Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/ceph/ceph_debug.h>
3
4#include <linux/module.h>
5#include <linux/fs.h>
6#include <linux/slab.h>
7#include <linux/string.h>
8#include <linux/uaccess.h>
9#include <linux/kernel.h>
10#include <linux/writeback.h>
11#include <linux/vmalloc.h>
12#include <linux/xattr.h>
13#include <linux/posix_acl.h>
14#include <linux/random.h>
15#include <linux/sort.h>
16#include <linux/iversion.h>
17#include <linux/fscrypt.h>
18
19#include "super.h"
20#include "mds_client.h"
21#include "cache.h"
22#include "crypto.h"
23#include <linux/ceph/decode.h>
24
25/*
26 * Ceph inode operations
27 *
28 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
29 * setattr, etc.), xattr helpers, and helpers for assimilating
30 * metadata returned by the MDS into our cache.
31 *
32 * Also define helpers for doing asynchronous writeback, invalidation,
33 * and truncation for the benefit of those who can't afford to block
34 * (typically because they are in the message handler path).
35 */
36
37static const struct inode_operations ceph_symlink_iops;
38static const struct inode_operations ceph_encrypted_symlink_iops;
39
40static void ceph_inode_work(struct work_struct *work);
41
42/*
43 * find or create an inode, given the ceph ino number
44 */
45static int ceph_set_ino_cb(struct inode *inode, void *data)
46{
47 struct ceph_inode_info *ci = ceph_inode(inode);
48 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
49
50 ci->i_vino = *(struct ceph_vino *)data;
51 inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
52 inode_set_iversion_raw(inode, 0);
53 percpu_counter_inc(&mdsc->metric.total_inodes);
54
55 return 0;
56}
57
58/**
59 * ceph_new_inode - allocate a new inode in advance of an expected create
60 * @dir: parent directory for new inode
61 * @dentry: dentry that may eventually point to new inode
62 * @mode: mode of new inode
63 * @as_ctx: pointer to inherited security context
64 *
65 * Allocate a new inode in advance of an operation to create a new inode.
66 * This allocates the inode and sets up the acl_sec_ctx with appropriate
67 * info for the new inode.
68 *
69 * Returns a pointer to the new inode or an ERR_PTR.
70 */
71struct inode *ceph_new_inode(struct inode *dir, struct dentry *dentry,
72 umode_t *mode, struct ceph_acl_sec_ctx *as_ctx)
73{
74 int err;
75 struct inode *inode;
76
77 inode = new_inode(dir->i_sb);
78 if (!inode)
79 return ERR_PTR(-ENOMEM);
80
81 inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
82
83 if (!S_ISLNK(*mode)) {
84 err = ceph_pre_init_acls(dir, mode, as_ctx);
85 if (err < 0)
86 goto out_err;
87 }
88
89 inode->i_state = 0;
90 inode->i_mode = *mode;
91
92 err = ceph_security_init_secctx(dentry, *mode, as_ctx);
93 if (err < 0)
94 goto out_err;
95
96 /*
97 * We'll skip setting fscrypt context for snapshots, leaving that for
98 * the handle_reply().
99 */
100 if (ceph_snap(dir) != CEPH_SNAPDIR) {
101 err = ceph_fscrypt_prepare_context(dir, inode, as_ctx);
102 if (err)
103 goto out_err;
104 }
105
106 return inode;
107out_err:
108 iput(inode);
109 return ERR_PTR(err);
110}
111
112void ceph_as_ctx_to_req(struct ceph_mds_request *req,
113 struct ceph_acl_sec_ctx *as_ctx)
114{
115 if (as_ctx->pagelist) {
116 req->r_pagelist = as_ctx->pagelist;
117 as_ctx->pagelist = NULL;
118 }
119 ceph_fscrypt_as_ctx_to_req(req, as_ctx);
120}
121
122/**
123 * ceph_get_inode - find or create/hash a new inode
124 * @sb: superblock to search and allocate in
125 * @vino: vino to search for
126 * @newino: optional new inode to insert if one isn't found (may be NULL)
127 *
128 * Search for or insert a new inode into the hash for the given vino, and
129 * return a reference to it. If new is non-NULL, its reference is consumed.
130 */
131struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino,
132 struct inode *newino)
133{
134 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(sb);
135 struct ceph_client *cl = mdsc->fsc->client;
136 struct inode *inode;
137
138 if (ceph_vino_is_reserved(vino))
139 return ERR_PTR(-EREMOTEIO);
140
141 if (newino) {
142 inode = inode_insert5(newino, (unsigned long)vino.ino,
143 ceph_ino_compare, ceph_set_ino_cb, &vino);
144 if (inode != newino)
145 iput(newino);
146 } else {
147 inode = iget5_locked(sb, (unsigned long)vino.ino,
148 ceph_ino_compare, ceph_set_ino_cb, &vino);
149 }
150
151 if (!inode) {
152 doutc(cl, "no inode found for %llx.%llx\n", vino.ino, vino.snap);
153 return ERR_PTR(-ENOMEM);
154 }
155
156 doutc(cl, "on %llx=%llx.%llx got %p new %d\n",
157 ceph_present_inode(inode), ceph_vinop(inode), inode,
158 !!(inode->i_state & I_NEW));
159 return inode;
160}
161
162/*
163 * get/construct snapdir inode for a given directory
164 */
165struct inode *ceph_get_snapdir(struct inode *parent)
166{
167 struct ceph_client *cl = ceph_inode_to_client(parent);
168 struct ceph_vino vino = {
169 .ino = ceph_ino(parent),
170 .snap = CEPH_SNAPDIR,
171 };
172 struct inode *inode = ceph_get_inode(parent->i_sb, vino, NULL);
173 struct ceph_inode_info *ci = ceph_inode(inode);
174 int ret = -ENOTDIR;
175
176 if (IS_ERR(inode))
177 return inode;
178
179 if (!S_ISDIR(parent->i_mode)) {
180 pr_warn_once_client(cl, "bad snapdir parent type (mode=0%o)\n",
181 parent->i_mode);
182 goto err;
183 }
184
185 if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
186 pr_warn_once_client(cl, "bad snapdir inode type (mode=0%o)\n",
187 inode->i_mode);
188 goto err;
189 }
190
191 inode->i_mode = parent->i_mode;
192 inode->i_uid = parent->i_uid;
193 inode->i_gid = parent->i_gid;
194 inode_set_mtime_to_ts(inode, inode_get_mtime(parent));
195 inode_set_ctime_to_ts(inode, inode_get_ctime(parent));
196 inode_set_atime_to_ts(inode, inode_get_atime(parent));
197 ci->i_rbytes = 0;
198 ci->i_btime = ceph_inode(parent)->i_btime;
199
200#ifdef CONFIG_FS_ENCRYPTION
201 /* if encrypted, just borrow fscrypt_auth from parent */
202 if (IS_ENCRYPTED(parent)) {
203 struct ceph_inode_info *pci = ceph_inode(parent);
204
205 ci->fscrypt_auth = kmemdup(pci->fscrypt_auth,
206 pci->fscrypt_auth_len,
207 GFP_KERNEL);
208 if (ci->fscrypt_auth) {
209 inode->i_flags |= S_ENCRYPTED;
210 ci->fscrypt_auth_len = pci->fscrypt_auth_len;
211 } else {
212 doutc(cl, "Failed to alloc snapdir fscrypt_auth\n");
213 ret = -ENOMEM;
214 goto err;
215 }
216 }
217#endif
218 if (inode->i_state & I_NEW) {
219 inode->i_op = &ceph_snapdir_iops;
220 inode->i_fop = &ceph_snapdir_fops;
221 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
222 unlock_new_inode(inode);
223 }
224
225 return inode;
226err:
227 if ((inode->i_state & I_NEW))
228 discard_new_inode(inode);
229 else
230 iput(inode);
231 return ERR_PTR(ret);
232}
233
234const struct inode_operations ceph_file_iops = {
235 .permission = ceph_permission,
236 .setattr = ceph_setattr,
237 .getattr = ceph_getattr,
238 .listxattr = ceph_listxattr,
239 .get_inode_acl = ceph_get_acl,
240 .set_acl = ceph_set_acl,
241};
242
243
244/*
245 * We use a 'frag tree' to keep track of the MDS's directory fragments
246 * for a given inode (usually there is just a single fragment). We
247 * need to know when a child frag is delegated to a new MDS, or when
248 * it is flagged as replicated, so we can direct our requests
249 * accordingly.
250 */
251
252/*
253 * find/create a frag in the tree
254 */
255static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
256 u32 f)
257{
258 struct inode *inode = &ci->netfs.inode;
259 struct ceph_client *cl = ceph_inode_to_client(inode);
260 struct rb_node **p;
261 struct rb_node *parent = NULL;
262 struct ceph_inode_frag *frag;
263 int c;
264
265 p = &ci->i_fragtree.rb_node;
266 while (*p) {
267 parent = *p;
268 frag = rb_entry(parent, struct ceph_inode_frag, node);
269 c = ceph_frag_compare(f, frag->frag);
270 if (c < 0)
271 p = &(*p)->rb_left;
272 else if (c > 0)
273 p = &(*p)->rb_right;
274 else
275 return frag;
276 }
277
278 frag = kmalloc(sizeof(*frag), GFP_NOFS);
279 if (!frag)
280 return ERR_PTR(-ENOMEM);
281
282 frag->frag = f;
283 frag->split_by = 0;
284 frag->mds = -1;
285 frag->ndist = 0;
286
287 rb_link_node(&frag->node, parent, p);
288 rb_insert_color(&frag->node, &ci->i_fragtree);
289
290 doutc(cl, "added %p %llx.%llx frag %x\n", inode, ceph_vinop(inode), f);
291 return frag;
292}
293
294/*
295 * find a specific frag @f
296 */
297struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
298{
299 struct rb_node *n = ci->i_fragtree.rb_node;
300
301 while (n) {
302 struct ceph_inode_frag *frag =
303 rb_entry(n, struct ceph_inode_frag, node);
304 int c = ceph_frag_compare(f, frag->frag);
305 if (c < 0)
306 n = n->rb_left;
307 else if (c > 0)
308 n = n->rb_right;
309 else
310 return frag;
311 }
312 return NULL;
313}
314
315/*
316 * Choose frag containing the given value @v. If @pfrag is
317 * specified, copy the frag delegation info to the caller if
318 * it is present.
319 */
320static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
321 struct ceph_inode_frag *pfrag, int *found)
322{
323 struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
324 u32 t = ceph_frag_make(0, 0);
325 struct ceph_inode_frag *frag;
326 unsigned nway, i;
327 u32 n;
328
329 if (found)
330 *found = 0;
331
332 while (1) {
333 WARN_ON(!ceph_frag_contains_value(t, v));
334 frag = __ceph_find_frag(ci, t);
335 if (!frag)
336 break; /* t is a leaf */
337 if (frag->split_by == 0) {
338 if (pfrag)
339 memcpy(pfrag, frag, sizeof(*pfrag));
340 if (found)
341 *found = 1;
342 break;
343 }
344
345 /* choose child */
346 nway = 1 << frag->split_by;
347 doutc(cl, "frag(%x) %x splits by %d (%d ways)\n", v, t,
348 frag->split_by, nway);
349 for (i = 0; i < nway; i++) {
350 n = ceph_frag_make_child(t, frag->split_by, i);
351 if (ceph_frag_contains_value(n, v)) {
352 t = n;
353 break;
354 }
355 }
356 BUG_ON(i == nway);
357 }
358 doutc(cl, "frag(%x) = %x\n", v, t);
359
360 return t;
361}
362
363u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
364 struct ceph_inode_frag *pfrag, int *found)
365{
366 u32 ret;
367 mutex_lock(&ci->i_fragtree_mutex);
368 ret = __ceph_choose_frag(ci, v, pfrag, found);
369 mutex_unlock(&ci->i_fragtree_mutex);
370 return ret;
371}
372
373/*
374 * Process dirfrag (delegation) info from the mds. Include leaf
375 * fragment in tree ONLY if ndist > 0. Otherwise, only
376 * branches/splits are included in i_fragtree)
377 */
378static int ceph_fill_dirfrag(struct inode *inode,
379 struct ceph_mds_reply_dirfrag *dirinfo)
380{
381 struct ceph_inode_info *ci = ceph_inode(inode);
382 struct ceph_client *cl = ceph_inode_to_client(inode);
383 struct ceph_inode_frag *frag;
384 u32 id = le32_to_cpu(dirinfo->frag);
385 int mds = le32_to_cpu(dirinfo->auth);
386 int ndist = le32_to_cpu(dirinfo->ndist);
387 int diri_auth = -1;
388 int i;
389 int err = 0;
390
391 spin_lock(&ci->i_ceph_lock);
392 if (ci->i_auth_cap)
393 diri_auth = ci->i_auth_cap->mds;
394 spin_unlock(&ci->i_ceph_lock);
395
396 if (mds == -1) /* CDIR_AUTH_PARENT */
397 mds = diri_auth;
398
399 mutex_lock(&ci->i_fragtree_mutex);
400 if (ndist == 0 && mds == diri_auth) {
401 /* no delegation info needed. */
402 frag = __ceph_find_frag(ci, id);
403 if (!frag)
404 goto out;
405 if (frag->split_by == 0) {
406 /* tree leaf, remove */
407 doutc(cl, "removed %p %llx.%llx frag %x (no ref)\n",
408 inode, ceph_vinop(inode), id);
409 rb_erase(&frag->node, &ci->i_fragtree);
410 kfree(frag);
411 } else {
412 /* tree branch, keep and clear */
413 doutc(cl, "cleared %p %llx.%llx frag %x referral\n",
414 inode, ceph_vinop(inode), id);
415 frag->mds = -1;
416 frag->ndist = 0;
417 }
418 goto out;
419 }
420
421
422 /* find/add this frag to store mds delegation info */
423 frag = __get_or_create_frag(ci, id);
424 if (IS_ERR(frag)) {
425 /* this is not the end of the world; we can continue
426 with bad/inaccurate delegation info */
427 pr_err_client(cl, "ENOMEM on mds ref %p %llx.%llx fg %x\n",
428 inode, ceph_vinop(inode),
429 le32_to_cpu(dirinfo->frag));
430 err = -ENOMEM;
431 goto out;
432 }
433
434 frag->mds = mds;
435 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
436 for (i = 0; i < frag->ndist; i++)
437 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
438 doutc(cl, "%p %llx.%llx frag %x ndist=%d\n", inode,
439 ceph_vinop(inode), frag->frag, frag->ndist);
440
441out:
442 mutex_unlock(&ci->i_fragtree_mutex);
443 return err;
444}
445
446static int frag_tree_split_cmp(const void *l, const void *r)
447{
448 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
449 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
450 return ceph_frag_compare(le32_to_cpu(ls->frag),
451 le32_to_cpu(rs->frag));
452}
453
454static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
455{
456 if (!frag)
457 return f == ceph_frag_make(0, 0);
458 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
459 return false;
460 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
461}
462
463static int ceph_fill_fragtree(struct inode *inode,
464 struct ceph_frag_tree_head *fragtree,
465 struct ceph_mds_reply_dirfrag *dirinfo)
466{
467 struct ceph_client *cl = ceph_inode_to_client(inode);
468 struct ceph_inode_info *ci = ceph_inode(inode);
469 struct ceph_inode_frag *frag, *prev_frag = NULL;
470 struct rb_node *rb_node;
471 unsigned i, split_by, nsplits;
472 u32 id;
473 bool update = false;
474
475 mutex_lock(&ci->i_fragtree_mutex);
476 nsplits = le32_to_cpu(fragtree->nsplits);
477 if (nsplits != ci->i_fragtree_nsplits) {
478 update = true;
479 } else if (nsplits) {
480 i = get_random_u32_below(nsplits);
481 id = le32_to_cpu(fragtree->splits[i].frag);
482 if (!__ceph_find_frag(ci, id))
483 update = true;
484 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
485 rb_node = rb_first(&ci->i_fragtree);
486 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
487 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
488 update = true;
489 }
490 if (!update && dirinfo) {
491 id = le32_to_cpu(dirinfo->frag);
492 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
493 update = true;
494 }
495 if (!update)
496 goto out_unlock;
497
498 if (nsplits > 1) {
499 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
500 frag_tree_split_cmp, NULL);
501 }
502
503 doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
504 rb_node = rb_first(&ci->i_fragtree);
505 for (i = 0; i < nsplits; i++) {
506 id = le32_to_cpu(fragtree->splits[i].frag);
507 split_by = le32_to_cpu(fragtree->splits[i].by);
508 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
509 pr_err_client(cl, "%p %llx.%llx invalid split %d/%u, "
510 "frag %x split by %d\n", inode,
511 ceph_vinop(inode), i, nsplits, id, split_by);
512 continue;
513 }
514 frag = NULL;
515 while (rb_node) {
516 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
517 if (ceph_frag_compare(frag->frag, id) >= 0) {
518 if (frag->frag != id)
519 frag = NULL;
520 else
521 rb_node = rb_next(rb_node);
522 break;
523 }
524 rb_node = rb_next(rb_node);
525 /* delete stale split/leaf node */
526 if (frag->split_by > 0 ||
527 !is_frag_child(frag->frag, prev_frag)) {
528 rb_erase(&frag->node, &ci->i_fragtree);
529 if (frag->split_by > 0)
530 ci->i_fragtree_nsplits--;
531 kfree(frag);
532 }
533 frag = NULL;
534 }
535 if (!frag) {
536 frag = __get_or_create_frag(ci, id);
537 if (IS_ERR(frag))
538 continue;
539 }
540 if (frag->split_by == 0)
541 ci->i_fragtree_nsplits++;
542 frag->split_by = split_by;
543 doutc(cl, " frag %x split by %d\n", frag->frag, frag->split_by);
544 prev_frag = frag;
545 }
546 while (rb_node) {
547 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
548 rb_node = rb_next(rb_node);
549 /* delete stale split/leaf node */
550 if (frag->split_by > 0 ||
551 !is_frag_child(frag->frag, prev_frag)) {
552 rb_erase(&frag->node, &ci->i_fragtree);
553 if (frag->split_by > 0)
554 ci->i_fragtree_nsplits--;
555 kfree(frag);
556 }
557 }
558out_unlock:
559 mutex_unlock(&ci->i_fragtree_mutex);
560 return 0;
561}
562
563/*
564 * initialize a newly allocated inode.
565 */
566struct inode *ceph_alloc_inode(struct super_block *sb)
567{
568 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
569 struct ceph_inode_info *ci;
570 int i;
571
572 ci = alloc_inode_sb(sb, ceph_inode_cachep, GFP_NOFS);
573 if (!ci)
574 return NULL;
575
576 doutc(fsc->client, "%p\n", &ci->netfs.inode);
577
578 /* Set parameters for the netfs library */
579 netfs_inode_init(&ci->netfs, &ceph_netfs_ops, false);
580
581 spin_lock_init(&ci->i_ceph_lock);
582
583 ci->i_version = 0;
584 ci->i_inline_version = 0;
585 ci->i_time_warp_seq = 0;
586 ci->i_ceph_flags = 0;
587 atomic64_set(&ci->i_ordered_count, 1);
588 atomic64_set(&ci->i_release_count, 1);
589 atomic64_set(&ci->i_complete_seq[0], 0);
590 atomic64_set(&ci->i_complete_seq[1], 0);
591 ci->i_symlink = NULL;
592
593 ci->i_max_bytes = 0;
594 ci->i_max_files = 0;
595
596 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
597 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
598 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
599
600 ci->i_fragtree = RB_ROOT;
601 mutex_init(&ci->i_fragtree_mutex);
602
603 ci->i_xattrs.blob = NULL;
604 ci->i_xattrs.prealloc_blob = NULL;
605 ci->i_xattrs.dirty = false;
606 ci->i_xattrs.index = RB_ROOT;
607 ci->i_xattrs.count = 0;
608 ci->i_xattrs.names_size = 0;
609 ci->i_xattrs.vals_size = 0;
610 ci->i_xattrs.version = 0;
611 ci->i_xattrs.index_version = 0;
612
613 ci->i_caps = RB_ROOT;
614 ci->i_auth_cap = NULL;
615 ci->i_dirty_caps = 0;
616 ci->i_flushing_caps = 0;
617 INIT_LIST_HEAD(&ci->i_dirty_item);
618 INIT_LIST_HEAD(&ci->i_flushing_item);
619 ci->i_prealloc_cap_flush = NULL;
620 INIT_LIST_HEAD(&ci->i_cap_flush_list);
621 init_waitqueue_head(&ci->i_cap_wq);
622 ci->i_hold_caps_max = 0;
623 INIT_LIST_HEAD(&ci->i_cap_delay_list);
624 INIT_LIST_HEAD(&ci->i_cap_snaps);
625 ci->i_head_snapc = NULL;
626 ci->i_snap_caps = 0;
627
628 ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
629 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
630 ci->i_nr_by_mode[i] = 0;
631
632 mutex_init(&ci->i_truncate_mutex);
633 ci->i_truncate_seq = 0;
634 ci->i_truncate_size = 0;
635 ci->i_truncate_pending = 0;
636 ci->i_truncate_pagecache_size = 0;
637
638 ci->i_max_size = 0;
639 ci->i_reported_size = 0;
640 ci->i_wanted_max_size = 0;
641 ci->i_requested_max_size = 0;
642
643 ci->i_pin_ref = 0;
644 ci->i_rd_ref = 0;
645 ci->i_rdcache_ref = 0;
646 ci->i_wr_ref = 0;
647 ci->i_wb_ref = 0;
648 ci->i_fx_ref = 0;
649 ci->i_wrbuffer_ref = 0;
650 ci->i_wrbuffer_ref_head = 0;
651 atomic_set(&ci->i_filelock_ref, 0);
652 atomic_set(&ci->i_shared_gen, 1);
653 ci->i_rdcache_gen = 0;
654 ci->i_rdcache_revoking = 0;
655
656 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
657 INIT_LIST_HEAD(&ci->i_unsafe_iops);
658 spin_lock_init(&ci->i_unsafe_lock);
659
660 ci->i_snap_realm = NULL;
661 INIT_LIST_HEAD(&ci->i_snap_realm_item);
662 INIT_LIST_HEAD(&ci->i_snap_flush_item);
663
664 INIT_WORK(&ci->i_work, ceph_inode_work);
665 ci->i_work_mask = 0;
666 memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
667#ifdef CONFIG_FS_ENCRYPTION
668 ci->fscrypt_auth = NULL;
669 ci->fscrypt_auth_len = 0;
670#endif
671 return &ci->netfs.inode;
672}
673
674void ceph_free_inode(struct inode *inode)
675{
676 struct ceph_inode_info *ci = ceph_inode(inode);
677
678 kfree(ci->i_symlink);
679#ifdef CONFIG_FS_ENCRYPTION
680 kfree(ci->fscrypt_auth);
681#endif
682 fscrypt_free_inode(inode);
683 kmem_cache_free(ceph_inode_cachep, ci);
684}
685
686void ceph_evict_inode(struct inode *inode)
687{
688 struct ceph_inode_info *ci = ceph_inode(inode);
689 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
690 struct ceph_client *cl = ceph_inode_to_client(inode);
691 struct ceph_inode_frag *frag;
692 struct rb_node *n;
693
694 doutc(cl, "%p ino %llx.%llx\n", inode, ceph_vinop(inode));
695
696 percpu_counter_dec(&mdsc->metric.total_inodes);
697
698 netfs_wait_for_outstanding_io(inode);
699 truncate_inode_pages_final(&inode->i_data);
700 if (inode->i_state & I_PINNING_NETFS_WB)
701 ceph_fscache_unuse_cookie(inode, true);
702 clear_inode(inode);
703
704 ceph_fscache_unregister_inode_cookie(ci);
705 fscrypt_put_encryption_info(inode);
706
707 __ceph_remove_caps(ci);
708
709 if (__ceph_has_quota(ci, QUOTA_GET_ANY))
710 ceph_adjust_quota_realms_count(inode, false);
711
712 /*
713 * we may still have a snap_realm reference if there are stray
714 * caps in i_snap_caps.
