kmem_cache_free(ceph_dentry_cachep, di);
1292 dentry->d_fsdata = NULL;
Sage Weil2817b002009-10-06 11:31:08 -07001293}
1294
1295static int ceph_snapdir_d_revalidate(struct dentry *dentry,
Al Viro0b728e12012-06-10 16:03:43 -04001296 unsigned int flags)
Sage Weil2817b002009-10-06 11:31:08 -07001297{
1298 /*
1299 * Eventually, we'll want to revalidate snapped metadata
1300 * too... probably...
1301 */
1302 return 1;
1303}
1304
Sage Weilb58dc412011-03-15 15:53:40 -07001305/*
1306 * When the VFS prunes a dentry from the cache, we need to clear the
1307 * complete flag on the parent directory.
1308 *
1309 * Called under dentry->d_lock.
1310 */
1311static void ceph_d_prune(struct dentry *dentry)
1312{
Sage Weil774ac212011-11-11 09:48:08 -08001313 dout("ceph_d_prune %p\n", dentry);
Sage Weilb58dc412011-03-15 15:53:40 -07001314
1315 /* do we have a valid parent? */
Sage Weil8842b3b2012-06-07 13:43:35 -07001316 if (IS_ROOT(dentry))
Sage Weilb58dc412011-03-15 15:53:40 -07001317 return;
1318
Yan, Zheng2f276c52013-03-13 19:44:32 +08001319 /* if we are not hashed, we don't affect dir's completeness */
Sage Weilb58dc412011-03-15 15:53:40 -07001320 if (d_unhashed(dentry))
1321 return;
1322
1323 /*
1324 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1325 * cleared until d_release
1326 */
David Howells2b0143b2015-03-17 22:25:59 +00001327 ceph_dir_clear_complete(d_inode(dentry->d_parent));
Sage Weilb58dc412011-03-15 15:53:40 -07001328}
Sage Weil2817b002009-10-06 11:31:08 -07001329
1330/*
1331 * read() on a dir. This weird interface hack only works if mounted
1332 * with '-o dirstat'.
1333 */
1334static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1335 loff_t *ppos)
1336{
1337 struct ceph_file_info *cf = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -05001338 struct inode *inode = file_inode(file);
Sage Weil2817b002009-10-06 11:31:08 -07001339 struct ceph_inode_info *ci = ceph_inode(inode);
1340 int left;
Sage Weilae598082011-05-12 14:28:05 -07001341 const int bufsize = 1024;
Sage Weil2817b002009-10-06 11:31:08 -07001342
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001343 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
Sage Weil2817b002009-10-06 11:31:08 -07001344 return -EISDIR;
1345
1346 if (!cf->dir_info) {
Yan, Zheng687265e2015-06-13 17:27:05 +08001347 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
Sage Weil2817b002009-10-06 11:31:08 -07001348 if (!cf->dir_info)
1349 return -ENOMEM;
1350 cf->dir_info_len =
Sage Weilae598082011-05-12 14:28:05 -07001351 snprintf(cf->dir_info, bufsize,
Sage Weil2817b002009-10-06 11:31:08 -07001352 "entries: %20lld\n"
1353 " files: %20lld\n"
1354 " subdirs: %20lld\n"
1355 "rentries: %20lld\n"
1356 " rfiles: %20lld\n"
1357 " rsubdirs: %20lld\n"
1358 "rbytes: %20lld\n"
1359 "rctime: %10ld.%09ld\n",
1360 ci->i_files + ci->i_subdirs,
1361 ci->i_files,
1362 ci->i_subdirs,
1363 ci->i_rfiles + ci->i_rsubdirs,
1364 ci->i_rfiles,
1365 ci->i_rsubdirs,
1366 ci->i_rbytes,
1367 (long)ci->i_rctime.tv_sec,
1368 (long)ci->i_rctime.tv_nsec);
1369 }
1370
1371 if (*ppos >= cf->dir_info_len)
1372 return 0;
1373 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1374 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1375 if (left == size)
1376 return -EFAULT;
1377 *ppos += (size - left);
1378 return size - left;
1379}
1380
1381/*
Sage Weil2817b002009-10-06 11:31:08 -07001382 * We maintain a private dentry LRU.
1383 *
1384 * FIXME: this needs to be changed to a per-mds lru to be useful.
1385 */
1386void ceph_dentry_lru_add(struct dentry *dn)
1387{
1388 struct ceph_dentry_info *di = ceph_dentry(dn);
1389 struct ceph_mds_client *mdsc;
Sage Weil2817b002009-10-06 11:31:08 -07001390
Al Viroa4555892014-10-21 20:11:25 -04001391 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001392 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1393 spin_lock(&mdsc->dentry_lru_lock);
1394 list_add_tail(&di->lru, &mdsc->dentry_lru);
1395 mdsc->num_dentry++;
1396 spin_unlock(&mdsc->dentry_lru_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001397}
1398
1399void ceph_dentry_lru_touch(struct dentry *dn)
1400{
1401 struct ceph_dentry_info *di = ceph_dentry(dn);
1402 struct ceph_mds_client *mdsc;
Sage Weil2817b002009-10-06 11:31:08 -07001403
Al Viroa4555892014-10-21 20:11:25 -04001404 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1405 di->offset);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001406 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1407 spin_lock(&mdsc->dentry_lru_lock);
1408 list_move_tail(&di->lru, &mdsc->dentry_lru);
1409 spin_unlock(&mdsc->dentry_lru_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001410}
1411
1412void ceph_dentry_lru_del(struct dentry *dn)
1413{
1414 struct ceph_dentry_info *di = ceph_dentry(dn);
1415 struct ceph_mds_client *mdsc;
1416
Al Viroa4555892014-10-21 20:11:25 -04001417 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
Sage Weil3d8eb7a2011-11-11 09:48:53 -08001418 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1419 spin_lock(&mdsc->dentry_lru_lock);
1420 list_del_init(&di->lru);
1421 mdsc->num_dentry--;
1422 spin_unlock(&mdsc->dentry_lru_lock);
Sage Weil2817b002009-10-06 11:31:08 -07001423}
1424
Sage Weil6c0f3af2010-11-16 11:14:34 -08001425/*
1426 * Return name hash for a given dentry. This is dependent on
1427 * the parent directory's hash function.