715 */
716 if (ci->i_snap_realm) {
717 if (ceph_snap(inode) == CEPH_NOSNAP) {
718 doutc(cl, " dropping residual ref to snap realm %p\n",
719 ci->i_snap_realm);
720 ceph_change_snap_realm(inode, NULL);
721 } else {
722 ceph_put_snapid_map(mdsc, ci->i_snapid_map);
723 ci->i_snap_realm = NULL;
724 }
725 }
726
727 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
728 frag = rb_entry(n, struct ceph_inode_frag, node);
729 rb_erase(n, &ci->i_fragtree);
730 kfree(frag);
731 }
732 ci->i_fragtree_nsplits = 0;
733
734 __ceph_destroy_xattrs(ci);
735 if (ci->i_xattrs.blob)
736 ceph_buffer_put(ci->i_xattrs.blob);
737 if (ci->i_xattrs.prealloc_blob)
738 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
739
740 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
741 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
742}
743
744static inline blkcnt_t calc_inode_blocks(u64 size)
745{
746 return (size + (1<<9) - 1) >> 9;
747}
748
749/*
750 * Helpers to fill in size, ctime, mtime, and atime. We have to be
751 * careful because either the client or MDS may have more up to date
752 * info, depending on which capabilities are held, and whether
753 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
754 * and size are monotonically increasing, except when utimes() or
755 * truncate() increments the corresponding _seq values.)
756 */
757int ceph_fill_file_size(struct inode *inode, int issued,
758 u32 truncate_seq, u64 truncate_size, u64 size)
759{
760 struct ceph_client *cl = ceph_inode_to_client(inode);
761 struct ceph_inode_info *ci = ceph_inode(inode);
762 int queue_trunc = 0;
763 loff_t isize = i_size_read(inode);
764
765 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
766 (truncate_seq == ci->i_truncate_seq && size > isize)) {
767 doutc(cl, "size %lld -> %llu\n", isize, size);
768 if (size > 0 && S_ISDIR(inode->i_mode)) {
769 pr_err_client(cl, "non-zero size for directory\n");
770 size = 0;
771 }
772 i_size_write(inode, size);
773 inode->i_blocks = calc_inode_blocks(size);
774 /*
775 * If we're expanding, then we should be able to just update
776 * the existing cookie.
777 */
778 if (size > isize)
779 ceph_fscache_update(inode);
780 ci->i_reported_size = size;
781 if (truncate_seq != ci->i_truncate_seq) {
782 doutc(cl, "truncate_seq %u -> %u\n",
783 ci->i_truncate_seq, truncate_seq);
784 ci->i_truncate_seq = truncate_seq;
785
786 /* the MDS should have revoked these caps */
787 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_RD |
788 CEPH_CAP_FILE_LAZYIO));
789 /*
790 * If we hold relevant caps, or in the case where we're
791 * not the only client referencing this file and we
792 * don't hold those caps, then we need to check whether
793 * the file is either opened or mmaped
794 */
795 if ((issued & (CEPH_CAP_FILE_CACHE|
796 CEPH_CAP_FILE_BUFFER)) ||
797 mapping_mapped(inode->i_mapping) ||
798 __ceph_is_file_opened(ci)) {
799 ci->i_truncate_pending++;
800 queue_trunc = 1;
801 }
802 }
803 }
804
805 /*
806 * It's possible that the new sizes of the two consecutive
807 * size truncations will be in the same fscrypt last block,
808 * and we need to truncate the corresponding page caches
809 * anyway.
810 */
811 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0) {
812 doutc(cl, "truncate_size %lld -> %llu, encrypted %d\n",
813 ci->i_truncate_size, truncate_size,
814 !!IS_ENCRYPTED(inode));
815
816 ci->i_truncate_size = truncate_size;
817
818 if (IS_ENCRYPTED(inode)) {
819 doutc(cl, "truncate_pagecache_size %lld -> %llu\n",
820 ci->i_truncate_pagecache_size, size);
821 ci->i_truncate_pagecache_size = size;
822 } else {
823 ci->i_truncate_pagecache_size = truncate_size;
824 }
825 }
826 return queue_trunc;
827}
828
829void ceph_fill_file_time(struct inode *inode, int issued,
830 u64 time_warp_seq, struct timespec64 *ctime,
831 struct timespec64 *mtime, struct timespec64 *atime)
832{
833 struct ceph_client *cl = ceph_inode_to_client(inode);
834 struct ceph_inode_info *ci = ceph_inode(inode);
835 struct timespec64 ictime = inode_get_ctime(inode);
836 int warn = 0;
837
838 if (issued & (CEPH_CAP_FILE_EXCL|
839 CEPH_CAP_FILE_WR|
840 CEPH_CAP_FILE_BUFFER|
841 CEPH_CAP_AUTH_EXCL|
842 CEPH_CAP_XATTR_EXCL)) {
843 if (ci->i_version == 0 ||
844 timespec64_compare(ctime, &ictime) > 0) {
845 doutc(cl, "ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
846 ictime.tv_sec, ictime.tv_nsec,
847 ctime->tv_sec, ctime->tv_nsec);
848 inode_set_ctime_to_ts(inode, *ctime);
849 }
850 if (ci->i_version == 0 ||
851 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
852 /* the MDS did a utimes() */
853 doutc(cl, "mtime %lld.%09ld -> %lld.%09ld tw %d -> %d\n",
854 inode_get_mtime_sec(inode),
855 inode_get_mtime_nsec(inode),
856 mtime->tv_sec, mtime->tv_nsec,
857 ci->i_time_warp_seq, (int)time_warp_seq);
858
859 inode_set_mtime_to_ts(inode, *mtime);
860 inode_set_atime_to_ts(inode, *atime);
861 ci->i_time_warp_seq = time_warp_seq;
862 } else if (time_warp_seq == ci->i_time_warp_seq) {
863 struct timespec64 ts;
864
865 /* nobody did utimes(); take the max */
866 ts = inode_get_mtime(inode);
867 if (timespec64_compare(mtime, &ts) > 0) {
868 doutc(cl, "mtime %lld.%09ld -> %lld.%09ld inc\n",
869 ts.tv_sec, ts.tv_nsec,
870 mtime->tv_sec, mtime->tv_nsec);
871 inode_set_mtime_to_ts(inode, *mtime);
872 }
873 ts = inode_get_atime(inode);
874 if (timespec64_compare(atime, &ts) > 0) {
875 doutc(cl, "atime %lld.%09ld -> %lld.%09ld inc\n",
876 ts.tv_sec, ts.tv_nsec,
877 atime->tv_sec, atime->tv_nsec);
878 inode_set_atime_to_ts(inode, *atime);
879 }
880 } else if (issued & CEPH_CAP_FILE_EXCL) {
881 /* we did a utimes(); ignore mds values */
882 } else {
883 warn = 1;
884 }
885 } else {
886 /* we have no write|excl caps; whatever the MDS says is true */
887 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
888 inode_set_ctime_to_ts(inode, *ctime);
889 inode_set_mtime_to_ts(inode, *mtime);
890 inode_set_atime_to_ts(inode, *atime);
891 ci->i_time_warp_seq = time_warp_seq;
892 } else {
893 warn = 1;
894 }
895 }
896 if (warn) /* time_warp_seq shouldn't go backwards */
897 doutc(cl, "%p mds time_warp_seq %llu < %u\n", inode,
898 time_warp_seq, ci->i_time_warp_seq);
899}
900
901#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
902static int decode_encrypted_symlink(struct ceph_mds_client *mdsc,
903 const char *encsym,
904 int enclen, u8 **decsym)
905{
906 struct ceph_client *cl = mdsc->fsc->client;
907 int declen;
908 u8 *sym;
909
910 sym = kmalloc(enclen + 1, GFP_NOFS);
911 if (!sym)
912 return -ENOMEM;
913
914 declen = ceph_base64_decode(encsym, enclen, sym);
915 if (declen < 0) {
916 pr_err_client(cl,
917 "can't decode symlink (%d). Content: %.*s\n",
918 declen, enclen, encsym);
919 kfree(sym);
920 return -EIO;
921 }
922 sym[declen + 1] = '\0';
923 *decsym = sym;
924 return declen;
925}
926#else
927static int decode_encrypted_symlink(struct ceph_mds_client *mdsc,
928 const char *encsym,
929 int symlen, u8 **decsym)
930{
931 return -EOPNOTSUPP;
932}
933#endif
934
935/*
936 * Populate an inode based on info from mds. May be called on new or
937 * existing inodes.
938 */
939int ceph_fill_inode(struct inode *inode, struct page *locked_page,
940 struct ceph_mds_reply_info_in *iinfo,
941 struct ceph_mds_reply_dirfrag *dirinfo,
942 struct ceph_mds_session *session, int cap_fmode,
943 struct ceph_cap_reservation *caps_reservation)
944{
945 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
946 struct ceph_client *cl = mdsc->fsc->client;
947 struct ceph_mds_reply_inode *info = iinfo->in;
948 struct ceph_inode_info *ci = ceph_inode(inode);
949 int issued, new_issued, info_caps;
950 struct timespec64 mtime, atime, ctime;
951 struct ceph_buffer *xattr_blob = NULL;
952 struct ceph_buffer *old_blob = NULL;
953 struct ceph_string *pool_ns = NULL;
954 struct ceph_cap *new_cap = NULL;
955 int err = 0;
956 bool wake = false;
957 bool queue_trunc = false;
958 bool new_version = false;
959 bool fill_inline = false;
960 umode_t mode = le32_to_cpu(info->mode);
961 dev_t rdev = le32_to_cpu(info->rdev);
962
963 lockdep_assert_held(&mdsc->snap_rwsem);
964
965 doutc(cl, "%p ino %llx.%llx v %llu had %llu\n", inode, ceph_vinop(inode),
966 le64_to_cpu(info->version), ci->i_version);
967
968 /* Once I_NEW is cleared, we can't change type or dev numbers */
969 if (inode->i_state & I_NEW) {
970 inode->i_mode = mode;
971 } else {
972 if (inode_wrong_type(inode, mode)) {
973 pr_warn_once_client(cl,
974 "inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
975 ceph_vinop(inode), inode->i_mode, mode);
976 return -ESTALE;
977 }
978
979 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
980 pr_warn_once_client(cl,
981 "dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
982 ceph_vinop(inode), MAJOR(inode->i_rdev),
983 MINOR(inode->i_rdev), MAJOR(rdev),
984 MINOR(rdev));
985 return -ESTALE;
986 }
987 }
988
989 info_caps = le32_to_cpu(info->cap.caps);
990
991 /* prealloc new cap struct */
992 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
993 new_cap = ceph_get_cap(mdsc, caps_reservation);
994 if (!new_cap)
995 return -ENOMEM;
996 }
997
998 /*
999 * prealloc xattr data, if it looks like we'll need it. only
1000 * if len > 4 (meaning there are actually xattrs; the first 4
1001 * bytes are the xattr count).
1002 */
1003 if (iinfo->xattr_len > 4) {
1004 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
1005 if (!xattr_blob)
1006 pr_err_client(cl, "ENOMEM xattr blob %d bytes\n",
1007 iinfo->xattr_len);
1008 }
1009
1010 if (iinfo->pool_ns_len > 0)
1011 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
1012 iinfo->pool_ns_len);
1013
1014 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
1015 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
1016
1017 spin_lock(&ci->i_ceph_lock);
1018
1019 /*
1020 * provided version will be odd if inode value is projected,
1021 * even if stable. skip the update if we have newer stable
1022 * info (ours>=theirs, e.g. due to racing mds replies), unless
1023 * we are getting projected (unstable) info (in which case the
1024 * version is odd, and we want ours>theirs).
1025 * us them
1026 * 2 2 skip
1027 * 3 2 skip
1028 * 3 3 update
1029 */
1030 if (ci->i_version == 0 ||
1031 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1032 le64_to_cpu(info->version) > (ci->i_version & ~1)))
1033 new_version = true;
1034
1035 /* Update change_attribute */
1036 inode_set_max_iversion_raw(inode, iinfo->change_attr);
1037
1038 __ceph_caps_issued(ci, &issued);
1039 issued |= __ceph_caps_dirty(ci);
1040 new_issued = ~issued & info_caps;
1041
1042 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
1043
1044#ifdef CONFIG_FS_ENCRYPTION
1045 if (iinfo->fscrypt_auth_len &&
1046 ((inode->i_state & I_NEW) || (ci->fscrypt_auth_len == 0))) {
1047 kfree(ci->fscrypt_auth);
1048 ci->fscrypt_auth_len = iinfo->fscrypt_auth_len;
1049 ci->fscrypt_auth = iinfo->fscrypt_auth;
1050 iinfo->fscrypt_auth = NULL;
1051 iinfo->fscrypt_auth_len = 0;
1052 inode_set_flags(inode, S_ENCRYPTED, S_ENCRYPTED);
1053 }
1054#endif
1055
1056 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
1057 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
1058 inode->i_mode = mode;
1059 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
1060 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
1061 doutc(cl, "%p %llx.%llx mode 0%o uid.gid %d.%d\n", inode,
1062 ceph_vinop(inode), inode->i_mode,
1063 from_kuid(&init_user_ns, inode->i_uid),
1064 from_kgid(&init_user_ns, inode->i_gid));
1065 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
1066 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
1067 }
1068
1069 /* directories have fl_stripe_unit set to zero */
1070 if (IS_ENCRYPTED(inode))
1071 inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
1072 else if (le32_to_cpu(info->layout.fl_stripe_unit))
1073 inode->i_blkbits =
1074 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
1075 else
1076 inode->i_blkbits = CEPH_BLOCK_SHIFT;
1077
1078 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
1079 (issued & CEPH_CAP_LINK_EXCL) == 0)
1080 set_nlink(inode, le32_to_cpu(info->nlink));
1081
1082 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
1083 /* be careful with mtime, atime, size */
1084 ceph_decode_timespec64(&atime, &info->atime);
1085 ceph_decode_timespec64(&mtime, &info->mtime);
1086 ceph_decode_timespec64(&ctime, &info->ctime);
1087 ceph_fill_file_time(inode, issued,
1088 le32_to_cpu(info->time_warp_seq),
1089 &ctime, &mtime, &atime);
1090 }
1091
1092 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
1093 ci->i_files = le64_to_cpu(info->files);
1094 ci->i_subdirs = le64_to_cpu(info->subdirs);
1095 }
1096
1097 if (new_version ||
1098 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
1099 u64 size = le64_to_cpu(info->size);
1100 s64 old_pool = ci->i_layout.pool_id;
1101 struct ceph_string *old_ns;
1102
1103 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
1104 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
1105 lockdep_is_held(&ci->i_ceph_lock));
1106 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
1107
1108 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
1109 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
1110
1111 pool_ns = old_ns;
1112
1113 if (IS_ENCRYPTED(inode) && size &&
1114 iinfo->fscrypt_file_len == sizeof(__le64)) {
1115 u64 fsize = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
1116
1117 if (size == round_up(fsize, CEPH_FSCRYPT_BLOCK_SIZE)) {
1118 size = fsize;
1119 } else {
1120 pr_warn_client(cl,
1121 "fscrypt size mismatch: size=%llu fscrypt_file=%llu, discarding fscrypt_file size.\n",
1122 info->size, size);
1123 }
1124 }
1125
1126 queue_trunc = ceph_fill_file_size(inode, issued,
1127 le32_to_cpu(info->truncate_seq),
1128 le64_to_cpu(info->truncate_size),
1129 size);
1130 /* only update max_size on auth cap */
1131 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1132 ci->i_max_size != le64_to_cpu(info->max_size)) {
1133 doutc(cl, "max_size %lld -> %llu\n",
1134 ci->i_max_size, le64_to_cpu(info->max_size));
1135 ci->i_max_size = le64_to_cpu(info->max_size);
1136 }
1137 }
1138
1139 /* layout and rstat are not tracked by capability, update them if
1140 * the inode info is from auth mds */
1141 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
1142 if (S_ISDIR(inode->i_mode)) {
1143 ci->i_dir_layout = iinfo->dir_layout;
1144 ci->i_rbytes = le64_to_cpu(info->rbytes);
1145 ci->i_rfiles = le64_to_cpu(info->rfiles);
1146 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
1147 ci->i_dir_pin = iinfo->dir_pin;
1148 ci->i_rsnaps = iinfo->rsnaps;
1149 ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
1150 }
1151 }
1152
1153 /* xattrs */
1154 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
1155 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
1156 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
1157 if (ci->i_xattrs.blob)
1158 old_blob = ci->i_xattrs.blob;
1159 ci->i_xattrs.blob = xattr_blob;
1160 if (xattr_blob)
1161 memcpy(ci->i_xattrs.blob->vec.iov_base,
1162 iinfo->xattr_data, iinfo->xattr_len);
1163 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
1164 ceph_forget_all_cached_acls(inode);
1165 ceph_security_invalidate_secctx(inode);
1166 xattr_blob = NULL;
1167 }
1168
1169 /* finally update i_version */
1170 if (le64_to_cpu(info->version) > ci->i_version)
1171 ci->i_version = le64_to_cpu(info->version);
1172
1173 inode->i_mapping->a_ops = &ceph_aops;
1174
1175 switch (inode->i_mode & S_IFMT) {
1176 case S_IFIFO:
1177 case S_IFBLK:
1178 case S_IFCHR:
1179 case S_IFSOCK:
1180 inode->i_blkbits = PAGE_SHIFT;
1181 init_special_inode(inode, inode->i_mode, rdev);
1182 inode->i_op = &ceph_file_iops;
1183 break;
1184 case S_IFREG:
1185 inode->i_op = &ceph_file_iops;
1186 inode->i_fop = &ceph_file_fops;
1187 break;
1188 case S_IFLNK:
1189 if (!ci->i_symlink) {
1190 u32 symlen = iinfo->symlink_len;
1191 char *sym;
1192
1193 spin_unlock(&ci->i_ceph_lock);
1194
1195 if (IS_ENCRYPTED(inode)) {
1196 if (symlen != i_size_read(inode))
1197 pr_err_client(cl,
1198 "%p %llx.%llx BAD symlink size %lld\n",
1199 inode, ceph_vinop(inode),
1200 i_size_read(inode));
1201
1202 err = decode_encrypted_symlink(mdsc, iinfo->symlink,
1203 symlen, (u8 **)&sym);
1204 if (err < 0) {
1205 pr_err_client(cl,
1206 "decoding encrypted symlink failed: %d\n",
1207 err);
1208 goto out;
1209 }
1210 symlen = err;
1211 i_size_write(inode, symlen);
1212 inode->i_blocks = calc_inode_blocks(symlen);
1213 } else {
1214 if (symlen != i_size_read(inode)) {
1215 pr_err_client(cl,
1216 "%p %llx.%llx BAD symlink size %lld\n",
1217 inode, ceph_vinop(inode),
1218 i_size_read(inode));
1219 i_size_write(inode, symlen);
1220 inode->i_blocks = calc_inode_blocks(symlen);
1221 }
1222
1223 err = -ENOMEM;
1224 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
1225 if (!sym)
1226 goto out;
1227 }
1228
1229 spin_lock(&ci->i_ceph_lock);
1230 if (!ci->i_symlink)
1231 ci->i_symlink = sym;
1232 else
1233 kfree(sym); /* lost a race */
1234 }
1235
1236 if (IS_ENCRYPTED(inode)) {
1237 /*
1238 * Encrypted symlinks need to be decrypted before we can
1239 * cache their targets in i_link. Don't touch it here.
1240 */
1241 inode->i_op = &ceph_encrypted_symlink_iops;
1242 } else {
1243 inode->i_link = ci->i_symlink;
1244 inode->i_op = &ceph_symlink_iops;
1245 }
1246 break;
1247 case S_IFDIR:
1248 inode->i_op = &ceph_dir_iops;
1249 inode->i_fop = &ceph_dir_fops;
1250 break;
1251 default:
1252 pr_err_client(cl, "%p %llx.%llx BAD mode 0%o\n", inode,
1253 ceph_vinop(inode), inode->i_mode);
1254 }
1255
1256 /* were we issued a capability? */
1257 if (info_caps) {
1258 if (ceph_snap(inode) == CEPH_NOSNAP) {
1259 ceph_add_cap(inode, session,
1260 le64_to_cpu(info->cap.cap_id),
1261 info_caps,
1262 le32_to_cpu(info->cap.wanted),
1263 le32_to_cpu(info->cap.seq),
1264 le32_to_cpu(info->cap.mseq),
1265 le64_to_cpu(info->cap.realm),
1266 info->cap.flags, &new_cap);
1267
1268 /* set dir completion flag? */
1269 if (S_ISDIR(inode->i_mode) &&
1270 ci->i_files == 0 && ci->i_subdirs == 0 &&
1271 (info_caps & CEPH_CAP_FILE_SHARED) &&
1272 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1273 !__ceph_dir_is_complete(ci)) {
1274 doutc(cl, " marking %p complete (empty)\n",
1275 inode);
1276 i_size_write(inode, 0);
1277 __ceph_dir_set_complete(ci,
1278 atomic64_read(&ci->i_release_count),
1279 atomic64_read(&ci->i_ordered_count));
1280 }
1281
1282 wake = true;
1283 } else {
1284 doutc(cl, " %p got snap_caps %s\n", inode,
1285 ceph_cap_string(info_caps));
1286 ci->i_snap_caps |= info_caps;
1287 }
1288 }
1289
1290 if (iinfo->inline_version > 0 &&
1291 iinfo->inline_version >= ci->i_inline_version) {
1292 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1293 ci->i_inline_version = iinfo->inline_version;
1294 if (ceph_has_inline_data(ci) &&
1295 (locked_page || (info_caps & cache_caps)))
1296 fill_inline = true;
1297 }
1298
1299 if (cap_fmode >= 0) {
1300 if (!info_caps)
1301 pr_warn_client(cl, "mds issued no caps on %llx.%llx\n",
1302 ceph_vinop(inode));
1303 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1304 }
1305
1306 spin_unlock(&ci->i_ceph_lock);
1307
1308 ceph_fscache_register_inode_cookie(inode);
1309
1310 if (fill_inline)
1311 ceph_fill_inline_data(inode, locked_page,
1312 iinfo->inline_data, iinfo->inline_len);
1313
1314 if (wake)
1315 wake_up_all(&ci->i_cap_wq);
1316
1317 /* queue truncate if we saw i_size decrease */
1318 if (queue_trunc)
1319 ceph_queue_vmtruncate(inode);
1320
1321 /* populate frag tree */
1322 if (S_ISDIR(inode->i_mode))
1323 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1324
1325 /* update delegation info? */
1326 if (dirinfo)
1327 ceph_fill_dirfrag(inode, dirinfo);
1328
1329 err = 0;
1330out:
1331 if (new_cap)
1332 ceph_put_cap(mdsc, new_cap);
1333 ceph_buffer_put(old_blob);
1334 ceph_buffer_put(xattr_blob);
1335 ceph_put_string(pool_ns);
1336 return err;
1337}
1338
1339/*
1340 * caller should hold session s_mutex and dentry->d_lock.
1341 */
1342static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1343 struct ceph_mds_reply_lease *lease,
1344 struct ceph_mds_session *session,
1345 unsigned long from_time,
1346 struct ceph_mds_session **old_lease_session)
1347{
1348 struct ceph_client *cl = ceph_inode_to_client(dir);
1349 struct ceph_dentry_info *di = ceph_dentry(dentry);
1350 unsigned mask = le16_to_cpu(lease->mask);
1351 long unsigned duration = le32_to_cpu(lease->duration_ms);
1352 long unsigned ttl = from_time + (duration * HZ) / 1000;
1353 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1354
1355 doutc(cl, "%p duration %lu ms ttl %lu\n", dentry, duration, ttl);
1356
1357 /* only track leases on regular dentries */
1358 if (ceph_snap(dir) != CEPH_NOSNAP)
1359 return;
1360
1361 if (mask & CEPH_LEASE_PRIMARY_LINK)
1362 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1363 else
1364 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1365
1366 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1367 if (!(mask & CEPH_LEASE_VALID)) {
1368 __ceph_dentry_dir_lease_touch(di);
1369 return;
1370 }
1371
1372 if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1373 time_before(ttl, di->time))
1374 return; /* we already have a newer lease. */
1375
1376 if (di->lease_session && di->lease_session != session) {
1377 *old_lease_session = di->lease_session;
1378 di->lease_session = NULL;
1379 }
1380
1381 if (!di->lease_session)
1382 di->lease_session = ceph_get_mds_session(session);
1383 di->lease_gen = atomic_read(&session->s_cap_gen);
1384 di->lease_seq = le32_to_cpu(lease->seq);
1385 di->lease_renew_after = half_ttl;
1386 di->lease_renew_from = 0;
1387 di->time = ttl;
1388
1389 __ceph_dentry_lease_touch(di);
1390}
1391
1392static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1393 struct ceph_mds_reply_lease *lease,
1394 struct ceph_mds_session *session,
1395 unsigned long from_time)
1396{
1397 struct ceph_mds_session *old_lease_session = NULL;
1398 spin_lock(&dentry->d_lock);
1399 __update_dentry_lease(dir, dentry, lease, session, from_time,
1400 &old_lease_session);
1401 spin_unlock(&dentry->d_lock);
1402 ceph_put_mds_session(old_lease_session);
1403}
1404
1405/*
1406 * update dentry lease without having parent inode locked
1407 */
1408static void update_dentry_lease_careful(struct dentry *dentry,
1409 struct ceph_mds_reply_lease *lease,
1410 struct ceph_mds_session *session,
1411 unsigned long from_time,
1412 char *dname, u32 dname_len,
1413 struct ceph_vino *pdvino,
1414 struct ceph_vino *ptvino)
1415
1416{
1417 struct inode *dir;
1418 struct ceph_mds_session *old_lease_session = NULL;
1419
1420 spin_lock(&dentry->d_lock);
1421 /* make sure dentry's name matches target */
1422 if (dentry->d_name.len != dname_len ||
1423 memcmp(dentry->d_name.name, dname, dname_len))
1424 goto out_unlock;
1425
1426 dir = d_inode(dentry->d_parent);
1427 /* make sure parent matches dvino */
1428 if (!ceph_ino_compare(dir, pdvino))
1429 goto out_unlock;
1430
1431 /* make sure dentry's inode matches target. NULL ptvino means that
1432 * we expect a negative dentry */
1433 if (ptvino) {
1434 if (d_really_is_negative(dentry))
1435 goto out_unlock;
1436 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1437 goto out_unlock;
1438 } else {
1439 if (d_really_is_positive(dentry))
1440 goto out_unlock;
1441 }
1442
1443 __update_dentry_lease(dir, dentry, lease, session,
1444 from_time, &old_lease_session);
1445out_unlock:
1446 spin_unlock(&dentry->d_lock);
1447 ceph_put_mds_session(old_lease_session);
1448}
1449
1450/*
1451 * splice a dentry to an inode.
1452 * caller must hold directory i_rwsem for this to be safe.
1453 */
1454static int splice_dentry(struct dentry **pdn, struct inode *in)
1455{
1456 struct ceph_client *cl = ceph_inode_to_client(in);
1457 struct dentry *dn = *pdn;
1458 struct dentry *realdn;
1459
1460 BUG_ON(d_inode(dn));
1461
1462 if (S_ISDIR(in->i_mode)) {
1463 /* If inode is directory, d_splice_alias() below will remove
1464 * 'realdn' from its origin parent. We need to ensure that
1465 * origin parent's readdir cache will not reference 'realdn'
1466 */
1467 realdn = d_find_any_alias(in);
1468 if (realdn) {
1469 struct ceph_dentry_info *di = ceph_dentry(realdn);
1470 spin_lock(&realdn->d_lock);
1471
1472 realdn->d_op->d_prune(realdn);
1473
1474 di->time = jiffies;
1475 di->lease_shared_gen = 0;
1476 di->offset = 0;
1477
1478 spin_unlock(&realdn->d_lock);
1479 dput(realdn);
1480 }
1481 }
1482
1483 /* dn must be unhashed */
1484 if (!d_unhashed(dn))
1485 d_drop(dn);
1486 realdn = d_splice_alias(in, dn);
1487 if (IS_ERR(realdn)) {
1488 pr_err_client(cl, "error %ld %p inode %p ino %llx.%llx\n",
1489 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1490 return PTR_ERR(realdn);
1491 }
1492
1493 if (realdn) {
1494 doutc(cl, "dn %p (%d) spliced with %p (%d) inode %p ino %llx.%llx\n",
1495 dn, d_count(dn), realdn, d_count(realdn),
1496 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1497 dput(dn);
1498 *pdn = realdn;
1499 } else {
1500 BUG_ON(!ceph_dentry(dn));
1501 doutc(cl, "dn %p attached to %p ino %llx.%llx\n", dn,
1502 d_inode(dn), ceph_vinop(d_inode(dn)));
1503 }
1504 return 0;
1505}
1506
1507/*
1508 * Incorporate results into the local cache. This is either just
1509 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1510 * after a lookup).
1511 *
1512 * A reply may contain
1513 * a directory inode along with a dentry.
1514 * and/or a target inode
1515 *
1516 * Called with snap_rwsem (read).
1517 */
1518int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1519{
1520 struct ceph_mds_session *session = req->r_session;
1521 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1522 struct inode *in = NULL;
1523 struct ceph_vino tvino, dvino;
1524 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1525 struct ceph_client *cl = fsc->client;
1526 int err = 0;
1527
1528 doutc(cl, "%p is_dentry %d is_target %d\n", req,
1529 rinfo->head->is_dentry, rinfo->head->is_target);
1530
1531 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1532 doutc(cl, "reply is empty!\n");
1533 if (rinfo->head->result == 0 && req->r_parent)
1534 ceph_invalidate_dir_request(req);
1535 return 0;
1536 }
1537
1538 if (rinfo->head->is_dentry) {
1539 struct inode *dir = req->r_parent;
1540
1541 if (dir) {
1542 err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1543 rinfo->dirfrag, session, -1,
1544 &req->r_caps_reservation);
1545 if (err < 0)
1546 goto done;
1547 } else {
1548 WARN_ON_ONCE(1);
1549 }
1550
1551 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1552 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1553 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1554 bool is_nokey = false;
1555 struct qstr dname;
1556 struct dentry *dn, *parent;
1557 struct fscrypt_str oname = FSTR_INIT(NULL, 0);
1558 struct ceph_fname fname = { .dir = dir,
1559 .name = rinfo->dname,
1560 .ctext = rinfo->altname,
1561 .name_len = rinfo->dname_len,
1562 .ctext_len = rinfo->altname_len };
1563
1564 BUG_ON(!rinfo->head->is_target);
1565 BUG_ON(req->r_dentry);
1566
1567 parent = d_find_any_alias(dir);
1568 BUG_ON(!parent);
1569
1570 err = ceph_fname_alloc_buffer(dir, &oname);
1571 if (err < 0) {
1572 dput(parent);
1573 goto done;
1574 }
1575
1576 err = ceph_fname_to_usr(&fname, NULL, &oname, &is_nokey);
1577 if (err < 0) {
1578 dput(parent);
1579 ceph_fname_free_buffer(dir, &oname);
1580 goto done;
1581 }
1582 dname.name = oname.name;
1583 dname.len = oname.len;
1584 dname.hash = full_name_hash(parent, dname.name, dname.len);
1585 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1586 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1587retry_lookup:
1588 dn = d_lookup(parent, &dname);
1589 doutc(cl, "d_lookup on parent=%p name=%.*s got %p\n",
1590 parent, dname.len, dname.name, dn);
1591
1592 if (!dn) {
1593 dn = d_alloc(parent, &dname);
1594 doutc(cl, "d_alloc %p '%.*s' = %p\n", parent,
1595 dname.len, dname.name, dn);
1596 if (!dn) {
1597 dput(parent);
1598 ceph_fname_free_buffer(dir, &oname);
1599 err = -ENOMEM;
1600 goto done;
1601 }
1602 if (is_nokey) {
1603 spin_lock(&dn->d_lock);
1604 dn->d_flags |= DCACHE_NOKEY_NAME;
1605 spin_unlock(&dn->d_lock);
1606 }
1607 err = 0;
1608 } else if (d_really_is_positive(dn) &&
1609 (ceph_ino(d_inode(dn)) != tvino.ino ||
1610 ceph_snap(d_inode(dn)) != tvino.snap)) {
1611 doutc(cl, " dn %p points to wrong inode %p\n",
1612 dn, d_inode(dn));
1613 ceph_dir_clear_ordered(dir);
1614 d_delete(dn);
1615 dput(dn);
1616 goto retry_lookup;
1617 }
1618 ceph_fname_free_buffer(dir, &oname);
1619
1620 req->r_dentry = dn;
1621 dput(parent);
1622 }
1623 }
1624
1625 if (rinfo->head->is_target) {
1626 /* Should be filled in by handle_reply */
1627 BUG_ON(!req->r_target_inode);
1628
1629 in = req->r_target_inode;
1630 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1631 NULL, session,
1632 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1633 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1634 rinfo->head->result == 0) ? req->r_fmode : -1,
1635 &req->r_caps_reservation);
1636 if (err < 0) {
1637 pr_err_client(cl, "badness %p %llx.%llx\n", in,
1638 ceph_vinop(in));
1639 req->r_target_inode = NULL;
1640 if (in->i_state & I_NEW)
1641 discard_new_inode(in);
1642 else
1643 iput(in);
1644 goto done;
1645 }
1646 if (in->i_state & I_NEW)
1647 unlock_new_inode(in);
1648 }
1649
1650 /*
1651 * ignore null lease/binding on snapdir ENOENT, or else we
1652 * will have trouble splicing in the virtual snapdir later
1653 */
1654 if (rinfo->head->is_dentry &&
1655 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1656 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1657 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1658 fsc->mount_options->snapdir_name,
1659 req->r_dentry->d_name.len))) {
1660 /*
1661 * lookup link rename : null -> possibly existing inode
1662 * mknod symlink mkdir : null -> new inode
1663 * unlink : linked -> null
1664 */
1665 struct inode *dir = req->r_parent;
1666 struct dentry *dn = req->r_dentry;
1667 bool have_dir_cap, have_lease;
1668
1669 BUG_ON(!dn);
1670 BUG_ON(!dir);
1671 BUG_ON(d_inode(dn->d_parent) != dir);
1672
1673 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1674 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1675
1676 BUG_ON(ceph_ino(dir) != dvino.ino);
1677 BUG_ON(ceph_snap(dir) != dvino.snap);
1678
1679 /* do we have a lease on the whole dir? */
1680 have_dir_cap =
1681 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1682 CEPH_CAP_FILE_SHARED);
1683
1684 /* do we have a dn lease? */
1685 have_lease = have_dir_cap ||
1686 le32_to_cpu(rinfo->dlease->duration_ms);
1687 if (!have_lease)
1688 doutc(cl, "no dentry lease or dir cap\n");
1689
1690 /* rename? */
1691 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1692 struct inode *olddir = req->r_old_dentry_dir;
1693 BUG_ON(!olddir);
1694
1695 doutc(cl, " src %p '%pd' dst %p '%pd'\n",
1696 req->r_old_dentry, req->r_old_dentry, dn, dn);
1697 doutc(cl, "doing d_move %p -> %p\n", req->r_old_dentry, dn);
1698
1699 /* d_move screws up sibling dentries' offsets */
1700 ceph_dir_clear_ordered(dir);
1701 ceph_dir_clear_ordered(olddir);
1702
1703 d_move(req->r_old_dentry, dn);
1704 doutc(cl, " src %p '%pd' dst %p '%pd'\n",
1705 req->r_old_dentry, req->r_old_dentry, dn, dn);
1706
1707 /* ensure target dentry is invalidated, despite
1708 rehashing bug in vfs_rename_dir */
1709 ceph_invalidate_dentry_lease(dn);
1710
1711 doutc(cl, "dn %p gets new offset %lld\n",
1712 req->r_old_dentry,
1713 ceph_dentry(req->r_old_dentry)->offset);
1714
1715 /* swap r_dentry and r_old_dentry in case that
1716 * splice_dentry() gets called later. This is safe
1717 * because no other place will use them */
1718 req->r_dentry = req->r_old_dentry;
1719 req->r_old_dentry = dn;
1720 dn = req->r_dentry;
1721 }
1722
1723 /* null dentry? */
1724 if (!rinfo->head->is_target) {
1725 doutc(cl, "null dentry\n");
1726 if (d_really_is_positive(dn)) {
1727 doutc(cl, "d_delete %p\n", dn);
1728 ceph_dir_clear_ordered(dir);
1729 d_delete(dn);
1730 } else if (have_lease) {
1731 if (d_unhashed(dn))
1732 d_add(dn, NULL);
1733 }
1734
1735 if (!d_unhashed(dn) && have_lease)
1736 update_dentry_lease(dir, dn,
1737 rinfo->dlease, session,
1738 req->r_request_started);
1739 goto done;
1740 }
1741
1742 /* attach proper inode */
1743 if (d_really_is_negative(dn)) {
1744 ceph_dir_clear_ordered(dir);
1745 ihold(in);
1746 err = splice_dentry(&req->r_dentry, in);
1747 if (err < 0)
1748 goto done;
1749 dn = req->r_dentry; /* may have spliced */
1750 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1751 doutc(cl, " %p links to %p %llx.%llx, not %llx.%llx\n",
1752 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1753 ceph_vinop(in));
1754 d_invalidate(dn);
1755 have_lease = false;
1756 }
1757
1758 if (have_lease) {
1759 update_dentry_lease(dir, dn,
1760 rinfo->dlease, session,
1761 req->r_request_started);
1762 }
1763 doutc(cl, " final dn %p\n", dn);
1764 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1765 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1766 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1767 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1768 struct inode *dir = req->r_parent;
1769
1770 /* fill out a snapdir LOOKUPSNAP dentry */
1771 BUG_ON(!dir);
1772 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1773 BUG_ON(!req->r_dentry);
1774 doutc(cl, " linking snapped dir %p to dn %p\n", in,
1775 req->r_dentry);
1776 ceph_dir_clear_ordered(dir);
1777 ihold(in);
1778 err = splice_dentry(&req->r_dentry, in);
1779 if (err < 0)
1780 goto done;
1781 } else if (rinfo->head->is_dentry && req->r_dentry) {
1782 /* parent inode is not locked, be careful */
1783 struct ceph_vino *ptvino = NULL;
1784 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1785 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1786 if (rinfo->head->is_target) {
1787 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1788 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1789 ptvino = &tvino;
1790 }
1791 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1792 session, req->r_request_started,
1793 rinfo->dname, rinfo->dname_len,
1794 &dvino, ptvino);
1795 }
1796done:
1797 doutc(cl, "done err=%d\n", err);
1798 return err;
1799}
1800
1801/*
1802 * Prepopulate our cache with readdir results, leases, etc.