1428 */
Sage Weile5f86dc2011-07-26 11:30:55 -07001429unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
Sage Weil6c0f3af2010-11-16 11:14:34 -08001430{
Sage Weil6c0f3af2010-11-16 11:14:34 -08001431 struct ceph_inode_info *dci = ceph_inode(dir);
1432
1433 switch (dci->i_dir_layout.dl_dir_hash) {
1434 case 0: /* for backward compat */
1435 case CEPH_STR_HASH_LINUX:
1436 return dn->d_name.hash;
1437
1438 default:
1439 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1440 dn->d_name.name, dn->d_name.len);
1441 }
1442}
1443
Sage Weil2817b002009-10-06 11:31:08 -07001444const struct file_operations ceph_dir_fops = {
1445 .read = ceph_read_dir,
Al Viro77acfa22013-05-17 16:52:26 -04001446 .iterate = ceph_readdir,
Sage Weil2817b002009-10-06 11:31:08 -07001447 .llseek = ceph_dir_llseek,
1448 .open = ceph_open,
1449 .release = ceph_release,
1450 .unlocked_ioctl = ceph_ioctl,
Yan, Zhengda819c82015-05-27 11:19:34 +08001451 .fsync = ceph_fsync,
Sage Weil2817b002009-10-06 11:31:08 -07001452};
1453
Yan, Zheng38c48b52015-01-14 13:46:04 +08001454const struct file_operations ceph_snapdir_fops = {
1455 .iterate = ceph_readdir,
1456 .llseek = ceph_dir_llseek,
1457 .open = ceph_open,
1458 .release = ceph_release,
1459};
1460
Sage Weil2817b002009-10-06 11:31:08 -07001461const struct inode_operations ceph_dir_iops = {
1462 .lookup = ceph_lookup,
1463 .permission = ceph_permission,
1464 .getattr = ceph_getattr,
1465 .setattr = ceph_setattr,
Andreas Gruenbacher2cdeb1e2016-04-14 00:30:17 +02001466 .setxattr = generic_setxattr,
1467 .getxattr = generic_getxattr,
Sage Weil2817b002009-10-06 11:31:08 -07001468 .listxattr = ceph_listxattr,
Andreas Gruenbacher2cdeb1e2016-04-14 00:30:17 +02001469 .removexattr = generic_removexattr,
Guangliang Zhao7221fe42013-11-11 15:18:03 +08001470 .get_acl = ceph_get_acl,
Sage Weil72466d02014-01-29 06:22:25 -08001471 .set_acl = ceph_set_acl,
Sage Weil2817b002009-10-06 11:31:08 -07001472 .mknod = ceph_mknod,
1473 .symlink = ceph_symlink,
1474 .mkdir = ceph_mkdir,
1475 .link = ceph_link,
1476 .unlink = ceph_unlink,
1477 .rmdir = ceph_unlink,
1478 .rename = ceph_rename,
1479 .create = ceph_create,
Miklos Szeredi2d83bde2012-06-05 15:10:25 +02001480 .atomic_open = ceph_atomic_open,
Sage Weil2817b002009-10-06 11:31:08 -07001481};
1482
Yan, Zheng38c48b52015-01-14 13:46:04 +08001483const struct inode_operations ceph_snapdir_iops = {
1484 .lookup = ceph_lookup,
1485 .permission = ceph_permission,
1486 .getattr = ceph_getattr,
1487 .mkdir = ceph_mkdir,
1488 .rmdir = ceph_unlink,
Yan, Zheng0ea611a2015-04-07 15:36:32 +08001489 .rename = ceph_rename,
Yan, Zheng38c48b52015-01-14 13:46:04 +08001490};
1491
Sage Weil52dfb8a2010-08-03 10:25:30 -07001492const struct dentry_operations ceph_dentry_ops = {
Sage Weil2817b002009-10-06 11:31:08 -07001493 .d_revalidate = ceph_d_revalidate,
Sage Weil147851d2011-03-15 14:57:41 -07001494 .d_release = ceph_d_release,
Sage Weilb58dc412011-03-15 15:53:40 -07001495 .d_prune = ceph_d_prune,
Sage Weil2817b002009-10-06 11:31:08 -07001496};
1497
Sage Weil52dfb8a2010-08-03 10:25:30 -07001498const struct dentry_operations ceph_snapdir_dentry_ops = {
Sage Weil2817b002009-10-06 11:31:08 -07001499 .d_revalidate = ceph_snapdir_d_revalidate,
Sage Weil147851d2011-03-15 14:57:41 -07001500 .d_release = ceph_d_release,
Sage Weil2817b002009-10-06 11:31:08 -07001501};
1502
Sage Weil52dfb8a2010-08-03 10:25:30 -07001503const struct dentry_operations ceph_snap_dentry_ops = {
Sage Weil147851d2011-03-15 14:57:41 -07001504 .d_release = ceph_d_release,
Sage Weilb58dc412011-03-15 15:53:40 -07001505 .d_prune = ceph_d_prune,
Sage Weil2817b002009-10-06 11:31:08 -07001506};