1803 */
1804static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1805 struct ceph_mds_session *session)
1806{
1807 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1808 struct ceph_client *cl = session->s_mdsc->fsc->client;
1809 int i, err = 0;
1810
1811 for (i = 0; i < rinfo->dir_nr; i++) {
1812 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1813 struct ceph_vino vino;
1814 struct inode *in;
1815 int rc;
1816
1817 vino.ino = le64_to_cpu(rde->inode.in->ino);
1818 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1819
1820 in = ceph_get_inode(req->r_dentry->d_sb, vino, NULL);
1821 if (IS_ERR(in)) {
1822 err = PTR_ERR(in);
1823 doutc(cl, "badness got %d\n", err);
1824 continue;
1825 }
1826 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1827 -1, &req->r_caps_reservation);
1828 if (rc < 0) {
1829 pr_err_client(cl, "inode badness on %p got %d\n", in,
1830 rc);
1831 err = rc;
1832 if (in->i_state & I_NEW) {
1833 ihold(in);
1834 discard_new_inode(in);
1835 }
1836 } else if (in->i_state & I_NEW) {
1837 unlock_new_inode(in);
1838 }
1839
1840 iput(in);
1841 }
1842
1843 return err;
1844}
1845
1846void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1847{
1848 if (ctl->page) {
1849 kunmap(ctl->page);
1850 put_page(ctl->page);
1851 ctl->page = NULL;
1852 }
1853}
1854
1855static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1856 struct ceph_readdir_cache_control *ctl,
1857 struct ceph_mds_request *req)
1858{
1859 struct ceph_client *cl = ceph_inode_to_client(dir);
1860 struct ceph_inode_info *ci = ceph_inode(dir);
1861 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1862 unsigned idx = ctl->index % nsize;
1863 pgoff_t pgoff = ctl->index / nsize;
1864
1865 if (!ctl->page || pgoff != ctl->page->index) {
1866 ceph_readdir_cache_release(ctl);
1867 if (idx == 0)
1868 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1869 else
1870 ctl->page = find_lock_page(&dir->i_data, pgoff);
1871 if (!ctl->page) {
1872 ctl->index = -1;
1873 return idx == 0 ? -ENOMEM : 0;
1874 }
1875 /* reading/filling the cache are serialized by
1876 * i_rwsem, no need to use page lock */
1877 unlock_page(ctl->page);
1878 ctl->dentries = kmap(ctl->page);
1879 if (idx == 0)
1880 memset(ctl->dentries, 0, PAGE_SIZE);
1881 }
1882
1883 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1884 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1885 doutc(cl, "dn %p idx %d\n", dn, ctl->index);
1886 ctl->dentries[idx] = dn;
1887 ctl->index++;
1888 } else {
1889 doutc(cl, "disable readdir cache\n");
1890 ctl->index = -1;
1891 }
1892 return 0;
1893}
1894
1895int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1896 struct ceph_mds_session *session)
1897{
1898 struct dentry *parent = req->r_dentry;
1899 struct inode *inode = d_inode(parent);
1900 struct ceph_inode_info *ci = ceph_inode(inode);
1901 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1902 struct ceph_client *cl = session->s_mdsc->fsc->client;
1903 struct qstr dname;
1904 struct dentry *dn;
1905 struct inode *in;
1906 int err = 0, skipped = 0, ret, i;
1907 u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1908 u32 last_hash = 0;
1909 u32 fpos_offset;
1910 struct ceph_readdir_cache_control cache_ctl = {};
1911
1912 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1913 return readdir_prepopulate_inodes_only(req, session);
1914
1915 if (rinfo->hash_order) {
1916 if (req->r_path2) {
1917 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1918 req->r_path2,
1919 strlen(req->r_path2));
1920 last_hash = ceph_frag_value(last_hash);
1921 } else if (rinfo->offset_hash) {
1922 /* mds understands offset_hash */
1923 WARN_ON_ONCE(req->r_readdir_offset != 2);
1924 last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1925 }
1926 }
1927
1928 if (rinfo->dir_dir &&
1929 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1930 doutc(cl, "got new frag %x -> %x\n", frag,
1931 le32_to_cpu(rinfo->dir_dir->frag));
1932 frag = le32_to_cpu(rinfo->dir_dir->frag);
1933 if (!rinfo->hash_order)
1934 req->r_readdir_offset = 2;
1935 }
1936
1937 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1938 doutc(cl, "%d items under SNAPDIR dn %p\n",
1939 rinfo->dir_nr, parent);
1940 } else {
1941 doutc(cl, "%d items under dn %p\n", rinfo->dir_nr, parent);
1942 if (rinfo->dir_dir)
1943 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1944
1945 if (ceph_frag_is_leftmost(frag) &&
1946 req->r_readdir_offset == 2 &&
1947 !(rinfo->hash_order && last_hash)) {
1948 /* note dir version at start of readdir so we can
1949 * tell if any dentries get dropped */
1950 req->r_dir_release_cnt =
1951 atomic64_read(&ci->i_release_count);
1952 req->r_dir_ordered_cnt =
1953 atomic64_read(&ci->i_ordered_count);
1954 req->r_readdir_cache_idx = 0;
1955 }
1956 }
1957
1958 cache_ctl.index = req->r_readdir_cache_idx;
1959 fpos_offset = req->r_readdir_offset;
1960
1961 /* FIXME: release caps/leases if error occurs */
1962 for (i = 0; i < rinfo->dir_nr; i++) {
1963 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1964 struct ceph_vino tvino;
1965
1966 dname.name = rde->name;
1967 dname.len = rde->name_len;
1968 dname.hash = full_name_hash(parent, dname.name, dname.len);
1969
1970 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1971 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1972
1973 if (rinfo->hash_order) {
1974 u32 hash = ceph_frag_value(rde->raw_hash);
1975 if (hash != last_hash)
1976 fpos_offset = 2;
1977 last_hash = hash;
1978 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1979 } else {
1980 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1981 }
1982
1983retry_lookup:
1984 dn = d_lookup(parent, &dname);
1985 doutc(cl, "d_lookup on parent=%p name=%.*s got %p\n",
1986 parent, dname.len, dname.name, dn);
1987
1988 if (!dn) {
1989 dn = d_alloc(parent, &dname);
1990 doutc(cl, "d_alloc %p '%.*s' = %p\n", parent,
1991 dname.len, dname.name, dn);
1992 if (!dn) {
1993 doutc(cl, "d_alloc badness\n");
1994 err = -ENOMEM;
1995 goto out;
1996 }
1997 if (rde->is_nokey) {
1998 spin_lock(&dn->d_lock);
1999 dn->d_flags |= DCACHE_NOKEY_NAME;
2000 spin_unlock(&dn->d_lock);
2001 }
2002 } else if (d_really_is_positive(dn) &&
2003 (ceph_ino(d_inode(dn)) != tvino.ino ||
2004 ceph_snap(d_inode(dn)) != tvino.snap)) {
2005 struct ceph_dentry_info *di = ceph_dentry(dn);
2006 doutc(cl, " dn %p points to wrong inode %p\n",
2007 dn, d_inode(dn));
2008
2009 spin_lock(&dn->d_lock);
2010 if (di->offset > 0 &&
2011 di->lease_shared_gen ==
2012 atomic_read(&ci->i_shared_gen)) {
2013 __ceph_dir_clear_ordered(ci);
2014 di->offset = 0;
2015 }
2016 spin_unlock(&dn->d_lock);
2017
2018 d_delete(dn);
2019 dput(dn);
2020 goto retry_lookup;
2021 }
2022
2023 /* inode */
2024 if (d_really_is_positive(dn)) {
2025 in = d_inode(dn);
2026 } else {
2027 in = ceph_get_inode(parent->d_sb, tvino, NULL);
2028 if (IS_ERR(in)) {
2029 doutc(cl, "new_inode badness\n");
2030 d_drop(dn);
2031 dput(dn);
2032 err = PTR_ERR(in);
2033 goto out;
2034 }
2035 }
2036
2037 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
2038 -1, &req->r_caps_reservation);
2039 if (ret < 0) {
2040 pr_err_client(cl, "badness on %p %llx.%llx\n", in,
2041 ceph_vinop(in));
2042 if (d_really_is_negative(dn)) {
2043 if (in->i_state & I_NEW) {
2044 ihold(in);
2045 discard_new_inode(in);
2046 }
2047 iput(in);
2048 }
2049 d_drop(dn);
2050 err = ret;
2051 goto next_item;
2052 }
2053 if (in->i_state & I_NEW)
2054 unlock_new_inode(in);
2055
2056 if (d_really_is_negative(dn)) {
2057 if (ceph_security_xattr_deadlock(in)) {
2058 doutc(cl, " skip splicing dn %p to inode %p"
2059 " (security xattr deadlock)\n", dn, in);
2060 iput(in);
2061 skipped++;
2062 goto next_item;
2063 }
2064
2065 err = splice_dentry(&dn, in);
2066 if (err < 0)
2067 goto next_item;
2068 }
2069
2070 ceph_dentry(dn)->offset = rde->offset;
2071
2072 update_dentry_lease(d_inode(parent), dn,
2073 rde->lease, req->r_session,
2074 req->r_request_started);
2075
2076 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
2077 ret = fill_readdir_cache(d_inode(parent), dn,
2078 &cache_ctl, req);
2079 if (ret < 0)
2080 err = ret;
2081 }
2082next_item:
2083 dput(dn);
2084 }
2085out:
2086 if (err == 0 && skipped == 0) {
2087 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
2088 req->r_readdir_cache_idx = cache_ctl.index;
2089 }
2090 ceph_readdir_cache_release(&cache_ctl);
2091 doutc(cl, "done\n");
2092 return err;
2093}
2094
2095bool ceph_inode_set_size(struct inode *inode, loff_t size)
2096{
2097 struct ceph_client *cl = ceph_inode_to_client(inode);
2098 struct ceph_inode_info *ci = ceph_inode(inode);
2099 bool ret;
2100
2101 spin_lock(&ci->i_ceph_lock);
2102 doutc(cl, "set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
2103 i_size_write(inode, size);
2104 ceph_fscache_update(inode);
2105 inode->i_blocks = calc_inode_blocks(size);
2106
2107 ret = __ceph_should_report_size(ci);
2108
2109 spin_unlock(&ci->i_ceph_lock);
2110
2111 return ret;
2112}
2113
2114void ceph_queue_inode_work(struct inode *inode, int work_bit)
2115{
2116 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2117 struct ceph_client *cl = fsc->client;
2118 struct ceph_inode_info *ci = ceph_inode(inode);
2119 set_bit(work_bit, &ci->i_work_mask);
2120
2121 ihold(inode);
2122 if (queue_work(fsc->inode_wq, &ci->i_work)) {
2123 doutc(cl, "%p %llx.%llx mask=%lx\n", inode,
2124 ceph_vinop(inode), ci->i_work_mask);
2125 } else {
2126 doutc(cl, "%p %llx.%llx already queued, mask=%lx\n",
2127 inode, ceph_vinop(inode), ci->i_work_mask);
2128 iput(inode);
2129 }
2130}
2131
2132static void ceph_do_invalidate_pages(struct inode *inode)
2133{
2134 struct ceph_client *cl = ceph_inode_to_client(inode);
2135 struct ceph_inode_info *ci = ceph_inode(inode);
2136 u32 orig_gen;
2137 int check = 0;
2138
2139 ceph_fscache_invalidate(inode, false);
2140
2141 mutex_lock(&ci->i_truncate_mutex);
2142
2143 if (ceph_inode_is_shutdown(inode)) {
2144 pr_warn_ratelimited_client(cl,
2145 "%p %llx.%llx is shut down\n", inode,
2146 ceph_vinop(inode));
2147 mapping_set_error(inode->i_mapping, -EIO);
2148 truncate_pagecache(inode, 0);
2149 mutex_unlock(&ci->i_truncate_mutex);
2150 goto out;
2151 }
2152
2153 spin_lock(&ci->i_ceph_lock);
2154 doutc(cl, "%p %llx.%llx gen %d revoking %d\n", inode,
2155 ceph_vinop(inode), ci->i_rdcache_gen, ci->i_rdcache_revoking);
2156 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
2157 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2158 check = 1;
2159 spin_unlock(&ci->i_ceph_lock);
2160 mutex_unlock(&ci->i_truncate_mutex);
2161 goto out;
2162 }
2163 orig_gen = ci->i_rdcache_gen;
2164 spin_unlock(&ci->i_ceph_lock);
2165
2166 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
2167 pr_err_client(cl, "invalidate_inode_pages2 %llx.%llx failed\n",
2168 ceph_vinop(inode));
2169 }
2170
2171 spin_lock(&ci->i_ceph_lock);
2172 if (orig_gen == ci->i_rdcache_gen &&
2173 orig_gen == ci->i_rdcache_revoking) {
2174 doutc(cl, "%p %llx.%llx gen %d successful\n", inode,
2175 ceph_vinop(inode), ci->i_rdcache_gen);
2176 ci->i_rdcache_revoking--;
2177 check = 1;
2178 } else {
2179 doutc(cl, "%p %llx.%llx gen %d raced, now %d revoking %d\n",
2180 inode, ceph_vinop(inode), orig_gen, ci->i_rdcache_gen,
2181 ci->i_rdcache_revoking);
2182 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2183 check = 1;
2184 }
2185 spin_unlock(&ci->i_ceph_lock);
2186 mutex_unlock(&ci->i_truncate_mutex);
2187out:
2188 if (check)
2189 ceph_check_caps(ci, 0);
2190}
2191
2192/*
2193 * Make sure any pending truncation is applied before doing anything
2194 * that may depend on it.
2195 */
2196void __ceph_do_pending_vmtruncate(struct inode *inode)
2197{
2198 struct ceph_client *cl = ceph_inode_to_client(inode);
2199 struct ceph_inode_info *ci = ceph_inode(inode);
2200 u64 to;
2201 int wrbuffer_refs, finish = 0;
2202
2203 mutex_lock(&ci->i_truncate_mutex);
2204retry:
2205 spin_lock(&ci->i_ceph_lock);
2206 if (ci->i_truncate_pending == 0) {
2207 doutc(cl, "%p %llx.%llx none pending\n", inode,
2208 ceph_vinop(inode));
2209 spin_unlock(&ci->i_ceph_lock);
2210 mutex_unlock(&ci->i_truncate_mutex);
2211 return;
2212 }
2213
2214 /*
2215 * make sure any dirty snapped pages are flushed before we
2216 * possibly truncate them.. so write AND block!
2217 */
2218 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
2219 spin_unlock(&ci->i_ceph_lock);
2220 doutc(cl, "%p %llx.%llx flushing snaps first\n", inode,
2221 ceph_vinop(inode));
2222 filemap_write_and_wait_range(&inode->i_data, 0,
2223 inode->i_sb->s_maxbytes);
2224 goto retry;
2225 }
2226
2227 /* there should be no reader or writer */
2228 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
2229
2230 to = ci->i_truncate_pagecache_size;
2231 wrbuffer_refs = ci->i_wrbuffer_ref;
2232 doutc(cl, "%p %llx.%llx (%d) to %lld\n", inode, ceph_vinop(inode),
2233 ci->i_truncate_pending, to);
2234 spin_unlock(&ci->i_ceph_lock);
2235
2236 ceph_fscache_resize(inode, to);
2237 truncate_pagecache(inode, to);
2238
2239 spin_lock(&ci->i_ceph_lock);
2240 if (to == ci->i_truncate_pagecache_size) {
2241 ci->i_truncate_pending = 0;
2242 finish = 1;
2243 }
2244 spin_unlock(&ci->i_ceph_lock);
2245 if (!finish)
2246 goto retry;
2247
2248 mutex_unlock(&ci->i_truncate_mutex);
2249
2250 if (wrbuffer_refs == 0)
2251 ceph_check_caps(ci, 0);
2252
2253 wake_up_all(&ci->i_cap_wq);
2254}
2255
2256static void ceph_inode_work(struct work_struct *work)
2257{
2258 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
2259 i_work);
2260 struct inode *inode = &ci->netfs.inode;
2261 struct ceph_client *cl = ceph_inode_to_client(inode);
2262
2263 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
2264 doutc(cl, "writeback %p %llx.%llx\n", inode, ceph_vinop(inode));
2265 filemap_fdatawrite(&inode->i_data);
2266 }
2267 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2268 ceph_do_invalidate_pages(inode);
2269
2270 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2271 __ceph_do_pending_vmtruncate(inode);
2272
2273 if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
2274 ceph_check_caps(ci, 0);
2275
2276 if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
2277 ceph_flush_snaps(ci, NULL);
2278
2279 iput(inode);
2280}
2281
2282static const char *ceph_encrypted_get_link(struct dentry *dentry,
2283 struct inode *inode,
2284 struct delayed_call *done)
2285{
2286 struct ceph_inode_info *ci = ceph_inode(inode);
2287
2288 if (!dentry)
2289 return ERR_PTR(-ECHILD);
2290
2291 return fscrypt_get_symlink(inode, ci->i_symlink, i_size_read(inode),
2292 done);
2293}
2294
2295static int ceph_encrypted_symlink_getattr(struct mnt_idmap *idmap,
2296 const struct path *path,
2297 struct kstat *stat, u32 request_mask,
2298 unsigned int query_flags)
2299{
2300 int ret;
2301
2302 ret = ceph_getattr(idmap, path, stat, request_mask, query_flags);
2303 if (ret)
2304 return ret;
2305 return fscrypt_symlink_getattr(path, stat);
2306}
2307
2308/*
2309 * symlinks
2310 */
2311static const struct inode_operations ceph_symlink_iops = {
2312 .get_link = simple_get_link,
2313 .setattr = ceph_setattr,
2314 .getattr = ceph_getattr,
2315 .listxattr = ceph_listxattr,
2316};
2317
2318static const struct inode_operations ceph_encrypted_symlink_iops = {
2319 .get_link = ceph_encrypted_get_link,
2320 .setattr = ceph_setattr,
2321 .getattr = ceph_encrypted_symlink_getattr,
2322 .listxattr = ceph_listxattr,
2323};
2324
2325/*
2326 * Transfer the encrypted last block to the MDS and the MDS
2327 * will help update it when truncating a smaller size.
2328 *
2329 * We don't support a PAGE_SIZE that is smaller than the
2330 * CEPH_FSCRYPT_BLOCK_SIZE.
2331 */
2332static int fill_fscrypt_truncate(struct inode *inode,
2333 struct ceph_mds_request *req,
2334 struct iattr *attr)
2335{
2336 struct ceph_client *cl = ceph_inode_to_client(inode);
2337 struct ceph_inode_info *ci = ceph_inode(inode);
2338 int boff = attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE;
2339 loff_t pos, orig_pos = round_down(attr->ia_size,
2340 CEPH_FSCRYPT_BLOCK_SIZE);
2341 u64 block = orig_pos >> CEPH_FSCRYPT_BLOCK_SHIFT;
2342 struct ceph_pagelist *pagelist = NULL;
2343 struct kvec iov = {0};
2344 struct iov_iter iter;
2345 struct page *page = NULL;
2346 struct ceph_fscrypt_truncate_size_header header;
2347 int retry_op = 0;
2348 int len = CEPH_FSCRYPT_BLOCK_SIZE;
2349 loff_t i_size = i_size_read(inode);
2350 int got, ret, issued;
2351 u64 objver;
2352
2353 ret = __ceph_get_caps(inode, NULL, CEPH_CAP_FILE_RD, 0, -1, &got);
2354 if (ret < 0)
2355 return ret;
2356
2357 issued = __ceph_caps_issued(ci, NULL);
2358
2359 doutc(cl, "size %lld -> %lld got cap refs on %s, issued %s\n",
2360 i_size, attr->ia_size, ceph_cap_string(got),
2361 ceph_cap_string(issued));
2362
2363 /* Try to writeback the dirty pagecaches */
2364 if (issued & (CEPH_CAP_FILE_BUFFER)) {
2365 loff_t lend = orig_pos + CEPH_FSCRYPT_BLOCK_SHIFT - 1;
2366
2367 ret = filemap_write_and_wait_range(inode->i_mapping,
2368 orig_pos, lend);
2369 if (ret < 0)
2370 goto out;
2371 }
2372
2373 page = __page_cache_alloc(GFP_KERNEL);
2374 if (page == NULL) {
2375 ret = -ENOMEM;
2376 goto out;
2377 }
2378
2379 pagelist = ceph_pagelist_alloc(GFP_KERNEL);
2380 if (!pagelist) {
2381 ret = -ENOMEM;
2382 goto out;
2383 }
2384
2385 iov.iov_base = kmap_local_page(page);
2386 iov.iov_len = len;
2387 iov_iter_kvec(&iter, READ, &iov, 1, len);
2388
2389 pos = orig_pos;
2390 ret = __ceph_sync_read(inode, &pos, &iter, &retry_op, &objver);
2391 if (ret < 0)
2392 goto out;
2393
2394 /* Insert the header first */
2395 header.ver = 1;
2396 header.compat = 1;
2397 header.change_attr = cpu_to_le64(inode_peek_iversion_raw(inode));
2398
2399 /*
2400 * Always set the block_size to CEPH_FSCRYPT_BLOCK_SIZE,
2401 * because in MDS it may need this to do the truncate.
2402 */
2403 header.block_size = cpu_to_le32(CEPH_FSCRYPT_BLOCK_SIZE);
2404
2405 /*
2406 * If we hit a hole here, we should just skip filling
2407 * the fscrypt for the request, because once the fscrypt
2408 * is enabled, the file will be split into many blocks
2409 * with the size of CEPH_FSCRYPT_BLOCK_SIZE, if there
2410 * has a hole, the hole size should be multiple of block
2411 * size.
2412 *
2413 * If the Rados object doesn't exist, it will be set to 0.
2414 */
2415 if (!objver) {
2416 doutc(cl, "hit hole, ppos %lld < size %lld\n", pos, i_size);
2417
2418 header.data_len = cpu_to_le32(8 + 8 + 4);
2419 header.file_offset = 0;
2420 ret = 0;
2421 } else {
2422 header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
2423 header.file_offset = cpu_to_le64(orig_pos);
2424
2425 doutc(cl, "encrypt block boff/bsize %d/%lu\n", boff,
2426 CEPH_FSCRYPT_BLOCK_SIZE);
2427
2428 /* truncate and zero out the extra contents for the last block */
2429 memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
2430
2431 /* encrypt the last block */
2432 ret = ceph_fscrypt_encrypt_block_inplace(inode, page,
2433 CEPH_FSCRYPT_BLOCK_SIZE,
2434 0, block,
2435 GFP_KERNEL);
2436 if (ret)
2437 goto out;
2438 }
2439
2440 /* Insert the header */
2441 ret = ceph_pagelist_append(pagelist, &header, sizeof(header));
2442 if (ret)
2443 goto out;
2444
2445 if (header.block_size) {
2446 /* Append the last block contents to pagelist */
2447 ret = ceph_pagelist_append(pagelist, iov.iov_base,
2448 CEPH_FSCRYPT_BLOCK_SIZE);
2449 if (ret)
2450 goto out;
2451 }
2452 req->r_pagelist = pagelist;
2453out:
2454 doutc(cl, "%p %llx.%llx size dropping cap refs on %s\n", inode,
2455 ceph_vinop(inode), ceph_cap_string(got));
2456 ceph_put_cap_refs(ci, got);
2457 if (iov.iov_base)
2458 kunmap_local(iov.iov_base);
2459 if (page)
2460 __free_pages(page, 0);
2461 if (ret && pagelist)
2462 ceph_pagelist_release(pagelist);
2463 return ret;
2464}
2465
2466int __ceph_setattr(struct mnt_idmap *idmap, struct inode *inode,
2467 struct iattr *attr, struct ceph_iattr *cia)
2468{
2469 struct ceph_inode_info *ci = ceph_inode(inode);
2470 unsigned int ia_valid = attr->ia_valid;
2471 struct ceph_mds_request *req;
2472 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
2473 struct ceph_client *cl = ceph_inode_to_client(inode);
2474 struct ceph_cap_flush *prealloc_cf;
2475 loff_t isize = i_size_read(inode);
2476 int issued;
2477 int release = 0, dirtied = 0;
2478 int mask = 0;
2479 int err = 0;
2480 int inode_dirty_flags = 0;
2481 bool lock_snap_rwsem = false;
2482 bool fill_fscrypt;
2483 int truncate_retry = 20; /* The RMW will take around 50ms */
2484 struct dentry *dentry;
2485 char *path;
2486 int pathlen;
2487 u64 pathbase;
2488 bool do_sync = false;
2489
2490 dentry = d_find_alias(inode);
2491 if (!dentry) {
2492 do_sync = true;
2493 } else {
2494 path = ceph_mdsc_build_path(mdsc, dentry, &pathlen, &pathbase, 0);
2495 if (IS_ERR(path)) {
2496 do_sync = true;
2497 err = 0;
2498 } else {
2499 err = ceph_mds_check_access(mdsc, path, MAY_WRITE);
2500 }
2501 ceph_mdsc_free_path(path, pathlen);
2502 dput(dentry);
2503
2504 /* For none EACCES cases will let the MDS do the mds auth check */
2505 if (err == -EACCES) {
2506 return err;
2507 } else if (err < 0) {
2508 do_sync = true;
2509 err = 0;
2510 }
2511 }
2512
2513retry:
2514 prealloc_cf = ceph_alloc_cap_flush();
2515 if (!prealloc_cf)
2516 return -ENOMEM;
2517
2518 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2519 USE_AUTH_MDS);
2520 if (IS_ERR(req)) {
2521 ceph_free_cap_flush(prealloc_cf);
2522 return PTR_ERR(req);
2523 }
2524
2525 fill_fscrypt = false;
2526 spin_lock(&ci->i_ceph_lock);
2527 issued = __ceph_caps_issued(ci, NULL);
2528
2529 if (!ci->i_head_snapc &&
2530 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2531 lock_snap_rwsem = true;
2532 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2533 spin_unlock(&ci->i_ceph_lock);
2534 down_read(&mdsc->snap_rwsem);
2535 spin_lock(&ci->i_ceph_lock);
2536 issued = __ceph_caps_issued(ci, NULL);
2537 }
2538 }
2539
2540 doutc(cl, "%p %llx.%llx issued %s\n", inode, ceph_vinop(inode),
2541 ceph_cap_string(issued));
2542#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
2543 if (cia && cia->fscrypt_auth) {
2544 u32 len = ceph_fscrypt_auth_len(cia->fscrypt_auth);
2545
2546 if (len > sizeof(*cia->fscrypt_auth)) {
2547 err = -EINVAL;
2548 spin_unlock(&ci->i_ceph_lock);
2549 goto out;
2550 }
2551
2552 doutc(cl, "%p %llx.%llx fscrypt_auth len %u to %u)\n", inode,
2553 ceph_vinop(inode), ci->fscrypt_auth_len, len);
2554
2555 /* It should never be re-set once set */
2556 WARN_ON_ONCE(ci->fscrypt_auth);
2557
2558 if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2559 dirtied |= CEPH_CAP_AUTH_EXCL;
2560 kfree(ci->fscrypt_auth);
2561 ci->fscrypt_auth = (u8 *)cia->fscrypt_auth;
2562 ci->fscrypt_auth_len = len;
2563 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2564 ci->fscrypt_auth_len != len ||
2565 memcmp(ci->fscrypt_auth, cia->fscrypt_auth, len)) {
2566 req->r_fscrypt_auth = cia->fscrypt_auth;
2567 mask |= CEPH_SETATTR_FSCRYPT_AUTH;
2568 release |= CEPH_CAP_AUTH_SHARED;
2569 }
2570 cia->fscrypt_auth = NULL;
2571 }
2572#else
2573 if (cia && cia->fscrypt_auth) {
2574 err = -EINVAL;
2575 spin_unlock(&ci->i_ceph_lock);
2576 goto out;
2577 }
2578#endif /* CONFIG_FS_ENCRYPTION */
2579
2580 if (ia_valid & ATTR_UID) {
2581 kuid_t fsuid = from_vfsuid(idmap, i_user_ns(inode), attr->ia_vfsuid);
2582
2583 doutc(cl, "%p %llx.%llx uid %d -> %d\n", inode,
2584 ceph_vinop(inode),
2585 from_kuid(&init_user_ns, inode->i_uid),
2586 from_kuid(&init_user_ns, attr->ia_uid));
2587 if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2588 inode->i_uid = fsuid;
2589 dirtied |= CEPH_CAP_AUTH_EXCL;
2590 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2591 !uid_eq(fsuid, inode->i_uid)) {
2592 req->r_args.setattr.uid = cpu_to_le32(
2593 from_kuid(&init_user_ns, fsuid));
2594 mask |= CEPH_SETATTR_UID;
2595 release |= CEPH_CAP_AUTH_SHARED;
2596 }
2597 }
2598 if (ia_valid & ATTR_GID) {
2599 kgid_t fsgid = from_vfsgid(idmap, i_user_ns(inode), attr->ia_vfsgid);
2600
2601 doutc(cl, "%p %llx.%llx gid %d -> %d\n", inode,
2602 ceph_vinop(inode),
2603 from_kgid(&init_user_ns, inode->i_gid),
2604 from_kgid(&init_user_ns, attr->ia_gid));
2605 if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2606 inode->i_gid = fsgid;
2607 dirtied |= CEPH_CAP_AUTH_EXCL;
2608 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2609 !gid_eq(fsgid, inode->i_gid)) {
2610 req->r_args.setattr.gid = cpu_to_le32(
2611 from_kgid(&init_user_ns, fsgid));
2612 mask |= CEPH_SETATTR_GID;
2613 release |= CEPH_CAP_AUTH_SHARED;
2614 }
2615 }
2616 if (ia_valid & ATTR_MODE) {
2617 doutc(cl, "%p %llx.%llx mode 0%o -> 0%o\n", inode,
2618 ceph_vinop(inode), inode->i_mode, attr->ia_mode);
2619 if (!do_sync && (issued & CEPH_CAP_AUTH_EXCL)) {
2620 inode->i_mode = attr->ia_mode;
2621 dirtied |= CEPH_CAP_AUTH_EXCL;
2622 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2623 attr->ia_mode != inode->i_mode) {
2624 inode->i_mode = attr->ia_mode;
2625 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2626 mask |= CEPH_SETATTR_MODE;
2627 release |= CEPH_CAP_AUTH_SHARED;
2628 }
2629 }
2630
2631 if (ia_valid & ATTR_ATIME) {
2632 struct timespec64 atime = inode_get_atime(inode);
2633
2634 doutc(cl, "%p %llx.%llx atime %lld.%09ld -> %lld.%09ld\n",
2635 inode, ceph_vinop(inode),
2636 atime.tv_sec, atime.tv_nsec,
2637 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2638 if (!do_sync && (issued & CEPH_CAP_FILE_EXCL)) {
2639 ci->i_time_warp_seq++;
2640 inode_set_atime_to_ts(inode, attr->ia_atime);
2641 dirtied |= CEPH_CAP_FILE_EXCL;
2642 } else if (!do_sync && (issued & CEPH_CAP_FILE_WR) &&
2643 timespec64_compare(&atime,
2644 &attr->ia_atime) < 0) {
2645 inode_set_atime_to_ts(inode, attr->ia_atime);
2646 dirtied |= CEPH_CAP_FILE_WR;
2647 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2648 !timespec64_equal(&atime, &attr->ia_atime)) {
2649 ceph_encode_timespec64(&req->r_args.setattr.atime,
2650 &attr->ia_atime);
2651 mask |= CEPH_SETATTR_ATIME;
2652 release |= CEPH_CAP_FILE_SHARED |
2653 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2654 }
2655 }
2656 if (ia_valid & ATTR_SIZE) {
2657 doutc(cl, "%p %llx.%llx size %lld -> %lld\n", inode,
2658 ceph_vinop(inode), isize, attr->ia_size);
2659 /*
2660 * Only when the new size is smaller and not aligned to
2661 * CEPH_FSCRYPT_BLOCK_SIZE will the RMW is needed.
2662 */
2663 if (IS_ENCRYPTED(inode) && attr->ia_size < isize &&
2664 (attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE)) {
2665 mask |= CEPH_SETATTR_SIZE;
2666 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2667 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2668 set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2669 mask |= CEPH_SETATTR_FSCRYPT_FILE;
2670 req->r_args.setattr.size =
2671 cpu_to_le64(round_up(attr->ia_size,
2672 CEPH_FSCRYPT_BLOCK_SIZE));
2673 req->r_args.setattr.old_size =
2674 cpu_to_le64(round_up(isize,
2675 CEPH_FSCRYPT_BLOCK_SIZE));
2676 req->r_fscrypt_file = attr->ia_size;
2677 fill_fscrypt = true;
2678 } else if (!do_sync && (issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
2679 if (attr->ia_size > isize) {
2680 i_size_write(inode, attr->ia_size);
2681 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2682 ci->i_reported_size = attr->ia_size;
2683 dirtied |= CEPH_CAP_FILE_EXCL;
2684 ia_valid |= ATTR_MTIME;
2685 }
2686 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2687 attr->ia_size != isize) {
2688 mask |= CEPH_SETATTR_SIZE;
2689 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2690 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2691 if (IS_ENCRYPTED(inode) && attr->ia_size) {
2692 set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2693 mask |= CEPH_SETATTR_FSCRYPT_FILE;
2694 req->r_args.setattr.size =
2695 cpu_to_le64(round_up(attr->ia_size,
2696 CEPH_FSCRYPT_BLOCK_SIZE));
2697 req->r_args.setattr.old_size =
2698 cpu_to_le64(round_up(isize,
2699 CEPH_FSCRYPT_BLOCK_SIZE));
2700 req->r_fscrypt_file = attr->ia_size;
2701 } else {
2702 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2703 req->r_args.setattr.old_size = cpu_to_le64(isize);
2704 req->r_fscrypt_file = 0;
2705 }
2706 }
2707 }
2708 if (ia_valid & ATTR_MTIME) {
2709 struct timespec64 mtime = inode_get_mtime(inode);
2710
2711 doutc(cl, "%p %llx.%llx mtime %lld.%09ld -> %lld.%09ld\n",
2712 inode, ceph_vinop(inode),
2713 mtime.tv_sec, mtime.tv_nsec,
2714 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2715 if (!do_sync && (issued & CEPH_CAP_FILE_EXCL)) {
2716 ci->i_time_warp_seq++;
2717 inode_set_mtime_to_ts(inode, attr->ia_mtime);
2718 dirtied |= CEPH_CAP_FILE_EXCL;
2719 } else if (!do_sync && (issued & CEPH_CAP_FILE_WR) &&
2720 timespec64_compare(&mtime, &attr->ia_mtime) < 0) {
2721 inode_set_mtime_to_ts(inode, attr->ia_mtime);
2722 dirtied |= CEPH_CAP_FILE_WR;
2723 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2724 !timespec64_equal(&mtime, &attr->ia_mtime)) {
2725 ceph_encode_timespec64(&req->r_args.setattr.mtime,
2726 &attr->ia_mtime);
2727 mask |= CEPH_SETATTR_MTIME;
2728 release |= CEPH_CAP_FILE_SHARED |
2729 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2730 }
2731 }
2732
2733 /* these do nothing */
2734 if (ia_valid & ATTR_CTIME) {
2735 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2736 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2737 doutc(cl, "%p %llx.%llx ctime %lld.%09ld -> %lld.%09ld (%s)\n",
2738 inode, ceph_vinop(inode),
2739 inode_get_ctime_sec(inode),
2740 inode_get_ctime_nsec(inode),
2741 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2742 only ? "ctime only" : "ignored");
2743 if (only) {
2744 /*
2745 * if kernel wants to dirty ctime but nothing else,
2746 * we need to choose a cap to dirty under, or do
2747 * a almost-no-op setattr
2748 */
2749 if (issued & CEPH_CAP_AUTH_EXCL)
2750 dirtied |= CEPH_CAP_AUTH_EXCL;
2751 else if (issued & CEPH_CAP_FILE_EXCL)
2752 dirtied |= CEPH_CAP_FILE_EXCL;
2753 else if (issued & CEPH_CAP_XATTR_EXCL)
2754 dirtied |= CEPH_CAP_XATTR_EXCL;
2755 else
2756 mask |= CEPH_SETATTR_CTIME;
2757 }
2758 }
2759 if (ia_valid & ATTR_FILE)
2760 doutc(cl, "%p %llx.%llx ATTR_FILE ... hrm!\n", inode,
2761 ceph_vinop(inode));
2762
2763 if (dirtied) {
2764 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2765 &prealloc_cf);
2766 inode_set_ctime_to_ts(inode, attr->ia_ctime);
2767 inode_inc_iversion_raw(inode);
2768 }
2769
2770 release &= issued;
2771 spin_unlock(&ci->i_ceph_lock);
2772 if (lock_snap_rwsem) {
2773 up_read(&mdsc->snap_rwsem);
2774 lock_snap_rwsem = false;
2775 }
2776
2777 if (inode_dirty_flags)
2778 __mark_inode_dirty(inode, inode_dirty_flags);
2779
2780 if (mask) {
2781 req->r_inode = inode;
2782 ihold(inode);
2783 req->r_inode_drop = release;
2784 req->r_args.setattr.mask = cpu_to_le32(mask);
2785 req->r_num_caps = 1;
2786 req->r_stamp = attr->ia_ctime;
2787 if (fill_fscrypt) {
2788 err = fill_fscrypt_truncate(inode, req, attr);
2789 if (err)
2790 goto out;
2791 }
2792
2793 /*
2794 * The truncate request will return -EAGAIN when the
2795 * last block has been updated just before the MDS
2796 * successfully gets the xlock for the FILE lock. To
2797 * avoid corrupting the file contents we need to retry
2798 * it.
2799 */
2800 err = ceph_mdsc_do_request(mdsc, NULL, req);
2801 if (err == -EAGAIN && truncate_retry--) {
2802 doutc(cl, "%p %llx.%llx result=%d (%s locally, %d remote), retry it!\n",
2803 inode, ceph_vinop(inode), err,
2804 ceph_cap_string(dirtied), mask);
2805 ceph_mdsc_put_request(req);
2806 ceph_free_cap_flush(prealloc_cf);
2807 goto retry;
2808 }
2809 }
2810out:
2811 doutc(cl, "%p %llx.%llx result=%d (%s locally, %d remote)\n", inode,
2812 ceph_vinop(inode), err, ceph_cap_string(dirtied), mask);
2813
2814 ceph_mdsc_put_request(req);
2815 ceph_free_cap_flush(prealloc_cf);
2816
2817 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2818 __ceph_do_pending_vmtruncate(inode);
2819
2820 return err;
2821}
2822
2823/*
2824 * setattr
2825 */
2826int ceph_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
2827 struct iattr *attr)
2828{
2829 struct inode *inode = d_inode(dentry);
2830 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2831 int err;
2832
2833 if (ceph_snap(inode) != CEPH_NOSNAP)
2834 return -EROFS;
2835
2836 if (ceph_inode_is_shutdown(inode))
2837 return -ESTALE;
2838
2839 err = fscrypt_prepare_setattr(dentry, attr);
2840 if (err)
2841 return err;
2842
2843 err = setattr_prepare(idmap, dentry, attr);
2844 if (err != 0)
2845 return err;
2846
2847 if ((attr->ia_valid & ATTR_SIZE) &&
2848 attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2849 return -EFBIG;
2850
2851 if ((attr->ia_valid & ATTR_SIZE) &&
2852 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2853 return -EDQUOT;
2854
2855 err = __ceph_setattr(idmap, inode, attr, NULL);
2856
2857 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2858 err = posix_acl_chmod(idmap, dentry, attr->ia_mode);
2859
2860 return err;
2861}
2862
2863int ceph_try_to_choose_auth_mds(struct inode *inode, int mask)
2864{
2865 int issued = ceph_caps_issued(ceph_inode(inode));
2866
2867 /*
2868 * If any 'x' caps is issued we can just choose the auth MDS
2869 * instead of the random replica MDSes. Because only when the
2870 * Locker is in LOCK_EXEC state will the loner client could
2871 * get the 'x' caps. And if we send the getattr requests to
2872 * any replica MDS it must auth pin and tries to rdlock from
2873 * the auth MDS, and then the auth MDS need to do the Locker
2874 * state transition to LOCK_SYNC. And after that the lock state
2875 * will change back.
2876 *
2877 * This cost much when doing the Locker state transition and
2878 * usually will need to revoke caps from clients.
2879 *
2880 * And for the 'Xs' caps for getxattr we will also choose the
2881 * auth MDS, because the MDS side code is buggy due to setxattr
2882 * won't notify the replica MDSes when the values changed and
2883 * the replica MDS will return the old values. Though we will
2884 * fix it in MDS code, but this still makes sense for old ceph.
2885 */
2886 if (((mask & CEPH_CAP_ANY_SHARED) && (issued & CEPH_CAP_ANY_EXCL))
2887 || (mask & (CEPH_STAT_RSTAT | CEPH_STAT_CAP_XATTR)))
2888 return USE_AUTH_MDS;
2889 else
2890 return USE_ANY_MDS;
2891}
2892
2893/*
2894 * Verify that we have a lease on the given mask. If not,
2895 * do a getattr against an mds.
2896 */
2897int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2898 int mask, bool force)
2899{
2900 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
2901 struct ceph_client *cl = fsc->client;
2902 struct ceph_mds_client *mdsc = fsc->mdsc;
2903 struct ceph_mds_request *req;
2904 int mode;
2905 int err;
2906
2907 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2908 doutc(cl, "inode %p %llx.%llx SNAPDIR\n", inode,
2909 ceph_vinop(inode));
2910 return 0;
2911 }
2912
2913 doutc(cl, "inode %p %llx.%llx mask %s mode 0%o\n", inode,
2914 ceph_vinop(inode), ceph_cap_string(mask), inode->i_mode);
2915 if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2916 return 0;
2917
2918 mode = ceph_try_to_choose_auth_mds(inode, mask);
2919 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2920 if (IS_ERR(req))
2921 return PTR_ERR(req);
2922 req->r_inode = inode;
2923 ihold(inode);
2924 req->r_num_caps = 1;
2925 req->r_args.getattr.mask = cpu_to_le32(mask);
2926 req->r_locked_page = locked_page;
2927 err = ceph_mdsc_do_request(mdsc, NULL, req);
2928 if (locked_page && err == 0) {
2929 u64 inline_version = req->r_reply_info.targeti.inline_version;
2930 if (inline_version == 0) {
2931 /* the reply is supposed to contain inline data */
2932 err = -EINVAL;
2933 } else if (inline_version == CEPH_INLINE_NONE ||
2934 inline_version == 1) {
2935 err = -ENODATA;
2936 } else {
2937 err = req->r_reply_info.targeti.inline_len;
2938 }
2939 }
2940 ceph_mdsc_put_request(req);
2941 doutc(cl, "result=%d\n", err);
2942 return err;
2943}
2944
2945int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
2946 size_t size)
2947{
2948 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
2949 struct ceph_client *cl = fsc->client;
2950 struct ceph_mds_client *mdsc = fsc->mdsc;
2951 struct ceph_mds_request *req;
2952 int mode = USE_AUTH_MDS;
2953 int err;
2954 char *xattr_value;
2955 size_t xattr_value_len;
2956
2957 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode);
2958 if (IS_ERR(req)) {
2959 err = -ENOMEM;
2960 goto out;
2961 }
2962
2963 req->r_feature_needed = CEPHFS_FEATURE_OP_GETVXATTR;
2964 req->r_path2 = kstrdup(name, GFP_NOFS);
2965 if (!req->r_path2) {
2966 err = -ENOMEM;
2967 goto put;
2968 }
2969
2970 ihold(inode);
2971 req->r_inode = inode;
2972 err = ceph_mdsc_do_request(mdsc, NULL, req);
2973 if (err < 0)
2974 goto put;
2975
2976 xattr_value = req->r_reply_info.xattr_info.xattr_value;
2977 xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len;
2978
2979 doutc(cl, "xattr_value_len:%zu, size:%zu\n", xattr_value_len, size);
2980
2981 err = (int)xattr_value_len;
2982 if (size == 0)
2983 goto put;
2984
2985 if (xattr_value_len > size) {
2986 err = -ERANGE;
2987 goto put;
2988 }
2989
2990 memcpy(value, xattr_value, xattr_value_len);
2991put:
2992 ceph_mdsc_put_request(req);
2993out:
2994 doutc(cl, "result=%d\n", err);
2995 return err;
2996}
2997
2998
2999/*
3000 * Check inode permissions. We verify we have a valid value for
3001 * the AUTH cap, then call the generic handler.
3002 */
3003int ceph_permission(struct mnt_idmap *idmap, struct inode *inode,
3004 int mask)
3005{
3006 int err;
3007
3008 if (mask & MAY_NOT_BLOCK)
3009 return -ECHILD;
3010
3011 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
3012
3013 if (!err)
3014 err = generic_permission(idmap, inode, mask);
3015 return err;
3016}
3017
3018/* Craft a mask of needed caps given a set of requested statx attrs. */
3019static int statx_to_caps(u32 want, umode_t mode)
3020{
3021 int mask = 0;
3022
3023 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME|STATX_CHANGE_COOKIE))
3024 mask |= CEPH_CAP_AUTH_SHARED;
3025
3026 if (want & (STATX_NLINK|STATX_CTIME|STATX_CHANGE_COOKIE)) {
3027 /*
3028 * The link count for directories depends on inode->i_subdirs,
3029 * and that is only updated when Fs caps are held.
3030 */
3031 if (S_ISDIR(mode))
3032 mask |= CEPH_CAP_FILE_SHARED;
3033 else
3034 mask |= CEPH_CAP_LINK_SHARED;
3035 }
3036
3037 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|STATX_BLOCKS|STATX_CHANGE_COOKIE))
3038 mask |= CEPH_CAP_FILE_SHARED;
3039
3040 if (want & (STATX_CTIME|STATX_CHANGE_COOKIE))
3041 mask |= CEPH_CAP_XATTR_SHARED;
3042
3043 return mask;
3044}
3045
3046/*
3047 * Get all the attributes. If we have sufficient caps for the requested attrs,
3048 * then we can avoid talking to the MDS at all.
3049 */
3050int ceph_getattr(struct mnt_idmap *idmap, const struct path *path,
3051 struct kstat *stat, u32 request_mask, unsigned int flags)
3052{
3053 struct inode *inode = d_inode(path->dentry);
3054 struct super_block *sb = inode->i_sb;
3055 struct ceph_inode_info *ci = ceph_inode(inode);
3056 u32 valid_mask = STATX_BASIC_STATS;
3057 int err = 0;
3058
3059 if (ceph_inode_is_shutdown(inode))
3060 return -ESTALE;
3061
3062 /* Skip the getattr altogether if we're asked not to sync */
3063 if ((flags & AT_STATX_SYNC_TYPE) != AT_STATX_DONT_SYNC) {
3064 err = ceph_do_getattr(inode,
3065 statx_to_caps(request_mask, inode->i_mode),
3066 flags & AT_STATX_FORCE_SYNC);
3067 if (err)
3068 return err;
3069 }
3070
3071 generic_fillattr(idmap, request_mask, inode, stat);
3072 stat->ino = ceph_present_inode(inode);
3073
3074 /*
3075 * btime on newly-allocated inodes is 0, so if this is still set to
3076 * that, then assume that it's not valid.
3077 */
3078 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
3079 stat->btime = ci->i_btime;
3080 valid_mask |= STATX_BTIME;
3081 }
3082
3083 if (request_mask & STATX_CHANGE_COOKIE) {
3084 stat->change_cookie = inode_peek_iversion_raw(inode);
3085 valid_mask |= STATX_CHANGE_COOKIE;
3086 }
3087
3088 if (ceph_snap(inode) == CEPH_NOSNAP)
3089 stat->dev = sb->s_dev;
3090 else
3091 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
3092
3093 if (S_ISDIR(inode->i_mode)) {
3094 if (ceph_test_mount_opt(ceph_sb_to_fs_client(sb), RBYTES)) {
3095 stat->size = ci->i_rbytes;
3096 } else if (ceph_snap(inode) == CEPH_SNAPDIR) {
3097 struct ceph_inode_info *pci;
3098 struct ceph_snap_realm *realm;
3099 struct inode *parent;
3100
3101 parent = ceph_lookup_inode(sb, ceph_ino(inode));
3102 if (IS_ERR(parent))
3103 return PTR_ERR(parent);
3104
3105 pci = ceph_inode(parent);
3106 spin_lock(&pci->i_ceph_lock);
3107 realm = pci->i_snap_realm;
3108 if (realm)
3109 stat->size = realm->num_snaps;
3110 else
3111 stat->size = 0;
3112 spin_unlock(&pci->i_ceph_lock);
3113 iput(parent);
3114 } else {
3115 stat->size = ci->i_files + ci->i_subdirs;
3116 }
3117 stat->blocks = 0;
3118 stat->blksize = 65536;
3119 /*
3120 * Some applications rely on the number of st_nlink
3121 * value on directories to be either 0 (if unlinked)
3122 * or 2 + number of subdirectories.
3123 */
3124 if (stat->nlink == 1)
3125 /* '.' + '..' + subdirs */
3126 stat->nlink = 1 + 1 + ci->i_subdirs;
3127 }
3128
3129 stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
3130 if (IS_ENCRYPTED(inode))
3131 stat->attributes |= STATX_ATTR_ENCRYPTED;
3132 stat->attributes_mask |= (STATX_ATTR_CHANGE_MONOTONIC |
3133 STATX_ATTR_ENCRYPTED);
3134
3135 stat->result_mask = request_mask & valid_mask;
3136 return err;
3137}
3138
3139void ceph_inode_shutdown(struct inode *inode)
3140{
3141 struct ceph_inode_info *ci = ceph_inode(inode);
3142 struct rb_node *p;
3143 int iputs = 0;
3144 bool invalidate = false;
3145
3146 spin_lock(&ci->i_ceph_lock);
3147 ci->i_ceph_flags |= CEPH_I_SHUTDOWN;
3148 p = rb_first(&ci->i_caps);
3149 while (p) {
3150 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
3151
3152 p = rb_next(p);
3153 iputs += ceph_purge_inode_cap(inode, cap, &invalidate);
3154 }
3155 spin_unlock(&ci->i_ceph_lock);
3156
3157 if (invalidate)
3158 ceph_queue_invalidate(inode);
3159 while (iputs--)
3160 iput(inode);
3161}
1#include <linux/ceph/ceph_debug.h>
2
3#include <linux/module.h>
4#include <linux/fs.h>
5#include <linux/slab.h>
6#include <linux/string.h>
7#include <linux/uaccess.h>
8#include <linux/kernel.h>
9#include <linux/writeback.h>
10#include <linux/vmalloc.h>
11#include <linux/xattr.h>
12#include <linux/posix_acl.h>
13#include <linux/random.h>
14#include <linux/sort.h>
15
16#include "super.h"
17#include "mds_client.h"
18#include "cache.h"
19#include <linux/ceph/decode.h>
20
21/*
22 * Ceph inode operations
23 *
24 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
25 * setattr, etc.), xattr helpers, and helpers for assimilating
26 * metadata returned by the MDS into our cache.
27 *
28 * Also define helpers for doing asynchronous writeback, invalidation,
29 * and truncation for the benefit of those who can't afford to block
30 * (typically because they are in the message handler path).
31 */
32
33static const struct inode_operations ceph_symlink_iops;
34
35static void ceph_invalidate_work(struct work_struct *work);
36static void ceph_writeback_work(struct work_struct *work);
37static void ceph_vmtruncate_work(struct work_struct *work);
38
39/*
40 * find or create an inode, given the ceph ino number
41 */
42static int ceph_set_ino_cb(struct inode *inode, void *data)
43{
44 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
45 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
46 return 0;
47}
48
49struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
50{
51 struct inode *inode;
52 ino_t t = ceph_vino_to_ino(vino);
53
54 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
55 if (inode == NULL)
56 return ERR_PTR(-ENOMEM);
57 if (inode->i_state & I_NEW) {
58 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
59 inode, ceph_vinop(inode), (u64)inode->i_ino);
60 unlock_new_inode(inode);
61 }
62
63 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
64 vino.snap, inode);
65 return inode;
66}
67
68/*
69 * get/constuct snapdir inode for a given directory
70 */
71struct inode *ceph_get_snapdir(struct inode *parent)
72{
73 struct ceph_vino vino = {
74 .ino = ceph_ino(parent),
75 .snap = CEPH_SNAPDIR,
76 };
77 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
78 struct ceph_inode_info *ci = ceph_inode(inode);
79
80 BUG_ON(!S_ISDIR(parent->i_mode));
81 if (IS_ERR(inode))
82 return inode;
83 inode->i_mode = parent->i_mode;
84 inode->i_uid = parent->i_uid;
85 inode->i_gid = parent->i_gid;
86 inode->i_op = &ceph_snapdir_iops;
87 inode->i_fop = &ceph_snapdir_fops;
88 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
89 ci->i_rbytes = 0;
90 return inode;
91}
92
93const struct inode_operations ceph_file_iops = {
94 .permission = ceph_permission,
95 .setattr = ceph_setattr,
96 .getattr = ceph_getattr,
97 .listxattr = ceph_listxattr,
98 .get_acl = ceph_get_acl,
99 .set_acl = ceph_set_acl,
100};
101
102
103/*
104 * We use a 'frag tree' to keep track of the MDS's directory fragments
105 * for a given inode (usually there is just a single fragment). We
106 * need to know when a child frag is delegated to a new MDS, or when
107 * it is flagged as replicated, so we can direct our requests
108 * accordingly.
109 */
110
111/*
112 * find/create a frag in the tree
113 */
114static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
115 u32 f)
116{
117 struct rb_node **p;
118 struct rb_node *parent = NULL;
119 struct ceph_inode_frag *frag;
120 int c;
121
122 p = &ci->i_fragtree.rb_node;
123 while (*p) {
124 parent = *p;
125 frag = rb_entry(parent, struct ceph_inode_frag, node);
126 c = ceph_frag_compare(f, frag->frag);
127 if (c < 0)
128 p = &(*p)->rb_left;
129 else if (c > 0)
130 p = &(*p)->rb_right;
131 else
132 return frag;
133 }
134
135 frag = kmalloc(sizeof(*frag), GFP_NOFS);
136 if (!frag) {
137 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
138 "frag %x\n", &ci->vfs_inode,
139 ceph_vinop(&ci->vfs_inode), f);
140 return ERR_PTR(-ENOMEM);
141 }
142 frag->frag = f;
143 frag->split_by = 0;
144 frag->mds = -1;
145 frag->ndist = 0;
146
147 rb_link_node(&frag->node, parent, p);
148 rb_insert_color(&frag->node, &ci->i_fragtree);
149
150 dout("get_or_create_frag added %llx.%llx frag %x\n",
151 ceph_vinop(&ci->vfs_inode), f);
152 return frag;
153}
154
155/*
156 * find a specific frag @f
157 */
158struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
159{
160 struct rb_node *n = ci->i_fragtree.rb_node;
161
162 while (n) {
163 struct ceph_inode_frag *frag =
164 rb_entry(n, struct ceph_inode_frag, node);
165 int c = ceph_frag_compare(f, frag->frag);
166 if (c < 0)
167 n = n->rb_left;
168 else if (c > 0)
169 n = n->rb_right;
170 else
171 return frag;
172 }
173 return NULL;
174}
175
176/*
177 * Choose frag containing the given value @v. If @pfrag is
178 * specified, copy the frag delegation info to the caller if
179 * it is present.
180 */
181static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
182 struct ceph_inode_frag *pfrag, int *found)
183{
184 u32 t = ceph_frag_make(0, 0);
185 struct ceph_inode_frag *frag;
186 unsigned nway, i;
187 u32 n;
188
189 if (found)
190 *found = 0;
191
192 while (1) {
193 WARN_ON(!ceph_frag_contains_value(t, v));
194 frag = __ceph_find_frag(ci, t);
195 if (!frag)
196 break; /* t is a leaf */
197 if (frag->split_by == 0) {
198 if (pfrag)
199 memcpy(pfrag, frag, sizeof(*pfrag));
200 if (found)
201 *found = 1;
202 break;
203 }
204
205 /* choose child */
206 nway = 1 << frag->split_by;
207 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
208 frag->split_by, nway);
209 for (i = 0; i < nway; i++) {
210 n = ceph_frag_make_child(t, frag->split_by, i);
211 if (ceph_frag_contains_value(n, v)) {
212 t = n;
213 break;
214 }
215 }
216 BUG_ON(i == nway);
217 }
218 dout("choose_frag(%x) = %x\n", v, t);
219
220 return t;
221}
222
223u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
224 struct ceph_inode_frag *pfrag, int *found)
225{
226 u32 ret;
227 mutex_lock(&ci->i_fragtree_mutex);
228 ret = __ceph_choose_frag(ci, v, pfrag, found);
229 mutex_unlock(&ci->i_fragtree_mutex);
230 return ret;
231}
232
233/*
234 * Process dirfrag (delegation) info from the mds. Include leaf
235 * fragment in tree ONLY if ndist > 0. Otherwise, only
236 * branches/splits are included in i_fragtree)
237 */
238static int ceph_fill_dirfrag(struct inode *inode,
239 struct ceph_mds_reply_dirfrag *dirinfo)
240{
241 struct ceph_inode_info *ci = ceph_inode(inode);
242 struct ceph_inode_frag *frag;
243 u32 id = le32_to_cpu(dirinfo->frag);
244 int mds = le32_to_cpu(dirinfo->auth);
245 int ndist = le32_to_cpu(dirinfo->ndist);
246 int diri_auth = -1;
247 int i;
248 int err = 0;
249
250 spin_lock(&ci->i_ceph_lock);
251 if (ci->i_auth_cap)
252 diri_auth = ci->i_auth_cap->mds;
253 spin_unlock(&ci->i_ceph_lock);
254
255 if (mds == -1) /* CDIR_AUTH_PARENT */
256 mds = diri_auth;
257
258 mutex_lock(&ci->i_fragtree_mutex);
259 if (ndist == 0 && mds == diri_auth) {
260 /* no delegation info needed. */
261 frag = __ceph_find_frag(ci, id);
262 if (!frag)
263 goto out;
264 if (frag->split_by == 0) {
265 /* tree leaf, remove */
266 dout("fill_dirfrag removed %llx.%llx frag %x"
267 " (no ref)\n", ceph_vinop(inode), id);
268 rb_erase(&frag->node, &ci->i_fragtree);
269 kfree(frag);
270 } else {
271 /* tree branch, keep and clear */
272 dout("fill_dirfrag cleared %llx.%llx frag %x"
273 " referral\n", ceph_vinop(inode), id);
274 frag->mds = -1;
275 frag->ndist = 0;
276 }
277 goto out;
278 }
279
280
281 /* find/add this frag to store mds delegation info */
282 frag = __get_or_create_frag(ci, id);
283 if (IS_ERR(frag)) {
284 /* this is not the end of the world; we can continue
285 with bad/inaccurate delegation info */
286 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
287 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
288 err = -ENOMEM;
289 goto out;
290 }
291
292 frag->mds = mds;
293 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
294 for (i = 0; i < frag->ndist; i++)
295 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
296 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
297 ceph_vinop(inode), frag->frag, frag->ndist);
298
299out:
300 mutex_unlock(&ci->i_fragtree_mutex);
301 return err;
302}
303
304static int frag_tree_split_cmp(const void *l, const void *r)
305{
306 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
307 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
308 return ceph_frag_compare(le32_to_cpu(ls->frag),
309 le32_to_cpu(rs->frag));
310}
311
312static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
313{
314 if (!frag)
315 return f == ceph_frag_make(0, 0);
316 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
317 return false;
318 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
319}
320
321static int ceph_fill_fragtree(struct inode *inode,
322 struct ceph_frag_tree_head *fragtree,
323 struct ceph_mds_reply_dirfrag *dirinfo)
324{
325 struct ceph_inode_info *ci = ceph_inode(inode);
326 struct ceph_inode_frag *frag, *prev_frag = NULL;
327 struct rb_node *rb_node;
328 unsigned i, split_by, nsplits;
329 u32 id;
330 bool update = false;
331
332 mutex_lock(&ci->i_fragtree_mutex);
333 nsplits = le32_to_cpu(fragtree->nsplits);
334 if (nsplits != ci->i_fragtree_nsplits) {
335 update = true;
336 } else if (nsplits) {
337 i = prandom_u32() % nsplits;
338 id = le32_to_cpu(fragtree->splits[i].frag);
339 if (!__ceph_find_frag(ci, id))
340 update = true;
341 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
342 rb_node = rb_first(&ci->i_fragtree);
343 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
344 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
345 update = true;
346 }
347 if (!update && dirinfo) {
348 id = le32_to_cpu(dirinfo->frag);
349 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
350 update = true;
351 }
352 if (!update)
353 goto out_unlock;
354
355 if (nsplits > 1) {
356 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
357 frag_tree_split_cmp, NULL);
358 }
359
360 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
361 rb_node = rb_first(&ci->i_fragtree);
362 for (i = 0; i < nsplits; i++) {
363 id = le32_to_cpu(fragtree->splits[i].frag);
364 split_by = le32_to_cpu(fragtree->splits[i].by);
365 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
366 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
367 "frag %x split by %d\n", ceph_vinop(inode),
368 i, nsplits, id, split_by);
369 continue;
370 }
371 frag = NULL;
372 while (rb_node) {
373 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
374 if (ceph_frag_compare(frag->frag, id) >= 0) {
375 if (frag->frag != id)
376 frag = NULL;
377 else
378 rb_node = rb_next(rb_node);
379 break;
380 }
381 rb_node = rb_next(rb_node);
382 /* delete stale split/leaf node */
383 if (frag->split_by > 0 ||
384 !is_frag_child(frag->frag, prev_frag)) {
385 rb_erase(&frag->node, &ci->i_fragtree);
386 if (frag->split_by > 0)
387 ci->i_fragtree_nsplits--;
388 kfree(frag);
389 }
390 frag = NULL;
391 }
392 if (!frag) {
393 frag = __get_or_create_frag(ci, id);
394 if (IS_ERR(frag))
395 continue;
396 }
397 if (frag->split_by == 0)
398 ci->i_fragtree_nsplits++;
399 frag->split_by = split_by;
400 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
401 prev_frag = frag;
402 }
403 while (rb_node) {
404 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
405 rb_node = rb_next(rb_node);
406 /* delete stale split/leaf node */
407 if (frag->split_by > 0 ||
408 !is_frag_child(frag->frag, prev_frag)) {
409 rb_erase(&frag->node, &ci->i_fragtree);
410 if (frag->split_by > 0)
411 ci->i_fragtree_nsplits--;
412 kfree(frag);
413 }
414 }
415out_unlock:
416 mutex_unlock(&ci->i_fragtree_mutex);
417 return 0;
418}
419
420/*
421 * initialize a newly allocated inode.
422 */
423struct inode *ceph_alloc_inode(struct super_block *sb)
424{
425 struct ceph_inode_info *ci;
426 int i;
427
428 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
429 if (!ci)
430 return NULL;
431
432 dout("alloc_inode %p\n", &ci->vfs_inode);
433
434 spin_lock_init(&ci->i_ceph_lock);
435
436 ci->i_version = 0;
437 ci->i_inline_version = 0;
438 ci->i_time_warp_seq = 0;
439 ci->i_ceph_flags = 0;
440 atomic64_set(&ci->i_ordered_count, 1);
441 atomic64_set(&ci->i_release_count, 1);
442 atomic64_set(&ci->i_complete_seq[0], 0);
443 atomic64_set(&ci->i_complete_seq[1], 0);
444 ci->i_symlink = NULL;
445
446 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
447 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
448
449 ci->i_fragtree = RB_ROOT;
450 mutex_init(&ci->i_fragtree_mutex);
451
452 ci->i_xattrs.blob = NULL;
453 ci->i_xattrs.prealloc_blob = NULL;
454 ci->i_xattrs.dirty = false;
455 ci->i_xattrs.index = RB_ROOT;
456 ci->i_xattrs.count = 0;
457 ci->i_xattrs.names_size = 0;
458 ci->i_xattrs.vals_size = 0;
459 ci->i_xattrs.version = 0;
460 ci->i_xattrs.index_version = 0;
461
462 ci->i_caps = RB_ROOT;
463 ci->i_auth_cap = NULL;
464 ci->i_dirty_caps = 0;
465 ci->i_flushing_caps = 0;
466 INIT_LIST_HEAD(&ci->i_dirty_item);
467 INIT_LIST_HEAD(&ci->i_flushing_item);
468 ci->i_prealloc_cap_flush = NULL;
469 INIT_LIST_HEAD(&ci->i_cap_flush_list);
470 init_waitqueue_head(&ci->i_cap_wq);
471 ci->i_hold_caps_min = 0;
472 ci->i_hold_caps_max = 0;
473 INIT_LIST_HEAD(&ci->i_cap_delay_list);
474 INIT_LIST_HEAD(&ci->i_cap_snaps);
475 ci->i_head_snapc = NULL;
476 ci->i_snap_caps = 0;
477
478 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
479 ci->i_nr_by_mode[i] = 0;
480
481 mutex_init(&ci->i_truncate_mutex);
482 ci->i_truncate_seq = 0;
483 ci->i_truncate_size = 0;
484 ci->i_truncate_pending = 0;
485
486 ci->i_max_size = 0;
487 ci->i_reported_size = 0;
488 ci->i_wanted_max_size = 0;
489 ci->i_requested_max_size = 0;
490
491 ci->i_pin_ref = 0;
492 ci->i_rd_ref = 0;
493 ci->i_rdcache_ref = 0;
494 ci->i_wr_ref = 0;
495 ci->i_wb_ref = 0;
496 ci->i_wrbuffer_ref = 0;
497 ci->i_wrbuffer_ref_head = 0;
498 ci->i_shared_gen = 0;
499 ci->i_rdcache_gen = 0;
500 ci->i_rdcache_revoking = 0;
501
502 INIT_LIST_HEAD(&ci->i_unsafe_writes);
503 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
504 INIT_LIST_HEAD(&ci->i_unsafe_iops);
505 spin_lock_init(&ci->i_unsafe_lock);
506
507 ci->i_snap_realm = NULL;
508 INIT_LIST_HEAD(&ci->i_snap_realm_item);
509 INIT_LIST_HEAD(&ci->i_snap_flush_item);
510
511 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
512 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
513
514 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
515
516 ceph_fscache_inode_init(ci);
517
518 return &ci->vfs_inode;
519}
520
521static void ceph_i_callback(struct rcu_head *head)
522{
523 struct inode *inode = container_of(head, struct inode, i_rcu);
524 struct ceph_inode_info *ci = ceph_inode(inode);
525
526 kmem_cache_free(ceph_inode_cachep, ci);
527}
528
529void ceph_destroy_inode(struct inode *inode)
530{
531 struct ceph_inode_info *ci = ceph_inode(inode);
532 struct ceph_inode_frag *frag;
533 struct rb_node *n;
534
535 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
536
537 ceph_fscache_unregister_inode_cookie(ci);
538
539 ceph_queue_caps_release(inode);
540
541 /*
542 * we may still have a snap_realm reference if there are stray
543 * caps in i_snap_caps.
544 */
545 if (ci->i_snap_realm) {
546 struct ceph_mds_client *mdsc =
547 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
548 struct ceph_snap_realm *realm = ci->i_snap_realm;
549
550 dout(" dropping residual ref to snap realm %p\n", realm);
551 spin_lock(&realm->inodes_with_caps_lock);
552 list_del_init(&ci->i_snap_realm_item);
553 spin_unlock(&realm->inodes_with_caps_lock);
554 ceph_put_snap_realm(mdsc, realm);
555 }
556
557 kfree(ci->i_symlink);
558 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
559 frag = rb_entry(n, struct ceph_inode_frag, node);
560 rb_erase(n, &ci->i_fragtree);
561 kfree(frag);
562 }
563 ci->i_fragtree_nsplits = 0;
564
565 __ceph_destroy_xattrs(ci);
566 if (ci->i_xattrs.blob)
567 ceph_buffer_put(ci->i_xattrs.blob);
568 if (ci->i_xattrs.prealloc_blob)
569 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
570
571 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
572
573 call_rcu(&inode->i_rcu, ceph_i_callback);
574}
575
576int ceph_drop_inode(struct inode *inode)
577{
578 /*
579 * Positve dentry and corresponding inode are always accompanied
580 * in MDS reply. So no need to keep inode in the cache after
581 * dropping all its aliases.
582 */
583 return 1;
584}
585
586void ceph_evict_inode(struct inode *inode)
587{
588 /* wait unsafe sync writes */
589 ceph_sync_write_wait(inode);
590 truncate_inode_pages_final(&inode->i_data);
591 clear_inode(inode);
592}
593
594static inline blkcnt_t calc_inode_blocks(u64 size)
595{
596 return (size + (1<<9) - 1) >> 9;
597}
598
599/*
600 * Helpers to fill in size, ctime, mtime, and atime. We have to be
601 * careful because either the client or MDS may have more up to date
602 * info, depending on which capabilities are held, and whether
603 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
604 * and size are monotonically increasing, except when utimes() or
605 * truncate() increments the corresponding _seq values.)
606 */
607int ceph_fill_file_size(struct inode *inode, int issued,
608 u32 truncate_seq, u64 truncate_size, u64 size)
609{
610 struct ceph_inode_info *ci = ceph_inode(inode);
611 int queue_trunc = 0;
612
613 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
614 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
615 dout("size %lld -> %llu\n", inode->i_size, size);
616 if (size > 0 && S_ISDIR(inode->i_mode)) {
617 pr_err("fill_file_size non-zero size for directory\n");
618 size = 0;
619 }
620 i_size_write(inode, size);
621 inode->i_blocks = calc_inode_blocks(size);
622 ci->i_reported_size = size;
623 if (truncate_seq != ci->i_truncate_seq) {
624 dout("truncate_seq %u -> %u\n",
625 ci->i_truncate_seq, truncate_seq);
626 ci->i_truncate_seq = truncate_seq;
627
628 /* the MDS should have revoked these caps */
629 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
630 CEPH_CAP_FILE_RD |
631 CEPH_CAP_FILE_WR |
632 CEPH_CAP_FILE_LAZYIO));
633 /*
634 * If we hold relevant caps, or in the case where we're
635 * not the only client referencing this file and we
636 * don't hold those caps, then we need to check whether
637 * the file is either opened or mmaped
638 */
639 if ((issued & (CEPH_CAP_FILE_CACHE|
640 CEPH_CAP_FILE_BUFFER)) ||
641 mapping_mapped(inode->i_mapping) ||
642 __ceph_caps_file_wanted(ci)) {
643 ci->i_truncate_pending++;
644 queue_trunc = 1;
645 }
646 }
647 }
648 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
649 ci->i_truncate_size != truncate_size) {
650 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
651 truncate_size);
652 ci->i_truncate_size = truncate_size;
653 }
654
655 if (queue_trunc)
656 ceph_fscache_invalidate(inode);
657
658 return queue_trunc;
659}
660
661void ceph_fill_file_time(struct inode *inode, int issued,
662 u64 time_warp_seq, struct timespec *ctime,
663 struct timespec *mtime, struct timespec *atime)
664{
665 struct ceph_inode_info *ci = ceph_inode(inode);
666 int warn = 0;
667
668 if (issued & (CEPH_CAP_FILE_EXCL|
669 CEPH_CAP_FILE_WR|
670 CEPH_CAP_FILE_BUFFER|
671 CEPH_CAP_AUTH_EXCL|
672 CEPH_CAP_XATTR_EXCL)) {
673 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
674 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
675 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
676 ctime->tv_sec, ctime->tv_nsec);
677 inode->i_ctime = *ctime;
678 }
679 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
680 /* the MDS did a utimes() */
681 dout("mtime %ld.%09ld -> %ld.%09ld "
682 "tw %d -> %d\n",
683 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
684 mtime->tv_sec, mtime->tv_nsec,
685 ci->i_time_warp_seq, (int)time_warp_seq);
686
687 inode->i_mtime = *mtime;
688 inode->i_atime = *atime;
689 ci->i_time_warp_seq = time_warp_seq;
690 } else if (time_warp_seq == ci->i_time_warp_seq) {
691 /* nobody did utimes(); take the max */
692 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
693 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
694 inode->i_mtime.tv_sec,
695 inode->i_mtime.tv_nsec,
696 mtime->tv_sec, mtime->tv_nsec);
697 inode->i_mtime = *mtime;
698 }
699 if (timespec_compare(atime, &inode->i_atime) > 0) {
700 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
701 inode->i_atime.tv_sec,
702 inode->i_atime.tv_nsec,
703 atime->tv_sec, atime->tv_nsec);
704 inode->i_atime = *atime;
705 }
706 } else if (issued & CEPH_CAP_FILE_EXCL) {
707 /* we did a utimes(); ignore mds values */
708 } else {
709 warn = 1;
710 }
711 } else {
712 /* we have no write|excl caps; whatever the MDS says is true */
713 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
714 inode->i_ctime = *ctime;
715 inode->i_mtime = *mtime;
716 inode->i_atime = *atime;
717 ci->i_time_warp_seq = time_warp_seq;
718 } else {
719 warn = 1;
720 }
721 }
722 if (warn) /* time_warp_seq shouldn't go backwards */
723 dout("%p mds time_warp_seq %llu < %u\n",
724 inode, time_warp_seq, ci->i_time_warp_seq);
725}
726
727/*
728 * Populate an inode based on info from mds. May be called on new or
729 * existing inodes.
730 */
731static int fill_inode(struct inode *inode, struct page *locked_page,
732 struct ceph_mds_reply_info_in *iinfo,
733 struct ceph_mds_reply_dirfrag *dirinfo,
734 struct ceph_mds_session *session,
735 unsigned long ttl_from, int cap_fmode,
736 struct ceph_cap_reservation *caps_reservation)
737{
738 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
739 struct ceph_mds_reply_inode *info = iinfo->in;
740 struct ceph_inode_info *ci = ceph_inode(inode);
741 int issued = 0, implemented, new_issued;
742 struct timespec mtime, atime, ctime;
743 struct ceph_buffer *xattr_blob = NULL;
744 struct ceph_string *pool_ns = NULL;
745 struct ceph_cap *new_cap = NULL;
746 int err = 0;
747 bool wake = false;
748 bool queue_trunc = false;
749 bool new_version = false;
750 bool fill_inline = false;
751
752 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
753 inode, ceph_vinop(inode), le64_to_cpu(info->version),
754 ci->i_version);
755
756 /* prealloc new cap struct */
757 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
758 new_cap = ceph_get_cap(mdsc, caps_reservation);
759
760 /*
761 * prealloc xattr data, if it looks like we'll need it. only
762 * if len > 4 (meaning there are actually xattrs; the first 4
763 * bytes are the xattr count).
764 */
765 if (iinfo->xattr_len > 4) {
766 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
767 if (!xattr_blob)
768 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
769 iinfo->xattr_len);
770 }
771
772 if (iinfo->pool_ns_len > 0)
773 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
774 iinfo->pool_ns_len);
775
776 spin_lock(&ci->i_ceph_lock);
777
778 /*
779 * provided version will be odd if inode value is projected,
780 * even if stable. skip the update if we have newer stable
781 * info (ours>=theirs, e.g. due to racing mds replies), unless
782 * we are getting projected (unstable) info (in which case the
783 * version is odd, and we want ours>theirs).
784 * us them
785 * 2 2 skip
786 * 3 2 skip
787 * 3 3 update
788 */
789 if (ci->i_version == 0 ||
790 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
791 le64_to_cpu(info->version) > (ci->i_version & ~1)))
792 new_version = true;
793
794 issued = __ceph_caps_issued(ci, &implemented);
795 issued |= implemented | __ceph_caps_dirty(ci);
796 new_issued = ~issued & le32_to_cpu(info->cap.caps);
797
798 /* update inode */
799 ci->i_version = le64_to_cpu(info->version);
800 inode->i_version++;
801 inode->i_rdev = le32_to_cpu(info->rdev);
802 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
803
804 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
805 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
806 inode->i_mode = le32_to_cpu(info->mode);
807 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
808 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
809 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
810 from_kuid(&init_user_ns, inode->i_uid),
811 from_kgid(&init_user_ns, inode->i_gid));
812 }
813
814 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
815 (issued & CEPH_CAP_LINK_EXCL) == 0)
816 set_nlink(inode, le32_to_cpu(info->nlink));
817
818 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
819 /* be careful with mtime, atime, size */
820 ceph_decode_timespec(&atime, &info->atime);
821 ceph_decode_timespec(&mtime, &info->mtime);
822 ceph_decode_timespec(&ctime, &info->ctime);
823 ceph_fill_file_time(inode, issued,
824 le32_to_cpu(info->time_warp_seq),
825 &ctime, &mtime, &atime);
826 }
827
828 if (new_version ||
829 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
830 s64 old_pool = ci->i_layout.pool_id;
831 struct ceph_string *old_ns;
832
833 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
834 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
835 lockdep_is_held(&ci->i_ceph_lock));
836 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
837
838 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
839 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
840
841 pool_ns = old_ns;
842
843 queue_trunc = ceph_fill_file_size(inode, issued,
844 le32_to_cpu(info->truncate_seq),
845 le64_to_cpu(info->truncate_size),
846 le64_to_cpu(info->size));
847 /* only update max_size on auth cap */
848 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
849 ci->i_max_size != le64_to_cpu(info->max_size)) {
850 dout("max_size %lld -> %llu\n", ci->i_max_size,
851 le64_to_cpu(info->max_size));
852 ci->i_max_size = le64_to_cpu(info->max_size);
853 }
854 }
855
856 /* xattrs */
857 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
858 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
859 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
860 if (ci->i_xattrs.blob)
861 ceph_buffer_put(ci->i_xattrs.blob);
862 ci->i_xattrs.blob = xattr_blob;
863 if (xattr_blob)
864 memcpy(ci->i_xattrs.blob->vec.iov_base,
865 iinfo->xattr_data, iinfo->xattr_len);
866 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
867 ceph_forget_all_cached_acls(inode);
868 xattr_blob = NULL;
869 }
870
871 inode->i_mapping->a_ops = &ceph_aops;
872
873 switch (inode->i_mode & S_IFMT) {
874 case S_IFIFO:
875 case S_IFBLK:
876 case S_IFCHR:
877 case S_IFSOCK:
878 init_special_inode(inode, inode->i_mode, inode->i_rdev);
879 inode->i_op = &ceph_file_iops;
880 break;
881 case S_IFREG:
882 inode->i_op = &ceph_file_iops;
883 inode->i_fop = &ceph_file_fops;
884 break;
885 case S_IFLNK:
886 inode->i_op = &ceph_symlink_iops;
887 if (!ci->i_symlink) {
888 u32 symlen = iinfo->symlink_len;
889 char *sym;
890
891 spin_unlock(&ci->i_ceph_lock);
892
893 if (symlen != i_size_read(inode)) {
894 pr_err("fill_inode %llx.%llx BAD symlink "
895 "size %lld\n", ceph_vinop(inode),
896 i_size_read(inode));
897 i_size_write(inode, symlen);
898 inode->i_blocks = calc_inode_blocks(symlen);
899 }
900
901 err = -ENOMEM;
902 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
903 if (!sym)
904 goto out;
905
906 spin_lock(&ci->i_ceph_lock);
907 if (!ci->i_symlink)
908 ci->i_symlink = sym;
909 else
910 kfree(sym); /* lost a race */
911 }
912 inode->i_link = ci->i_symlink;
913 break;
914 case S_IFDIR:
915 inode->i_op = &ceph_dir_iops;
916 inode->i_fop = &ceph_dir_fops;
917
918 ci->i_dir_layout = iinfo->dir_layout;
919
920 ci->i_files = le64_to_cpu(info->files);
921 ci->i_subdirs = le64_to_cpu(info->subdirs);
922 ci->i_rbytes = le64_to_cpu(info->rbytes);
923 ci->i_rfiles = le64_to_cpu(info->rfiles);
924 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
925 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
926 break;
927 default:
928 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
929 ceph_vinop(inode), inode->i_mode);
930 }
931
932 /* were we issued a capability? */
933 if (info->cap.caps) {
934 if (ceph_snap(inode) == CEPH_NOSNAP) {
935 unsigned caps = le32_to_cpu(info->cap.caps);
936 ceph_add_cap(inode, session,
937 le64_to_cpu(info->cap.cap_id),
938 cap_fmode, caps,
939 le32_to_cpu(info->cap.wanted),
940 le32_to_cpu(info->cap.seq),
941 le32_to_cpu(info->cap.mseq),
942 le64_to_cpu(info->cap.realm),
943 info->cap.flags, &new_cap);
944
945 /* set dir completion flag? */
946 if (S_ISDIR(inode->i_mode) &&
947 ci->i_files == 0 && ci->i_subdirs == 0 &&
948 (caps & CEPH_CAP_FILE_SHARED) &&
949 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
950 !__ceph_dir_is_complete(ci)) {
951 dout(" marking %p complete (empty)\n", inode);
952 i_size_write(inode, 0);
953 __ceph_dir_set_complete(ci,
954 atomic64_read(&ci->i_release_count),
955 atomic64_read(&ci->i_ordered_count));
956 }
957
958 wake = true;
959 } else {
960 dout(" %p got snap_caps %s\n", inode,
961 ceph_cap_string(le32_to_cpu(info->cap.caps)));
962 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
963 if (cap_fmode >= 0)
964 __ceph_get_fmode(ci, cap_fmode);
965 }
966 } else if (cap_fmode >= 0) {
967 pr_warn("mds issued no caps on %llx.%llx\n",
968 ceph_vinop(inode));
969 __ceph_get_fmode(ci, cap_fmode);
970 }
971
972 if (iinfo->inline_version > 0 &&
973 iinfo->inline_version >= ci->i_inline_version) {
974 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
975 ci->i_inline_version = iinfo->inline_version;
976 if (ci->i_inline_version != CEPH_INLINE_NONE &&
977 (locked_page ||
978 (le32_to_cpu(info->cap.caps) & cache_caps)))
979 fill_inline = true;
980 }
981
982 spin_unlock(&ci->i_ceph_lock);
983
984 if (fill_inline)
985 ceph_fill_inline_data(inode, locked_page,
986 iinfo->inline_data, iinfo->inline_len);
987
988 if (wake)
989 wake_up_all(&ci->i_cap_wq);
990
991 /* queue truncate if we saw i_size decrease */
992 if (queue_trunc)
993 ceph_queue_vmtruncate(inode);
994
995 /* populate frag tree */
996 if (S_ISDIR(inode->i_mode))
997 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
998
999 /* update delegation info? */
1000 if (dirinfo)
1001 ceph_fill_dirfrag(inode, dirinfo);
1002
1003 err = 0;
1004out:
1005 if (new_cap)
1006 ceph_put_cap(mdsc, new_cap);
1007 if (xattr_blob)
1008 ceph_buffer_put(xattr_blob);
1009 ceph_put_string(pool_ns);
1010 return err;
1011}
1012
1013/*
1014 * caller should hold session s_mutex.
1015 */
1016static void update_dentry_lease(struct dentry *dentry,
1017 struct ceph_mds_reply_lease *lease,
1018 struct ceph_mds_session *session,
1019 unsigned long from_time)
1020{
1021 struct ceph_dentry_info *di = ceph_dentry(dentry);
1022 long unsigned duration = le32_to_cpu(lease->duration_ms);
1023 long unsigned ttl = from_time + (duration * HZ) / 1000;
1024 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1025 struct inode *dir;
1026
1027 spin_lock(&dentry->d_lock);
1028 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1029 dentry, duration, ttl);
1030
1031 /* make lease_rdcache_gen match directory */
1032 dir = d_inode(dentry->d_parent);
1033
1034 /* only track leases on regular dentries */
1035 if (ceph_snap(dir) != CEPH_NOSNAP)
1036 goto out_unlock;
1037
1038 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
1039
1040 if (duration == 0)
1041 goto out_unlock;
1042
1043 if (di->lease_gen == session->s_cap_gen &&
1044 time_before(ttl, di->time))
1045 goto out_unlock; /* we already have a newer lease. */
1046
1047 if (di->lease_session && di->lease_session != session)
1048 goto out_unlock;
1049
1050 ceph_dentry_lru_touch(dentry);
1051
1052 if (!di->lease_session)
1053 di->lease_session = ceph_get_mds_session(session);
1054 di->lease_gen = session->s_cap_gen;
1055 di->lease_seq = le32_to_cpu(lease->seq);
1056 di->lease_renew_after = half_ttl;
1057 di->lease_renew_from = 0;
1058 di->time = ttl;
1059out_unlock:
1060 spin_unlock(&dentry->d_lock);
1061 return;
1062}
1063
1064/*
1065 * splice a dentry to an inode.
1066 * caller must hold directory i_mutex for this to be safe.
1067 */
1068static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1069{
1070 struct dentry *realdn;
1071
1072 BUG_ON(d_inode(dn));
1073
1074 /* dn must be unhashed */
1075 if (!d_unhashed(dn))
1076 d_drop(dn);
1077 realdn = d_splice_alias(in, dn);
1078 if (IS_ERR(realdn)) {
1079 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1080 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1081 dn = realdn; /* note realdn contains the error */
1082 goto out;
1083 } else if (realdn) {
1084 dout("dn %p (%d) spliced with %p (%d) "
1085 "inode %p ino %llx.%llx\n",
1086 dn, d_count(dn),
1087 realdn, d_count(realdn),
1088 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1089 dput(dn);
1090 dn = realdn;
1091 } else {
1092 BUG_ON(!ceph_dentry(dn));
1093 dout("dn %p attached to %p ino %llx.%llx\n",
1094 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1095 }
1096out:
1097 return dn;
1098}
1099
1100/*
1101 * Incorporate results into the local cache. This is either just
1102 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1103 * after a lookup).
1104 *
1105 * A reply may contain
1106 * a directory inode along with a dentry.
1107 * and/or a target inode
1108 *
1109 * Called with snap_rwsem (read).
1110 */
1111int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1112 struct ceph_mds_session *session)
1113{
1114 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1115 struct inode *in = NULL;
1116 struct ceph_vino vino;
1117 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1118 int err = 0;
1119
1120 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1121 rinfo->head->is_dentry, rinfo->head->is_target);
1122
1123#if 0
1124 /*
1125 * Debugging hook:
1126 *
1127 * If we resend completed ops to a recovering mds, we get no
1128 * trace. Since that is very rare, pretend this is the case
1129 * to ensure the 'no trace' handlers in the callers behave.
1130 *
1131 * Fill in inodes unconditionally to avoid breaking cap
1132 * invariants.
1133 */
1134 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1135 pr_info("fill_trace faking empty trace on %lld %s\n",
1136 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1137 if (rinfo->head->is_dentry) {
1138 rinfo->head->is_dentry = 0;
1139 err = fill_inode(req->r_locked_dir,
1140 &rinfo->diri, rinfo->dirfrag,
1141 session, req->r_request_started, -1);
1142 }
1143 if (rinfo->head->is_target) {
1144 rinfo->head->is_target = 0;
1145 ininfo = rinfo->targeti.in;
1146 vino.ino = le64_to_cpu(ininfo->ino);
1147 vino.snap = le64_to_cpu(ininfo->snapid);
1148 in = ceph_get_inode(sb, vino);
1149 err = fill_inode(in, &rinfo->targeti, NULL,
1150 session, req->r_request_started,
1151 req->r_fmode);
1152 iput(in);
1153 }
1154 }
1155#endif
1156
1157 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1158 dout("fill_trace reply is empty!\n");
1159 if (rinfo->head->result == 0 && req->r_locked_dir)
1160 ceph_invalidate_dir_request(req);
1161 return 0;
1162 }
1163
1164 if (rinfo->head->is_dentry) {
1165 struct inode *dir = req->r_locked_dir;
1166
1167 if (dir) {
1168 err = fill_inode(dir, NULL,
1169 &rinfo->diri, rinfo->dirfrag,
1170 session, req->r_request_started, -1,
1171 &req->r_caps_reservation);
1172 if (err < 0)
1173 goto done;
1174 } else {
1175 WARN_ON_ONCE(1);
1176 }
1177
1178 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1179 struct qstr dname;
1180 struct dentry *dn, *parent;
1181
1182 BUG_ON(!rinfo->head->is_target);
1183 BUG_ON(req->r_dentry);
1184
1185 parent = d_find_any_alias(dir);
1186 BUG_ON(!parent);
1187
1188 dname.name = rinfo->dname;
1189 dname.len = rinfo->dname_len;
1190 dname.hash = full_name_hash(parent, dname.name, dname.len);
1191 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1192 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1193retry_lookup:
1194 dn = d_lookup(parent, &dname);
1195 dout("d_lookup on parent=%p name=%.*s got %p\n",
1196 parent, dname.len, dname.name, dn);
1197
1198 if (!dn) {
1199 dn = d_alloc(parent, &dname);
1200 dout("d_alloc %p '%.*s' = %p\n", parent,
1201 dname.len, dname.name, dn);
1202 if (dn == NULL) {
1203 dput(parent);
1204 err = -ENOMEM;
1205 goto done;
1206 }
1207 err = 0;
1208 } else if (d_really_is_positive(dn) &&
1209 (ceph_ino(d_inode(dn)) != vino.ino ||
1210 ceph_snap(d_inode(dn)) != vino.snap)) {
1211 dout(" dn %p points to wrong inode %p\n",
1212 dn, d_inode(dn));
1213 d_delete(dn);
1214 dput(dn);
1215 goto retry_lookup;
1216 }
1217
1218 req->r_dentry = dn;
1219 dput(parent);
1220 }
1221 }
1222
1223 if (rinfo->head->is_target) {
1224 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1225 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1226
1227 in = ceph_get_inode(sb, vino);
1228 if (IS_ERR(in)) {
1229 err = PTR_ERR(in);
1230 goto done;
1231 }
1232 req->r_target_inode = in;
1233
1234 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1235 session, req->r_request_started,
1236 (!req->r_aborted && rinfo->head->result == 0) ?
1237 req->r_fmode : -1,
1238 &req->r_caps_reservation);
1239 if (err < 0) {
1240 pr_err("fill_inode badness %p %llx.%llx\n",
1241 in, ceph_vinop(in));
1242 goto done;
1243 }
1244 }
1245
1246 /*
1247 * ignore null lease/binding on snapdir ENOENT, or else we
1248 * will have trouble splicing in the virtual snapdir later
1249 */
1250 if (rinfo->head->is_dentry && !req->r_aborted &&
1251 req->r_locked_dir &&
1252 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1253 fsc->mount_options->snapdir_name,
1254 req->r_dentry->d_name.len))) {
1255 /*
1256 * lookup link rename : null -> possibly existing inode
1257 * mknod symlink mkdir : null -> new inode
1258 * unlink : linked -> null
1259 */
1260 struct inode *dir = req->r_locked_dir;
1261 struct dentry *dn = req->r_dentry;
1262 bool have_dir_cap, have_lease;
1263
1264 BUG_ON(!dn);
1265 BUG_ON(!dir);
1266 BUG_ON(d_inode(dn->d_parent) != dir);
1267 BUG_ON(ceph_ino(dir) !=
1268 le64_to_cpu(rinfo->diri.in->ino));
1269 BUG_ON(ceph_snap(dir) !=
1270 le64_to_cpu(rinfo->diri.in->snapid));
1271
1272 /* do we have a lease on the whole dir? */
1273 have_dir_cap =
1274 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1275 CEPH_CAP_FILE_SHARED);
1276
1277 /* do we have a dn lease? */
1278 have_lease = have_dir_cap ||
1279 le32_to_cpu(rinfo->dlease->duration_ms);
1280 if (!have_lease)
1281 dout("fill_trace no dentry lease or dir cap\n");
1282
1283 /* rename? */
1284 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1285 struct inode *olddir = req->r_old_dentry_dir;
1286 BUG_ON(!olddir);
1287
1288 dout(" src %p '%pd' dst %p '%pd'\n",
1289 req->r_old_dentry,
1290 req->r_old_dentry,
1291 dn, dn);
1292 dout("fill_trace doing d_move %p -> %p\n",
1293 req->r_old_dentry, dn);
1294
1295 /* d_move screws up sibling dentries' offsets */
1296 ceph_dir_clear_ordered(dir);
1297 ceph_dir_clear_ordered(olddir);
1298
1299 d_move(req->r_old_dentry, dn);
1300 dout(" src %p '%pd' dst %p '%pd'\n",
1301 req->r_old_dentry,
1302 req->r_old_dentry,
1303 dn, dn);
1304
1305 /* ensure target dentry is invalidated, despite
1306 rehashing bug in vfs_rename_dir */
1307 ceph_invalidate_dentry_lease(dn);
1308
1309 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1310 ceph_dentry(req->r_old_dentry)->offset);
1311
1312 dn = req->r_old_dentry; /* use old_dentry */
1313 }
1314
1315 /* null dentry? */
1316 if (!rinfo->head->is_target) {
1317 dout("fill_trace null dentry\n");
1318 if (d_really_is_positive(dn)) {
1319 ceph_dir_clear_ordered(dir);
1320 dout("d_delete %p\n", dn);
1321 d_delete(dn);
1322 } else {
1323 if (have_lease && d_unhashed(dn))
1324 d_add(dn, NULL);
1325 update_dentry_lease(dn, rinfo->dlease,
1326 session,
1327 req->r_request_started);
1328 }
1329 goto done;
1330 }
1331
1332 /* attach proper inode */
1333 if (d_really_is_negative(dn)) {
1334 ceph_dir_clear_ordered(dir);
1335 ihold(in);
1336 dn = splice_dentry(dn, in);
1337 if (IS_ERR(dn)) {
1338 err = PTR_ERR(dn);
1339 goto done;
1340 }
1341 req->r_dentry = dn; /* may have spliced */
1342 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1343 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1344 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1345 ceph_vinop(in));
1346 d_invalidate(dn);
1347 have_lease = false;
1348 }
1349
1350 if (have_lease)
1351 update_dentry_lease(dn, rinfo->dlease, session,
1352 req->r_request_started);
1353 dout(" final dn %p\n", dn);
1354 } else if (!req->r_aborted &&
1355 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1356 req->r_op == CEPH_MDS_OP_MKSNAP)) {
1357 struct dentry *dn = req->r_dentry;
1358 struct inode *dir = req->r_locked_dir;
1359
1360 /* fill out a snapdir LOOKUPSNAP dentry */
1361 BUG_ON(!dn);
1362 BUG_ON(!dir);
1363 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1364 dout(" linking snapped dir %p to dn %p\n", in, dn);
1365 ceph_dir_clear_ordered(dir);
1366 ihold(in);
1367 dn = splice_dentry(dn, in);
1368 if (IS_ERR(dn)) {
1369 err = PTR_ERR(dn);
1370 goto done;
1371 }
1372 req->r_dentry = dn; /* may have spliced */
1373 }
1374done:
1375 dout("fill_trace done err=%d\n", err);
1376 return err;
1377}
1378
1379/*
1380 * Prepopulate our cache with readdir results, leases, etc.
1381 */
1382static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1383 struct ceph_mds_session *session)
1384{
1385 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1386 int i, err = 0;
1387
1388 for (i = 0; i < rinfo->dir_nr; i++) {
1389 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1390 struct ceph_vino vino;
1391 struct inode *in;
1392 int rc;
1393
1394 vino.ino = le64_to_cpu(rde->inode.in->ino);
1395 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1396
1397 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1398 if (IS_ERR(in)) {
1399 err = PTR_ERR(in);
1400 dout("new_inode badness got %d\n", err);
1401 continue;
1402 }
1403 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1404 req->r_request_started, -1,
1405 &req->r_caps_reservation);
1406 if (rc < 0) {
1407 pr_err("fill_inode badness on %p got %d\n", in, rc);
1408 err = rc;
1409 }
1410 iput(in);
1411 }
1412
1413 return err;
1414}
1415
1416void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1417{
1418 if (ctl->page) {
1419 kunmap(ctl->page);
1420 put_page(ctl->page);
1421 ctl->page = NULL;
1422 }
1423}
1424
1425static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1426 struct ceph_readdir_cache_control *ctl,
1427 struct ceph_mds_request *req)
1428{
1429 struct ceph_inode_info *ci = ceph_inode(dir);
1430 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1431 unsigned idx = ctl->index % nsize;
1432 pgoff_t pgoff = ctl->index / nsize;
1433
1434 if (!ctl->page || pgoff != page_index(ctl->page)) {
1435 ceph_readdir_cache_release(ctl);
1436 if (idx == 0)
1437 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1438 else
1439 ctl->page = find_lock_page(&dir->i_data, pgoff);
1440 if (!ctl->page) {
1441 ctl->index = -1;
1442 return idx == 0 ? -ENOMEM : 0;
1443 }
1444 /* reading/filling the cache are serialized by
1445 * i_mutex, no need to use page lock */
1446 unlock_page(ctl->page);
1447 ctl->dentries = kmap(ctl->page);
1448 if (idx == 0)
1449 memset(ctl->dentries, 0, PAGE_SIZE);
1450 }
1451
1452 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1453 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1454 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1455 ctl->dentries[idx] = dn;
1456 ctl->index++;
1457 } else {
1458 dout("disable readdir cache\n");
1459 ctl->index = -1;
1460 }
1461 return 0;
1462}
1463
1464int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1465 struct ceph_mds_session *session)
1466{
1467 struct dentry *parent = req->r_dentry;
1468 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1469 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1470 struct qstr dname;
1471 struct dentry *dn;
1472 struct inode *in;
1473 int err = 0, skipped = 0, ret, i;
1474 struct inode *snapdir = NULL;
1475 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1476 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1477 u32 last_hash = 0;
1478 u32 fpos_offset;
1479 struct ceph_readdir_cache_control cache_ctl = {};
1480
1481 if (req->r_aborted)
1482 return readdir_prepopulate_inodes_only(req, session);
1483
1484 if (rinfo->hash_order && req->r_path2) {
1485 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1486 req->r_path2, strlen(req->r_path2));
1487 last_hash = ceph_frag_value(last_hash);
1488 }
1489
1490 if (rinfo->dir_dir &&
1491 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1492 dout("readdir_prepopulate got new frag %x -> %x\n",
1493 frag, le32_to_cpu(rinfo->dir_dir->frag));
1494 frag = le32_to_cpu(rinfo->dir_dir->frag);
1495 if (!rinfo->hash_order)
1496 req->r_readdir_offset = 2;
1497 }
1498
1499 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1500 snapdir = ceph_get_snapdir(d_inode(parent));
1501 parent = d_find_alias(snapdir);
1502 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1503 rinfo->dir_nr, parent);
1504 } else {
1505 dout("readdir_prepopulate %d items under dn %p\n",
1506 rinfo->dir_nr, parent);
1507 if (rinfo->dir_dir)
1508 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1509 }
1510
1511 if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2 &&
1512 !(rinfo->hash_order && req->r_path2)) {
1513 /* note dir version at start of readdir so we can tell
1514 * if any dentries get dropped */
1515 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1516 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1517 req->r_readdir_cache_idx = 0;
1518 }
1519
1520 cache_ctl.index = req->r_readdir_cache_idx;
1521 fpos_offset = req->r_readdir_offset;
1522
1523 /* FIXME: release caps/leases if error occurs */
1524 for (i = 0; i < rinfo->dir_nr; i++) {
1525 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1526 struct ceph_vino vino;
1527
1528 dname.name = rde->name;
1529 dname.len = rde->name_len;
1530 dname.hash = full_name_hash(parent, dname.name, dname.len);
1531
1532 vino.ino = le64_to_cpu(rde->inode.in->ino);
1533 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1534
1535 if (rinfo->hash_order) {
1536 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1537 rde->name, rde->name_len);
1538 hash = ceph_frag_value(hash);
1539 if (hash != last_hash)
1540 fpos_offset = 2;
1541 last_hash = hash;
1542 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1543 } else {
1544 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1545 }
1546
1547retry_lookup:
1548 dn = d_lookup(parent, &dname);
1549 dout("d_lookup on parent=%p name=%.*s got %p\n",
1550 parent, dname.len, dname.name, dn);
1551
1552 if (!dn) {
1553 dn = d_alloc(parent, &dname);
1554 dout("d_alloc %p '%.*s' = %p\n", parent,
1555 dname.len, dname.name, dn);
1556 if (dn == NULL) {
1557 dout("d_alloc badness\n");
1558 err = -ENOMEM;
1559 goto out;
1560 }
1561 } else if (d_really_is_positive(dn) &&
1562 (ceph_ino(d_inode(dn)) != vino.ino ||
1563 ceph_snap(d_inode(dn)) != vino.snap)) {
1564 dout(" dn %p points to wrong inode %p\n",
1565 dn, d_inode(dn));
1566 d_delete(dn);
1567 dput(dn);
1568 goto retry_lookup;
1569 }
1570
1571 /* inode */
1572 if (d_really_is_positive(dn)) {
1573 in = d_inode(dn);
1574 } else {
1575 in = ceph_get_inode(parent->d_sb, vino);
1576 if (IS_ERR(in)) {
1577 dout("new_inode badness\n");
1578 d_drop(dn);
1579 dput(dn);
1580 err = PTR_ERR(in);
1581 goto out;
1582 }
1583 }
1584
1585 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1586 req->r_request_started, -1,
1587 &req->r_caps_reservation);
1588 if (ret < 0) {
1589 pr_err("fill_inode badness on %p\n", in);
1590 if (d_really_is_negative(dn))
1591 iput(in);
1592 d_drop(dn);
1593 err = ret;
1594 goto next_item;
1595 }
1596
1597 if (d_really_is_negative(dn)) {
1598 struct dentry *realdn;
1599
1600 if (ceph_security_xattr_deadlock(in)) {
1601 dout(" skip splicing dn %p to inode %p"
1602 " (security xattr deadlock)\n", dn, in);
1603 iput(in);
1604 skipped++;
1605 goto next_item;
1606 }
1607
1608 realdn = splice_dentry(dn, in);
1609 if (IS_ERR(realdn)) {
1610 err = PTR_ERR(realdn);
1611 d_drop(dn);
1612 dn = NULL;
1613 goto next_item;
1614 }
1615 dn = realdn;
1616 }
1617
1618 ceph_dentry(dn)->offset = rde->offset;
1619
1620 update_dentry_lease(dn, rde->lease, req->r_session,
1621 req->r_request_started);
1622
1623 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1624 ret = fill_readdir_cache(d_inode(parent), dn,
1625 &cache_ctl, req);
1626 if (ret < 0)
1627 err = ret;
1628 }
1629next_item:
1630 if (dn)
1631 dput(dn);
1632 }
1633out:
1634 if (err == 0 && skipped == 0) {
1635 req->r_did_prepopulate = true;
1636 req->r_readdir_cache_idx = cache_ctl.index;
1637 }
1638 ceph_readdir_cache_release(&cache_ctl);
1639 if (snapdir) {
1640 iput(snapdir);
1641 dput(parent);
1642 }
1643 dout("readdir_prepopulate done\n");
1644 return err;
1645}
1646
1647int ceph_inode_set_size(struct inode *inode, loff_t size)
1648{
1649 struct ceph_inode_info *ci = ceph_inode(inode);
1650 int ret = 0;
1651
1652 spin_lock(&ci->i_ceph_lock);
1653 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1654 i_size_write(inode, size);
1655 inode->i_blocks = calc_inode_blocks(size);
1656
1657 /* tell the MDS if we are approaching max_size */
1658 if ((size << 1) >= ci->i_max_size &&
1659 (ci->i_reported_size << 1) < ci->i_max_size)
1660 ret = 1;
1661
1662 spin_unlock(&ci->i_ceph_lock);
1663 return ret;
1664}
1665
1666/*
1667 * Write back inode data in a worker thread. (This can't be done
1668 * in the message handler context.)
1669 */
1670void ceph_queue_writeback(struct inode *inode)
1671{
1672 ihold(inode);
1673 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1674 &ceph_inode(inode)->i_wb_work)) {
1675 dout("ceph_queue_writeback %p\n", inode);
1676 } else {
1677 dout("ceph_queue_writeback %p failed\n", inode);
1678 iput(inode);
1679 }
1680}
1681
1682static void ceph_writeback_work(struct work_struct *work)
1683{
1684 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1685 i_wb_work);
1686 struct inode *inode = &ci->vfs_inode;
1687
1688 dout("writeback %p\n", inode);
1689 filemap_fdatawrite(&inode->i_data);
1690 iput(inode);
1691}
1692
1693/*
1694 * queue an async invalidation
1695 */
1696void ceph_queue_invalidate(struct inode *inode)
1697{
1698 ihold(inode);
1699 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1700 &ceph_inode(inode)->i_pg_inv_work)) {
1701 dout("ceph_queue_invalidate %p\n", inode);
1702 } else {
1703 dout("ceph_queue_invalidate %p failed\n", inode);
1704 iput(inode);
1705 }
1706}
1707
1708/*
1709 * Invalidate inode pages in a worker thread. (This can't be done
1710 * in the message handler context.)
1711 */
1712static void ceph_invalidate_work(struct work_struct *work)
1713{
1714 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1715 i_pg_inv_work);
1716 struct inode *inode = &ci->vfs_inode;
1717 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1718 u32 orig_gen;
1719 int check = 0;
1720
1721 mutex_lock(&ci->i_truncate_mutex);
1722
1723 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1724 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1725 inode, ceph_ino(inode));
1726 mapping_set_error(inode->i_mapping, -EIO);
1727 truncate_pagecache(inode, 0);
1728 mutex_unlock(&ci->i_truncate_mutex);
1729 goto out;
1730 }
1731
1732 spin_lock(&ci->i_ceph_lock);
1733 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1734 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1735 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1736 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1737 check = 1;
1738 spin_unlock(&ci->i_ceph_lock);
1739 mutex_unlock(&ci->i_truncate_mutex);
1740 goto out;
1741 }
1742 orig_gen = ci->i_rdcache_gen;
1743 spin_unlock(&ci->i_ceph_lock);
1744
1745 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1746 pr_err("invalidate_pages %p fails\n", inode);
1747 }
1748
1749 spin_lock(&ci->i_ceph_lock);
1750 if (orig_gen == ci->i_rdcache_gen &&
1751 orig_gen == ci->i_rdcache_revoking) {
1752 dout("invalidate_pages %p gen %d successful\n", inode,
1753 ci->i_rdcache_gen);
1754 ci->i_rdcache_revoking--;
1755 check = 1;
1756 } else {
1757 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1758 inode, orig_gen, ci->i_rdcache_gen,
1759 ci->i_rdcache_revoking);
1760 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1761 check = 1;
1762 }
1763 spin_unlock(&ci->i_ceph_lock);
1764 mutex_unlock(&ci->i_truncate_mutex);
1765out:
1766 if (check)
1767 ceph_check_caps(ci, 0, NULL);
1768 iput(inode);
1769}
1770
1771
1772/*
1773 * called by trunc_wq;
1774 *
1775 * We also truncate in a separate thread as well.
1776 */
1777static void ceph_vmtruncate_work(struct work_struct *work)
1778{
1779 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1780 i_vmtruncate_work);
1781 struct inode *inode = &ci->vfs_inode;
1782
1783 dout("vmtruncate_work %p\n", inode);
1784 __ceph_do_pending_vmtruncate(inode);
1785 iput(inode);
1786}
1787
1788/*
1789 * Queue an async vmtruncate. If we fail to queue work, we will handle
1790 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1791 */
1792void ceph_queue_vmtruncate(struct inode *inode)
1793{
1794 struct ceph_inode_info *ci = ceph_inode(inode);
1795
1796 ihold(inode);
1797
1798 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1799 &ci->i_vmtruncate_work)) {
1800 dout("ceph_queue_vmtruncate %p\n", inode);
1801 } else {
1802 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1803 inode, ci->i_truncate_pending);
1804 iput(inode);
1805 }
1806}
1807
1808/*
1809 * Make sure any pending truncation is applied before doing anything
1810 * that may depend on it.
1811 */
1812void __ceph_do_pending_vmtruncate(struct inode *inode)
1813{
1814 struct ceph_inode_info *ci = ceph_inode(inode);
1815 u64 to;
1816 int wrbuffer_refs, finish = 0;
1817
1818 mutex_lock(&ci->i_truncate_mutex);
1819retry:
1820 spin_lock(&ci->i_ceph_lock);
1821 if (ci->i_truncate_pending == 0) {
1822 dout("__do_pending_vmtruncate %p none pending\n", inode);
1823 spin_unlock(&ci->i_ceph_lock);
1824 mutex_unlock(&ci->i_truncate_mutex);
1825 return;
1826 }
1827
1828 /*
1829 * make sure any dirty snapped pages are flushed before we
1830 * possibly truncate them.. so write AND block!
1831 */
1832 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1833 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1834 inode);
1835 spin_unlock(&ci->i_ceph_lock);
1836 filemap_write_and_wait_range(&inode->i_data, 0,
1837 inode->i_sb->s_maxbytes);
1838 goto retry;
1839 }
1840
1841 /* there should be no reader or writer */
1842 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1843
1844 to = ci->i_truncate_size;
1845 wrbuffer_refs = ci->i_wrbuffer_ref;
1846 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1847 ci->i_truncate_pending, to);
1848 spin_unlock(&ci->i_ceph_lock);
1849
1850 truncate_pagecache(inode, to);
1851
1852 spin_lock(&ci->i_ceph_lock);
1853 if (to == ci->i_truncate_size) {
1854 ci->i_truncate_pending = 0;
1855 finish = 1;
1856 }
1857 spin_unlock(&ci->i_ceph_lock);
1858 if (!finish)
1859 goto retry;
1860
1861 mutex_unlock(&ci->i_truncate_mutex);
1862
1863 if (wrbuffer_refs == 0)
1864 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1865
1866 wake_up_all(&ci->i_cap_wq);
1867}
1868
1869/*
1870 * symlinks
1871 */
1872static const struct inode_operations ceph_symlink_iops = {
1873 .get_link = simple_get_link,
1874 .setattr = ceph_setattr,
1875 .getattr = ceph_getattr,
1876 .listxattr = ceph_listxattr,
1877};
1878
1879int __ceph_setattr(struct inode *inode, struct iattr *attr)
1880{
1881 struct ceph_inode_info *ci = ceph_inode(inode);
1882 const unsigned int ia_valid = attr->ia_valid;
1883 struct ceph_mds_request *req;
1884 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1885 struct ceph_cap_flush *prealloc_cf;
1886 int issued;
1887 int release = 0, dirtied = 0;
1888 int mask = 0;
1889 int err = 0;
1890 int inode_dirty_flags = 0;
1891 bool lock_snap_rwsem = false;
1892
1893 prealloc_cf = ceph_alloc_cap_flush();
1894 if (!prealloc_cf)
1895 return -ENOMEM;
1896
1897 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1898 USE_AUTH_MDS);
1899 if (IS_ERR(req)) {
1900 ceph_free_cap_flush(prealloc_cf);
1901 return PTR_ERR(req);
1902 }
1903
1904 spin_lock(&ci->i_ceph_lock);
1905 issued = __ceph_caps_issued(ci, NULL);
1906
1907 if (!ci->i_head_snapc &&
1908 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1909 lock_snap_rwsem = true;
1910 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1911 spin_unlock(&ci->i_ceph_lock);
1912 down_read(&mdsc->snap_rwsem);
1913 spin_lock(&ci->i_ceph_lock);
1914 issued = __ceph_caps_issued(ci, NULL);
1915 }
1916 }
1917
1918 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1919
1920 if (ia_valid & ATTR_UID) {
1921 dout("setattr %p uid %d -> %d\n", inode,
1922 from_kuid(&init_user_ns, inode->i_uid),
1923 from_kuid(&init_user_ns, attr->ia_uid));
1924 if (issued & CEPH_CAP_AUTH_EXCL) {
1925 inode->i_uid = attr->ia_uid;
1926 dirtied |= CEPH_CAP_AUTH_EXCL;
1927 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1928 !uid_eq(attr->ia_uid, inode->i_uid)) {
1929 req->r_args.setattr.uid = cpu_to_le32(
1930 from_kuid(&init_user_ns, attr->ia_uid));
1931 mask |= CEPH_SETATTR_UID;
1932 release |= CEPH_CAP_AUTH_SHARED;
1933 }
1934 }
1935 if (ia_valid & ATTR_GID) {
1936 dout("setattr %p gid %d -> %d\n", inode,
1937 from_kgid(&init_user_ns, inode->i_gid),
1938 from_kgid(&init_user_ns, attr->ia_gid));
1939 if (issued & CEPH_CAP_AUTH_EXCL) {
1940 inode->i_gid = attr->ia_gid;
1941 dirtied |= CEPH_CAP_AUTH_EXCL;
1942 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1943 !gid_eq(attr->ia_gid, inode->i_gid)) {
1944 req->r_args.setattr.gid = cpu_to_le32(
1945 from_kgid(&init_user_ns, attr->ia_gid));
1946 mask |= CEPH_SETATTR_GID;
1947 release |= CEPH_CAP_AUTH_SHARED;
1948 }
1949 }
1950 if (ia_valid & ATTR_MODE) {
1951 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1952 attr->ia_mode);
1953 if (issued & CEPH_CAP_AUTH_EXCL) {
1954 inode->i_mode = attr->ia_mode;
1955 dirtied |= CEPH_CAP_AUTH_EXCL;
1956 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1957 attr->ia_mode != inode->i_mode) {
1958 inode->i_mode = attr->ia_mode;
1959 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1960 mask |= CEPH_SETATTR_MODE;
1961 release |= CEPH_CAP_AUTH_SHARED;
1962 }
1963 }
1964
1965 if (ia_valid & ATTR_ATIME) {
1966 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1967 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1968 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1969 if (issued & CEPH_CAP_FILE_EXCL) {
1970 ci->i_time_warp_seq++;
1971 inode->i_atime = attr->ia_atime;
1972 dirtied |= CEPH_CAP_FILE_EXCL;
1973 } else if ((issued & CEPH_CAP_FILE_WR) &&
1974 timespec_compare(&inode->i_atime,
1975 &attr->ia_atime) < 0) {
1976 inode->i_atime = attr->ia_atime;
1977 dirtied |= CEPH_CAP_FILE_WR;
1978 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1979 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1980 ceph_encode_timespec(&req->r_args.setattr.atime,
1981 &attr->ia_atime);
1982 mask |= CEPH_SETATTR_ATIME;
1983 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1984 CEPH_CAP_FILE_WR;
1985 }
1986 }
1987 if (ia_valid & ATTR_MTIME) {
1988 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1989 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1990 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1991 if (issued & CEPH_CAP_FILE_EXCL) {
1992 ci->i_time_warp_seq++;
1993 inode->i_mtime = attr->ia_mtime;
1994 dirtied |= CEPH_CAP_FILE_EXCL;
1995 } else if ((issued & CEPH_CAP_FILE_WR) &&
1996 timespec_compare(&inode->i_mtime,
1997 &attr->ia_mtime) < 0) {
1998 inode->i_mtime = attr->ia_mtime;
1999 dirtied |= CEPH_CAP_FILE_WR;
2000 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2001 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2002 ceph_encode_timespec(&req->r_args.setattr.mtime,
2003 &attr->ia_mtime);
2004 mask |= CEPH_SETATTR_MTIME;
2005 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2006 CEPH_CAP_FILE_WR;
2007 }
2008 }
2009 if (ia_valid & ATTR_SIZE) {
2010 dout("setattr %p size %lld -> %lld\n", inode,
2011 inode->i_size, attr->ia_size);
2012 if ((issued & CEPH_CAP_FILE_EXCL) &&
2013 attr->ia_size > inode->i_size) {
2014 i_size_write(inode, attr->ia_size);
2015 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2016 inode->i_ctime = attr->ia_ctime;
2017 ci->i_reported_size = attr->ia_size;
2018 dirtied |= CEPH_CAP_FILE_EXCL;
2019 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2020 attr->ia_size != inode->i_size) {
2021 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2022 req->r_args.setattr.old_size =
2023 cpu_to_le64(inode->i_size);
2024 mask |= CEPH_SETATTR_SIZE;
2025 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2026 CEPH_CAP_FILE_WR;
2027 }
2028 }
2029
2030 /* these do nothing */
2031 if (ia_valid & ATTR_CTIME) {
2032 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2033 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2034 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2035 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2036 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2037 only ? "ctime only" : "ignored");
2038 inode->i_ctime = attr->ia_ctime;
2039 if (only) {
2040 /*
2041 * if kernel wants to dirty ctime but nothing else,
2042 * we need to choose a cap to dirty under, or do
2043 * a almost-no-op setattr
2044 */
2045 if (issued & CEPH_CAP_AUTH_EXCL)
2046 dirtied |= CEPH_CAP_AUTH_EXCL;
2047 else if (issued & CEPH_CAP_FILE_EXCL)
2048 dirtied |= CEPH_CAP_FILE_EXCL;
2049 else if (issued & CEPH_CAP_XATTR_EXCL)
2050 dirtied |= CEPH_CAP_XATTR_EXCL;
2051 else
2052 mask |= CEPH_SETATTR_CTIME;
2053 }
2054 }
2055 if (ia_valid & ATTR_FILE)
2056 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2057
2058 if (dirtied) {
2059 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2060 &prealloc_cf);
2061 inode->i_ctime = current_time(inode);
2062 }
2063
2064 release &= issued;
2065 spin_unlock(&ci->i_ceph_lock);
2066 if (lock_snap_rwsem)
2067 up_read(&mdsc->snap_rwsem);
2068
2069 if (inode_dirty_flags)
2070 __mark_inode_dirty(inode, inode_dirty_flags);
2071
2072 if (ia_valid & ATTR_MODE) {
2073 err = posix_acl_chmod(inode, attr->ia_mode);
2074 if (err)
2075 goto out_put;
2076 }
2077
2078 if (mask) {
2079 req->r_inode = inode;
2080 ihold(inode);
2081 req->r_inode_drop = release;
2082 req->r_args.setattr.mask = cpu_to_le32(mask);
2083 req->r_num_caps = 1;
2084 err = ceph_mdsc_do_request(mdsc, NULL, req);
2085 }
2086 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2087 ceph_cap_string(dirtied), mask);
2088
2089 ceph_mdsc_put_request(req);
2090 if (mask & CEPH_SETATTR_SIZE)
2091 __ceph_do_pending_vmtruncate(inode);
2092 ceph_free_cap_flush(prealloc_cf);
2093 return err;
2094out_put:
2095 ceph_mdsc_put_request(req);
2096 ceph_free_cap_flush(prealloc_cf);
2097 return err;
2098}
2099
2100/*
2101 * setattr
2102 */
2103int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2104{
2105 struct inode *inode = d_inode(dentry);
2106 int err;
2107
2108 if (ceph_snap(inode) != CEPH_NOSNAP)
2109 return -EROFS;
2110
2111 err = setattr_prepare(dentry, attr);
2112 if (err != 0)
2113 return err;
2114
2115 return __ceph_setattr(inode, attr);
2116}
2117
2118/*
2119 * Verify that we have a lease on the given mask. If not,
2120 * do a getattr against an mds.
2121 */
2122int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2123 int mask, bool force)
2124{
2125 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2126 struct ceph_mds_client *mdsc = fsc->mdsc;
2127 struct ceph_mds_request *req;
2128 int err;
2129
2130 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2131 dout("do_getattr inode %p SNAPDIR\n", inode);
2132 return 0;
2133 }
2134
2135 dout("do_getattr inode %p mask %s mode 0%o\n",
2136 inode, ceph_cap_string(mask), inode->i_mode);
2137 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2138 return 0;
2139
2140 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2141 if (IS_ERR(req))
2142 return PTR_ERR(req);
2143 req->r_inode = inode;
2144 ihold(inode);
2145 req->r_num_caps = 1;
2146 req->r_args.getattr.mask = cpu_to_le32(mask);
2147 req->r_locked_page = locked_page;
2148 err = ceph_mdsc_do_request(mdsc, NULL, req);
2149 if (locked_page && err == 0) {
2150 u64 inline_version = req->r_reply_info.targeti.inline_version;
2151 if (inline_version == 0) {
2152 /* the reply is supposed to contain inline data */
2153 err = -EINVAL;
2154 } else if (inline_version == CEPH_INLINE_NONE) {
2155 err = -ENODATA;
2156 } else {
2157 err = req->r_reply_info.targeti.inline_len;
2158 }
2159 }
2160 ceph_mdsc_put_request(req);
2161 dout("do_getattr result=%d\n", err);
2162 return err;
2163}
2164
2165
2166/*
2167 * Check inode permissions. We verify we have a valid value for
2168 * the AUTH cap, then call the generic handler.
2169 */
2170int ceph_permission(struct inode *inode, int mask)
2171{
2172 int err;
2173
2174 if (mask & MAY_NOT_BLOCK)
2175 return -ECHILD;
2176
2177 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2178
2179 if (!err)
2180 err = generic_permission(inode, mask);
2181 return err;
2182}
2183
2184/*
2185 * Get all attributes. Hopefully somedata we'll have a statlite()
2186 * and can limit the fields we require to be accurate.
2187 */
2188int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2189 struct kstat *stat)
2190{
2191 struct inode *inode = d_inode(dentry);
2192 struct ceph_inode_info *ci = ceph_inode(inode);
2193 int err;
2194
2195 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2196 if (!err) {
2197 generic_fillattr(inode, stat);
2198 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2199 if (ceph_snap(inode) != CEPH_NOSNAP)
2200 stat->dev = ceph_snap(inode);
2201 else
2202 stat->dev = 0;
2203 if (S_ISDIR(inode->i_mode)) {
2204 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2205 RBYTES))
2206 stat->size = ci->i_rbytes;
2207 else
2208 stat->size = ci->i_files + ci->i_subdirs;
2209 stat->blocks = 0;
2210 stat->blksize = 65536;
2211 }
2212 }
2213 return err;
2214}